... 1990-2006 data or 1990-2016 data. It was a mistake, and DRBS show no evidence to the contrary, to use DLR's evidence to argue that higher MWs don’t reduce employment, and even more to argue that DLR's results were more credible and should supplant the earlier evidence. 19/19

To sum up, DRBS throw spaghetti at the wall to see what sticks, even at the cost of burying DLR, which used to be Dube’s and Reich’s favorite child. One fact remains: the border discontinuity approach shows disemployment effects when using MSCZs, whether with DLR’s... 18/19

If DRBS is so concerned about heterogeneous/dynamic treatment effects, why do they assume that minimum wage effects remain the same every year up to 6 years after treatment? 17/19

…in federal MW states. With employment declining in the “clean control,” DRBS estimates are likely biased toward zero.
Another reason their “clean diff-in-diff” approach may not be as clean is that every event has a different post-period length, ranging from 1 to 6 years. 16/19

DRBS uses federal MW states as clean controls and excludes from their windows years of MW increases one year before the event. However, evidence (Clemens & Wither, 2019) shows that minimum wage changes have dynamic effects extending for a few years... 15/19

… and their subset of events does not even align with Cengiz et al.—10 of their 40 events for the 1990-2016 period are “too small” to appear in Cengiz et al. 14/19

In a final attempt to salvage DLR’s result—though not DLR’s border design—DRBS introduce a “clean” event-study design riddled with subjective choices in defining events and control groups. They make ad hoc decisions about bundling events, ignoring federal MW increases, ... 13/19

If we calculate the own-wage elasticity (OWE) and focus on the specification that uses log population weights, note that it barely changes from −.241/.165 = −1.461 for the 1990-2016 period, to −0.170/.114 = −1.491 for the 2000-2016 period. 12/19

… or log emp weights—which ensure that results are not disproportionately influenced by outliers—again shows significant disemployment effects. We discussed log weights in our paper, so it’s surprising (or maybe just cherry-picking) that DRBS ignored these results. 11/19

In our view, DRBS’s only important critique of our work is that the MSCZs pair-period estimation yields weaker employment *and earnings* effects of minimum wages for the 2000-2016 period. What DRBS does not show, however, is that weighting matters. Using log population... 10/19

… argue that only pre-1993 data are affected by parallel trends violations. In our paper we showed that our results are robust to excluding these years. 9/19

DRBS’s concern with data from the full 1990s decade is new—and seems to have arisen only in response to our replication, as a goalpost-shifting strategy. Cengiz et al. (2019), which DRBS now considers the benchmark in minimum wage studies, … 8/19

… the relevant variable in DLR and JNR is log employment, and iii) with our CBP data, log emp in their “ever treated” and “never treated” groups follow similar trends, with a 99.8% correlation in levels and a 95.7% correlation in one-year first differences (see plot in levels).7/19

Even if we assume that the comparison is valid (it’s not), their analysis is deficient because i) they don’t even bother to do the analysis with our data, ii) they pollute the analysis by looking at emp-to-pop ratios, which are not justified because... 6/19

Their parallel trends analysis is invalid. Their comparison of “ever treated” vs “never treated” (federal MW states) groups makes no sense because, in the pair-approach analysis, federal MW increases (six since 1990) generate minimum wage differentials within MSCZs and BCPs. 5/19

… in the 1990s, (2) that removing 1990s data yields no significant disemployment effects, and (3) that a “clean” diff-in-diff event study again shows no disemployment effects. We tackle each of these below. 4/19

Instead of engaging with our main point, DRBS throws DLR under the bus (saying “it is best to dispense with TWFE-style regressions,” including DLR’s approach) and (1) claims, without using our provided data, that our results are a consequence of parallel trends violations… 3/19

DRBS ignores the central issue of our replication: that changing the definition of local areas from border county pairs (BCPs) to multi-state commuting zones (MSCZs)—using DLR’s border discontinuity design and data—yields significant disemployment effects of minimum wages. 2/19