Regression on time-series and cross-sectional data

[Charlie Skene]

Hi,

While writing an essay on transitions to democracy in SSA mainly using logit models for survival and emergence I included a standard regression in the appendix to just further a small point. My question regards the use of this regression. So I have data from 43 countries in SSA over the years from 1960-2016, looking at the effect of several independent variables on polity2's 20 point scale for levels of democracy. I have read that using standard regression is not allowed. I've looked it up but struggle to find explicit statistical details for why I can't run regression on cross-sectional/time-series data. As an undergraduate I would appreciate any advice possible.

Many thanks, Charlie

[Carlo Lazzaro]

Charlie:
in general you cannot use -logit- for panel data analysis, because there are estimators that can take both cross-sectional and time dimension into account, such as -xtlogit-.
You may want to take a look at -xt- and -xtlogit- entries in Stata .pdf manual.

[Charlie Skene]

HI Carlo,

Thanks for that, yeah I should've expanded on the fact that for the logit I did use xtlogit and accounted for the time dimension. I was more concerned about my use of OLS regression.

Best,

Charlie

[Carlo Lazzaro]

Charlie:
if you refer, as it seems to panel data regression, it's rare (but possible, though), that (pooled) OLS outperforms -xtreg-.
For instance, it happens when the F test appearing as a footnote of the -xtreg, fe- outcome table fails to reach statistical significance: if that were the case, switch to pooled OLS.