Tuesday, April 05, 2016
Why The Gross Gender Gap in Earnings Is Mostly A Useless Figure
Or today's statistical raised eyebrow from yours truly (why yours truly?).
I came across an article on Fortune.com about the gender wage gap between white men on the one hand and women in various ethnic and racial categories on the other*. Now those are very important things to study, as are the gender wage gaps within each racial and ethnic group. Or the racial or ethnic gaps within a gender category. Or doing the same analysis for women in all those categories vis-a-vis, say, Asian-American or African-American men's earnings.
But the gross wage gap doesn't really tell us anything but one thing: The total difference in the amounts workers from two different categories earn over their working lives, and even this only if the total is properly calculated (which isn't as easy as it might look). And it really cannot be interpreted the way a net gender gap in wages can be: As possible evidence of labor market discrimination against one or more groups. I feel that the Fortune.com article is slipping and sliding in that direction.
Are you still reading? Probably not, if the weather is as nice there as here (white snow!). Now how to make this more exciting? Let's do awful pedagogy.
Suppose Mary has spent 20 dollars on apples, Anthony 50 dollars and Evelyn 40 dollars. Who has purchased the most apples?
You can't get the answer because you don't know how much each of them paid per apple. They might even all have paid different prices if they didn't shop in the same store.
The gross gender gap is like that example. The net gender gap would be an example where you are told the price of apples for Mary, Anthony and Evelyn. That example would let you conclude which of the three has the most apples or the least apples etc.
Or using econo-babble, what we ideally wish to compare are the lifetime earnings of two imaginary (average) individuals who differ in nothing but the characteristics we deem relevant. In the Fortune.com article those characteristics would be gender, race and ethnicity and any significant interaction terms between them**. Everything else should be exactly the same: age, length of working life, average working hours per week, education levels, experience, local labor market conditions, the industry where the individuals work and so on, possibly also the number of minor children the employees have and so on.
But the study the Fortune.com article describes doesn't standardize for those other things, or at least that is my reading. Consider age. The average age of white men in the labor force is higher than the average age of Latinas. The linked study computes lifetime earnings by assuming that the length of one's working life is 40 years and then multiplies the current median earnings of Latinas and white men by forty. Thus, some part of the calculated lifetime earnings difference is because Latinas, on average, are relatively young workers in the US labor markets and young workers earn less than older workers.
There are also educational differences between the groups the study compares, though not between men and women overall, and those differences should also be controlled for. The same goes for all the so-called non-discriminatory variables which affect earnings if we wish to compare the remaining wage gaps from the is-this-discrimination-? angle.***
All this is about The Proper Way of addressing the gender gap in earnings. I cannot tell what the correctly calculated monetary lifetime differences between white men and the studied racial and ethnic groups of women might be, though I believe that the direction of the difference and the overall ranking of the sizes of the lifetime differences would not be affected.****
So why am I boring you with this? I don't want to feed the rabid anti-feminists and other eager critics who insist on telling me that there is no gender gap in wages, silly women, and if there is, then it is because those brave men work 24 hours per day fighting dangerous crocodiles, work that women just don't want (being most eager to be cleaning ladies, of course). And focusing on just the gross gaps in earnings does leave the door open for that.
* In theory, the gap between white men and white women could be calculated from the data, too, but I don't have the labor market percentages of the various female groups listed and am too tired to look them up.
** Economists have been doing intersectionality of a sort for ever! The interaction terms allow for the possibility that race or ethnicity might affect the earnings of women and men differently or that gender might affect ethnic and racial differences.
*** Note that many of these corrections would reduce the calculated lifetime earnings differences, but not all of them. The assumption that all workers can spend forty years in the labor force is less likely to be true for women than for men (those damn kids) and may differ between women of different ethnicity, and it's always possible that controlling for a specific non-discriminatory variable could increase the net differences over the working years.
To make things even more twisted together, some variables which I list here as non-discriminatory may themselves be a consequence of discrimination of a different sort. This may apply to education if the school system funding and teaching quality is discriminatory on the basis of race/ethnicity/gender. The industry in which someone works might not be a wholly free choice if young women are steered into traditionally female but poorly paid industries by cultural norms or their parents. This steering could differ by race and/or ethnicity if cultural norms differ between those demographic groups.
**** With one possible exception: The lifetime net earnings difference between white men and Asian-American women could be smaller than the life time net earnings difference between white men and white women. See the last graph in this article which gets further into the interesting stuff but still not far enough. Then read this.