The growth of the "gig" economy generates worker flexibility that, some have speculated, will favor women. We explore one facet of the gig economy by examining labor supply choices and earnings among more than a million rideshare drivers on Uber in the U.S. Perhaps most surprisingly, we find that there is a roughly 7% gender earnings gap amongst drivers. The uniqueness of our data—knowing exactly the production and compensation functions—permits us to completely unpack the underlying determinants of the gender earnings gap. We find that the entire gender gap is caused by three factors: experience on the platform (learning-by-doing), preferences over where/when to work, and preferences for driving speed. This suggests that, as the gig economy grows and brings more flexibility in employment, women’s relatively high opportunity cost of non-paid-work time and gender-based preference differences can perpetuate a gender earnings gap even in the absence of discrimination.What they are saying is that in an environment where there can be no discrimination, there is still a wage gap. Based on the Uber example, it is 7%.
The myth of gender wage gap is one of those zombie notions which is hard to kill. It is endlessly recycled by the perpetually aggrieved and those convinced that they are victims of a vast societal conspiracy. Research around the world has repeatedly shown that the more you take into account relevant causal factors (field of study, field of endeavor, hours worked, continuity of work, etc.) the ever smaller the purported gap becomes. In other words, the more you compare like-to-like, the more it appears that men and women are paid the same for the same work. In the US, Claudia Goldin's is a useful body of research to reference. She is a feminist in good standing whose research is consistent with the above summary.
There are two items of note. Of course there is always going to be variability and indeed discrimination the more you get down to the level of individuals. The average is not the individual. So if you are looking at all the work of a million people over ten years, you might find, indeed, will find, that men and women on average are indeed paid the same for the same work. That elevated view tends to overlook that at the granular level there is all sorts of variance going on. Two equal individuals might be paid differently despite all relevant variables being identical. But the variance is random. One was hired at the top of a boom and the other at the bottom of a recession and given the stickiness of wages, there is a measurable difference. One was hired by a "gut feel for the market" boss and another was hired by a boss who rigorously indexes to market conditions. And indeed one was hired with discriminatory practices in their favor and the other by a boss with discriminatory behaviors against them.
But all that noise is random. The female hired by a discriminating male boss might have her income constrained but so might the male hired by a discriminating female boss.
So the finding that at an average system level there is no discrimination based on like-for-like does not preclude that there is indeed bias and active discrimination at the individual level. It just means that you are as often a beneficiary of that random bias and discrimination as you are a victim. For both males and females. Indeed, bias and discrimination might be common, but as long as it is randomly distributed, it will not show up in the aggregate numbers.
The second item of note is that even the most rigorous multivariate analysis of gender wage gaps typically show some small residual gap, from recollection, usually 2-5%. Once you have closed the gap from the claimed 30% to 2%, it is easy to dismiss the residual as simply a function of a yet-to-be identified variable(s) or an insufficiently large data set. And broadly, I think that is right. But it is annoying to have that loose end.
All of the above is reasonably put to bed. We will continue to learn more but the naive claim that there is a 30% gender wage gap based on systemic discrimination can be dismissed.
What Cook et al have done is very clever. They recognized that data from Uber drivers provided a great test case for what we think we know. Uber drivers choose when to work, what rides to accept, how often they work, etc. There is no human involvement at all - it is entirely driven by algorithms. The work is standard and measured and the outcome is also entirely driven by performance and algorithms. There is no opportunity for personalized discrimination or even bias.
If there is no possibility for bias and discrimination then the default assumption is that male and female drivers should earn exactly the same. But that is not what they found. There is a 7% gap. Still.
How can that be? My first assumption was that there might be some non-human systemic bias in the algorithms. Maybe women are more defensive in the locations they pick-up and the times that they drive? No. The researchers looked at that and though there were some small variances, it was not a measurable difference to the outcome. Men did indeed drive more late night lucrative shifts than women but women drove more shifts during the lucrative Sunday afternoon window. Yes there were differences in time preferences but no, there was no systemic difference because of those preferences.
Eventually they narrowed the causes down to three factors. In increasing order of impact on the 7% variance:
Preferences over where/when to work - Men and women do exhibit different preferences of when to work and what locations they prefer to work. But the difference explainable by the "when" is small. More relevant is the "where". And it isn't actually so much the geographical "where" as it is the type of route. Airport runs are apparently the most lucrative routes and men compete heavily for those runs. Overall the combination of where and when only explains about 20% of the 7% wage gap.Just fascinating. A system free of discrimination still generates a material difference in productivity based on three personal choices 1) how fast you drive, 2) how familiar you become with the system, and 3) choices about when and where to work (primarily choices about which routes to work).
Men have longer work durations with the platform - The six month attrition rate of drivers on Uber is about 20% higher for women than men. Overall, the average male driver has worked longer with the Uber platform than the average female driver. Not only do men use the platform over longer periods of time, but they use it more intensively, men driving 50% more trips per week than women. With practice duration comes increasing performance perfection. The more experienced you are with the platform, its algorithms and its reward systems, the better you are at choosing the rides which optimize your financial returns for your effort. This experience edge explains 30% of the 7% wage gap.
Men drive faster - It is a small difference, but men were completing more trips per hour than women. They were more productive. The gap is tiny. Men drive 2% faster than women. But over a shift, that 2% accumulates into a material difference in overall number of trips completed. In fact, that small 2% increase in speed accounts for 50% of the variance in the 7% gap.
This does not explain the 2-5% remaining gap in the big wage gap studies. I still suspect that the great bulk of the residual gap is primarily due to overlooked variables and sample sizes.
It does put a stake in the ground regarding the magnitude of personal choices. Driving skill is close to a commodity market and yet even there, personal choices, experience, and small edges in productivity can create a sustained and long term impact.
Just to play the numbers out, let's say Uber drivers are only driving to make ends meet and that is $20,000 a year. With a 7% margin of improvement, males are making $1,400 more each year than women. If they invest all that each year at 5% yield, at the end of a 20 year run, female Uber drivers will have only covered their annual living expenses but the average male Uber driver will have an additional $50,000 or so in the bank.
Fair? Fair doesn't come in to it. It is a neutral platform. It is the consequence of many small personal decisions with nothing precluding any individual from optimizing those factors with the greatest impact on their personal performance against their own personal preferences.
The 7% figure does not address the 2-5% residual gap in multivariate studies but it does suggest that a significant fraction of that 2-5% is due to individual choices and actual performance and reduces further the likelihood of non-randomized bias and discrimination.
Freakonomics also did a podcast on this study with one of the researchers.
Posts
No comments:
Post a Comment