20
GENDER GAPS AND LIFETIME INEQUALITY: AN EMPIRICAL ANALYSIS OF MICRO-DATA FROM EUROPE 1 PLATON TINIOS & ANTIGONE LYBERAKI University of Piraeus Panteion University GREECE GREECE IFA Conference, Prague May 2012

5 tinios gender ifa-tinios-2012

  • Upload
    ifa2012

  • View
    137

  • Download
    2

Embed Size (px)

DESCRIPTION

 

Citation preview

Page 1: 5 tinios gender ifa-tinios-2012

GENDER GAPS AND LIFETIME INEQUALITY:

AN EMPIRICAL ANALYSIS OF MICRO-DATA FROM EUROPE

1

PLATON TINIOS & ANTIGONE LYBERAKI University of Piraeus Panteion University GREECE GREECE

IFA Conference, Prague May 2012

Page 2: 5 tinios gender ifa-tinios-2012

A new –lifetime - perspective on an old issue

• Attempt to examine empirical implications of a novel and innovative micro dataset on an old issue of importance for gender balance

• The old issue: ‘the’ Gender gap ▫ Systematic differences in life chances between men & women ▫ Observed in many dimensions, across countries, domains etc.

• The new data: SHARELIFE ▫ SHARE {w1 (2004) w2 (2007)} – European interdisciplinary

panel of people 50+ (comparable to HRS in US, ELSA in UK)) ▫ SHARELIFE (2009) – Retrospective lifetime questions for SHARE sample:

Childhood, education, health, family, work, etc for entire life.

• The new perspective ▫ How are gender gaps generated and how do they solidify in older

ages? Cross-country, cross-cohort, various dimensions Key (ultimate) motivation: Palliative role of welfare states??

▫ A tour d’ horizon - attempt to quantify effects of different dimensions Clarify issues – See the ‘lie of the land’

2

Page 3: 5 tinios gender ifa-tinios-2012

Outline: Halfway there…

1. Gender gaps and the life time perspective

▫ Conceptual discussion.

2. Derive alternative measures of the gender gap appropriate to older populations aged 50+

3. Vector of starting deprivation into scalar index of starting position

4. Examine socioeconomic mobility patterns

5. Attempt to disaggregate gender gap by a means of a reduced form equation

3

Page 4: 5 tinios gender ifa-tinios-2012

What is the gender gap? • Gender gap is an achievement gap

▫ women are underpaid, undervalued and overworked. ▫ Gender inequities in own income – economic independence deficit ▫ Why doesn’t wage gap lead to rise in demand?

• In general gender gaps are shrinking over time • What does it depend on?

▫ Observables (education etc) / discrimination ▫ Oaxaca (1973) method attempts to decompose

But: occupational segregation complicates Bergmann (1974) overcrowding depresses wages

• In career terms – snowball effect of low participation, few hours, lower wages ▫ Evidence of polarisation among women ▫ Cumulative earning gap (Luxembourg Income Study) ▫ Could solidify further once retire from the labour market

• SHARELIFE should provide data to examine issues as well as a new perspective:. ▫ E.g. interdisciplinary insights + effect of Welfare State.

4

Page 5: 5 tinios gender ifa-tinios-2012

Conceptual issues

• GENDER GAPS: ▫ Multidimensional. Prominence to remuneration ▫ Here: Cumulative (over life) or end-state.

• Research strategy: DETERMINANTS ▫ Split into different stages of life: Initial, education, work/family, pension, health, Social protection as a palliative influence through the life course

‘Worlds of welfare capitalism’ -- chart the heyday of the Welfare Staete

• 1st approach: Examine total effect on end- result ▫ OLS of variables on end state Gender gap.

≅ Reduced form. Gives overview of total significance, though not causal (yet)

▫ Could examine each stage in turn as stages cumulate over the life course, (plus partial attempts at amelioration) Approach adopted in FRB (initial and work/family).

Axel Börsch-Supan et al 2011. Ongoing work. E.g. early health⇒ Education⇒ Work chances⇒ End-state

5

Page 6: 5 tinios gender ifa-tinios-2012

How to define an over-50 Gender gap? • Gender gap usually identified as difference in wages and/or

earnings. ▫ End-state to be examined relatively well-defined

• In a sample of people 50+, situation is more nuanced: ▫ Some have never worked Depending on conditions some decades ago.

▫ Earnings exist for those who work. ▫ Pensions cumulate past differences and correct depending on welfare state structure and parameters of the

preceding periods ▫ Selection between earnings/pensions endogenous

Depends on personal preferences and welfare state parameters. (retirement ages)

• In household level micro-data: ▫ Savings also cumulate – income from property, rents, business Categories of income accrue to household. (some social assistance)

Equivalence scales force gender equality by definition!

6

Page 7: 5 tinios gender ifa-tinios-2012

Two alternative courses of action

I. Define hybrid ‘Personal income’ ▫ Personal Income= Personal Income from Pensions +

Personal Income from Employment + Equivalent income from other household level income sources Not so much an ‘earnings gap’ but a ‘disposition of resources

gap’ ▫ Wider than gender gap -- cumulative effect over time ▫ In addition to earnings encompasses other disadvantages ▫ Narrower than gender gap --intra-household sharing Rich housewives still rich, if household is rich

II. Decompose into (a) participation gap (b) retirement gap (c) earnings gap for those working (d) pensions gap for pensioners/ retired.

▫ Richer diagnosis – less easy to interpret. Two discrete choices – two alternative earnings

determinations ▫ More complex as a description

7

Page 8: 5 tinios gender ifa-tinios-2012

Overview of empirical work presented

1. Examination of end-state gaps • Under the two alternative definitions

2. Social Mobility analysis ▫ Derivation of initial state ‘deprivation index’ ▫ From initial conditions to end-conditions.

3. Reduced form OLS equation explaining end-gap*** ▫ (Equivalent to Oaxaca-type study) ▫ Groups of variables ‘Pure gender’ effect – dummy Effect of initial conditions Education, work/family, pension

▫ Quantify effect of different groups using predicted values for country groups and cohorts For average values For the poorest 20%

8

Page 9: 5 tinios gender ifa-tinios-2012

1: Crude gender gaps in 2007/9:

Personal income, 50+ population

9

1619

33

38

33 3437

35

39

45 46

13

27

17

23

3942

33

39

48

4143

46

18

50

21

0

10

20

30

40

50

SE DK NL DE BE FR CH AT IT ES GR PL CZ

Gender Gap based on Mean Personal Income

Gender Gap based on Median Personal Income

Gap= 1- (Average for women/ Average for men) Highest in South; smaller in North/Transition Median gap – indication significant difference in distribution by gender Gender gap widest in group 65-80. 80+ falls (influence of widows’ pensions)

Page 10: 5 tinios gender ifa-tinios-2012

Personal Income distribution by gender:

Two extremes GR, SE)

10

010

2030

0 5000 10000 15000 20000 0 5000 10000 15000 20000

male female

Greece Greece

Perc

ent

personal incomeGraphs by male or female

05

1015

0 10000 20000 30000 40000 0 10000 20000 30000 40000

male female

Sweden Sweden

Perce

nt

personal incomeGraphs by male or female

Gini M=0.437 F=0.574

Gini M=0.330 F=0.342

High prevalence of zero incomes -- non-declaration? Non-working spouse in employee or pensioner household?

Page 11: 5 tinios gender ifa-tinios-2012

Detailed participation and separate

earnings and pension gaps

Participation gap Earnings and pensions gap

(for those with +ve values)

11

0

10

20

30

40

50

SE DK NL DE BE FR CH AT IT ES GR PL CZ

Labour income Pension income

Pension or labour income

1,2 1,2

9,9 9,5

16,4

6,7 7,8

13,8 16,8

34,4

19,4

5,6

21,8 21,4

34,8

40,7

24,2

35,0

43,1

30,2

38,8

32,9 34,6

27,3

22,5

0

5

10

15

20

25

30

35

40

45

SE DK NL DE BE FR CH AT IT ES GR PL CZ

Gender Participation Gap (M-W) to Personal Income sources (in percentage points)

Gender Earnings Gap (among those receiving either income source)

Page 12: 5 tinios gender ifa-tinios-2012

Comparison of more familiar concepts:

earnings and pension gaps

• Define for two groups relatively close in age. ▫ Earnings 50-64 ▫ Pensions 65-80

• Could also be seen as ‘a look into the future of the pension gap’.

• No clear pattern ▫ Earnings>Pensions

DK, BE,CH,AT,CZ, ES ▫ Pensions> Earnings

FR,IT,GR, PL ▫ Almost the same

SE,DE

• Need to understand differences

12

0

10

20

30

40

50

60

SE DK NL DE BE FR CH AT IT ES GR PL CZ

Gender Income Gap by income source among

those receiving each source

Labour income: Persons 50-64

Pension income: Persons 65-80

Pension and labour income: Persons 50+

Page 13: 5 tinios gender ifa-tinios-2012

2. Mobility analysis

• Childhood deprivation index ▫ Described in Lyberaki, Tinios, Georgiadis 2011

Related to end-state persistent poverty + soc. protection Index of relative deprivation by country at age 10. Weight more deprivation of those qualities more widely enjoyed

Constructed by 11 indicators (housing, family indicators)

• Gives an idea of starting point ▫ As chiefly family deprivation - gender balanced. ▫ Cohort sensitive; though ▫ However, the two sexes have different chances to alter

personal position • Chart personal changes in rank

▫ Change of quintile in distribution ▫ Spearman rank order coefficients

13

Page 14: 5 tinios gender ifa-tinios-2012

Mobility – Change in

quintile ranks

Considerable change over time

From starting point of gender equality (household status same for brother & sister)

To end point

Males move up

Females down

Biggest difference in South

Spearman rank-order coefficient (on percentiles)

Biggest difference in Continent

Smallest in North

14

51,338,6

53,5

32,6

53,1

32,346,3 41,4

25,036,0

21,2

42,1

22,6

43,231,3 36,9

0

25

50

75

100

M F M F M F M F

Nordics Continental Southern Eastern

Upward mobility Remained stable Downward mobility

Page 15: 5 tinios gender ifa-tinios-2012

3. Description of Reduced form OLS

equation • Dependent: Log (Personal income in 2007/9) • ‘Pure gender effect’ = Gender dummy • ‘Indirect gender effect’ = through gender differences in other

determinants of income ▫ (Assuming as 1st approx. coefficients same for men and women).

1. Initial effects – Cohort, Index, + GNP pc at 1960. 2. Education – years of schooling, higher dummy 3. Family status – Never married, widowed, divorced, married,

Number of children 4. Work – years, years squared 5. Health – bad health during life, at end 6. Pensions - pensioner, years since pension 7. Country group dummies. Log of GDP pc in 1970 Approach similar to Oaxaca decomposition - proceed step by step

15

Page 16: 5 tinios gender ifa-tinios-2012

Pooled equation Pooled equation – presumes independent variables operate in the same way for men and women.

Use as first approximation

Reasonable fit

Intuitive results

“pure gender effect” – implies a gap of 33% if all else is equal.

Employment crucial.

All stages appear to have to ‘expected’ effect

Country dummies included in lieu of social protection. Effect not easy to interpret.

16 Y=Log(Personal income) Pooled

Variables Coef. Std. Err.

Age: 50-64 years 0,4631** 0,0430

Age: 65-79 years

Age: 80 years 0,1807** 0,0374

Female -0,4049** 0,0355

Childhood non deprivation index:

0 to 1 (no deprivation) 0,6276** 0,1243

Years in education 0,0432** 0,0048

Dummy: Higher Education 0,4067** 0,0431

Single 0,0743 0,0652

Married

Divorced 0,0444 0,0639

Widowed 0,5488** 0,0375

Number of Children 0,0207* 0,0109

Years in employment 0,0485** 0,0041

Squared term: Years in employment -0,0005** 0,0001

Dummy: Pensioner 1,5025** 0,0497

Years in retirement -0,0090** 0,0013

Ever had physical injury to

disability -0,0850* 0,0454

Less than good health -0,1630** 0,0347

Country-specific

Nordic

Continental -0,0936** 0,0343

Southern -0,2374** 0,0610

Transition -0,1134 0,1496

Log of GDP per capita 1970 0,8777** 0,0729

Constant term -1,6337* 0,7340

# Observations 23113

R2 0.323

Page 17: 5 tinios gender ifa-tinios-2012

OLS equation by

gender By gender– presumes variables operate differently for men and women.

Obviously so.

Higher explanatory power for women.

Initial conditions less important

Education more

Employment crucial (non-linear for W)

Some variables have opposite sign M/F. (family variables)

Some evidence of dampening by group – women benefit more from initial high GDP.

17 Y=Log(Personal income) Males Females

Variables Coef.

Std.

Err. Coef. Std. Err.

Age: 50-64 years 0,2686** 0,0513 0,4965** 0,0630

Age: 80 years 0,0898 0,0513 0,2089** 0,0491

Childhood deprivation

index: 0 to 1 0,9018** 0,1647 0,3925** 0,1772

Years in education 0,0327** 0,0067 0,0513** 0,0065

Dummy: Higher Education 0,3104** 0,0565 0,4787** 0,0627

Single -0,2142** 0,0915 0,2643** 0,0931

Married

Divorced -0,4089** 0,1078 0,3102** 0,0762

Widowed 0,1414** 0,0548 0,5698** 0,0450

Number of Children 0,0182 0,0148 0,0210 0,0147

Years in employment 0,0204** 0,0081 0,0452** 0,0052

Squared employment -0,0001 0,0001 -0,0004** 0,0001

Dummy: Pensioner 1,0146** 0,0707 1,8432** 0,0691

Years in retirement -0,0159** 0,0024 -0,0066** 0,0015

Ever had physical injury to

disability -0,1039* 0,0554 -0,0498 0,0711

Less than good health -0,1510** 0,0439 -0,1713** 0,0496

Country-specific

Continental -0,0315 0,0452 -0,1238** 0,0486

Southern -0,2944** 0,0868 -0,1453 0,0828

Transition -0,5211** 0,1986 -0,1327 0,2132

Log of GDP per capita 1970 0,6701** 0,0973 1,0112** 0,1031

Constant term 1,3380** 0,9899 -3,4974** 1,0391

# Observations 10404 12709

R2 0.245 0.339

Page 18: 5 tinios gender ifa-tinios-2012

First results at decomposing differences

• Use common Loading factors (pooled equation) i.e. no separate effect, (as in Neumark (1988) – non discrimination). How different are explanatory variables vs ‘pure’ gender effect (gender dummy)

• Look at effect of different ‘endowments’ in four country groups. • Leave different loading factors for future (> complex interpretation) as in

usual Oaxaca models. • Results considerably different by country group and cohort.

▫ Most equalising effect in Nordics

18

12

2733

1825

3631 35

48

1623 24

30,3

24,9

53,6

38,936,0

44,4

24,5

21,9

44,3

30,5

12,7

35,5

0

10

20

30

40

50

60

50-

64

65-

80

80+ 50-

64

65-

80

80+ 50-

64

65-

80

80+ 50-

64

65-

80

80+

Nordics Continental Southern Transition

% explained by differences in employment by differences in education

Page 19: 5 tinios gender ifa-tinios-2012

Separate equations for work and

pensions: preliminary observations In the earnings equation: • Family variables are very significant but of opposite sign by gender. • Favourable initial conditions affect men more than women • Wider differences M/W in effect of initial conditions • Years in employment has no apparent influence. Education has a very strong

influence. • Being in a rich country affects men more than women • Collecting a pension depresses income from earnings In the pensions equation: • Overall gender effects are comparable to earnings. • Women have systematically lower pensions than men in the Continental countries

(social insurance systems?). In Transition countries (ceteris paribus) women are better off.

• Education is far more important for women (due to participation effects?). • Number of children exerts a strong negative effect on women – presumably

accounting for dropping out of the labour market and other constraints on working. • Early deprivation has smaller effect, confined to men. It appears that the social

protection system to some extent corrects for initial disadvantage. • Years of employment have a non-linear effect which diminishes with years • Early retirement is associated with lower pensions.

19

Page 20: 5 tinios gender ifa-tinios-2012

“There’s gold in them thar hills…” (Virginia gold rush – 1840s’)

• First results are encouraging • Long way in a short space. Attempt to be

parsimonious with numbers!! • There is much to be explained + much still left out.

▫ Easy to lose the wood from the trees. • The wood: • The exercise using retrospective data of the

European 50+ population may chart the success or failure of social protection system.

• How are today’s pension gaps related to past events? What are the prospects for the future??

20