This is amazing post that I’ll need to read more than once! Thank you!
I have a very minor nitpick. MySpace was probably more important than you make it sound:
“After the acquisition, MySpace continued its exponential growth. In January 2006, the site was signing up 200,000 new users a day. A year later, it was registering 320,000 users a day, and had overtaken Yahoo! to become the most visited website in the United States. ComScore said that a key driver of the site's success in the US was high "engagement levels", with the average MySpace user viewing over 660 pages a month.”
IIUC, diff in diff assumes parallel trends, meaning that mental health trends in “thefacebook” colleges would have been parallel to trends in non-“thefacebook” colleges if it had not been for “thefacebook” arrival.
To support this assumption, you write “All of them show flat pre-trends, a common heuristic to support parallel trends in Y(0) post treatment.”
However that’s not sufficient to prove parallel trends. That’s where you would have liked more falsifications.
Amazing information
This is amazing post that I’ll need to read more than once! Thank you!
I have a very minor nitpick. MySpace was probably more important than you make it sound:
“After the acquisition, MySpace continued its exponential growth. In January 2006, the site was signing up 200,000 new users a day. A year later, it was registering 320,000 users a day, and had overtaken Yahoo! to become the most visited website in the United States. ComScore said that a key driver of the site's success in the US was high "engagement levels", with the average MySpace user viewing over 660 pages a month.”
IIUC, diff in diff assumes parallel trends, meaning that mental health trends in “thefacebook” colleges would have been parallel to trends in non-“thefacebook” colleges if it had not been for “thefacebook” arrival.
To support this assumption, you write “All of them show flat pre-trends, a common heuristic to support parallel trends in Y(0) post treatment.”
However that’s not sufficient to prove parallel trends. That’s where you would have liked more falsifications.
Do I understand correctly?