Has anyone dealt with analysis of experiment/trial/ab test where recruitment occurs over let’s say 2 weeks and the length of exposure to the treatment depends on how early you enter the study. So for instance, 2 weeks if join the first day and 1 day if join the last. Trying to think through appropriate metrics here.
Ryan Travis
@proportional_hazard
- 8 Followers
- 39 Following
- 5 Posts
Katie BauerIt's resolution season and I'm here to offer you unsolicited advice about how to help your data team play a bigger role at your company. https://wrongbutuseful.substack.com/p/elbows-of-data
Elbows of data
It seems like it happened slowly, then all at once. For some time it was the consensus that new data tools and technologies were going to be the thing that finally helped data teams break through, get executive buy-in, and drive the successful outcomes that we all knew they could. But in the past year or so, I’ve been seeing dissent. Commentators and thought-leader-types started asking if we’ve focused too much on technology, at the cost of staying in touch with the needs of the businesses we work in. Now there’s a healthy contingent of folks calling for us to focus on
I was reading Gelman’s review of Imbens and Rubin (2015) and it’s not possible for me to disagree more. Chapter 4 starts with randomization inference and builds up to model based inference. It’s nice to see what needs to be true, rather than just assuming a random sample.
