In my opinion, Arindrajit Dube is one of the best economists around in figuring out what we should control for and why in order to achieve real econometric identification. The contrasting pole is simply to throw in a bunch of controls until you have produced the numbers you want. In my view, we do not teach what should be controlled for and how enough, so people pick it up on the fly. Arindrajit has picked it up, and is a master: Arindrajit Dube: Minimum wages and the distribution of family incomes in the United States: "I find that a 10 percent increase in the minimum wage reduces poverty among the nonelderly population by 2.1 percent and 5.3 percent across the range of specifications in the long run...
...There are also reductions in shares earning below 125 percent and below 75 percent of the poverty threshold. For my preferred model with the richest set of controls, the falls in shares below 75 percent, 100 percent, and 125 percent of the poverty threshold are 5.6 percent, 5.3 percent, and 3.4 percent, respectively, from a 10 percent increase in the minimum wage.... Since minimum wage policies are not randomly assigned across states, it is important to account for any bias that may arise from differences across states raising the minimum wage as compared to those which do not.... There is a strong regional clustering.... Economists disagree... on the best way to account for such biases.... I report results using eight different specifications with alternative...state-level controls that subsume most of the approaches.... Starting with the classic model that assumes all states are on parallel trends (known as the two-way fixed effects model), I progressively add regional controls (division-period effects), state-specific linear trends, and state-specific business cycle effects. This exercise allows the evolution of family income distribution to differ across states in many different ways. Moreover, all of these controls have been shown to be important....
All of these specifications find that increases in the minimum wage reduce the nonelderly poverty rate.... The specification with all three sets of control... performs the best in a variety of falsification tests...
#shouldread
