In my work, I used OLS model to run the regressors on Return on Equity.And the result is as below
While variable of interest is pt, which is post*treatment in difference-in-difference setting. While column (1) run pt on ROE without any other control variable,column (2) adding some firm and country control variables, (3) adding some industry control variable, (4) control for industry*year fixed effect, (5) control for region*year fixed effects, (6) running for non-US sample. All regressions are ran with firm and year fixed effect if not stated elsewhere.
As can be seen from the Table above, the sign of variable of interest pt flip in column 5, when I run the regression using firm and year x region fixed effect? I am wondering what does it mean and how to deal with that from your experience ? From this topic, controlling for year seems to be more strict compared to yearxindustry or yearxregion, so can we ignore the result of column (5) and (4) while we already had the result in column (2)? Normally, flipping in sign can be caused due to multicolinearity or Simpson's Paradox, but in this case, it is caused by controlling for year*industry fixed effect.
And whether in my case, the result is driven by US sample based on column 6 in this Table (column 6 has the same list of variable in column 2)?