Dependant variable with two dimensions and explanatory variable with only one. xtreg areg or xtscc

[Jeanne Sif]

Hello,

I would like to know if it is possible to use the panel model, and hence xtreg command, on Stata if none of my explanatory variable depends on time dimesion. My dependant variable depends on time and individu (extraction). My explanatory variables (treatment_questionnaire ctb study_level study_discipline) only depend on invididu and I add a fixed effect on time.

here is my code (nb= individu, instant = time)

xtset nb instant
xtreg extraction treatment_questionnaire ctb study_level study_discipline i.instant, robust

Does areg command with absorb(isntant) is more convenient (I used it but I obtain results completly different)

areg extraction treatment_questionnaire ctb study_level study_discipline, absorb(instant) robust

I also undeline that time dimension in my sample is higher than the number of individu, this is why i also tested the xtscc command

tsset nb instant
xtscc extraction treatment_questionnaire ctb study_level study_discipline i.instant

Thank in advance!!

Best regards

J.

[Maxence Morlet]

It is all very confusing, not too sure I understand the problem, but bottom line is that if your regressors are time-invariant, you cannot use fixed-effects because they will be dropped in the transformation. The same goes for absorbed regession (areg).

Also, for the community contributed command xtscc, the ratio T/N needs to tend to infinity. What is the size of T and N in your data?

[Jeanne Sif]

Thank you for your return.

I only put a fixed effect on time, not on individu (that indeed is dropped). The dependant variable "extraction" varies with time.
In my sample T=300 and N=150

J;

[Carlo Lazzaro]

Jeanne:
welcome to this forum.
You may want to take a look at -xtgls- and/or -xtregar-.

[Jeanne Sif]

Thank you Carlo Lazzaro for your recommandations.
As I note, all these models require to have at least on explanatory variables dependant on both dimensions (time t and individu i). Given the fact that in my regression only the dependant variables evolves with time I'm not sure that the Panel approach is correct...If no, how can I consider the time dimension.

y_{i,t}= c + \beta X_i +\epsilon_{i,t}

Thank you.

[Carlo Lazzaro]

Jeanne:
1) are you dealing with a panel or a repeated cross-sectional dataset?
2) if only the regeressand changes with time, are all your predictors=K?

[Jeanne Sif]

Carlo Lazzaro, sorry my question was very confusing. I now understand my problem: the model was completly wrong because I didn't consider a key explanatory variable.
Now I'm sure to work with a panel model ! One explanatory variable depends on a stock that evolves across time and individu. But now I have another issue because with explanatory variable is a stock that evolves with my dependant variable. I have thus an autoregressive model.
I have to clarify some important points before to ask addiitonal questions...

Thank you for your understanding,

[Carlo Lazzaro]

Jeanne:
be also sure that you do not have reverse causation (endogeneity) problem.