Time dummies are the same as year fixed effects but time dummies can also be used for xtreg re regressions even though they're fixed?

[DavidHarmsen]

Good afternoon readers,

Time dummies are the same as year fixed effects (both are used as i.year) but I just read that time dummies can also be used for xtreg re regressions even though time dummies control for fixed year effects?

For example:

xtreg gender weight gender_weight i.year, re

So with the above code you would run a random effect regression but still control for fixed year effects?

When xtreg fe or xtreg re is used, the regression automatically implements firm fixed effects right? So that also sounds like contradiction, to me at least.

Can someone clarify the above?

[DavidHarmsen]

Are fixed year effects always time dummies but are time dummies not always fixed effects. So far that's my conclusion after further research.

If my first post is too unclear, I can make it more clear probably.

[DavidHarmsen]

Xtreg fe controls for fixed effects but many people still use i.year i.industry when they're using xtreg fe. Using i.year i.industry shouldn't be necessay if xtreg fe controls for all fixed effects. Am I correct? Why else should anyone want to use xtreg fe instead of OLS with i.firm i.year and i.industry?

[Richard Williams]

Fixed effects models are designed to control for the effects of time-invariant variables with time invariant effects. The value of year obviously varies across time for individuals. Firm and industry can change across time if individuals move to different firms or industries.

You may want to read up on fixed effects models. My own brief notes are at

http://www3.nd.edu/~rwilliam/xsoc739...xedEffects.pdf

http://www3.nd.edu/~rwilliam/xsoc63993/Panel.pdf .

The Allison book that my notes refer to is excellent:

http://www.amazon.com/Effects-Regres.../dp/0761924973

[DavidHarmsen]

Very useful advice! I appreciate it very much. I haven't read (and found) yet why someone should use xtlogit instead of xtreg. I read that xtlogit can be used for things besides panel data but I didn't read when someone should prefer xtlogit above xtreg if it's going to be used for panel data.

"Other methods (e.g. random effects) can be used when we want to estimate the effects of variables like sex and race, but then the method is no longer controlling for omitted variables"

If I need to use re (Random effects) for gender, race, education based on the well-known Hausman test, how can I then control for omitted variables like risk aversion, intelligence, background of parents (e.g. education/wealth)? Fixed effects model cannot be used if someone wants to analyze gender, race, education degree so both models cannot control for the omitted variables of gender, race, education degree

By the way, when should someone control for fixed firm effects? Only when individuals move to different firms or when firm-specific data is analyzed like cash holdings? Xtset automatically controls for fixed firm effecs, so it might be better to exclude the fixed firm effects. On the other hand, it probably cannot be harmful to the regression output like giving biased results. Are fixed firm, industry and year effects the only ones used or do people also use individual fixed effects when the regressions are mostly focused on things like race and gender?

For such a great helpful book I never expected a price below 50 dollars!

[Richard Williams]

Allison's book or something similar can answer most of your questions. I would read it before getting too much further in your analysis. A gentle introduction to Stata would be another good book to read if you are new to Stata.