The need for Time Fixed Effects

[Ahmad Alsaraireh]

Hi Everyone,

I am Using Fixed Effect Model and i am confused about adding time fixed effect or not. the model i ran is xtreg y x1 x2 x3 i.year, fe and i have found the year dummies to be highly significant (at .01). Furthermore, when i use testparm i.year command, i have found prob of F to be 0.000 meaning that they are not different from zero. in this case, can any one please help me with the next step ( Shall i include them in my model or not? )

[daniel klein]

[...]i have found prob of F to be 0.000 meaning that they are not different from zero.

The Null states 1.year == 2.year == ... ==k.year == 0, so a significant result suggests the coefficients are jointly different from zero - which matches the t-test results you report. The decision whether or not to include the variables in the model should depend more on theoretical reasoning than on the result of a significance test. We cannot give advice without knowing what y, x1, x2, etc. are and what the scientific question is that you are trying to answer here.

Best
Daniel

[Ahmad Alsaraireh]

Hello Daniel,

Thank you for your Note, actually, i am trying to answer the following question, does firm family ownership affect firm performance? and the y is the Tobin's Q, Family ownership is a dummy variable that equals to 1 for the family firm and 0 otherwise. moreover, i have included some control variables, such as size, growth, leverage, ROA, and tangibility.


Many Thanks
Ahmad

[Marcos Almeida]

Hi "ahmadsar",

Please re-register with name and family name (click on the "contact us" button below to the right).

I got the impression your question demands a review on the main assumptions and principles related to panel data, and I fear that would be unfeasible to summarize in just a short message (at least to me).

That said, I don't know how you "xtset" your data, but you can "xtset" including a time variable ("year") this way:

. xtset id year, yearly

Concerning fixed versus random effects, I gather you should do the "modeling" task and test for both. Also, dutifully perform the postestimations.

When working with the modeling, you can insert in the "fixed section" some time varying covariates ("years of ownership" at each Tobin's Q, for example), of course.

Best,

Marcos

[Ahmad Alsaraireh]

Hi Marcos,

Actually, i did the xtset the way you mentioned. and after sunning both fixed and random effect i ran hausman test to choose the right model and it gave me that i should use the fixed effects. after that i included the year dummies that are 14 years (minus one) and ran the testparm command, in this command i got that all time effects are jointly significant. does that suggest that i need to include them in my model? and if yes, as far as i know, i will not be able to include any macro variables in the model as they don\t vary across firms. So, how can deal with this problem?

Many thanks for your help

Ahmad Alsaraireh

[Marcos Almeida]

In fact, you haven't yet fully showed the commands, neither your results in Stata...

That said, if you are mainly interested in evaluating the effect of "year", perhaps you could perform the commands "margins" and "marginsplot", like that:

. margins, dydx(family_ownership) over(year)
. marginsplot

Hopefully it helps!

Best,

Marcos

[daniel klein]

Which variables to include and which not to include depends on the mechanism you have in mind for the "effect" for family ownership. I interpret your question as being manly interested in the causal effect of family ownership, that is the coefficient of the respective indicator variable. IN that case, and if you are running a linear model, you should control for anything that varies within firm and predicts the outcome and is correlated with family ownership. Note that you only need to control for those predictors that satisfy all three conditions Any predictor that is not correlated with family ownership, cannot mess up inference in a linear model, and thus, need not be controlled for. You should probably neither control for predictors that are consequences rather than causes of family ownership in a first step.

Best
Daniel