In yesterday’s comments, Mai Naem mobile asked a great question:
No doubt the opposition from this administration didnt help but isnt it possible that the improving economy and lower unemployment rates contributed a little to the decreased applications.
Fundamentally what time scales do we look and expect for things to change?
In the study that Paul Shafer and I conducted, we expected enrollments to change on a weekly basis. Everything else we thought (with some justification I think) would be either immutable (urban percentage, Trump vote percentage etc) or changes so slowly over the course of a fourteen week open enrollment window (economy/labor market) to be assumed to be a constant for a single open enrollment period. We fixed everything other than enrollment as a constant within the 2016 or 2017 open enrollment period while allowing for whatever variation could happen between open enrollment periods.
This choice makes sense for this study. We cared that the trends pre-January 20, 2016 and 2017 were constant and we were just looking for deviation from trend in 2017.
Other questions will require different assumptions. I think the question that Mai Naem is poking towards: “What explains variation in enrollment year over year” requires significantly different assumptions about changes. I’m working on a manuscript, that we need to get done in the next twelve days, that looks at variation in enrollment within counties over a six year period. There, we need to assume a lot more changes happen year over year; states expand Medicaids, states move to Healthcare.gov or to their own exchanges, states Broad Load or Silver Load. Those decisions happen every year and have to be accounted for every year.
Another question that has different time variant qualities is “what factors will lead to 2020 enrollment being different than 2019 enrollment?” Here, the economy is a major factor (along with Medicaid expansion). All else being equal, we should expect a lower unemployment rate, a higher prime age labor force participation rate and increasing median compensation to lead to a smaller pool of people eligible for ACA enrollment. That is something that can be notably measured year over year while doing that type of analysis week over week may not produce enough movement to be worthwhile.
Figuring out what things move, and at what time scales is one of the harder analytical challenges. It is a damn good question.