Measuring job quality and job satisfaction

Erik Schokkaert�, Luc Van Ootegemy, Elsy Verhofstadtzx

First and preliminary draft, August 2009

Abstract

Job quality is a multi-dimensional concept, that has become prominenton the agenda of policy-makers. There is no consensus about how tomeasure and how to monitor it. In this paper we compare often usedobjective and subjective indicators of job quality. We argue that objectiveindicators are �too objective�, as they neglect interindividual di¤erences inpreferences, while subjective job satisfaction is �too subjective�as it alsore�ects di¤erences in aspirations. We propose an alternative measure ofjob quality in terms of equivalent incomes that does respect individualpreferences but rules out aspirations. We illustrate our approach withFlemish data on school-leavers (SONAR) using the information on the �rstjob of the 1978 birth cohort. We compare the results for the equivalentincome indicator with the results of objective and subjective indicators.

Key words: job quality, job satisfactionJEL classi�cation: J28, J80

1 Introduction

�Full employment� has always been one of the most prominent objectives ofeconomic and social welfare policy. In recent decades, there has been a growingawareness that not only the number, but also the quality of jobs is important.In Europe, it is at the top of the policy agenda ever since the Lisbon, Niceand Stockholm summits (2000 and 2001). Monitoring �job quality�is not easy,however, and the speci�cation of what constitutes an adequate measure of jobquality is still in abeyance.There is a large consensus, both in academic and in policy circles, that job

quality is a multidimensional concept. As an example of the former, Clark (2005)shows that changes in income and hours of work are insu¢ cient to describe job

�CORE, UCLouvain; Department of Economics, KULeuven.yHABE, University College Ghent; Sherppa, Ghent University; Higher Institute of Labour

Studies (HIVA), KULeuven.zHABE, University College Ghent; Sherppa, Ghent University - Steunpunt Loopbanen.xThe authors acknowledge support from the Flemish Ministries of Science and Technology

and Education (PBO 1997 and 1998) and the Interuniversity Attraction Poles Program -Belgian Science Policy (Contract no. P5/21).

1

quality and that employees (in seven OECD countries) rate job security, inter-esting work and autonomy as essential dimensions of a good job. As an exampleof the latter, in 2001 a list of key indicators was included in the EmploymentGuidelines of the European Commission, identifying not less than ten main el-ements of quality in work (Davoine and Erhel, 2006). The multidimensionalnature of the concept raises a huge informational challenge, as it is necessaryto collect data on a (possibly wide) range of characteristics. More importantly,however, it also raises a di¢ cult indexing problem. In some exceptional situa-tions, the information on a vector of characteristics may seem su¢ cient to derivea quality ranking of jobs. This will be the case if one job dominates another, i.e.if it scores better on all relevant dimensions. More generally, however, if thereis no dominance, one will have to evaluate trade-o¤s, and therefore it is neces-sary to create an overall index, aggregating several dimensions of job quality.Two approaches to the indexing problem have been adopted in the recent past.The �rst constructs an objective summary index by applying a uniform set ofweights to the di¤erent dimensions, where �uniform�means that the weightsvector is the same for all jobs and individuals. The second assumes that subjec-tive measures of job satisfaction o¤er a natural way to aggregate the di¤erentjob dimensions taking into account interindividual di¤erences in preferences.The objective indicator approach starts from a set of job dimensions that

are judged to be relevant by the analyst or by the policy-maker. To get at anoverall index, one either uses statistical data reduction techniques, such as prin-cipal components or regression analysis (Davoine and Erhel, 2006; Kallebergand Vaisey, 2005; Jencks et al., 1988), or one applies a set of a priori-weightsthat are chosen by the analyst. This latter approach is predominantly used inpolicy oriented work. While equal weighting or simple averaging is most pop-ular (Heintz et al., 2005; Global Policy Network, 2006; Tangian, 2005, 2007)),some indices also implement di¤erential weights (European Job Quality Index,Leschke and Watt, 2008).A priori weighting leads to measures which are relatively easy to compute

and to interpret. However, it completely neglects the simple fact of life thatdi¤erent individuals have di¤erent ideas about the relative importance of dif-ferent job characteristics. As Tangian (2005, p. 12) writes: �A young womenwith a small child may pay more attention to time factors, a middle-aged manmay be most interested in career prospects, and a disabled worker may be moreconcerned with physical factors. Therefore, assigning a higher weight to careerprospects we favor the middle-aged man and discriminate both the woman andthe disabled worker. Higher weights of certain questions are advantageous forthose who are most interested in them and disadvantageous for those who arenot.�1 At this moment there is no scienti�c or social consensus about an ex-

1Tangian (2005) formulates the problem in a clear way. However, his �solution� is lesssatisfactory. He continues: �Thereby unequal question weights result in a factual inequalityof individuals. Therefore, the problem of weighting questions is closely linked to the oneof weighting individuals. Since individual weights are usually assumed equal (= one voterone vote), regardless of education, experience, or intelligence, the question weights should beassumed equal as well. Any deviation from equal weights is a source of debate, and to avoid

2

plicitly spelled-out objective theory of job quality, that would justify a set ofa priori-weights neglecting completely these interindividual di¤erences in pref-erences. Yet it is clear that normative or conceptual questions (�what is anormatively attractive measure of job quality?�) as such cannot be settled bydata analysis. One needs a theoretical framework to interpret the data andsummarize them in one single index.A second approach is to use subjective job satisfaction as a one-dimensional

proxy for job quality. It is adopted in a growing number of policy documents(e.g. Diaz-Serrano and Cabral Vieira, 2005; D�Addio et al., 2007; Eurofound,2006; SVR, 2007) and academic papers (e.g. Green, 2006; Hamermesh, 2001;Leontaridi and Sloane, 2001; Levy-Garboua and Montmarquette, 2004; Ritterand Anker, 2002). This approach starts from the assumption �that people areable to balance out the various aspects of job characteristics to come up with anoverall assessment of job quality.�(Kalleberg & Vaisey, 2005, p.434). If overallsatisfaction with one�s job is related in a meaningful way to satisfaction withthe various dimensions of that job, it can act as an aggregator that takes intoaccount the individual�s own evaluations.Yet it may be misleading to assume a one-to-one relationship between sub-

jective job satisfaction and job quality. The work on the determinants of jobsatisfaction indicates the central role of �adaptation�, �expectations�and �relativedeprivation�(the concept of a reference level to compare with) in explaining jobsatisfaction (Landy and Trumbo, 1976). One may argue that such di¤erences inaspiration levels should not induce di¤erences in measured job quality. Greenand Tsitsianis (2005) cautiously say that when job satisfaction is rising (falling)�we could conclude that workers�well-being is rising (falling), conditional onthe assumption that their norms are changing little or not at all�(p. 408, ouritalics). The same point is made by Hamermesh (2001), Levy-Garboua andMontmarquette (2004) and Munoz de Bustillo et al. (2005). Along the samelines, Brown et al. (2007) argue that policy evaluation with subjective job sat-isfaction measures may be highly misleading, as �many of those who do reporthigh job satisfaction and good employment relations are, on objective groundsin low quality jobs, and express satisfaction only against a low benchmark levelof norms and expectations�(Brown et al., 2007, p. 966).It seems that we are facing a dilemma here. At one side, we have a series

of objective indicators that implement a largely arbitrary weighting schemeneglecting individual preferences. These indicators are perhaps �too objective�.At the other side, we have measures of subjective job satisfaction which do takeinto account individual preferences, but at the same time also re�ect di¤erencesin expectations and in aspiration levels, that have not much to do with jobquality. These indicators are perhaps �too subjective�. Is there a way out ofthis dilemma? In this paper we argue that there is. In section 2, we introduce a

it equal weights are accepted whenever possible. In statistics it is also a tradition to acceptthe equal distribution (weights) by default, unless no other information is available.�This isa strange reasoning. In this setting equal weighting is not less arbitrary than any unequalweighting scheme, as this would be to the advantage of those individuals that consider thevarious job dimensions to be equally relevant.

3

conceptual framework borrowed from the social choice literature, to get a bettertheoretical insight into the precise nature of the dilemma. We will discuss thepros and the cons of both the objective indicator and the subjective satisfactionapproaches and propose a possible third approach, which respects individualpreferences but corrects for the e¤ects of expectations and aspirations. In section3 we present the Flemish dataset SONAR containing detailed information aboutthe quality of the �rst job of individuals. Section 4 discusses the estimates of ajob satisfaction equation estimated with these SONAR-data. In sections 5 and6, we then compute and compare the di¤erent measures of job quality that havebeen introduced in section 2. Section 7 concludes.Two caveats are in order. First, the discussion on job satisfaction and job

quality is closely related to the general debate on happiness and well-being. Infact, some consider job satisfaction as a proxy for workers�utility and thus fortheir well-being in general (e.g. Clark and Oswald, 1996; Frey and Stutzer,2002). The objective approaches to job quality are more in line with the inspi-ration of alternative approaches to well-being like the functionings-capabilitiesapproach of Sen (1985, 1992, 1999) and Nussbaum (2000, 2006).2 In this paper,we focus on measuring job quality as such and we will not link explicitly ourdiscussion to the broader setting of life satisfaction and well-being.3

Second, we focus on the question of deriving a normatively attractive mea-sure of job quality, which can be useful for monitoring the consequences ofdi¤erent policies. Another strand of the literature focuses on job satisfactionas an explanatory factor for crucial aspects of labour market behaviour (Free-man, 1978). Clark (2001a) o¤ers a non-exhaustive overview of research papers,showing that �happiness measures predict observable future behaviours or out-comes�. More speci�cally, job satisfaction in�uences productivity, absenteeismand turnover (e.g. Hall, 1994), it has an e¤ect on job search, job resignationor mobility in general (e.g. Clark 2001b, Delfgauw, 2007) and it goes togetherwith higher motivation and stronger commitment (De Witte, 2001). Our criti-cal analysis of subjective job satisfaction as a measure of job quality, does notin the least undermine the relevancy of these empirical �ndings. Nor do theseempirical �ndings undermine the relevancy of the normative analysis in thispaper.

2 Satisfaction, preferences and job characteris-tics: a conceptual framework

We �rst introduce some relevant concepts and notation, and we then formalizea set of interesting requirements that a good measure of job quality arguablyshould satisfy. After a discussion of the logical relations (and con�icts) betweenthese requirements we discuss the pros and cons of di¤erent speci�c measures of

2 In fact, Nussbaum�s theory of capabilities is one of the most ambitious attempts to buildan �objective� theory of well-being that is available in the present literature.

3Such a broader analysis can be found in Schokkaert (2007) and Fleurbaey et al. (2009).

4

job quality. Finally, these di¤erent measures are illustrated for a linear speci�-cation of the job satisfaction equation that will be implemented empirically inthe next sections.

2.1 Concepts and notation

Suppose that the job of individual i is characterized by a vector of objective jobcharacteristics Ci. We will sometimes partition this vector as (Yi; Di), whereYi is income (or wage) and Di is a vector of other objective job characteristics(such as degree of autonomy or creativity). Individual i has preferences overthese vectors, and we express this in the usual terminology with a well-de�nedpreference ordering Ri (with indi¤erence denoted by Ii and strict preferencedenoted by Pi). We assume that preferences are monotonic, i.e. we measureall the job characteristics such that Ci > C 0i implies CiRiC 0i. Individual prefer-ences re�ect the relative importance of the job characteristics for the individual.Of course, di¤erent individuals may have di¤erent relative weighting schemes,i.e. di¤erent preferences. We parameterize these di¤erences using a vector ofconditioning personality characteristics Z, such that Ri = R(Zi). Therefore,two individuals with the same values for the personality characteristics Z alsohave identical preferences. Note that the preference ordering is a purely ordinalconcept. Following standard microeconomics, it can be represented by a utilityfunction Ui, such that CiRiC 0i () Ui(Ci) > Ui(C 0i): Given the ordinal natureof preferences, all positive monotonic transformations of Ui are equally accept-able as a representation of the preference ordering Ri. Preferences determinethe indi¤erence curves in Figure 1.4 These indi¤erence curves show that thevector of job characteristics A is preferred by individual i to the vector of jobcharacteristics B and that the opposite is true for individual j. However, at thisstage we do not have su¢ cient information to compare the situation of bothindividuals, i.e. to derive an overall and interpersonally comparable measure ofjob quality.The subjective job satisfaction expressed by individual i is denoted by Si.

We do accept that their overall subjective job assessment at a given point intime is consistent with their preferences. However, subjective job satisfactionwill also be in�uenced by the aspirations (the reference levels) of individuali. Denoting these by Ai, we write Si as a function S(Ci; Ri; Ai). Consis-tency with preferences at given aspiration levels can then be written as CiRiC 0i() S(Ci; Ri; Ai) > S(C 0i; Ri; Ai). Comparing this expression with the de�n-ition of the utility function, one sees immediately that subjective measures ofjob satisfaction solve the problem of comparing individual situations by impos-ing one common cardinalization of the utility function for all persons, i.e. byintroducing an interpersonally comparable labeling of the indi¤erence curves inFigure 1. Of course, as soon as we can attach evaluative �numbers�to the indif-

4We focus here on orderings of the bundles of job characteristics only. These preferencesmust be seen as conditional on the amounts of other (private and public) goods available tothe individual. Separability assumptions are needed if we want the indi¤erence map in Figure1 to be independent of the amounts of these other commodities.

5

Figure 1: Individual preferences and job satisfaction

ference curves, we can unambiguously rank the individuals. It becomes possibleto compare the quality of job A (on the indi¤erence curve 1 of individual i) withthe quality of job B - and also to compare the quality of job A for individuali with the quality of that same job as evaluated by individual j. For the laterdiscussion, it is essential to see that the speci�c cardinalization (the speci�cvalue of the subjective measure of job satisfaction) does depend on the aspira-tion levels Ai. Aspiration levels may be di¤erent for di¤erent individuals and, asfor preferences, we parameterize these di¤erences using a vector of personalitycharacteristics �i, such that Ai = A(�i). Two individuals with the same valuesfor the personality characteristics �i also have identical aspiration levels. Notethat the vectors Zi and �i may partly be overlapping. Di¤erences in Zi (andtherefore Ri) lead to di¤erences in the slopes of the indi¤erence curves in Figure1, di¤erences in �i (and therefore Ai) lead to di¤erences in the cardinalizationof satisfaction, i.e. in the labeling of the indi¤erence curves.In order to measure job quality, we consider global situations (Ci; Ri; Ai),

or, equivalently, (Ci; Zi; �i). We denote job quality experienced by individual iby Qi and we say that Q(Ci; Ri; Ai) > Q(C 0i; R0i; A0i), if the job quality impliedby the situation (Ci; Ri; Ai) is at least as good as the job quality implied bythe situation (C 0i; R

0i; A

0i). We have opted deliberately for this general formula-

tion, because it allows us to compare di¤erent speci�c measures in a unifyingconceptual framework.

6

2.2 Requirements for a good measure of job quality5

The di¢ culty of measuring job quality is closely related to the multidimensionalnature of the concept. In this context it seems very natural to assume that thequality of job A cannot be lower than the quality of job B, if job A is at leastas good on all dimensions than job B. In more formal terms, this dominancecondition can be written as follows:

Condition 1 Dominance. If Ci > Cj, then Q(Ci; Ri; Ai) > Q(Cj ; Rj ; Aj).

Condition 1 is only helpful in a limited set of job comparisons, as it willbe rather exceptional that Ci > Cj . It will therefore only result in a (very)partial ordering. The measures that we will introduce in the next subsectionyield complete orderings, and imposing condition 1 then means that the com-plete ordering implied by the measure is an extension of the partial dominanceordering. At �rst sight, condition 1 may seem obvious. Indeed, how to explainto individual j that his job is of better quality than the one of individual i, if itscores worse on all relevant dimensions? However, note carefully that condition1 implies that all the information on interindividual di¤erences in preferencesand aspiration levels is neglected in the measure of job quality. In a moment,we will see that this has rather strong (and arguably undesirable) consequences.This brings us to a second possible requirement. In a situation in which

di¤erent individuals have di¤erent ideas about what is important in their job,i.e. have di¤erent preferences, it seems natural to require that a measure of jobquality should be consistent with these preferences in a comparison of two jobsituations for a given individual. How to convince individual i that the qualityof job B is higher than the quality of job A if she herself prefers job A overjob B? Moreover, it is equally natural to extend this idea to a comparison ofthe job situations of two individuals with identical preferences and aspirationlevels. If two individuals with exactly the same preferences and aspirationsagree that job A is better than job B, should a measure of job quality then notfollow these common ideas and feelings of the two individuals? We formalizethis requirement as

Condition 2 Respect of individual preferences.

(a) If CiRiC 0i, then Q(Ci; Ri; Ai) > Q(C 0i; Ri; Ai).(b) If Ri = Rj = R and Ai = Aj = A, then CiRCj () Q(Ci; R;A) >

Q(Cj ; R;A):

Condition 2 keeps aspirations constant. What if aspirations di¤er or change?Suppose that individual i at a certain point in time prefers job A (a creative jobwith a high wage) over job B (a boring job with a lower wage). As a result ofchanges in the economic environment (e.g. unemployment increases), she losesher job A but �nds a new job B. Of course, even in this period of crisis, she

5The conditions described in this section are the same as described in more general termsin Fleurbaey et al. (2009).

7

still prefers a creative job with a high wage over a boring job with a lower wage.Yet, she has adapted her aspiration levels downwards - and she is quite satis�edthat she could �nd job B. Indeed, in the light of the overall economic situation,her satisfaction with job B in a crisis period could even be higher than hersatisfaction with job A in a period of full employment. Would this imply that thequality of (the boring low-wage) job B for individual i is higher than the qualityof the (creative high-wage) job A? This conclusion would be quite misleadingif we want to use our measure to monitor the success of employment policiesin creating good jobs. Or consider two individuals with the same preferences:they both prefer the creative high-wage job A to the boring low-wage job B.However, the �rst individual comes from a deprived family and the second one isthe daughter of a professor and a succesful entrepreneur. The job expectationsfrom these two individuals may be very di¤erent and the �rst individual maybe more satis�ed in job B (a big leap forward compared to his family) than thesecond individual in job A. Would this mean that the quality of job B for that�rst individual is higher than the quality of job A for the second individual?Remember that they both prefer A to B. The poor individual would feel evenbetter-o¤ in job A, the rich individual would feel even more terrible in job B.These examples suggest that not correcting for changes and/or di¤erences inaspiration levels might in some circumstances lead to misleading conclusions.We therefore formulate a third requirement:

Condition 3 Independence of aspiration levels.

(a) If CiRiC 0i, then Q(Ci; Ri; Ai) > Q(C 0i; Ri; A0i).(b) If Ri = Rj = R, then CiRCj () Q(Ci; R;Ai) > Q(Cj ; R;Aj):

Condition 3 is stronger than condition 2, i.e. independence of aspirationlevels implies respect for individual preferences, but not vice versa. This will beimportant for the evaluation of subjective job satisfaction as a measure of jobquality. Equally important is the relationship between condition 2 and condition1. We mentioned already that imposing the dominance condition implies acomplete disregard of interindividual di¤erences in preferences - and thereforea complete disregard of the intuition behind condition 2. As a matter of fact, ithas been proven that both conditions are incompatible.6 This is easily seen fromFigure 1. Denote the job quality of the vectors A-D with subscripts. Imposingboth conditions 1 and 2 would then result in QB > QC (dominance), QC > QD(respect for preferences), QD > QA (dominance) and QA > QB (respect forpreferences). The existence of a cycle shows that it is not possible to constructa measure of job quality satisfying at the same time the dominance condition andrespecting individual preferences. Since there is a con�ict between conditions 1and 2, and condition 3 implies condition 2, it is clear that it is also impossible to

6The dominance condition was proposed by Sen (1985) to get at a partial ordering ofbundles of functionings (or capabilities). Its incompatibility with (even minimal) respect forpreferences has been proven and discussed in a series of papers by Brun and Tungodden (2004),Fleurbaey (2007) and Pattanaik and Xu (2007).

8

reconcile the dominance condition with independence of aspiration levels. Thistension will also be relevant for our comparison of di¤erent measures.To solve the con�ict, we may formulate a weaker version of condition 1:

Condition 4 Subset dominance. Consider a subregion T of the space of jobdimensions and assume that Ci; Cj 2 T: If Ci > Cj, then Q(Ci; Ri; Ai) >Q(Cj ; Rj ; Aj).

We will see that there is no con�ict between subset dominance and respectfor preferences. In fact, the combination of these two conditions leads to apreference-sensitive measure (that we will call �equivalent income�) that alsosatis�es independence of aspiration levels.

2.3 Comparing di¤erent measures

We are now well equipped to discuss and compare the pros and cons of thevarious measures of job quality that have been proposed in the literature. We�rst look at subjective job satisfaction, we then interpret the objective measuresand we �nally introduce a new preference-sensitive but aspiration-insensitivemeasure of job quality, that satis�es the condition of subset dominance.

2.3.1 Job satisfaction as an aggregative measure

The de�nition of subjective job satisfaction as a measure of job quality isstraightforward:

De�nition 5 Subjective indicators. QS(Ci; Ri; Ai) = S(Ci; Ri; Ai):

We can now evaluate these subjective indicators in the light of the require-ments that have been introduced in the previous subsection. First, subjectiveindicators do satisfy respect for preferences if the job satisfaction measure is con-sistent with preferences. This is the main justi�cation for their use. However,secondly, subjective job satisfaction does not satisfy independence of aspira-tion levels. As the example in the previous subsection has shown, if aspira-tion levels do change, it is possible to have at the same time CiRiC 0i and yetS(Ci; Ri; Ai) < S(C

0i; Ri; A

0i). And, if their aspiration levels di¤er, two individ-

uals i and j in the same �objective� situation Ci = Cj = C may both preferjob A over job B and yet the job satisfaction of the individual in job A may belower than the job satisfaction of the individual in job B. As we described in theintroduction, this issue has been discussed extensively in the literature and itwas the main argument for Brown et al. (2007) to claim that subjective job sat-isfaction is a misleading measure for monitoring employment policies. Thirdly,and less well known in the literature, subjective indicators do not satisfy thedominance condition. As we have seen, the dominance condition is incompatiblewith respect for preferences. Moreover, because of changes or di¤erences in theaspiration levels it is easily possible to have Ci > C 0i and yet Si < S0i. Returningto Figure 1, the cardinal labeling of the indi¤erence curves is in�uenced by the

9

aspiration levels. If individual i has more ambitious aspirations, his subjectivesatisfaction with job B may be lower than the subjective job satisfaction of theless ambitious individual j in job C.

2.3.2 Objective indicators of job quality

Objective indicators impose a set of dimensional weights on statistical or a priorigrounds, and hence they do not take into account (di¤erences in) individualpreferences at all. Although these indicators are usually not interpreted interms of standard economic theory, they can easily be interpreted within ourconceptual framework. Imposing weights implies imposing one speci�c referencepreference ordering (one set of indi¤erence curves in Figure 1) and a speci�ccardinalization of the utility function. Alternatively, one could say that objectiveindicators pick a reference individual (with speci�c reference values of R and A)and then evaluate jobs on the basis of the job satisfaction of that (arbitrarilychosen) reference individual. To give an example: equal weighting methodsimplicitly impose linear indi¤erence curves with slope -1. Denoting the referencevalues by a superscript r, we de�ne the objective indicators as

De�nition 6 Objective indicators. QO;r(Ci; Ri; Ai) = S(Ci; Rr; Ar):

We can easily check whether these measures satisfy our requirements. First,it is obvious that objective indicators simply ignore all information about in-dividual preferences and therefore cannot satisfy respect for preferences norindependence of aspiration levels (which implies respect for preferences). Sec-ondly, however, they do satisfy the dominance condition, as long as they re�ecta monotonic reference preference ordering, i.e. implement positive weights forall the dimensions.

2.3.3 A new preference-sensitive measure: equivalent incomes

We can now reformulate in more formal terms the dilemma that was alreadysketched in the introduction. Objective indicators do satisfy the dominancecondition, but ignore information about preferences and basically impose oneverybody the view of one arbitrarily chosen reference individual. Subjective jobsatisfaction does respect preferences but is also contaminated by expectationsand aspirations. Moreover, it does not satisfy dominance. Is there an attractivesolution in between these two extremes? Remember that it is pointless to lookfor a measure that would satisfy respect for preferences and dominance at thesame time: these two requirements are incompatible. However, there is one(and only one) approach that satis�es respect for preferences and subset dom-inance. This is the equivalence approach, proposed and defended in a series ofpublications by Fleurbaey (2007, 2008) and used for interpreting the subjectivehappiness-literature in Fleurbaey et al. (2009).The basic idea of the approach is illustrated in Figure 2, where we have in-

come Y on the horizontal axis and another job characteristic D on the verticalaxis. Suppose we want to compare the quality of job A for individual i with

10

Figure 2: Equivalent incomes

the quality of job B for individual j. We �rst pick a reference value Dr for D.Respect for preferences then implies that the quality for individual i of the jobvector A�is equal to the quality of job vector A, since they are both on the sameindi¤erence curve. An analogous reasoning holds for jobs B and B�and individ-ual j. Now, suppose that we can compare the job vectors A�and B�(di¤eringonly in income) on the basis of income. In that case the �equivalent incomes�Y �Ai and Y

�Bj can be interpreted as an interpersonally comparable measure of

job quality. In general terms, this measure is de�ned as follows:

De�nition 7 Equivalent income. QEI;r(Ci; Ri; Ai) = Y �i (Ci; Ri; Ai) with Y�i

implicitly de�ned by (Yi; Di)Ii(Y �i ; Dr).

How reasonable is it to suppose that job vectors A�and B�can be comparedon the basis of incomes only? Keeping to the two-dimensional example, is itsu¢ cient that Yi > Yj to draw the conclusion that (Yi; D) is better for indi-vidual i than (Yj ; D) is for individual j for all possible values D? This is notfully convincing. Suppose that D re�ects a low value of D. Suppose also thatindividual j does not care very much about income but does care a lot about D(her marginal rate of substitution between D and Y is large), while the oppositeis true for individual i. Could one then not argue that the quality of (Yi; D) forindividual j is lower than that of (Yj ; D) for individual i? A constructive wayto think about the issue is to phrase the following normative question: is therea reference value Dr such that a comparison of the situations of two individualscan be based on a comparisons of their incomes only? In some cases, the ques-

11

tion has an attractive answer. Suppose one takes for Dr the best possible valueof D. Nobody in society has a job with a higher value. In that case it wouldseem that individual i cannot legitimately complain that she is worse o¤ than jbecause she attaches more value to D. She has the best possible job along thatdimension anyway. If Dr refers to the best possible value of D, one might saythat in a comparison of the job vectors (Yi; Dr) and (Yj ; Dr) the only variablethat should legitimately matter is income.7

Let us emphasize that it can be shown (for a formal proof, see e.g. Fleurbaeyet al., 2009) that the equivalence approach is the only possible approach thatsatis�es respect for preferences and subset dominance. It does not satisfy thedominance condition: in the example of Figure 2, job B is dominating job A,but Y �Ai > Y

�Bj . However, the equivalent income does satisfy subset dominance,

with as the relevant subset the horizontal line through Dr.8 Note also thatequivalent income boils down to another cardinalization of the utility function.This cardinalization has the advantage that it satis�es independence of aspira-tion levels. Yet, this does come at a cost. Measured job quality will depend onthe values chosen for Dr. As we described, the choice of the reference vector isnot fully arbitrary and can be justi�ed on normative grounds. The more con-vincing this normative reasoning, the more convincing is the equivalent incomeapproach.

Table 1 summarizes our discussion of the characteristics of the three �fami-lies�of measures. In the next sections we will show that these insights are notonly theoretically, but also empirically relevant.

Table 1. Characteristics of the di¤erent measures of job quality

Subjective jobsatisfaction

Objective indi-cators

Equivalentincome

Dominance No Yes NoSubset domi-nance

No Yes Yes

Respect for pref-erences

Yes No Yes

Independence ofaspirations

No No Yes

7Fleurbaey (2005, 2006) has discussed the issue for income-health combinations.8 In fact, it can easily be seen that respect for preferences and subset dominance can only

be reconciled if the �subset� reduces to a line. Indeed, it it were not a line, one could alwaysconstruct a case of incompatibility as in Figure 1 (see, e.g., Fleurbaey et al., 2009).

12

2.4 A linear illustration

The various theoretical concepts can easily be illustrated for a linear equation,that will also be used in our empirical work. Suppose we start from observationson overall job satisfaction Si, which we assume to be consistent with preferencesin the sense de�ned above. A �rst approach is then simply to equate job qualityand job satisfaction, i.e. QSi = Si:To introduce our other concepts, we specify the empirical relationship be-

tween job satisfaction, job characteristics, preferences and aspirations in thefollowing way:

Si = �0 + �1Yi + �02Di + �

03�i + �

04YiZi + Z

0i�5Di + "i (1)

where "i is a disturbance and the ��s and �5 are coe¢ cients to be estimated.Note that the marginal rate of substitution between Yi and job characteristicDik is given by

MRSY;Dk

i =�2k + Z

0i�5k

�1 + �04Zi; (2)

where �5k is the k-th column of the coe¢ cients matrix �5. This expressionshows how di¤erences (changes) in Zi a¤ect the slope of the indi¤erence curves.The interaction coe¢ cients �4 and �5 play a crucial role in this regard. Notealso that di¤erences (changes) in the aspiration parameters �i only a¤ect thecardinalization of subjective job satisfaction and not the marginal rates of sub-stitution.Objective indicators of job quality neglect di¤erences in personal character-

istics and basically pick one reference set of indi¤erence curves and one referenceset of aspiration levels.9 Denoting these by the superscript r, we get

QO;ri (Ci; Ri; Ai) = �0 + �1Yi + �02Di + �

03�

r + �04YiZr + Zr0�5Di + "

r (3)

Di¤erent choices of the reference characteristics will lead to di¤erent measures.10

Finally, we have our new preference-sensitive measure. Denoting the refer-ence values for the dimensions as Dr, we compute the equivalent income indi-cator from the following equation

�0 + �1Y�i + �

02D

r + �03�i + �04Y

�i Zi + Z

0i�5D

r + "i =

�0 + �1Yi + �02Di + �

03�i + �

04YiZi + Z

0i�5Di + "i

9Since these reference aspiration levels are kept constant in comparisons between jobs, andenter eq. (3) as a simple level shift, they have no in�uence on the relative ranking of the jobs.10The approach by Jencks et al. (1988) is closely related to eq. (3). They conducted a

Survey of Job Characteristics, asking each worker to rate his or her job on a ratio scale inwhich an average job scores 100, regressed these scores on 48 job characteristics and used thecoe¢ cients to construct an Index of Job Desirability. The index incorporates 13 non-monetaryjob characteristics and some measures related to income. They themselves interpret their indexas re�ecting the preferences of an average worker. They also use their regression results toderive average values of the marginal rates of substitution in eq. (2).

13

resulting in

QEI;ri (Ci; Ri; Ai) = Yi +�02(Di �Dr) + Z 0i�5(Di �Dr)

�1 + �04Zi(4)

Note that the aspiration parameters (and the stochastic disturbance term) dis-appear from this equation (as they should). However, since the preference char-acteristics Zi and the interaction coe¢ cients �4 and �5 do appear in eq. (4),the equivalent income is preference-sensitive and takes into account preferencedi¤erences.

3 Data

We use a survey database for Flanders (SONAR). This database had been cre-ated to study the transition from education to the labour market, which makesit exceptionally rich on labour market information for school-leavers in their�rst working experiences. It provides information on the labour market statusfrom the moment one leaves school until the moment of the last interview aswell as on a whole range of socio-economic variables. As a consequence, we haveinformation which is often not available in other samples (e.g. information onhaving a driving license and club membership). The �rst job is de�ned as the�rst paid employment after leaving the educational system. It is a job withtenure of at least one month and for at least one hour a day and one day perweek.Data collection is achieved in two phases. First, a survey of a large and

heterogeneous sample of young adults of the same age is taken at regular inter-vals. Next, the same age group is surveyed again at a later stage (longitudinalfollow-up design).11 This study will use the longitudinal data of the birth co-hort 1978. These young Flemish adults were questioned a �rst time when theywere 23 years old and a second time when they were 26 years old. The sampleswere randomly selected and trained interviewers performed the oral interviewsat the interviewees�home address. The dataset is thus based on self-reportedinformation of the respondents.In the SONAR-questionnaire the workers were asked for their satisfaction

with 12 aspects of their job and with the job as a whole. The response-scalevaries from (1) �very unsatis�ed�to (5) �very satis�ed�. All individuals who haveonce started working thus had to give a retrospective evaluation of their jobsatisfaction at the beginning of their �rst job.12 We will use satisfaction withthe job as a whole in our analysis (one item measurement of �global satisfaction�,

11The response rate for the follow up interviews was about 70%.12For some respondents the beginning of the �rst job is longer ago than for others. This

di¤erence could a¤ect the recollection of the data. To control for the time span since therespondent started to work we include an extra variable in the model: the duration of �recallbias�. This variable expresses the number of months that passed since the start of the �rst job(date of the interview minus starting date of the �rst job). This variable was not signi�cant inour estimations and was therefore excluded from the �nal model presented in the next section.

14

see: Spector, 1997). Respondents were fairly satis�ed with their �rst labourmarket experience: almost three-quarter of the respondents were very or rathersatis�ed with their work as a whole. In the remainder we will talk about jobsatisfaction, although each time we mean satisfaction with the �rst job. In theremainder of this section, we will give a compact description of all the variablesthat are included in the explanatory model. In the next section, we present theestimation results of a �nal model where non-signi�cant variables are left out.The signi�cant variables get some more detailed description here.As characteristics of the �rst job we included the net monthly wage in full

time equivalents.13 The following job characteristics are dummy-variables: shiftwork, contract type, full-time or part-time work, private or public sector, com-pany size, whether or not receiving extra legal advantages, formal training dur-ing the �rst job and learning new skills during the job. Besides these there isalso a self-assessment of the worker on several characteristics of his job. Therespondents were asked to evaluate 19 items on a 4-point scale, ranging from�completely agree�, �rather agree�, �rather disagree�, to �completely disagree�. Us-ing factor analysis we reduced this set to four variables: (1) demanding workconcerning mental e¤ort and time pressure, (2) creative, responsible and variedjob, (3) physically demanding work and (4) autonomy.Several personality characteristics are included for the measurement of (dif-

ferences in) preferences or aspiration levels. Besides gender, educational level ofthe respondent and educational level of the mother we have information on thenumber of children and the nationality of the grandmother (from the mother�sside) of the respondent. The latter is often used for determining the ethnicalbackground.Next to these �traditional�conditioning variables, the dataset also allows to

include some characteristics which are often unknown. A �rst such variable isthe pollster�s impression of the respondent.14 We also have the respondents�answers on questions related to their motivation to work and their locus of con-trol (on a 4-point scale). Both sets of items were reduced using factor analysis.For �motivation to work�this generated 4 variables (motivation because of dis-advantages of not working, motivation from the content of work, motivationfrom material aspects and motivation because of advantages of working). Forlocus of control, �locus of control internal�and �locus of control external�indicatewhether the respondents believe that they themselves or some higher power (orother people) are responsible for their position in life. The third variable forlocus of control is labelled �own in�uence�and indicates to which extent the re-spondents believe to have their life in own hands and are able to execute plans.In this factor the focus is more on being able to do things, while the internallocus of control factor rather focus on who is responsible for achievements.The variable �number of o¤ers�counts the number of items respondents are

13The respondents were asked about their net monthly wage which we divided by thepercentage of employment.14The pollsters have to score the respondents to the account that they are energetic, active,

calm, friendly, cheerful, open, optimistic and motivated to answer. Using factor analysis wecreate a single variable indicating how others perceive the respondents.

15

willing to accept regarding work (15 items: eg. lower wage, part-time work,shift work, work for which one must travel). A last set of variables is related tosearch behaviour. On the one hand we include the number of used channels ororganisations. On the other hand we include dummies concerning the momentwhen one started to search for a job (never searched, start searching before orafter leaving school).

4 Estimating the job satisfaction equation

Since the dependent variable is measured on a �ve points-scale from low to highsatisfaction we opt for an ordinal regression analysis. Such a model is based onan underlying latent variable (Verbeek, 2000). We started the estimation witha model including all job characteristics (Y and D in equation (1)) and personalcharacteristics (� and Z in equation (1)) that are presented in section 3. Aswe consider all personal characteristics as potentially in�uencing aspirationsas well as preferences, the vectors � and Z coincide in our application. Weassume that the direct (level) e¤ects capture di¤erences in aspirations, and thatthe interaction e¤ects with the personal characteristics capture di¤erences inpreferences. Table 2 presents the estimation results of the �nal model wherenon-signi�cant variables are left out.

Table 2. Ordered logit regression results of job satisfaction in the�rst job

As the literature on job satisfaction suggests (e.g. Furnham, 1997; Gazioglu& Tansel, 2006), net income has a positive and strong e¤ect on job satisfaction.In line with the summary of Knoop & Schouteten (2006), we see a strong impactof the di¤erent job characteristics on job satisfaction. A creative, responsibleand varied job boosts job satisfaction whereas a physically demanding job hasa negative e¤ect on job satisfaction. Working in a large company, having muchautonomy and learning new skills contribute to a higher job satisfaction.Di¤erences in aspirations, the direct e¤ects of the personality characteristics,

are mainly captured by the motivation to work and the educational level. Peoplemotivated by the content of their work have a higher satisfaction. A highereducation reduces job satisfaction. This is in line with the hypothesis that ahigher educational level increases expectations (Verhofstadt et al., 2007). The�number of o¤ers� a respondent is willing to make for a job not surprisinglyincreases the job satisfaction.There is no signi�cant gender e¤ect on job satisfaction. Our sample consists

of people in their �rst job and as Clark (1997) argues, the gender di¤erencesshould appear only when workers are getting older. But it is rather surprisingthat also the interaction e¤ects between gender and the job characteristics are

16

insigni�cant. This means that men and women not only have the same aspira-tions but also the same preferences regarding their �rst job.15 The same holdsfor the interactions between wage and all conditioning variables: none of thecoe¢ cients �4 is signi�cantly estimated.This brings us to the analysis of the signi�cant interaction e¤ects. A men-

tally demanding job is harder for those who are more calm, friendly, open,motivated etc but is preferred when the locus of control is more internal. Thenegative e¤ect of a physically demanding job on job satisfaction is lower forthose motivated by the content and for those who are willing to make o¤ers fortheir job. The positive e¤ect of having a creative and responsible job is higherfor those motivated by content but lower for those motivated by materialism.The preference for creative and responsible jobs of respondents with a tertiarydegree increases their satisfaction. Large companies (which in general createmore job satisfaction) are not the place to be when you are motivated by con-tent. Those who are willing to o¤er much for their job or have a locus of controlthat is internal �do not like�autonomy.

5 Indicators of job quality

First, we will explain the calculation of the di¤erent job quality concepts asthey are presented in section 2. Then we will discuss the correlation betweenthe indicators. Finally we will compare the resulting job rankings.

5.1 Calculation of the indicators

The subjective indicator, reported job satisfaction (QS), respects individualpreferences but is also depending on individual aspirations (cfr table 1).In our sample the distribution of job satisfaction is as follows:

- Very unsatis�ed: 4.3%- Rather unsatis�ed: 9.6%- Neutral: 13.6%- Rather satis�ed: 48.9%- Very satis�ed: 23.6%

Objective indicators do not depend on individual aspirations but do notrespect individual preferences either. As shown in eq. (3), we have to pick ref-erence values for the preference variables. Objective indicators ��x�preferencesin a more or less arbitrary way. We have calculated three alternatives. The �rstone is an equal weighting method (QO;EW ) which imposes for all individualsequal (linear) indi¤erence curves (with slope -1). To make it possible to sum updi¤erent dimensions, the data for the di¤erent dimensions were normalised giv-ing �gures between 0 and 1.16 The objective indicator QO;EW is then a simple

15This result might be speci�c for the �rst job. It is quite well possible that men and womenhave di¤erent preferences with respect to their job when getting older.16Normalised value = (value - minimum)/ (maximum �minimum). The minimum value is

the value of the worst performer and the maximum is the value of the best performer.

17

average of the following dimensions:17

- Net monthly wage in full time equivalents- Non-standard forms of employment (average of shift work, temporary

contracts and part-time work)- Company size (as a proxy for collective interest representation and

participation)- Demanding work concerning mental e¤ort and time pressure- Creative, responsible and varied job- Physically demanding work- Autonomy- Learning new skills during �rst job

Two other objective indicators make use of the estimated job satisfactionequation, and pick reference levels for preferences and aspirations as presentedin equation (3). Recall that the choice of reference does not matter for theaspirations. However, the implied speci�c choice for the preference structure is ofcrucial importance. To show this, we take two types of reference individuals andevaluate jobs on the basis of the job satisfaction of these reference individuals:- QO;r1 Reference �individual average�: average values of Z and � in the

sample.- QO;r2:Reference �individual extreme�: average values for all Z and �-

variables except for motivation to work where we take as reference values thehighest motivation from the content of work and the lowest motivation frommaterial aspects. Motivation to work has signi�cant interaction e¤ects and thusa large impact on the preferences.Our indicator QO;r1 is closely related to the measure proposed by Jencks et

al. (1988).The equivalent income indicator (equation (4) QEI;rrespects individual pref-

erences but does not depend on individual aspirations. For the reasons discussedbefore, we use as reference values Dr the characteristics of a perfect job:

- Minimum demanding work wrt mental e¤ort and time pressure- Maximum creative, responsible and varied job- Large company- Minimum physically demanding work- Maximum autonomy- Learning new skills during �rst job

17This is similar as in the JQI (Leschke and Watt, 2008). The dimensions overlap to theextent that this was possible given the di¤erent nature of our data. We opted to include 8dimensions in stead of 6 in the JQI (wage, non standard forms of employment, working time,working conditions, skills, collective interest). Our �creative, demanding mentally and physi-cally, and autonomy�variables overlap with �working time and working condition�but get agreater weight. The results for our job satisfaction estimations show that these dimensions areimportant, so including them increases the comparability between the equal weights indicatorand the indicators that are based on the regression results.

18

5.2 Correlation between the di¤erent indicators

Table 3 presents the correlations between the di¤erent indicators. The correla-tion among the subjective indicator and the equal weights objective indicatoris the lowest (0.420). This indicator is the only one which does not use theestimated satisfaction model to calculate �job quality�. The objective indicatormaking use of the estimation results and weighting the job characteristics ac-cording to the personality characteristics of an average worker, correlates muchmore highly with the subjective indicator (0.670). This correlation is even higherthan the correlation between subjective job satisfaction and the equivalent in-come indicator (0.625). Since the latter rules out aspirations but respects prefer-ences, one could perhaps have expected a higher correlation with the subjectiveindicator. But we have to keep in mind that di¤erences between QO;r1 andQEI;r do not only result from the respect for preferences in the latter, but arealso sensitive to the choice of a reference level with perfect job characteristics(in the equivalent income case) versus the choice of reference individuals in theobjective case (personal characteristics are put equal to the average in QO;r1.The impact of the choice of the reference individual is illustrated when lookingat QO;r2: choosing a more extreme preference pattern (high motivation fromcontent, lowest motivation from material aspects) decreases the correlation withsubjective job satisfaction considerably (from 0.670 to 0.537).The comparison of the three objective indicators shows that the choice of

the relevant dimensions to include as well as a weighting scheme to aggregatethem, has serious consequences for the de�nition of what is a �good job�. Inmost cases, policy makers prefer an equal weighting scheme because it has the�avour of being �neutral�and use is made of the job characteristics that happento be available. We have argued that this assumption of neutrality is mistaken.Our empirical results show that the supposedly neutral weighting scheme withequal weights is more correlated with the extreme (motivation) reference case(0.821) than with the average weights reference case (0.674).The equivalent income indicator correlates most with the objective indicator

with an average reference individual (0.823). As said before, sources of di¤er-ence between both indicators are the di¤erent treatment of preferences and thedi¤erent choices for the references. The correlation between the equivalent in-come indicator and the alternative reference objective indicator (0.644) is lowerbecause individual preferences are more related to average preferences (as inQO;r1) than to more extreme preferences (as in QO;r2).

Table 3: Correlations between the di¤erent indicators

5.3 Comparison of the di¤erent indicators

In order to compare the di¤erent indicators in more detail, the distribution ofthe reported job satisfaction is applied to the other indicators such that thesame proportion of the sample is within each category. So for each indicator

19

4.3% of the sample is in the lowest category (1), . . . and 23.6% is in the highestcategory (5). The following tables analyse the movements between categorieswhen di¤erent indicators are compared.Table 4 and 5 compare objective indicators, the �rst compares equal weights

with average weights, the second compares average weights with the extreme(motivation) reference. In general, most of the shifts are those out of the cate-gory 3. Only 26% of the middle ranked jobs according the equal weights methodalso get that classi�cation according to the objective indicator with an averagereference individual. Once again we see that equal weighting is not as neutralas sometimes assumed. Even more, table 5 shows that any other weighting ofpreferences does make a di¤erence. Although the correlation between the twoobjective indicators with a reference individual is high (see table 3) table 5 in-dicates that the choice of the reference still has a notable impact on the rankingof the jobs.

Table 4: Cross tabulation of objective indicator using equal weightsand of objective indicator with average reference individual.Table 5: Cross tabulation of objective indicator with averagereference individual and objective indicator with reference

individual with a high motivation from the content of work and alow motivation from material aspects.

The following tables 6-9 compare the subjective indicator (the reported jobsatisfaction) to the alternative job quality indicators. The general correlationresults (table 3) are re�ected here by looking at the �gures on the diagonaland by assessing the strength of the shifts between di¤erent categories. Ingeneral, the numbers on the diagonal are small, indicating that the di¤erentmethods yield rather di¤erent rankings of jobs. The low correlation betweenthe subjective indicator and the equal weighting case shows up in table 6 wheremore people shift from one category to another. When comparing the objectiveindicators, we saw that most of the shifts were out of category 3. Here, wesee that most shifts are out of the �rst three categories. This suggests thatespecially (wrong) aspirations make (young) workers more unhappy about theirwork than they should be based on their job characteristics and preferences.

Table 6: Cross tabulation of subjective job satisfaction andobjective indicator using equal weights.

Table 7: Cross tabulation of subjective job satisfaction andobjective indicator with average reference individual.

Table 8: Cross tabulation of subjective job satisfaction and objectiveindicator with reference individual with a high motivation from the

content of work and a low motivation from material aspects.

20

Table 9: Cross tabulation of subjective job satisfaction andequivalent income indicator.

The correlations already indicated that the equivalent income indicator givesa di¤erent classi�cation of jobs compared to the equal weights method andthat the similarities are highest when compared to the objective indicator withaverage reference individual. Tables 10 and 11 show this in more detail. Againwe notice that the di¤erences between the classi�cations of the indicators aresituated more among the lower quality categories than among the higher ones.Individual preferences as used for the equivalent income indicator are closer

to average preferences (table 11) than to more extreme preferences (table 12).Therefore the probability of changing more than one category is higher in thecomparison with the extreme preference indicator. In particular, there is arather high fraction of shifts from category 1 and 2 to category 3 and 4 respec-tively in table 12. 21% of workers saying to be very unsatis�ed are classi�ed ashaving a job of average quality when using an extreme preference scheme, whilethis is only 13% when using average preferences. For those saying to be �ratherunsatis�ed�the shift to the category 4 is twice as large in table 12 (34%) thanin table 11 (17%).

Table 10: Cross tabulation of objective indicator using equal weightsand equivalent income indicator.

Table 11: Cross tabulation of objective indicator with averagereference individual and equivalent income indicator.

Table 12: Cross tabulation of objective indicator with referenceindividual with a high motivation from the content of work and alow motivation from material aspects and equivalent income

indicator.

6 Characteristics of poorest and best jobs

In the previous section we compared the shifts between the job quality categoriesfor di¤erent indicators. To get a better insight into the relevant di¤erences, table13 o¤ers a description of some background characteristics, job characteristicsand the sector of employment for the �poorest� (1) and �best� (5) job qualitycategories according to the di¤erent indicators. The �rst column (�all�) givesthe fraction (of men/women, of the di¤erent education levels, of certain jobtypes and of sector of employment) or averages (for wages and factor scores ofjob characteristics) for the whole sample.

Table 13: Characteristics of poorest and best jobs according to thedi¤erent indicators

21

We see that the �worst� jobs on average provide a lower wage; are morementally and especially physically demanding; are less creative, responsible andvaried; provide a lower level of autonomy; require more shift work; are mostof the time in the private sector and have less learning opportunities. On thecontrary, the �best�jobs are more situated in the public sector; provide higherwages; are less demanding; more creative and varied; give more autonomy andare jobs in which one learns new skills. This general picture holds for all indi-cators.The objective indicator only considers the job characteristics and not the

preferences nor the aspirations of the individual workers that are included inthe subjective responses. This is re�ected in the table, as the dichotomy be-tween poorest and best quality jobs (as described in the previous paragraph) issharper for the objective indicator than for the subjective indicator. The dif-ferent de�nitions of good and bad jobs according to the di¤erent indicators arealso re�ected in the pro�le of the workers occupying these jobs. Lower (higher)educated are more often in jobs with bad (good) characteristics. The equiva-lent income indicator does not depend on aspirations but does take preferencesinto account, resulting in a more mixed picture but still respecting the generaldistinction between good and bad job characteristics.The results of the ordered logit model (table 2) show an aspiration e¤ect of

the educational level (higher educated workers are less satis�ed because theyexpect more from their work, cfr supra). Both the objective and the equiva-lent income indicator correct for the impact of aspirations. Indeed, we observechanges in the distribution over the educational levels both in the poorest andbest jobs. People with a lower (higher) education are found relatively less (more)in a �very good job�with the objective indicator than with subjective job sat-isfaction. Taking into account preferences (as is the case with the equivalentindicator) yields results which are in between these two extremes.

7 Conclusion

�More and better jobs�is one of the key objectives to make the EU �the mostcompetitive and dynamic knowledge-driven economy by the year 2010�. Thisimplies that the measurement of job quality is a very important question forthe purpose of policy. However the multidimensional nature of the conceptraises a di¢ cult indexing problem as it is necessary to aggregate data on awide range of job characteristics. Mainly two approaches have been used in therecent past for constructing an indicator of job quality: objective and subjectiveindicators. The objective indicator approach starts from a set of job dimensionsthat are judged to be relevant by the analyst or by the policy-maker and appliesa weighting to these dimensions. This a priori weighting leads to measures whichare relatively easy to compute and to interpret. However, objective indicatorsneglect the fact that di¤erent individuals have di¤erent ideas about the relativeimportance of di¤erent job characteristics. The second approach uses subjectivejob satisfaction as a one-dimensional proxy for job quality. Yet this may be

22

misleading since job satisfaction does not only re�ect job characteristics butalso individual expectations and aspirations.We proposed a way out of the dilemma between �too objective� and �too

subjective�indicators by introducing the equivalent income indicator which re-spects individual preferences but corrects for the e¤ects of expectations andaspirations.We illustrate the impact of the di¤erent indicators by using data for Flemish

youngsters in their �rst job (SONAR data). It turns out that the ranking of jobsis very di¤erent depending on the indicator used (objective, subjective or equiv-alent income indicator). The dichotomy in job characteristics between poorestand best quality jobs is sharper for the objective indicators than for subjec-tive job satisfaction. The equivalent income indicator, which does not dependon aspirations but takes preferences into account, yields a more mixed picture.A striking example of di¤erent expectations can be found in the educationallevel. A higher educational level increases expectations and so reduces reportedjob satisfaction (Verhofstadt et al., 2007). Therefore, according to the subjec-tive indicator, higher educated will have a job of lower quality. The objectiveand equivalent income indicators rule out this impact. The equivalent incomeindicator has the additional advantage of respecting the preference di¤erencesbetween lower and higher educated workers.

23

REFERENCESBrown, A., A. Charlwood, and C. Spencer D. Forde. 2007. Job quality

and the economics of New Labour: a critical appraisal using subjective data.Cambridge Journal of Economics 31:941-971.

Brun, B. and B. Tungodden. 2004. Non-welfaristic theories of justice: Is the"intersection approach" a solution to the indexing impasse? Social Choice andWelfare 22, no. 1:49-60.

Clark, A. 1997. Job satisfaction and gender: why are women so happy atwork? Labour Economics 4:341-372.

Clark, A. 2001a. Economic foundations of subjective well-being. Universityof Orléans: Working Paper.

Clark, A. 2001b. What really matters in a job? Hedonic measurement usingquit data. Labour Economics 8:223-242.

Clark, A. 2005. Your money or your life: changing job quality in OECDcountries. British Journal of Industrial Relations 43, no. 3:377-400.

Clark, A. and A. Oswald. 1996. Satisfaction and comparison income. Jour-nal of Public Economics 61, no. 3:359-381.

D�Addio, A., T. Eriksson, and P. Frijters. 2007. An analysis of the determi-nants of job satisfaction when individuals�baseline satisfaction levels may di¤er.Applied Economics 39, no. 19:2413-2423.

Davoine, L. and C. Erhel. 2006. Monitoring employment quality in Eu-rope: European employment strategy indicators and beyond. CES: Document deTravail No. 66.

De Witte, H. 2001. Arbeidstevredenheid: een zinvol concept? OverWerk4:18-21.

Delfgauw, J. 2007. The e¤ect of job satisfaction on job search: not justwhether, but also where. Labour Economics 14, no. 3:299-317.

Diaz-Serrano, L. and J. Cabral Vieira. 2005. Low pay, higher pay andjob satisfaction within the European Union: empirical evidence from fourteencountries. IZA: Discussion Paper 1558.

Eurofound. 2006. Measuring job satisfaction in surveys. ERM: ComparativeAnalytical Report.

Fleurbaey, M. 2005. Health, wealth and fairness. Journal of Public Eco-nomic Theory 7, no. 2:253-284.

Fleurbaey, M. 2006. Health, equity and social welfare. Annales d�Economieet Statistique 83-84:21-60.

Fleurbaey, M. 2007. Social choice and the indexing dilemma. Social Choiceand Welfare 29, no. 4:633-648.

Fleurbaey, M. 2008. Fairness, responsibility and welfare. Oxford: OxfordUniversity Press.

24

Fleurbaey, M., E. Schokkaert, and K. Decancq. 2009. What good is happi-ness? CORE: Discussion Paper 2009/17.

Freeman, R. 1978. Job satisfaction as an economic variable. AmericanEconomic Review 68, no. 2:135-141.

Frey, B. and A. Stutzer. 2002. What can economists learn from happinessresearch? Journal of Economic Literature 40, no. 2:402-435.

Furnham, A. 1997. The psychology of behaviour at work. The individual inthe organization. Hove East Sussex: Psychology Press.

Gazioglu, S. and A. Tansel. 2006. Job satisfaction in Britain: individualand job related factors. Applied Economics 38:1163-1171.

Global Policy Network. 2006. Proposal for calculation of the Good JobsIndex (GJI) for all countries. Mimeo.

Green, F. 2006. Demanding work: the paradox of job quality in the a­ uenteconomy : Princeton University Press.

Green, F. and N. Tsitsianis. 2005. An investigation of national trends injob satisfaction in Britain and Germany. British Journal of Industrial Relations43, no. 3:401-429.

Hall, R. 1994. Sociology of work: perspectives, analysis and issues.: PineForge Press.

Hamermesh, D. 2001. The changing distribution of job satisfaction. Journalof Human Resources 36, no. 1:1-30.

Heintz, J., J. Wicks-Lim, and R. Pollin. 2005. Work environment index.Technical Background Paper WP107: Political Economy Research Institute,University of Massachusets Amherst.

Jencks, C., L. Perman, and L. Rainwater. 1988. What is a good job? A newmeasure of labor market success. American Journal of Sociology 6:1322-1357.

Kalleberg, A. and S. Vaisey. 2005. Pathways to a good job: perceived workquality among the machinists in North-America. British Journal of IndustrialRelations 43, no. 3:431-454.

Knoop, F. and R. Schouteten. 2006. Arbeidstevredenheid als maat voorkwaliteit van de arbeid . Tijdschrift voor Arbeidsvraagstukken 22, no. 4.

Landy, F. and D. Trumbo. 1976. Psychology of work behaviour.: DorseyPress.

Leontaridi, R. and P. Sloane. 2001. Measuring the quality of jobs: promotionprospects, low pay and job satisfaction. Amsterdam: LoWER Working PaperNo. 7.

Leschke, J. and A. Watt. 2008. Putting a number on job quality? Construct-ing a European Job Quality Index.: ETUI-REHS WP 2008.03.

Levy-Garboua, L. and C. Montmarquette. 2004. Reported job satisfaction:what does it mean? Journal of Socio-Economics 33, no. 2:135-151.

25

Munoz de Bustillo Llorente, R. and E. Fernandez Macias . 2005. Job sat-isfaction as an indicator of the quality of work. Journal of Socio-Economics34:656-673.

Nussbaum, M. 2000. Women and human development: the capabilities ap-proach. Cambridge: Cambridge University Press.

Nussbaum, M. 2006. Frontiers of justice: disability, nationality, speciesmembership. Cambridge, Mass.: Harvard University Press.

Pattanaik, P. and Y. Xu. 2007. Minimal relativism, dominance, and stan-dard of living comparisons based on functionings. Oxford Economic Papers, no.59:354-374.

Ritter, J. and R. Anker. 2002. Good jobs, bad jobs: worker�s evaluations in�ve countries. International Labour Review 141, no. 4:331-358.

Schokkaert, E. 2007. Capabilities and satisfaction with life. Journal ofHuman Development 8, no. 3:415-430.

Sen, A. 1985. Commodities and capabilities. Amsterdam: North-Holland.

Sen, A. 1999. Development as freedom. New York: Knopf.

Sen, A. 1992. Inequality re-examined. Oxford: Clarendon Press.

Spector, P. 1997. Job satisfaction. Application, assessment, causes, andconsequences. London: SAGE.

Tangian, A. 2005. A composite indicator of working conditions in the EU-15for policy monitoring and analytical purposes. Hans Böckler Stifting, Düsseldorf:WSI Discussion Paper 135.

Tangian, A. 2007. Is �exible work precarious? A study based on the 4thEuropean Survey of working conditions 2005. Hans Böckler Stifting, Düsseldorf:WSI Discussion Paper 135.

Verbeek, M. 2000. A guide to modern econometrics. New York: Wiley.

Verhofstadt, E., H. De Witte, and E. Omey. 2007. Higher educated workers:better jobs but less satis�ed? International Journal of Manpower 28, no. 2:135-151.

26

Table 2. Ordered logit regression results of job satisfaction in the first job

Coeff. Std. Err.

Neth montly wage in FTE 0,545 *** 0,188

Demanding work wrt mental effort en time pressure -0,035 0,047

Creative, responsible and varied job 1,499 *** 0,076

Company more than 50 employees 0,335 *** 0,088

Physically demanding work -0,611 *** 0,072

Autonomy 0,696 *** 0,072

Learning new skills during first job 0,358 *** 0,108

Number of used search channels -0,03 * 0,016

Motivation to work from content 0,137 ** 0,061

Motivation to work from material aspects 0,081 * 0,047

Number of offers 0,029 ** 0,014

Respondent primary education 1,078 *** 0,266

Respondent lower secondary education 0,262 0,178

Respondent higher secondary education (ref)

Respondent lower tertiary education -0,397 *** 0,120

Respondent higher tertiary education -0,683 *** 0,151

Impression respondent x mentally demanding work -0,131 *** 0,047

Locus of control internal x mentally demanding work 0,100 ** 0,044

Motivation from content x creative, responsible job 0,095 ** 0,041

Motivation from content x large company -0,165 * 0,088

Motivation from content x physically demanding work 0,089 ** 0,041

Motivation from material aspects x creative, responsible job -0,137 *** 0,049

Number of offers x physically demanding work 0,042 *** 0,014

Primary education x creative, responsible job 0,491 ** 0,270

Lower secondary education x creative, responsible job 0,029 0,173

Lower tertiary education x creative, responsible job 0,526 *** 0,125

Higher tertiary education x creative, responsible job 0,252 ** 0,141

Locus of control internal x autonomy -0,092 ** 0,044

Number of offers x autonomy -0,033 ** 0,014

* significant at 10% level, ** significant at 5% level, *** significant at 1% level

Number of observations: 2211

Log likelihood = -2205.7623

Table 3: Correlations between the different indicators

QS Q

O,EW Q

O,r1 Q

O,r2 Q

EI,r

QS 1

QO,EW

0,420 1

QO,r1

0,670 0,674 1

QO,r2

0,537 0,821 0,862 1

QEI,r

0,625 0,525 0,823 0,644 1

Note: all correlations are significant at 0.01 level

Table 4: Cross tabulation of objective indicator using equal weights and of objective indicator with average reference individual.

QO,EW / QO,r1

1 2 3 4 5

1 0,43 0,31 0,19 0,07 0,00

2 0,15 0,29 0,29 0,26 0,01

3 0,06 0,20 0,26 0,43 0,05

4 0,01 0,06 0,12 0,59 0,23

5 0,00 0,01 0,03 0,49 0,48

Table 5: Cross tabulation of objective indicator with average reference individual and objective indicator with reference individual with a high motivation from the content of work and a low motivation from material aspects.

QO,r1 /

QO,r2

1 2 3 4 5

1 0,61 0,36 0,03 0,00 0,00

2 0,14 0,40 0,30 0,15 0,00

3 0,02 0,21 0,36 0,41 0,00

4 0,00 0,03 0,11 0,70 0,16

5 0,00 0,00 0,00 0,32 0,68

Table 6: Cross tabulation of subjective job satisfaction and objective indicator using equal weights.

QS / Q

O,EW 1 2 3 4 5

1 0,21 0,31 0,24 0,22 0,00

2 0,12 0,21 0,19 0,40 0,09

3 0,07 0,17 0,23 0,41 0,12

4 0,02 0,06 0,12 0,54 0,27

5 0,00 0,05 0,08 0,52 0,35

Table 7: Cross tabulation of subjective job satisfaction and objective indicator with average reference individual.

QS / Q

O,r1 1 2 3 4 5

1 0,30 0,41 0,18 0,11 0,00

2 0,18 0,30 0,27 0,24 0,00

3 0,06 0,19 0,27 0,45 0,03

4 0,01 0,04 0,12 0,62 0,21

5 0,00 0,01 0,03 0,40 0,56

Table 8: Cross tabulation of subjective job satisfaction and objective indicator with reference individual with a high motivation from the content of work and a low motivation from material aspects.

QS / Q

O,r2 1 2 3 4 5

1 0,26 0,33 0,17 0,24 0,00

2 0,17 0,20 0,26 0,36 0,02

3 0,06 0,18 0,25 0,44 0,07

4 0,01 0,06 0,11 0,58 0,24

5 0,00 0,03 0,06 0,44 0,47

Table 9: Cross tabulation of subjective job satisfaction and equivalent income indicator.

QS / Q

EI,r 1 2 3 4 5

1 0,33 0,31 0,21 0,13 0,02

2 0,18 0,30 0,27 0,21 0,03

3 0,05 0,16 0,24 0,49 0,06

4 0,00 0,06 0,12 0,61 0,21

5 0,00 0,01 0,03 0,43 0,53

Table 10: Cross tabulation of objective indicator using equal weights and equivalent income indicator.

QO,EW /

QEI,r

1 2 3 4 5

1 0,24 0,40 0,20 0,16 0,00

2 0,14 0,19 0,24 0,37 0,05

3 0,06 0,17 0,25 0,46 0,06

4 0,02 0,07 0,12 0,53 0,25

5 0,01 0,01 0,06 0,52 0,41

Table 11: Cross tabulation of objective indicator with average reference individual and equivalent income indicator.

QO,r1 /

QEI,r

1 2 3 4 5

1 0,47 0,37 0,13 0,03 0,00

2 0,17 0,40 0,27 0,17 0,00

3 0,05 0,21 0,34 0,39 0,01

4 0,00 0,03 0,12 0,70 0,15

5 0,00 0,00 0,00 0,32 0,68

Table 12: Cross tabulation of objective indicator with reference individual with a high motivation from the content of work and a low motivation from material aspects and equivalent income indicator.

QO,r2 /

QEI,r

1 2 3 4 5

1 0,34 0,38 0,21 0,05 0,01

2 0,14 0,23 0,26 0,34 0,02

3 0,06 0,18 0,25 0,45 0,06

4 0,02 0,07 0,13 0,58 0,21

5 0,00 0,00 0,02 0,46 0,52

Table 13: Characteristics of poorest and best jobs according to the different indicators

All QS Q

O,r1 Q

EI,r

Variables 1 5 1 5 1 5

Men 50,30 40,00 45,27 51,52 46,69 45,26 57,09

Women 49,70 60,00 54,73 48,48 53,31 54,74 42,91

Respondent primary education 3,49 4,76 4,99 5,05 0,92 6,32 2,49

Respondent lower secondary education 7,63 13,33 6,88 18,18 2,57 4,21 6,51

Respondent higher secondary education 43,61 46,67 35,11 61,62 27,21 35,79 42,91

Respondent lower tertiary education 26,92 19,05 34,42 8,08 41,18 38,95 26,05

Respondent higher tertiary education 18,35 16,19 18,59 7,07 28,13 14,74 22,03

Net monthly wage in FTE 1108 1044 1152 995 1209 1059 1171

Demanding work wrt mental eff en time press 0 0,075 -0,047 0,465 0,011 0,365 -0,091

Creative, responsible and varied job 0 -1,431 0,754 -1,983 1,017 -1,822 0,916

Physically demanding work 0 0,321 -0,137 0,896 -0,390 0,340 -0,174

Autonomy 0 -0,507 0,273 -0,647 0,395 -0,443 0,268

Work in shifts (%) 20,95 31,07 16,08 42,42 9,67 37,89 11,31

Company more than 50 employees (%) 47,83 47,12 48,42 40,40 53,49 47,37 49,81

Private secor (%) 82,25 91,35 72,76 94,95 67,77 92,63 75,48

Parttime job (%) 13,40 20,95 12,05 14,14 11,21 9,47 13,22

Temporary contract (%) 54,39 62,50 52,99 73,74 50,55 63,16 50,00

Formal training during the first job (%) 26,19 13,33 30,64 9,09 35,85 20,00 33,14

Learning new skills during first job (%) 74,28 25,71 83,99 21,21 91,73 29,47 91,00

NACE classification of economic activities

Primary sector 1,46 0,95 1,55 2,02 1,29 0,00 1,15

Manufacturing 21,88 33,33 15,15 44,44 13,24 36,84 18,20

Electricity, gas and water supply or Construction 5,20 0,95 5,34 5,05 4,96 5,26 7,09

Wholesale and retail trade; repair of motor vehicles, motorcycles and personal and household goods 15,10 15,24 11,88 20,20 9,56 13,68 12,64

Hotels and restaurants 4,18 5,71 3,27 7,07 0,92 5,26 4,02

Transport, storage and communication 5,64 3,81 6,88 4,04 4,60 6,32 5,36

Financial intermediation 4,02 5,71 4,30 1,01 5,70 5,26 3,26

Real estate, renting and business activities 13,28 9,52 11,19 7,07 13,97 13,68 13,03

Public administration and defence, compulsary social security 4,38 3,81 5,85 2,02 6,07 3,16 5,36

Education 8,73 3,81 15,66 0,00 23,71 3,16 16,28

Health and social work 11,04 7,62 14,11 1,01 11,95 5,26 8,43

Other community, social and personal service activities or activities of households 4,22 7,62 4,13 4,04 4,04 1,05 4,79

Badly defined activities or missing information 0,85 1,90 0,69 2,02 0,00 1,05 0,38