So, I have a relatively large panel dataset (26,000 observations) with data on speech by individuals belonging to political parties. The data is grouped by individual-week (i.e. an observation for individuals for each week) over the course of a number of years. I am interested in the effect of time (particularly proximity to particular events) on the type of speech used. At this stage I am probably just going to use a simple linear regression.
I am going to use individual fixed-effects to take into account unobserved heterogeneity across individuals, however, I'm wondering at which level to cluster my standard errors. I think it makes sense to cluster my standard errors by political party given the likely within-party correlation but I'm wondering if I should also (or) cluster them by individual?
My limited understanding suggests I should probably also cluster my standard errors at individual level but I'm not sure....
