Linear mixed model overfitting

Question

I am trying to determine whether it is statistically sensible to model a single outcome variable as predicted by a predictor with a high number of repeated measures.

For example, the outcome variable for each subject is a single numeric value. The predictor variable has 80 values related to the one outcome variable.

The time predictor variable is based on a time series, and the idea was behind this was too avoid data loss about the time series due to mean aggregation. However, I am wondering if this approach of maintaining the predictor across time only confounds the model further. Even if I were to include some dummy variable (time) to account for each of the 80 events, wouldn’t the model likely be overfit?

Thanks in advanced!

Answer 1

If the predictor variable is a time series (autocorrelated), but your observations are created by independent events, you should have no residual autocorrelation, and thus an OLS should be fine, no need to do anything.

Note that with high autocorrelation in a predictor, any model misfit will result in residual autocorrelation, so if you find this check for misfit.