(1/n)
When I was a young scientist-in-training, I learned that confounder = variable related to both treatment and outcome.
It's not wrong, but...
(2/n)
Feb 22, 2019 • 5 tweets • 2 min read
Hello again, fellow diff-in-diff enthusiasts.
I see you out there living your best quasi-experimental life.
But causal inference is still so HARD.
Which causal assumptions are necessary? Reasonable? Should we match treated and control? Test for parallel trends? (1/5)
Include unit and time fixed effects? Fit a non-linear model? Adjust for potential confounders? How should we pick comparison groups? Are synthetic controls better? What about permutation inference? (2/5)
Jul 27, 2018 • 18 tweets • 9 min read
Do you use diff-in-diff? Then this thread is for you.
You’re no dummy. You already know diverging trends in the pre-period can bias your results.
But I’m here to tell you about a TOTALLY DIFFERENT, SUPER SNEAKY kind of bias.
