Open yulinHAO1995 opened 5 years ago
Thanks! You can always cluster at district level, no clustering makes little sense.
Are those regressions at the individual level? It would probably make more sense to run them at the aggregated country wave level. For outcome, I have to think again. Initially I though about the continuous propaganda indicator as the outcome , weighting number of people with how many of the technologies they observed. This would be problematic if the number of people varies between districts. Is that the case? If yes, we should use the share of people who report at least one fighting or propaganda technology as the respective outcomes.
Besides, I like the regressions. Maybe we should try with different outcomes. The share with at least one. But also the first option, with district FE it might not be problematic if the number differs
Generally it looks like military activity is influenced much more. That is interesting. It might be that for propaganda, we also need to distinguish more by category. But lets see once we know about the issues above. Thank you!
Are those regressions at the individual level? It would probably make more sense to run them at the aggregated country wave level. For outcome, I have to think again. Initially I though about the continuous propaganda indicator as the outcome , weighting number of people with how many of the technologies they observed. This would be problematic if the number of people varies between districts. Is that the case? If yes, we should use the share of people who report at least one fighting or propaganda technology as the respective outcomes.
1.) lets get started with just fixed effects. You can then also add all variables that we consider exogenous (wheather, e.g.) or time-invariant variables like ruggedness and elevation, interacted with the wave dummies (to allow for them having a different effect at different times). I rather have only fixed effects in there first, controls make things complicated. We can add them to the main spec. Time trends might be interesting, like a district specific linear time trend 2.) Based On your descriptive statistics, it looks like there is much more variation for individual technologies then when we aggregate them. So we might also want to run this for the individual technologies
I constructed 2 measures, leading to competing stories. tables can be found in : Data\Stata\tables