Harmonic (logistic) Regression

Question

I can find examples of people using spectral analysis to break a time series into trigonometric components and then using those components with a multiple regression. Examples:

https://onlinecourses.science.psu.edu/stat501/node/364

Is time of the day (predictor in regression) a categorical or a continuous variable?

Fit a sinusoidal term to data

https://ncss-wpengine.netdna-ssl.com/wp-content/themes/ncss/pdf/Procedures/NCSS/Harmonic_Regression.pdf

Does this extend to other forms of regression, like logistic regression? The outcome variable I'm working with is binary, but there's a pronounced periodic trend over time if it's summarized as a proportion. I was able to identify the periodic components and run a logistic model with the same kind of trig transformations seen in the above examples. The model predicts the actual proportions in the data really well, but I'm a bit leery of this approach since I literally can't find an example of someone doing this with a logistic regression.

Is there any reason I should avoid doing this? Have you seen any examples of people doing this before?

Answer 1

There is no reason you cannot use harmonic regressors with logistic regression. This: Does a logistic regression end up modelling linear relationships? references papers doing something like that.

Without knowing more of your context we cannot say much more, the details of how to implement this might vary depending on sample size and other details. If you add more context I can try to say more.