Multilevel modeling and time variables.

[Nikos Korompos]

Hi,

I have a model, in which most variables are individual-level, however, one variable changes depending on the month of birth (so it's a kind of ecological variable). Do you think I could employ a multi-level model nested in month and then in year of birth, or is this completely irrelevant?

Best,

Nikos

[Clyde Schechter]

Well, there are at most 12 levels to the month variable (if I understand you correctly). So you are talking about creating a level in a multilevel model for a variable whose entire universe has only 12 categories. I wouldn't say you can't do it, but it seems to me to make more sense to just include i.month among the regressors.

In any case, month would not be nested in year, it would be crossed with it, assuming that January in 2000 is to be considered equivalent to January in 2001, etc. Again, depending on how many different birth-years are represented in your data, you might be better off simply including i.year_of_birth among the regressors More important, the use of either i.year_of_birth or a year_of_birth level in a model implies that the effect of year of birth is a haphazard shock to the outcome variable. That may be a reasonable model, but if you have a large number of years and long time periods, then, depending on what your outcome variable is, modeling of continuous time trends might make more sense. Really it all depends on what these variables are and how they can reasonably be assumed to vary over time.