Economic incentives and gender identity

Transcription

1 Economic incentives and gender identity Andrea Ichino European University Institute and University of Bologna Martin Olsson Research Institute of Industrial Economics (IFN) Barbara Petrongolo Queen Mary University London and CEP (LSE) Peter Skogman Thoursie Stockholm University NIESR, Jan 2018

2 Motivation Role played by gender identity norms has attracted increasing attention Gender identity norms dictate appropriate behavior for men and women. utility cost of deviating from prescribed norm Norms of behavior feed into gender gaps (Fortin 2005, 2015; Bertrand, 2010; Bertrand et al 2015, 2016; Bursztyn et al 2017). Norms are typically slowly-evolving, and their persistence correlates quite well with slowing convergence in gender trends. Gender gaps induced by norms cannot be easily removed. This is often rationale for advocating gender quotas (Bertrand et al, 2016)

3 This paper By the definition of identity norms, individuals concerned are willing to bear a cost to adhere to prescribed behavior. e.g. couple sticking to male breadwinner model may forgo household income if the wife has high labor market returns However, the observed allocation of male and female effort in the household and labor market may combine the effect of both norms and economic incentives. Natural test of norms: measure the amount of household income that couples are willing to forgo in order to comply with the male breadwinner model in response to shocks e.g. changes in labor market returns for each spouse. This is in turn related to the substitutability of spousal roles in the provision of home services.

4 Approach Combine exogenous change in monetary cost of adhering to norms & information on changes in individual behavior along the line predicted by the norm. Exploit variation in net wages generated by EITC in Sweden in 2007, coupled with information on husbands and wives home production available in Swedish registry data. Swedish tax system is individual based; EITC generates independent changes in mg tax rates of husbands and wives. Home production concept: Temporary Parental Leave, which either spouse can take in order to care for a sick child. Parents shares of TPL positively correlated with overall home production shares in time use data (Eriksson and Nermo 2010).

5 From home production to gender norms Household model in which spouses jointly choose time investments in market work and home production so as to max disposable income, subject to home production technology. Following change in respective tax rates, parents reallocate home production. For given tax change, gain in family income increases with the substitutability between spousal inputs in home production. Substitutability of spousal inputs is key parameter that captures the strength of gender norms. home production is care for 3-11 year olds; no biological comparative advantages. Elasticity of substitution inversely related to norm intensity. Asymmetric elasticity to positive and negative tax changes revealing of type of norms.

6 Related work - Gender norms Gender labor market outcomes correlated with self-reported gender norms (Fortin 2005, 2015) Marriage penalty of deviating from male breadwinner model (Bertrand, Kamenica and Pan, 2015); focus on skilled marriage penalty (Bertrand, Cortes, Olivetti and Pan, 2016) High-achieving women may avoid career-enhancing actions if these damage their marriage prospects (Bursztyn, Fujiwara, Pallais, 2017) Our paper: direct test of the monetary penalty a couple is willing to pay in order to adhere to their adopted norm. Emphasis on heterogeneous behavior within-gender, along a few demographics With assortative mating, important dimension of gender norms is variation in decisions and behavior across couples with direct consequences on children s outcomes rather than simply across men and women.

7 Related work - Home production and taxes Determinants of home production studied in both macro and micro literatures Growth literature has highlighted relationship between taxes and time use as important factor driving structural transformation and employment growth (Prescott, 2004; Rogerson, 2007; Ngai and Pissarides 2007) Focus on cross-country/time-series variation in time use data Labor supply literature has focused on causal impact of taxes on labor supply (earnings) combining tax reforms with earnings data from administrative sources Gelber (2014) closely related: response of spouses earned income to tax changes in Sweden Our identification is similar to Gruber and Saez (2002) and Gelber (2014) and complements their work with a focus on home production which offers direct perspective on gender norms

8 Plan of the talk Motivation Model Data and institutions Econometric framework Results

9 The household model Households enjoy home-produced public good H: [ H = (1 s)h f β 1 β + sh β 1 β m ] β β 1, H j is share of time in home production for spouse j = m, f. s is the weight of male time into home production. β is the elasticity of substitution between spousal inputs. Each couple is characterized by a certain value of β, representing preferences about the combination of spousal Norms as preferences on how to provide the household public good

10 The household s optimization problem In the market, one unit of time of spouse j produces P j and earns W j = P j Labor supply is 1 H j Gross earnings are Y j = P j (1 H j ). Earnings taxed at T (Y j ), affected by EITC reform. Optimization: [ max H m,h f (1 s)h β 1 β f + sh β 1 β m ] s. to Y = [Y m T (Y m )] + [Y f T (Y f )]

11 First-order conditions FOCs require ( ) H m s β [ ] (1 τf )P β f = H f 1 s (1 τ m )P m τ j = T (Y j ) > 0 denotes the marginal tax rate Optimal H j falls with own pre-tax wage (1 τ j )P j and rises with pre-tax wage of the other spouse. Opportunity cost of producing H: C = (1 τ m )P m H m + (1 τ f )P f H f (assuming for simplicity proportional taxation)

12 Equilibrium H " C P " (1 τ " ) H " * E * H $ * P $ (1 τ $ ) P " (1 τ " ) H $

13 Equilibrium: Cut in τ f (rise in wife s net wage) H " C P " (1 τ " ) H ". E. H " * E * H $. H $ * P $ (1 τ $ ) P " (1 τ " ) H $

14 Equilibrium: strong and weak norms C P " (1 τ " ) H " C P " (1 τ " ) H ". E. H " * E * H $. H $ * P $ (1 τ $ ) P " (1 τ " ) H $

15 Traditional norms (only respond to τ m cuts) H " C P " (1 τ " ) H " # E # E. H $ # H $

16 Untraditional norms (only respond to τ f cuts) H " C P " (1 τ " ) E. H " * E * H $ * P $ (1 τ $ ) P " (1 τ " ) H $

17 Summary of cases Strength of norm decreases with β Let β + = d(h m h f ) d(τ m τ f ) d(τm τ f )>0 and β = d(h m h f ) d(τ m τ f ) d(τm τ f )<0. a couple conforms to a traditional gender identity norm if β + < β ; a couple conforms to a untraditional gender identity norm if β + > β.

18 From model to data Take log of FOC: (h mt h ft ) = β (σ ft σ mt ) + (u mt u ft ) h j = ln H j σ j = ln(1 τ j ) is (log) net-of-tax income share for spouse j Introduce income effects: (h mt h ft ) = β (σ ft σ mt ) + η( θ mt θ ft ) + (u mt u ft ) where θ j = ln[y jt T jt (Y jt )] dy H jdτ j P m (1 τ j )H j

19 Data and sample LOUISE register from Statistics Sweden for , covering demographics, income variables and transfers for population merge to multi-generational register to link parents and children and TPL register (Swedish Social Insurance) providing information on start and end dates of TPL spells. Select couples: Live together and earn > base amount of e4,000. Their youngest child is 3 years old in 2006 and their oldest is 11 years old in 2007.

20 Summary statistics Mean SD Min Max Males: Age Education (%) Labor earnings in ke ,949.5 Capital earnings in ke ,001.8 Benefit payments in ke Marginal tax rate (%) Days of TPL Females: Age Education (%) Labor earnings in ke Capital earnings in ke ,823.6 Benefit payments in ke Marginal tax rate (%) Days of TPL Couples: No. children aged 4-10 in Age of youngest child in Combined days of TPL Male-female gap in TPL Male-female gap in taxes (%) No. Obs. 172,125

21 The Swedish EITC Reform Introduced in January 2007, salient as of January 2007 payslips Cut marginal tax rates by points on earnings between e12k-30k, from baseline of about 33%. Altered the within-couple tax gap depending on earnings of spouses Obtain simulated change in marginal tax rates: τ j07 τ j06 = T j07(y j06 ) T j06(y j06 ) σ j07 σ j06 = ln[1 τ j07 ] ln[1 τ j06 ] tax change that would have occurred at constant earnings

22 Tax schedule before ad after the EITC Marginal tax rate Labor earnings

23 Tax schedule vs male earnings distribution p25 p50 p75 Marginal tax rate Labor earnings

24 Tax schedule vs female earnings distribution p25 p50 p75 Marginal tax rate Labor earnings

25 Distribution of simulated changes in tax-gap Change in the gap between male and female tax rates Change in the gap between female and male (log) net-of-tax shares

26 Treated + couples: tax gap increases Marginal tax rate T (Y f ) T (Y f ) Y f Y m T (Y m ) Labor earnings

27 Treated couples: tax gap decreases Marginal tax rate T (Y f ) Y f Y m T (Y m ) T (Y m ) Labor earnings

28 Control couples Marginal tax rate T (Y f ) T (Y m ) Y f Y m T (Y m ) T (Y m ) Labor earnings

29 Temporary Parental Leave Available to either parent to care for sick child (0-11 year old) during normal working hours. Replacement ratio of 80% (capped). Widely used on average 1.3 days per year per employee Good proxy for division of work within household Swedish Standard of Living Survey contains time use information, linked with earnings and TPL: father s share of TPL strongly correlated with father s share of other home production tasks. TPL is form of home production visible to employers may signal spousal specialization and shape employers beliefs about productivity and work attachment.

30 TPL statistics at baseline Share of couples with Total TPL=0 18.4% Male TPL=0 and Female TPL>0 30.0% Male TPL>0 and Female TPL=0 8.3% Male TPL > Female TPL 21.2% Male TPL = Female TPL 0 4.8% Male TPL < Female TPL 56.5% Distribution of the male-female TPL gap in days 25th percentile -5 50th percentile -1 75th percentile 0

31 Estimation (h mt h ft ) = β (σ ft σ mt ) + η( θ mt θ ft ) + (u mt u ft ) OLS estimates biased because u mt u ft : 1. is correlated to tax changes because of progressivity exploit EITC as exogenous source of variation in σ jt and θ jt 2. captures impact of income trends of spouses

32 Empirical strategy: ITT Conditional on base-year earnings and pre-treatment characteristics of spouses, the simulated change in tax gap is a valid instrument for the actual change However, the monotonicity assumption is violated: EITC incentives induce treated employees to work more, and as a consequence they end-up in a higher tax bracket Focus on ITT, i.e. use simulated tax changes as regressors

33 Actual and predicted net-of-tax share changes: Men Panel A: Men Change in marginal tax Real gross labor earnings (2006 values) Actual change in marginal tax Actual change in marginal tax Simulated change in marginal tax 06-07

34 Actual and predicted net-of-tax share changes: Women Panel B: Women Change in marginal tax Real gross labor earnings (2006 values) Actual change in marginal tax Actual change in marginal tax Simulated change in marginal tax 06-07

35 Controlling for baseline earnings (I) ITT specification: (h m07 h f 07 ) = β ( σ f 07 σ m07 ) + η ( θ m07 θ f 07 ) + (ũ m07 ũ f 07 ) Need to control for base-year earnings of spouses, which would have an impact on TPL over and above the EITC i.e. ũ m07 ũ f 07 = ρ m y m06 ρ f y f 06 + ν 07 But overcontrolling for baseline earnings would kill identification when useful variation stems from one tax reform, systematically related to earnings (Gruber and Saez, 2002) Adopt a quasi diff-in-diff strategy (h m07 h f 07 ) = β ( σ f 07 σ m07 ) + η ( θ m07 θ f 07 ) + (h m07 h f 07 no reform) + ν 07 Need to estimate (h im07 h if 07 no reform)

36 Controlling for baseline earnings (II) Two strategies 1. Use parametric prediction of (h m07 h f 07 no reform) Estimate the relationship between TPL and earnings in a period without tax changes ( ) and assume that, absent EITC, such relationship would hold during sample period (Gelber 2014). ten-piece spline in log earnings, with knots at the deciles of the distribution obtain prediction G (h m06 h f 06 ) = (h m07 h f 07 no reform) dep var: ˆ (hm07 h f 07 ) = (h m07 h f 07 ) G (h m06 h f 06 ) 2. Use nonparametric prediction based on matching For each spouse in main sample, select spouses in 2005 in the same bin (1/25) of earnings distribution and same municipality use their observed M (h m06 h f 06 ) = (h m07 h f 07 no reform) dep var: ˆ (hm07 h f 07 ) = (h m07 h f 07 ) M (h m06 h f 06 ) Implement placebo tests for both

37 Empirical specification Uniform β ˆ (h m07 h f 07 ) = β ( σ f 07 σ m07 ) + η ( θ im07 θ if 07 ) + γ f X f 07 + γ m X m07 + u 07. Asymmetric βs ˆ (h m07 h f 07 ) = β + ( σ f 07 σ m07 ) ( σf 07 σ m07 )>0 + β ( σ f 07 σ m07 ) ( σf 07 σ m07 )<0 + η ( θ im07 θ if 07 ) + γ f X f 07 + γ m X m07 + u 07. β + : away from traditional allocation ( against the tyde ) β : towards traditional allocation ( with the tyde )

38 Results: Baseline estimates Dep var is ˆ log TPL gap Spec: Gelber Matching Placebo G Placebo M β (0.092) (0.094) (0.095) (0.098) β (0.116) (0.119) (0.120) (0.123) β (0.222) (0.228) (0.226) (0.232) P-value diff Mean 2006 level No. Obs. 172, , , ,592

39 Heterogeneous effects Couple s characteristics often related to gender norms: Family income shares (Bertrand et al, 2015) Marital status Husband s education Country of origin (Fernandez 2007, Fernandez and Fogli 2009) Shares of standard parental leave Results presented based on Gelber (2014) residualization; results based on matching are similar.

40 Male vs female breadwinner Breadwinner: Husband Wife Column diff Dep var: ˆ log TPL gap β (0.127) (0.248) (0.294) β (0.152) (0.469) (0.502) β (0.365) (0.328) (0.467) Row diff (0.403) (0.613) (0.719) Mean 2006 level No. Obs. 139,656 32,469

41 By marital status Married Unmarried Column diff Dep var: ˆ log TPL gap β (0.117) (0.172) (0.200) β (0.164) (0.221) (0.255) β (0.275) (0.401) (0.475) Row diff (0.363) (0.502) (0.590) Mean 2006 level No. Obs. 119,795 52,330

42 By husband s education Low High Column diff Dep var: ˆ log TPL gap β (0.110) (0.207) (0.225) β (0.149) (0.270) (0.279) β (0.251) (0.539) (0.580) Row diff (0.326) (0.669) (0.700) Mean 2006 level No. Obs. 133,290 38,356

43 By country of origin ROW Scandinavia Column diff Dep var: ˆ log TPL gap β (0.399) (0.104) (0.411) β (0.577) (0.144) (0.589) β (0.744) (0.242) (0.773 ) Row diff (1.045) (0.316) (1.076) Mean 2006 level No. Obs. 10, ,855

44 By age of youngest child Age Column diff Dep var: ˆ log TPL gap β (0.136) (0.135) (0.181) β (0.180) (0.184) (0.230) β (0.317) (0.315) (0.432) Row diff (0.405) (0.410) (0.536) Mean 2006 level No. Obs. 92,891 78,755

45 Conclusions Test of gender identity norms based on the response of husbands and wives home production to a shock in their post-tax wages. Concept of norm closely related to the elasticity of substitution between spouses inputs in home production asymmetries in such elasticity are informative about the specific norm (if any) that a couple wants to adhere to. Exploit variation in post-tax wages introduced by Swedish EITC and TPL response of spouses. Elasticity of substitution not significantly different from one low Heterogeneous elasticity along a number of dimensions arguably correlated to gender norms Suggestive evidence of the relevance of both traditional and untraditional norms in different groups of the Swedish population.