Labor Economics. Discrimination on the Labor Market. Sébastien Roux/ Labor Economics ENSAE. May 23rd 2014

Transcription

1 Labor Economics Discrimination on the Labor Market Sébastien Roux/ Labor Economics ENSAE May 23rd 2014 Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

2 Introduction Introduction Why study discrimination in an economic class? The economic theory provides a benchmark situation : discrimination can be stated when reality differs from the benchmark. Example : wage differentials are supposed to be related to individual productivity. Existence of wage differentials not related to productivity differential but to other characteristics can be interpreted as discrimination. The economic theory may help to analyze the consequences of discrimination (welfare analysis) and how it can be suppressed or fought. A crucial question : how to measure discrimination? Need to have good data describing the attributes that are supposed to be discriminated against. Measure of outcome differentials, same difficulties as in the treatment literature. How to measure the counterfactual situation? Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

3 Introduction Legal definition Legal definition : Discrimination consists in treating an individual or a group of individuals less favourably because of their possession of an attribute, compared with someone without that attribute in same circumstances. In France, creation of the Halde (Haute Autorité de Lutte contre les Discriminations et pour l Egalité, now Défenseur des Droits) claims in 2012, by criterion Origin ,5% Health-disability % Sex 303 3,7% Pregnancy 388 4,7% Union activity 364 4,5% Age 400 4,8% Religious beliefs 157 1,9% Sexual orientation 198 2,4% Family-Marital Status 265 3,2% Physical Appearance 143 1,7% 50% of the claims are about employment, mostly about the career. Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

4 Introduction Outline Focus here on race and gender discriminations, because main studies are available on these attributes. 1 Taste-based Discrimination 2 Statistical Discrimination 3 Some empirical results Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

5 Taste-based discrimination Theories : a benchmark framework Null hypothesis (no discrimination) : preference and skill differences explain the outcome differentials. Discrimination is the residual difference, when accounting for these effects. Differences in preferences : Most often relate to gender preferences (preference for leisure) Need to look closer to pre-market differences Differences in comparative advantages Historically, physical strength. Skills related to parental investment, or family network. Differences in human capital investment Pre-labor market discrimination. Schools or neighborhoods. Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

6 Taste-based discrimination Theoretical Considerations Two classes of model : Competitive models in which agents act individually (mainstream) Collective models in which one group acts collectively against another Two types of discriminations in competitive models : Prejudice : taste by some members of the majority group (see Becker) In these models, discrimination should vanish through competition. Statistical discrimination in the presence of imperfect information about the skills or behaviors of the members of the minority group. Consistent with the persistence of long-term group differentials. Definition of discrimination : Situation in which persons who provide labor market services and who are equally productive are treated unequally in a way that is related to an observable characteristic Problems with defining equally productive (e.g. physical beauty in entertainment industry). Distinction between current labor market discrimination given a set of characteristics and the effects of prior discrimination on these characteristics (e.g. gender-specific education). Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

7 Taste-based discrimination Taste-based Discrimination Becker (1971) defines a situation in which some employers of majority group A are reluctant to hire members of minority group B. Employers maximize their utility U U = pf (N B + N A ) w A N A w B N B dn B (1) where p is the price, N g is employment of members of group g(g = A, B), w g is wage paid to member of group g and d is the prejudice parameter that valuates the cost for employers A to employ group B members. Then employer A hire group B members if w A w B d. Let G ( d; d ) the CDF of d in the population of employers, d is a parameter (e.g. the mean). The fraction of firms that hire B workers is G ( w A w B ; d ). The optimal number of workers hired is determined by : pf (N A ) = w A pf (N B ) = w B + d This is formally equivalent to introduce a market premium for a worker attribute Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

8 Taste-based discrimination Aggregating all firms in the economy, wages are solutions to these equations NA d ( wa, w B ; d ) = NA s (w A) NB d ( wa, w B ; d ) = NB s (w B) Wage differentials arise if d is sufficiently large for the demand for B workers when w B = w A to be lower than the supply. The price on the attribute is determined by the preferences of the least prejudiced employer who hires B workers. Discriminating employers have lower profits than non-discriminating ones. On the long run, with free entry condition, discrimination should vanish. Other types of discrimination could happen : Employee discrimination : taste of A employees to prefer working with same-group employees. Consumer discrimination : Consumers in group A get less utility if they purchase from a group B member. This lowers the labor market payoff for group B members who work in occupations with consumer contact. Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

9 Taste-based discrimination Taste-based discrimination with costly search Black(Journal of Labor Economics, 1995). Let γ be the type-b workers fraction. All workers are equally productive, have the same leisure preferences and direct search costs, c per period. There are two types of employers, p and u. p-employers, share θ, hire only A workers at wage w A p. u-employers, share 1 θ simply maximise profits, hire both type of workers, paying them w g u, g = A, B. The utility a worker gets from being employed is the sum of the wage and a match-specific component α. The worker know this value prior to accepting or rejecting an offer. The employer knows only the cdf F of this component. Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

10 Taste-based discrimination Worker s strategies Strategy of A workers : Value of search U A U A = θe max { w A p + α, U A} + (1 θ) E max { w A u + α, U A} c This leads to : θ α A p U A = ( w A p + α ) f (α) dα + (1 θ) ( α w A A u u + α ) f (α) dα c 1 θf ( ) αp A (1 θ) F (α A u ) where αj A = ur A wj A (j = u, p), ur A is the reservation utility such that U A = ur A. This is the solution to equation c = θ α A p Comparative statics : ( w A p + α ur A ) ( f (α) dα + (1 θ) w A u + α ur A ) f (α) dα 0 < du A r /dw A p < 1 0 < du A r /dw A u < 1 du A r /dθ 0, as w A p w A u Sequential search, expected number of searches : ν A = { θ [ 1 F ( )] [ ( )]} ur A wp A + (1 θ) 1 F u A r wu A 1 Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31 α A u

11 Taste-based discrimination Strategy of type-b workers : Value of search U B U B = θu B + (1 θ) E max { w B u + α, U B} c This leads to : U B = (1 θ) ( α w B B u + α ) f (α) dα c (1 θ) (1 F (α B )) where α B = ur B wu B, ur B is the reservation utility for type-b workers, it is such that U B = ur B = wu B + α B and is given by : c 1 θ = Comparative statics : ( w B u + α u B ) r f (α) dα α B 0 < du B r /dw B u = 1 du B r /dθ < 0 Expected number of searches : ν B = { (1 θ) [ 1 F ( ur B Ambiguous effect of θ on ν B : empirical investigation. wu B )]} 1 Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

12 Taste-based discrimination Employer s behavior Profit for a non-discriminating firm u to hire a type-g worker : π g u = ( V w g u ) [ 1 F ( u g r w g u )], where V is the worker s marginal product. F.O.C : V w g u m ( u g r w g u ) = 0, where m = f / (1 F ), sufficient if m ( u g r w g u ) 1 < 0 Same condition for discriminating firms (who hire only type-a workers) : V wp A m ( ur A wp A ) = 0 Hence w A p = w A u, and π A p = π A u = π A. Only u firms hire type-b workers : w B = w B u Comparative statics : 0 < dw B /du B r < 1 0 < dw A /du A r < 1 Since, du B r /dθ < 0, then w B < w A for θ > 0. Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

13 Taste-based discrimination Equilibrium Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

14 Taste-based discrimination Role of Entry How is determined θ? Idea : Entrepreneurs differentiate not only in their discriminating behavior, but also on their entrepreneurial ability. M potential entrepreneurs, a share ρ refusing to hire type-b workers. Let τ the entrepreneurial ability, and φ (τ) the fixed cost of operating a firm, φ (τ) < 0. Let G (φ) be the induced distribution of fixed costs with supports [ φ, φ ]. The distribution is the same for discriminating or not entrepreneurs. Π p (φ) = N (1 γ) ν A π A M [ρg (φ p ) + (1 ρ) G (φ u )] φ Π u (φ) = N { γν B π B + (1 γ) ν A π A} M [ρg (φ p ) + (1 ρ) G (φ u )] φ φ p and φ u are determined such that Π j (φ j ) = 0, j = p, u. Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

15 Taste-based discrimination ρg(φ θ = p) ρg(φ p)+(1 ρ)g(φ u) Comparative statics : φ u > φ p, the least profitable non-discriminating firm can work with a higher cost than the least profitable discriminating firm. Hence ( θ < ρ Π ) ( u φ 0 > Π ) p φ 0 dφ p /dur B > 0, and dφ u /dur B < 0. Hence, dθ/dur B > 0, an increase in the reservation utility of type-b workers increase the share of discriminating employers. The economic equilibrium consists in ( θ, ur B ), which consists in the intersection of the curve θ = σ ( ur B ), given by firm s entry behavior and the curve h (θ), that describes type-b workers search behavior (ur B is determined as a function of θ). Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

16 Taste-based discrimination Equilibrium Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

17 Taste-based discrimination Effect of a change in ρ or γ Effect of an increase in the share ρ of discriminating entrepreneurs Shift of σ ( ) to the right Hence dw B dρ = dw B du B r du B r dθ dθ dρ ( ) < 0 Since dαb dρ = 1 dw B du B dur B r dθ dθ dρ < 0, the expected value of the satisfaction of minority workers decrease. Effect of an increase in the fraction of type-b workers γ Shift of σ ( ) to the left Hence dw B dγ = dw B du B r du B r dθ dγ ( ) > 0 Since dαb dγ = 1 dw B du B dur B r dθ satisfaction of minority workers decrease. dθ dθ dγ < 0, the expected value of the This is a rationale for minority groups to congregate in communities. Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

18 Statistical Discrimination Statistical Discrimination Idea : Firms have limited information about the skills of applicants. Hence incentive to use all observable characteristics to statistically discriminate among workers if these characteristics are correlated with performance. Two different approaches : Role of self-confirming stereotypes (see Coate and Loury, 1993) This model authorizes to examine the effect of such public policy as affirmative action. Higher uncertainty on the measure of productivity about a group of individuals (e.g. related to cultural differences). This may decrease match qualities, the return to job matching and the return to performance in the B group (inducing an underinvestment in skill acquisition) Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

19 Statistical Discrimination Stereotypes and Statistical Discrimination Coate and Loury, American Economic Review (1993.5) Employers are matched to a pool of workers, each belonging to an identified group, A if majority, B if minority. Each firm has two jobs. Task 0 is unskilled and can be performed by anyone. Task 1 requires a qualified worker. Firms pay a wage premium w to workers who do task 1. The net return to the firms of assigning a worker to task 1 is x q if he is qualified and x u if he is not. Employers observe group membership and a noisy signal T [0, 1] about worker s qualification. F q (T ) (resp. F u ) is the cdf of T when a worker is qualified (resp. unqualified). Likelihood ratio ϕ (T ) = f u (T ) /f q (T ) supposed to be non-increasing. Firms form a posterior probability that the worker is qualified upon the signal observed and a prior belief π g that a member of this group is qualified. Qualification is the result of an ex ante investment, at cost c, drawn from a distribution G. Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

20 Statistical Discrimination Sequence of actions nature chooses worker s types (A or B) workers make investment decisions (at cost c) workers match with employers test results T is observed employers make assignment decisions payoffs are received At equilibrium, employer s beliefs in about the likelihood of a group s members being qualified will affect their assignment decision. This perspective affects also the investment decision of workers. The equilibrium is then a pair of employer belief s which are self-confirming. A discriminatory equilibrium is one in which workers from one group are less likely believed to be qualified. Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

21 Statistical Discrimination Employer s behavior Consider a worker belonging to group g, π g is the prior probability to be qualified. Observing T, the posterior probability for the worker to be qualified is ξ (π g, T ) = π g f q (T ) π g f q (T ) + (1 π g ) f u (T ) = π g π g + (1 π g ) ϕ (T ) Hence, the payoff from assigning this to task 1 is ξ (π g, T ) x q (1 ξ (π g, T )) x u 0 or r = xq x u 1 πg π g ϕ (T ) Monotonicity of ϕ ( ) allows to define { s (π g ) = min T [0, 1] r 1 π } g ϕ (T ) π g s ( ) is decreasing in π g. Hence more optimistic beliefs will be reflected in easier standards. Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

22 Statistical Discrimination Worker s investment decisions The probability to task 1 assignment is 1 F q (s) when qualified and 1 F u (s) when not. The expected benefit of investment is β (s) = w [F u (s) F q (s)] A worker invest only if β (s) c. Hence the proportion of workers who become qualified is π = G (β (s)) Since β (0) = β (1) = 0 and ϕ is monotonous, β is a single peak function, increasing (decreasing) when ϕ > (<)1. Under minor assumptions, G (β ( )) is also a single peak function. Equilibrium definition : An equilibrium is a pair of beliefs (π A, π B ) satisfying π g = G (β (s (π g ))), g = A, B Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

23 Statistical Discrimination Multiple equilibria Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

24 Statistical Discrimination Affirmative action policy Affirmative action force employers to equalize the probability to be task 1 for any group. ρ (s, π) = π [1 F q (s)] + (1 π) [1 F u (s)] is this probability. P (s, π) = π [1 F q (s)] x q (1 π) [1 F u (s)] x u is the expected payoff. Employer s program becomes max [(1 γ) P (s A, π A ) + γp (s B, π B )] s A,s B subject to ρ (s A, π A ) = ρ (s B, π B ) The equilibrium under affirmative action consists in the beliefs (π A, π B ) and the standards (s A, s B ) such that (s A, s B ) solve the firm s program, given (π A, π B ) π g = G (β (s g )), g = A, B Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

25 Statistical Discrimination Theoretical consequences of affirmative action Let ˆρ (s) = ρ (G (β (s)), s), the expected fraction of the group to be assigned in task 1 at equilibrium. Under affirmative action, you have ˆρ (s A ) = ˆρ (s B ) If ˆρ ( ) is non-increasing on [0, 1] then s A = s B, which leads to π A = π B and there is no discrimination. Affirmative action policy completes its objective!! But, ˆρ ( ), is not necessarily non-increasing (especially for s < s). Then, there can be patronizing equilibria in which employers hold negative stereotypes about B workers and where these stereotypes are worsened by affirmative action. Idea : since firms believe that type-b workers are less productive, and because they must employ them (affirmative action), they propose them lower s B. This may diminish F u (s B ) F q (s B ) and reduce the payoff of training. In this case, affirmative action needs to be permanent!!! Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

26 Measuring discrimination The Oaxaca-Blinder decomposition Idea : Decompose a variable into an explained and an unexplained component. Consider wages of groups A and B : W iat = β At X iat + ɛ iat W ibt = β Bt X ibt + ɛ ibt Then E (W iat W ibt ) = β At (X iat X ibt ) + X ibt (β At β Bt ) The first term is the explained component, the second term is the unexplained component. The unexplained component is often defined as the discrimination share. Misleading in that it depends very much on the control variables. Moreover discrimination may also affect X ibt. But, when no other data are available, best way to proceed. Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

27 Measuring discrimination Gender Wage Differentials in France Meurs-Ponthieux (Economie et Statistique, 2006) Gender monthly wage differentials : in 1990, 26.2%, in 2002, 25.3%. Use of Oaxaca-Random decomposition : ( ) ) E (W iat W ibt ) = ˆβ (X iat X ibt )+X ibt ˆβ β Bt +X iat (β iat ˆβ Year Total Diff Unexpl Diff Expl Diff Human Capital Working Time W M W F (β ˆβ) M X M + ( ) X M X F ˆβ Educ Exp week part- ( ) ˆβ β F X F time Source : LFS Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

28 Measuring discrimination Origin-based differentials Aeberhardt, Fougère, Pouget, Rathelot (The Journal of Population Economics, 2010) Use of the information on the nationality at birth of the parents. From FQP, Outcome differentials between individuals whose parents were born in France and those whose parents were born in Africa : Wage differential is 15% and the employment probability is 12% those whose parents were born in South Europe : Wage differential is 3% and the employment probability is 1.5% Idea : accounting for discrimination on employment probability, what are the unexplained differentials? Origin Empl. Pr. Expl. Unexpl. Wage Expl Unexpl. Diff. Diff. Africa S.Europe Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

29 Measuring discrimination Audit studies Famous Example : Goldin and Rouse (1996),difficult to generalize. From the 70s, some orchestra have been using a screen for auditioning candidates from the jury. In such orchestra, the proportion of women increased from 0.1 in 1970 to 0.2 in Is this related to this blind audition system? Estimated model : P ijt = α + βf i + γb jt + δ (F i B jt ) + θ 1 X it + θ 2 Z jt where P is the probability that person i is pre-selected or hired in the audition with orchestra j at time t F is an indicator for female auditionner B jt is an indicator for blind audition X and Z are controls for person and audition characteristics, possibly individual fixed effects. Results : Use of screens increases by 50% the probability for women to be pre-selected. Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

30 Measuring discrimination Field Experiment study Recent example : Duguet and Petit (Annales d Economie et de Statistique, 2005) Idea : Apply for jobs by sending resume by mail or fax. Total control on resume s contents. Bertrand and Mullainathan (2003) explore this idea by proposing resumes differentiating names, some being black-named. Duguet and Petit adapt this idea by sending resumes with different profiles with respect to age (25 or 37), marital status (single or married) and child (3 or none). Other descriptive characteristics related to education or past occupations (157 different profiles). Each profile is doubled, differentiating gender. Resumes sent to firms in the financial sector. The authors explore the roles of employer s birth anticipation as an explanation for discrimination. Outcomes : Probability to obtain an interview Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31

31 Measuring discrimination Direct results to correspondance tests : Type Neither invited Both invited Discrimination Discrimination against women against men 25, single, childless , single, childless , married, 3 children Model : E (Y ik X i ) = P 0 k + X iβ k, k = m, f. Hence E (D i X i ) = P 0 m P 0 f + X i (β m β f ) (β m β f ) corresponds here to conditional discrimination. Results : On average, no significant discrimination : range from 0.7% to 4.1% on the probability to get an interview Strong conditional discrimination : high skilled young women without children are significantly discriminated. Same results when the job includes paid training. These results are counterbalanced by institutional features : greater chances for young women to get interviews in public-owned firms (La Poste) or in credit organisms (phone operators preferring women). Cahuc, Kramarz, Roux, Wasmer/ Labor Economics (ENSAE) Labor Economics May 23rd / 31