It may not seem that way, but any gender pay gap is a macroeconomic issue. This is because we mostly measure that gap with average national statistics, and most policy approaches that begin with the assertion that this is evidence of discrimination are at the national level as well.
Jordan Peterson is the public intellectual of the season. In one of his many videos he notes at the 6:00 mark that “Like if you’re a social scientist, worth your salt, you never do a univariate analysis”.† Hold that thought.
In general discussion, the factoid that most people know about the gender pay gap is that women are paid some percentage less than men. Sometimes that’s derived from averages across the gender, and sometime from medians. The current best estimate is 12%, although politicians and advocates prefer older estimates with larger values.
Either way, that factoid is evidence of univariate analysis, because the only variable used to generate the difference is gender.
Economists have had trouble with this for a long time because when you start including other variables into the analysis, you find out that most/all of the pay gap is explained by variables that have little to do with discrimination. If that’s the case, then the discrimination explanation is a “Just-so Story”. Peterson and others make this point.
Economists don’t have a problem addressing discrimination through non-market measures. But they definitely do not want a market modified by such a measure when the market is not the source of the discrimination. That’s the problem with these results and the typical responses to the possibility of discrimination. In this case, this often amounts to some policy to extract money out of employers to transfer to the disadvantaged group (good economics students will recognize that there’s an incidence problem there, since firms may be able to pass some or all of that along to their customers). It’s OK to have a policy to divert some money from owners and customers provided that the owners and customers are the source of the problem. If they’re not, it’s not efficient for the economy as a whole, and we should be looking harder for other forms of redress.
So there’s a new working paper by Cook, Diamond, Hall, List and Oyer entitled “The Gender Earnings Gap In the Gig Economy: Evidence from Over a Million Rideshare Drivers”.‡ They studied Uber.
Uber has done a lot to design a system in which men and women should be compensated equally. And yet … they’re not.
Prior to this paper, the primary explanation for the remaining 12% gap is that women prefer more flexible scheduling. To the extent that employers are unwilling to work around that preference, it would constitute a solid motivation for regulatory redress from the employers.
But that’s not what this new paper finds. Uber’s system applies the same pay formula regardless of gender, without negotiations, or role for tenure or work intensity. Customers might be discriminating against women, but this doesn’t show up in the data. What they’re left with is a 7% pay gap that does not seem to involved employer discrimination. (Do note that doesn’t mean that the other 5% of the measured gap is not from discrimination).
Also note that assertion of the authors “… We know of no prior work that fully decomposes the gender earnings gap in any setting”, and “… We are not aware of prior cross-sectional wage regressions that have precisely and entirely eliminated the gender pay gap in virtually any context”. Those are fairly bald claim that they’ve accounted for more sources of the gap than previous authors.
What they did find was 3 factors that can explain the pay gap. Now, these could reflect differences in preferences between men and women. And you can argue that those differences in preferences are driven by social conditioning. Even so, that implies that redress should not come from employers and customers, but rather from some other broader section of society.
So what are those 3 factors?
First, men drive faster. Half of the gender pay gap is men putting in more trips in the same period of time. What’s really interesting about this cause is that the private sector has already compensated for this by charging higher auto insurance rates to men.
Also, it’s not a factor that would be relevant in most professions. So what we’re seeing is a 12% univariate pay gap, reduced to 7% by a firm’s compensation system, further reduced to 3% due to a job-related skill that’s already compensated for.
Second, male drivers make more trips with Uber. They demonstrate there is a learning curve associated with Uber driving, that is persists for years, and that men climb it faster by taking more trips. This gap opens up quite early in a worker’s tenure with Uber, and still persists two years after hire: it takes women 16 months to accumulate the same experience that men do in a year.
Third, this learning curve appears to be related to the choice between agreeing to make a distant pick-up and rejecting it in the hope something nearby shows up. So, if men drive more often, they establish this skill earlier and better than women do. The authors provide evidence that men and women learn this skill at the same rate.
Do note that none of this implies that there is not a problem. It just says that the source of the problem may be in the personal preferences of men and women. There may be a role for social policy to alleviate that, but it’s hard to see one. Think about what’s going on here: Uber has done a lot to make the job as flexible as possible, but the flexibility that women seem to care about the most is not in how to do the job day-to-day but rather whether to do the job at all on a given day.
However, there is also a perception that women work more than men “because they work two jobs”. The American Time Use Survey data from the Bureau of Labor Statistics suggest that the difference is only … about 18 minutes per day. That’s not nothing, but on the other hand, it’s 2% of the day, and there is little basis for arguing that the elasticity of pay with respect to time is around 6 (the value needed to convert a 2% difference in time into a 12% difference in pay). Could it explain the 5% difference between the nationwide gap and Uber’s gap? Maybe.
† Do note that when we get into the time series section of the class (towards the end of the month) it may appear that we are doing univariate analysis, when in fact we are doing indirect multivariate analysis. The reason is that if a variable depends on many other variables, and we relate that variable to its own lags, we are indirectly relating it to all those other variables. Basically, it’s a short cut.
‡ John List is a good role model for the SUU economics student. His vita is huge, and he has tons of heavily cited papers rather than just a few big ones. Best of all, he went to the University of Wisconsin at Stevens Point (an SUU sized, rural, school) as an undergraduate, and the University of Wyoming for graduate school. That’s a path comparable to the one you’re on right now.