Falsification Exercise: Econometrics and the Minimum Wage

I love the concept of a "falsification exercise." It basically means, take an analysis that supposedly shows a causal effect of some cause (say, a minimum wage increase) on some effect (say, a change in the income share of the lower class). Apply the same analysis to measure the causal effect on some other effect, where there's no conceivable theoretical justification for a causal relationship. This is an incredibly useful concept. Maybe your statistical Rube Goldberg machine shows that x causes y, but it also shows that x causes a, b, c, z, and triple-z, none of which any conceivable theory would have predicted. So maybe we should doubt that your clever economic analysis is telling us what you think its telling us.

Here's an excerpt from the excellent book Minimum Wages by Neumark and Wascher:

[T]he upper-tail evidence constitutes what is often referred to as a falsification exercise. That is, if theory predicts an effect of x (the minimum wage) on y (lower-tail inequality), and such evidence is found, researchers often also explore whether there is an effect of x on another variable, z (upper-tail inequality), which is conceptually related to y but for which theory does not predict an effect on z. If no evidence suggesting an effect of x on z is found, the evidence of an effect of x on y is viewed as more convincing, and vice versa. The point of the Autor, Katz, and Kearney analysis is that, in this case, the falsification exercise fails.

More here:

Autor, Katz, and Kearney also cast serious doubt on previous research that emphasized the importance of minimum wages for changes in wage inequality. The most striking evidence they present is that the minimum wage is strongly correlated with upper-tail wage inequality as well as with lower-tail wage inequality. Indeed, in simple regressions of the 90/50 or 50/10 wage gaps on the real minimum wage, the estimated coefficient on the real minimum wage is larger for the 90/50 gap than for the 50/10 gap (-0.44 vs. -0.27, with both significant). In more complete regression models that account for a time trend, the relative supply of more- and less-educated workers, and aggregate economic conditions, a significant relationship between these gaps and the real minimum wage persists, although the estimated coefficient on the minimum wage is about two-thirds lower for the 90/50 gap than for the 50/10 gap. Autor, Katz, and Kearney conclude that correlations between tween the minimum wage and both upper- and lower-tail inequality measures suggest that the "time series correlation between minimum wages and inequality is unlikely to provide an accurate account of the causal effect of the minimum wage on earnings inequality. Indeed, we view the relationship between the minimum wage and upper tail inequality as potential evidence of spurious causation".

This is good econometrics. Really, it borders on philosophy of science.