The document discusses the impact of statistical discrimination on gender wage gaps among young workers, emphasizing how fertility-related decisions can lead to disparities in hiring and wages. It highlights trends such as delayed fertility and demographic changes that serve as potential explanations for these discrimination patterns across multiple countries. The findings suggest employers may exhibit bias based on fertility timing, with evidence showing a correlation between delayed fertility and reduced gender wage gaps.

1. Statistical discrimination at young age:
Statistical discrimination at young age:
evidence from young workers across four decades and 56 countries
Joanna Tyrowicz [FAME|GRAPE, University of Warsaw & IZA ]
Lucas van der Velde [FAME|GRAPE & Warsaw School of Economics]
Congress of the European Economic Association
August 2022

2. Statistical discrimination at young age:
Motivation
Motivation – textbook case for statistical discrimination
Fertility (-related absences) as premise for gender inequality
fertility plans → hiring decisions
(Becker et al., 2019)
child bearing → wage loss among mothers (not fathers)
(Landais & Kleven, 2019; Cukrowska-Torzewska & Matysiak, 2017; Pertold-Gebicka, 2014)

3. Statistical discrimination at young age:
Motivation
Motivation – textbook case for statistical discrimination
Fertility (-related absences) as premise for gender inequality
fertility plans → hiring decisions
(Becker et al., 2019)
child bearing → wage loss among mothers (not fathers)
(Landais & Kleven, 2019; Cukrowska-Torzewska & Matysiak, 2017; Pertold-Gebicka, 2014)
Demographic trends: ↑ age at first birth and ↓ # of births
⇒ less reasons for statistical discrimination

4. Statistical discrimination at young age:
Motivation
Motivation – textbook case for statistical discrimination
Fertility (-related absences) as premise for gender inequality
fertility plans → hiring decisions
(Becker et al., 2019)
child bearing → wage loss among mothers (not fathers)
(Landais & Kleven, 2019; Cukrowska-Torzewska & Matysiak, 2017; Pertold-Gebicka, 2014)
Demographic trends: ↑ age at first birth and ↓ # of births
⇒ less reasons for statistical discrimination
What we do
study gender wage gaps among labor market entrants
explore the role of delayed fertility

5. Statistical discrimination at young age:
Motivation
Motivation – textbook case for statistical discrimination
Fertility (-related absences) as premise for gender inequality
fertility plans → hiring decisions
(Becker et al., 2019)
child bearing → wage loss among mothers (not fathers)
(Landais & Kleven, 2019; Cukrowska-Torzewska & Matysiak, 2017; Pertold-Gebicka, 2014)
Demographic trends: ↑ age at first birth and ↓ # of births
⇒ less reasons for statistical discrimination
What we do
study gender wage gaps among labor market entrants
explore the role of delayed fertility → implicit test of statistical discrimination

6. Statistical discrimination at young age:
Motivation
Our contribution
We uncover a link from timing of fertility to (adjusted) gender wage gaps

7. Statistical discrimination at young age:
Motivation
Our contribution
We uncover a link from timing of fertility to (adjusted) gender wage gaps
Comparable measures of AGWG (across c & t) for entrants

8. Statistical discrimination at young age:
Motivation
Our contribution
We uncover a link from timing of fertility to (adjusted) gender wage gaps
Comparable measures of AGWG (across c & t) for entrants
Causal evidence: several instruments
Duration of compulsory education (multiple reforms)

9. Statistical discrimination at young age:
Motivation
Our contribution
We uncover a link from timing of fertility to (adjusted) gender wage gaps
Comparable measures of AGWG (across c & t) for entrants
Causal evidence: several instruments
Duration of compulsory education (multiple reforms)
Military conscription (many changes)

10. Statistical discrimination at young age:
Motivation
Our contribution
We uncover a link from timing of fertility to (adjusted) gender wage gaps
Comparable measures of AGWG (across c & t) for entrants
Causal evidence: several instruments
Duration of compulsory education (multiple reforms)
Military conscription (many changes)
New IV: international variation in “pill” admission
(in the US: Goldin & Katz, 2002; Bailey, 2006; Oltmans-Ananat & Hungerman, 2012)

11. Statistical discrimination at young age:
Data & method
What we would like to do
We would like to estimate the following regression
AGWGc,t = βi + β × Fertilityc,t + γXc,t + ϵc,t

12. Statistical discrimination at young age:
Data & method
What we would like to do
We would like to estimate the following regression
AGWGc,t = βi + β × Fertilityc,t + γXc,t + ϵc,t
Fertility: use mean age at first birth
TFR is noisy → we want the “risk” by employers at 20 < age < 30

13. Statistical discrimination at young age:
Data & method
What we would like to do
We would like to estimate the following regression
AGWGc,t = βi + β × Fertilityc,t + γXc,t + ϵc,t
Fertility: use mean age at first birth
TFR is noisy → we want the “risk” by employers at 20 < age < 30
AGWG: obtain own estimates
→ adjust raw GWG for 20 < age < 30
But: fertility decisions endogenous to labor force participation & AGWG

14. Statistical discrimination at young age:
Data & method
What we would like to do
We would like to estimate the following regression
AGWGc,t = βi + β × Fertilityc,t + γXc,t + ϵc,t
Fertility: use mean age at first birth
TFR is noisy → we want the “risk” by employers at 20 < age < 30
AGWG: obtain own estimates
→ adjust raw GWG for 20 < age < 30
But: fertility decisions endogenous to labor force participation & AGWG → need to instrument

15. Statistical discrimination at young age:
Data & method
(I) Fertility data
We use mean age at first birth (MAB) as a measure of fertility
Direct link to probability of becoming a parent
Less noisy than alternatives
Total fertility rate, age specific fertility, childlessness
Data collected from a variety of sources
Eurostat, UNECE, OECD, Human Fertility Database + bureaus of statistics + papers

16. Statistical discrimination at young age:
Data & method
(II) Measuring the adjusted gender wage gap
Nopo decomposition
A flexible non-parametric approach based on exact matching
Reliable even when when small set of covariates
Reliable even when cannot correct for selection bias
AGWG within common support

17. Statistical discrimination at young age:
Data & method
(II) Measuring the adjusted gender wage gap
Nopo decomposition
A flexible non-parametric approach based on exact matching
Reliable even when when small set of covariates
Reliable even when cannot correct for selection bias
AGWG within common support
We need individual level data

18. Statistical discrimination at young age:
Data & method
(II) Measuring the adjusted gender wage gap
Collecting individual level data
1 Harmonized data sources:
IPUMS + LISSY + EU (SILC, SES, ECHP)
2 Longitudinal data
Canada, Germany, Korea, Russia, Sweden, the UK, Ukraine and the US
3 Labor Force Surveys and Household Budget Surveys:
Albania, Argentina, Armenia, Belarus, Chile, Croatia, France, Hungary, Italy, Poland,
Serbia, the UK and Uruguay
4 LSMS (The World Bank):
Albania, B& H, Bulgaria, Kazakhstan, Kyrgistan, Serbia and Tajikistan

19. Statistical discrimination at young age:
Data & method
(II) Measuring the adjusted gender wage gap
Collecting individual level data
1 Harmonized data sources:
2 Longitudinal data
3 Labor Force Surveys and Household Budget Surveys:
4 LSMS (The World Bank):
In total:
– unbalanced panel 56 countries from early 1980s onwards
– ∼ 1258 measures of the Adjusted GWG
details

20. Statistical discrimination at young age:
Data & method
(III) Instruments
Compulsory schooling ⇒ fertility (Black et al. 2008, Cygan-Rehm and Maeder 2013)
Source: Compulsory schooling: UNESCO + papers for earlier years

21. Statistical discrimination at young age:
Data & method
(III) Instruments
Compulsory schooling ⇒ fertility (Black et al. 2008, Cygan-Rehm and Maeder 2013)
Source: Compulsory schooling: UNESCO + papers for earlier years
Military conscription ⇒ the timing of family formation
Source: Mulligan and Shleifer (2005) + Military Balance

22. Statistical discrimination at young age:
Data & method
(III) Instruments
Compulsory schooling ⇒ fertility (Black et al. 2008, Cygan-Rehm and Maeder 2013)
Source: Compulsory schooling: UNESCO + papers for earlier years
Military conscription ⇒ the timing of family formation
Source: Mulligan and Shleifer (2005) + Military Balance
Authorization of contraceptive pills ⇒ female education, family and labor supply
(US: Goldin and Katz 2002, Bailey 2006, Ananat and Hungerman 2012)
Source: Finlay, Canning and Po (2012)

23. Statistical discrimination at young age:
Data & method
(III) Instruments - a small bit of history
The pill first invented in 1940s in the UK, the first approved patent in the US in 1960

24. Statistical discrimination at young age:
Data & method
(III) Instruments - a small bit of history
The pill first invented in 1940s in the UK, the first approved patent in the US in 1960
Adoption timing varied a lot, even in Europe

25. Statistical discrimination at young age:
Data & method
(III) Instruments - a small bit of history
The pill first invented in 1940s in the UK, the first approved patent in the US in 1960
Adoption timing varied a lot, even in Europe
Eastern European countries were forerunners
Portugal and Spain lagged behind (late 60’s and 70’s)
The latest: Norway

26. Statistical discrimination at young age:
Data & method
(III) Instruments - a small bit of history
The pill first invented in 1940s in the UK, the first approved patent in the US in 1960
Adoption timing varied a lot, even in Europe
Admission ̸= access (→ timing)

27. Statistical discrimination at young age:
Data & method
(III) Instruments - a small bit of history
The pill first invented in 1940s in the UK, the first approved patent in the US in 1960
Adoption timing varied a lot, even in Europe
Admission ̸= access (→ timing)
E.g. former socialist countries: admitted but unavailable
Prescriptions vs otc
The UK originally admitted it only for married women

28. Statistical discrimination at young age:
Data & method
(III) Instruments - a small bit of history
The pill first invented in 1940s in the UK, the first approved patent in the US in 1960
Adoption timing varied a lot, even in Europe
Admission ̸= access (→ timing)
Until today persistent differences in use as contraceptive
∼ 38% in W. Europe; ∼ 14% E. Europe but 48% (!) in Czech Republic

29. Statistical discrimination at young age:
Data & method
Estimation procedure
AGWGi,s,t = α + β × time + γ [
MABi,t + ξs + ϵi,s,t
MABi,t = ϕ + θPILLi,t + ϱEDUi,t + µCONSCRi,t + ςM FERTi,t + εi,t
Variation in pill authorizaton: one data-point for each country
We use 2SLS for panel data as in Baltagi and coauthors (1981, 1992, 2000)
It is a random effects model (FGLS)
but... instrumentation is different
Additional instruments are redundant in White sense
→ standard errors adjusted to unbalanced panels

30. Statistical discrimination at young age:
Results
Raw correlation between MAB and AGWG
AGWGc,t = 0.88 − 0.028 MABc,t + ϵc,t
(0.046) (0.001)
More descriptives

31. Statistical discrimination at young age:
Results
The effect of delayed fertility on AGWG - IVs
AGWG estimates βIV
β
(1) (2) (3) (4)
Fertility timing -0.025*** -0.033*** -0.025*** -0.019**
(0.0062) (0.0093) (0.0065) (0.0097)
Observations 1106 1161 1114 1170
R-squared 0.28 0.28 0.28 0.76
F-statistic 25286.6 461.1 15667.1
More demanding AGWG

32. Statistical discrimination at young age:
Results
The effect of delayed fertility on AGWG - Robustness checks
HDFE Quantile Regression Heterogeneous fertility
(1) (2) (3) (4) (5) (6)
Q25 Q50 Q75 Intercepts Slopes
MAB -0.012 *** -0.023 *** -0.022 *** -0.032 ***
[-0.02,-0.00] [-0.03,-0.01] [-0.03,-0.01] [-0.04,-0.02]
MAB< Q25 0.133 *** -0.018
[0.07,0.20] [-0.05,0.02]
MAB ∈ [Q25, Q75] 0.027 -0.019
[-0.03,0.08] [-0.05,0.01]
MAB> Q75 -0.019
[-0.05,0.01]

33. Statistical discrimination at young age:
Results
Finding a benchmark
How do our estimates compare to differences in costs faced by employers?

34. Statistical discrimination at young age:
Results
Finding a benchmark
How do our estimates compare to differences in costs faced by employers?
We need information on
Probability of becoming a parent
→ Age-specific fertility rates
Differences in cost of childbearing / rearing
→ Time-use surveys / ISSP
→ Diff-in-Diff : men vs women, parents vs childless

35. Statistical discrimination at young age:
Results
Benchmarking statistical gender discrimination: USA

36. Statistical discrimination at young age:
Results
Benchmarking statistical gender discrimination: Western Europe

37. Statistical discrimination at young age:
Summary
Summary
Do employers discriminate statistically? Tentatively yes
Delayed fertility among youth → GWG ↓

38. Statistical discrimination at young age:
Summary
Summary
Do employers discriminate statistically? Tentatively yes
Delayed fertility among youth → GWG ↓
IV estimates ∼ −0.03 (out of AGWG ∼ 0.12 on average)
Estimates stable and robust across model specifications

39. Statistical discrimination at young age:
Summary
Summary
Do employers discriminate statistically? Tentatively yes
Delayed fertility among youth → GWG ↓
IV estimates ∼ −0.03 (out of AGWG ∼ 0.12 on average)
Estimates stable and robust across model specifications
IV and OLS similar, but F-statistics strong

40. Statistical discrimination at young age:
Summary
Summary
Do employers discriminate statistically? Tentatively yes
Delayed fertility among youth → GWG ↓
IV estimates ∼ −0.03 (out of AGWG ∼ 0.12 on average)
Estimates stable and robust across model specifications
IV and OLS similar, but F-statistics strong
Benchmarking: ∆c × π “explains away” AGWG sometimes
→ employers may receive signals correctly, but rarely do

41. Statistical discrimination at young age:
Summary
Questions or suggestions?
Thank you!
w: grape.org.pl
t: grape org
f: grape.org
e: lvandervelde[at]grape.org.pl

42. Statistical discrimination at young age:
Summary
Availability of individual level database
back

43. Statistical discrimination at young age:
Summary
Adjusted vs raw gender wage gap
back

44. Statistical discrimination at young age:
Summary
Trends in gender wage gaps
All age groups Youth
Raw GWG Adjusted GWG Raw GWG Adjusted GWG
(1) (2) (3) (4)
Year -0.160 -0.0308 -0.164** -0.158**
(0.101) (0.0662) (0.0773) (0.0705)
Observations 1,151 1,151 1,128 1,128
R-squared 0.204 0.117 0.105 0.108
Mean value 16.28 17.60 7.93 12.23
back

45. Statistical discrimination at young age:
Summary
Evolution of the adjusted gender wage gap
back

46. Statistical discrimination at young age:
Summary
Robustness check: only AGWG estimates with ind. & occ.
AGWG estimates βIV
β
(1) (2) (3) (4)
Fertility timing -0.026*** -0.032*** -0.025*** -0.020***
(0.0051) (0.0069) (0.0053) (0.0060)
Observations 825 864 834 873
R-squared 0.28 0.28 0.28 0.62
F-statistic 7058.1 379.5 3894.7
Clustering SE Yes Yes Yes Yes
Time trends Yes Yes Yes Yes
back

47. Statistical discrimination at young age:
Summary
Robustness check: Fertility levels
Adjusted estimates Raw estimates
Youth All Youth All
Fertility level -0.016 0.013 0.0026 0.019
(0.026) (0.022) (0.036) (0.033)
Observations 1241 1301 1241 1301
R-squared 0.74 0.83 0.62 0.80
Clustering SE Yes Yes Yes Yes
Time trends Yes Yes Yes Yes
Notes: Fertility levels are measured using Total Fertility Rate in the country.
back