Wednesday 10 March 2010

Wage discrimination: the evidence

LemmusLemmus in the comments asks about the evidence on wage differences between males and females: specifically, whether maternity risk shouldn't provide a wage gap even after correcting for everything else. Rather than answer there, it seemed worth posting what evidence we do have, especially since I spend a week lecturing on it in my Econ and Current Policy Issues class.

It turns out that maternity risk is one of the things you can control for in discrimination studies. The nicest example I know of is Pascale Petit's 2007 piece “The effects of age and family constraints on gender hiring discrimination: A field experiment in the French financial sector.” Labour Economics 14: 371-391. (ungated)

In Petit's field experiment, employers advertising positions are sent applications from three different experience-paired candidates. The first pair is aged 25, single and childless; the second, aged 37, either single or divorced, and childless; the third, aged 37, married with three kids. In the paired samples, the only difference is gender. Petit says the fertility rate in France of women aged 25-29 is 130% and that of those 35-39 was 50% (though much lower for those who already have three children). He finds that maternity risk explains all differences.
We find significant discrimination against women with a high probability of maternity for highly qualified administrative jobs. In the other cases, unequal treatment between genders is not significant. So, controlling for female probability of maternity, we find no significant unequal treatment between genders. We conclude that female employment suffers more from their probability of maternity (and their anticipated career interruptions) than family responsibilities alone. So, statistical discrimination due to female probability of maternity exists on the French labor market. An appropriate economic policy may correct it by reducing firms’ cost due to maternity leave.
So yes, LemmusLemmus, where a firm expects to bear high costs of maternity risk (in France, 45 days pay at employer's expense), men will have an easier time finding work or will get higher salaries. It would be a tough call to say this is discrimination though. In the French case, the employer can reasonably expect 1.5 spells of mandatory 45-days' additional paid holiday for a hired 25 year old female over the subsequent 4 years, that the timing of that time off work will be at the employee's control, and that hiring temporary replacement can also be difficult. If there's some public policy goal served by having paid parental leave, the burden of such leave should fall on the state rather than being imposed through regulation on the employer if we wish to avoid having firms avoid hiring women.

It would be very interesting to see whether pay gaps found across countries correlated with the burden faced by firms in case of employee maternity risk. I don't know that anybody's written that article yet though.

The Petit piece is an example of one of the two main types of studies looking at discrimination: an audit study. The typical audit study sends out matched sets of job applications and checks whether men or women, those with white or black sounding names, or whatever other "irrelevant" difference affects the number of interview requests. The second kind of study runs wage regressions on whatever set of employee characteristics we might have at hand.

James Heckman nicely goes through the advantages and disadvantages of each type of technique. Audit studies may find that some firms act in discriminatory fashion, but so long as they're only a small part of the market, they can't affect aggregate wages or employment outcomes (they're not the marginal firm). Regression studies essentially assume that unobserved variables are distributed equally across races or genders; since correcting for observables that are correlated with race reduces the racial pay gap from somewhere around 20% to somewhere around 5% (see Table 5 of O'Neill, here), it would seem somewhat heroic to assume that unobservable differences were distributed equally across groups when observables so clearly are not. We then can't say that the remaining 5% represents discrimination. Audit studies suffer from a similar problem: explicitly setting the observables equal makes the selection happen on the unobservables. So an audit study using white sounding and black sounding names tells you much more about the candidate than just race: it says something about the employee's likely cultural background, and if that background is correlated with performance, then we've got problems again.

In sum, audit studies find plenty of discrimination at a minority of firms (see Raich and Rich, 2002), but regression studies show only small pay gaps once we've corrected for other observables. Even those gaps are more readily explained by differences in unobservables than in discrimination: think about maternity risk, think about racial differences in high school versus GED achievement and whether the study has equated a GED with a high school diploma...

The more interesting recent work is less about horrible racists and sexists discriminating because they hate people (which is stupid and loses them money) and more about whether statistical discrimination can be consistent with profit maximization but also lead to negative self-reinforcing traps for minorities. Glen Loury's work here is nice. In his model, employers expect that, because blacks have less human capital than whites on average, employers will expect that any prospective black employee giving a good performance in interview is more likely to have achieved that result due to chance than due to underlying skill; consequently, black workers are not only more likely to be sorted into jobs requiring less human capital given any achieved level of human capital, they'll also have much lower incentive to accumulate human capital since the returns to those investments are lower. Policy conclusions coming from this work are ambiguous though: affirmative action policies can also reduce the return to human capital by turning employment into something of a lottery; whether policies like affirmative action improve outcomes then is ambiguous. Subsidies for minority enrollment in higher education are more likely to be beneficial.

I've linked to a few pieces above; here are the other ones I have on my reading list for the economics of discrimination (second year undergrad class, so I've tried to avoid anything way too technical):

Week 9: Economics of Discrimination

Required Reading:

Harford, Tim. 2008. “The dangers of rational racism.” Chapter 6, pp. 145-165 in The Logic of Life.

Recommended Readings:

Coyne, Christopher, Justin Isaacs, Jeremy Schwartz and Anthony Carilli. 2007. “Put me in, Coach, I’m ready to play.” Review of Austrian Economics 20: pp. 237-246.

Heckman, James. 1998. “Detecting Discrimination.” Journal of Economic Perspectives. 12:2, pp. 101-116. Ignore the math appendix.

Supplemental Readings:

Barro, Robert J. 1998. “So you want to hire the beautiful. Well, why not?” Business Week 16 March, p. 18.

Becker, Gary. 1993. “Nobel lecture: the economic way of looking at behaviour.” The Journal of Political Economy. 101:3 (June), pp. 385-409.

DeLeire, Thomas. 2000. “The Unintended Consequences of the Americans with Disabilities Act.” Regulation, 23:1, pp. 21-24.

The Economist. 2008. “Nearer to Overcoming: Black America.” 387:8579 (10 May), p. 38.

Ginther, Donna and Shulamit Kahn. 2006. “Does science promote women? Evidence from academia 1973-2001.” NBER Working paper # 12691 available at http://www.nber.org/papers/w12691

Holzer, Harry and David Neumark. 2000. “Assessing Affirmative Action.” Journal of Economic Literature 38 (September) 483-568.

“Nearer to overcoming.” 2008. The Economist 8 May. Available at http://www.economist.com/world/na/PrinterFriendly.cfm?story_id=11326407

O’Neill, June. 1990. “The role of human capital in earnings differences between black and white men.” Journal of Economic Perspectives 4:4 (Autumn). pp. 25-45.

Petit, Pascale. 2007. “The effects of age and family constraints on gender hiring discrimination: A field experiment in the French financial sector.” Labour Economics 14: 371-391.

Posner, Richard. 1992. “Racial Discrimination.” Chapter 26, pp. 651-63 in Economic Analysis of Law: Fourth Edition.

Riach, P.A. and J. Rich. 2002. “Field experiments of discrimination in the market place.” The Economic Journal 112 (November) pp. F480-F518.

Roback, Jennifer. 1986. "The Political Economy of Segregation: The Case of Segregated Streetcars." Journal of Economic History 56:4, pp. 893-917.

Sowell, Thomas. 2004. “The economics of discrimination.” Chapter 6, pp. 161-191 in Applied Economics: Thinking Beyond Stage One.

3 comments:

  1. That's much more than I asked for. Thank you very much!

    ReplyDelete
  2. I aim to please when the costs on me are low; I have the two hour lecture version of this in my back pocket :>

    ReplyDelete
  3. Yes, audit studies find to much discrimination compared to wage gaps. Leave mkvey on table, goes to profit from hiring shunned workers

    ReplyDelete