Show this thread
Repeat after me: staggered rollout designs don't provide causal identification in cases of obvious endogenous enactment.
-
-
-
Think about what the parallel trends assumption in your staggered DiD's mean. It's not that the trend won't break, it's that there would be parallel outcome paths without the treatment. If you're working in an endogenous adoption setting you need to prove why that's the case.
Show this thread -
Take, for example, the setting of voter ID laws. Assume that the party that prefers lower turnout (shocker which one) adopts such laws when she demographic trends turning against their favor in present or upcoming elections. No change is consistent with an effect!
Show this thread -
New conversation -
-
-
Matching on the pre-period outcome trends though would fix this issue though right? Doesn’t fix the potential for confounding policies however. Kind of interesting because I feel like IR gets shat on for doing this but somehow there are tons of AP papers that get away with it.
-
Matching on pre-period outcome trends would not fix this is the policy is adopted in anticipation of forward changes.
-
Also in that case you would just pick it up with the pre-trends most likely.
-
Right, parallel trends is about the **post-period** trends being parallel and not the pre-period. But matching on pre-period trends would be sufficient if we assume that pre-period trends are what actors are making their decisions off of.
-
Probably right, if they're expectations are correct.
-
In general I'm in support of matching on pretends.
End of conversation
New conversation -
-
-
I've been saying we should randomly assign policies but no state legislators will listen to me.
-
Me too! Would be harder at state level. But right now the way admin law works is you have to prove ex ante cba of regulations, which is mostly not possible given the expertise in government. Swap that out with some real random assignment and strict ex post cba review in my view.
- 1 more reply
New conversation -
-
-
agree 100%, but yet so many well-published papers get away with it!!
- End of conversation
New conversation -
-
-
I say this all the time. This is basically an example of parallel trends being violated bc absent the entry of whatever you’re studying, expected potential outcomes would’ve diverged anyway.
-
So unless I'm not understanding this correctly, that would affect this paper, no? Asking because I just taught this yesterday https://www.journals.uchicago.edu/doi/abs/10.1086/694293?mobileUi=0 …
-
There us far more subtlety in the identifying assumptions of diff in diff that we acknowledge. Even “accidental” laws like I use in my RI study doesn’t guarantee parallel trends. Maybe rapes were going to fall anyway.
-
Even if you don't believe her D-in-D, she still showed (quite persuasively) that the birth control pill did not lead to the outcomes previously thought.
-
Right - the recoding of the laws and the endogeneity of birth control to contemporaneous abortion policy seems like a major contribution to me
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.