For One More Year with You : Changes in Compulsory Schooling, Education and the Distribution of Wages in Europe

Transcription

1 For One More Year with You : Changes in Compulsory Schooling, Education and the Distribution of Wages in Europe Margherita Fort Giorgio Brunello and Guglielmo Weber PRELIMINARY WORK European University Institute, Max Weber Post-Doctoral Programme Florence, January 18, 2007 p. 1/31

2 Motivation Large literature on average returns to schooling Less is known on the heterogeneity of the returns Are returns lower or higher for individuals at the lowest quartile of the distribution of earnings? Florence, January 18, 2007 p. 2/31

3 Are returns higher in the lowest quartile? Y = F(A,E) where: Y earnings; A ability; E education If ability and education are complements for individual productivity (and wages), the returns are lower for the less able A Y E > 0 Florence, January 18, 2007 p. 3/31

4 Motivation II Implications for education and inequality Do education reforms which affect compulsory years of education reduce or increase earnings inequality? Difficult question but relevant for policy Florence, January 18, 2007 p. 4/31

5 Our Paper We study the effects of compulsory schooling reforms introduced in Europe in the 1960s and 1970s on the distribution of earnings. By exploiting the exogenous changes in education induced by the reforms, we assess the causal effect of education on the distribution of earnings for individuals whose schooling level changed due to the reforms (compliers) Florence, January 18, 2007 p. 5/31

6 Outline of the Talk Education & Income: Association & Causality Moving On From Average Returns The Identification Strategy in a Nutshell Data Issues Preliminary Results (Intention-to-Treat Parameters) Florence, January 18, 2007 p. 6/31

7 What the other have done Association U.S. : returns to education increase dramatically over the quantiles of the conditional distribution of wages [Buchinsky (2004)]; (ii) Europe: stronger education-related earnings increment for individuals who receive higher hourly wages conditional on their observed characteristics [Pereira & Martins (2004)] Florence, January 18, 2007 p. 7/31

8 What the other have done Causality U.S. : ability & education act as complements in determining wages, after controlling for endogeneity [Arias et al. (2001), data on twins] UK: substitutability between education and cognitive & non-cognitive ability [Denny et al. (2004), controls for cognitive ability] Florence, January 18, 2007 p. 8/31

9 Identification of Endogenous Quantile Treatment Effects (QTEs) Abadie et al. (2002): move from the approach of Angrist & et al. (1996); propose an IV estimation method for QTEs; application to JPTA Chesher (2003): discuss identification in recursive non linear structural model (based on exclusion restriction); estimation developed by Ma & Koenker (2004) (weighted average derivative estimators) and Arias et al. (2001) (control variate approach) Chernozhucov & Hansen (2005, 2006): propose an IV QTE model and estimation (analogue to two stage least squares estimator) Florence, January 18, 2007 p. 9/31

10 Identification of Endogenous Quantile Treatment Effects (QTEs) (continued) Abadie et al. (2002): binary treatment, binary instrument setting; identification and estimation of QTEs for compliers; authors do not consider estimation of the potential outcomes distributions for compliers Chesher (2003): continuous treatment variable & continuous instrument; identification and estimation of QTEs;! Chesher (2005) extension to the case of a discrete treatment variable (interval identification) Chernozhucov & Hansen (2005, 2006): use of marginal independence conditions Florence, January 18, 2007 p. 10/31

11 Main Features of This Project Identification: (Fuzzy) Regression Discontinuity Design (RDD) exploiting exogenous variation in schooling induced by school reforms rolled out in Europe in the 20th century. Data: European Community Household Panel (ECHP, 2001), International Social Survey Programme (ISSP, ), Survey on Health Ageing and Retirement in Europe (SHARE, 2004) and OECD, ILO (participation and unemployment rates, time series by gender ). Florence, January 18, 2007 p. 11/31

12 The Identification Strategy in a Nutshell Hahn et. al.(2001), Card & Lee (2006), Imbens & Rubin (1997) Many European countries increased the minimum school leaving age (MSLA) in 1960s-1970s. Date of birth (randomly assigned) determines whether an individual had to stay longer in school. The causal effect of education of wages is identified for those individuals whose education changed due to the effect of changes in MSLA (compliers). The changes in wages experienced by these individuals are proportional to the causal effect of education on wages provided that (i) they cannot be explained by any other event, except the reform in MSLA; (ii) there are no defiers. Florence, January 18, 2007 p. 12/31

13 Reforms in Minimum School Leaving Age (MSLA), Europe, 20th century 1st cohort Change in Country Reform affected MSLA yrs school. Denmark France Italy Also in Austria (1962), Belgium (1971), Finland (1970s), Germany (1960s), Greece (1975), Ireland (1972), the Netherlands (1975), Portugal (1960s), Spain (1970), Sweden (1962), UK ( ). Positive effect on average years of schooling (+0.3) and imperfect compliance documented for several countries. Florence, January 18, 2007 p. 13/31

14 Wish List (and Bones of Contention ) A (sizeable) micro-data set on several European countries Pooling (surveys, countries) Comparable individuals around the cutoff point Country specific institutional & labour market features, age, period (businness cycle at entry and at the time wages are observed) Data quality Coding issues (wage, education); missing data Florence, January 18, 2007 p. 14/31

15 Empirical Analysis Binary treatment: D 1( years of schooling compulsory years of schooling) Binary instrument: Z 1(T 0),T =cohort- c k, c k year in which the reform was introduced in coutryk Countries with 1-2 years increase in MSLA: Austria, Denmark, France, Germany, Ireland, Netherlands, Portugal, Spain, Sweden Countries with 3-4 years increase in MSLA: Belgium, Finland, Greece, Italy Countries where the change in MSLA was introduced in the late 50- early 60s: Austria, France, Germany, Italy, Portugal, Sweden Florence, January 18, 2007 p. 15/31

16 Empirical Analysis Analysis limited to individuals aged more than 25 who are employed at the time of the interview,t [ 9, +9], i.e. born between 1932 and 1975, aged between 25 and 67 at the time of the interview Dependent variable: logarithm of deflated (2000 prices) gross monthly earnings in purchaising power standards (PPSs) Controls (X): country & gender specific quadratic trend int; survey, country dummies; GDP per capita; employment protection legislation; unemployment rate at the (estimated) time of entry into the labour market and at the time the wage is observed; net wage dummy. Florence, January 18, 2007 p. 16/31

17 Data: Summary Statistics on Key Variables N n t t E[Y] E[D] N n t t E[Y] E[D] A 2, , IR 2, , B 1, , IT 3, , DK 3, , N 3, , FI 2, , PO 1, , FR 2, , SP 3, , GE 3, , SW 3, , GR 1, , LegendNsample size (employed individuals,t [ 9, +9]);n t sample size byt;ydeflated (2000 prices) gross monthly earnings in PPSs,D 1(yearsofschooling compulsoryyearsof schooling) Florence, January 18, 2007 p. 17/31

18 The Effect of MSLA Changes on Education Country group Features E[D Z = 1,X d ] E[D Z = 0,X d ](s.e.) Males Females 1-2 years increase in MSLA & reform 0.03 (0.01) 0.01 (0.01) (A,Dk,Fr,Ge,Ir,Ne,Po,Sp,Swe) 3-4 years increase in MSLA & reform 0.06 (0.01) 0.05 (0.02) (Be,Fi,Gr,It) Employed 1-2 years increase in MSLA & reform 0.01 (0.01) 0.01 (0.01) (A,Dk,Fr,Ge,Ir,Ne,Po,Sp,Swe) 3-4 years increase in MSLA & reform 0.04 (0.02) 0.04 (0.02) (Be,Fi,Gr,It) Florence, January 18, 2007 p. 18/31

19 The Effect of MSLA Changes on Education Country group Features E[D Z = 1,X d ] E[D Z = 0,X d ] 1-2 years increase in MSLA & reform 0.02 (0.007) (A,Dk,Fr,Ge,Ir,Ne,Po,Sp,Swe) 3-4 years increase in MSLA & reform 0.05 (0.012) (Be,Fi,Gr,It) Employed 1-2 years increase in MSLA & reform 0.01 (0.009) (A,Dk,Fr,Ge,Ir,Ne,Po,Sp,Swe) 3-4 years increase in MSLA & reform 0.04 (0.015) (Be,Fi,Gr,It) Florence, January 18, 2007 p. 19/31

20 The Effect of MSLA Changes on Education Treatment: D 1(yearsofschooling compulsoryyearsof schooling) Instrument: Z 1(T 0),T =cohort- c k, was introduced in coutryk Proportion of Individuals with High Qualification (Observed and Predicted ) All countries c k year in which the reform Year of birth Year of birth of first cohort affected by the reform Effect (s.e.) 0.04 (0.01) Observed Predicted (smpl>0) Predicted (smpl<0) Florence, January 18, 2007 p. 20/31

21 The Effect of MSLA Changes on Labour Force Participation Proportion of Employed Women Year of birth Year of birth of first cohort affected by the reform Austria,observed Belgium,observed Denmark,observed Finland,observed France,observed Germany,observed Greece,observed Austria,fitted Belgium,fitted Denmark,fitted Finland,fitted France,fitted Germany,fitted Greece,fitted No clear evidence of effects on the proportion of employed individuals. Florence, January 18, 2007 p. 21/31

22 The Effect of MSLA Changes on Labour Force Participation Proportion of Employed Women Year of birth Year of birth of first cohort affected by the reform Ireland,observed Italy,observed Netherlands,observed Portugal,observed Spain,observed Sweden,observed UK,observed Ireland,fitted Italy,fitted Netherlands,fitted Portugal,fitted Spain,fitted Sweden,fitted UK,fitted No clear evidence of effects on the proportion of employed individuals (exception: UK). Florence, January 18, 2007 p. 22/31

23 The Effect of MSLA Changes on Labour Force Participation Proportion of Employed Men Year of birth Year of birth of first cohort affected by the reform Austria,observed Belgium,observed Denmark,observed Finland,observed France,observed Germany,observed Greece,observed Austria,fitted Belgium,fitted Denmark,fitted Finland,fitted France,fitted Germany,fitted Greece,fitted No clear evidence of effects on the proportion of employed individuals. Florence, January 18, 2007 p. 23/31

24 The Effect of MSLA Changes on Labour Force Participation Proportion of Employed Men Year of birth Year of birth of first cohort affected by the reform Ireland,obs. Italy,obs. Netherlands,obs. Portugal,obs. Spain,obs. Sweden,obs. UK,obs. Ireland,fit. Italy,fit. Netherlands,fit. Portugal,fit. Spain,fit. Sweden,fit. UK,fit. No clear evidence of effects on the proportion of employed individuals. Florence, January 18, 2007 p. 24/31

25 Controlling for Differences in (Pre-Reform ) Observables over the Distribution of Earnings ỹ i = y i β(τ 2 )X i if Q Y (τ 1 X i,z i ) y i Q Y (τ 2 X i,z i ), where Q Y ( X i,z i ) estimated conditional quantile τ 1 τ 2 Pre reform Log(earnings) Uomini Blue: Denmark; Red: Spain Residuals Uomini Blue: Denmark; Red: Spain Percent Percent Log(wage in PPS) Generalized Residual Percent Percent Percent Percent Florence, January 18, 2007 p. 25/31

26 The Effect of MSLA Changes on Distribution of Earnings Belgium Cumulative distribution function at decile 0.20 Belgium Cumulative distribution function at decile 0.90 Belgium Year of birth Year of birth of first cohort affected by the reform Year of birth Year of birth of first cohort affected by the reform Observed (mean) avg_f_20 Predicted Fitted values Observed (mean) avg_f_90 Predicted Fitted values Florence, January 18, 2007 p. 26/31

27 The Effect of MSLA Changes on Distribution of Earnings Finland Cumulative distribution function at decile 0.20 Finland Cumulative distribution function at decile 0.90 Finland Year of birth Year of birth of first cohort affected by the reform Year of birth Year of birth of first cohort affected by the reform Observed (mean) avg_f_20 Predicted Fitted values Observed (mean) avg_f_90 Predicted Fitted values Florence, January 18, 2007 p. 27/31

28 In Short... Empirical evidence on the role of education in reducing income inequality is ambiguous This study seems to be the first trying to tackle the issue in the case of Europe, exploiting changes in MSLA introduced in the 20th century Preliminary results suggest that changes in MSLA : increase the education level of individuals in countries where the change in MSLA was 3/4 years reduce the proportion of individuals at the lower deciles in some countries has generally negligible effects on the proportion of individuals at the higher deciles Florence, January 18, 2007 p. 28/31

29 What next? Robustness checks (measure of earnings; set of countries considered; reforms considered) Discussion of the internal validity of the identification strategy Assessment of the causal effects of education on the distribution of earnings for compliers Florence, January 18, 2007 p. 29/31

30 Data: Summary Statistics on Key Variables All Individuals E[Z = 0] E[Z = 1] E[Z = 0] E[Z = 1] Austria Ireland Belgium Italy Denmark Netherlands Finland Portugal France Spain Germany Sweden Greece Florence, January 18, 2007 p. 30/31

31 Data: Summary Statistics on Key Variables Employed Individuals E[Z = 0] E[Z = 1] E[Z = 0] E[Z = 1] Austria Ireland Belgium Italy Denmark Netherlands Finland Portugal France Spain Germany Sweden Greece Florence, January 18, 2007 p. 31/31