INTEGRATING CAPABILITY AND HAPPINESS APPROACHES IN … · subjective wellbeing (SWB) provide distinctive information, not contained in the happiness indicator, while together with

INTEGRATING CAPABILITY AND HAPPINESS APPROACHES IN WELFARE ASSESSMENT: A

THEORETICAL AND EMPIRICAL INVESTIGATION

Submitted by

Hamid Hasan

MPhil, MSc, BSc

A thesis submitted in total fulfilment of the requirements for the degree of

Doctor of Philosophy

School of Economics

Faculty of Business, Economics and Law

La Trobe University

Bundoora, Victoria 3086

Australia

September, 2012

Supervised by Prof. Sisira Jayasuriya, Prof. Gary Magee, and Dr. Hayat Khan

ii | P a g e

Table of Contents List of Tables ...................................................................................................... v List of Figures ................................................................................................... vii Abstract .............................................................................................................viii Statement of Authorship ................................................................................... xi Acknowledgements .......................................................................................... xii

Chapter 1 Introduction ....................................................................................... 1 1.1 Motivation and Background .................................................................... 1

1.2 Sen’s Capabilities Approach .................................................................. 5

1.2.1 Functionings ....................................................................................... 6

1.2.2 Conversion Efficiency ......................................................................... 7

1.2.3 Freedom ............................................................................................. 7

1.3 Issues in Operationalising Sen’s Capabilities Approach and

Approach of the Thesis ........................................................................ 10

1.3.1 Measurement of Functionings .......................................................... 11

1.3.2 Selection of Functionings .................................................................. 12

1.3.3 Aggregation of and Trade-off between Functionings ........................ 14

1.3.4 Measurement of Ability to Achieve a Functioning ............................. 14

1.3.5 Measurement of Freedom to Achieve a Functioning ........................ 15

1.3.6 Circularity in Capability Dimensions .................................................. 17

1.3.7 Problem of Incorporating Unobserved Human Diversity ................... 18

1.4 Contributions, Objectives and Structure of the Thesis.......................... 19

Chapter 2 Measuring Happiness and Capabilities ......................................... 23 2.1 Introduction .......................................................................................... 23

2.2 Background .......................................................................................... 24

2.3 Empirical Capability Literature: A Brief Review .................................... 28

2.3.1 Measurement of Achieved Functionings ........................................... 28

2.3.2 Measurement of Capabilities or Achievable Functionings ................ 29

2.3.3 Measurement of Conversion Efficiency of the Utilisation Function ... 31

2.4 Methodology ......................................................................................... 32

iii | P a g e

2.4.1 Selection of a Functioning ................................................................ 32

2.4.2 Measurement of Capability ............................................................... 33

2.4.3 A Comparison of Dimensions of Wellbeing ....................................... 36

2.5 The Data .............................................................................................. 37

2.6 Results and Discussion ........................................................................ 42

2.7 Conclusions and Policy Implications .................................................... 47

Appendix A2 ...................................................................................................... 50

Chapter 3 Modeling Happiness with Capabilities .......................................... 51 3.1 Introduction .......................................................................................... 51

3.2 A Brief Review of Empirical Literature .................................................. 52

3.3 Methodology ......................................................................................... 54


3.5 Conlusions and Policy implications ...................................................... 56

Appendix A3 ....................................................................................................... 58

Chapter 4 Robustness of the Link between Happiness and Capabilities .... 60

4.1 Introduction .......................................................................................... 60

4.2 Methodology ......................................................................................... 62

4.3 Estimation Process and Sensitivity of Estimation Results to

Structural Models ................................................................................. 64

4.3.1 Stage-I – PLS Path Analysis ............................................................. 64

4.3.2 Stage-II – Regression Analysis ......................................................... 72


4.5 Conclusion and Policy Implications: ..................................................... 73

Appendix A4 ....................................................................................................... 75

Chapter 5 Interaction of Capability Dimensions: A Theoretical Analysis.... 80 5.1 Introduction .......................................................................................... 80

5.2 Selection of the Model .......................................................................... 81

5.3 Dynamic modeling using Bootstrapping ............................................... 82

5.4 Theoretical Dynamics of the Model ...................................................... 83

5.5 Conclusion and Policy Implications ...................................................... 85

Appendix A5 ....................................................................................................... 87

iv | P a g e

Chapter 6 Capability Dimensions and Their Correspondence with Income and Education ................................................................................................... 89

6.1 Introduction .......................................................................................... 89

6.2 A Brief Review of Literature: ................................................................. 90

6.3 Ordered Choice Models ....................................................................... 90

6.4 Description of Variables ....................................................................... 93

6.5 Model Estimation and Results .............................................................. 96


6.7 Conclusion and Policy Implications: ..................................................... 98

Appendix A6 ......................................................................................................100

Chapter 7 Testing the Existence of Hedonic Adaption to Income in PSES Panel .................................................................................................................107

7.1 Introduction ........................................................................................ 107

7.2 Hedonic Adaptation Models: A Brief Literature Review ...................... 107

7.3 Model Estimation ................................................................................ 109

7.4 Results and Discussion ...................................................................... 110

7.5 Conclusion and Policy Implications .................................................... 112

Appendix A7 ......................................................................................................113

Chapter 8 Conclusions....................................................................................121 8.1 The Perspective ................................................................................. 121

8.2 Findings of the Thesis and Their Implications .................................... 125

8.3 Limitations and Future Directions ....................................................... 126

Bibliography.....................................................................................................129 Appendix B.......................................................................................................144 Appendix C.......................................................................................................164 Appendix D.......................................................................................................169

v | P a g e

List of Tables

Table 1.1: Indicator: Sense of freedom to achieve 39

Table 1.2: Indicator: Sense of Ability to achieve 40

Table 1.3: Indicator: Sense of achievement 41

Table 1.4: Indicator: Happiness 41

Table 2.1 District ranking 46

Table A3.1: OLS estimates: Dependent variable = Happiness 58

Table A3.2: 3SLS estimates 58

Table A3.3: Ordered Logit estimates (dependent variable= Happiness) 59

Table A4.1: OLS estimates-Dependent variable = Happiness 75

Table A4.2: 3SLS estimates 75

Table A4.3: Ordered Logit estimates: Dependent variable = Happiness 75

Table A4.4 Estimates under different structural forms 75

Table A4.5: Comparison of PLS-PM and LISREL 78

Table 5.1: Scenario 1. If you target E (when E/R<β and R is fixed) F=

αE/β would increase at a slower rate than E.

85

Table 5.2: Scenario 2. If you target R (when E/R> β and E is fixed) F=

αE/ β would increase at a faster rate than E.

85

Table A5.1: Policy targets at district level 87

Table A5.2: Correlation Analysis 88

Table A6.1: Marginal Effects (Discrete Changes) for freedom: comparison

of ordered logit models

100

Table A6.2: Marginal Effects (Discrete Changes) for efficiency:

comparison of ordered logit models

101

Table A6.3: Differences between the marginal effects on Efficiency (e)

and Freedom (r)

102

Table A6.4: Ordered logit models: estimation and post estimation results 104

Table 7.1: Random Effects Ordered Probit Models 111

Table A7.1: Model 1 with current and lag nominal income 113

Table A7.2: Model 2 with differenced nominal income 114

Table A7.3: Model 3 with current and lag real income(Layard, 2006) 115

Table A7.4: Model 4 with first order difference of real income 116

vi | P a g e

Table A7.5: Model 5 with current and differenced real income (Clark et al,

2006)

117

Table A7.6: Model 6 with differenced happiness and differenced real

income (Ferrer-i-Carbonell and Van Praag, 2008)

118

Table A7.7: Model 7 with autoregressive happiness, current, and lag real

income (Botton and Truglia, 2011)

119

Table A7.8: Model 8 with autoregressive happiness and differenced real

income

120

vii | P a g e

List of Figures

Graph 2.1: Boxplots for sense of ability to achieve (SATA), sense of

achievement (SA), sense of freedom to achieve (SFTA), and

happiness (HAPP) based on individual data.

43

Graph 2.2: Histograms for sense of ability to achieve (SATA), sense of


happiness (HAPP).

43

Graph 2.3: Boxplots for sense of ability to achieve (SATA), sense of


happiness (HAPP) based on district level data.

44

Graph 2.4: Histograms for sense of ability to achieve (SATA), sense of


happiness (HAPP) based on district level data.

44

Figure 4.1: Latent Variable Model 1 (Path Diagram 1) for Happiness 66




Figure A4.1: Happiness rankings against Base case 76

Figure A4.2: Sense of Achievement (SA) rankings against Base case 76

Figure A4.3: Sense of Ability to Achieve (SATA) rankings against Base

case

76

Figure A4.4: Sense of Freedom to Achieve (SFTA) rankings against Base

case

77

Figure A4.5: SA, SATA, and SFTA rankings against HAPPINESS 77

Graph 5.1: Policy emphasis regions for the low-efficiency scenario 84

Graph 5.2: Policy emphasis regions for the low-freedom scenario 84

Graph A6.1: Behaviour of predicted probabilities of extreme categories for

freedom and efficiency against education and income

103

viii | P a g e

Abstract

Traditionally, affluence and poverty have been measured by a money-metric

measure to reflect the level of economic wellbeing. Sen’s seminal work on

capabilities, among others, brought a new dimension to the debate on human

welfare, primarily building up on the idea that a money-metric does not provide a

reasonable measure of human wellbeing. Evidence suggests that poverty is not

only associated with lack or low level of income, but also with different forms of

deprivations (for example, illness, illiteracy, malnutrition, and deprivation of rights

and freedoms). These deprivations inspired Sen’s well-known capabilities

framework.

The two competing approaches that added other dimensions to overcome the

shortcomings of the one-dimensional resource-based approach are the

Happiness Approach (HA) and the Capabilities Approach (CA). HA revived

utilitarianism but with an empirical application, while the capabilities approach

(CA) criticised utilitarianism as the sole criterion for human wellbeing

assessment.

The capabilities approach revolves around three key concepts; functioning,

freedom, and conversion efficiency. This provides a broader and rich

informational space to policy makers but faces challenges on the operational side

(including selection, measurement, and aggregation of capabilities).

This thesis integrates Sen’s CA with HA and investigates whether or not there

is any correspondence between the two. The focus is on addressing issues

related to selection, measurement, and aggregation of capabilities in a

parsimonious manner. It is argued that most of these issues can greatly be

minimised if we focus on capabilities of “being achieved” which is an overall

functioning. It identifies these capabilities to be the sense-of-achievement (SA),

sense-of-freedom-to-achieve (SFTA), and sense-of-ability-to-achieve (SATA)

which measure Sen’s functioning, freedom and conversion efficiency,

respectively. The three new terms SA, SFTA, and SATA are introduced in the

thesis to overcome problems in the measurment of functioning, freedom, and

ix | P a g e

conversion efficiency.They are measured using subjective indicators in a unique

questionnaire about mental wellbeing in the Pakistan Socio-Economic Survey

(PSES). Each capability dimension comprises of 3 indicators which are

aggregated using alternative weighting scheme and latent modeling structures.

Using tools of EDA (Exploratory Data Analysis) and non-parametric tests for

the equality of distributions, it is shown that the PSES capability dimensions of

subjective wellbeing (SWB) provide distinctive information, not contained in the

happiness indicator, while together with the happiness indicator they provide

additional insights about SWB. The thesis ranks districts (policy units) in Pakistan

by happiness and different dimensions of capabilities and shows that they are

quite distinct from each other. This classification therefore provides policy makers

with a broader informational space for wellbeing assessment.

Happiness is then regressed on capabilities under different controls to check

for any correspondence between the two. Results strongly support the hypothesis

that capabilities of “being achieved” (functioning, capabilities, and efficiency) are

the most important and stable (in terms of size, sign and significance)

determinants of happiness with respect to size, sign, and significance of

coefficients across various model formulations. This stability is also robust to

estimation methods used and to latent structural relations.

The thesis also explores interaction between different dimensions of

capabilities through bootstrapping and formalises them in the form of theoretical

policy regions. The study of dynamics through bootstrapping is one of the crucial

contributions of the thesis. Application of the theoretical models to the PSES data

reveals that most of the policy units are characterised as low freedom (relative to

efficiency) which requires focus on freedom with increasing emphasis on

efficiency when functionings improve.

While investigating correspondence between objective variables (education

and income) with capability dimensions, the thesis finds that education and

income are key determinants of capability dimensions. The partial effects of

education and income on freedom and efficiency show that education plays a

x | P a g e

more important role than income for the majority, highlighting the importance of

provision of education in enhancing freedom and efficiency.

The thesis also addresses the issue of happiness adaptation to income for a

wide variety of models and finds that evidence in favour of happiness adaptation

to income is quite weak, yet consistent with bulk of the existing evidence. The

thesis, to the best of my knowledge, is the first to investigate the issue for a

developing country.

For what it is worth, this is also the first study that introduces the terms SA,

SFTA, and SATA to operationalize capabilities approach, measures the three

dimensions of capabilities (freedom, functioning and efficiency) simultaneously

and links them with happiness. This provides a strong case for useful extensions

to the frequently used General Household Questionnaire on the one hand, and a

parsimonious manner of addressing issues related to operationalisation of

capabilities and their correspondence with happiness on the other.

xi | P a g e

Statement of Authorship

Except where reference is made in the text of the thesis, this thesis contains

no material published elsewhere or extracted in whole or in part from a thesis

submitted for the award of any other degree or diploma.

No other person’s work has been used without due acknowledgement in the

main text of the thesis.

The thesis has not been submitted for the award of any degree or diploma in

any other tertiary institution.

Some of the work in this thesis is a result of collaboration with my co-

supervisor Dr. Hayat Khan.

Hamid Hasan

24 September 2012

xii | P a g e

In the name of Allah (the God), the most gracious, the most merciful.

Acknowledgements

I would like to express my deep gratitude and thanks to Allah Al-Mighty (the

Only One God) for His unconditional supervision and support at every instant.

I am grateful to my supervisors, Prof Gary Magee and Prof Sisira Jayasuriya

for their trust and confidence on my competence. I am deeply indebted to my co-

supervisor, Dr Hayat Khan, for stimulating suggestions, encouragement, and

collaboration.

I would like to thank Dr David Prentice for feedback on earlier draft of my

thesis.

I also extend my heartfelt gratitude to teachers, colleagues, and students of

International Institute of Islamic Economics (IIIE), Islamabad for their respect and

moral support. It is an honour for me to thank my mentor, Dr Sayyid Tahir, for his

continued encouragement to pursue my PhD, and to my teacher, Dr Asad

Zaman, for his thought provoking lectures on econometrics, Islamic economics,

and critique of modern economic thoughts and policies.

I would like to express my special thanks to my colleague at IIIE and at La

Trobe University, Dr Nauman Ejaz, for his all-round support especially the proof

reading of final draft of my thesis. I am grateful to Abubakar Memon in Lahore

University of Management Sciences (LUMS) for his help and guidance in data

collection and management.

This thesis would not have been possible unless the support, patience, and

sacrifice of my family. To them I dedicate this thesis.

1 | P a g e

Chapter 1

Introduction

“Indeed, We sent aforetime Our Messengers with clear proofs and

sent down with them the Book and the Balance (of right and wrong),

that men may stand forth in justice; […]”

(Qur’an, 57:25)

1.1 Motivation and Background

Traditionally, affluence and poverty have been measured by a money-metric

measure to reflect the level of economic wellbeing. Sen’s seminal work on

capabilites, among others, brought a new dimension to the debate on human

welfare, primarily building up on the idea that a money-metric does not provide a

reasonable measure of human wellbeing.

The evidence of rising income with no corresponding increase in happiness

(the Easterlin Paradox, 1974) has led to extensive research on happiness (the

happiness approach (HA)). Similarly, the evidence indicates that poverty is not

only associated with lack or low level of income, but also with different forms of

deprivations, for example, illness, illiteracy, malnutrition, deprivation of rights and

freedoms (Sen, 1979). The latter inspired Sen’s well-known capabilities

framework.

Both approaches conclude that a one-dimensional resource-based approach

is not sufficient to assess multidimensional human wellbeing; “It is thus

interesting to note that although motivated by contrasting empirical problems, that

is, poverty vs opulence, both approaches have reached similar conclusions about

the complex ‘technology’ of transforming resources into human wellbeing” (Bruni,

et al 2008, p. 5). HA revived utilitarianism but with an empirical application while

the capabilities approach (CA) criticised utilitarianism as the sole criterion for

human wellbeing assessment.

2 | P a g e

The CA, as compared to the HA, not only has provided a wider informational

space for evaluation of human wellbeing but also has revived the dimension of

justice (as opposed to equality) overlooked in the policy design of a utilitarian

social planner.

Justice has been a moot topic in economics and has probably never received

the attention in mainstream economics that it deserves, despite the seminal work

on justice by John Rawls (A Theory of Justice; 1971, and Justice as Fairness- A

Restatement; 2001). Later on, Amartya Sen’s work on justice (which culminated

in 2009 in the book, The Idea of Justice), forcefully revived the concept of justice

in mainstream economic debates.1 Both, Rawls and Sen, are critical of the

utilitarian approach2 to justice. Rawls’ work, however, concentrates mainly on

institutions while Sen’s work focuses on individuals. Sen (1983, p.167) compares

his work with Rawls’:

“The capability approach shares with John Rawls the rejection of

the utilitarian obsession with one type of mental reaction, but differs

from Rawls’ concentration on primary goods by focusing on

capabilities of human beings rather than characteristics of goods

they possess”.

This thesis is motivated by the fact that though there is extreme concern over

justice in all divine religions, as shown by the epigram at the beginning of this

chapter, it is largely neglected in mainstream economics, particularly at the policy

level. 1 Sen’s ideas can be traced back to John Rawls, Karl Marx, Adam Smith, and Aristotle among others. 2 “Utilitarianism can be factorised into welfarism (the demand that evaluation of any social state be based exclusively on the utilities generated in that state), sum-ranking (aggregating individual utilities by simply summing them), and consequentialism (judging the rightness of actions, policies, and other choices exclusively on the basis of the consequent states of affairs)” (Sen, 1991, p. 16). Sen (1979, p. 538) discusses the conditions to be satisfied by utilitarianism: “Utilitarianism – the classical approach to welfare economics – satisfies Pareto-inclusive welfarism, and in the usual applications is combined with the use of interpersonally comparable and cardinal individual utilities. It was the disquiet about interpersonal comparability and cardinality of utilities (expressed in such works as Robbins, 1932, methodological critique) that led to the addition of ordinalism and non-comparable utilities as further features to be satisfied. The so-called ‘new welfare economics’ accepted all these properties as legitimate”.

3 | P a g e

Although, Sen’s capabilities approach (Sen, 1985b) along with his Weak

Equity Axiom (Sen, 1973) provides a framework of justice for an assessment of

human wellbeing, it faces a challenge on the operationalisation side in practice.

The happiness approach to measuring wellbeing, in contrast, provides a

workable framework of subjective wellbeing assessment3 using the tools of

psychology.

Sen (1984, p. 188) criticises the happiness interpretation of utility4 on the

grounds that it is a mental state that does not take into account other aspects of

wellbeing and other mental activities, and that it is subject to mental conditioning.

The benefits of measuring the other mental activities are two-fold: “[these] are of

direct relevance to a person’s wellbeing [and] involve valuation of one’s life” (Sen,

1984, p. 189). This thesis is primarily interested in subjective measures of other

mental activities that are directly related to capabilities vis-à-vis objective

measures of welfare and their correspondence with happiness.

Sen’s and his followers’ critique of the HA sometime lead to the misconception

that Sen is uninterested in happiness. Sen’s characterization in fact is about

replacing the happiness approach to welfare assessment, not happiness itself,

with one that is grounded in capabilities. Anand (2011, p.175) maintains that

“[t]he key is to be careful about causality as happiness does not necessarily

confirm the rational value of something, though with that caveat in mind, Sen

concludes that there is much of interest in happiness, even for post-utilitarian

economics”. Sen (1985b) himself quite clearly treats happiness as a derived

notion from a functioning and does not consider happiness as a general stand-

alone concept as conceived by Kahneman et al (1997) in terms of objective

3 The notable earlier efforts by economists to measure utility include Francis Y. Edgeworth (1881), Irving Fisher (1892), and Ragnar Frisch (1932). These efforts were abandoned in 1930’s with “Lionel Robbins’s (1932) argument against using interpersonal welfare comparisons, which were necessary to use the statistical measurement approach”. (Colander; 2007, p. 222) Afterwards, ordinal axiomatic approach to utility was pursued with no connection to policy. With the emergence of behavioural economics particularly experimental economics, utility measurement has got a second wind. The work of Daniel Kahneman and his colleagues was further responsible for bringing this issue to the forefront of this debate. 4 The other interpretations of utility, Sen (1984) mentions and criticises, are desire fulfilment and choice.

4 | P a g e

happiness (total utility) or by Bentham in terms of subjective happiness

(experienced utility).5 Sen also uses happiness as a functioning like being-happy.

Building up on the theoretical framework given in Sen (1985b), this thesis

postulates that happiness depends on functioning, which in turn, depends on

freedom and conversion efficiency (given the conversion factors and resources),

along with some other alternative specifications.

This thesis is not obviously the first attempt to make such an effort. It is

however the first to measure Sen’s capabilities in all dimensions (identified by the

approach as functioning, freedom and conversion efficiency) and investigate their

correspondence with happiness. This is done in a parsimonious manner which

minimizes problems on the operationalization side of the CA. This marks another

important contribution of the thesis.

For what it is worth, even if Sen’s was uninterested in happiness, it is still

interesting to see if capabilities have any correspondence with happiness. Any

evidence in favour of the link could in this case be treated as evidence against

Sen’s alleged rejection of happiness.

Sen has been criticized on using ambigious terminology for capability (see, for

example, Cohen, 1993; Des Gasper, 2002) and various uses of the term

capability (Qizilbash, 2005). The thesis attempts to remove misconcetions

regarding the nomenclature used in the CA and provides a clear understanding of

these terms. The CA has also been severely criticized by Sugden (1993) and

Roemer (1996), among others, on issues in operationalization of the CA. These

5 Kahneman and Kruger (2006) divide utility into two types: decision utility which is based on revealed preferences and has been used in economics and decision research, and experienced utility which is further divided into moment-based and memory-based. The moment-based utility captures moment to moment responses and these are summed up to obtain total utility. Kahneman and his colleagues advocate this approach and it is based on Experience Sampling Method (ESM) and Day Reconstruction Method (DRM). The memory-based approach depends on retrospective report (life satisfaction) and it is the standard method for measuring subjective happiness. Kahneman and Kruger (2006) criticise this approach as it is affected by current mood, earlier questions in the survey, and duration. But Kahneman’s approach is problametic in its implementation on a large scale; the ESM and DRM can be done on a very small sample of the population since the former is conducted in a laboratory whereas the latter needs continuous reflections of various activities during a day.

5 | P a g e

issues are discussed in section 1.3 in detail. Some of the criticism on the CA is

summarized in Clark (2006) including problems associated with identification of

capabilities list, hence problems of inter-personal comparison of well-being and

determination of weights, high information requirements for the measurement of

functionings, and even high information requirement for the measurement of

capabilities because of their counterfactual nature.

Before introducing detailed objectives of the thesis and highlighting

contribution of the thesis, it is convenient to start with an introduction of Sen’s

capabilities and some of the issues that it faces on the operationalization side.

The rest of the chapter is organized as follows. Section 1.2 breifly introduces key

aspects of the CA, followed by a secton, 1.3, on issues related to

operationilzation of the approach and the solution suggested by the thesis.

Section 1.4 highlights contribution of the thesis in the light of discussion in

sections 1.2 and 1.3 and identify objectives set out for the thesis. The chapter

concludes with a chapter plan to achieve these objectives.

1.2 Sen’s Capabilities Approach

The capabilities approach is a normative framework for assessing individual

wellbeing and social arrangements, and for designing policies for social change

and justice. The approach was pioneered by Amartya Sen (1984, 1985a, 1985b,

1987a, 1987b, 1990, 1992, 1993, 1999) and further developed by Martha

Naussbaum (2000, 2005).6 The five distinct factors in the CA identified in Sen’s

writings are: achievement (functioning), freedom to achieve, ability to achieve

(conversion efficiency), conversion factors, and resources. Conversion factors

(such as age and education) and resources (such as income) are well-known in

the traditional literature (see, for example, Kuklys, 2005). Sen’s contribution

therefore revolves around three key concepts; functioning, freedom, and

conversion efficiency.

6 See, for example, Robeyns (2005, 2011) for the theoretical survey and philosophical discussion on the capabilities approach.

6 | P a g e

Before introducing these terms, it might be convenient to point out that, along

with conversion factors and resources, this thesis focuses on capabilities of

“being achieved” which is an overall functioning. It argues that doing so greatly

minimises problems associated with operationalising the CA approach. It

identifies these capabilities to be the sense-of-achievement (SA), sense-of-

freedom-to-achieve (SFTA), and sense-of-ability-to-achieve (SATA) which

measure Sen’s functioning, freedom and conversion efficiency, respectively.

The following sections explain these concepts briefly.

1.2.1 Functionings

A person can be considered to be in two types of states; state of beings and

state of doings. Collectively these states are called functionings, i.e., functionings

are “beings and doings” of a person. For example, beings include; being-

educated, being-healthy, being-nourished, being-sheltered, being-happy,

whereas doings include; studying, travelling, caring for a child, voting in an

election, taking part in debate, donating money to charity, and so on.7

These states of functionings have intrinsic or intrinsic and instrumental values.

In contrast, resources (means) have only instrumental value. In simple words,

functionings are achievements of a person. The state of being can be considered,

for the sake of understanding, as a ‘stock’, whereas the state of doing can be

termed as a ‘flow’. For example, the flow of exercise (the doing of exercise) adds

to the stock of health (being-healthy). Similarly, reading adds to being-literate.

However, this distinction between stock and flow may not be straight forward in

practice.

Functionings result from either choice or constraint. Whereas the former is

referred to as a refined functioning, the latter is simply called a functioning.

Human diversity and multiplicity of functionings lead to four methodological

problems (Kuklys, 2005, p. 21): “the selection of the relevant functionings, the

measurement of these functionings at the individual level, the aggregation of

7 Some of the examples are drawn from Robeyns (2011).

7 | P a g e

these functionings into a composite measure of individual welfare, and finally, the

aggregation of individual welfare to social welfare”. Section 1.3 discusses these

issues in detail along with implications of the classification of functionings being

exogenous and endogenous. It also sugests a solution to minimize these

problems, along with others.

1.2.2 Conversion Efficiency

Conversion efficiency is the ability of a person to convert his/her resources into

functionings given his/her freedom. The conversion efficiency is influenced by

three types of conversion factors; individual/personal, social, and environmental

(Kuklys, 2005; Robeyns, 2005). These conversion factors are illlustrated in the

following example by Robeyns (2011, p. 6):“How much [conversion efficiency] a

bicycle [a resource] contributes to a person’s mobility [a functioning] depends on

that person’s physical condition (a personal conversion factor), the social mores

including whether women are socially allowed to ride a bicycle (a social

conversion factor), and the availability of decent roads or bike paths (an

environmental conversion factor)”.

1.2.3 Freedom

Positive freedom (henceforth freedom) in terms of the range of choices and

autonomy is the basic requirement for justice and measurement of standard of

living, and it is the principal component of Sen’s capabilities approach.8 It has

instrumental as well as intrinsic value, and evaluation on the basis of freedom

has the potential of providing an encompassing measure of wellbeing; “According

to Sen’s capabilities approach, economic and social arrangements should be

evaluated in terms of the freedoms enjoyed by those who live in them” (Alkire,

2005). Sen (1990) discusses freedom as a focal personal feature for ethical

judgment on the lives of persons and compares it to primary goods and liberties

(Rawls), rights (Nozick), resources (Dworkin), among others. In this context, he

distinguishes between means and what people can obtain from these means. He

argues (p.115):

8 For a discussion on positive freedom and negative freedom see Sen (1987b), among others.

8 | P a g e

“Since the conversion of these primary goods and resources into

freedom to select a particular life and to achieve may vary from

person to person, equality in holdings of primary goods or

resources can go hand in hand with serious inequalities in actual

freedoms enjoyed by different persons”.

The notion of individual freedom has two aspects in Sen’s capabilities

approach: the opportunity aspect and the process aspect (Sen, 2002). The

opportunity aspect is concerned with the advantage one has as compared to

others (Sen, 1985a) and with one’s ability to achieve what one values irrespective

of the process through which that achievement comes about, while the process

aspect is concerned with the process of choice itself (Sen, 2009). The first aspect

is termed ‘Capability’ while the second is called ‘Agency’ in Sen’s writings. Sen

(1985a, p.5) states:

‘‘Wellbeing’ is concerned with a person’s achievement: how ‘well’ is

his or her ‘being’? ‘Advantage’ refers to the real opportunities that

the person has, especially compared with others […]. The freedom

to achieve wellbeing is closer to the notion of advantage than

wellbeing itself”.

The concept of capabilities emphasises the opportunity to achieve the best

with the availability of multiple opportunities. Sen (2002, p.509) writes:

“To conclude this section on concepts of freedom, we have to be

concerned with at least two distinct aspects of freedom, viz. (i) the

opportunity aspect, and (ii) the process aspect. The opportunity

aspect must pay particular attention to the opportunity of achieving

the best that can be achieved, but may extend that concern by

taking some supplementary note of the range of opportunities

offered. The process aspect, being concerned with the freedom of

the person’s decisions, must take note of both (iia) the scope for

autonomy in individual choices, and (iib) immunity from interference

by others”.

9 | P a g e

Sen argues that it is the responsibility of the society to provide freedom to

achieve functionings. Sen (1992, p.148) writes:

“In dealing with responsible adults, it is more appropriate to see the

claims of individuals on the society (or the demand of equity or

justice) in terms of freedom to achieve rather than actual

achievements. If the social arrangements are such that a

responsible adult is given no less freedom (in terms of set

comparisons) than others, it is possible to argue that no unjust

inequality may be involved”.

This does not mean that individuals do not have a responsibility to change

their status for a better life. Sen (1999, p. 53) argues:

“The people have to be seen, in this perspective, as being actively

involved – given the opportunity – in shaping their own destiny, and

not just as passive recipients of the fruits of cunning development

programs”.

The possession of commodities does not correctly represent the opportunity-

freedom. Sen (2002, p.519) argues:

“[…] opportunity-freedom cannot be sensibly judged merely in

terms of possession of commodities, but must take note of the

opportunity of doing things and achieving results one has reason to

value”.

Sen (1985, p.203) defines human agency as;

“[it is] people’s ability to act on behalf of goals that matter to them”.

Capability is, therefore, a freedom oriented concept. Qizilbash (2011, p. 27)

explains capability as: “[…] Sen’s use of the term ‘capability’ refers to a range of

lives from which a person can choose one, and that if one has to list things which

make a life good these are best understood as (valuable) functionings. The

capability approach – as I understand it – sees wellbeing in terms of an

10 | P a g e

evaluation of functionings – and the quality of life is seen in terms of the freedom

to choose between lives”. (Italics in original)

Measurement of freedom has been the most difficult task in operationalising

the capabililties approach. That is why most of the empirical work has

concentrated on measuring functionings. Section 1.3 reviews these issues in

detail.

The above account explains Sen’s capabilities approach (CA) and its

essentials. The main thrust of the approach is on the freedom to achieve

functionings. The thesis measures freedom to achieve subjectively, discussed in

the methodology section of chapter 2, and refers to it as sense-of-freedom-to-

achieve (SFTA).

The following section identifies some problems in operationalising capabilities,

discusses solution(s) proposed/applied in the literature, and suggests alternative

solutions where previously suggested or used solution(s) is (are) not appropriate.

1.3 Issues in Operationalising Sen’s Capabilities Approach and Approach of the Thesis

The term ‘operationalisation’ is a broader term then ‘quantification’. Comim

(2001, p.1) defines four stages or alternatives for operationalising the capabilities

approach:

i) theoretical inclusion: elaboration of theoretical concepts with potential

empirical significance;

ii) measurement: transformation of these theoretical concepts into empirical

variables;

iii) application: use of these variables in qualitative empirical analysis; and

iv) quantification: use of these variables in quantitative empirical analysis.

The measurement of capabilities (stage ii) also includes clarification of

concepts and terms (Comim, et al 2008). There are many technical issues in

11 | P a g e

operationalising capabilities (see, for example, Robyens, 2006; Kuklys, 2005). I

will focus here on those that are important from point of view of the thesis.

Human diversity and multiplicity of functionings lead to four methodological

problems (Kuklys, 2005, p. 21): “the selection of the relevant functionings, the

measurement of these functionings at the individual level, the aggregation of

these functionings into a composite measure of individual welfare, and finally, the

aggregation of individual welfare to social welfare”. The following discusses these

issues along with measurement of ability and freedom to achieve a functioning,

This is followed by circularity in different dimensions of capabilities and a

discussion on the problem of unobserved human diversity.

1.3.1 Measurement of Functionings

Kuklys (2005, p.33) reviews the following problems in the measurement of

functionings:

i) absence of a unit of measurement;

ii) missing natural aggregator;

iii) measurement error; and

iv) relative judgement and anchoring problem (in case of subjectively reported

indicators).

In traditional poverty or inequality analysis, monetary units are used to

measure and aggregate welfare, but in case of functionings there is no such unit

of measurement and natural aggregator since functionings are usually measured

on ordinal scales. The problem of measurement error is likely to arise since

functionings are not directly measurable and often they are represented by more

than one variables. In case of subjective indicators, recall error may also arise

with problems of relative comparison (comparison with a reference group) and

anchoring (different understanding of ordinal scale).

12 | P a g e

1.3.2 Selection of Functionings

Functionings are selected in one of the following methods in the literature

(Kuklys, 2005, p. 21):

i) ad hoc method (selection according to the normative views of a

researcher);

ii) mechanical method (for example, using factor analysis);

iii) selection criteria method (proposing a list of functioning satisfying a

criteria); and

iv) participatory method (deciding functionings with the consultation of

concerned individuals).

Some of these problems may have arisen due to the implicit assumption that

a capabilities set is exogenously given to an individual. Because of this

assumption, many researchers (For example, Rawls, Qizilbash, Nussbaum) have

proposed lists of functionings including human rights (a top-down approach: A

macro level list like human rights is implemented at micro level). Though the

assumption is made in order to simplify multiplicity of functionings, it makes the

issue more complicated as no such list may obtain universal acceptance. Hence

there is too often a chance of incorrectly estimating capability deprivations. Since,

Sen puts a great emphasis on human diversity, he does not approve a list of

functionings (Sen 1993; 2004a):

“To have [...] a fixed list, emanating entirely from pure theory, is to

deny the possibility of fruitful public participation on what should be

included and why”. (Sen, 2004a, p.77)

Consequently he implicitly assumes that a capabilities set is endogenously

given. Therefore a universal list of functionings is not applicable to all individuals.

Burchardt (2009, p.16) argues “Utilitarianism treats tastes and preferences as

exogenous, but the capabilities approach should not follow suit, since the ‘menu’

of options available to an individual not only influences his actual choice but can

also shape the formation of his goals and preferences”.

13 | P a g e

The second problem with a list of functionings is that it cannot distinguish

between voluntary and involuntary functionings (e.g., starving versus fasting)

because they are observationally equivalent. In other words, a list lacks refined

functionings (i.e., functionings with their alternatives).Instead of proposing a list of

functionings, some studies select functionings on the basis of a theory while

others rely on mechanical or atheoretical selection of functionings.

However, there are a few studies (for example, Alkire, 2002; Robeyns, 2003,

2005) that implicitly assume endogenous capabilities sets. These studies show

how to select capabilities and outline procedures to carry out that selection. This

approach, called bottom-up, takes into account the endogeneity of capabilities

sets as well as human diversity. These methods are useful when conducting a

new household survey and asking individuals about their capabilities sets as

most of the available household surveys do not contain such information. A

bottom-up list of functionings might be in conflict with a top-down list of

functionings (e.g., human rights standards). Burchardt and Vizard (2011)

attempts to resolve this conflict by suggesting a two-stage selection procedure.

The study suggests a top-down approach at stage-1 to derive a minimum core

capabilities list from international human rights framework, and a bottom-up

approach at stage-2 to derive a minimum core capabilities list from deliberations

and consultations. If there is still a conflict between these two lists, they suggest

giving priority to stage-1 list.

The approaches mentioned above (top-down, bottom-up, and two-stage) are

useful only when conducting a new household survey that incorporates capability

dimensions in the questionnaire. However, the last two approaches may be

applicable to only a small group of individuals. Another approach, say quasi-

endogenous, analyses satisfaction with existing rules and regulations. For

example satisfaction with political freedom, religious freedom, and freedom of

expression (Anand et al 2011). The present study shows how an existing large-

scale survey may be used to measure capabilities sets using existing subjective

questions related to capability dimensions.

This thesis avoids this problem altogather by choosing capabilities of an

overall functioning, “being achieved” which makes the list of individual

14 | P a g e

functionings subjective in nature. This approach has the advantage of taking into

account only functionings that are valued by a particular individual which is at the

heart of the CA. One advantage of assuming the capabilities set as an

exogenously given list however is that it has a lower probability of suffering from

the adaptation problem, as opposed to an endogenously determined capabilities

set, since an endogenously determined list may be affected by individual

preferences.

1.3.3 Aggregation of and Trade-off between Functionings

There may be three levels of aggregations of functionings and at each level a

weighting scheme is used. At first, indicators need to be aggregated to create a

functioning (since a functioning is not directly observable). Since an individual

can be identified with a number of functionings at a time, the second stage

involves aggregation of functionings at the individual level. This is followed by

aggregation of functionings over a group of individuals (where policy units contain

more than one individual). The literature uses a variety of weighting schemes,

including scaling method (arbitrary selection), those derived through statistical or

econometric procedures (factor analysis, principal component analysis, or

structural equation modeling), and social choice procedure (participatory

approach). Whereas the aggregation at the third stage is straight forward and

poses less of a problem, aggregation at the first two stages need attention. The

thesis avoids the second stage aggregation problem through selecting

capabilities of an overall function of “being achieved” which is equivalent to letting

the individual decide weight. For aggregation at stage 1, the thesis uses scaling

method as a base case and checks its robustness against the structural equation

modeling (SEM) technique of Partial Least Square Path Modeling (PLS-PM).

Chapter 4 discusses the rationale behind using PLS-PM as an alternative and

why using statistical techniques such as Factor analysis and Principal

Component Analysis are not appropriate.

1.3.4 Measurement of Ability to Achieve a Functioning

The literature has paid least attention to the issue of how to measure the

ability of an individual so as to convert resources into functionings, despite the

fact that it is the most essential component of the capabilities approach. Most

15 | P a g e

researchers have considered conversion factors and neglected the ‘ability to

convert’ (conversion efficiency), with an implicit assumption that conversion

factors are sufficient to capture conversion efficiency. Although conversion

factors have level effects on conversion efficiency, yet they do not distinguish

different conversion efficiencies across same levels of conversion factors. For

example, the conversion factor, age, has different effects on conversion efficiency

depending on which age group an individual belongs to, but there is no

mechanism that differentiates various degrees of conversion efficiency within an

age group.

Some researchers have used techniques of technical efficiency analysis,

which are usually applied to firms, like data envelope analysis (DEA) or efficient

frontier (Binder and Broekel, 2011; Ramos and Silber, 2005) to address this

problem. In order to apply DEA or efficient frontier techniques which are suitable

for the analysis of firms, these studies assume that individuals behave like firms.

This assumption has been criticised by Sen (1985b, p.15) as it is inappropriate to

consider an individual as a firm or a factory9. Another drawback of this approach

is that it fails to distinguish between voluntary and involuntary choices.

The analysis in this thesis applies a subjective approach to deal with this

problem and uses sense-of-ability-to-achieve (SATA), as will be described later in

the methodology section of Chapter 2, of an individual as a proxy for his/her

conversion efficiency given the conversion factors.

1.3.5 Measurement of Freedom to Achieve a Functioning

The freedom to achieve is concerned with the availability of choices and the

process of choosing, whereas ability to achieve is concerned with the ability to

convert resources into a functioning. Former is related to multiple functionings in

a capabilities set, but the latter is related to a single functioning from a

9 “It may indeed be illuminating up to a point to see functionings as ‘commodities’ produced by the household, but this analogy can also be misleading since functionings are features of the state of existence of a person, and not detached objects, that the person or the household happens to ‘produce’ and ‘own’.”

16 | P a g e

capabilities set. Sen (1987a, p. 15) explains ability to achieve by the following

example:

“However, two persons possessing identical bicycles [resources],

may have very different ability to move about [a functioning], if one

happens to be disabled and the other not”.

It is difficult to measure freedom to achieve a functioning since the capabilities set

is unobservable. Robeyns (2000, p.21) comments:“If we have information on the

achieved functionings and the capability set, we can then deduce whether the

non-achievement of a certain functioning is the result of a free choice to forego

this particular functioning or simply because it was not available in the capability

set. The problem, obviously, is that it will be extremely difficult to gather

information on the capability set so that the question how to assess whether a

choice was ‘free’ or not remains unsolved”.

Anand et al (2011) attempt to solve this problem by constructing self-reported

freedom in various domains, like freedom of political expression, freedom of

political participation, freedom of religion, and freedom of thought. It seems that

this study considers negative freedoms while Sen emphasises positive freedom

in the capabilities approach and contrasts it with negative freedom (Sen, 1987b,

p.6):

“This contrast, which has been discussed particularly by Isaiah

Berlin, is quite important since the two ways of characterising

freedom may yield very different assessments”.

The negative freedom is related to the freedom from interference, whereas

positive freedom is concerned with freedom to do something. The former is

external to an individual, whereas the latter is internal to an individual. For

example, political freedom and religious freedom are related to negative freedom.

Associating negative freedom with capabilities means that we are assuming an

exogenously given capabilities set, and hence may be contradicting the basic

premise of the capabilities approach.

17 | P a g e

The thesis attempts to solve this problem by taking sense of freedom to

achieve (SFTA), as will be described later in section 2.1, as a measure of positive

freedom. This freedom is also associated with the overall functioning ‘being-

achieved’. Another important advantage of this solution is that it somehow

captures the agency aspect of freedom as well since agency and wellbeing are

mutually related. Sen (1985a, p.187) remarks:

“For an integrated person it is likely- possibly even inevitable- that

the person’s wellbeing will be influenced by his or her agency role.

This does not, however, imply that the wellbeing information itself

could capture the important features of agency, or act as its

informational surrogate. In fact, some types of agency roles, e.g.,

those related to fulfilling obligations, can quite possibly have a

negative impact on the person’s wellbeing. Even when the impact is

positive, the importance of the agency aspect has to be

distinguished from the importance of the impact of agency on

wellbeing”. (Italics in original)

I would argue that the freedom measured from an exogenously given

capabilities set or from a quasi-endogenous set may not capture the agency

aspect of freedom since it is entirely an internal attribute of an individual.

1.3.6 Circularity in Capability Dimensions

Sen (1985a, p. 202) highlights mutual dependence and simultaneity between

functionings and capabilities:

“This might look like introducing a circularity in the relationship

between functionings and capabilities, and between wellbeing and

wellbeing freedom. But what it, in fact, does is to force us to see

these concepts as mutually dependent, taking note of the

simultaneity of the relationships involved”.

Binder and Coad (2011, p.328) discuss the circularity problem in functionings:

“The more functionings one looks at, the more interdependencies between them

and the resources side can be expected (this also pertains to conversion factors

18 | P a g e

that might be considered either resources or functionings in different contexts)”.

They suggest using panel vector autoregression (VAR) to solve this problem

econometrically as panel VAR takes into account interdependencies between

resources and functionings (by simultaneous structure), and also incorporates

feedback effect (by time lags). According to Binder and Coad (2011, p. 328), the

problem becomes highly complex once we take into account these factors

(interdependencies and time lags): “While existing research mainly focuses on a

limited set of functionings, it neglects the complex interaction between these and

other variables, especially their intertemporal development. We may need to

consider several different time lags to appreciate the richer structure of the

dynamics of individual functioning achievement and possible feedback effects”.

The panel VAR approach in the context of capabilities, suggests that as number

of functionings increases, the interdependencies would also increase, and more

equations and time lags would be needed to capture the dynamics. The situation

would be even more complicated when we include not only functionings but also

ability to achieve and freedom to achieve dimensions of capability.

The solution of overall functioning of “being-achieved” suggested in the thesis

avoids this complication to a great extent.

1.3.7 Problem of Incorporating Unobserved Human Diversity

Standard economic analysis assumes homogeneity of individuals whereas the

capabilities approach emphasises heterogeneity of individuals as each individual

has a unique (endogenous) capabilities set. Sen (1992, p.ix) argues:

“Human diversity is no secondary complication (to be ignored, or to

be introduced ‘later on’); it is a fundamental aspect of our interest in

equality”.

Sen (1992, p.3) further argues that:

“The pervasive diversity of human beings intensifies the need to

address the diversity of focus in the assessment of equality”.

19 | P a g e

Anand et al (2011) attempt to model unobserved heterogeneity using

generalised linear latent and mixed model (GLLAMM). This thesis does not apply

GLLAMM due to data limitations and the sampling design used.

To account for human diversity, aggregation is done over groups of

individuals, as opposed to over individuals, to allow for a certain degree of human

diversity (that solves the problem of order of aggregation as well), and then these

groups (districts) of individuals are ranked for policy recommendations. Since the

internal attributes like agency, freedom and, functioning, as argued earlier, are

position-dependent, the aggregation strategy used here is valid.

1.4 Contributions, Objectives and Structure of the Thesis

In an attempt to integrates the HA with the CA, the thesis first focuses on

addressing methodological problems associated with operationalization of the

CA. It argues that these problems can be greatly minimized if we consider

capabilities of “being achieved,” which is an overall functioning. It introduces SA

(sense of Acheivement), SATA (Sense of ability to achieve), and SFTA (sense of

freedom to achieve) as subjective measures of Sen’s three dimensions of

capability – freedom, efficiency, and functioning – respectively. To the best of our

knowledge, this thesis is the first to introduce capabilities of an ovrall function of

“being achieved”. More importantly, it is also the first attempt which measures

Sen’s capabilities in all dimensions and explore their correspondence with

happiness10. This also allows us to study interaction between different

dimensions of capabilities and draw some policy conclusion.

Sen has been criticized on using ambigious terminology for capability (see, for

example, Cohen, 1993) and various uses of the term capability (Qizilbash, 2005).

The thesis attempts to remove misconcetions regarding the nomenclature used in

the CA and provides a clear understanding of these terms. Clark (2002) also

attempts to clarify the concept of capability.

10 These key factors have never been analysed simultaneously particularly in empirical capability literature due to operationalization problems due to multiplicity of functionings (Sugden, 1993, criticizes the CA on operationalisation issue).

20 | P a g e

In the context of our discussion above the thesis attempts to achieve the

following objectives:

i) The main objective of the thesis is to bridge the gap between Sen’s

capabilities approach and the happiness approach in assessing human

wellbeing.

ii) The next objective is to address problems on operationalization side of the

CA and develop subjective measures of Sen’s capabilities in all

dimensions.

iii) The third objective is to explore whether the subjective measures of each

dimension of capability contains information distinct from the other and the

happiness indicator using tool of EDA (exploratory data analysis).

iv) The fourth objective is to estimate a link between capabilities and

happiness and check its robustness using a variety of various weighting

schemes and models formulations.

v) The fifth objective is to understand the interaction between different

dimensions of capabilities and formalize it in the form of a stylzed

theortical model.

vi) The sixth objective is to use the theoretical model for policy analysis and

apply that to our selected sample.

vii) The seventh objective is to addresses issues related to operaztionlisation

of policy implications as our capability dimensions are latent in nature.

viii) The last objective of the thesis is to address the issue of adaptation of

happiness to income, for the first time, for developing countries in general

and for Pakistan in particular by creating a two-period panel for a random

effects ordered probit model.

The rest of the thesis is divided into seven chapters. Chapter 2 attempts to

achieve the second and third objective.. It argues that selecting “being-achieved”

as an overall functioning would greatly minimise, if not solve, the difficulties

21 | P a g e

posed by multiplicity of functioning. It identifies sense-of-achievement (SA),

sense-of-freedom-to-achieve (SFTA), and sense-of-ability-to-achieve (SATA)

which measure Sen’s functioning, freedom and conversion efficiency

respectively, as capabilities of “being-achieved”. They are measured using

subjective indicators in a unique questionnaire about mental wellbeing in the

Pakistan Socio-Economic Survey (PSES). Each capability dimension comprises

of three indicators which are aggregated using a weighting scheme. This chapter

uses scaling method (equal weights) as a base case.

The chapter also compares distributions of capability dimensions with each

others and with the distribution of subjective happiness using tools of EDA

(exploratory data analysis) and non-parametric tests for the equality of

distributions. It is shown that the PSES capability dimensions of subjective

wellbeing (SWB) provide distinctive information while together with the happiness

indicator they provide additional insights about SWB. Rankings of districts (our

policy unit) on the basis of capability dimensions and happiness provide

distinctive information which can be used for policy implementation.

Chapters 3 and 4 address the fourth objective. Whereas chapter 3 models

happiness in a capabilities framework using various formulations and estimation

methods (single- and simultaneous equation models) and demonstrates that

capabilities are the most important and stable determinants (in terms of size,

sign, and significance) of happiness, chapter 4 checks robustness of the analysis.

It discusses the main alternatives to the scaling method used and checks its

robustness to changes in weights and latent structural relations. Similarly

rankings of districts are checked for robustness to weighting schemes and latent

structures. It turns out that equal weights used in the scaling method (base case)

and district rankings are robust to these changes, and hence conclusions derived

in chapter 3 are valid.

Chapter 5 focuses on objectives (v) and (vi). It explores interaction between

different dimensions of capabilities and its impact on happiness using

bootstraping. This is formalized in the form of a stylized theoritical model which is

solved to identify different theoritical policy regions in terms of freedom relative to

efficiency or efficiency relative to freedom. The model when applied to districts,

22 | P a g e

our policy unit, in Pakistan concludes that most of the districts fit the low-freedom

(relative to efficiency) scenario in a policy region, which requires focus on

freedom with increasing emphasis on efficiency as a functioning, improves.

Chapter 6 addresses the seventh objective and operationalises the policy

targeting of the last chapter by analysing the determinants of freedom and

efficiency using competing ordered choice (logit) models. It is shown that income

(education) has a larger effect at higher (medium) levels of freedom and

efficiency. Both have the same effect at lower levels of freedom and efficiency.

Hence, at lower levels, it does not matter which policy targeting is employed.

Since most respondents are at medium levels of both freedom and efficiency, the

role of education is particularly valuable.

Chapter 7 examines the issue of happiness adaptation to income, our last

objective, through formal econometric analysis. It uses eight models which are

used in the literature to test adaptation and inertia (a related concept to

adaptation). The results provide weak evidence in favour of adaptation. This is

consistent with the majority of findings in the literature, but inconsistent with some

of the findings from studies using long panels.

Chapter 8 concludes the thesis by placing the findings in a wider context and

drawing their implications, and presenting a discussion of limitations and future

directions for research.

23 | P a g e

Chapter 2

Measuring Happiness and Capabilities

“O you who believe! Stand out firmly for justice, as witnesses to

Allah, even as against yourselves, or your parents, or your kin, and

whether it be (against) rich or poor; For Allah can best protect both.

Follow not the lusts (of your hearts), lest you swerve, and if you

distort (justice) or decline to do justice, verily Allah is well

acquainted with all that you do”.

(Qur’an, 4:135)

“Valuing a life and measuring the happiness generated in that life

are two different exercises”.

Amartya Sen (1985b)

2.1 Introduction

After addressing issues related to the selection, measurement and

aggregation of functioning, this chapter addresses the question of whether or not

different dimensions of capability contain information distinct from each other and

from the subjective happiness indicator. The chapter begins with a brief

background to the relevance and importance of Sen’s capabilities. It argues that

multiplicity of functioning poses serious measurement problems, which can

greatly be minimised if we select the overall functioning of “being achieved”. It

identifies sense-of-achievement (SA), sense-of-freedom-to-achieve (SFTA), and

sense-of-ability-to-achieve (SATA) which measures Sen’s functioning, freedom

and conversion efficiency respectively, as capabilities of “being-achieved”. They

are measured using subjective indicators in a unique questionnaire about mental

wellbeing in the Pakistan Socio-Economic Survey (PSES). Each capability

dimension comprises of 3 indicators which are aggregated using a weighting

scheme. This chapter uses scaling method (equal weights) as a base case.

24 | P a g e

Using tools of EDA (Exploratory Data Analysis) and non-parametric tests for

the equality of distributions, it is shown that the PSES capability dimensions of

subjective wellbeing (SWB) provide distinctive information, not contained in the

happiness indicator, while together with the happiness indicator they provide

additional insights about SWB. The thesis ranks districts in Pakistan by

happiness and different dimensions of capabilities and shows that they are quite

distinct from each other. This classification therefore provides policy makers with

a broader informational space for wellbeing assessment.

2.2 Background

Induced by growing dissatisfaction with resource-based measures of

wellbeing, particularly Gross Domestic Product (GDP), the inclination of

economics towards moral philosophy and development ethics is relatively recent

(see, for example, Sen, 1988, 1999, 2009). The dissatisfaction is not new and

concerns have been raised in previous research, for example, in the early

writings of Denis Goulet (1931-2006).11 What is new is an increasing interest in

complementing resource-based measures of wellbeing with alternative indicators

(for example, social capital index, human development index, gross national

happiness index). There is nothing wrong with GDP per se as long as it is

restricted to the purpose for which it was developed. Problems arose when policy

makers and governments started using it as a measure of human wellbeing

primarily because of its simplicity. To overcome the issues associated with GDP

as a measure of human wellbeing, Mahbub ul Haq constructed the Human

Development Index (HDI) as an indicator of human wellbeing.12

Although HDI is a crude measure it is as simple and transparent as GDP. The

purpose was two-fold: to provide a single numerical value to policy makers like

GDP and to initiate a debate on human development issues (see, for example,

Blanchfower and Oswald, 2005). Since the publication of the first human

development report in 1990, there has been a major shift in the thinking in the

11 He is considered to be the father of development ethics. See Goulet (2006). 12 The idea of HDI was proposed and implemented by Mahbub ul Haq, the founder of UNDP Human Development Report, in 1990 based on Amartya Sen’s capabilities approach.

25 | P a g e

development paradigm; from commodity-centred development to people-centred

development. This was recognised in the Stiglitz Commission Report (2009)

which emphasised the shift from measuring economic production to measuring

wellbeing of people.13 This approach is summed up in the following quote:

"Human development, as an approach, is concerned with what I

take to be the basic development idea: namely, advancing the

richness of human life, rather than the richness of the economy in

which human beings live, which is only a part of it." (Amartya Sen)

There are two major approaches to measuring the wellbeing of people in the

literature – objective (impersonal/ external evaluation) approach using cardinal

measures and subjective (personal/ self evaluation) approach using ordinal

measures (see, for example, Krueger, 2009).14 Human Development Indices

capture most of the objective measures of wellbeing while the happiness

indicator has been an important measure of subjective wellbeing.15 Sen criticises

commodity-based approaches to welfare assessment on the following grounds

(Sen, 1985b); first, these approaches do not take human diversity into

consideration but instead assume homogeneity which is a gross simplification;

second, they are not focusing on an individual’s abilities or disabilities but on

what an individual possess or reveals to prefer; and last, these approaches are

subject to adaptability, i.e., individuals adjust to their circumstances and do not

show their true wellbeing in terms of possessions and preferences.

Sen’s theoretical work in the field of welfare economics is all-encompassing

as it gives importance to objective as well as subjective measures and adds new 13 A commission on the measurement of economic performance and social progress was set up in France in 2008 comprising twenty two renowned economists and social scientists, headed by Professor Joseph Stiglitz, Amartya Sen, and Jean Paul Fitoussi, including Bina Argarwal, Francois Bourguignon, and Nicholas Stern. Their report is available on www.stiglitz-sen-fitoussi.fr. A similar position was taken by Sen a couple of decades ago in 1984 (see, Sen, 1984). 14 See Kristoffersen (2010) for issues concerning cardinal and ordinal measures. 15 HDI covers three dimensions of wellbeing – health, education, and living standards (see HDR, 2010, p. 13). Recently, Alkire and Santos (2010) proposed a multidimensional poverty index (MPI), which modifies the HDI to meet the requirements of Millennium Development Goals (MDGs). Kahneman, in the 1960’s and, later, Easterlin (1974) developed an alternative approach in the form of subjective wellbeing or happiness.

http://www.stiglitz-sen-fitoussi.fr/

26 | P a g e

dimensions – the capabilities dimension – of human wellbeing. Although

capabilities and happiness are closely related to each other, they however are

distinct from each other as indicated by Dasgupta (1993, p. 3):“Two aspects of

personhood have alternated in dominating the thinking of social philosophers

over the centuries, each true in itself, but each quite incomplete without the other

[…] If one vision sees us doing things, the other sees us residing in states of

being. Where the former leads one to the language of freedom and rights, the

latter directs one to a concern with welfare and happiness.”

The capabilities approach encompasses both doing (e.g., freedom) and being

(e.g., happiness) and hence captures additional insights about SWB. There is

however no summary statistic or index that ranks policy units on the basis of

capabilities or incorporates such information in the existing measures of

subjective wellbeing. This sort of ranking is important since the distribution of

happiness does not necessarily imply the distribution of capabilities. Functioning

and capabilities have intrinsic as well as policy/instrumental value as they provide

information on mental health and have implication for happiness (Sen, 1985b).16

The empirical literature to date has been focusing more on individual

dimensions of capabilities, functioning or freedom in particular. With the

exception of a few papers, such as Anand et al (2011), most of these studies use

objective indicators to quantify capabilities. The BHPS (British Household Panel

Survey) and the German Socio-Economic Panel Survey (GSOEP) use a 12-

questions General Health Questionnaire (GHQ) which has information on the

freedom aspect of “being achieved”.17 Alkire (2005, p.10) makes similar

observations:“With respect to the measurement of freedom as indicated above, I

observe that the literature to date has focused upon the measurement of

functionings, and left process freedoms – and indeed opportunity freedoms –

largely unaddressed thus far”.

16 Some recent literature estimates the impact of capabilities on happiness (life satisfaction), such as Anand et al (2011) and Burchardt (2005). 17 Anand et al (2011) developed their own survey instrument to measure the freedom aspect of capabilities.

27 | P a g e

The information, contained in the GHQ on the freedom aspect of “being

achieved” has not been capitalised as yet perhaps because it lacks other

complementary information. PSES is the first survey that collects information on

all aspect of capabilities in a parsimonious and generalised manner and the

current study therefore has the advantage of being the first, to our knowledge, to

analyse happiness vis-à-vis all aspects of capabilities.

We focus on capabilities of a single functioning, “being achieved”, for reasons

discussed in the methodology section, and measure Sen’s capabilities in three

dimensions, namely functioning, freedom, and conversion efficiency as in Sen

(1985b) through subjective indicators in a unique questionnaire about mental

wellbeing in the Pakistan Socio-Economic Survey (PSES).18 These indicators

are: sense-of-achievement (SA), sense-of-freedom-to-achieve (SFTA), and

sense-of-ability-to-achieve (SATA) which measure Sen’s functioning, freedom

and conversion efficiency, respectively. These indicators are based on

individuals’ perception of “being achieved”.

Using statistical techniques, we show that the capability indicators contain

information distinct from each other and from our happiness indicator19. We use

the above-mentioned indicators to rank districts in Pakistan and construct a

composite index of these three indicators, called Subjective Capability Index

(SCI). Our capability rankings, individual and that of SCI, turn out to be quite

different from the happiness rankings. This provides further support to the idea of

having capability based rankings. We also show how these differences could be

used to identify the policy-focus appropriate for each district (chapter 4).

The rest of the chapter is organised as follows. Section 3 reviews the

empirical literature on operationalisation of capabilities briefly. Section 4 provides

details on methodology of the thesis. This section provides details on the

selection of functioning, the nature of our capability measurement, and the 18 Kuklys (2005, p.34) notes: “There is no requirement that indicators have to be objective when evaluating welfare according to the capabilities approach.” 19 Happiness in this paper is considered to be one of the measures of SWB. Some, particularly those in economics, treat happiness and SWB as synonymous, while literature in psychology treats happiness as a narrower concept than SWB (see, for example, Bruni and Porta (eds.), 2007)

28 | P a g e

statistical method employed in assessing its relevance and importance. Section 5

gives details of the data and questions used to measure different dimensions of

capabilities and happiness. Section 6 report results of the exercise and, finally,

Section 7 concludes the chapter and identifies some relevant policy implications.

2.3 Empirical Capability Literature: A Brief Review20

Operationalisation of Sen’s capabilities approach is one of the most difficult, if

not impossible, aspects of his framework (Comim, 2001). Studies on

measurement abound but I review some of the representative studies in this area

(see Kuklys, 2005, for a comprehensive survey of empirical capability literature). I

use and follow Sen (1985b) seminal multi-dimensional framework of the

capabilities approach as an organising and unifying principle for the literature

review. According to his approach, capabilities are inherently unobservable and

manifest themselves by indicators whereas functionings are unobserved and

hence measured with error. Therefore, in most of the literature, capabilities and

functionings are represented by latent variables (see, for example, Anand et al

2011).

The three aspects of the capabilities approach have been the focus of

measurement in the literature: achieved functionings, capabilities or achievable

functionings, and conversion efficiency of utilisation function. I review the

literature in the same order.

2.3.1 Measurement of Achieved Functionings

Kuklys (2005) measures two functionings – “being-healthy” and “being

well-sheltered”, each in turn measured by a range of indicators. The independent

variables are resources, such as income or education, and conversion factors are

age, marital status, or region of living. She uses MIMIC (Multiple-Indicators

Multiple-Causes) models (a special case of SEM) to analyse these two

20 For survey of theoretical literature related to the capabilities approach see Robeyns (2000, 2005). Surprisingly, there is no attempt to develop a theoretical model within Sen’s capabilities approach framework except the notable attempt by Kuklys (2005) for disabled individuals which I discuss in this review. Alkire (2005) reviews the empirical issues in measuring freedoms.

29 | P a g e

functionings. These functionings are treated as latent endogenous variables. She

estimates the model parameters using the individual as a unit of analysis in the

British Household Panel Survey (BHPS), and establishes that resources such as

income and education have little impact on functionings achievement while

conversion factors such as gender, age, and region of living have a significant

impact.

2.3.2 Measurement of Capabilities or Achievable Functionings

Anand and Hees (2006) demonstrate that it is possible to design a

questionnaire to distinguish between capabilities and functionings. Using the

survey instrument they develop the data required for the capabilities approach

and try to measure satisfaction or happiness with capabilities. They examine the

capabilities approach in the following seven dimensions: happiness, sense of

achievement, health, intellectual stimulation, social relation, environment, and

personal projects. They use ordinal logistic regression models and Spearman

rank correlations for the analysis of survey results. One of their notable findings is

that higher income levels are associated with lower capability satisfactions. This

may indicate a trade-off between objective improvement and subjective

dissatisfaction. Another important finding is that people use their own capabilities

to make judgments about the distribution of opportunities within society, except in

the areas of heath and the environment.

Anand et al (2005) develops a new survey instrument to elicit information

about capabilities at the individual level. The paper finds that many capability

indicators are highly correlated with happiness after controlling for socio-

demographic and personal variables.

Krishnakumar (2007), and Krishnakumar and Ballon (2008) propose a

theoretical framework that encompasses all important features of the capabilities

approach and provides a basis for structural equation modeling using real world

data. Krishnakumar (2007) considers the following three capabilities and uses

UNDP and World Bank databases in addition to other datasets for measurement;

“knowledge”, “health”, and “political freedom”. The indicators used for each

capability are: adult literacy rate and gross enrollment ratio for “knowledge”, life

30 | P a g e

expectancy at birth, infant mortality rate and under-five mortality rate for “health”,

political rights, civil liberties, and voice and accountability for “political freedom”.

Alongwith these indicators she uses a broad range of possible observed

exogenous variables for structural and measurement parts of the model.

She finds highly significant coefficients for most of the indicators of

capabilities and hence concludes that the selected indicators reflect the latent

dimension satisfactorily.

The political exogenous factors turn out to be insignificant in the measurement

model but some of them are significant in the structural model. The interactions

among the latent variables in the structural model show a positive and a

significant impact of health on education which in turn has a positive effect on

political rights. She also concludes from the results that greater political freedom

leads to better health status.

Based on her estimations of the model, she computes an aggregate capability

index (ACI) as a weighted average of the factors scores using the inverse of their

variance as weights. She then compares her ranking of countries on the basis of

ACI with that of HDI and finds a strong correlation between the two indices.

Krishnakumar and Ballon (2008) estimate two basic capabilities – knowledge

and living conditions – relating to children in Bolivia. Their results show strong

interdependence between the two basic capabilities. The results also highlight

the significance of the role played by demand and supply factors for acquiring

these capabilities.

Kuklys (2005) develops a method for the estimation of capabilities sets which

takes into account differential needs of individuals especially in the context of

disabled individuals. The estimation results show that a disabled individual needs

1.56 times the income of a healthy individual to achieve the same level of income

satisfaction, i.e., his consumption set is only 1/1.56 the consumption set of a

healthy individual. In other words, the capabilities set of this individual is only

64% of the capabilities set of non-disabled. In this way, she shows how a welfare

measure adjusted for a disability reflects a correct picture of social welfare.

31 | P a g e

All of the above studies claim to measure capability but, in fact, measure

functionings. Aggregating capabilities (functionings) across households involves

a number of problems as discussed in Kuklys (2005). One of these studies

(Krishnakumar and Ballon; 2008) considers functionings as indicators for

capabilities without taking into account the freedom aspect. Kuklys (2005) argues

that functioning may be considered a latent variable with many indicators.

2.3.3 Measurement of Conversion Efficiency of the Utilisation Function

Some of the studies attempt to measure efficiency with which individuals

convert their resources into achieved functioning.

For instance, Binder and Broekel (2011) use non-convex order-m frontier

estimation in a two-stage method to assess the conversion efficiency of the

basket of “basic functionings” namely: “being happy”, “being educated”, and

“being healthy” for the British Household Panel Survey (BHPS) wave 2006

dataset. They find that 76.64 % of the individuals in the sample are not able to

transform their resources into functioning achievement as efficiently as the best

23 %. Their results also show that the average inefficient individual achieves

about 33 % less functioning achievement than an efficient individual with same

resources and that resource-based welfare measures do not give correct level of

human welfare. They argue that a measure of conversion inefficiency reflects

diverse welfare-reducing institutional constraints on individuals.

Due to important unobservable conversion factors like intelligence, aptitude,

temperament, inheritance, and personal traits, an individual can not be

considered a “firm” and the methods used in technical efficiency measurement

may not be applied. Although there are few unobservable factors in the

measurement of technical efficiency of a firm, yet there are a number of issues in

these methods, for example, estimation of a large number of parameters (in

some studies more than a thousand), problems in aggregation of inputs and

outputs, unreliability of efficiency ranking due to overlapping of confidence

intervals (Jensen, 2000), assumption of homogenous output, and non-testable

assumptions like half-normal distribution.

32 | P a g e

2.4 Methodology

This section discusses our selection of functioning and the advantages of

subjective measures of capabilities over objective measures. It also describes the

methods employed to demonstrate that the new indicators embody information

not contained in the happiness indicator.

2.4.1 Selection of a Functioning

We consider being-achieved as an overall functioning, primarily because the

capabilities approach is ultimately concerned with the ability to achieve

combinations of valued functioning, as stated by Sen (2009, p. 233):

“Even though it is often convenient to talk about individual

capabilities […], it is important to bear in mind that the capability

approach is ultimately concerned with the ability to achieve

combinations of valued functionings”.

Some of the several reasons for taking a single overall functioning as a proxy

for combinations of valued functionings are:

i) Since the extent or nature of freedom (opportunity and process) is different

for different functionings, taking more than one functionings at a time

would be problematic since it would be very difficult to isolate freedom

associated with each functioning. That is why Alkire (2005, p.15)

contends:“Thus I argue that autonomy or process freedoms must be

evaluated with respect to each basic functioning. The reason for this is that

the autonomies required for a woman to decide to seek paid employment,

to be nourished, to plan her family, to vote, to attend literacy courses may

be present in varying degrees and it is precisely these variations that may

identify the ‘freedom’ associated with a particular functioning or a particular

deprivation”.

ii) Because of the complexity associated with measuring capabilities, it is

easier to analyse one functioning at a time in all its important capability

dimensions (functionings, freedom and conversion efficiency). There is an

apparent trade-off: taking multiple functionings only in one dimension or

33 | P a g e

taking a single functioning in all its dimensions. By taking a single specific

functioning we can avoid the problem of aggregating multiple functionings,

it however may create a problem of omitted functioning bias. Kuklys (2005)

highlights (a) the selection of relevant functionings, (b) measurement of

functionings at the individual level, (c) aggregation of functionings into a

composite measure of individual welfare, and (d) aggregation of a

functioning across individuals, as methodological problems in measuring

functionings. These problems however can be avoided, or greatly

minimised at least, when we consider an overall functioning – “being

achieved” – which gives a sense of achievement in life.

2.4.2 Measurement of Capability

There are at least two distinct ways to measure capability dimensions in the

empirical literature on capabilities (Anand et al, 2011): direct measurement of

capability dimensions by self-reported questionnaire consistent with theory (e.g.

Anand and Martin, 2006, Anand et al, 2011, and Ramos and Silber, 2005), and

indirect measurement by constructing latent variables for capability dimensions

(e.g. Kuklys, 2005, Krishnakumar, 2007, and Krishnakumar and Ballon, 2008).21

We resort to a method that lies somewhere in-between the two: capabilities’ are

measured directly from self-reported questionnaire by categorising questions on

the basis of the literature on capabilities, rather than measuring them

atheoretically, using latent variable modeling techniques.22

In this study capability is measured subjectively,23 primarily because of the

fact that the PSES questionnaire that we are using is subjective in nature. It is

21 The method used to measure SWB dimensions by self-reports is commonly referred to as Experience Sampling Method as opposed to Kahneman’s Day Reconstruction Method. See Kahneman and Krueger (2006) for discussion. 22 Robeyns (2011) argues that “Moving from ideal theory to non-ideal theory and empirical applications makes the selection of relevant capabilities even more complicated […], ranging from substantive proposals with elaborate theoretical underpinnings,[…], to the atheoretical practice that an investigator should simply conduct a survey in order to collect rich data (or use an existing survey) and let a statistical technique, such as factor analysis, ‘decide’.” 23 On the reliability of SWB measures, Krueger and Schkade (2007) found that both the overall life satisfaction measures and the affective experiences measures derived from the Day Reconstruction Method showed test-retest correlations in the range of 0.5-0.7

34 | P a g e

also worthwhile to alert readers to the following problems associated with

objective measurement:24

i) Objective measurements depend on revealed preferences and not on

actual choices; it is therefore not obvious whether a preference is

voluntary or involuntary. This is because both are observationally

equivalent (for example, voluntary and involuntary types of unemployment

are indistinguishable, but each type requires a different policy response).25

The subjective measures of freedom, functioning, efficiency and

happiness, have intrinsic value and may have instrumental value as in the

case of capabilities, while objective measures such as income and

education have instrumental or derivative significance.26

ii) Sen himself argues in favour of self-reflective and deliberate judgment of

people about the valuation of their lives. What is valuable for an individual

cannot be judged without considering his/her views about it.27 It is,

therefore, the mental state that determines the behaviour of an individual;

a person committing suicide in the presence of all luxuries of life simply

shows that he viewed his life as worthless. Sen (1991, p. 20) commenting

on the connection between welfare, preference, and freedom writes:

“If individual preference is what counts, then role of ‘the good

of the individual’ has to be derivative, unless, of course, the

good of the individual is simply defined as the fulfillment of

what the individual prefers (no matter what his or her

motives may be)”. (Italics in original).

which, they conclude, are sufficiently high to support much of the research on SWB. On the validity of SWB measures, see Bruni et al (2008, p. 74-75). 24 Anand et al (2005) also measure capabilities subjectively for the UK but by using a different set of questions. They however do not explicitly distinguish between functioning, freedom, and efficiency. 25 Sen (1973) discusses at length the problems with the revealed preferences approach. See Sen (1971), among others, for discussion on the difference between observed and unobserved choices, and weaknesses of rationality axioms. 26 See Sen (1991) on these issues. 27 In the case of prisoners’ dilemma, for example, what is individually desirable may not be optimal.

35 | P a g e

Sen (1977) critically examines behavioural foundations of economic theory

and shows self-interest as an unrealistic premise.

i) Objective wellbeing is a mean to attain subjective wellbeing as an end. A

measure is more useful if measured by output (subjective wellbeing) rather

than by inputs (objective wellbeing measures) alone, since preferences

are state-dependent and a state is largely dependant on the mental

state.28

ii) Subjective measurement encompasses a number of factors which are

difficult to measure objectively. Commenting on the direct welfare effects

of an act of choice, Sen (1997a, p. 748) writes:

“The person’s wellbeing may be affected directly by the

process of choice […], and this requires that the reflective

utility function (and the person’s conception of her self-

interest) be defined not just over culmination outcomes

(such as final commodity vectors, as in standard

consumer theory), but inter alia also over choice

processes and their effects.”

Hausman and McPherson (2009, p.1) argue that preference-satisfaction basis

of wellbeing is questionable. They argue that:“Yet it is obvious that people’s

preferences are not always self-interested and that false beliefs may lead people

to prefer what is worse for them even when people are self-interested. So welfare

is not preference satisfaction, and hence it appears that cost-benefit analysis and

welfare economics in general rely on a mistaken theory of wellbeing.”

Further, instability and inconsistency in preferences are discussed in Sugden

(2010) and Bykvist (2010) as weaknesses in preference satisfaction as a criterion

of wellbeing.

28 Sen (1991) criticises exclusive reliance on mental states for measuring welfare. See Robeyns (2011) on the issue of capabilities and utilitarianism.

36 | P a g e

Objective measurement is more vulnerable to the problem of

endogeneity/simultaneity than subjective assessments. Objective achievements

can have a feedback effect and may be different from actual achievement since

actual achievement varies from person to person (because each person has

different goals in life). Van Ootegem and Spillemaeckers (2008, p. 19) elaborate

on this point: “Resources (however defined) are only a means (however

important) to produce wellbeing. Moreover, the effect of income on wellbeing will

be distinct for different people. Therefore, a comprehensive theory of wellbeing

starts from the possibilities and opportunities (capabilities) at an individual’s

disposal.”

2.4.3 A Comparison of Dimensions of Wellbeing

We compare the distributions of capabilities and happiness with each other

using the tools of exploratory data analysis (EDA)29: boxplots and histograms,

and the formal statistical tests for equality of distributions. We regress happiness

on different dimensions of capabilities, under different controls, to see how

important capabilities are in determining happiness.30

This is followed by a ranking exercise, where we rank each district by

capabilities and happiness. These rankings are obtained using the following

procedure:

i) The self-reported score for each question (j) is added up to obtain a score

of a dimension (D) for each individual (i) in the survey. Since there are J

questions in each dimension, these are summed up to obtain scores for

that dimension, i.e.,

(1)J

i jij

D Q= ∑

29 “Unless exploratory data analysis uncovers indications, usually quantitative ones, there is likely to be nothing for confirmatory data analysis to consider.” (Tukey, 1977, p.3) 30 We use the Kruskal-Wallis equality-of-populations rank test to test the relevance of each question in a dimension, since the response to each question is measured on a discrete (ordinal) scale, while the distribution of each dimension is compared with each other using the Kolmogorov-Smirnov two-sample test since each dimension is measured on a continuous scale.

37 | P a g e

All these dimensions are rescaled between zero and one to measure

deprivation using the following formula:31

[ min ]= (2)[max min ]

ii

D DRSDD D−

−

where RSD is rescaled D.

ii) The score for each dimension in a district is obtained by taking a simple

average of the RSD scores over all individuals in a district. Outliers are

identified through boxplots for each district and dropped before computing

the district averages.32

The average district scores are used to rank districts in all capability

dimensions and their averages are used to construct a composite index,

Subjective Capability Index (SCI), which is a simple average of SA, SFTA

and SATA.

The capability rankings are compared with our happiness rankings,

which are further used to identify district-based policy focus.

2.5 The Data

We use the Pakistan Socio-Economic Survey (PSES) 2002 dataset (at

individual level)33. This is a unique dataset which has information on capability

dimensions. PSES surveys all urban and rural areas of the four provinces of

Pakistan (Punjab, Sind, Baluchistan, and NWFP34) defined as such by the 1981

population census excluding FATA (Federally Administered Tribal Areas), military

restricted areas, districts of Kohistan, Chitral, Malakand, and protected areas of

NWFP. The population of the excluded areas constitutes about 4 percent of the

total population. 31 This formula has been extensively used in HDRs. 32 Using median to exclude outliers, as an alternative, does not alter our conclusions as the two scores are very close to each other in our data. 33 The PSES (2002) is based on round II of the PSES. The sample design for round II is based on the sample design of round I conducted in 1998. Details of the sample design are given in Arif et al (2001) and Siddiqui and Hamid (2003). 34 NWFP is now known as Khyber-Pakhtoonkhwa.

38 | P a g e

A two stage stratified sample design is adopted for the 1998-99 PSES.

Enumeration blocks in urban areas and Mouzas/Dehs/villages in rural areas are

taken as primary sampling units (PSUs). Households within the sampled PSUs

are taken as secondary sampling units (SSUs). Within a PSU, a sample of 8

households from urban areas and 12 households from rural areas is selected.

Households covered during round I of the PSES are revisited during round II in

2000-01. After some adjustment due to attrition, the total sample for round II of

the PSES turns out to be 4021 households (2577 rural and 1444 urban).

The dataset comprises of 6749 individuals who directly responded to the

subjective questionnaire (21 questions), after list-wise (subject-wise) deletion of

the missing values. Since the number of missing values is very low (around 2%)

and their pattern is random (i.e., missing at random), deleting them in this way

will not cause any statistical problems such as bias.

The PSES uses twelve questions about mental wellbeing used by the British

Household Panel Survey (BHPS) and adds nine more valuable questions. These

additional nine questions are very important for subjectively measuring

achievement (functioning), freedom to achieve, and ability to achieve (conversion

efficiency). In fact the BHPS questions help to measure sense of freedom only,

whereas the additional nine questions in the PSES help to measure the ability to

achieve and achievement subjectively, which are important dimensions of

capabilities ignored by other surveys.

In what follows we describe the questions used in PSES to quantify different

dimensions of capabilities and happiness. At the outset, it is important to

appreciate the fact that the questions posed under each indicator adequately

serve the purpose of “being-achieved’ in a generalised sense.

i) Sense-of-freedom-to-achieve (SFTA) consists of three senses of

freedoms: freedom of action, freedom of decision making, and freedom of

problem solving. The following survey questions approximately define

these senses:

39 | P a g e

Table 1.1 Indicator: Sense of freedom to achieve Question Statement Categories

Q.1 Have you recently felt that you are playing a useful part in things?35

Q.2 Have you recently felt capable of making decisions about things?

Q.3 Have you been able to face your problems?

1. More so than usual

2, 3…

4. Much less usual

The sense of freedom to act and participate captures whether or not

people are allowed to engage in useful activities they value. The question

about playing a useful part in things shows one’s freedom to do useful

activities that matter to one’s interest (the agency aspect). The agency

aspect is concerned with seeking goals, performing religious duties, or

fulfilling social responsibilities.

The question about being capable to make decisions reflects freedom

in decision making. The reasons for the importance of perceived freedom

are given below:

First, this is a very important question as far as the democratic election

process is concerned. An election process can be shown transparent

amidst imposed implicit decisions on majority of voters by, for example,

feudal lords particularly in rural areas. Although it affects their sense of

freedom in decision making, yet it is not reflected in any objective criterion.

Second, freedom in decision making is also a major concern in gender

and ethnic issues. In some societies females are not encouraged to make

decisions about their careers. This adversely affects the freedom of

women to achieve. In some regions, minority ethnic groups similarly do not

have the freedom to proceed in their preferred careers. On the contrary,

some systems favour a minority elite class. This severely affects the sense

of freedom in the majority though legally everyone has equal freedom. 35 “The process aspect, being concerned with the freedom of the person’s decisions, must take note of both (iia) the scope for autonomy in individual choices, and (iib) immunity from interference by others” (Sen, 2002).

40 | P a g e

This fact cannot be captured by an objective criterion since written

documents and laws do not discriminate between the elite (the minority)

and the non-elite (the majority) classes. The biasedness towards minority

elite class can not be overcome by providing equal freedom to all because

of unequal and unjust initial endowment; “But here the choice is not

independent of previous conditions of inequality. Identical capability sets

do not afford the same real chance, in practice, of achieving valuable

functionings, and the reason for this difference is aspirations formed in

previous unequal and unjust conditions” (Burchardt, 2009, p. 9).

The last question regarding the ability to face up to problems reflects

decision making ability in an adverse situation.

ii) Sense-of-ability-to-achieve (SATA) is based on the following survey

questions:

Table 1.2 Indicator: Sense of Ability to achieve Question Statement Categories

Q.1 Do you normally accomplish what you want to? Q.2 Do you feel you can manage situations even when

they do not turn out as expected? Q.3 Do you feel confident that in case of a crisis you

will be able to cope with it?

1. Most of the time 2, 3... 4. Hardly ever

These questions address the sense of ability at three levels of difficulty

– from a normal situation to a situation of crisis. SATA is a proxy for

physical and psychological ability of an individual to convert his/her

material and non-material resources into achievement. Accomplishment is

one of the five components of wellbeing proposed in wellbeing theory by

Seligman (2011) in the field of positive psychology.36

36 The other four are: positive emotion, engagement, relationships, and meaning and purpose.

41 | P a g e

iii) Sense of achievement (SA) is based on the following survey questions:

Table 1.3 Indicator: Sense of achievement Question Statement Categories

Q.1 Do you think you have achieved the standard of living and the social status that you had expected?37

Q.2 How do you feel about the extent to which you have achieved success and are getting ahead?38

Q.3 Do you feel life is interesting?

1. Very much 2, 3… 4. Not so much

The first question covers one of the dimensions of HDI – access to a

decent standard of living – but in a subjective way. It complements HDI by

adding information about level of satisfaction with standard of living. This

level of satisfaction also takes into account aspirations and feelings about

the relative standard of living. The last two questions support these

feelings.

The first question regarding standard of living may be subject to the

same criticism as a happiness indicator, e.g., adaptation problem.

Including expectations somewhat minimises the effect of adaptation as it

asks about the living standard relative to expectations, unless there is

reason to believe that expectations by themselves suffer from adaptation

(see chapter 8 for the analysis of adaptation problem).

iv) Happiness (HAPP) is based on the following survey questions:

Table 1.4 Indicator: Happiness Question Statement Categories Q.1 Have you been feeling reasonably happy, recently

considering all difficulties? Q.2 Compared with the past, do you feel your life is Q.3 On the whole, how happy are you with the kind of

things you have been doing in recent years?

1. More so than usual 2, 3… 4. Much less usual 1. Very happy 2, 3… 4. Not so happy

37 ‘Functionings [achievements] are, in a sense, more directly related to living conditions, since they are different aspects of living conditions’. (Sen, 1987) 38 “[…] opportunity-freedom cannot be sensibly judged merely in terms of possession of commodities, but must take note of the opportunity of doing things and achieving results one has reason to value” (Sen, 2002).

42 | P a g e

These three questions ask about happiness – in general, in the past,

and in the present, and give us a reliable overall picture of happiness. In

some studies, happiness indicator is constructed by the twelve questions

in general health questionnaire (GHQ).39 Some questions in GHQ are not

directly relevant to happiness but rather to some other dimensions of

SWB.

It is important to note that the first two indicators in Table 1.1 extract

information on SFTA that are recent in nature and the last indicator complements

it with an overall sense of freedom to achieve. Similarly, in Table 1.2, the first

indicator manifests the SATA relevant to the normal course of life, and indicators

2 and 3 complement it with the SATA under unexpected circumstances (which

would be effected by any recent experience as well as other past experiences). A

similar pattern is observed by the SA indicators and the happiness indicators, in

Table 1.3 and Table 1.4 respectively, as they extract information relevant to the

recent past and those relevant during the normal course of life which serves the

purpose of modeling capabilities of “being achieved” in a consistent manner.

The following section shows how each indicator is different from the other

indicator by comparing the distributions graphically and statistically.

2.6 Results and Discussion

A comparison of the boxplots in Graph 2.1 reveals that the three dimensions

of capabilities have different distributions.40 The middle 50% of the data for SATA

(efficiency) is located very tightly around 0.4 whereas middle 50% of the data for

SA (functioning) is well spread out between 0.2 and 0.4. The middle 50% of the

data for SFTA (freedom), on the other hand, is clustered around 0.6. There is no

overlap of the middle 50% of SFTA and the first two dimensions (SATA and SA).

39 GHQ is a part of the British Household Panel Survey (BHPS). However, happiness is measured by a single question in HDR, 2010. 40 The Kruskal-Wallis equality-of-populations rank test is applied to each indicator of a dimension. The null hypothesis of equality of distributions is rejected in each case at 1% significance level. This means that no distribution is redundant.

43 | P a g e

Graph 2.1: Boxplots for sense of ability to achieve (SATA), sense of achievement (SA), sense of freedom to achieve (SFTA), and happiness (HAPP) based on individual data.

Graph 2.2: Histograms for sense of ability to achieve (SATA), sense of achievement (SA), sense of freedom to achieve (SFTA), and happiness (HAPP).

44 | P a g e

Graph 2.3: Boxplots for sense of ability to achieve (SATA), sense of achievement (SA), sense of freedom to achieve (SFTA), and happiness (HAPP) based on district level data.

Graph 2.4: Histograms for sense of ability to achieve (SATA), sense of achievement (SA), sense of freedom to achieve (SFTA), and happiness (HAPP) based on district level data.

45 | P a g e

This may indicate that SFTA contains information not shared by the other two

dimensions.

Similarly, the middle 50% of the data on happiness (HAPP) is located firmly

between 0.4 and 0.6 and shares a small proportion with the first three capability

indicators. Similar conclusions can be drawn from a comparison of the

histograms in Graph 2.2; all histograms show distinct patterns.

Repeating the same exercise for district level averages of the data shows that

distinction in SFTA, SATA, SA, and HAPP is even more pronounced (See Graph

2.3 and Graph 2.4). There is no overlap of the middle 50% data across the four

subjective wellbeing dimensions in the boxplots.

Results of the data analysis can be summarized in the following three main

conclusions

i. Each of the dimensions being measured has different characteristics –

thus they measure different things and are independent of each other

in terms of underlying information.

ii. District rankings on the basis of happiness and the three dimensions of

capabilities are significantly different from each other.

iii. Sense of Freedom-to-Achieve (SFTA) stands out clearly in terms of

sharing its distribution with other dimensions and happiness. It shares

no part of its distribution with the other two capability dimensions while

it marginally shares its distribution with happiness. This is in contrast

with other two capability dimensions and happiness. They share

relatively a large portion of their distributions with each other.

Formally testing the equality of distributions hypothesis using Kolmogorov-

Smirnov two-sample test provides further support to our conclusions from the

comparisons of boxplots and histograms. The null hypothesis of equality of

distributions is rejected in each case at less than 1% significance level. It implies

that each distribution provides useful information about wellbeing not contained in

the other distributions.

46 | P a g e

Table 2.1: District ranking District Happ SCI Eff Func Frdm BADIN 56 52 54 40 56 THARPARKAR 55 53 48 54 54 MEKRAN 54 56 56 53 53 SAWAT 53 54 55 52 49 JHELUM 52 47 40 48 39 JACOBABAD 51 55 53 56 32 RAWALPINDI 50 51 51 49 50 LORALAI 49 8 1 12 29 SHIKARPUR 48 50 44 55 20 KALAT 47 27 31 30 31 KARAK 46 38 6 51 52 SIBI 45 49 46 47 35 JHANG 44 48 43 44 44 MANSEHRA 43 9 8 34 2 HYDERABAD 42 46 47 35 38 KARACHI 41 44 36 37 47 THATTA 40 31 42 23 25 DADU 39 37 35 43 18 SARGODHA 38 41 34 38 43 LARKANA 37 25 24 27 46 R.Y.KHAN 36 26 27 19 36 FAISAL ABAD 35 42 38 36 41 MUZAFFARGARH 34 40 20 50 37 BANNU 33 20 21 21 26 SAHIWAL 32 45 49 29 40 LEIAH 31 30 23 45 34 NAWAB SHAH 30 12 7 26 9 SANGHAR 29 21 39 11 14 D.G.KHAN 28 16 12 39 12 GUJRAT 27 13 3 22 21 SHEIKHUPURA 26 28 19 41 42 QUETTA 25 17 17 20 23 KHUSHAB 24 34 28 46 22 GUJRANWALA 23 24 29 18 24 BAHAWALPUR 22 10 16 7 19 MULTAN 21 18 13 32 30 MIRPUR KHAS 20 36 41 33 28 KASUR 19 23 26 10 48 SUKKUR 18 35 52 24 16 OKARA 17 29 32 25 33 KOHAT 16 43 45 42 17 T.T. SINGH 15 7 9 9 6 BAHAWALNAGAR 14 15 22 31 7 KHAIR PUR 13 39 50 14 45 BHAKKAR 12 14 18 16 10 PESHAWAR 11 6 14 8 4 DIR 10 32 33 13 55 SIALKOT 9 33 30 15 51 LAHORE 8 11 11 17 8 MIANWALI 7 22 25 28 15 RAJANPUR 6 19 37 4 27 ATTOCK 5 5 10 3 11 VEHARI 4 2 2 5 3 ISLAMABAD 3 1 4 6 1 ABBOTTABAD 2 3 15 1 5 MARDAN 1 4 5 2 13 Happ=Happiness, SCI= Subjective Capability Index, Eff=efficiency=SATA, Func=functioning=SA, and Frdm=freedom=SFTA

47 | P a g e

The fact that each dimension of capability contains information distinct from

the others and from that contained in the happiness indicator logically leads us to

expect different district rankings. This ranking is given in Table 2.1. The average

absolute difference in rankings between SCI, efficiency, functioning and freedom

from happiness rankings is 9, 11.9, 8.3 and 10.5, respectively. Given that there

are a total of 56 districts, these differences in rankings are significantly different

from zero and provide further support to the earlier results.41 Capability

dimensions, therefore, do matter.

2.7 Conclusions and Policy Implications

Happiness is an important measure of subjective wellbeing. It is however

a derived notion which among other things depends on Sen’s capabilities.

The distribution of happiness does not automatically imply the distribution of

capabilities. It is therefore useful to rank policy units on the basis of

capabilities to correctly identify unit-specific policy focus (see chapter 4 for

policy analysis).

While happiness is an important dimension of SWB, capabilities have a

standalone value as well as highlighted in epigram on the first page. This also

resonates well with the Authentic Happiness (AH) and Wellbeing (WB)

theories in positive psychology which distinguish happiness from other

subjective wellbeing dimensions (Seligman, 2011). The AH theory considers

happiness as being uni-dimensional whereas the WB theory regards it as a

multi-dimensional concept with accomplishment (or achievement) as one of

its dimensions.

This chapter identifies SA (sense-of-achievement), SFTA (sense-of-

freedom-to-achieve) and SATA (sense-of-ability-to-achieve) as capabilities of

41 Rank tests may not be valid here since these tests check independence of distributions whereas the important point to note here is the absolute differences in ranking which matters in case of policy implications. The thesis therefore uses absolute difference in ranking instead of rank test to quantify the difference. Note that the absolute difference is significantly different from zero. This conclusion is also supported by the boxplot analysis above and the Kolmogorov-Smirnov two-sample test results.

48 | P a g e

“being achieved”, which represents Sen’s functioning, freedom and

conversion efficiency, respectively. Focusing on capabilities of being achieved

addresses important issues related to operationalising the capabilities

approach. Data analysis shows that each of the dimension being measured

has different characteristics- thus they measure different things and are

independent of each other in terms of underlying information.

For policy purpose, it is important that we study feelings of individuals who

are the ultimate targets of policies. Layard (2006, p. C29) aptly comments

about the importance of knowing the feelings of people for policy makers: “At

present our policies are based far too much on policy-makers’ judgements

about how they would feel in a given situation, rather than detailed studies of

how people actually feel”.

Since an individual’s achievements depend on personal goals, which vary

from person to person, these achievements therefore can not be measured

objectively as they are individual-specific. The most appropriate way,

perhaps, to quantify this is to ask a person about his/her sense of

achievement. PSES is the only survey that collects such information on all

aspects of capabilities (functioning, freedom, and efficiency). This data

contains distinctive information not present in the happiness indicator and

could be used to rank policy units and identify unit-based policy focus.

We do not insist that the questions used in PSES are the ones that should

be used in future research/surveys. These questions can be improved in a

number of ways to capture additional aspects of capabilities; the questions for

example, more or less, ask about an individuals’ assessment of his/her

happiness and capabilities at a given level of resources without any reference

to a reference group/state. Asking for example a question like “How happy do

you think you are…..” is not the same as “How happy do you think you are

relative to those living in Islamabad (a relatively high income, developed

city)”. The purpose of this study is to demonstrate that capability dimensions

provide information distinct from those contained in the happiness indicators

which provide another good reason to have capability-based rankings for

policy implications.

49 | P a g e

This sort of ranking is important since the distribution of happiness does

not necessarily imply the distribution of capabilities. Functioning and

capabilities have intrinsic as well as policy/instrumental value as they provide

information on mental health and have implications for happiness (Sen,

1985b). The next chapter moves on to investigate the existence, importance

and stability of the claimed association between happiness and different

dimensions of capabilities.

50 | P a g e

Appendix A2 Boxplot

The box plot was first introduced by John Tukey in 1977 as a tool of

Exploratory Data Analysis (EDA). Its recent revival is due to advancement in

computer graphics. It summarises the data by 5 descriptive statistics – the

median, the upper and the lower quartiles (75th percentile and 25th percentile), the

maximum, and the minimum values. The difference between the upper quartile

and the lower quartile gives inter-quartile range (IQR) which is represented by a

box and it shows a spread of middle 50% of the data. As IQR is not affected by

outliers, it gives a better distribution of data. A line in the box indicates median or

central value when data are ordered. The vertical lines extending from the box,

called whiskers, end at the minimum and maximum values represented by short

horizontal bars on each whisker. If there are outliers, these bars are extended to

1.5 times the IQR. Any data point beyond this limit indicates an outlier.

The box plot gives useful information about skewness of a distribution as

indicated by the position of the box (if the box is positioned towards the lower end

then the distribution is positively skewed and vice versa. If a box is completely

located at one of the ends, then 75% of the data points are contained in the box),

kurtosis (peakedness) of the distribution as indicated by the size of the box (if the

box is thin relative to the whiskers then it indicates that a large number of data

are contained in a very small portion of the sample and hence distribution has a

thinner peak), and presence of outliers.

51 | P a g e

Chapter 3

Modeling Happiness with Capabilities

“[…] On that Day will men proceed in groups sorted out, to be

shown the deeds that they (had done). Then shall anyone who has

done an atom’s weight of good, see it. And anyone who has done

an atom’s weight of evil, shall see it”.

(Qur’an, 99: 6-8)

“Quality of life depends on people’s objective conditions and

capabilities.

Steps should be taken to improve measures of people’s health,

education, personal activities and environmental conditions. In

particular, substantial effort should be devoted to developing and

implementing robust, reliable measures of social connections,

political voice, and insecurity that can be shown to predict life

satisfaction”.

(Stiglitz et al, 2009, p. 15)

3.1 Introduction

The last chapter presented evidence that capabilities provide distinct

information not contained in the happiness indicator. Generally, happiness is

regressed over quantitative variables like education and income, and

demographic controls. This chapter adds capabilities of “being achieved” to

the list of determinants of happiness, using a lead from Sen (1985b). The

study demonstrates that capabilities are the most important and stable

determinants (in terms of size, sign, and significance) of happiness. Sen (in

Bruni et al eds., 2008, p.26-27) argues that functionings have intrinsic value

and most often lead to happiness:

52 | P a g e

“Finally, aside from the recognition that happiness is valuable in

itself, we must take note of the fact that the achievement of other

things that we do value (and have reason to value) very often

influences our sense of happiness – generated by that fulfilment.

[…] This can be of great ‘evidential’ interest in checking whether

people are succeeding or failing to get what they value and have

reason to value”.

“It is also important to be clear that the evidential role of happiness

does not entail that a thing is valuable only to the extent that it

yields happiness. Happiness may be linked with success, but the

metric of happiness need not be a particularly good guide to the

force and extent of our valuations in general. To confound the two

was part of the old utilitarian trap into which we have reason not to

fall. We have to keep distinct issues separate, and yet we must

take note of the way the achievement or failure of what we have

reason to value may in fact influence our happiness”. [Italics in

original].

3.2 A Brief Review of Empirical Literature

There is a vast empirical literature on happiness modeling, for example,

Easterlin (1974, 2004), Clark and Oswald (1994), Diener et al. (1995),

Veenhoven (1999), Di Tella (2001), Frey and Stutzer (2000, 2002), Kahneman

(1999), Kahneman et al. (2004), Vaan Prag and Ferri-i-Carbonell (2004),

Powdthavee (2010), among others. Our focus here would be on the empirical

literature that analyses happiness with capabilities. We first review the empirical

literature on happiness with capabilities and then comment on the general form of

the happiness model used in the happiness literature. At the end, we propose a

happiness function that incorporates capabilities.

There is not much empirical literature that explores happiness with

capabilities. This may be due to Sen’s critique of happiness as a measure of well

being. In spite of this, there are theoretical discussions about potential

53 | P a g e

reconciliation of these approaches in order to better understand human well

being and able to derive policy implications (see, for example, Comim (2005)).

The following happiness function is extensively used in happiness research

(Blanchflower and Oswald, 2011):

where h is self-reported happiness level measured with error (ε) , u is the

unobserved true happiness level determined by income (y), demographic

factors (z), and time (t).

Estimation of the happiness function, given above, may pose fewer problems

for a time-series/panel analysis than for a cross-sectional analysis; in case of

panel analysis, we can control for unobservable personality traits whereas in

cross-section analysis they may act as confounding factors and introduce bias in

the results. The time-series/panel analysis also obtains the advantage of the

finding that happiness is stationary over time (the Easterlin Paradox).

Anand et. al. (2011) attempts to model happiness with capabilities by

simultaneous equation latent variable models. Happiness is determined by ten

capability domains given the socioeconomic and personality factors. The study

applies a generalized linear latent and mixed model (GLLAMM) to allow for

unobserved individual heterogeneity and possible endogeneity.

Anand and Hees (2006) demonstrate that it is possible to design a

questionnaire to distinguish between capabilities and functionings. Using the

survey instrument they develop the data required for the capabilities approach

and try to measure satisfaction or happiness with capabilities. One of their

notable findings is that higher income levels are associated with lower capability

satisfactions. This study does not use capabilities as determinants of happiness

rather it explores satisfaction or happpiness with capabilities.

Anand et al (2005) develops a new survey instrument to elicit information

about capabilities at the individual level. The paper finds that many capability

54 | P a g e

indicators are highly correlated with happiness after controlling for socio-

demographic and personal variables.

3.3 Methodology

The present chapter attempts to show that unobservable personality traits can

be controlled to some degree in a cross-sectional analysis if capability

dimensions, as opposed to capability domains, are included in the happiness

equation in a subjective manner. The other advantage of including the capability

dimensions is that happiness can be explained in a better way by both subjective

and objective indicators of wellbeing with control variables. The capabilities

augmented happiness function is given below:

where Q is the capabilities function and it is determined by functioning (F),

freedom (R), and efficiency (E). The following section estimates various forms

of this capabilities augmented happiness function.

The variable “Happiness” is created by adding the responses to three

happiness questions and then standardising it. Initially OLS is used to estimate

the capabilites augmented happiness function. Later on alternative estimation

techniques like 3SLS and Ordered Logit, are used to check robustness of

estimation methods using the following simultaneous equation model:

( )( , , , , )

h u FF f R E y z t

==

3.4 Results and Discussion42

Table A3.1 reports results of OLS regressions, regressing happiness on the different dimensions of capabilities, under different controls. Similar to the capabilities scores, objective variables (income and education) are rescaled to lie between 0 and 1 as in equation (2). All variables, except dummies, are standardised43. Doing so does not affect standard errors but makes interpretation 42 All estimations are done using STATA 11.0. 43 Hartwig and Dearing (1979, p.57-58) write: "Thus, symmetrising distributions of variables by means of re-expression prior to the analysis of relationships between

55 | P a g e

more convenient.44 Coefficients on capabilities and objective variables are directly comparable. Notice that, in the absence of the three dimensins of capabilities (efficiency, functioning, freedom), the model explains only 8 percent of the variations in happiness (See Table A3.1). Introducing capabilities as determinants of happiness explains another 57 percent of the variations. These results strongly support the hypothesis that functioning, capabilities, and efficiency are the most important and stable determinants (in terms of size, sign, and significance) of happiness with respect to size, sign, and significance of coefficients across various model formulations. The stability is also robust to estimation methods used. Including other controls does not alter our conclusions.45 The same conclusions, in terms of sign and significance, hold when we instead resort to a simultaneous equation model, where happiness is determined by functioning and functioning by efficiency, freedom and other controls (see Table A3.2). Using ordered logit estimations (Table A3.346) or beta regressions (not reported) also does not change our main conclusions.47

Simultaneous equation model better represents Sen’s capability formulation (see, Sen, 1985b) as opposed to the simple OLS. Testing endogeneity of functioning using the Durbin–Wu–Hausman test (augmented regression test for endogeneity), see Davidson and MacKinnon, 1993, also favors the simultaneous equation model. The single equation model, however, shows better fit in terms of explained variations in happiness (around 65% and 45% respectively). This may be due to measurement error in capability dimensions, which are measured by proxy questions not formulated for this purpose or individual-specific

variables not only contributes to the analysis of non-linear relationships but also provides a solid basis for measures of explained variance and statistical significance[...]For example, a set of values can be reexpressed in terms of standard deviations from the mean, i.e., ‘standardised,’ by subtracting the mean from each value and dividing by the standard deviation." 44 The coefficient in this case would imply responses of the dependent variable to a one standard deviation increase in independent variable. 45 Diagnostic tests indicate that residuals do not significantly depart from normality, regressors do not suffer from the problem of high multicollinearity, and models are not misspecified. The heteroscedasticity-consistent standard errors are used as in some models residuals are not homoscedastic. Our results do not change when we use robust regressions recommended by Zaman et al (2001) and Atkinson (2009) implying the absence of outlier effects. 46 Results reported in Table A3.3 assume parallel-line regressions. We use the Brant test to find that the parallel-line assumption is violated. We apply the Ordinal Generalised Linear Model, OGLM, (which is appropriate when the parallel-line assumption is violated) and find no difference in our qualitative results and significance (apart from education in regression 3 which become insignificant). OGLM regression gives slightly smaller coefficient relative to our ordered logit estimations. 47 Beta regression assumes beta-distribution which is appropriate for variables bounded by 0 and 1.

56 | P a g e

characteristics not captured by capabilities and other independent variables per se.

3.5 Conlusions and Policy implications

Our results are consistent with the conclusions drawn by Ferrer-i-Carbonell

and Frijters (2004); qualitative results are similar whether one uses ordered logit

(probit) model or ordinary least squares in case of happiness data:

“In this paper, we found that assuming cardinality or ordinality of the answers

to general satisfaction questions is relatively unimportant to results. What matters

to the estimates is how one takes account of time-invariant unobserved factors”.

Income and education have a consistent positive level effect on happiness.

OLS estimates reveal that this is true only for males. This however does not

mean that education and income are not important as the underlying questions

largely represent happiness and capabilities at given resources. This is because

the questions posed do not explicitly ask for such comparisons. Q3 on happiness

for example is not the same as when you add to it “relative to those in the middle

class” or “those living in Islamabad (a relatively modern, developed area)”.

The fact that we mostly obtain significant coefficients points indicates that

people do value objective differences even when they are not explicitly asked to

make such comparisons. The rural-urban results show that education has a

consistently positive level effect on happiness in rural areas, whereas income has

consistently positive level effect in urban areas. This again points towards social

comparisons. This is an interesting result which implies that education contributes

towards happiness in rural areas whereas income contributs towards happiness

in urban areas. This is however not surprising as education is relatively scarce

and a symbol of social-status in rural areas and income likewise in urban areas.

Resorting to a simultaneous equation model, as in Table A3.2, however reveals

that education and income are equally important in rural areas and education

slightly more important in urban areas.

57 | P a g e

The stability of freedom, efficiency, and functioning as determinants of

happiness calls for public policy to enhance these determinants besides income

and education for improving happiness level.

58 | P a g e

Appendix A3

Table A3.1: OLS estimates: Dependent variable = Happiness (0) (1) (2) (3) (4) (5) Male Female Rural Urban Constant 0.09* - - 0.01 0.01 0.08 0.07* 0.06* 0.04** 0.10 Efficiency - 0.08 0.08 0.08 0.08 0.09 0.09 0.09 0.07 0.11 Functioning - 0.51 0.50 0.50 0.50 0.50 0.47 0.52 0.50 0.50 Freedom - 0.35 0.34 0.34 0.34 0.34 0.34 0.33 0.35 0.32 Income 0.13* 0.02 0.02* 0.02* 0.01** 0.02** 0.01 0.001 0.02** Education 0.21* 0.02** 0.02* 0.02* 0.02* 0.03 * 0.01 0.02** 0.02 D(gender) -

0.22* -

0.03** -0.03**

-0.03**

-0.02 -0.05**

D(urban) -0.04 0.01 -0.01 -0.03 0.004 D(balochistan) 0.03* 0.02 -0.02 0.07* 0.06** -0.01 D(nwfp)_ -0.01 0.01 0.08* -0.06** 0.04 -0.03 D(Punjab) 0.02 -0.11 -0.12 -0.10 -0.07 -0.15 R^2 0.08 0.65 0.65 0.65 0.65 0.65 0.6 0.7 0.65 0.67 Happiness, capabilities and objective variables are standardised. Coefficients highlighted in bold are insignificant, those marked with a *(**) significant at 5(10)%, and all others significant at 1%. D(.) are dummy variables. Prob> F = 0.000 in all above models. The values for Variance Inflation factor (VIF) in all above models are less than 2 indicating the absence of a serious multicollinearity problem. Table A3.2: 3SLS estimates (1) (2) (3) (4) (5) Male Female Rural Urban Equation 1 (dependent variable: Happiness) Functioning 1.058 1.054 1.043 1.043 1.039 1.069 1.023 1.030 1.051 R^2 0.45 0.46 0.46 0.46 0.47 0.35 0.55 0.45 0.48 Equation 2 (dependent Variable: Functioning) Constant - - 0.09 0.09 0.13 0.05* 0.04 0.74 0.17 Efficiency 0.33 0.31 0.32 0.33 0.33 0.26 0.38 0.33 0.32 Freedom 0.44 0.44 0.43 0.43 0.42 0.43 0.42 0.43 0.43 Income 0.059 0.055 0.055 0.052 0.067 0.043 0.042 0.059 Education 0.034 0.056 0.056 0.058 0.067 0.046 0.042 0.065 D(gender) -0.17 -0.17 -0.17 -0.17 -0.18 D(urban) -0.001 -0.01 -0.01 -0.01 D(balochistan) 0.004 -0.08** 0.10 0.06 -0.06 D(nwfp)_ -0.06* -0.04 -0.10 -0.02 -0.11 D(Punjab) -.055 -0.07* -0.06* 0.01 -0.13 R^2 0.42 0.43 0.44 0.44 0.44 0.37 0.51 0.43 0.45 All variables standardised, except dummies. Coefficients highlighted in bold are insignificant, those marked with a *(**) significant at 5(10)%, and all others significant at 1%. D(.) are dummy variables. P-values for Chi-square for all simultaneous equation models above are zero.

59 | P a g e

Table A3.3: Ordered Logit estimates (dependent variable= Happiness) (1) (2) (3) (4) (5) Male Female Urban Rural Efficiency 1.00 0.96 0.99 0.99 1.10 1.01 1.27 1.28 0.96 Functioning 6.36 6.32 6.29 6.29 6.34 5.90 6.76 6.07 6.50 Freedom 6.31 6.28 6.27 6.27 6.17 5.54 7.02 5.88 6.40 Income 1.63 1.57 1.54 1.21* 1.67** 1.06 1.47** 0.52 Education 0.12 0.17** 0.16** 0.19* 0.24* 0.08 0.19 0.16 D(Gender) -0.10* -0.10* -0.12* - -0.14** -0.09 D(urban) 0.02 -0.04 -0.07 -0.01 - - D(balochistan) 0.12 -0.07 0.37 0.02 0.20** D(nwfp)_ 0.02 0.2* -0.23** -0.08 0.11 D(Punjab) -0.33 -0.33 -0.32 -0.44 -0.23 Cut1 0.73 0.76 0.71 0.72 0.49 0.23 0.94 0.45* 0.58 Cut2 3.06 3.08 3.03 3.04 2.83 2.48 3.42 2.69 2.97 Cut3 4.75 4.78 4.72 4.73 4.53 4.23 5.06 4.39 4.68 Cut4 6.09 6.12 6.06 6.07 5.88 5.57 6.43 5.75 6.02 Cut5 9.37 9.40 9.35 9.35 9.18 8.49 10.20 8.80 9.47 Cut6 10.76 10.79 10.74 10.75 10.57 9.88 11.61 10.15 10.90 Cut7 12.22 12.26 12.21228 12.22 12.05 11.20 13.25 11.56 12.46 Obs 6749 3371 3378 2464 4285 LR statistic 7028 7040 7044 7044 7100 3104 4014 2685 4370 Pseudo R^2 0.3 0.3 0.3 0.3 0.3 0.26 0.34 0.3 0.3

Coefficients highlighted in bold are insignificant, those marked with a *(**) significant at 5(10)%, and all others significant at 1%. D(.) are dummy variables. Where cut1, cut2,…,cut7 are thresholds with no intrinsic value for our purpose. They are estimated for the purpose of coding and assumed to be the same for all individuals.

60 | P a g e

Chapter 4

Robustness of the Link between Happiness and Capabilities

“Allah commands you to render back your trusts to those to whom they are due; And when you judge between people, that you judge with justice; Verily how excellent is the teaching which He (Allah) gives you. Truly, Allah is Ever All-Hearer, All-Seer.”

(Qur’an, 4:58)

4.1 Introduction

In chapter 2, I used equal weights as a base case to generate aggregate

scores for our composite variables (capability dimensions and happiness),

following the approach adopted in constructing of the human development index

(HDI) by United Nation Development Programs (UNDP). This approach has been

defended in many studies (see, for example, Nguefack-Tsague et al, 2011).

There are, however, two issues worth noting (Haenlein and Kaplan, 2004): first,

assigning equal weights lacks theoretical justification, and second, some

indicators might be more directly relevant, and reliable, to the composite variable

then others, and should be assigned larger weights.

The advantage of using scaling methods is that we can use any theoretical

framework to link composite variables generated by the scaling. But a

disadvantage is that these weights have nothing to do with the theoretical

structure and that the weights are completely arbitrary.

The primary objective of this chapter is to check robustness of results

obtained in chapters 2 and 3; the base case. To this end, this chapter identifies

alternative ways of selecting the weighting schemes and check sensitivity of

regression coefficients and district rankings to various weighting schemes vis-à-

vis the base case which assumes equal weights. It also estimates alternative

general structural froms, underlying the alternative weighting shcemes, and

discuss their plausibility vis-à-vis the base case simultaneous equation model.

61 | P a g e

A method is valid for our purpose if it satisfies the following criteria:

i) The determination of weights takes into account the relative importance of

indicators,

ii) the weights incorporate assumed theoretical structure, and

iii) the theoretical structure is unaffected by the determination of weights.

The main alternatives to scaling methods that do not have the two issues (no

theoretical justification and degree of relevance of indicators to the composite

variable) mentioned above are: factor analysis (FA), principal component analysis

(PCA), and the structural equation modeling (SEM) methods.48 The FA and the

PCA are data reduction techniques and satisfy the first criterion only.

There are the two main techniques for SEM; the Linear Structural Relations

(LISREL) method due to Joreskog (1973) and the Partial Least Square (PLS)

Path modeling (as opposed to PLS regression) due to Wold (1966, 1982, and

1985).49 LISREL strongly satisfies the first two criteria but it weakly satisfies the

third criterion since structural relationship between latent variables50 may not

stick to the user given (assumed) structure but change according to the

covariance criterion in contrast to PLS Path Modeling (PLS-PM).51 PLS-PM

strongly satisfies the three criteria and hence is selected as an alternative to

check the robustness of scaling method used in the base case.

48 For critical analysis of structural equations and path models, see for example, Freedman (1987). Wold (1987, p.202) responds to Freedman’s rejection of the structural assumptions of path models and argues “Since the parameters of causal chain systems can be estimated by ordinary least squares (OLS), strong stochastic assumptions are not necessary”. 49 See Hammer (2006) for a detailed comparison between these two approaches. Also see Tenenhaus et al (2005) for detail discussion on PLS Path modeling. 50 See appendix A4.2 and A4.3 for the definitions of unobserved and latent variables. These definitions are given for the sake of understanding and completeness. As far as the analysis is concerned, it does not matter which definition is used. 51 Given the structure and the data in this chapter, LISREL changes the assumed structure and hence does not satisfy the second criterion as well.

62 | P a g e

4.2 Methodology

The following discussion briefly describes LISREL and PLS-PM methods

since they are close substitutes and an understanding of LISREL, a relatively

common method among economists as compared to PLS-PM, helps in

understanding the other.

LISREL is a merger of simultaneous equations models developed in

econometrics and factor analysis models developed in psychometrics. It is an

interpretative (explanatory) and confirmatory covariance-based technique which

usually requires a strong theoretical foundation in order to model causal

relationships and is based on the assumption that manifest variables are

multivariate normal. The estimators may be seriously biased if the underlying

distribution is far from being multivariate normal. Generally a small departure

from multivariate normality leads to inflate the chi-square value. The main cause

of violation of this assumption is the use of dichotomous or ordinal variables. The

most commonly used estimation method is MLE; but it is not robust in the

presence of ordinal or non-normal data. The idiosyncratic error term needs to be

i.i.d. normal. Huberet al (2004) suggest the use of the Laplace approximated MLE

(LAMLE) in this situation. An important limitation of the LISREL approach is the

assumption that all latent variables are continuous.

PLS-PM, in contrast to LISREL, is a predictive and exploratory variance-

based technique and hence it does not require a strong theoretical basis and is

distribution free. It has been extensively used in natural sciences particularly in

Chemometrics. It is useful when the sample is small, there are missing values,

and when there is a high multicollinearity problem, and is robust against skewed

distributions of manifest variables, multicollineariy, and misspecification of

structural model (see Cassel et al, 1999).

Ringle et al (2009) have shown in a simulation study that LISREL is a better

method when its prerequisites are met. Otherwise, the PLS-PM provides a viable

approximation of model parameters.52 As pointed out earlier, the capabilities

52 See table A4.13 for a comparison between PLS-PM and LISREL methods.

63 | P a g e

approach is a conceptual framework and not a theory (see Robeyns, 2005),

therefore, it is not advisable to use methods of confirmatory analysis in the initial

stage. Once a theory is formulated then it is strongly recommended to apply

confirmatory methods. Since most of the theories in social sciences are

conjectures, it is better to perform exploratory analysis before applying

confirmatory methods. Freedman (2005, p.193) clarifies this point and

writes:“There is no way to infer the ‘right’ model from the data unless there is

strong prior theory to limit the universe of possible models. (More technically,

diagnostics and specification tests usually have good power only against

restricted classes of alternatives). That kind of strong theory is rarely available in

the social sciences.”

Pirouz (2006) mentions the following key advantages of partial least squares:

PLS (i) is able to model multiple dependent as well as multiple independent

variables, (ii) can handle high multicollinearity53, (iii) is robust in the presence of

data noise and missing data, (iv) creates independent latent variables directly on

the basis of cross products involving response variable(s), (v) it gives stronger

predictions, (vi) allows for reflective and formative latent variables, (vii) is

applicable to small sample, (viii) does not require strong distributional

assumptions, and (ix) can handle a range of variables; nominal, ordinal,

continuous, and so on.

Some of the disadvantages of partial least squares include: (i) difficulty in

interpreting loadings of independent latent variables, (ii) distributional properties

of estimates are not known, (iii) failure to obtain significance unless bootstrapped,

and (iv) lack of model test statistics particularly the lack of a global optimising

criterion for goodness of fit (Trinchera and Russolillo, 2010).54

53 This is true in case of reflective indicators (when indicators covary and each indicator is an effect of the composite variable) but if indicators are formative (when indicators do not covary and the composite indicator is an effect of these indicators) then PLS regression should be preferred to OLS regression to counter the problem of multicollinearity. Similarly, if multicollinearity in structural model is high (as shown, for example, by variance inflation factor), then PLS regression should be used. 54 However, a goodness of fit (GoF) index may be used when measurement models are reflective.

64 | P a g e

LISREL has been used in the empirical capability literature by many researchers.

Kuklys (2005) measures functionings by various indicators using the LISREL

approach. She critically reviewes other methods used in the literature for

measuring functionings like scaling, fuzzy sets theory, factor analysis, principal

component analysis, and time series clustering.

Due to the limitations of factor analysis, principal component analysis, and

LISREL in the context of this chapter, I use PLS Path Modeling to measure capability

dimensions. However, the disadvantages of PLS-PM mentioned above, pose no

serious problem in the present analysis since it is used only for generating latent

variable scores in the first-stage. The inferential analysis will be done in the second-

stage using standard single or simultaneous-equation estimation methods.

In the first stage, exploratory PLS Path modeling approach is applied to figure

out the causal mechanism and estimate scores for the latent variables. In the

second stage, a confirmatory method is used to test the Stage-I model and

estimate the population parameters.

4.3 Estimation Process and Sensitivity of Estimation Results to Structural Models

The capability dimensions are measured and modeled in two stages:55 Stage-

I formulates a latent variable model in the PLS framework and obtains latent

variable scores (LVS) for each capability dimension using qualitative indicators.

The advantage of using PLS path modeling is to generate indices for indicators

and convert ordinal variables into continuous variables so as to use them in

stage-II; regression analysis.

4.3.1 Stage-I – PLS Path Analysis

The latent variable scores are constructed for every individual in the sample

by PLS path modeling for the four latent variables – functioning, freedom,

conversion efficiency, and happiness using their indicators.

55 The Australian Curriculum Assessment and Reporting Authority (ACARA) is also applying a two-stage method similar to this for the construction of Index of Community Socio-Educational Advantage (ICSEA) for comparing school performance in Australia.

65 | P a g e

I follow the general-to-simple methodology to model structural relationships

between latent variables. The term general-to-simple used here should not be

confused with the term general-to-specific (G2S) used and pioneered, among

others, by David F. Hendry in time-series econometric modeling. The difference

between the two is that G2S starts with a generalized model that contains all

relevant variables and reduces it to a specific model by dropping insignificant

variables/lags whereas the general-to-simple method, as used here, starts with a

general model in terms of containing all theoretically possible interactions

between latent variables and reduces it to a simple/parsimonious model by

deleting interactions between latent variables one by one if latent variable scores

remain robust to these deletions.

Before going into technical details of the PLS-path model, it is convenient to

start with a graphical representation of alternative models considered here in

fugues 4.1 to 4.4. A rectangle in the diagram represent a block of manifest

exogenous or endogenous variables (indicators). Exogenous manifest variables

are labeled as X and manifest endogenous variables as Y. Each manifest block is

of dimension LxN where L (3 here) is the number of indicators in each masifest

variable and N is the number of observations.

The exogenous and endogenous manifest block X and Y are are

manifestations of latent variables ( )Xξ and ( )Yη respectively represented by

eclipses. The ellipses show variables in the inner or structural model and

rectangles show variables in the outer or measurement model. Errors are not

shown.

In each of the following diagrams, there are three manifest variables

(indicators) in each block of endogeous (Y) and exogenous (X) manifest

variables. Figure 4.1 shows a general latent variable model for happiness with all

possible interactions. Each manifest variable has three indicators. X here is R

(freedom) with three manifest indicators (questions asked) and our endogenous

manifest variables are E, F and H. This specification assumes that freedom (R)

affects H directly as well as indirectly. The indirect effect feed through E and F, E

also has a direct impact on happiness and an indirect impact through F.

66 | P a g e

Figure 4.1 Latent Variable Model 1 (Path Diagram 1) for Happiness

Figure 4.2 drops the interaction between efficiency and freedom which

means that efficiency is now an exogenous manifest variable that affect

happiness directly and through its impact indirectly on F.


Figure 4.3 drops the direct link between E and R implying that E and R now

has an indirect impract on H scores. Notice that this is the simultaneous equation

model we used the previous chapter.

3Y (H Indicators)

1Y (E Indicators)

X (R Indicators)

2Y (F Indicators)

η (E)

ξ (R)

η (F) η (H)

2Y (H Indicators)

1X (E Indicators)

2X (R Indicators)

1Y (F Indicators)

ξ (E)

ξ (R)

η (F) η (H)

67 | P a g e


Figure 4.4 reduces the interaction further and now assume that all the three

dimensions (E, F and R) are exognous manifest variables which has direct

interaction with H (the only endogenous manifest variable). This is the simple

OLS model we used in chapter 3..


ξ (F)

1X

(E Indicators)

3X (R Indicators)

Y (H Indicators)

2X (F Indicators)

ξ (E)

ξ (R)

η (H)

2Y (H Indicators)

1X (E Indicators)

2X (R Indicators)

1Y (F Indicators) ξ (E)

ξ (R)

η (F) η (H)

68 | P a g e

The objective of SEM is to generate estimates that link happiness and

capability dimensions.56 The estimation of latent variables is carried out by the

interaction of the following three parts of the structural equation model:

i) The measurement model

The measurement model has three modes: (a) Reflexive mode, (b)

Formative mode, and (c) A combination of (a) and (b) called MIMIC (multiple-

indicators-multiple-causes) mode.

a) In reflexive mode the latent variables affect manifest variables and the

manifest variables covary and are highly correlated. Simple

regressions are used to obtain the loading coefficients using the

following equations:

(1)x x

y y

xy

ξ εη ε

= Λ += Λ +

where Λ shows the matrix of loading coefficients relating latent

variables to manifest variables.

The following equations are obtained, called predictor

specifications, by imposing the assumptions on the above equations in

order to assure desirable properties of least squares estimation

(Trinchera and Russolillo, 2010).

Assumptions:( )=0, ( , ) 0( )=0, ( , ) 0

Predictor specification:( | ) ( | )

x x

y y

x

y

E EE E

E xE y

ε ε ξε ε η

ξ ξη η

==

= Λ= Λ

56 This section is partly based on Lauro and Vinzi (2004).

69 | P a g e

b) In formative mode, manifest variables affect a latent variable and the

manifest variables should not be highly correlated and should not

covary. Multiple regressions are used to obtain the loading coefficients

using the following equations:

= (1 )=

x

y

XY

ξ

η

ξ δ

η δ

′Π +

Π +

where Π represents matrix of the regression coefficients and δ is the

error term. The following predictor specifications are obtained when

assumptions are imposed on the above equations:

Assumptions:( )=0, ( , ) 0( )=0, ( , ) 0

Predictor specifications:( | ) ( | )

x x

y y

x

y

E EE E

E xE y

δ δ ξδ δ η

ξ ξη η

=

=

= Π= Π

As far as the other assumptions are concerned, Lauro and Vinzi (2004, p.

204) remark:“[Predictor specification] avoids the classical i.i.d. assumptions for

which the observations need to be jointly ruled by a specified multivariate

distribution being also independently distributed. At the same time, it leads to

consistent estimates and minimum variance predictions.”

Lauro and Vinzi (2004, p. 204) while discussing the difference between

variance-based PLS-PM and covariance-based LISREL further write:“This is

consistent with the predictive objective of PLS that, differently from covariance

structure models, does not aim at minimising the residual covariance matrix by

reproducing the observed covariances. Consequently, the residual covariance

structure:

( ) ( ) ( )E E Eε δεε θ δδ θ ζζ′ ′ ′= = = Ψ

70 | P a g e

is not restricted and PLS aims at minimising the trace (variances) of Ψ and, in

case of reflective indicators, also the trace of εθ while the trace of δθ is

minimised in case of formative indicators.”

ii) Weight relations

a) Initial (outer) weights: The PLS-PM begins with arbitrary weights to

estimate exogenous and endogenous latent variables as linear

combinations of corresponding standardised manifest variables:

ˆ

ˆXY

ξ

η

ξ ω

η ω

=

=

The latent variables obtained are then standardised to have unit

variances. For simplicity, the initial estimates of latent variables are

denoted by hv :

That is, h jhjhv w x= ∑

b) Inner weights: Using the initial estimates for latent variables, weighted

aggregates of the adjacent latent variables ( hz ) are obtained:

That is h hh hz e v′= ∑

The inner weights ( hhe ′ ) are determined by one of the three

weighting schemes – centroid (uses sign of the correlations), factor

(uses correlation coefficients), or path (uses multiple regression

coefficients if latent variables are endogenous and simple regression

coefficients – or correlations since variables are standardised – if

variables are exogenous). However, it does not really matter in practice

which weighting scheme is used.

c) Outer weights: In case of reflexive indicators, weights are determined

by simple regressions (correlations):

71 | P a g e

( , )jh jh hw corr x z=

In case of formative indicators, weights are determined by multiple

regressions:

1( )h h h hw X X X z−′ ′=

The process iterates between inner and outer weights computation

and stops when convergence between hv and hz is achieved.

iii) Structural equation

This may be specified as

= + (2)η η ξ ζΒ Γ +

where

η is a k x 1 vector of endogenous latent variables,

ξ is a h x 1 vector of exogenous latent variables,

ζ is a k x 1 vector of errors,

Β is a k x k matrix of regression (path) coefficients, and

Γ is a k x h matrix of regression (path) coefficients, and it is

assumed that ( | ) 0E ζ ξ =

Equation (1) or (1’) shows relationship between manifest variables

(indicators) and the latent (composite) variables, while equation (2) shows the

relationship between latent variables. These two constitute the measurement

model. Equation (1) is used when indicators are reflective, i.e., they are the

effect of a latent variable whereas equation (1’) is used when indicators are

formative, i.e., they are affecting a latent variable. Equation (2) shows the

relationship between endogenous latent variables and exogenous latent

variables. Since the objective of Stage-I in the present study is to obtain latent

72 | P a g e

variable scores to be used in Stage-II, it does not matter which mode

(reflexive or formative) is used since latent variable scores obtained from both

modes are very highly correlated. However, reflexive mode is a better

representative of interactions between a latent variable and their indicators in

the present study since they are subjective responses.

4.3.2 Stage-II – Regression Analysis

Models estimated in the previous chapters with base-case wights are

reestimated with the LVSs generated by by the four PLS-Path models from

Stage-I with income and education as controls.57 Moreover, the chapter also

estimate structural equations underlying each model of the four models

presented above.


Tables A4.1 to A4.3 report the results based on four latent variable models

alongside our base case using single equation (OLS), simultaneous equation

(3SLS), and discrete choice (ordered logit) models. Estimates from these four

models and our base case results are quite similar in each of these three

econometric models/techniques. They show that results are not sensitive to the

choice of an econometric technique. This evidence also supports the validity of

results obtained in the last chapter using the scaling method.

Thus estimation results obtained in chapter 2 are robust with respect to latent

structural models. The empirical exercise in this chapter also sustains the

conclusion of the last chapter that freedom, functioning, and efficiency are stable

determinants (in terms of size, sign, and significance) of happiness.

Having said that, it might not be appropriate to use the latent variable scores

generated by latent structures in any functional form other than the one

underlying each model. In order to investigate whether or not this introduces any

serious bias in our estimates, estimates based on models underlying each latent

structure are reported in Table A4.4. Four main conclusions can be derived from 57 SmartPLS and MS Excel solver are used to generate LVS.

73 | P a g e

the table. They are: (i) capabilities are still important determinants of happiness,

(ii) freedom might have direct and indirect impact on happiness, (iii) efficiency

affect happiness indirectly through functioning, and (iii) when the first two

equations are reduced to a single equation through substitution, the impact of

capabilities on happiness is more or less similar in size. It is worth mentioning

that model 3 (our base case simaltanious equation model) results in standard

errors lower that the other two samaltanious equation models. This model is also

close to Sen’s theoritical formulation of capabilities (Sen 1985b)

The LSVs obtained for HAPP, SFTA, SATA, and SA from the four models

described in the previous section are used to rank districts. These ranksings are

also robust with respect to latent structural model. Figures A4.1 to A4.4

summarise the trends in ranking for HAPP, SFTA, SATA, and SA across four

latent structures. A close examination of these trends shows a consistent ranking

across the various latent structures relative to the base case. The consistency of

rankings demonstrates robustness of the base case to latent variable models.

Figure A4.5 compares district rankings for SFTA, SATA, and SA against HAPP.

Rankings for each dimension seem to be different from the happiness ranking.

A one-tailed one-sample t-test is used to test the statistical significance of

absolute differences in the size of rankings. Results are reported in Appendix D.

In all cases, the null hypothesis of average absolute difference of 3 is rejected in

favour of an alternative hypothesis of difference greater than 3. This shows that

these rankings are statistically different.These differences in rankings provide

evidence that the capability dimensions provide distinct information from the

happiness indicator.

4.5 Conclusion and Policy Implications:

The robustness of estimates to structural relations indicates that

circularity or simultaneity between capability dimensions is not an issue in the

present study. Hence the simplest structure (Latent Models 3 or 4) may be

used for policy design. i.e., policy makers can use the determinants of

efficiency, freedom, and functioning in order to directly affect happiness. This

structure helps policy makers to clearly monitor the marginal effects of

74 | P a g e

capability dimensions on happiness. It is also important to note that targeting

happiness alone by its determinants (other than functioning, freedom and

conversion efficiency) does not affect the district rankings in capabilility

dimensions since happiness indicator and capability dimensions provide

statistically significant distinct district rankings.

75 | P a g e

Appendix A4

A4.1 Sensitivity of the estimates to various models

Table A4.1: OLS estimates-Dependent variable = Happiness Variable model-1 model-2 model-3 model-4 base-case Efficiency 0.08 0.08 0.08 0.09 0.10 Functioning 0.53 0.52 0.53 0.52 0.53 Freedom 0.33 0.33 0.32 0.33 0.39 The base-case uses scaled scores whereas all other models use pls scores. All variables are scaled between 0 and 1. All coefficients significant at 1% level of significance Table A4.2: 3SLS estimates Variable model-1 model-2 model-3 model-4 base-case Equation 1 (dependent variable: Happiness) Functioning 1.05 1.0481 1.039 1.024 1.04 Equation 2 (dependent Variable: Functioning) Freedom 0.45 0.44 0.44 0.43 0.43 Efficiency 0.33 0.34 0.34 0.38 0.34 All coefficients were significant at 1% level of significance and R2 greater than 0.42 in all cases Table A4.3: Ordered Logit estimates: Dependent variable = Happiness Variable model-1 model-2 model-3 model-4 base-case Efficiency 0.25 0.25 0.25 0.31 0.27 Functioning 1.72 1.72 1.71 1.66 1.66 Freedom 1.00 1.00 1.00 1.02 1.05 Pseudo R^2 0.3 0.23 0.23 0.23 0.23 All coefficients are significant at 1% level of significance

Table A4.4: Estimates under different structural forms Variable model-1 (3SLS) model-2(3SLS) model-3(3SLS) Efficiency -0.06 (0.03) -0.001 (0.02) Functioning 0.86 (0.07) 0.72 (0.03) 1.04 (0.01) Freedom 0.21 (0.03) 0.26 (0.01) R2 0.60 0.64 0.50 Freedom 0.36 (0.01) 0.36 (0.01) 0.44 (0.009) Efficiency 0.41 (0.01) 0.42(0.01) 0.35 (0.009) R2 0.45 0.45 0.45 Freedom 0.49 (0.01) R2 0.24 Coefficients highlighted in bold are insignificant at 5%, all other are significant at 1% level of significance. Numbers reported in () are standard errors.

76 | P a g e

Figure A4.1 Happiness rankings against Base case

0

10

20

30

40

50

60

LORA

LAI

BAHA

WAL

PUR

D.I.K

HAN

QUE

TTA

ISLA

MAB

ADBH

AKKA

RM

IRPU

R KH

ASDA

DULA

HORE

LEIA

HKA

RAK

LARK

ANA

RAJA

NPUR

GUJ

RANW

ALA

SIAL

KOT

ABBO

TTAB

ADKA

SUR

PESH

AWAR

ATTO

CKSA

RGO

DHA

BADI

NSU

KKUR

BANN

USA

HIW

ALVE

HARI

THAT

TASA

WAT

FAIS

AL A

BAD

SANG

HAR

Districts

Rank

base-casemodel-1model-2model-3model-4

Figure A4.2 Sense of Achievement (SA) rankings against Base case

0102030405060

BAH

AWAL

NAG

ARKA

RAC

HI

ISLA

MAB

ADJH

ELU

MQ

UET

TAR

.Y.K

HAN

SIAL

KOT

VEH

ARI

ATTO

CK

SIBI

THAR

PAR

KAR

GU

JRAN

WAL

AM

IRPU

R K

HAS

LEIA

HM

EKR

ANM

ULT

AND

ADU

ABBO

TTAB

ADSA

HIW

ALKH

AIR

PU

RKA

SUR

T.T.

SIN

GH

THAT

TASH

EIKH

UPU

RA

KAR

AKJH

ANG

OKA

RA

SAN

GH

ARH

YDER

ABAD

Districts

Rank

ing


Figure A4.3 Sense of Ability to Achieve (SATA) rankings against Base case

0

10

20

30

40

50

60

LORA

LAI

D.G

.KHA

NLE

IAH

KARA

CHI

SIBI

BHAK

KAR

ISLA

MAB

ADRA

JANP

URR.

Y.KH

ANG

UJRA

TKH

USHA

BSA

HIW

ALAB

BOTT

ABAD

SAW

ATD.

I.KHA

NSI

ALKO

TKA

SUR

T.T.

SIN

GH

JHAN

GLA

RKAN

AKO

HAT

SUKK

URDA

DUNA

WAB

SHA

HM

ARDA

NKA

RAK

PESH

AWAR

SANG

HAR

JACO

BABA

D

Districts

Rank

ing


77 | P a g e

Figure A4.4 Sense of Freedom to Achieve (SFTA) rankings against Base case

0

10

20

30

40

50

60

LOR

ALA

IIS

LAM

AB

AD

MIR

PU

R K

HA

SJH

ELU

MK

OH

AT

THA

RP

AR

KA

RB

AH

AW

ALP

UR

LAH

OR

ELA

RK

AN

AA

BB

OTT

AB

AD

D.I.

KH

AN

MU

LTA

NK

AR

AC

HI

SH

EIK

HU

PU

RM

AN

SE

HR

AFA

ISA

L A

BA

DS

AR

GO

DH

AD

AD

UK

AR

AK

SA

WA

TS

UK

KU

RS

HIK

AR

PU

RK

HU

SH

AB

SA

HIW

AL

SIA

LKO

TD

IRV

EH

AR

IN

AW

AB

SH

AH

JHA

NG

Districts

Rank

ing


Figure A4.5 SA, SATA, and SFTA rankings against HAPPINESS

05

1015202530354045

LOR

ALAI

SIBI

BAH

AWAL

PUR

BAN

NU

MIR

PUR

KH

AS

DAD

U

BHAK

KAR

JHEL

UM

THAR

PAR

KAR

LAR

KAN

A

R.Y

.KH

AN

BAD

IN

SUKK

UR

SIAL

KOT

GU

JRAT

MU

ZAFF

ARG

ARH

T.T.

SIN

GH

ATTO

CK

OKA

RA

happ_rankingsa_rankingsata_rankingsfta_ranking

78 | P a g e

Table A4.5: Comparison of PLS-PM and LISREL

Criterion PLS-PM LISREL

Objective Prediction oriented Parameter oriented

Approach Variance based Covariance based

Assumptions Predictor specification (non parametric)

Typically multivariate normal distribution and independent observations (parametric)

Parameter estimates

Consistent as indicators and sample size increase

Consistent

Latent variable scores

Explicitly estimated Indeterminate

Epistemic relationship between a latent variable and its measures

Can be modeled in either formative or reflective mode

Typically only with reflective indicators (however procedures to consider formative indicators exist)

Implications Optimal for prediction accuracy

Optimal for parameter accuracy

Model complexity

Large complexity (e.g. 100 constructs and 1000 indicators)

Small to moderate complexity (e.g. less than 100 indicators)

Sample size Power analysis based on the portion of the model with the largest number of predictors. Minimal recommendations range from 30 to 100 cases.

Ideally based on power analysis of specific model – minimal recommendations range from 100 to 800.

Source: Andreas Hammer (2006).

79 | P a g e

A4.2 Unobserved variables

According to Kementa (1991), unobservable variables in econometrics are

represented in one of the following three ways: (i) variables with measurement

errors; (ii) proxy variables; and (iii) intrinsically latent variables. The last type of

unobservable variables is characterised by a number of indicators (manifest or

observed variables) or a number of observable causes. Because of the nature of

the capabilities approach (CA), I use the last representation of unobservable

variables in subsequent modeling. There are many definitions – formal and

informal – being used in the literature for a latent variable. Bollen (2002) reviews

some of these definitions. However, for empirical purpose, it does not matter

which definition of latent variables is used.

80 | P a g e

Chapter 5

Interaction of Capability Dimensions: A Theoretical Analysis

“O mankind! We created you from a single (pair) of a male and a female, and made you into nations and tribes, that you may know each other (not that you may despise each other). Surely the most honoured of you in the Sight of Allah is (he who is) the most righteous of you. And Allah has full knowledge and is well acquainted (with all things)”.

(Qur’an, 49:13)

5.1 Introduction

The previous chapters show that capability dimensions are stable

determinants of happiness in terms of size, sign, and significance. The analysis in

this chapter first looks at the correlations between capability dimensions and

happiness at aggregate level and at disaggregated levels; province level, district

level, and rural-urban level. It also looks at the correlations among these

variables for gender and for various years of education at aggregate level. The

objective of the correlation analysis is to figure out pattern(s) in correlations.

These patterns, in turn, help us understanding the interactions between capability

dimensions.

With some digression, the chapter then analyse how capability dimensions

interact with each other. The present chapter attempts to clarify linkages between

the three dimensions of capabilities – freedom (SFTA), efficiency (SATA), and

functioning (SA). Theoretical linkages are established on the basis of PLS-PM.

The objective of this chapter is to study the dynamics of capability dimensions.58

58 The latent variable model considered in this chapter is robust to happiness as latent variable scores, with and without happiness, are very similar. Thus happiness can safely be omitted in order to focus on interaction of capability dimensions. This assertion can be strengthened empirically by near one-to-one correspondence between functioning and happiness in simultaneous equation model in chapter 3.

81 | P a g e

5.2 Selection of the Model

The dynamic analysis in this chapter builds up on one of the theoretical

models that is extensively used in the literature (also estimated in chapters 3 and

4) to understand the interaction between different dimensions of capabilities. It

models happiness as a function of functioning, and functioning as a function of

freedom and conversion efficiency (model 3 of chapter 4). This is how advocates

of the CA, including Sen, characterized the CA (Sen, 1985b). Moreover, this is

the simplest model one can start with to study interactions between capabilities

and their impact on happiness. Having said that, it is also important to highlight

that, in the cotext of capabilities of “being achieved”, this specification can be

assumed a priori as one would expect happiness to depend on one’s sense of

acheivment (SA) which in turn is determined by other dimensions of capabilities

and controls.

From the point of view of the theoritical analysis in this chapter, it is also

important to note that the second model in chapter 4 reduces to model 3 as the

analysis in this model mainly relies, as will be clear, on interaction between F and

the other two dimensions of capabilities (i.e. E and R). This is because all these

factors have a positive impact on happiness through direct interaction and

happiness can as such be “ignored” in theortical discussion. Emperical analysis

however does include it in its estimation. More importantly, E and R are latent

variables which can be targeted through objective variables whereas functioning

(F) cannot be. This is something that this thesis looks into in the next chapter.

The idea is to study interaction in a way that helps us identify some policy

relevance in terms of objective targets. This is parsimoniously captured by the

empirically plausible model we estimated in the previous chapters.

The simple and partial correlation analyses (see Table A5.2 in the appendix

A5) shows that happiness (HAPP) and sense of achievement (SA) are highly

correlated in majority of cases whereas sense of freedom to achieve (SFTA) and

sense of ability to achieve (SATA) are weakly correlated in majority of cases as

compared to other correlations at all levels irrespective of gender and years of

education. The correlation results of this chapter reinforce the model developed

in the previous chapter. It is, therefore, correct to include SFTA and SATA

82 | P a g e

together as independent variables since they are weakly correlated in the

regression of SA on SFTA and SATA. The regression analysis in the last chapter

and the correlation analysis in the present chapter clearly show that HAPP and

SA are highly correlated. Given the strong correspondence between SA and

HAPP, the next section looks at the interaction between SA, SFTA, and SATA,

assuming that this interaction is also true for HAPP through feedback from SA.

Hence the dynamic analysis of capability dimensions is justified statistically as

well as theoretically.

5.3 Dynamic modeling using Bootstrapping

The following regression is run by drawing 1000 random samples with

replacement to obtain bootstrap estimates of a and b.

H = f(F)+ error

F=aR + bE + error (1)

The bootstrap estimates show a negative and highly significant relationship

between the coefficients (partial effects) of freedom (a) and efficiency (b):

a – bα β= (2)

The above relationship between partial effects also holds in case of all

districts as shown by the bootstrapping results for each district (see Appendix B).

This points towards the fact that these partical effects are sbstitutes which has

important policy implications. This relationship is used to understand theoretical

dynamics of the model with the purpose of deriving some policy lessons. It

identifies different policy regions (E, R, RE and ER representing policy focus on

E, R, primarily on R with increasing emphasis on E as F increases, and primarily

on E with increasing emphasis on R as F increases, respectively) under

alternative scenarios and applies it to the data.

The model is then applied to PSES data. Results show that most districts fit

the low-freedom scenario (as opposed to low efficiency scenario) and most of

them are located in the RE policy region.

83 | P a g e

5.4 Theoretical Dynamics of the Model

Substituting equation (2) in the deterministic form of equation (1) gives the

following general expressions for a and b in terms of the ratio of capability

dimensions:

a (3)

b

E FR RER

FRER

α β

β

α

β

− =

−

− = −

(4)

Dividing equation (3) by equation (4) gives the ratio of partial effects of R and E.

(5)

E Fa R R

FbR

α β

α

− =

−

The following three equations (6, 7, and 8) give the change in partial effects

due to change in E, R, and F respectively.

( )

( )( )

1

2 2

3

( / ) / (6)( )

( / ) ( ) (7)( )

( ) (( / )( )

a b RFE F RR

a b F ER F R

F Ra bF

α αψαα

α β αψα

α β αψ

∂= = =

∂ − −

∂ −= =

∂ −

− − −∂= =

∂ ( ) ( )2 2

) ( ) (8)E F R E

F R F R

β α βα α

− −=

− −

Assuming α , β >0, and (F/R-α )≠0,

1ψ could be >0 (when F/R>α ), and <0 (when F/R<α ).

2ψ could be >0 (when F/E>α /β), =0 (when F/E=α /β), and <0 (when

F/E<α /β).

3ψ could be >0 (when E/R<β), =0 (when E/R=β), and <0 (when E/R>β)

84 | P a g e

For the sake of theoretical analysis, it is assumed that F, R, and E are

continuous variables. This is a plausible assumption since the variables

underlying the categorical indicators of F, R, and E are continuous.

The above information is used to indentify different policy regions which are

summarized in the following graphs and and in two possible scenarios. This is

further elaborated on in the next section.

Graph 5.1: Policy emphasis regions for the low-efficiency scenario

E/R<β which implies ∂(a/b)/∂F>0

Graph 5.2: Policy emphasis regions for the low-freedom scenario

E/R>β which implies ∂(a/b)/∂F<0

a/b

F

αR -E/R

αEH/β

Region R Region RE Region E

a = 0 b = 0

ba

F

αR

αEL/β

Region E Region ER Region R

a = 0 b = 0

-E/R

85 | P a g e

Table 5.1: Scenario 1. If you target E (when E/R<β and R is fixed) F= αE/β would increase at a slower rate than E. As a result E/R<β

F Sign of a and b Appropriate Policy Target

Region E (Low F) F<αE/β a<0

b>0 E

Region ER (middle F) αR>F>αE/β a>0

b>0 E and R with increasing emphasis on

R as F increases

Region R (High F) F>αR a>0

b<0 R

Table 5.2: Scenario 2. If you target R (when E/R>β and E is fixed) F= αE/β would increase at a faster rate than E. As a result E/R>β

F Sign of a and b Appropriate Policy Target

Region R (Low F)

F<αR

a>0 b<0 R

Region RE (Middle F)

αR<F<αE/β

a>0 b>0

E and R with increasing emphasis on E as F increases

Region E (High F) F>αE/β a<0

b>0 E

The above analysis is applied to all the districts and policy emphasis region is

identified for each district.

5.5 Conclusion and Policy Implications

Our empirical results show that functioning has more or less 1-to-

1.correspondence with happiness. As a result , the impact of any policy on F is

transferred to happiness 1-to-1. The previous used bootstrapping to model

interaction between capabilities and found that the partical effects of freedom and

efficiency on functioning are substitutes. This relationship was used to identify

different policy regions with different policy emphasis. Policy emphasis was

shown to depend on a policy units’s (district here) level of efficiency relative to

freedom (i.e. E/R) and the level of achieved function (F). This is summarized as

below

86 | P a g e

i) A district with relatively lower F having E/R less (more) than a threshold

should target E(R) as targeting R(E) would decrease F further.

ii) A district with relatively better F, in region ER (RE), having an E/R less

(more) than the minimum threshold should target both E and R with

increasing emphasis on R(E). This is because the effectiveness of

targeting E(R) declines as F increases and that of R(E) increases as F

improves. This is similar to having decreasing returns to policy. As F

improves and crosses to region III, the policy emphasis should be

completely shifted to R(E) as the diminishing returns to targeting E(R) lead

to a negative effect on F.

In a nutshell, there are four policy target regions: E, R, ER, or RE. I repeated

the boostrapping exercise at district level data to calculate α and β which

togather with E and R sort these districts into different policy regions. Table A5.1

report these values and identify the policy region a distric falls in. Sixteen districts

are found to fall in policy region E, thirty-five districts in policy region RE, and six

districts in policy region R. There is no district in the last region RE (i.e., E and R

with increasing emphasis on R as F increases). Thus a majority of the districts

are classified as low-freedom.

The next chapter analyses the main determinants of R and E so that policy

targeting can be operationalised in practice using their determinants.

87 | P a g e

Appendix A5

Table A5.1: Policy targets at district level

Low-Efficiency districts E/R<β Policy

District F E R α β E/R αE/β αR Region MARDAN 0.82 0.67 1.13 0.81 0.91 0.60 0.60 0.92 E PESHAWAR 0.81 0.64 1.09 0.82 0.84 0.59 0.63 0.89 E DADU 0.95 0.73 0.96 0.80 0.86 0.76 0.67 0.77 R KOHAT 0.89 0.81 1.18 0.65 0.73 0.69 0.72 0.76 R BANNU 0.76 0.48 0.93 0.58 0.65 0.51 0.42 0.54 R KALAT 1.17 0.98 1.18 0.88 0.99 0.83 0.87 1.03 R

Low Freedom districts E/R>β Policy

District F E R α β E/R αE/β αR Region OKARA 0.68 0.68 0.78 0.73 0.77 0.87 0.64 0.57 E GUJRAT 1.08 1.07 0.97 0.76 0.90 1.10 0.91 0.74 E SIALKOT 1.15 0.96 0.91 0.77 0.82 1.05 0.90 0.70 E BAHAWALPUR 1.33 1.15 1.13 0.58 0.54 1.02 1.23 0.65 E BAHAWALNAGAR 1.45 1.11 0.83 0.38 0.37 1.34 1.12 0.31 E JACOBABAD 0.55 0.45 0.79 0.57 0.48 0.57 0.54 0.45 E SHIKARPUR 0.98 0.65 0.82 0.66 0.58 0.79 0.73 0.54 E SUKKUR 0.89 0.76 0.87 0.68 0.65 0.88 0.80 0.59 E LARKANA 0.98 0.82 1.11 0.61 0.63 0.73 0.79 0.68 E SANGHAR 0.60 0.54 0.66 0.54 0.49 0.82 0.60 0.36 E NAWAB SHAH 0.85 0.69 0.59 0.60 0.53 1.17 0.78 0.36 E D.I.KHAN 1.17 0.96 1.08 0.21 0.37 0.89 0.53 0.22 E QUETTA 1.23 1.39 1.38 0.53 0.60 1.00 1.22 0.73 E LORALAI 1.32 1.57 1.60 0.52 0.72 0.98 1.14 0.83 E RAWALPINDI 1.18 1.33 1.31 0.94 1.00 1.01 1.26 1.24 R KHAIR PUR 0.90 0.73 1.21 0.76 0.39 0.60 1.42 0.92 R ATTOCK 1.08 1.34 1.05 0.71 0.54 1.28 1.77 0.74 RE JHELUM 1.21 1.25 1.25 0.70 0.69 1.00 1.27 0.87 RE ISLAMABAD 1.26 1.20 1.46 0.69 0.37 0.82 2.22 1.00 RE SARGODHA 0.96 0.98 1.03 0.65 0.47 0.95 1.35 0.67 RE MIANWALI 1.10 1.22 1.20 0.67 0.55 1.01 1.49 0.80 RE KHUSHAB 1.04 1.03 0.98 0.45 0.44 1.05 1.07 0.44 RE BHAKKAR 1.27 1.25 0.95 0.44 0.27 1.32 2.03 0.41 RE LAHORE 1.03 1.28 1.12 0.59 0.54 1.15 1.40 0.66 RE KASUR 0.88 0.94 1.04 0.54 0.50 0.90 1.00 0.56 RE SHEIKHUPURA 0.81 0.91 1.05 0.47 0.33 0.87 1.26 0.49 RE GUJRANWALA 1.04 0.99 0.87 0.55 0.44 1.14 1.23 0.48 RE FAISAL ABAD 0.80 0.95 0.90 0.62 0.59 1.06 0.99 0.56 RE T.T. SINGH 0.83 0.91 0.91 0.58 0.62 1.00 0.85 0.53 RE JHANG 0.74 0.90 0.61 0.63 0.48 1.48 1.19 0.38 RE MULTAN 0.95 1.18 0.99 0.52 0.51 1.19 1.20 0.52 RE VEHARI 1.13 0.95 0.74 0.54 0.41 1.28 1.24 0.40 RE SAHIWAL 0.89 1.00 0.88 0.68 0.73 1.13 0.93 0.60 RE D.G.KHAN 1.22 1.38 1.20 0.75 0.67 1.15 1.55 0.90 RE

88 | P a g e

Table A5.1: Policy targets at district level (contd.)

Low Freedom districts E/R>β Policy

District F E R α β E/R αE/β αR Region LEIAH 1.04 1.34 1.02 0.54 0.40 1.31 1.80 0.55 RE MUZAFFARGARH 0.95 1.04 1.04 0.54 0.50 1.00 1.11 0.56 RE RAJANPUR 1.06 1.16 0.87 0.70 0.67 1.33 1.22 0.61 RE R.Y.KHAN 1.15 1.14 0.77 0.48 0.44 1.47 1.25 0.37 RE HYDERABAD 0.57 0.67 0.76 0.48 0.53 0.88 0.61 0.37 RE BADIN 0.79 0.83 1.03 0.63 0.41 0.80 1.27 0.65 RE THARPARKAR 1.03 0.79 1.15 0.58 0.41 0.68 1.12 0.67 RE THATTA 0.80 0.80 1.00 0.36 0.31 0.80 0.96 0.37 RE MIRPUR KHAS 1.03 0.66 1.29 0.67 0.41 0.51 1.09 0.86 RE KARACHI 1.34 1.29 0.98 0.53 0.47 1.31 1.44 0.52 RE DIR 0.73 0.57 0.79 0.65 0.49 0.72 0.75 0.51 RE SAWAT 0.67 0.97 0.83 0.55 0.40 1.16 1.32 0.46 RE MANSEHRA 0.99 1.04 1.02 0.46 0.14 1.02 3.37 0.47 RE ABBOTTABAD 0.91 0.99 1.12 0.48 0.35 0.88 1.38 0.54 RE KARAK 0.80 0.66 0.97 0.25 0.16 0.69 1.04 0.24 RE SIBI 1.06 1.26 1.40 0.42 0.46 0.90 1.15 0.59 RE MEKRAN 1.02 0.90 0.97 0.66 0.35 0.92 1.72 0.65 RE

Table A5.2: Correlation Analysis Correlation\partical correlation

between happiness and Conditional on

Functioning (unconditional) 0.75 Freedom 0.62 Efficiency 0.63 Freedom and Efficiency 0.55

Freedom (unconditional) 0.65 Functioning 0.43 Efficiency 0.53 Functioning and Efficiency 0.40

Efficiency (unconditional) 0.54 Functioning 0.20 Freedom 0.35 Functioing and Freedom 0.11

89 | P a g e

Chapter 6

Capability Dimensions and Their Correspondence with Income and Education

“Allah commands justice, the doing of good, and giving to kith and

kin, and He forbids all indecent deeds, and evil and rebellion: He

instructs you, that you may receive admonition.”

(Qur’an, 16:90)

6.1 Introduction

The theoretical analysis in chapter 5 shows the significance of targeting

freedom or efficiency or both for the enhancement of functioning and hence

happiness. Since freedom and efficiency are latent variables, we need to target

their main determinants. It is assumed and many studies show (for example,

Kuklys, 2005, and Anand et. al. 2005, among others) that income and education

play important role in enhancing freedom and efficiency.

To see the impact of education and income on R and E given conversion

factors59, the latent scores are converted into ordinal scale for better

interpretation of results in terms of probabilities and then ordered logit models are

applied which are estimated by the Maximum Likelihood method.60 The reason

for preferring logistic distribution to the normal distribution is that the distributions

of freedom and efficiency both have heavy tails relative to the normal distribution.

Though applying logit or probit does not have a substantial difference in terms of

probabilities, there are differences in estimates due to different distributional

assumptions.61

59 The conversion factors are age, gender, marital status, and region of living. 60 OLS is not applicable in this situation since OLS estimates are biased and inefficient when dependent variable is ordinal. 61 The logit coefficients are approximately 1.8 times greater than that of probit; since the standard deviation for probit model is 1 whereas it is 1.81 for logit model.

90 | P a g e

6.2 A Brief Review of Literature:

There are a number of theoretical discussions on the importance of education

for capability enhancement (See, for example, Sen, 1992, 1999; Nussbaum,

1997; Robeyns, 2006; Unterhalter, 2003 2005; Unterhalter et al. 2007; Walker,

2005, 2006; Saito, 2003; Watts and Bridges, 2006; Tao, 2010). These studies

theoretically show the intrinsic as well as instrumental value of education and

schooling for capability expansion.

Among the empirical studies, Kuklys (2005) shows that education and

income have little impact on the functionings of ‘being-healthy’ and ‘being-well-

sheltered’ in case of the UK.

Wigley and Akkoyunlu-Wigley (2006) shows that average years of schooling

(educational attainment) has a siginificant positive effect on health functioning

(life expectancey) independent of per capita income effect in the panel data of 35

developing countries.

Using the survey data in South Africa, Clark (2005) concludes that material

resources have positive impact on functioning but the extent of impact depends

on how efficiently resources are converted into a functioning.

In a survey article, Robeyns (2005) summarizes that education plays an

important role in enhancing a functioning and the associated capability. The

effect of income on a functioning, however, depends on which functioning or

functionings are analyzed. Most of the studies in this survey provide with a clear

evidence that income and education complement capability-based information.

In what follows, I will model freedom and efficiency using income and education

as main determinants with conversion factors as controls.

6.3 Ordered Choice Models

The ordered choice models for freedom and efficiency estimates are:

*

*

( | )

( | )i i i i

i i i i

R r X z

E e X z

ε

η

= +

= +

91 | P a g e

The specific forms of above models are given in section 6.4. The general

text-book form of these models, following Greene and Hensher (2008), is

given below:

,...1 ,* Nixy iii =+′= εβ (1)

where y* is a latent continuous variable observed as a discrete variable y

through a censoring mechanism like,

1*

2*

1

1*

if ...

, if 2

, if 1

−>=

=≤<=

≤=

Ji

i

ii

yJ

yyy

µ

µµ

µ

where μs are thresholds to be estimated but they have no intrinsic value.

They are estimated for the purpose of coding and assumed to be same for

all individuals. It is assumed that the model follows the standard

assumptions with appropriate dimensions of matrices. ε follows a standard

logistic distribution with mean 0 and variance equal to π2/3. The ordered

choice models are better represented in terms of probabilities because of

the limited nature of the dependent variable.

Four ordered logit models are compared for robustness of results. These four

models are distinguished on the basis of the following two important

assumptions:62

i) Parallel lines assumption: Slopes coefficients are the same or regression

lines are parallel across response categories.

The general form of the first three models is:

62 The discussion in this section is based on Williams (2010).

1J ..., 2, , 1 j ,)][exp(1

)exp()( −=

++

+=>

jij

jiji X

XjyP

βαβα

92 | P a g e

These models are distinguished by the way they violate the parallel lines

assumption.

a) Totally constrained generalised (TCG) model: A special case of the

generalised model, the ordered logit, is obtained if βj = β;63 all the slope

coefficients are constrained to be parallel.

b) Totally unconstrained generalised (TUG) model: When there is no

constraint on slope coefficients.

c) Partially constrained generalised (PCG) model: When some slope

coefficients are constrained to be parallel but some are free from this

constraint.

ii) Errors are homosecdastic64. All of the above models assume

homoscedasticity of errors. Since these models are non-linear, variance of

residuals also affects partial coefficients in the model and, in the present

situation, heteroscedasticity is one of the reasons for violation of parallel

lines assumption. If we correctly model heteroscedasticity then there is a

chance that violation of parallel lines assumption can be avoided. The

following model, called Heterogeneous choice (HC) model, attempts to

model heteroscedasticity;

where iσ is the variance adjustment factor for the ith individual. As we

can see, when iσ = 1, model 4 becomes model 1. The original model,

therefore, is:

* , 1... ,i i iy x i Nα σε′= + = (2)

63 The simplest case, logit model, is obtained when J = 2. 64 The assumption of homoscedasticity of errors also needs to be tested since most often data are cross-sectional where this assumption may not be valid.

exp[( ) / ]( ) , j 1 , 2, ..., J 1

1 [exp[( ) / ]]j i i

ij i i

XP y j

Xα β σ

α β σ+

> = = −+ +

93 | P a g e

Since yi* is unobserved, its scale is fixed by scaling the coefficients so

that the variance of the residuals is π2/3 (in case of logit link function) or 1

(in case of probit link function).65 The α’s in equation (2) and β’s in

equation (1) are related in the following way:

/ j 1,...,j j Jβ α σ= =

If σ is same for all cases then the variances of residuals as well as βs

are also same across cases.

To model heteroscedasticity, we make use of the above relationship

between the homoscedasticity and parallel lines assumptions. We first test

for violation of parallel lines assumption by the Brant test (after the totally

constrained ordered logit model). The slope coefficients violating this

assumption are then included in the variance equation of the

heterogeneous choice model.

6.4 Description of Variables

Each of the estimated ordered choice models consists of seventeen variables.

Their precise definitions are given below:

Dependent Variable for model 1 (Table A6.1):

R: an ordinal variable which takes discrete values from 1 to 3.

Dependent Variable for model 2 (Table A6.2):

E: an ordinal variable which takes discrete values from 1 to 3.

Independent Variables:

EDUCATION is a variable which is set equal to 1 for primary education, 2

for secondary education, 3 for higher secondary education, 4 for post

65 We can either set first threshold (cut-point), μ0 , equal to zero or the intercept term equal to zero for identification purposes. They are opposite in signs.

94 | P a g e

secondary (intermediate) education, 5 for under-graduate education, and 6 for

post-graduate education.

The variable EDUCATION is classified into the following five dummy

variables to analyse their impact separately using primary education as a

reference (base) category:

EDU2 = 1 for secondary education and

= 0 otherwise.

EDU3 = 1 for higher-secondary education and

= 0 otherwise.

EDU4 = 1 for intermediate education and

= 0 otherwise.

EDU5 = 1 for under-graduate education and

= 0 otherwise.

EDU6 = 1 for post-graduate education and

= 0 otherwise.

INCOME is a continuous variable which is converted into three ordinal

categories in order to capture distributional effect of income.66 Those at the

lowest 20% income are set equal to 1, those at the middle 60% income are

set equal to 2, and those at the highest 20% income are set equal to 3.

The variable INCOME is classified into the following two dummy variables

to analyse their impact separately setting lowest income class as a reference

category:

66 Generally, ordinals are converted into cardinals to improve estimation efficiency but here I convert cardinals into ordinals for the sake of distributional concerns. I preferr economic criteria to statistical criteria with some cost in terms of efficiency loss.

95 | P a g e

MIDDLE = 1 for middle income class, and

= 0 otherwise.

RICHEST = 1 for upper income class, and

= 0 otherwise.

Similalry, AGE is a continuous variable which is converted into three

ordinal categories in order to analyse distributional impact of age. Those

having age equal to or above 18 but below 30 years are set equal to 1 (young

age), those between 30 and 60 years are set equal to 2 (middle age), and

those above 60 years are set equal to 3 (old age).

The variable AGE is classified into the following two dummy variables to

analyse their impact separately setting young age (AGEY) as a reference

category:

AGEM = 1 for medium age, and

= 0 otherwise.

AGEO = 1 for old age, and

= 0 otherwise.

MARITAL_STATUS is a categorical variable which is set equal to 0 for

marital status not married (msn), 1 for marital status married (msm), 2 for

marital status separated (mss). It is classified into the following two dummy

variables to analyse their impact separately setting not married (msn) as the

reference category:

MSM = 1 for marital status married, and

= 0 otherwise.

MSS = 1 for marital status separated, and

= 0 otherwise.

96 | P a g e

Similarly, PROVINCE is a categorical variable which is set equal to 1 for

the province of Punjab, 2 for the province of Sind, 3 for the province of NWFP,

and 4 for the province of Baluchistan. The variable is classified into the

following three categories to analyse their impact separately setting Punjab as

a reference category:

PROV2 = 1 for the province of Sind, and

= 0 otherwise.

PROV3 = 1 for the province of NWFP, and

= 0 otherwise.

PROV4 = 1 for the province of Baluchistan, and

= 0 otherwise.

Finaly, UR, GEND and LIT are binary variables. UR takes a value of 1 for

urban regions and 0 otherwise, GEND takes a value of 1 for males and 0

otherwise, and LIT takes a value of 1 for literates and 0 otherwise.

6.5 Model Estimation and Results

The following specifications for ordered choice model *i i iy xβ ε′= + given in

equation (1) are considered for estimation:

Model 1:

*1 2 3 4 5

6 7 8

9 10 11 12

13 14 15 16 17

2 3 4 5 6 2 3 4

R EDU EDU EDU EDU EDU MIDDLE RICHEST GEND

AGEM AGEO MSM MSSPROV PROV PROV UR LIT

α α α α αα α αα α α αα α α α α ε

= + + + ++ + + + + + ++ + + + + +

97 | P a g e

Model 2:

*1 2 3 4 5

6 7 8

9 10 11 12

13 14 15 16 17

2 3 4 5 6 2 3 4

E EDU EDU EDU EDU EDU MIDDLE RICHEST GEND

AGEM AGEO MSM MSSPROV PROV PROV UR LIT

β β β β ββ β ββ β β ββ β β β β ε

= + + + ++ + + + + + ++ + + + + +

Discrete changes (marginal effects/partial effects) are computed for the four

models and reported in Table A6.1 and Table A6.2.

Long (1997, p. 75) describes discrete change as:“The discrete change in the

probability for a change of δ in kx equals

Pr( 1| ) Pr( 1| , ) Pr( 1| , )k kk

y x y x x y x xx

δ∆ == = + − =

∆

The discrete change can be interpreted as:

For a change in the variable kx from kx δ+ , the predicted probability of

an event changes by Pr( 1| )k

y xx

∆ =∆ , holding all other variables constant.”

The consistency of signs and sizes of coefficients for all the variables in the

four models above are compared. It turns out that the least consistent model is

the totally unconstrained model and then the totally constrained model. The most

consistent model is the partially constrained model and then the heterogeneous

choice model.

To figure out the differences between the partial effects of all variables on

efficiency and freedom, the results of most consistent model – the partially

constrained model – are compared (see Table A6.3).


The analysis (see Tables A6.1-A6.3 for estimation results and Table A6.4 for

post-estimation and diagnostic results) shows that income (education) has a

larger effect at higher (medium) levels of freedom and efficiency. Both have the

same effect at lower levels of freedom and efficiency. Hence, at lower levels, it

98 | P a g e

does not matter which policy targeting is employed. Since most respondents are

at medium levels of both freedom and efficiency, the role of education is

particularly valuable.

Graph A6.1 shows the behaviour of predicted probabilities for extreme

categories against income and education. The first part of the graph may indicate

an inverse correspondence between higher income and probability of reporting

lower freedom, and a positive correspondence between higher income and

probability of reporting higher freedom. The last part of the graph may show an

inverse correspondence between higher education and probability of reporting

lower freedom, and a positive correspondence between higher education and

probability of reporting higher freedom. The difference between the upper part

(income correspondence) and the lower part (education correspondence) of the

graph is that there is sharpness in the fall or rise in the upper part whereas there

is slowness in the fall or rise in the lower part. This may be due to steady (fast)

effect of education (income) on probability of reporting lower or higher freedom.

“This could be supplemented with an EDA style analysis where we can look

for high contrasts – this is also the object of the logit modeling, but EDA provides

a more transparent methodology. This would involve subdividing the population

for each dependent variable (each of the four capabilities can be be considered

as a dependent variable) into High Medium and Low and looking for variables (or

combinations of variables) which are maximally different between the high and

low categories. This will generate hypotheses about which variables are of

importance for which capability that can then be compared with our logit results.

6.7 Conclusion and Policy Implications:

Education policy may play a pivotal role in enhancing capability dimensions

particularly at medium capability levels. Educating people not noly enhances

capability dimensions with happiness but also increases income-earning

opportunities. Government should ensure equal effective freedom for all to obtain

education. The public and private education sectors should increase awareness

about the importance of eduation so that everyone has reason to value

education. This is the duty of a government to eliminate all sorts of barriers to

99 | P a g e

getting education. A quality education leads to empowerment, confidence in

decision making, better conversion of resources into achievements, social skills

etc. One of the fruits of education is tolerance which leads to a better community.

100 | P a g e

Appendix A6

Table A6.1: Marginal Effects (Discrete Changes) for freedom: comparison of ordered logit models EQUATION VARIABLES TCG MODEL TUG MODEL PCG MODEL HC MODEL

1 EDU2 -0.024 -0.0314 -0.0219 -0.0251 EDU3 -0.0550*** -0.0670*** -0.0514*** -0.0550*** EDU4 -0.0971*** -0.0792*** -0.0910*** -0.0962*** EDU5 -0.113*** -0.125*** -0.104*** -0.111*** EDU6 -0.123*** -0.0645 -0.113*** -0.120*** MIDDLE -0.0314* -0.0276* -0.0308* -0.0279* RICHEST -0.106*** -0.111*** -0.105*** -0.103*** GEND 0.0351*** 0.0506*** 0.0528*** 0.0535*** AGEM 0.0573*** 0.0659*** 0.0588*** 0.0568*** AGEO 0.173*** 0.192*** 0.184*** 0.181*** MSM -0.0283 -0.0275 -0.0288 -0.0255 MSS 0.00223 0.00837 0.00187 0.0112 PROV2 0.0301** 0.0264* 0.0295** 0.0317** PROV3 -0.0154 -0.0375** -0.0383** -0.0364* PROV4 -0.0760*** -0.119*** -0.120*** -0.118*** UR -0.0420*** -0.0513*** -0.0512*** -0.0525*** LIT -0.0430** -0.0323* -0.0437*** -0.0423**

2 EDU2 0.0128 0.0321 0.0114 0.0133 EDU3 0.0258*** 0.0556** 0.0236*** 0.0257*** EDU4 0.0274*** 0.0143 0.0256*** 0.0273*** EDU5 0.0194 0.0658* 0.0200* 0.0201 EDU6 0.00946 -0.0549 0.0135 0.0121 MIDDLE 0.0183* 0.00893 0.0176* 0.0162* RICHEST 0.0502*** 0.0565** 0.0478*** 0.0489*** GEND -0.0203*** -0.0539*** -0.0588*** -0.0612*** AGEM -0.0317*** -0.0460*** -0.0316*** -0.0313*** AGEO -0.129*** -0.164*** -0.154*** -0.148*** MSM 0.0173 0.0155 0.0172 0.0154 MSS -0.0013 -0.0206 -0.00106 -0.00666 PROV2 -0.0183* -0.0112 -0.0175* -0.0192** PROV3 0.00852 0.0556*** 0.0573*** 0.0539** PROV4 0.0315*** 0.117*** 0.119*** 0.120*** UR 0.0243*** 0.0444*** 0.0447*** 0.0471*** LIT 0.0238*** -0.00234 0.0235*** 0.0234**

3 EDU2 0.0112 -0.000691 0.0104 0.0118 EDU3 0.0291** 0.0113 0.0278** 0.0293** EDU4 0.0697*** 0.0649** 0.0654*** 0.0689*** EDU5 0.0934*** 0.0597* 0.0837*** 0.0909*** EDU6 0.114*** 0.119** 0.0991** 0.108** MIDDLE 0.0130* 0.0186 0.0132* 0.0117* RICHEST 0.0562*** 0.0547*** 0.0576*** 0.0546*** GEND -0.0148*** 0.00321 0.00596 0.00761 AGEM -0.0256*** -0.0199* -0.0272*** -0.0255*** AGEO -0.0442*** -0.0280** -0.0303** -0.0332*** MSM 0.011 0.012 0.0115 0.0101

101 | P a g e

Table A6.1: Marginal Effects (Discrete Changes) for freedom: comparison of ordered logit models (contd.) EQUATION VARIABLES TCG MODEL TUG MODEL PCG MODEL HC MODEL

3 MSS -0.000934 0.0123 -0.000811 -0.00457 PROV2 -0.0118** -0.0153* -0.0120** -0.0125** PROV3 0.00687 -0.0182* -0.0191* -0.0176 PROV4 0.0445*** 0.00156 0.000632 -0.00179 UR 0.0178*** 0.00689 0.0065 0.00547 LIT 0.0191** 0.0346** 0.0202** 0.0190** Observations 6749 6749 6749 6749 Standard errors in parentheses *** p<0.001, ** p<0.01, * p<0.05

Table A6.2: Marginal Effects (Discrete Changes) for efficiency: comparison of ordered logit models EQUATION VARIABLES TCG MODEL TUG MODEL PCG MODEL HC MODEL

1 EDU2 -0.0228 -0.0451* -0.0227 -0.0219 EDU3 -0.0407* -0.0576** -0.0405* -0.0412* EDU4 -0.0646*** -0.0811** -0.0628** -0.0654*** EDU5 -0.0929*** -0.103*** -0.0876*** -0.0943*** EDU6 -0.124*** -0.127*** -0.118*** -0.123*** MIDDLE -0.0134 -0.024 -0.0246 -0.025 RICHEST -0.0987*** -0.114*** -0.117*** -0.120*** GEND -0.0488*** -0.0454*** -0.0492*** -0.0542*** AGEM 0.0358*** 0.0297* 0.0350*** 0.0343** AGEO 0.0805*** 0.0648** 0.0798*** 0.0783*** MSM -0.0234 -0.00693 -0.0232 -0.0231 MSS -0.0121 0.0121 -0.0117 -0.0134 PROV2 0.114*** 0.130*** 0.133*** 0.128*** PROV3 0.148*** 0.169*** 0.168*** 0.170*** PROV4 -0.00713 -0.0514** -0.0494** -0.0504** UR 0.0153 0.00164 0.00426 -0.000482 LIT -0.0554*** -0.0495** -0.0549*** -0.0552***

2 EDU2 0.00628 0.0453 0.00565 0.00606 EDU3 0.00938*** 0.039 0.00821*** 0.00943*** EDU4 0.00804* 0.0409 0.00614 0.00795* EDU5 -0.00228 0.0285 -0.00324 -0.00293 EDU6 -0.0313 -0.00663 -0.0316 -0.0296 MIDDLE 0.00447 0.034 0.0354* 0.0291 RICHEST 0.0203*** 0.0597** 0.0677*** 0.0626** GEND 0.0160*** 0.00775 0.0150*** 0.0177*** AGEM -0.0111*** 0.000231 -0.0100*** -0.0106*** AGEO -0.0378** -0.011 -0.0360** -0.0364** MSM 0.0086 -0.0206 0.00805 0.00848 MSS 0.00362 -0.0488 0.00323 0.00396 PROV2 -0.0522*** -0.0850*** -0.0886*** -0.0789*** PROV3 -0.0785*** -0.120*** -0.118*** -0.129*** PROV4 0.00224 0.0774*** 0.0733*** 0.0872*** UR -0.00505 0.0215 0.0164 0.0234 LIT 0.0160*** 0.00628 0.0145*** 0.0159***

102 | P a g e

Table A6.2: Marginal Effects (Discrete Changes) for efficiency: comparison of ordered logit models (contd.) EQUATION VARIABLES TCG MODEL TUG MODEL PCG MODEL HC MODEL

3 EDU2 0.0165 -0.000215 0.017 0.0158 EDU3 0.0313* 0.0186 0.0323* 0.0317* EDU4 0.0566* 0.0402 0.0566* 0.0574* EDU5 0.0952** 0.0747* 0.0909** 0.0972** EDU6 0.155*** 0.133** 0.150*** 0.152*** MIDDLE 0.00891 -0.00998 -0.0107 -0.00408 RICHEST 0.0784*** 0.0539** 0.0494** 0.0578*** GEND 0.0328*** 0.0377*** 0.0342*** 0.0364*** AGEM -0.0247*** -0.0300** -0.0250*** -0.0237** AGEO -0.0427*** -0.0538*** -0.0437*** -0.0419*** MSM 0.0148 0.0275* 0.0152 0.0147 MSS 0.00847 0.0368 0.0085 0.00944 PROV2 -0.0620*** -0.0454*** -0.0442*** -0.0488*** PROV3 -0.0692*** -0.0493*** -0.0496*** -0.0411*** PROV4 0.00489 -0.0259* -0.0238 -0.0368** UR -0.0103 -0.0231** -0.0206* -0.0230* LIT 0.0394*** 0.0433** 0.0404*** 0.0392*** Observations 6749 6749 6749 6749 Standard errors in parentheses *** p<0.001, ** p<0.01, * p<0.05

Table A6.3: Differences between the marginal effects on Efficiency (e) and Freedom (r)

Prob of reporting high e is Prob of reporting high r is higher in the separated than unmarried. lower in separated than unmarried. lower in the middle income class than in the lower income class.

higher in the middle income class than in the lower income class.

lower in the province of Baluchistan than Punjab.

higher in the province of Baluchistan than Punjab.

lower in urban area than in rural area. higher in urban area than in rural area.

Prob of reporting medium e is Prob of reporting medium r is higher in separated than unmarried. lower in separated than unmarried. higher in male than female. lower in male than female. lower for a post-graduate than a primary passed.

higher for a post-graduate than a primary passed.

lower for an under-graduate than a primary passed.

higher for an under-graduate than a primary passed.

lower in the province of NWFP than Punjab. higher in the province of NWFP than Punjab.

Prob of reporting low e is Prob of reporting low r is higher in the province of NWFP than Punjab.

lower in the province of NWFP than Punjab.

higher in urban region than in rural area. lower in urban area than in rural area. lower in male than female. higher in male than female. lower in separated than unmarried. higher in separated than unmarried.

103 | P a g e

Graph A6.1: Behaviour of predicted probabilities of extreme categories for freedom and efficiency against education and income

0.2

.4.6

Pr(

ordi

nal_

r==1

)

4 6 8 10 12 14lnincome

0.1

.2.3

.4P

r(or

dina

l_r=

=3)

4 6 8 10 12 14lnincome

0.2

.4.6

Pr(

ordi

nal_

r==1

)

0 5 10 15RECODE of eduyrs

0.1

.2.3

.4P

r(or

dina

l_r=

=3)


0.1

.2.3

.4.5

Pr(

ordi

nal_

e==1

)


0.2

.4.6

Pr(

ordi

nal_

e==3

)


0.1

.2.3

.4.5

Pr(

ordi

nal_

e==1

)

4 6 8 10 12 14lnincome

0.2

.4.6

Pr(

ordi

nal_

e==3

)

4 6 8 10 12 14lnincome

104 | P a g e

Table A6.4: Ordered logit models: estimation and post estimation results Brant Test of Parallel Regression Assumption Variable chi2 p>chi2 df All 46.58 0.000 17 EDU2 2.30 0.129 1 EDU3 1.65 0.199 1 EDU4 0.97 0.324 1 EDU5 0.48 0.487 1 EDU6 0.26 0.613 1 MIDDLE 3.50 0.061 1 RICHEST 5.07 0.024 1 GEND 0.16 0.691 1 AGEM 0.45 0.502 1 AGEO 0.75 0.385 1 MSM 2.04 0.153 1 MSS 1.91 0.167 1 PROV2 8.29 0.004 1 PROV3 6.74 0.009 1 PROV4 11.59 0.001 1 UR 3.71 0.054 1 LIT 0.17 0.681 1 A significant test statistic provides evidence that the parallel regression assumption has been violated. Heteroskedastic Ordered Logistic Regression Log likelihood = -5839.3047 Number of obs = 6749

LR chi2(23) = 622.97 Prob > chi2 = 0.0000 Pseudo R2 = 0.0506

ordinal_e Coef. Std. Err. z P>|z| [95% Conf. Interval] Choice

EDU2 .1152158 .0995571 1.16 0.247 -.0799125 .3103441 EDU3 .2230184 .0955305 2.33 0.020 .0357821 .4102546 EDU4 .3760108 .130934 2.87 0.004 .1193848 .6326367 EDU5 .5859458 .1583429 3.70 0.000 .2755995 .8962921 EDU6 .8371831 .1774952 4.72 0.000 .4892989 1.185067 MIDDLE .0631649 .0679963 0.93 0.353 -.0701055 .1964352 RICHEST .5457731 .0844917 6.46 0.000 .3801724 .7113737 GEND .276025 .0497549 5.55 0.000 .1785071 .3735428 AGEM -.1767239 .0561101 -3.15 0.002 -.2866977 -.0667501 AGEO -.362559 .0975322 -3.72 0.000 -.5537186 -.1713993 MSM .1148992 .0780635 1.47 0.141 -.0381024 .2679008 MSS .0696912 .1291134 0.54 0.589 -.1833665 .3227488 PROV2 -.5350432 .0646368 -8.28 0.000 -.661729 -.4083573 PROV3 -.6520132 .0866833 -7.52 0.000 -.8219092 -.4821171 PROV4 .0327854 .0678501 0.48 0.629 -.1001983 .1657692 UR -.069228 .0485 -1.43 0.153 -.1642861 .0258302 LIT .2880069 .0739176 3.90 0.000 .143131 .4328828

Variance MIDDLE -.0576003 .0439631 -1.31 0.190 -.1437663 .0285657 RICHEST -.0977793 .0484558 -2.02 0.044 -.1927509 -.0028077 PROV2 .0697518 .0365596 1.91 0.056 -.0019037 .1414073 PROV3 .1522705 .0473748 3.21 0.001 .0594177 .2451234 PROV4 -.1978944 .0464874 -4.26 0.000 -.2890081 -.1067808 UR -.0643511 .0301084 -2.14 0.033 -.1233626 -.0053397

/cut1 -.9400576 .1449483 -6.49 0.000 -1.224151 -.655964 /cut2 1.805146 .1785374 10.11 0.000 1.45522 2.155073

105 | P a g e

Table A6.4: Ordered logit models: estimation and post estimation results (contd.) Brant Test of Parallel Regression Assumption Variable chi2 p>chi2 df All 81.98 0.000 17 EDU2 0.88 0.348 1 EDU3 2.44 0.118 1 EDU4 0.11 0.736 1 EDU5 1.58 0.209 1 EDU6 3.67 0.056 1 MIDDLE 0.48 0.487 1 RICHEST 0.11 0.742 1 GEND 11.47 0.001 1 AGEM 1.51 0.219 1 AGEO 5.55 0.018 1 MSM 0.03 0.869 1 MSS 0.42 0.516 1 PROV2 0.37 0.541 1 PROV3 10.13 0.001 1 PROV4 24.37 0.000 1 UR 4.04 0.045 1 LIT 2.64 0.104 1 A significant test statistic provides evidence that the parallel regression assumption has been violated. Heteroskedastic Ordered Logistic Regression Log likelihood = -5038.3231 Number of obs = 6749


ordinal_r Coef. Std. Err. z P>|z| [95% Conf. Interval] Choice

EDU2 .1538384 .1205402 1.28 0.202 -.082416 .3900928 EDU3 .3565238 .1165392 3.06 0.002 .1281112 .5849364 EDU4 .7096086 .1628199 4.36 0.000 .3904875 1.02873 EDU5 .8695747 .19092 4.55 0.000 .4953784 1.243771 EDU6 .94329 .243064 3.88 0.000 .4668933 1.419687 MIDDLE .1633764 .0762269 2.14 0.032 .0139744 .3127784 RICHEST .6647335 .0942373 7.05 0.000 .4800318 .8494352 GEND -.1685582 .0554707 -3.04 0.002 -.2772788 -.0598377 AGEM -.3412338 .0675403 -5.05 0.000 -.4736104 -.2088572 AGEO -.8223599 .1212459 -6.78 0.000 -1.059997 -.5847224 MSM .1450608 .0932749 1.56 0.120 -.0377547 .3278764 MSS -.0576178 .1517658 -0.38 0.704 -.3550732 .2398377 PROV2 -.1825806 .0648708 -2.81 0.005 -.3097251 -.0554362 PROV3 .0503948 .0791881 0.64 0.525 -.104811 .2056006 PROV4 .4747063 .0903275 5.26 0.000 .2976677 .6517449 UR .2273596 .0590255 3.85 0.000 .1116717 .3430474 LIT .2546307 .0857258 2.97 0.003 .0866111 .4226503

Variance EDU6 .2348807 .1048693 2.24 0.025 .0293408 .4404207 GEND .1092384 .0281622 3.88 0.000 .0540414 .1644353 AGEO .0897603 .0601566 1.49 0.136 -.0281445 .207665 PROV3 -.1258841 .0426526 -2.95 0.003 -.2094816 -.0422865 PROV4 -.2222577 .0441466 -5.03 0.000 -.3087835 -.1357319 UR -.0546672 .0289263 -1.89 0.059 -.1113616 .0020273

/cut1 -.6574786 .1522282 -4.32 0.000 -.9558404 -.3591168 /cut2 3.035594 .22315 13.60 0.000 2.598228 3.47296

106 | P a g e

Table A6.4: Ordered logit models: estimation and post estimation results (contd.) Brant Test of Parallel Regression Assumption Variable chi2 p>chi2 df All 113.60 0.000 17 EDU2 0.05 0.815 1 EDU3 0.52 0.472 1 EDU4 0.43 0.514 1 EDU5 0.32 0.569 1 EDU6 0.76 0.383 1 MIDDLE 12.19 0.000 1 RICHEST 4.55 0.033 1 GEND 23.53 0.000 1 AGEM 0.06 0.806 1 AGEO 1.35 0.246 1 MSM 0.13 0.722 1 MSS 0.01 0.930 1 PROV2 0.93 0.334 1 PROV3 4.21 0.040 1 PROV4 42.25 0.000 1 UR 20.11 0.000 1 LIT 1.25 0.264 1 A significant test statistic provides evidence that the parallel regression assumption has been violated. Heteroskedastic Ordered Logistic Regression Log likelihood = -5544.3033 Number of obs = 6749


ordinal_f Coef. Std. Err. z P>|z| [95% Conf. Interval] Choice

EDU2 .111394 .0739999 1.51 0.132 -.033643 .2564311 EDU3 .2560761 .0741231 3.45 0.001 .1107975 .4013546 EDU4 .36284 .1038919 3.49 0.000 .1592156 .5664643 EDU5 .5512935 .1282865 4.30 0.000 .2998566 .8027304 EDU6 .8750519 .1468414 5.96 0.000 .587248 1.162856 MIDDLE .1543813 .0514791 3.00 0.003 .0534842 .2552785 RICHEST .5646526 .0721858 7.82 0.000 .4231711 .7061341 GEND -.3022565 .0409203 -7.39 0.000 -.3824588 -.2220541 AGEM -.1381221 .0424502 -3.25 0.001 -.2213229 -.0549213 AGEO -.0649684 .0691031 -0.94 0.347 -.200408 .0704712 MSM .1598344 .0589537 2.71 0.007 .0442872 .2753815 MSS -.0650207 .0957897 -0.68 0.497 -.2527651 .1227238 PROV2 -.1655213 .0426862 -3.88 0.000 -.2491846 -.081858 PROV3 -.2960365 .0520776 -5.68 0.000 -.3981066 -.1939663 PROV4 .0673878 .0503712 1.34 0.181 -.031338 .1661137 UR .0254493 .0363644 0.70 0.484 -.0458237 .0967223 LIT .2047445 .0536028 3.82 0.000 .099685 .309804

Lnsigma RICHEST -.0660859 .0533127 -1.24 0.215 -.1705768 .038405 MIDDLE -.1460222 .0500673 -2.92 0.004 -.2441524 -.047892 GEND -.1373073 .0301311 -4.56 0.000 -.1963632 -.0782514 PROV3 -.1425898 .0505224 -2.82 0.005 -.2416118 -.0435677 PROV4 -.3556016 .0486151 -7.31 0.000 -.4508854 -.2603179 UR -.1429956 .0317037 -4.51 0.000 -.2051337 -.0808574

/cut1 -.3672903 .100374 -3.66 0.000 -.5640197 -.170561 /cut2 1.850253 .1703033 10.86 0.000 1.516465 2.184041

107 | P a g e

Chapter 7

Testing the Existence of Hedonic Adaption to Income in PSES Panel

“Do not usurp one another’s property by unjust means, nor offer it to

the judges, so that you may devour, knowingly and unjustly, a

portion of the goods of others”

(Qur’an, 2:188)

7.1 Introduction

One of the explanations to the Easterlin Paradox (1974) is that people adapt

their happiness to income over time and this explanation is termed as hedonic

adaptation or hedonic treadmill in the literature (Brickman and Campbell, 1971).

For a brief literature review on hedonic adaptation, see for example, Bottan and

Truglia (2011) and Di Tella et al (2010). Diener et al (2006) propose five

revisions in the original treadmill model. These revisions are, to wit: non-neutral

set points67, individual set points, multiple set points, and individual differences in

adaptation. These revisions allow us to explain incomplete or non-adaptation in

the data. The following section presents a variety of formulations discussed in the

literature that allow for hedonic adaptation to income68.

7.2 Hedonic Adaptation Models: A Brief Literature Review

Many formulations are proposed in the literature to study hedonic adaptation.

Based on empirical findings, Layard (2006) proposes the following happiness

function with a negative effect of the lagged income to allow for adaptation:

1( , ) (1)it it itU U y y hβ −= −

67 The set point is a term in psychology for a genetically determined hedonic or happiness point to which a person converges after a positive or negative shock. 68 A distinction is made between hedonic and eudemonic approaches to happiness in psychology. The former relates to pleasure as a stand alone concept where as the latter not only considers happiness but also the sources and processes that lead to happiness.

108 | P a g e

where itU is the happiness of the ith individual at time t, ity is the real

household income of the ith individual at time t, 1ity − is one period lag of

real household income of the ith individual at time t-1, and h is number of

hours of work.

Another formulation considered in Layard (2006) is to allow for loss-

aversion69:

[(1 ) , ] (2)it it itU U y y hβ β= − + ∆

Based on Layard’s explanation, loss-aversion can be defined as:

0 0| | (3)it itit y it yU U∆ < ∆ >∆ >> ∆

That is, the effect of a unit change in income on happiness is greater when

income falls than when income rises. The asymmetry of happiness response to

changes in income is an important finding attributed to Kahneman and his

colleagues, and has many important policy implications.

Somewhat similar to model in (1), Clark et al (2006) considers the following

formulation with current real income and change in real income:70

1 2 1ln( ) ln( ) (4)it it it it itU y y y Zβ β γ−= + +

where Z indicates demographic variables.

Ferrer-i-Carbonell and Van Praag (2008) consider many modifications of the

following general specification:71

1 1 1( ) ( ) (5)it it it it it it itU U y y Z Z Zβ δ γ− − −− = − + + −

69 The asymmetry of income comparison by higher income group and lower income group is termed as loss aversion by Kahneman and Tversky (1979) in their prospect theory. 70 This is not the exact specification used in Clark et al (2006). I have modified it to suit for a two-period panel. 71 The original specification given in Ferrer-i-Carbonell and Van Praag (2008) is for more than two time periods.

109 | P a g e

To allow for loss-aversion, for instance, they consider the following

specification:

1 1 2 1( ) (6)it it it it it it itU U y y Z Z Zβ β δ γ+ −− −− = ∆ + ∆ + + −

where 0 0 for | and for | (7)it itit it y it it yy U y U+ −

∆ > ∆ <∆ ∆ ∆ ∆

Bottan and Truglia (2011) test whether happiness is autoregressive and use

models similar to the following formulation:

1 1 2 1 1 2 1ln ln (8)it it it it it itU U y y Z Zα β β γ γ− − −= + + + +

where a positive coefficient of lagged happiness variable would show that

happiness is inertial.

7.3 Model Estimation

The above models are estimated using two period panel data from the

Pakistan Socio-Economic Survey (PSES) phase 1 and phase 2. The life

happiness question is not available in phase 1. However, there is a question in

phase 2 asking about happiness relative to the past. That question is used to

make a surrogate for life happiness question in phase 1. The estimation is run

using the sample common in both phases. Since the happiness question has

only three categories, it is considered ordinal and an ordered probit panel model

is used for estimation. The unobservable individual traits are considered random

and assumed to be uncorrelated with included variables in the model. These

assumptions are plausible since there is very high heterogeneity in individuals’

responses, and hence a random effects model is preferred to the fixed effects

model. The theoretical formulation (Crouchley’s formulation) of ordered choice

models with random effects for panel data are given in Greene and Hensher

(2008).

*

*1

2

if

~ (0,1), stochastic error term.

~ (0, ), random effect term independent of for all t.

it it i it

it j it j

it

i it

y x u

y j yN

u N

β ε

µ µ

ε

σ ε

−

′= + +

= ≤ <

110 | P a g e

7.4 Results and Discussion72

Table 7.1 sumarises the results of the random effects ordered probit models

(detailed results are reported in subsequent tables). The coefficient on lag

income is negative but statistically insignificant in model 1(see Table A7.1).

Economic criterion suggests that there is an evidence of adaptation to income but

statistical criterion does not endorse that conclusion. Hence current happiness

depends only on current income and is not affected by the previous level of

income.

The coefficient for first-order difference of nominal household income is

positive and significant in model 2 (Table A7.2). This indicates a positive effect of

income changes on happiness. It may indicate adaptation to income if we restrict

coefficients of current and lagged incomes to be the same.

Model 3 and model 4 show similar results to the above models but with real

income (Table A7.3 and Table A7.4).

The coefficients for current and first-order difference incomes are positive but

insignificant in model 5 (Table A7.5). Moreover, the log likelihood is flat at the

estimates.

The dependent variable is change in happiness in model 6 and the coefficient

of first-order difference real income is positive and significant but log likelihood is

flat at current estimates (Table A7.6). This shows that change in income has a

positive effect on change in happiness.

The coefficient for current income is positive and significant, the coefficient of

lagged income is negative but insignificant, and the coefficient of lagged

happiness is positive and significant but the log likelihood is flat at current

estimates in model 7 (Table A7.7). The positive coefficient of lagged happiness

would indicate inertia in happiness. Since the time periods are two years apart

and it might be the case that the gap is too long so that happiness dissipates over

72 All estimations are done by NLOGIT 4.0 (LIMDEP 9.0) econometric software developed by William Greene.

111 | P a g e

Table 7.1: Random Effects Ordered Probit Models Model Number Probability Index Function

(Correct prediction) Remarks

Model 1 39.96% No adaptation effect Constant -.24624842 GENDER -.05327971 AGE .00127252 EDU .00404028* UR .05011964 LNNY .05924241 LAGLNNY -.01630208

Model 2 39% No adaptation effect Constant .21565893 GENDER -.05533392 AGE .00128470 EDU .00474809 UR .06995613 DLNNY .03779831

Model 3 39.5% No adaptation effect Constant .00525406 GENDER -.05387937 AGE .00128347 EDU .00423288* UR .05478879* LNRY .05416031 LAGLNRY -.02147177

Model 4 39.3% No adaptation effect Constant .21566188 GENDER -.05533209 AGE .00128464 EDU .00474778 UR .06995819 DLNRY .03783129

Model 5 39.5% No adaptation effect Constant .00525420 GENDER -.05387935 AGE .00128347 EDU .00423288* UR .05478879* LNRY .03268851 DLNRY .02147178

Model 6 10% No adaptation effect DLNRY .06460951

Model 7 36.3% Inertia LNRY .05134365 LAGLNRY -.02478057 LAGHAPP .09698894

Model 8 35.6% Inertia DLNRY .04010989 LAGHAPP .19844387

Note: Coefficients highlighted in bold are insignificant, those marked by a * significant at 10%, and all other significant at 5%. All the relevant statistics are given in appendix A7.2- A7.8.

112 | P a g e

this interval to its previous level and hence shows inertia. If that is the case it

would indicate an adaptation effect. The other extreme case is also possible – the

gap is too short – and the happiness would take time to adjust to its previous

level after the passage of a long time, and hence would depict real inertia.

The coefficient of differenced real income is positive and significant, and the

coefficient of lagged happiness is positive and significant in model 8 (Table A7.8).

It has the same interpretation as model 7.

7.5 Conclusion and Policy Implications

A comparison of the eight models, estimated in the last section, is made on

the basis of percentage of correct predictions. All models show correct

predictions in the range of 35 % to 40% except model 6 with only 10% correct

predictions. These models may provide a weak evidence for hedonic adaptation

to income, although inconsistent with the findings on long panels like GSOEP and

BHPS, is yet consistent with most of the findings in the literature (see, Clark et al

2006 for a review of this evidence). Easterlin et. al (2010) attempts to resolve the

paradox. The study finds that happiness and income are directly related in the

short term but they are not related in the long term (for a period of more than 10

years).Since present study uses a very short panel, it confirms Easterlin et. al

(2010) findings. However, the findings in the present study should be taken with

caution since the panel is relatively short and the happiness in phase 1 is

measured with a surrogate. The evidence of inertia in some models remains

inconclusive unless supported by evidence from a longer panel.

In order to increase happiness during a short term (usually less than 10

years), policy makers should ensure increment in income after regular intervals

and take steps to minimize unemployment and to control inflation (unemployment

reduces nominal income whereas inflation diminishes real income) since both

have negative impact on happiness (Gandelman and Murillo, 2009; Di Tella and

MacCulloch, 2001, 2006, 2008; Frey and Stutzer, 2002; Di Tella, MacCulloch,

and Oswald, 2001, 2003; Wolfers, 2001; Oswald, 1997; Clark and Oswald, 1994).

113 | P a g e

Appendix A7

Table A7.1: Model 1 with current and lag nominal income Random Effects Ordered Probability Model Maximum Likelihood Estimates Model estimated: Aug 24, 2011 at 10:40:15PM. Dependent variable HAPP Weighting variable None Number of observations 8912 Iterations completed 26 Log likelihood function -9502.817 Number of parameters 9 Info. Criterion: AIC = 2.13461 Finite Sample: AIC = 2.13461 Info. Criterion: BIC = 2.14177 Info. Criterion:HQIC = 2.13705 Restricted log likelihood -9556.548 McFadden Pseudo R-squared .0056225 Chi squared 107.4632 Degrees of freedom 1 Prob[ChiSqd > value] = .0000000 Underlying probabilities based on Normal Unbalanced panel has 6684 individuals.

Variable Coefficient Standard Error b/St.Er. P[|Z|>z] Mean of X Index function for probability Constant -.24624842 .23672815 -1.040 .2982 GENDER -.05327971 .02663758 -2.000 .0455 .49865350 AGE .00127252 .00091692 1.388 .1652 39.5031418 EDU .00404028 .00237896 1.698 .0894 2.92706463 UR .05011964 .03100220 1.617 .1060 .36243268 LNNY .05924241 .02119863 2.795 .0052 8.06168833 LAGLNNY -.01630208 .02144922 -.760 .4472 7.94832831

Threshold parameters for index model Mu(01) 1.09741327 .02987742 36.731 .0000

Std. Deviation of random effect Sigma .50089211 .06523264 7.679 .0000

Summary of Marginal Effects for Ordered Probability Model (probit) Variable Y=00 Y=01 Y=02 Y=03 Y=04 Y=05 Y=06 Y=07 *GENDER .0184 -.0039 -.0145 AGE -.0004 .0001 .0003 EDU -.0014 .0003 .0011 *UR -.0173 .0036 .0137 LNNY -.0205 .0044 .0161 LAGLNNY .0056 -.0012 -.0044

Cross tabulation of predictions. Row is actual, column is predicted. Model = Probit. Prediction is number of the most probable cell. Actual Row Sum 0 1 2 3 4 5 6 7 8 9

0 3553 1466 2087 0 1 3258 1179 2079 0 2 2060 773 1287 0

Col Sum 8871 3418 5453 0 0 0 0 0 0 0 0 Correction prediction = 39.96%

114 | P a g e

Table A7.2: Model 2 with differenced nominal income Random Effects Ordered Probability Model Maximum Likelihood Estimates Model estimated: Aug 24, 2011 at 10:46:36PM. Dependent variable HAPP Weighting variable None Number of observations 8912 Iterations completed 21 Log likelihood function -9504.805 Number of parameters 8 Info. Criterion: AIC = 2.13483 Finite Sample: AIC = 2.13483 Info. Criterion: BIC = 2.14120 Info. Criterion:HQIC = 2.13700 Restricted log likelihood -9556.702 McFadden Pseudo R-squared .0054305 Chi squared 103.7949 Degrees of freedom 1 Prob[ChiSqd > value] = .0000000 Underlying probabilities based on Normal Unbalanced panel has 6684 individuals.

Variable Coefficient Standard Error b/St.Er. P[|Z|>z] Mean of X Index function for probability Constant .21565893 .04247525 5.077 .0000 GENDER -.05533392 .02686137 -2.060 .0394 .49865350 AGE .00128470 .00092726 1.385 .1659 39.5031418 EDU .00474809 .00236440 2.008 .0446 2.92706463 UR .06995613 .02989275 2.340 .0193 .36243268 DLNNY .03779831 .01852213 2.041 .0413 -4.59699954 Threshold parameters for index model Mu(01) 1.11037966 .03005751 36.942 .0000


Summary of Marginal Effects for Ordered Probability Model (probit) Variable Y=00 Y=01 Y=02 Y=03 Y=04 Y=05 Y=06 Y=07 *GENDER .0189 -.0040 -.0149 AGE -.0004 .0001 .0003 EDU -.0016 .0003 .0013 *UR -.0238 .0049 .0189 DLNNY -.0129 .0028 .0102


0 3553 1059 2494 0 1 3258 829 2429 0 2 2060 549 1511 0

Col Sum 8871 2437 6434 0 0 0 0 0 0 0 0 Correction prediction = 39%

115 | P a g e

Table A7.3: Model 3 with current and lag real income(Layard, 2006)

Random Effects Ordered Probability Model Maximum Likelihood Estimates Model estimated: Aug 24, 2011 at 10:50:14PM. Dependent variable HAPP Weighting variable None Number of observations 8912 Iterations completed 25 Log likelihood function -9503.642 Number of parameters 9 Info. Criterion: AIC = 2.13479 Finite Sample: AIC = 2.13480 Info. Criterion: BIC = 2.14196 Info. Criterion:HQIC = 2.13723 Restricted log likelihood -9556.551 McFadden Pseudo R-squared .0055364 Chi squared 105.8187 Degrees of freedom 1 Prob[ChiSqd > value] = .0000000 Underlying probabilities based on Normal Unbalanced panel has 6684 individuals.

Variable Coefficient Standard Error b/St.Er. P[|Z|>z] Mean of X Index function for probability Constant .00525406 .14570139 .036 .9712 GENDER -.05387937 .02669553 -2.018 .0436 .49865350 AGE .00128347 .00091965 1.396 .1628 39.5031418 EDU .00423288 .00237834 1.780 .0751 2.92706463 UR .05478879 .03111123 1.761 .0782 .36243268 LNRY .05416031 .02126830 2.547 .0109 3.74536596 LAGLNRY -.02147177 .02150644 -.998 .3181 3.63249806 Threshold parameters for index model Mu(01) 1.10084501 .02992165 36.791 .0000


Summary of Marginal Effects for Ordered Probability Model (probit) Variable Y=00 Y=01 Y=02 Y=03 Y=04 Y=05 Y=06 Y=07 *GENDER .0186 -.0040 -.0146 AGE -.0004 .0001 .0003 EDU -.0015 .0003 .0011 *UR -.0189 .0039 .0149 LNRY -0.187 .0040 .0147 LAGLNRY -.0074 -.0016 -.0058


0 3553 1369 2184 0 1 3258 1119 2139 0 2 2060 711 1349 0


116 | P a g e

Table A7.4: Model 4 with first order difference of real income Random Effects Ordered Probability Model Maximum Likelihood Estimates Model estimated: Aug 24, 2011 at 10:54:32PM. Dependent variable HAPP Weighting variable None Number of observations 8912 Iterations completed 21 Log likelihood function -9504.801 Number of parameters 8 Info. Criterion: AIC = 2.13483 Finite Sample: AIC = 2.13483 Info. Criterion: BIC = 2.14120 Info. Criterion:HQIC = 2.13700 Restricted log likelihood -9556.702 McFadden Pseudo R-squared . .0054309 Chi squared 103.8027 Degrees of freedom 1 Prob[ChiSqd > value] = .0000000 Underlying probabilities based on Normal Unbalanced panel has 6684 individuals.

Variable Coefficient Standard Error b/St.Er. P[|Z|>z] Mean of X Index function for probability Constant .21566188 .04247544 5.077 .9712 GENDER -.05533209 .02686147 -2.060 .0394 .49865350 AGE .00128464 .00092727 1.385 .1659 39.5031418 EDU .00474778 .00236443 2.008 .0446 2.92706463 UR .06995819 .02989296 2.340 .0193 .36243268 DLNRY .03783129 .01852156 2.043 .0411 -4.59701241

Threshold parameters for index model Mu(01) 1.11038526 .03005771 36.942 .0000


Summary of Marginal Effects for Ordered Probability Model (probit) Variable Y=00 Y=01 Y=02 Y=03 Y=04 Y=05 Y=06 Y=07 *GENDER .0189 -.0040 -.0149 AGE -.0004 .0001 .0003 EDU -.0016 .0003 .0013 *UR -.0238 .0049 .0189 DLNRY -.0129 .0028 .0102


0 3553 1059 2494 0 1 3258 829 2429 0 2 2060 549 1511 0


117 | P a g e

Table A7.5: Model 5 with current and differenced real income (Clark et al, 2006) Random Effects Ordered Probability Model Maximum Likelihood Estimates Model estimated: Aug 25, 2011 at 10:57:34AM. Dependent variable HAPP Weighting variable None Number of observations 8912 Iterations completed 22 Log likelihood function -9503.642 Number of parameters 9 Info. Criterion: AIC = 2.13479 Finite Sample: AIC = 2.13480 Info. Criterion: BIC = 2.14196 Info. Criterion:HQIC = 2.13723 Restricted log likelihood -9556.413 McFadden Pseudo R-squared .0055221 Chi squared 105.5421 Degrees of freedom 1 Prob[ChiSqd > value] = .0000000 Underlying probabilities based on Normal Unbalanced panel has 6684 individuals.

Variable Coefficient Standard Error b/St.Er. P[|Z|>z] Mean of X Index function for probability Constant .00525420 .14570140 .036 .9712 GENDER -.05387935 .02669553 -2.018 .0436 .49865350 AGE .00128347 .00091965 1.396 .1628 39.5031418 EDU .00423288 .00237834 1.780 .0751 2.92706463 UR .05478879 .03111123 1.761 .0782 .36243268 LNRY .03268851 .02192674 1.491 .1360 3.74536596 DLNRY .02147178 .02150644 -.998 .3181 -4.59701241 Threshold parameters for index model Mu(01) 1.10084503 .02992165 36.791 .0000


Summary of Marginal Effects for Ordered Probability Model (probit) Variable Y=00 Y=01 Y=02 Y=03 Y=04 Y=05 Y=06 Y=07 *GENDER .0186 -.0040 -.0146 AGE -.0004 .0001 .0003 EDU -.0015 .0003 .0011 *UR -.0189 .0039 .0149 LNRY -0.113 .0024 .0089 DLNRY -.0074 -.0016 -.0058


0 3553 1369 2184 0 1 3258 1119 2139 0 2 2060 711 1349 0


118 | P a g e

Table A7.6: Model 6 with differenced happiness and differenced real income (Ferrer-i-Carbonell and Van Praag, 2008)

Random Effects Ordered Probability Model Maximum Likelihood Estimates Model estimated: Aug 25, 2011 at 00:24:27PM. Dependent variable RECDHAPP Weighting variable None Number of observations 8911 Iterations completed 16 Log likelihood function -15475.60 Number of parameters 5 Info. Criterion: AIC = 3.47449 Finite Sample: AIC = 3.47449 Info. Criterion: BIC = 3.47847 Info. Criterion:HQIC = 3.47585 Restricted log likelihood -15798.65 McFadden Pseudo R-squared .0204483 Chi squared 646.1109 Degrees of freedom 1 Prob[ChiSqd > value] = .0000000 Underlying probabilities based on Normal Unbalanced panel has 6683 individuals.

Variable Coefficient Standard Error b/St.Er. P[|Z|>z] Mean of X Index function for probability DLNRY .06460951 .02070915 3.120 .0018 -4.48541966 Threshold parameters for index model Mu(01) .40177395 .01451199 27.686 .0000 Mu(02) 1.67017051 .03379934 49.414 .0000 Mu(03)

2.47588435 .04932703 50.193 .0000

Std. Deviation of random effect Sigma 1.25533547 .04608544 27.239 .0000

Summary of Marginal Effects for Ordered Probability Model (probit) Variable Y=00 Y=01 Y=02 Y=03 Y=04 Y=05 Y=06 Y=07 DLNRY -.0161 .0005 .0062 .0045

.0049


0 891 891 0 0 0 0 1 1446 1446 0 0 0 0 2 4198 4196 0 2 0 0 3 1450 1450 0 0 0 0 4 886 886 0 0 0 0

Col Sum 8871 8869 0 0 0 0 0 0 0 0 0 Correction prediction = 10%

119 | P a g e

Table A7.7: Model 7 with autoregressive happiness, current, and lag real income (Botton and Truglia, 2011)


Variable Coefficient Standard Error b/St.Er. P[|Z|>z] Mean of X Index function for probability LNRY .05134365 .01658286 3.096 .0020 3.74536596 LAGLNRY -.02478057 .01651411 -1.501 .1335 3.63249806 LAGHAPP .09698894 .01413740 6.860 .0000 .71947935

Threshold parameters for index model Mu(01) .98813603 .02676482 36.919 .0000

Std. Deviation of random effect Sigma .196318D-06 117518.292 .000 1.0000

Summary of Marginal Effects for Ordered Probability Model (probit) Variable Y=00 Y=01 Y=02 Y=03 Y=04 Y=05 Y=06 Y=07 LNRY -.0198 .0042 .0157 LAGLNRY .0096 -.0020 -.0076 LAGHAPP -.0375 .0079 .0296


0 3553 2657 896 0 1 3258 2693 565 0 2 2060 1367 693 0


120 | P a g e

Table A7.8: Model 8 with autoregressive happiness and differenced real income


Variable Coefficient Standard Error b/St.Er. P[|Z|>z] Mean of X Index function for probability DLNRY .04010989 .01654700 2.424 .0154 -4.59701241 LAGHAPP .19844387 .01057528 18.765 .0000 .71947935

Threshold parameters for index model Mu(01) .94529198 .02416635 39.116 .0000

Std. Deviation of random effect Sigma .840935D-08 .267270D+07 .000 1.0000

Summary of Marginal Effects for Ordered Probability Model (probit) Variable Y=00 Y=01 Y=02 Y=03 Y=04 Y=05 Y=06 Y=07 DLNRY -.0158 .0040 .0118 LAGHAPP -.0781 .0197 .0584


0 3553 2712 841 0 1 3258 2812 446 0 2 2060 1411 649 0


121 | P a g e

Chapter 8

Conclusions

“Successful indeed are the believers,- Those who humble

themselves in their prayers; Who avoid vain talk; Who are active in

giving zakat (charity); […]Those who faithfully observe their trusts

and their covenants; And who guard their prayers;- These will be

the heirs, Who will inherit Paradise: They will dwell therein (for

ever)”.

(Quran, 23:1-11)

“Measures of both objective and subjective wellbeing provide key

information about people’s quality of life. Statistical offices should

incorporate questions to capture people’s life evaluations, hedonic

experiences and priorities in their own survey”.

(Stiglitz et al, Report by the Commission on the

Measurement of Economic Performance and Social

Progress, 2009, p. 16)

8.1 The Perspective

Objective measures of wellbeing have been the focus of research in

economics for a long time and an objectivist approach to welfare seems to be

preferred over a subjectivist approach. This may be because the former is

generally considered to be scientific whereas the latter is usually viewed with

skepticism.73 Reservations about using a subjectivist approach have been diluted

due to some recent findings that show stability of subjective responses through

time (Krueger and Schkade, 2008) and consistency of objective and subjective

wellbeing measures (Oswald and Wu, 2010). It has also become more widely

73 Difficulties in objectivist and subjectivist views of wellbeing are discussed in Sen (1984). “Being ‘well off’ may help, other things given, to have ‘wellbeing’, but there is a distinctly personal quality in the latter absent in the former” (Sen, 1984, p. 195).

122 | P a g e

accepted that happiness research in economics provides new avenues for

effective policy making.

As seen from the quote at the beginning of this chapter, the human

development (HD) approach to measurement of human wellbeing has

demonstrated its value and has paved the way for specific recommendations for

integration of subjective wellbeing measures with standard economic measures.

The happiness approach and the capabilities approach challenge the

orthodox objectivist approach of wellbeing in economics by making a distinction

between means and ends for the evaluation of wellbeing. The present thesis has

been motivated by this debate on human wellbeing and attempts to integrate

happiness approach with the capabilities approach, as both claim to be intrinsic in

their space rather than instrumental, in order to assess human wellbeing using a

broader informational base. An integrated approach can take a step further than

taken by the HD approach, which is an important step in its own right, and

attempts to include people’s life evaluation, as recommended by the Stiglitz et al

commission, in the space of happiness and capabilities. The HD approach has

focused on material capabilities because the basic purpose of creating a HD

index is to highlight the status of misery, other than income poverty, in poor

countries, and “[t]he issue of capabilities – specifically ‘material’ capabilities – is

particularly important in judging the standard of living of people in poor countries”

(Sen, 1987b, p. 85).

This thesis attempts to model happiness using capability dimensions and

objective indicators of wellbeing with demographic controls. It is argued that this

approach is reasonable when the set of capabilities is endogenous to individuals.

This is consistent with Sen’s emphasis on positive freedom and human diversity,

the fundamental premises of the capabilities approach, and provides a way to

subjectively evaluate capability dimensions of an individual and to incorporate

positive freedom in the measurement of wellbeing. A possible objection to

measuring capability dimensions subjectively is that it may be subject to the

problem of adaptive preferences like the happiness indicator. We attempt to

counter this objection by invoking Sen’s (1993) approach of positional objectivity

– personal invariance and positional dependence – and by using districts

123 | P a g e

(positions), rather than individuals, in drawing policy implications. The task of

integrating both approaches involves measuring and modeling happiness and

capability dimensions. Measuring happiness presents fewer challenges now due

to the substantial work done in this area by researchers in psychology. However,

modeling happiness remains problematic as it is a very complex phenomenon

(Blanchflower and Oswald, 2011). Further, the measurement and modeling of

capability dimensions continues to pose challenges to researchers in this area

because of difficulties in the operationalisation of the capabilities approach.

In this thesis we argue that problems related to operatationalisation can

substantially be minimised if we consider capability dimensions of an overall

functioning of “being achieved”; Sense-of-achievement (SA as functioning),

sense-of-ability-to-achieve (SATA as conversion efficiency) and sense-of-

freedom-to-achieve (SFTA as freedom). The thesis uses a unique survey

(PSES) that collects information on indicators of these dimensions.

The thesis attempts to bridge the gap between Sen’s capabilities approach

and the happiness approach in assessing human wellbeing. Some of the

literature on well-being contrasted hedonic approach with capability approach

(Bruni, et. al., 2009). This literature indicates that capability provides opportunities

or chances whereas happiness is the result or outcome of these opportunities

and there is interdependence between the two (Veenhoven, 2010). However,

there are only a few attempts to empirically integrate these two notions of well-

being (for example, Anand et. al. 2011). These two approaches criticize

mainstream welfare economics independently and attempt to design policies to

improve well-being. These policies, however, may not be same or compatible

with each other. Sen at various places clearly indicate possible synergy between

the two; Sen (1985) shows happiness dependence on functioning whereas Sen

(1992) indicates being-happy as a functioning. In this way, Sen's capabilities

approach provides a theoretical framework for happiness approach, which

otherwise, depends exclusively on empirical and atheoretic methods.

The thesis measures three dimensions of capability – freedom, efficiency, and

functioning – simultaneously. This shows the contribution of each dimension in

124 | P a g e

capability measurement. To the best of our knowledge, this thesis is the first to do

so in this respect.

It decomposes subjective wellbeing (SWB) information into four components:

the three dimensions of capabilities – freedom to achieve, ability to achieve, and

functioning (achievement) – and the happiness indicator. Using the tools of

exploratory data analysis (EDA), it shows that each dimension of capability

contains information distinct from each other and the happiness indicator. The

disaggregation of SWB into four components guides us to the sources of SWB,

which in turn, can provide policy-relevant information to formulate priorities in

addressing capability dimensions and happiness issues. The decomposition of

SWB also provides a valid comparison between happiness and capability

dimensions.

The thesis uses an overall functioning of being-achieved to minimise the

problems of aggregation of functionings for an individual, like absence of a unit of

measurement or missing a natural aggregator, and the problem of selection of

functioning(s) from alternatives available to an individual as an exogenous list or

endogenously determined list.

It develops a theoretical model to analyse dynamics of capability dimensions

and derive policy focus regions in terms of freedom relative to efficiency or

efficiency relative to freedom.

It addresses the issue of adaptation of happiness to income for the first time in

the case of Pakistan by creating a two-period panel. It is also for the first time for

any developing country.

The thesis examines the probability distributions of SA, SATA, and SFTA for

possible differences by comparing these dimensions with each other and with the

happiness indicator using the tools of EDA (exploratory data analysis). The

differences are further investigated by ranking the districts according to the

scores of each capability dimension and the happiness indicator.

We empirically estimate and establish the relevance and importance of

capability dimensions for happiness, explore the robustness of this link to

125 | P a g e

alternative latent structure formulations, and develop a theoretical model under

alternative policy scenarios in terms of relative focus on freedom or efficiency.

We then discuss policy implications using education and income as policy

variables for freedom and efficiency. Finally, the thesis explores the issue of

hedonic adaptation to income using various formulations given in the literature.

8.2 Findings of the Thesis and Their Implications

In order to achieve the objectives stated in chapter 1, the thesis analyses the

issues by applying various exploratory and confirmatory techniques. The findings

and their implications are summarised below:

i) Using the tools of exploratory data analysis (EDA), the present work

shows that the capability dimensions provide information distinct from

each other and from the happiness indicator, providing a broader

informational base for the assessment of human wellbeing and for justice.

ii) When the differences in information obtained from each capability

dimension and happiness are analysed by ranking districts, the rankings

differ significantly. This illustrates the need to tailor policy to specific

district-level circumstances.

iii) Modeling happiness as a derived notion, with capability dimensions as

determinants shows that these dimensions are stable determinants of

happiness in terms of size, sign, and significance. This is an important

finding with implications for explaining a range of phenomena like reverse

migration, and altruistic and philanthropic behaviours.

iv) A stylised theoretical model of capabilities is developed that is

implemented with bootstrapping to explore alternative policy scenarios.

The results indicate when policy should be focused more on efficiency

relative to freedom or vice versa at the district level. Our application of the

theoretical models to the PSES data reveals that most of the policy units

are characterised as low freedom (relative to efficiency) which requires

focus on freedom with increasing emphasis on efficiency when

functionings improve.

126 | P a g e

v) Our results confirm previous findings that education and income are key

determinants of capability dimensions. This provides a means to address

efficiency and freedom objectives, usually not direct targets of policy, by

focusing policy on education and income targets. The partial effects of

education and income on freedom and efficiency show that education

plays a more important role than income for the majority, highlighting the

importance of provision of education in enhancing freedom and efficiency

for a majority of the population.

vi) Our estimation results from various happiness models show that evidence

in favour of happiness adaptation to income is quite weak, yet consistent

with bulk of the existing evidence (see, for example, Clark et al, 2006, for a

review of this evidence) that income raises happiness. The evidence

against income adaptation may be valid for countries with low per capita

income (Frey and Sustzer, 2002, p.428) or in a short-term (Easterlin et al.,

2010).

8.3 Limitations and Future Directions

There are several limitations of the present work that we note here together

with directions for future research. The general limitation of the study is that it

uses observational data and hence no causal inference can be drawn from the

analysis. Other important limitations and issues for future research include:

i) Due to nonavailability of long panel data on subjective questions from

Pakistan for our empirical study, the present work relies upon a very short

set of panel data. The capability dimensions are measured from the

subjective questions already given in the survey and provide an

approximation of the original concepts. There is a need to develop a

questionnaire specially designed for capability measurement including a

wider range of categories that can facilitate greater use of standard

econometric techniques. The existing limited category scales would give

an impression of stationary time series for happiness, and hence they

might lead to an incorrect long run relationship between happiness and

macroeconomic variables. The capability of being-achieved is measured at

127 | P a g e

a given information set of an individual. A randomised control experiment

to study the impact of information on capability enhancement can enable a

counterfactual analysis to be conducted to provide a better measurement

of a capabilities set. Happiness can be a derived notion from capability

dimensions, like overall happiness. But it can also be a functioning, like

being happy, or both. Our analysis is based on the first notion of

happiness assuming capability dimensions cause happiness, but there

might be an endogeneity problem due to the possibility of the other cases.

ii) The present analysis is limited to ranking of districts on the basis of

happiness and capability dimensions. Future work can compare the

ranking of districts based on objective measures of wellbeing like income

and education, and human development index.

iii) Education is considered a resource in the thesis, consistent with most of

the literature. It would be interesting to explore the implications of

considering education as a functioning of being-educated. A refined

functioning can be developed from the information given in many of the

existing large scale surveys about reasons for leaving education.

iv) The limitations of analysing the problem of adaptation to income by

empirical models available in the literature using a two-period panel can be

overcome by using long panel data, such the BHPS 18 waves panel data.

The problem can be more formally analysed in a theoretical dynamic

framework using difference or differential simultaneous-equation models.

A general limitation of the capabilities approach is that it does not directly take

into account the mental state in the evaluation of human wellbeing. We have

emphasised in this study that subjective and objective measures of wellbeing

should go hand in hand to provide a complete assessment of human wellbeing

and illustrated how such an approach can be empirically implemented.

While data constraints impose significant limitations on the scope of the study

and the empirical results, it is hoped that this study would stimulate and

128 | P a g e

encourage further research along these lines on this issue of major theoretical

interest and fundamental policy relevance.

129 | P a g e

Bibliography

Aaron B.T., Stephen G.W. and Leona S.A., 2006. Loss of Power in Logistic,

Ordinal Logistic, and Probit Regression When an Outcome Variable Is Coarsely

Categorised, Educational and Psychological Measurement, 66: 228.

Alkire, S., 2002. Dimensions of Human Development, World Development, 30(2):

181-205. Elsevier Science Ltd.

Alkire, S., 2005. Measuring the Freedom Aspects of Capabilities. Paper given to

the American Economic Association Conference, (Mimeo Harvard University).

Alkire, S., 2007. Measuring Agency: Issues and Possibilities. Indian Journal of

Human Development, 1(1): 169-175.

Alkire, S. and Santos, M.E., 2010. Acute Multidimensional Poverty: A New Index

for Developing countries. OPHI Working Paper 38, UNDP HDRO Background

Paper 2010/11.

Anand, P., 2011. Reviews. Capabilities and Happiness, edited by Luigino Bruni,

Flavio Comim and Maurizio Pugno. Oxford University Press, 2008. Economics

and Philosophy, 27: 175-215.

Anand, P. and Martin, van H., 2006. Capabilities and Achievement: An Empirical

Study. The Journal of Socio-Economics, 35: 268-285.

Anand, P., Hunter, G., Carter, I., Dowding, K., Guala, F. and Hees, M., 2005. The

Measurement of Human Capabilities. (Available at URL:

www.oecd.org/dataoecd/63/37/38363699.pdf).

Anand, P., Krishnakumar, J. and Tran, N.B., 2011. Measuring Welfare: Latent

Variable Models for Happiness and Capabilities in the Presence of Unobservable

Heterogeneity. Journal of Public Economics, 95: 205-215.

Anderson, G., Crawford, I. and Leicester, A., 2011. Welfare Rankings from

Multivariate Data, a Non-Parametric Approach. Journal of Public Economics, 95:

247-252.

http://www.oecd.org/dataoecd/63/37/38363699.pdf

130 | P a g e

Arif, G.M., Ali, M.S., Nasir, Z.M. and Arshad, N., 2001. An Introduction to the

1998-99 Pakistan Socio-Economic Survey (PSES). MIMAP Technical Paper

Series No. 4, Pakistan Institute of Development Economics, Islamabad, Pakistan.

Arrow, K., 1999. Amartya K. Sen’s Contributions to the Study of Social Welfare.

Scandinavian Journal of Economics, 101(2): 163-172.

Atkinson, A.B., 1999. The Contributions of Amartya Sen to Welfare Economics.

Scandinavian Journal of Economics, 101(2): 173-190.

Atkinson, A.C., 2009. Econometric Applications of the Forward Search in

Regression: Robustness, Diagnostics, and Graphics. Econometric Reviews,

28(1): 21-39.

Bentham, J., 1789. An Introduction to the Principle of Morals and Legislations.

Oxford, UK: Blackwell.

Bentler, P.M. and Huba, G.J., 1982. On The Usefulness of Latent Variable

Causal Modeling in Testing Theories of Naturally Occurring Events (Including

Adolescent Drug Use): A Rejoinder to Martin. Journal of Personality and Social

Psychology, Vol 43(3): 604-611.

Binder, M. and Broekel, T., 2011. Applying a Nonparametric Efficiency Analysis

to Measure Conversion Efficiency in Great Britain. Journal of Human

Development and Capabilities, 12(2): 257-281.

Binder, M. and Coad, A., 2011. Disentangling the Circularity in Sen’s Capability

Approach – An Analysis of the Co-evolution of Functioning Achievement and

Resources. Social Indicators Research, 103(3): 327-355.

Blanchflower, D.G. and Oswald, A.J., 2005. Policy Forum: Some Policy

Implications of Behavioural Economics, Happiness and the Human Development

Index: The Paradox of Australia. The Australian Economic Review 38(3): 307–18.

Blanchflower, D.G. and Oswald, A.J., 2011. International Happiness. NBER

Working Paper No: 16668.

131 | P a g e

Bollen, K.A., 2002. Latent Variables in Psychology and the Social Sciences.

Annual Review Psychology, 53: 605-34.

Bottan, N.L. and Truglia, R.P., 2011. Deconstructing the Hedonic Treadmill: Is

Happiness Autoregressive? Journal of Socio-Economics, 40: 224-236.

Brickman, P. and Campbell, D.T., 1971. Hedonic Relativism and Planning the

Good Society. In M. H. Appley (ed.), Adaptation Level Theory: A Symposium:

287-302. New York: Academic Press.

Bro, R. and Elden, L., 2009. PLS Works. Journal of Chemometrics, 23(2): 69-71.

Bruni, L., Comim, F. and Pugno, M. (eds.), 2008. Capabilities and Happiness.

Oxford. U.K: Oxford University Press.

Bruni, L. and Porta, P.L., 2007. Handbook on the Economics of Happiness.

Edward Elgar Publishing.

Burchardt, T., 2005. Are One Man's Rags Another Man's Riches? Identifying

Adaptive Expectations Using Panel Data. Social Indicators Research, 74: 57-102.

Burchardt, T., 2009. Agency Goals, Adaptation and Capability Sets. Journal of

Human Development and Capabilities, 10(1): 3-19.

Burchardt, T. and Vizard, P., 2011. Operationalising the capability approach as a

Basis for Equality and Human Rights Monitoring in Twenty-first-century Britain.

Journal of Human Development and Capabilities, 12(1): 91-119.

Bykvist, K., 2010. Can Unstable Preferences Provide a Stable Standard of

Wellbeing? Economics and Philosophy 26: 1-26.

Cassel, C., Hackl, P. and Westlund, A.H., 1999. Robustness of Partial Least-

Squares Method for Estimating Latent Variable Quality Structures. Journal of

Applied Statistics, 26(4): 435–446.

Clark, A.E., Frijters, P. and Shields, M.A., 2006. Income and Happiness:

Evidence, Explanations and Economic Implications. Working Paper No: 2006-24,

Paris-Jourdan Sciences Economiques.

http://www.google.com/search?tbo=p&tbm=bks&q=inauthor:%22Luigino+Bruni%22

http://www.google.com/search?tbo=p&tbm=bks&q=inauthor:%22Pier+Luigi+Porta%22

http://journals.cambridge.org/action/displayJournal?jid=EAP

132 | P a g e

Clark, A.E and Georgellis, Y., 2010. Back to Baseline in Britain: Adaptation in the

BHPS. Working Paper No: 2010-02, Paris-Jourdan Sciences Economiques.

Clark, Andrew and Oswald, Andrew J. (1994). Unhappiness and Unemployment. Economic Journal 104(5): 648–59.

Clark, D. A. 2005. Sen’s capability approach and the many spaces of human

well-being. Journal of Development Studies. 41(8), 1339-1368.

Colander, D., 2007. Edgeworth’s Hedonimeter and the Quest to Measure Utility.

Journal of Economic Perspective, 21(2): 215-225.

Comim, F., 2001. Operationalising Sen’s Capability Approach. Paper Prepared

for the Conference on Justice and Poverty: Examining Sen’s Capability

Approach. Cambridge University Press.

Comim, F., Qizilbash, M. and Alkire, S. (ed.), 2008. The Capability Approach-

Concepts, Measures, and Applications. Cambridge University Press.

Cora, J.M.M. and Joop J.H., 2005. Sufficient Sample Sizes for Multilevel

Modeling. Methodology, 1(3): 86-92.

Dasgupta, P., 1993. An Inquiry into Wellbeing and Destitution. Oxford: Clarendon

Press.

Davidson, R. and J. G. MacKinnon. 1993. Estimation and Inference in

Econometrics. New York: Oxford University Press

Diener, E., Lucas, R. and Scollon, C.N., 2006. Beyond the Hedonic Treadmill:

Revising the Adaptation Theory of Well-Being. American Psychologist, 61: 305-

314.

Di Tella, R., Haisken-De N.J. and MacCulloch, R., 2010. Happiness Adaptation to

Income and to Status in an Individual Panel. Journal of Economic Behaviour and

Organisation, 76: 834-852.

Di Tella, Rafael, and Robert J. MacCulloch. (2006). Some Uses of Happiness

Data inEconomics. Journal of Economic Perspectives 20(1): 25-46.

http://www.stata.com/bookstore/estimation-inference-econometrics/

http://www.stata.com/bookstore/estimation-inference-econometrics/

http://dx.doi.org/10.1007/978-90-481-2350-6_5

http://dx.doi.org/10.1007/978-90-481-2350-6_5

http://dx.doi.org/10.1007/978-90-481-2350-6_5

133 | P a g e

Di Tella, Rafael, and Robert J. MacCulloch. (2008). Gross national happiness as

an answer to the Easterlin Paradox? Journal of Development Economics 86: 22-

42.

Di Tella, Rafael, Robert J. MacCulloch and Andrew J. Oswald. (2001).

Preferences over Inflation and Unemployment: Evidence from Surveys of

Happiness. American Economic Review 91(1): 335-341.

Di Tella, Rafael, Robert J. MacCulloch and Andrew J. Oswald. (2003). The

Macroeconomics of Happiness. Review of Economics and Statistics

85(4): 809-827.

Dworkin, R., 2000. Sovereign Virtue. The Theory and Practice of Equality.

Cambridge: Harvard University Press.

Easterlin, R., 1974. Does Economic Growth Improve the Human Lot? In Paul A.

David and Melvin W. Reder, eds., Nations and Households in Economic Growth:

Essays in Honor of Moses Abramovitz. New York: Academic Press, Inc.

Easterlin, R., 2004. The Economics of Happiness. Daedalus, 133 (2).

Easterlin, R. and Angelescu, L., 2009. Happiness and Growth the World Over:

Time Series Evidence on The Happiness-Income Paradox. Discussion Paper

Series. IZA DP No. 4060. The Institute for the Study of Labor (IZA), Bonn,

Germany.

Easterlin Richard A., McVey, Laura Angelescu and Malgorzata Switek, et

al. (2010). The Happiness-Income Paradox Revisited. Proceedings of the

National Academy of Sciences, vol. 107 (52), 22463-22468

Edgeworth, F.Y., 1881. Mathematical Psychics: An Essay on the Application of

Mathematics to The Moral Sciences. London: C.K. Paul Publisher.

http://www.archive.org/search.php?query=publisher%3A%22London+C.K.+Paul%22

134 | P a g e

Ferrer-i-Carbonell, A. and Frijters, P., 2004. How Important is Methodology for

the Estimates of the Determinants of Happiness? Economic Journal, 114 (497):

641-659.

Ferrer-i Carbonell, A. and van Praag, B.M.S., 2008. Do People Adapt to Changes

in Income and Other Circumstances? The Discussion is not Finished Yet. Mimeo.

Festinger, L., 1954. A Theory of Social Comparison Processes. Human

Relations, 7(2): 117-140.

Fisher, I., 1892. Mathematical Investigations in the Theory of Value and Prices.

Cosimo Classics Publisher.

Freedman, D.A., 1987. As Others See Us: A Case Study in Path Analysis.

Journal of Educational Statistics, 12(2): 101-128.

Freedman, D.A., 2005. Statistical Models: Theory and Practice. Cambridge

University Press.

Frey, B.S. and Stutzer, A., 2002. What Can Economists Learn from Happiness

Research? Journal of Economic Literature, 40: 402-435.

Frisch, R., 1932. New Methods of Measuring Marginal Utility. Porcupine Pr.

Publisher.

Gandelman, N. and Murillo, R. H. The impact of Inflation and Unemployment on

Subjective Personal and Country Evaluations, Federal Reserve Bank of St. Louis

Review, May/June 2009, 91(3), pp. 107-26.

Gasper, D. and Troung, T., 2010. Movements of the ‘We’: International and

Transnational Migration and the Capabilities Approach. Journal of Human

Development and Capabilities, 11(2): 339-357.

Goulet, D., 2006. Development Ethics at Work: Explorations, 1960-2002.

Routledge, London.

135 | P a g e

Greene, W.H. and Hensher, D.A. 2008. Ordered Choices and Heterogeneity in

Attribute Processing. Working Paper, ITLS-WP-08-16.

Haenlein, M. and Kaplan, A.M., 2004. A Beginner’s Guide to Partial Least

Squares Analysis. Understanding Statistics, 3(4): 283-297.

Hammer, A., 2006. Enabling Successful Supply Chain Management –

Coordination, Collaboration, and Integration for Competitive Advantage. Doctoral

Thesis, University of Mannheim.

Haq, M., 1995. Reflections on Human Development. Oxford University Press.

Haq, R. and Zia, U., 2008. Dimensions of Wellbeing and the Millennuum

Development Goals. The Pakistan Development Review, 47(4): 851-876.

Haq, R. (2009). Measuring Human Wellbeing in Pakistan: Objective versus

Subjective Indicators. European Journal of Social Sciences, 9(3): 516-532.

Hartwig, F. and Dearing, B.E., 1979. Exploratory Data Analysis. SAGE.

Hausman, D.M. and McPherson, M.S., 2009. Preference Satisfaction and

Welfare Economics. Economics and Philosophy, 25: 1-25.

Huber, P., Ronchetti, E. and Victoria-Feser, M.P., 2004. Estimation of

Generalised Latent Trait Models. Journal of The Royal Statistical Society, Series

B 66: 893—908.

Human Development Report (HDR, 2010)

Jensen, U., 2000. Is it Efficient to Analyse Efficiency Rankings? Empirical

Economics, 25: 189-208.

Joreskog, K.G., 1973. Analysis of Covariance Structures. In Multivariate Analysis-

III, P.R. Krishnaiah (ed): 263–285. New York: Academic Press.

Joreskog, K.G., 2000. Latent Variable Scores and Their Uses. (Available at URL:

http://www.ssicentral.com/lisrel/techdocs/lvscores.pdf)

http://www.ssicentral.com/lisrel/techdocs/lvscores.pdf

136 | P a g e

Kahneman, D., Wakker, P. and Sarin, R., 1997. Back to Bentham? Explorations

of Experienced Utility. Quaterly Journal of Economics, 112: 375-405.

Kahneman, D. and Krueger. A.B., 2006. Developments in the Measurement of

Subjective Wellbeing. Journal of Economic Perspective 20(1): 3-24.

Kahneman, D. and Tversky, A., 1979. Prospect Theory: An Analysis of Decision

under Risk. Econometrica: 47, 263-291.

Kementa, J., 1991. Latent Variables in Econometrics. Statistica Neerlandica,

45(2): 73-84.

Kesebir, P. and Diener, E., 2008. In Defense of Happiness: Why Policymakers

Should Care about Subjective Wellbeing. In Capabilities and Happiness, Bruni, et

al (eds). Oxford University Press.

Krishnakumar, J. and Ballon, P., 2008. Estimating Basic Capabilities: A Structural

Equation Model Approach Applied to Bolivian Data. World Development, 36(6):

992-1010.

Krishnakumar, J., 2007. Going beyond Functionings to Capabilities: an

Econometric Model to Explain and Estimate Capabilities. Journal of Human

Development, 8(1): 39-63.

Kristoffersen, I., 2010. The Metrics of Subjective Wellbeing: Cardinality, Neutrality

and Additivity. Economic Record, 86(272): 98-123.

Krueger, A.B. and Schkade, D., 2008. The Reliability of Subjective Well-Being

Measures. Journal of Public Economics, 92: 1833-1845.

Krueger, A.B., 2009. Measuring the Subjective Wellbeing of Nations: National

Accounts of Time Use and Wellbeing. University of Chicago Press.

Kuklys, W., 2005. Amartya Sen’s Capability Approach – Theoretical Insights and

Empirical Applications. Springer, Berlin et al.

Kumbhakar, S.C. and Lovell, C.A. 2000. Stochastic Frontier Analysis. Cambridge

University Press.

137 | P a g e

Lauro, C. and Vinzi, E.V., 2004. Some Contributions to PLS Path Modeling and a

System for the European Customer Satisfaction. Dipartimento di Matematica e

Statistica Università “Federico II” di Napoli.

Layard, R., 2006. Happiness and Public Policy: A Challenge to the Profession.

The Economic Journal, 116 (March): C24-C33.

Long J.S. and Freese, J., 2006. Regression Models for Categorical Dependent

Variables Using Stata, Second edition. Texas: Stata Press Publication.

Long, J.S., 1997. Regression Models of Categorical and Limited Dependent

Variables. SAGE.

Nussbaum, M. 1997. Education and Democratic citizenship: Capabilities and

Quality Education. Journal of Human Development. 7(3), 385-396.

Nussbaum, M., 2000. Women and Human Development: the Capabilities

Approach. Cambridge University Press.

Nussbaum, M., 2005. Well-Being, Contracts and Capabilities. In Rethinking Well-

Being, L. Manderson (ed): 27-44. Perth: API Network.

Nguefack-Tsague, G., Klasen, S. and Zucchini, W., 2011. On Weighting the

Components of Human Development Index: A Statistical Justification. Journal of

Human Development and Capabilities, 12(2): 183-202.

NLOGIT 4.0., 2007. New York: Econometric Software, Inc.

Nozick, R., 1974. Anarchy, State, and Utopia. New York: Basic Books.

Olsson, U., 1979. Maximum Likelihood Estimation of the Polychoric Correlation

Coefficient. Psychometrika, 44 (4): 443-460.

Oswald, Andrew J. (1997). Happiness and Economic Performance. Economic

Journal107(5): 1815–31.

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.96.62&rep=rep1&type=pdf

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.96.62&rep=rep1&type=pdf

138 | P a g e

Oswald, A.J. and Wu, S., 2010. Objective Confirmation of Subjective Measures of

Human Wellbeing: Evidence from The USA. Science, 327: 576-579.

Pakistan National Human Development Report (2003).

Pirouz, D.M., 2006. An Overview of Partial Least Squares. (Available at SSRN:

http://ssrn.com/abstract-1631359).

Qizilbash, M., 2011. Sugden’s Critique of the Capability Approach. Utilitas, 23(1):

25-51.

Quran, with English translation by Abdullah Yusuf Ali. Kingdom of Saudi Arabia:

King Fahd Holy Quran Printing Complex.

Permanyer, I., 2011. Assessing the Robustness of Composite Indices Rankings.

Review of Income and Wealth, 57(1): 306-326.

Porta, P.L. and Bruni, L. (eds),2007. Handbook on the Economics of Happiness.

UK: Edward Elgar Publishing Limited.

Pressman, S. and Summerfield, G., 2000. The Economic Contributions of

Amartya Sen. Review of Political Economy, 12(1): 89-113.

Pressman, S. and Summerfield, G., 2002. Sen and Capabilities. Review of

Political Economy, 14(4): 429-434.

Rawls, J., 1971. A Theory of Justice. Cambridge, Massachusetts: Belknap Press.

Rawls, J. 2001. Justice as Fairness: A Restatement. Cambridge, Massachusetts:

Belknap Press.

Ramos, X. and Silber, J., 2005. On the Application of Efficiency Analysis to the

Study of the Dimensions of Human Development. Review of Income and Wealth,

51(2): 285-310.

Ringle, C.W. and Sven, A., 2005. SmartPLS 2.0 (http://www.smartpls.de).

SmartPLS, Hamburg, Germany.

http://ssrn.com/abstract-1631359

http://www.smartpls.de/

139 | P a g e

Robeyns, I., 2000. An Unworkable Idea or a Promising Alternative? Sen’s

Capability Approach Re-Examined. Discussion Paper Series (DPS) 00.30. Center

for Economic Studies, Katholieke Universiteit Leuven.

Robeyns, I., 2003. Sen’s Capability Approach and Gender Inequality: Selecting

Relevant Capabilities. Feminist Economics, 9(2-3): 61-92.

Robeyns, I., 2005. The Capability Approach: a Theoretical Survey. Journal of

Human Development, 6(1): 93-117.

Robeyns, I., 2006. The Capability Approach in Practice. Journal of Political

Philosophy, 17(3): 351-376.

Robeyns, I., 2006. Three models of education: Rights, Capability and Human

Capital. Theory and Research in Education. 4(1), 69-84.

Robeyns, I., 2011. The Capability Approach. In the Stanford Encyclopaedia of

Philosophy (Summer Edition), Edward N. Zalta (ed). (Available at URL:

http://plato.stanford.edu/archives/sum2011/entries/capability-approach)

Robbins, L., 1932. An Essay on the Nature and Significance of Economic

Science. London: Macmillan.

Roemer, J., 1996. Theories of Distributive Justice. Cambridge MA: Harvard

University Press.

Ryan, R.M. and Deci, E.L., 2000. Self-Determination Theory and the Facilitation

of Intrinsic Motivation, Social Development, and Wellbeing. American

Psychologist, 55: 68-78.

Sabir, M., 2003. Gender and Public Spending on Education in Pakistan: A Case

Study of Disaggregated Benefit Incidence. Conference Paper No. 48.

Saito, M., 2003. Amartya Sen’s Capability Approach to Education: A Critical

Exploration. Journal of Philosophy of Education, 37(1): 17-33.

Seligman, M.E.P., 2011. Flourish: A Visionary New Understanding of Happiness

and Wellbeing. New York: Free Press.

http://plato.stanford.edu/archives/sum2011/entries/capability-approach

140 | P a g e

Sen, A.K., 1971. Choice Functions and Revealed Preference. The Review of

Economic Studies, 38(3): 307-317.

Sen, A.K., 1973. Behaviour and the Concept of Preference. Economica, 40(159):

241-259.

Sen, A.K., 1977. Rational Fools: A Critique of the Behavioural Foundations of

Economic Theory. Philosophy and Public Affairs, 6(4): 317-344.

Sen, A.K., 1979. Personal Utilities and Public Judgements: Or What’s Wrong

With Welfare Economics. The Economic Journal, 89(355): 537-558.

Sen, A.K., 1983. Poor, Relatively Speaking. Oxford Economic Papers, 35: 153-

169.

Sen, A.K., 1984. Resources, Values and Development. Oxford: Basil Blackwell.

Sen, A.K., 1985a. Wellbeing, Agency and Freedom: The Dewey Lectures 1984.

Journal of Philosophy, 82 (4): 169-221.

Sen, A.K., 1985b. Commodities and Capabilities. North-Holland.

Sen, A.K., 1987a. Standard of Living. Cambridge University Press.

Sen, A.K., 1987b. Freedom of Choice: Concept and Content. World Institute for

Development Economics Research of the United Nations University, WP 25.

Sen, A.K., 1988. On Ethics and Economics. Blackwell Publishing.

Sen, A.K., 1990. Justice: Means versus Freedoms. Philosophy and Public Affairs,

19(2): 111-121.

Sen, A.K., 1991. Welfare, Preference and Freedom. Journal of Econometrics, 50:

15-29.

Sen, A.K., 1992. Inequality Reexamined. Oxford: Oxford University Press.

Sen, A.K., 1993. Positional Objectivity. Philosophy and Public Affairs, 22(2): 126-

145.

141 | P a g e

Sen, A.K., 1997a. Maximisation and the Act of Choice. Econometrica, 65(4): 745-

779.

Sen, A.K., 1997b. Editorial: Human Capital and Human Capability. World

Development, 25(12): 1959-1961.

Sen, A.K., 1999. Development as Freedom. Oxford University Press.

Sen, A.K., 2002. Rationality and Freedom. Massachusetts: Belknap Press.

Sen, A.K., 2004a. Dialogue: Capabilities, Lists and Public Reason: Continuing the

Conversation. Feminist Economics, 10(3): 77–80.

Sen, A.K., 2004b. Elements of a Theory of Human Rights. Philosophy and Public

Affairs, 32(4): 315-356.

Sen, A.K., 2009. The Idea of Justice. ALLAN LANE, Penguin Group.

Siddiqui, R. and Hamid, S., 2003. Correlates of Poverty: Gender Dimensions.

Final report, Pakistan Institute of Development Economics, Islamabad, Pakistan.

StataCorp. 2009. Stata Statistical Software: Release 11. TX: StataCorp LP.

Stiglitz, J., Sen, A. and Fitoussi, J.P., 2009. Report by the Commission on the

Measurement of Economic Performance and Progress. (Available at URL:

http://www.stiglitz-sen-fitoussi.fr/documents/rapport_anglais.pdf).

Sugden, R., 1993. Welfare, Resources, and Capabilities: A Review of Inequality

Reexamined by Amartya Sen. Journal of Economic Literature, 31: 1947-1962.

Sugden, R., 2010. Opportunity as Mutual Advantage, Economics and Philosophy,

26(1): 47-68.

Tao, S. 2010. Applying the capability approach to school improvement

interventions in Tanzania. EdQual working paper no: 22, Institute of Education,

University of London, U.K.

Tenenhaus, M., Esposito, V.V., Chatelin, Y.M. and Lauro C., 2005. PLS Path

Modeling. Computational Statistics and Data Analysis, 48: 159-205.

http://www.stiglitz-sen-fitoussi.fr/documents/rapport_anglais.pdf

http://0-journals.cambridge.org.alpha2.latrobe.edu.au/action/displayJournal?jid=EAP

142 | P a g e

Trinchera, L. and Roussolillo, G., 2010. On the use of Structural Equation Models

and PLS Path Modeling to Build Composite Indicators. Working Paper No: 30-

2010. Macerata University, Department of Studies on Economic Development

(DiSSE).

Tukey, J.W., 1977. Exploratory Data Analysis. Addison-Wesley Publishing

Company.

Unterhalter, E, 2003. Education, Capabilities, Social justice. Background paper

prepared for the Education for All Global Monitoring Report 2003/4. Gender and

Education for All: The Leap to Equality. UNESCO. 2004/ED/EFA/MRT/PI/76.

Unterhalter, E., Vaughan, R., and Walker, M. 2007. The Capability Approach and

Education. Prospero, November.

Unterhalter, E. 2003. The Capability Approach and Gendered Education: An

Examination of South African Complexities. Theory and Research in Education,

1(1), 7-22.

Unterhalter, E. 2005. Global inequality, capabilities, social justice and the

Millennium Development Goal for gender equality in education. International

Journal of Education Development, 25(2), 111-122.

Van Ootegem, L. and Spillemaeckers, S., 2008. With a Focus on Well-being and

Capabilities. Contribution to the 2008 Conference of the HDCA, New Delhi / 10-

13 September.

Vikander, N., 2007. Kahneman’s Objective Happiness and Sen’s Capabilities: a

Critical Comparison. (Available at URL: www.tinbergen.nl/~vikander).

Walker, M. 2005. Amartya Sen’s capability approach and education. Educatonal

Action Research. 13(1).

Walker, M. 2006. Towards a capability based theory of social justice in education,

Journal of Education Policy. 21(2), 163-185.

http://www.tinbergen.nl/~vikander

143 | P a g e

Watts, M. and Bridges, D. 2006. Enhancing Students’ Capabilities? UK higher

education and the widening participation agenda, in S. Deneulin et al. (eds),

Transforming Unjust Structures: The Capability Approach. Dordrecht: Springer

Verlag.

Williams, R., 2010. Generalized Ordered Logit Models. (Available at URL:

http://www.nd.edu/~rwilliam/xsoc73994/MSS2010-Handout.pdf).

Wold, H., 1966. Estimation of Principal Components and Related Models by

Iterative Least Squares. In Multivariate Analysis, P.R. Krishnaiah (ed). New York:

Academic Press.

Wold H., 1982. Soft Modeling: The Basic Design and Some Extensions. In

Systems under Indirect Observation, Jöreskog-Wold (eds). North-Holland.

Wold, H., 1985. Partial Least Squares. In Encyclopedia of Statistical Sciences

(Vol. 6), S. Kotz and N.L. Johnson (eds): 581-591. New York: Wiley.

Wold, H., 1987. Response to D.A. Freedman. Journal of Educational Statistics,

12(2): 202-205.

Wolfers, Justin. (2003). Is Business Cycle Volatility Costly? Evidence from

Surveys of Subjective Well-being. International Finance 6(1): 1-26.

Zaman, A., Rousseeuw, P.J. and Orhan, M., 2001. Econometric Applications of

High-Breakdown Robust Regression Techniques. Economics Letters, 71: 1-8.

http://www.nd.edu/~rwilliam/xsoc73994/MSS2010-Handout.pdf

144 | P a g e

Appendix B

Linear regressions and bootstrap estimates at district level District Attock Linear regression

Number of obs = 92 F(2, 90) = 36.10 Prob > F = 0.0000 R-squared = 0.3743 Root MSE = .8549

Robust sa_model5 Coef. Std. Err. t P>|t| [95% Conf. Interval] sfta_model5 .4805633 .1033797 4.65 0.000 .2751816 .685945 sata_model5 .4251866 .1087287 3.91 0.000 .2091781 .6411951

Source SS df MS Number of obs = 1000 F(1, 998) = 478.30

Model 3.67379658 1 3.67379658 Prob > F = 0.0000 Residual 7.66552687 998 .007680889 R-squared = 0.3240

Total 11.3393234 999 .011350674 Adj R-squared = 0.3233 Root MSE = .08764

b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.5363017 .0245221 -21.87 0.000 -.5844224 -.4881809 _cons .7068227 .0106643 66.28 0.000 .6858957 .7277496 District Rawalpindi Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.9981162 .0117512 -84.94 0.000 -1.021176 -.9750563 _cons .9436537 .0064903 145.39 0.000 .9309175 .9563899

145 | P a g e

District Jhelum Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.6884998 .0176965 -38.91 0.000 -.7232264 -.6537731 _cons .6969305 .0113373 61.47 0.000 .6746827 .7191782 District Islamabad Linear regression


Robust sa_model5 Coef. Std. Err. t P>|t| [95% Conf. Interval] sfta_model5 .6490491 .1048449 6.19 0.000 .4346172 .863481 sata_model5 .098822 .1648041 0.60 0.553 -.2382403 .4358842




b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.3716876 .0166409 -22.34 0.000 -.4043429 -.3390324 _cons .687428 .0031095 221.08 0.000 .6813262 .6935299 District Sargodha Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.4708502 .0267677 -17.59 0.000 -.5233777 -.4183227 _cons .6495741 .0140957 46.08 0.000 .6219134 .6772348

146 | P a g e

District Mianwali Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.545593 .0195224 -27.95 0.000 -.5839026 -.5072834 _cons .6667504 .0128261 51.98 0.000 .6415811 .6919197 District Khushab Linear regression


Robust sa_model5 Coef. Std. Err. t P>|t| [95% Conf. Interval] sfta_model5 .1452141 .1817076 0.80 0.429 -.223306 .5137341 sata_model5 .6914681 .3180874 2.17 0.036 .046357 1.336579




b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.4356864 .01095 -39.79 0.000 -.4571741 -.4141988 _cons .4493825 .0083913 53.55 0.000 .432916 .4658491 District Bhakkar Linear regression




Model .685652385 1 .685652385 Prob > F = 0.0000 Residual 2.03732158 998 .002041404 R-squared = 0.2518



147 | P a g e

District Lahore Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.5391262 .0288566 -18.68 0.000 -.5957528 -.4824996 _cons .5892244 .0101387 58.12 0.000 .5693289 .60912 District Kasur Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.5030738 .0246592 -20.40 0.000 -.5514635 -.454684 _cons .5385184 .0121748 44.23 0.000 .5146273 .5624095 District Sheikhpura Linear regression







148 | P a g e

District Okara Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.773656 .0247984 -31.20 0.000 -.822319 -.724993 _cons .7318292 .0102874 71.14 0.000 .7116418 .7520165 District Gujranwala Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.443567 .023135 -19.17 0.000 -.4889659 -.3981682 _cons .54701 .0099904 54.75 0.000 .5274053 .5666147 District Gujrat Linear regression







149 | P a g e

District Sialkot Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.8170956 .0270847 -30.17 0.000 -.8702451 -.763946 _cons .7719887 .0123139 62.69 0.000 .7478246 .7961529 District Faisalabad Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.5935131 .0191486 -31.00 0.000 -.6310891 -.555937 _cons .6186717 .0088317 70.05 0.000 .6013409 .6360026 District T.T. Singh Linear regression







150 | P a g e

District Jhang Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.4767562 .028501 -16.73 0.000 -.532685 -.4208274 _cons .6308671 .0112141 56.26 0.000 .6088612 .6528731 District Multan Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.511929 .0219401 -23.33 0.000 -.554983 -.468875 _cons .5212069 .006939 75.11 0.000 .5075902 .5348237 District Vihari Linear regression







151 | P a g e

District Sahiwal Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.7337746 .0264933 -27.70 0.000 -.7857637 -.6817856 _cons .6811638 .0146286 46.56 0.000 .6524576 .7098701 District D.G. Khan Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.6688425 .0364882 -18.33 0.000 -.7404449 -.5972401 _cons .7492074 .0190893 39.25 0.000 .7117476 .7866671 District Leiah Linear regression







152 | P a g e

District Muzaffargarh Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.5026088 .030437 -16.51 0.000 -.5623367 -.4428809 _cons .536909 .0034423 155.97 0.000 .530154 .543664 District Rajanpur Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.6658128 .0179765 -37.04 0.000 -.701089 -.6305367 _cons .6965505 .010448 66.67 0.000 .676048 .717053 District Bhawalpur Linear regression







153 | P a g e

District Bhawalnagar Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.374179 .0129166 -28.97 0.000 -.3995257 -.3488322 _cons .377276 .0073747 51.16 0.000 .3628043 .3917477 District R.Y.Khan Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.4354749 .020953 -20.78 0.000 -.476592 -.3943579 _cons .4802194 .0139534 34.42 0.000 .452838 .5076007 District Jacobabad Linear regression







154 | P a g e

District Khairpur Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.3901101 .0221885 -17.58 0.000 -.4336516 -.3465687 _cons .761184 .005036 151.15 0.000 .7513017 .7710664 District Shikarpur Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.5840106 .0219767 -26.57 0.000 -.6271365 -.5408847 _cons .6630462 .0112418 58.98 0.000 .6409859 .6851065 District Sukkur Linear regression







155 | P a g e

District Larkana Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.6332575 .0280895 -22.54 0.000 -.6883788 -.5781361 _cons .6111206 .0123772 49.37 0.000 .5868323 .6354088 District Dadu Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.8614513 .0226513 -38.03 0.000 -.9059009 -.8170017 _cons .79928 .0122227 65.39 0.000 .7752949 .8232652 District Hyderabad Linear regression







156 | P a g e

District Badin Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.4102443 .0164377 -24.96 0.000 -.4425007 -.377988 _cons .6321478 .0035446 178.34 0.000 .6251922 .6391035 District Sanghar Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.4858864 .0234736 -20.70 0.000 -.5319498 -.4398231 _cons .5391661 .0116104 46.44 0.000 .5163825 .5619496 District Tharparkar Linear regression







157 | P a g e

District Thatta Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.3065503 .0321471 -9.54 0.000 -.3696339 -.2434667 _cons .3648106 .0115799 31.50 0.000 .3420868 .3875343 District Mirpur Khas Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.4060406 .0227952 -17.81 0.000 -.4507726 -.3613085 _cons .6698069 .0102273 65.49 0.000 .6497375 .6898763 District Nawab Shah Linear regression







158 | P a g e

District Karachi Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.4707405 .0258109 -18.24 0.000 -.5213903 -.4200907 _cons .5260098 .0151618 34.69 0.000 .4962572 .5557625 District Dir Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.4921728 .0264605 -18.60 0.000 -.5440974 -.4402483 _cons .6454203 .0117271 55.04 0.000 .6224077 .6684329 District Swat Linear regression







159 | P a g e

District Mansehra Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 .140898 .0321998 4.38 0.000 .0777109 .2040851 _cons .457577 .0111177 41.16 0.000 .4357603 .4793937 District Abbottabad Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.3458151 .0361692 -9.56 0.000 -.4167915 -.2748388 _cons .4822987 .0151329 31.87 0.000 .4526028 .5119947 District Mardan Linear regression







160 | P a g e

District Peshawar Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.8418049 .0283193 -29.73 0.000 -.897377 -.7862327 _cons .8205182 .0151433 54.18 0.000 .7908019 .8502345 District Kohat Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.7306785 .0376252 -19.42 0.000 -.804512 -.656845 _cons .6493686 .0187388 34.65 0.000 .6125966 .6861407 District Karak Linear regression


Robust sa_model5 Coef. Std. Err. t P>|t| [95% Conf. Interval] sfta_model5 .1592298 .1101543 1.45 0.158 -.0654314 .383891 sata_model5 .568799 .1076584 5.28 0.000 .3492282 .7883697





161 | P a g e

District D.I. Khan Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.3711288 .020603 -18.01 0.000 -.411559 -.3306985 _cons .2058009 .0087268 23.58 0.000 .1886758 .2229259 District Bannu Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.6535491 .0279017 -23.42 0.000 -.7083017 -.5987965 _cons .5827526 .0172712 33.74 0.000 .5488607 .6166446 District Quetta Linear regression







162 | P a g e

District Sibi Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.4583713 .0243975 -18.79 0.000 -.5062475 -.410495 _cons .4186689 .0169474 24.70 0.000 .3854123 .4519256 District Kalat Linear regression






b1 Coef. Std. Err. t P>|t| [95% Conf. Interval] b2 -.9902915 .0236565 -41.86 0.000 -1.036714 -.9438692 _cons .880068 .0042011 209.48 0.000 .8718239 .8883121 District Mekran Linear regression







163 | P a g e

District Loralai Linear regression


Robust sa_model5 Coef. Std. Err. t P>|t| [95% Conf. Interval] sfta_model5 .2935504 .2732228 1.07 0.292 -.2661211 .8532219 sata_model5 .3016196 .2986575 1.01 0.321 -.3101525 .9133917





164 | P a g e

Appendix C

Ordered logit results: dependent variable: Happiness (HAP1) Iteration 0: log likelihood = -6334.067 Iteration 1: log likelihood = -4786.642 Iteration 2: log likelihood = -4480.9951 Iteration 3: log likelihood = -4472.2724 Iteration 4: log likelihood = -4472.2495 Iteration 5: log likelihood = -4472.2495

Ordered logistic regression Number of obs = 6749 LR chi2(10) = 3723.64 Prob > chi2 = 0.0000 Log likelihood = -4472.2495 Pseudo R2 = 0.2939

HAP1 Coef. Std. Err. z P>|z| [95% Conf. Interval] sasumz .7216531 .0399419 18.07 0.000 .6433684 .7999378

satasumz .276773 .037305 7.42 0.000 .2036565 .3498896 sftasumz 1.260697 .0377428 33.40 0.000 1.186722 1.334671 incomez .0835347 .0298112 2.80 0.005 .0251058 .1419635

eduz -.0291005 .0325675 -0.89 0.372 -.0929316 .0347307 GEND .0483538 .0604498 0.80 0.424 -.0701257 .1668334

UR -.1110327 .0634394 -1.75 0.080 -.2353716 .0133063 balochistan -.264735 .1129763 -2.34 0.019 -.4861644 -.0433055

nwfp .0445073 .1021958 0.44 0.663 -.1557929 .2448074 punjab -.2678903 .0729062 -3.67 0.000 -.4107839 -.1249968

/cut1 -5.284152 .1484268 -5.575063 -4.993241 /cut2 -2.529737 .1252372 -2.775198 -2.284277 /cut3 2.537041 .1246432 2.292745 2.781337

Iteration 0: log likelihood = -6334.067 Iteration 1: log likelihood = -4792.8265 Iteration 2: log likelihood = -4491.4578 Iteration 3: log likelihood = -4483.002 Iteration 4: log likelihood = -4482.9793 Iteration 5: log likelihood = -4482.9793




eduz -.024649 .0323659 -0.76 0.446 -.088085 .038787 GEND .0537417 .0602751 0.89 0.373 -.0643953 .1718787

UR -.1363482 .0626777 -2.18 0.030 -.2591943 -.0135022 /cut1 -5.143078 .1399286 -5.417333 -4.868822 /cut2 -2.396629 .1154224 -2.622853 -2.170405 /cut3 2.663689 .1154382 2.437435 2.889944

165 | P a g e





eduz -.0075023 .031383 -0.24 0.811 -.0690118 .0540072 GEND .0448545 .0601232 0.75 0.456 -.0729848 .1626938

/cut1 -4.922645 .0954927 -5.109808 -4.735483 /cut2 -2.176732 .0547188 -2.283979 -2.069485 /cut3 2.880673 .0592272 2.76459 2.996756





eduz -.0014298 .0303095 -0.05 0.962 -.0608354 .0579757 /cut1 -4.944829 .0908595 -5.12291 -4.766747 /cut2 -2.199292 .0457336 -2.288929 -2.109656 /cut3 2.858333 .0509639 2.758446 2.958221

166 | P a g e




satasumz .2648886 .0362121 7.31 0.000 .1939142 .335863 sftasumz 1.264591 .0372861 33.92 0.000 1.191512 1.337671

/cut1 -4.941217 .0908219 -5.119224 -4.763209 /cut2 -2.195262 .0456714 -2.284777 -2.105748 /cut3 2.853311 .0508346 2.753677 2.952945





eduz -.1015571 .0560164 -1.81 0.070 -.2113472 .0082331 UR -.0619335 .0949928 -0.65 0.514 -.248116 .1242489

balochistan -.3095051 .1718988 -1.80 0.072 -.6464206 .0274103 nwfp -.0773822 .1513369 -0.51 0.609 -.3739971 .2192326

punjab -.4222614 .1094961 -3.86 0.000 -.6368698 -.2076529 /cut1 -5.546819 .2173463 -5.97281 -5.120828 /cut2 -2.694861 .1816205 -3.05083 -2.338891 /cut3 2.719304 .1801768 2.366164 3.072445

167 | P a g e




satasumz .3645814 .0500191 7.29 0.000 .2665458 .462617 sftasumz 1.049115 .0482846 21.73 0.000 .9544794 1.143752 incomez .0636295 .0453125 1.40 0.160 -.0251814 .1524403

eduz .0099161 .0398355 0.25 0.803 -.06816 .0879922 UR -.1575638 .0863296 -1.83 0.068 -.3267666 .0116391

balochistan -.207988 .1511975 -1.38 0.169 -.5043297 .0883538 nwfp .2086693 .1399466 1.49 0.136 -.0656211 .4829596

punjab -.1225849 .0982989 -1.25 0.212 -.3152472 .0700775 /cut1 -5.089204 .2009317 -5.483023 -4.695385 /cut2 -2.399327 .1696358 -2.731807 -2.066847 /cut3 2.39256 .1694141 2.060514 2.724605





eduz -.0153027 .0442354 -0.35 0.729 -.1020026 .0713971 GEND .1430822 .0968868 1.48 0.140 -.0468124 .3329767

balochistan -.5491627 .1604023 -3.42 0.001 -.8635454 -.23478 nwfp -.154845 .1574325 -0.98 0.325 -.4634071 .153717

punjab -.3925083 .1120282 -3.50 0.000 -.6120795 -.1729371 /cut1 -5.05259 .1769595 -5.399424 -4.705756 /cut2 -2.36585 .1195186 -2.600102 -2.131598 /cut3 2.469549 .1196247 2.235089 2.704009

168 | P a g e




satasumz .2515357 .048952 5.14 0.000 .1555916 .3474799 sftasumz 1.327874 .0485003 27.38 0.000 1.232816 1.422933 incomez .0301419 .0518374 0.58 0.561 -.0714575 .1317413

eduz -.0582403 .0481521 -1.21 0.226 -.1526166 .036136 GEND .0019212 .0780399 0.02 0.980 -.1510341 .1548766

balochistan .032628 .1587191 0.21 0.837 -.2784557 .3437117 nwfp .2211692 .1349714 1.64 0.101 -.0433698 .4857083

punjab -.135728 .0968859 -1.40 0.161 -.3256209 .0541648 /cut1 -5.056046 .1455195 -5.341259 -4.770833 /cut2 -2.250235 .1058316 -2.457661 -2.042809 /cut3 2.96361 .1100859 2.747846 3.179374

169 | P a g e

Appendix D

One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] sfta_a~2 57 .4912281 .1005088 .7588244 .2898848 .6925713

mean = mean(sfta_absd12) t = -24.9607 Ho: mean = 3

degrees of freedom = 56

Ha: mean < 3 Ha: mean != 3 Ha: mean > 3 Pr(T < t) = 0.0000 Pr(|T| > |t|) = 0.0000 Pr(T > t) = 1.0000

One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] sfta_a~3 57 2.45614 .4136382 3.1229 1.627523 3.284757









mean = mean(sfta_absd15) t =-2.8099 Ho: mean = 3



One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] happ_~12 57 .6842105 .1204573 .9094326 .4429056 .9255155

mean = mean(happ_absd12) t = -19.2250 Ho: mean = 3



170 | P a g e

One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] happ_~13 57 2.017544 .2924394 2.207869 1.431717 2.60337

mean = mean(happ_absd13) t =-3.3595 Ho: mean = 3



One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] happ_~14 57 1.22807 .2832163 2.138236 .6607197 1.795421




One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] happ_~15 57 1.052632 .1447084 1.092524 .7627459 1.342517




One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] sata_~12 57 .4736842 .0972854 .7344887 .2787981 .6685703

mean = mean(sata_absd12) t = -25.9681 Ho: mean = 3



One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] sata_~13 57 2.22807 .2951334 2.228208 1.636847 2.819293




171 | P a g e

One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] sata_~14 57 1.561404 .2515834 1.899413 1.057421 2.065386




One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] sata_~15 57 1.070175 .1657739 1.251565 .7380904 1.40226

mean = mean(sata_absd15) t =-11.6413 Ho: mean = 3



One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] sfta_a~2 57 .4912281 .1005088 .7588244 .2898848 .6925713












172 | P a g e





One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] happ_~_d 57 4.438596 .5751852 4.342553 3.286362 5.590831

mean = mean(happ_sata_d) t = 2.5011 Ho: mean = 3



One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] hap~sa_d 57 4.403509 .6927207 5.229926 3.015823 5.791195

mean = mean(happ_sa_d) t = 2.0261 Ho: mean = 3



One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] ha~fta_d 57 5.614035 .68768 5.19187 4.236447 6.991624

mean = mean(happ_sfta_d) t = 3.8012 Ho: mean = 3



One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] sa_sat~d 57 4.438596 .5751852 4.342553 3.286362 5.590831

mean = mean(sa_sata_d) t = 2.5011 Ho: mean = 3



173 | P a g e

One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] sa_sft~d 57 6.122807 .894872 6.756135 4.330163 7.915451

mean = mean(sa_sfta_d) t = 3.4897 Ho: mean = 3



One-sample t test Variable Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] sata_~_d 57 6.596491 .8425033 6.36076 4.908754 8.284228

mean = mean(sata_sfta_d) t = 4.2688 Ho: mean = 3



Documents

INTEGRATING CAPABILITY AND HAPPINESS APPROACHES IN … · subjective wellbeing (SWB) provide distinctive information, not contained in the happiness indicator, while together with