Panel data - Fixed effect individual slope

[Cheol-Min Kang]

Dear Statalist,

I am currently using Stata 13, working on a panel data set.
I have used xtreg,fe command and am content with the results.

I would like to, however, in a next step, try to estimate fixed effect individual slope model that is described in Wooldridge(2010) in chapter 11.7.2
I have seen there is a xtfeis command by Volker Ludwig that has been presented in the 2010 German stata Users meeting.
But I cannot find the ado files.

How can I do this in Stata?
I rely upon the kindness of strangers. A kindness much appreciated. Best regards,

Kang

[Carlo Lazzaro]

Kang:
I would e-mail the author directly at the e-mail address reported in the front page of his slide presentation on this topic (please, see http://www.stata.com/meeting/germany...y10_ludwig.pdf).

[Cheol-Min Kang]

Does anyone know if there is another command to estimate fixed effect individual slope that is described in Wooldridge(2010) in chapter 11.7.2 ?
Any help would be really appreciated.

[Sergio Correia]

Hi Cheol-Min,

I don't have my copy of Wooldridge in hand; could you tell us what exactly is that model doing? Is it just a fixed effect ols plus a different fixed slope per group? or a random or mixed model or something more complex?

Best,
S

[Cheol-Min Kang]

SergioCorreia:

Thank you for your reply.
I would like to estimate fixed effect model that allow for different slope per group.

Regards, Kang

[Sergio Correia]

Hi Kang,

If as you said it's just a fixed effect model, you can use -reghdfe-.
For instance, using the usual auto dataset:

sysuse auto
reghdfe price weight , absorb(i.foreign##c.length)
reg price weight i.foreign##c.length

reghdfe, as shown in the second line, basically generalizes areg so you can include fixed intercepts (i.foreign) as well as the slopes with respect to the variable length. Note that i. and c. are used to indicate which is the group and which is the variable related to the slope. Also note the ## which in stata notation means to add both intercept and slope (# would mean just slope).

The third line is how you would do it manually with regress just to check that it gives the same result.

To install it, I would advise you to get it with the code below, as works better than the SSC version.

cap ado uninstall reghdfe
net from https://raw.githubusercontent.com/sergiocorreia/reghdfe/master/package/
net install reghdfe

HTH,
Sergio

[Cheol-Min Kang]

Dear Sergio

I really appreciate your help.
I have one more question.
By using "reghdfe", is it possible to estimate slopes of each group and the average slope of entire group?

Regards, Kang

[Sergio Correia]

You can always take the average of the slopes for whatever purpose you have (do you want to test if the individual slopes are needed?), but given that these slopes are individual fixed effects, the concept of the avg slope as a parameter doesn't make that much sense to me.

That said, I know -areg- and even -reghdfe- will report an intercept (check the FAQs for how to interpret it) so maybe in the future it would be interesting decomposing the average slope from the individual deviations from it (and report an F test of those).

[giuseppe piroli]

ciao Sergio,
non capisco come ottenere le slope individuali con reghdfe

[Carlo Lazzaro]

Giuseppe (in English, please!)
is wondering how to get individual slopes via -reghdfe-.

[giuseppe piroli]

you're right Carlo, thanks.
Giuseppe

[Sergio Correia]

Sorry for the delayed reply. You just need to use something like .. absorb(id##c.time)

Example:

sysuse xtline1
reghdfe calories, a(person##c.day, save)