I have a dataset that records COVID outcomes (cases, deaths) daily in a tally for the 50 states of the USA. I want to find the influence of various state-level continuous (e.g. percent of seniors in a state, percent of smokers in a state, GDP per capita) and categorical (e.g. whether a state is coastal) variables on these state-level outcomes, along with interactions (e.g. whether a state's GDP per capita moderates the effect of coastal-ness on a state's cases).

I've already done cross sectional univariate multiple OLS regression on the data but the issue is that the relationships seem to change over time -- e.g. some variables will be significant in the first half of 2020 but not in the second half, and vice versa. What I've done at this point, which is admittedly crude, is just take the different tallies for the cases at different dates (at the halfway point of the year, at the end of the year) and do the regressions on that (in R).

I want a method that integrates time and the fact that I have all this daily data that seems to have important statistical information that I'm not analysing at the moment.

  • $\begingroup$ Data quality is a big problem with COVID, and the quality of some types of data changed over time. See this report, for example. $\endgroup$
    – EdM
    Jul 8, 2022 at 16:50

1 Answer 1


What you are describing sounds to me dangerously close to "I want to take an imagined relationship between variables and then look for a narrow enough interval to make it true by coincidence."

I would advise you to try to define clearly how what you are doing is different from this, and perhaps that will lead you to the answer you are looking for!

  • $\begingroup$ I don't think that captures what I'm trying to do. There are very real theoretical reasons why a relationship would change over time (for example we would expect whether a state is coastal to affect cases greatly during the beginning of the pandemic and much less so later on) and I would have no reason not to explain these issues clearly in the research. For example this paper on partisanship and COVID outcomes showed the same variable (whether a governor was republican or not) had opposite effects halfway through 2020: sciencedirect.com/science/article/pii/S0749379721001355 $\endgroup$ Jul 8, 2022 at 18:17
  • $\begingroup$ @torontobizphd yeah no I get that there could be real effects in there. I was just curious how you think you can tell them apart from the spurious ones! $\endgroup$
    – kqr
    Jul 8, 2022 at 18:33
  • $\begingroup$ I mean as long as you're clear about the time periods you're talking about and are honest about them being more influential at some times rather than others I don't see what the problem would be. $\endgroup$ Jul 8, 2022 at 18:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.