Time fixed effects

[evilbert]

I am estimating a model using STATA and want to use country and time fixed effects. I'm not having any trouble with the standard xtreg fixed effects for country but i was wondering if there is a quicker way to use time fixed effects rather than entering dummy variables for each year?

also wouldn't this reduce the power of the test if i've got say 40 dummy variables?

any advice would be much appreciated

[breakz]

Not sure about reducing power of the test, but if you want to create dummies for panel data (looks like you do, countries and years) use the 'xi' command.

Check the help file for more info, but if you've got a variable coded as an explanatory variable (call it indvar = {1, 2,...,10}), you simply type it as i.indvar and STATA will create 9 dummy variables (and drop one, obviously).

[evilbert]

cheers for that...

i managed to carry out my regressions using a series of time dummies and country fixed effects.

I was wondering if anyone might have an interpretation of the weakness of results if the panel data has about 500 observation and after including the time dummies, there are about 40 explanatory variables. I would think myself that such a large amount of independent variables would bias the result somehow...at least providing an artificially high r-squared.

essentially, i suspect the results to be somewhat off the mark & am wondering if the use of so many explanatory variables (about 35 time dummies) might be the cause of my strange results

any ideas?

[notacolour]

Smaller sample size and a greater number of independent variables will decrease the likelihood of finding statistically significant results, but should not (when all other assumptions are met, which of course is never true) bias the results.

Measures of goodness of fit aren't useful at all for evaluating a model; one reason is what you said: tossing in lots of fixed effects is going to inflate your R-squared.

Are your results statistically significant? Do they change significantly when you add time or country fixed effects? Looking into that might help you understand what's going on.

[evilbert]

the main problem is that when time fixed effects are used variables become too significant if that makes sense. vairables that were insignificant without using time fixed effects have become greatly significant and often change sign as well.

when using the time fixed effects i also end up up completely contrasting results (all significant) depending on whether Panel Corrected Standard Errors are used or if default stardard errors are used. Such huge differences between the 2 standard error measures are not as present when the time dummies are excluded....

i'm at a bit of a loss to try and explain why

[notacolour]

This is pretty much completely off-topic for this board (given that it has no relation to admissions whatsoever), so I'll send you a PM about this.