normal R2 versus adjusted R2 and Within R2 versus Areg R2

[Victoria Rogers]

At the moment, I'm dealing with the following problem:

In general it's better to use adj. R2 instead of (normal) R2.
However, the adj. R2 isn't given after using -xtreg fe- or -xtreg re-
After searching for several hours I couldn't find a solution for that, only found how to calculate it by hand and I need to run lots of regressions, so I'm inclined to use the (normal) R2 (unless someone is willing to give a solution how to get the adj. R2 in Stata). Maybe someone can explain why the adj. R2 isn't given, considering that it's a lot better than the (normal) R2.

In case of -xtreg re- I must use the overall R2. In the case of -xtreg fe- some sources claim that people must use the within R2, while other sources claim it's better to use the R2 of areg because it's also based on the variance explained by dummies (on contrary with the within R2 of -xtreg fe-). What's considered to be the best method these days?

[Carlo Lazzaro]

Victoria:
you may want to take a look at Akaike's information criterion and Bayesian information criterion by typing -estat ic- after performing panel data analysis; -help estat ic- will point you to the related entry in Stata 13.1 .pdf manual.

Kind regards,
Carlo

[Carlo Lazzaro]

Victoria,
in my previous reply I forgot to highlight that -estat ic-

is not appropriate after xtreg with the be, pa, or re option

but only after xtreg, fe (please, see -help estat ic -).
Kind regards,
Carlo

[Victoria Rogers]

I received the error code 321: "likelihood information not found in last estimation results" & after clicking on the error "requested action not valid after most recent estimation command" ...I used:

xtreg DV IV, re 
estat ic

[Carlo Lazzaro]

Victoria:
please see my previous # 3.

KInd regards,
Carlo

[Victoria Rogers]

Thank you Carlo, somehow I didn't see your post even though I always refresh the page before I reply.

I will use/try it after -xtreg fe-
I tried it without success, could you explain me how I can get the adjusted R-squared with -estat ic- after the regression?
On the Internet I also saw that other researchers had the same problem:

According to Maarten L. Buis, who posted a lot on statalist:
"Stata does not accept neither "estat" alone nor "estat ic" after svy and I could not find an "estat (svy)" command which would produce any of these statistics (respectively AIC/BIC for the ordered logit; AIC/BIC/adj. R-squared for the linear regression)"


Hopefully someone knows how to get the adjusted R2 in case of -xtreg fe- and -xtreg re-

[Victoria Rogers]

...and how to get the adjusted R-squared after a regression with robust standard errors. (R2=R-squared in my previous post)

[Jeph Herrin]

You may find this thread helpful

http://www.statalist.org/forums/foru...random-effects

[Victoria Rogers]

@Jeph, thank you for the suggestion but that thread isn't about the adjusted R-squared.

The normal R-squared is pretty useless, because it always increases after adding more variables.
Does noone know how to get the adjusted R-squared?

Hopefully someone knows how to get the adjusted R-squared in case of -xtreg fe- and -xtreg re-.

[Aspen Chen]

I don't know if there is a simple answer concerning the use of R-squared values for random effect estimators. But for fixed effect models, you can find the adjusted R-squared values after the estimation by typing -ereturn list-. The value you are looking for is in the scalar e(r2_a).

[Victoria Rogers]

Thank you very much Aspen. Could you also explain me why that adjusted R-squared isn't just given in the regression output? For example, above the 3 (normal) R-squared (within, between, overall)?

Could someone also help with the adjusted R-squared for the random effect models?

[Aspen Chen]

No. I did not write the program and decide what to display, so I am afraid I can't.

But you can always access the value as a stand-alone scalar like other scalars. Or if you have multiple models and like to summarize them, -esttab- can incorporate the adjusted R-squared with the ar2 option.

[Victoria Rogers]

Thank you again.

Could someone also help with the adjusted R-squared for the random effect models?

[Carlo Lazzaro]

Victoria:
unfortunately, I did not find any guidance about ths topic.
Kind regards,
Carlo

[Victoria Rogers]

Ok thank you for trying Carlo. I've also searched again without succes. It's strange that there's no publicly available information about the adjusted R-squared for random effect models.