![]() If possible, researchers show that outcomes in the treatment and the control group prior to the treatment moved in parallel, which supports the assumption of parallel trends over the introduction of the treatment. The plausibility of this identifying assumption depends on the specific setting to which DiD estimation is applied. The assumption that the treatment class would have experienced a counterfactual achievement gain identical to the observed achievement gain in the control class is illustrated by the dotted line in Fig. 1.2. In terms of the hypothetical example, the identifying assumption is that both classes would have experienced the same increase in test scores over the school year in the absence of afternoon lessons. The identification assumption of the DiD approach is that the group-specific trends in the outcome of interest would be identical in the absence of treatment. The remaining difference between these group-specific differences must then reflect the causal effect of interest. However, as long as this difference is constant over time (in the absence of treatment), it can be differenced out by deducting group-specific means of the outcome of interest. That is, the group-specific means might differ in the absence of treatment. The two groups might be observationally different. The idea behind the DiD identification strategy is simple. Stylized exposition of identification in the DiD model. The Mariel Boatlift study uses the comparison cities to estimate the counterfactual average, E, i.e., what the unemployment rate in Miami would have been if the Mariel immigrants had not come.įig. 1.2. In practice, we know that the Mariel immigration happened in Miami in 1980, so that the only values of E, we get to see are for c = Miami and t > 1980. The unemployment rate in city c in year t is E, with no immigration wave, and E if there is an immigration wave. As in the union example, let Y 0 i be i’s employment status in the absence of immigration and let Y 1 i be i’s employment status if the Mariel immigrants come to i’s city. The rationale for this double-differencing strategy can be explained in terms of restrictions on the conditional mean function for potential outcomes in the absence of immigration. Standard errors are shown in parentheses. An important component of this identification strategy is the selection of comparison cities that can be used to estimate what would have happened in the Miami labor market absent the Mariel immigration.Ī Notes: Adapted from Card ( 1990, Tables 3 and 6). In particular, Card asks whether the Mariel immigration, which increased the Miami labor force by about 7% between May and September of 1980, reduced the employment or wages of non-immigrant groups. As in our earlier examples, the object of research on immigration is to find some sort of comparison that provides a compelling answer to “what if” questions about the consequences of immigration.Ĭard’s study used a sudden large-scale migration from Cuba to Miami known as the Mariel Boatlift to make comparisons and answer counterfactual questions about the consequences of immigration. See Friedberg and Hunt (1995) for a survey of research on this question. Anecdotal evidence for this claim includes newspaper accounts of hostility between immigrants and natives in some cities, but the empirical evidence is inconclusive. Some observers have argued that immigration is undesirable because low-skilled immigrants may displace low-skilled or less-educated US citizens in the labor market. The DD approach is explained here using Card’s (1990) study of the effect of immigration on the employment of natives as an example. An early example in labor economics is Lester (1946), who used the differences-in-differences technique to study employment effects of minimum wages. The DD method has been used in hundreds of studies in economics, especially in the last two decades, but the basic idea has a long history. This approach, which is transparent and often at least superficially plausible, is well-suited to estimating the effect of sharp changes in the economic environment or changes in government policy. Krueger, in Handbook of Labor Economics, 1999 The differences-in-differences (DD) modelĭifferences-in-differences strategies are simple panel-data methods applied to sets of group means in cases when certain groups are exposed to the causing variable of interest and others are not.
0 Comments
Leave a Reply. |