[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: Fixed effects with reference category vs. average fixed effects
in a dyadic panel with 20 destinaton, 100 origin countries and 18 time periods I get fixed effects that are all statistically significant if I do reg y x cdum1-cdum20, robust cluster(origin). One destination dummy will be omitted and the fixed effects should represent the difference from the omitted reference category (the one with the highest fixed effect in my case). If, instead, I compute the average fixed effects, subtract this from the original dependent variable to get y*, and then re-run the regression (following the advice of David Drukker, Stata Corp., from 6 Sept. 2002) as reg y* x cdum1-cdum20, nocons robust cluster(origin) then the fixed effects represent the deviation from the average fixed effect. However, all the 20 cdum variables are now clearly statistically insignificant. I don't seem to have made a mistake, but I don't understand how the fixed effects can be highly significant relative to an omitted reference category, but then clearly insignificant relati!
ve to the average effect.
As usual, any help highly welcome. Thanks, Eric
* For searches and help try: