Triple differences part 5: Presenting the event study plots
Corrected code and suggested practice
A few months ago, I was writing about triple differences (here, here, here, here). And there was a slight error in the code that I generated so I wanted to show it again now updated.
If you’ll recall, I said there were two schools of thought about the assumptions necessary for identification with triple differences. One said that you needed two parallel trends assumptions — in other words, you needed two unbiased diff-in-diffs. But if you have an unbiased diff-in-diff, you don’t need to run triple differences. You just run diff-in-diff and perhaps then show the placebo diff-in-diff as a falsification. There isn’t really anything to gain other wise because the triple difference is just the true diff-in-diff minus the placebo diff-in-diff, and if the placebo diff-in-diff is unbiased, you’re subtracting out a zero.
The other school of thought is that you need only need one parallel trends assumption with triple differences and that’s a “parallel biases” assumption. That is, the true…
Keep reading with a 7-day free trial
Subscribe to Scott's Substack to keep reading this post and get 7 days of free access to the full post archives.