ONLINE APPENDIX:
Combining Parenthood and Work:
Transmission Channels and Heterogeneous Returns to Early Public Childcare
Review of Economics of the Household
Eric Schuss
*BIBB
Mohammed Azaouagh
University of Duisburg-Essen
1 Outline
This Online Appendix provides additional statistics and empirical results that supplement the main paper. In Appendix A, we provide additional descriptive statistics about the process of merging individual level data and spatial data at the municipality level and display the propensity score by socioeconomic characteristics. In Appendix B, we provide additional empirical results from the estimation of marginal treatment effects and show further technical robustness checks.
Appendix A Additional Information Regarding Data and De- scriptive Statistics
A.1 Merging Individual Level Data and Spatial Data at the Municipality Level
Because the empirical analysis of the main paper is restricted to mothers with young children, it is possible that the final sample is selective regarding residential choices and regions in Germany.
As a consequence, it is possible that the final sample at the individual level and the spatial data at the municipality level are not comparable regarding childcare coverage. To support the argu-
*
All correspondence to Eric.Schuss@bibb.de, Federal Institute for Vocational Education and Training (BIBB),
Robert-Schuman-Platz 3, 53175 Bonn, Tel.: +49 (228) 107 1937
ment that this is not the case, Figure A.1 compares the distribution of childcare coverage of the final sample and the distribution of the spatial data at the municipality level in 2002 and 2010.
Despite some minor deviations, the figure indicates that the two distributions are comparable.
(a) Merged Data at the Individual Level
2002 2010
(b) Spatial Data at the Municipality Level
2002 2010
Figure A.1 – Initial and Post-Reform Distribution of Childcare Coverage (in Percent) in Individual Level Data and Spatial Data at the Municipality Level
Note: Figure A.1 displays the distribution of initial childcare coverage in 2002 and childcare coverage in 2010. Panel a shows the distributions for the final sample, in which individual SOEP data and spatial data at the municipality level are merged.
Panel b displays the distributions for spatial data at the municipality level.
Source: SOEP; Statistical Offices of the German Federal States; Statistical Office of the Federal Employment Agency; own illustration.
A.2 Propensity Score by Socioeconomic Status
In Figure A.2, the propensity score is plotted on childcare coverage to illustrate differences by education, relationship status, and ethnic origin in the utilization of public childcare.
Panel a underlines that the interaction of education background and relationship status mat- ters. While the relationship between the propensity score and childcare coverage is the steepest for highly educated and lone mothers, the slope is distinctly less steep for lone mothers who are low educated.
In Panel b, the predicted probability of utilizing public childcare of natives and Union migrants is almost the same in the range of coverage between zero and 25 percent. However, there is a large gap from non-Union mothers to both natives and mothers from another country of the EU. Thus, the gap in childcare utilization between natives and foreigners is mainly driven by immigrants from third countries.
(a) By Education and Relationship (b) By Ethnic Origin
Figure A.2 – Propensity Score by Socioeconomic Status
Note: The figures plot the propensity score on childcare coverage. The calculation of the propensity score is based on probit estimations of the first stage (equation (2) of the main text). In the first stage, the treatment dummyDit is regressed on childcare coverage, on the interaction between coverage and the post-period time dummyP ostt, dummies for survey years, dummies for federal states, further spatial characteristics given in Table 3 of the main text, and covariates summarized in Table 2 of the main text.
Source: SOEP; Statistical Offices of the German Federal States; Statistical Office of the Federal Employment Agency; own illustration.
Table B.1 – Parameter Estimates: Marginal Returns to Early Public Childcare
(1) Employed (2) Full-Time Employed (3) Part-Time Employed
Panel A: Women’s Socioeconomic Characteristics:
Age 0.005
∗∗(0.002) 0.002 (0.001) 0.003 (0.002)
Married
−0.096∗∗(0.045) -0.034 (0.026)
−0.055∗∗(0.028)
Lone Parent -0.078 (0.067) -0.004 (0.034) -0.069 (0.052)
German Citizenship
−0.032∗(0.018) -0.025 (0.016) -0.006 (0.027)
Foreign-Born 0.014 (0.054) 0.001 (0.039) 0.021 (0.042)
Non-EU -0.024 (0.058) -0.009 (0.040) -0.017 (0.041)
Schooling Years 0.026
∗∗∗(0.007) 0.009
∗∗(0.004) 0.015
∗∗∗(0.006) Pre-Birth Employed 0.116
∗∗∗(0.021) 0.014 (0.013) 0.104
∗∗∗(0.021) Having at Least One Boy
−0.079∗∗∗(0.021) -0.002 (0.012)
−0.074∗∗∗(0.022) under the Age of Three
Min. 2 Children Aged
<3 0.054 (0.107) 0.004 (0.056) 0.0566 (0.088) Further Child Aged 3 - 6
−0.093∗∗∗(0.024)
−0.046∗∗∗(0.012)
−0.042∗∗(0.018) Further Child Aged 7 - 10
−0.096∗∗∗(0.021)
−0.030∗(0.018)
−0.061∗∗∗(0.021) Further Child Aged 11 - 15 -0.029 (0.033)
−0.041∗(0.022) 0.013 (0.032) Further Child Aged
≥16 -0.014 (0.062) 0.030 (0.033) -0.050 (0.066)
Panel B: Interactions with PS:
Age*PS -0.005 (0.013) -0.009 (0.008) 0.006 (0.011)
Married*PS 0.107 (0.173) 0.137 (0.123) -0.032 (0.118)
Lone Parent*PS -0.058 (0.265) -0.094 (0.148) 0.051 (0.190)
German Citizenship*PS 0.153 (0.130) 0.091 (0.103) 0.064 (0.128)
Foreign-Born*PS -0.033 (0.222) 0.148 (0.153) -0.193 (0.180)
Non-EU -0.124 (0.262) -0.046 (0.161) -0.115 (0.200)
Schooling Years*PS 0.038 (0.033) -0.012 (0.020) 0.0531
∗∗(0.024)
Pre-Birth Employed*PS 0.127 (0.115) 0.085 (0.067) 0.052 (0.110)
Having at Least One Boy 0.289
∗∗∗(0.107) -0.075 (0.077) 0.354
∗∗∗(0.116) under the Age of Three*PS
Min. 2 Children
<3*PS -0.714 (0.873) -0.017 (0.532) -0.681 (0.686) Further Child Aged 3 - 6*PS 0.045 (0.118) 0.046 (0.077) -0.012 (0.098) Further Child Aged 7 - 10*PS 0.313
∗∗(0.133) 0.096 (0.123) 0.214 (0.152) Further Child Aged 11 - 15*PS 0.006 (0.230) 0.266 (0.172) -0.261 (0.229) Further Child Aged
≥16*PS -0.454 (0.387)
−0.361∗∗(0.148) -0.066 (0.389)
p1 0.001 (2.272) 0.001 (0.803) 0.001 (1.453)
p2 1.646 (1.696) 1.279 (1.415) 0.318 (1.415)
p3 -1.860 (1.608) -0.216 (1.354) -1.625 (1.258)
Additional Controls Yes Yes Yes Yes Yes Yes
ATE -0.348 (0.617) 0.653 (0.429)
−0.978∗(0.513)
Test for Heterogeneity 0.057 (0.111) 0.110 (0.087) -0.064 (0.126)
Observations 4,057 4,057 4,057
Individuals 2,841 2,841 2,841
Note: ∗p<10%,∗∗p<5%,∗∗∗p<1%; bootstrapped standard errors with 50 replications in parentheses.
Table B.1 displays marginal treatment returns from public childcare utilization for the three indicators of maternal labor supply (employment, full-time and part-time employment) regarding sample means of covariates. Estimations are based on the first stage in selection equation (2) and the outcome equation (11) of the main text. The corresponding first stage is given in Table 4 of the main text where the set of variablesZ̃iktinstruments public childcare utilizationDit. The outcome equation is based on OLS, controls for time dummies, federal state dummies, and covariates given in Table 2, and estimates a third-order polynomial degree by parametric estimation. Panel B gives the interactions between individual characteristics and the propensity score PS.
Source: SOEP; Statistical Offices of the German Federal States; Statistical Office of the Federal Employment Agency; own illustration.
(a) Second-Order Polynomial Degree
Full-Time Employment Part-Time Employment
(b) Exclusive Consideration of Common Support
Full-Time Employment Part-Time Employment
(c) District-Specific Time Trends
Full-Time Employment Part-Time Employment
Figure B.1 – Econometric Specification Checks
Note: Figure B.1 displays the MTE curve for the probability of being full-time and part-time employed based on the first-stage in equation (2) and the outcome equation (11) of the main text. The first-stage applies probit estimations and regresses the utilization of public childcare (Dit=1) on spatial characteristics given in Table 3 of the main text, dummies for survey years, dummies for federal states, and covariates summarized in Table 2 of the main text. The outcome equation based on OLS also controls on time dummies, federal state dummies, and variables given in Table 2 of the main text and estimates a third-order polynomial degree by parametric estimation. Bootstrapped standard errors with 50 replications are clustered at the individual level. The graphs respectively plots marginal returns from utilizing public childcare on resistance UD for sample means of covariates. Panel a is based on estimations with polynomial degree two and Panel b exclusively runs estimations for the common support with propensity scores from 0 to 80.3 percent. Panel c adds interactions between spatial characteristics and post-period dummies.
Source:SOEP; Statistical Offices of the German Federal States; Statistical Office of the Federal Employment Agency; own illustration.
Appendix B Further Empirical Analyses
B.1 Further Robustness Checks
This section has a more detailed focus on the econometric specification of the baseline results of the main text. First, Table B.1 displays coefficients of relevant socioeconomic characteristics and its interactions with the propensity score of the second stage of the baseline MTE results displayed in Figure 6 of the main text. The lower part also displays the results of the test for heterogeneity, which is insignificant due to the choice of third-order polynomial degree.
Figure B.1 tackles some further econometric issues related to the MTE framework. In Panel a, we demonstrate that the effect pattern regarding unobserved resistance is robust if the polynomial degree is chosen as two. The choice of third-order polynomial degree seems to be reasonable to capture the full range of differential and flexible reactions. Panel b exclusively considers the common support, so that 133 observations at the tails are excluded from analysis. In Panel c, district-specific time trends with interaction between spatial characteristics and post-period time dummies are added to strengthen again the validity of the common trend assumption.
B.2 Empirical Results Conditional on Employment
Figure B.2 shows the results from estimating marginal treatment effects from the childcare ex- pansion for mothers conditional on employment.
B.3 Marginal Returns to Early Public Childcare on Re-Entry into Labor Mar- ket and on Partners’ Employment
Figure B.3 examines two research questions: First, do the childcare reforms reduce depreciation costs of human capital during a time-out from the labor market? Second, do mothers’ partners also shape their employment in consequence of better childcare availability?
In the upper row of Figure B.3, we display marginal returns for hourly re-entry wages and the
ratio of pre-birth wages and re-entry wages. Panel a uncovers a distinct positive selection on
gains with a negative sloped MTE curve. For mothers with a low resistance of U
D< 0.30, the
utilization of public childcare significantly increases re-entry wages.
(a) F ull-Time Emplo ymen t (b) P art-Time Emplo ymen t (c) Hours W ork ed Figure B.2 – Marginal Retur ns to Early Public Childcare Conditional on Emplo ymen t
FigureB.2displaystheMTEcurveforfull-timeemployment,part-timeemploymentandhoursworkedconditionalonemployment.Estimationsarebasedonthefirst-stageinequation theoutcomeequation(11)ofthemaintext.Thefirst-stageappliesprobitestimationsandregressestheutilizationofpublicchildcare(Dit=1)onspatialcharacteristicsgiveninTable maintext,dummiesforsurveyyears,dummiesforfederalstates,andcovariatessummarizedinTable2ofthemaintext.TheoutcomeequationbasedonOLSalsocontrolsontime federalstatedummies,andvariablesgiveninTable2ofthemaintextandestimatesathird-orderpolynomialdegreebyparametricestimation.Bootstrappedstandarderrorswith areclusteredattheindividuallevel.ThegraphsrespectivelyplotmarginalreturnsfromutilizingpublicchildcareonresistanceUDforsamplemeansofcovariates. SOEP;StatisticalOfficesoftheGermanFederalStates;StatisticalOfficeoftheFederalEmploymentAgency;ownillustration.(a) Hourly Re-Entry Wage (b) Ratio of Re-Entry Wage and Pre-Birth Wage
(c) Partner: Full-Time Employment (d) Partner: Part-Time Employment
Figure B.3 – Marginal Effects on the Re-Entry into Labor Market and on Partners’ Employment
Note: Panel a plots marginal returns for hourly wages from utilizing public childcare on resistance UD for sample means of covariates. Panel b models the ratio of pre-birth wage and current wage. Note that both outcomes are only observable for sample conditional on employment (see also Table 3 of the main text). Both MTE curves are based on the first-stage in equation (2) and the outcome equation (11). The first-stage applies probit estimations and regresses the utilization of public childcare (Dit=1) on spatial characteristics given in Table 3 of the main text, dummies for survey years, dummies for federal states, and covariates summarized in Table 2 of the main text. The outcome equation based on OLS also controls on time dummies, federal state dummies, and variables given in Table 2 of the main text and estimates a third-order polynomial degree by parametric estimation. Bootstrapped standard errors with 50 replications are clustered at the individual level.
Instead, Panel c and d model the effect of childcare utilization on partners’ current employment status, while the underlying regressions control for mothers’ and partners’ socioeconomic covariates. Note that for the extensive margin no significant effects are found. This is why the MTE curve for a father’s general employment status is omitted.
Source: SOEP; Statistical Offices of the German Federal States; Statistical Office of the Federal Employment Agency; own illustration.