Does Unobserved Heterogeneity Matter A Panel Data Analysis

Does Unobserved Heterogeneity Matter? A Panel Data Analysis of the Gender Pay Gap Amynah Gangji, Kristian Orsini and Salimata Sissoko Dulbea (Université Libre Bruxelles) DULBEA 1

Outlines Ø This paper provides evidences on the effect of controlling for unobserved individual heterogeneity in estimating the (un)adjusted gender pay differentials. Ø Using the European Community Household Panel (ECHP), we present a cross-country comparison of the gender pay gap over time (1994 -2001) based on cross-section and panel data estimation techniques. DULBEA 2

Data base The European Community Household Panel (ECHP) * Annual interview of a representative set of European households. Length: 1994 -2001 * Covers many topics, subdivided into 2 groups: - Data related to households (financial situation, housing conditions, durable consumption, children, …) - Data related to individuals (economic activities, income, education and training, health, social relations, migrations, satisfaction with various aspects of life, etc. ) => valuable source of information: allow countries comparison thanks to a standardised methodology as well as to study changes over time at the micro level. DULBEA 3

Sample *Countries: Belgium, Denmark, Ireland, Italy, United Kingdom, Germany, Spain, Ireland, the Netherlands *Selected individuals: Wage-earners aged between 20 to 60 years, working in the private sector and not self-employed BE: 7. 226 observations and 2001 individuals GE: 24. 896 observations, 5759 individuals *Years: 1994 -2001 *Unbalanced panel data (BE: 36%; GE: 33%; UK, IT, DK: 30%) DULBEA 4

Countries Ø The select countries differ according to their welfare state model: üThe conservative-corporatist model (Belgium, France, Germany, Italy, The Netherlands and Spain) üThe Scandinavian model (Denmark) üThe liberal welfare model (Ireland UK). Furthermore, we can identify Italy and Spain as representative of the Mediterranean mode DULBEA 5

Countries Ø According to Gornick they also differ regarding the level of public support to child care arrangements, maternity and parental leave provision. üHigh support (Denmark, France and Belgium) üMedium support (West-Germany, Italy and the Netherlands) üLow support (Ireland, Spain and United. Kingdom) DULBEA 6

Methodology Wage equations Estimated model: ln Wit = β 0 + β 1 Xit + β 2 Zi + αi + εit Xit: time-varying explanatory variables, Zi : time-invariant explanatory variables, αi : unobserved heterogeneity term constant over time. E(αi)= 0, σ²α εit: error term E(εit)= 0, σ²ε, uncorrelated with X, Z and α DULBEA 7

Methodology How to estimate the wage equation? • Ordinary Least Square (OLS) bias since do not take into account the unobserved heterogeneity term • Fixed effects model (FE) unbiased estimators but inefficient since do not take into account time-invariant variables • Random Generalised Least Square (GLS) only if E(αi | Xit, Zi) = 0 (Hausman test) • Random Generalised Least Square with instrumentation (IV/GLS) unbiased and efficient estimators DULBEA 8

A. Gangji, K. Orsini and S. Sissoko Why Are Inequalities so Unequal? Methodology Wage equations 1. Fixed effects model: ln Wit - ln Wi. = β (Xit - Xi. )+ (εit - εi. ) where Wi. = (1/Ti) , Xi. = (1/Ti) 2. Computation of the mean individual residuals 3. Between effect model If E(αi | Xit, Zi) 0 =>Instrumentation of Zi Proper instruments: Xi. strongly correlated with the Zi but uncorrelated with αi DULBEA 9

A. Gangji, K. Orsini and S. Sissoko Why Are Inequalities so Unequal? Methodology Wage equations 4. Computation of variances estimations to compute weights. where σ²ε is the residual variance from the fixed effects model and σ²be the residual variance from the between regressions. 5. Estimation of the GLS equation Yit – θi Yi. = (Xit – θi Xi. )β + (1 - θi ) Zi γ + (εit - θi ε i. ) DULBEA 10

A. Gangji, K. Orsini and S. Sissoko Why Are Inequalities so Unequal? Methodology Hausman test Null hypothesis: E(αi | Xit, Zi) = 0 Distributed as a chi-squared with a degree of freedom equal to the number of instruments used minus the number of time-invariant instrumented variables. If the test reject the null hypothesis, need to use new combination of instruments until the test accept the null hypothesis. DULBEA 11

Average gender pay gap (1994 & 2001) DULBEA 12

A. Gangji, K. Orsini and S. Sissoko Why Are Inequalities so Unequal? Estimation Variables used • • • Actual tenure <1 year of tenure, 1 -5 years, 6 -10 years (reference), 11 -15 years, 16 years and more. potential experience & potential experience squared: =age-age when started first job - tenure Type of contract: - permanent employment (reference) - fixed-term, short-term contract, casual work with no contract DULBEA 13

Why Are Inequalities so Unequal? A. Gangji, K. Orsini and S. Sissoko Estimation Variables used • Occupations - ISCO 1 -DIGIT: 1. Legislators senior officials & managers 2. Professionals 3. Technicians & associate professionals 4. Clerks (reference) 5. Service workers&shops & market sales workers 6. Skilled agricultural&fishery workers 7. Craft and related trades workers 8. Plant & machine operators & assemblers 9. Elementary occupation DULBEA 14

Estimation Variables used • Industries - ISCO 1 -DIGIT: 1. Mining and quarrying (C) 2. Manufacturing (D) (reference) 3. Electricity, gas and water supply (E) (reference) 4. Construction (F) 5. Wholesale and retail trade; repair of motor vehicles, motorcycles and personal and household goods (G) 6. Hotels and restaurants (H) 7. Transport, storage and communication (I) 8. Financial intermediation (J) 9. Real estate, renting and business activities (K) DULBEA 15

Why Are Inequalities so Unequal? A. Gangji, K. Orsini and S. Sissoko Estimation Variables used • Enterprise’s size: • Medium size (100 -500 employees) Large size (>500 employees) Level of education: Small size (1 -99 employees) (reference) Education low (ISCED 0 -2)(reference) Education medium (ISCED 3) Education high (ISCED 5 -7) • Job status Supervisory Intermediate, Non-supervisory (reference) DULBEA 16

DULBEA 17

DULBEA 18

DULBEA 19

DULBEA 20

DULBEA 21

Results Ø Strong positive effect of education upon wage üBE, SP, UK: men better remunerated üFR, GE, IT: women better remunerated for an upper secondary education, üIRL: women better remunerated for tertiary education üDK: women better remunerated Positive influence of experience: concave relation Ø For all countries, except Spain, the return of experience is larger for men. Ø DULBEA 22

Results An additional year of experience leads to from about 1. 5% higher wage in Belgium, Denmark, France and Spain to 2. 1 -2. 6% in Germany, Ireland, the Netherlands and UK Ø Wage increases with level of tenure. Ø The variables relative to job status (supervisory), size of the company and type of contract show, when significant, a positive relation between wage and degree of supervision of an employee, the size of firm or the permanent nature of an employment contract. . Ø DULBEA 23

Conclusion Ø Controlling for individual heterogeneity, we observe an increase of the male-female differential, the rates of the return of experience and education as well as a reduction of penalty due to low tenure, low skilled occupations, fixed employment contract and relatively small enterprise size. DULBEA 24