One of the great pleasures of writing this column is explaining economic research results to readers who are experts at something other than economic research. After all, we are all experts at something, and if my wife is to be trusted on these matters, most subjects are more interesting than economics. This week I thought I’d try to explain how economists use theories and data to learn something about the world. My example will be the often timely gender wage gap.

Everyone knows the statistic that women earn roughly 72 cents for every dollar men earn. Is this evidence of gender discrimination at work? Theory and common sense pretty clearly suggest that education, job tenure, occupation and effort also influence how much we earn. Thus, there may be more to the story than discrimination. So, if we have data on hundreds of people, we could use fairly straightforward statistical models to determine which of these things mattered most.

This approach may also tell us what doesn’t matter. Because the scientific method really only disproves relationships, proof only occurs after repeated efforts to disprove something fails. Now, it turns out that some of these things are easy to measure, like gender, age, and education, while effort might be nearly impossible to count. That’s fine though, because the typical statistical techniques can account for the size of unmeasured effects. Even more useful is the model’s ability to account jointly for two variables that are related. For that, it is best to go right to an example.

In a simple model, we might wish to test whether wages and gender were correlated, while separately testing whether wages and education were correlated. In these cases, we find that, on average, women make less than men and, separately, we find that better educated people earn more. Then combining gender and education, the effect of education grows, and the effect of gender shrinks. But, the overall model leaves less unexplained variation in wages. It is still a very incomplete picture of the issue.

As we add more data to our equation we learn more things. We confirm that occupation, choice of major, job tenure, hours worked all affect earnings. Also, each time we add these extra data about people, we find the size of the gender effect shrinks. In fact, there are quite a few studies that report little or no gender wage gap when accounting for education, tenure, occupation and hours worked. This is important to know, not simply because it helps unmask duplicitous politicians, but also because deeper research can help us identify and face the real problems instead of trying to remedy fictitious ones.

So, if a well-performed study reveals little or no gender wage gap, does this mean there’s no gender-based discrimination? Of course not. What it means is that when gender discrimination exists, it occurs in places other than labor markets. Thus, we would wish to look for it in schools, or in policies or cultural practices that influence occupational or educational choices. This would focus more research and analysis of these matters, and maybe tell us where and if a policy intervention is needed.

Besides the very obvious value of this sort of analysis to this particular question, the statistical modeling approach I just described has revolutionized public policy research in the past half century. The combination of broadly available data with tremendous computing power means that advanced statistical modeling is now just as necessary as an element of graduate training in sociology, anthropology and political science as it has been in economics for many decades.

Happily, these tools not only make better research in other social sciences, they also increase the value of carefully constructed studies that don’t rely on statistics. Good quantitative studies make qualitative studies more, rather than less important, and that may be its greatest contribution.