Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: st: RE: Adj-R2 becomes too low with XTIVREG2?


From   "Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: RE: Adj-R2 becomes too low with XTIVREG2?
Date   Fri, 27 Mar 2009 13:51:08 -0000

Erasmo,

You've got a good question here.

The rationale for -xtivreg2-'s behaviour is straightforward. The formula for adjusted R2 is:

Adjusted R2 = 1- (1 - R2 )((n - 1)/(n - k - 1))

The VCV doesn't figure in this, so I don't see why the dof adjustment for adjusted R2 should change simply because the user asked for a cluster-robust VCV.  Thus -xtivreg2- reports the same R2 and adjusted R2 whether or not -cluster- is specified.

Why official -xtreg, fe- should change the dof adjustment when someone changes the way the VCV is calculated - but not the way R2 is calculated! - is a mystery to me.

Cheers,
Mark

> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu 
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of 
> Erasmo Giambona
> Sent: Friday, March 27, 2009 1:15 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: Re: st: RE: Adj-R2 becomes too low with XTIVREG2?
> 
> Dear Statlist,
> 
> Sorry to be here again, but I think this might be interesting 
> to everybody.
> 
> I think I have actually found the source of the apparent discrepancy.
> It turns out that xtreg does not include the number of groups in the
> denominator of the formula to calculate the adj-r2 if one clusters at
> the group level. xtivreg2 always includes the number of groups in the
> denominator of the formula whether or not one clusters at the group
> level.
> 
> I really hope Mark Schaffer could comment on this.
> 
> Regards,
> 
> Erasmo
> 
> 
> 
> On Fri, Mar 27, 2009 at 10:39 AM, Erasmo Giambona 
> <e.giambona@gmail.com> wrote:
> > Dear Statalist,
> >
> > I posted a message a few days ago about the Adj-R2 with xtivreg2.
> > Following a suggestion of Nick Cox, I posted a code (and data) that
> > exactly replicates my commands.
> >
> > I will try to reformulate my question with the hope that 
> somebody can
> > provide some suggestions on how to interpret my output. 
> Here it is: I
> > fit exactly the same model with xtreg and xtivreg2. R2 and 
> Adj-R2 are
> > about 0.20 with xtreg. R2 and Adj-R2 are respectively about 0.20 and
> > 0.02 with xtreg and xtivreg2. I am puzzled by the drop in the Adj-R2
> > in this latter case. My question is: could this drop by explained by
> > the poor properties of the Adj-R2 with IV regressions?
> >
> > I would also be happy to pass my data and code to anybody who can
> > provide further insights.
> >
> > Thanks and regards,
> >
> > Erasmo
> >
> >
> >
> >
> > On Wed, Mar 25, 2009 at 6:19 PM, Erasmo Giambona 
> <e.giambona@gmail.com> wrote:
> >> Here is the exact code I am using based on data (also 
> attached) from
> >> help q_cross.
> >>
> >> Thanks,
> >>
> >> Erasmo
> >>
> >> On Wed, Mar 25, 2009 at 5:46 PM, Nick Cox 
> <n.j.cox@durham.ac.uk> wrote:
> >>> I am just asking for more information on behalf of those who might
> >>> answer this question.
> >>>
> >>> I understand that your dataset is too big to post here, 
> but that's all
> >>> the more reason to strain to reproduce the behaviour you report in
> >>> something mutually accessible, say one of the datasets 
> downloadable via
> >>>
> >>> . help q_cross
> >>>
> >>> I can't see anywhere in your reply the precise commands 
> you used. That
> >>> means a do file, or equivalent. It means exact code. It 
> means something
> >>> reproducible. You don't need more "hints": just do what I say!
> >>>
> >>> Nick
> >>> n.j.cox@durham.ac.uk
> >>>
> >>> Erasmo Giambona
> >>>
> >>> Thanks Nick.
> >>>
> >>> It is a very large dataset, but I would still be happy to 
> pass it to
> >>> Mark Scaffer.
> >>>
> >>> In any case, I will also try to provide more details about the
> >>> commands and the data.
> >>>
> >>> I am using an unbalanced panel dataset of firms over a 
> period of about
> >>> 10 years. For both xtreg and xtivreg2 I am fitting 
> exactly the same
> >>> model. In both cases, I cluster the standard errors at 
> the firm level
> >>> and I use fe i(firm). In the iv-model, three of the independent
> >>> variables are assumed endogenous.
> >>>
> >>> I hope I have not missed anything else.
> >>>
> >>> Any hints would be appreciated,
> >>>
> >>> Erasmo
> >>>
> >>>
> >>> On Wed, Mar 25, 2009 at 5:06 PM, Nick Cox 
> <n.j.cox@durham.ac.uk> wrote:
> >>>> -xtivreg2- is a user-written command from SSC (Mark Schaffer).
> >>>>
> >>>> For Mark, or anybody else, to have anything much to work 
> with here,
> >>> tell
> >>>> us about the data and the precise commands you used.
> >>>>
> >>>> Ideally, reproduce your results on a dataset accessible 
> to all, or let
> >>>> Mark privately have a copy of your data _and_ your commands.
> >>>>
> >>>> Otherwise there is little to go on here. It's not even 
> clear that you
> >>>> are fitting precisely the same model, or equivalent models.
> >>>>
> >>>> More attention to age-old advice in the FAQ, often 
> repeated on the
> >>> list,
> >>>> would have yielded a question easier to answer.
> >>>>
> >>>> Nick
> >>>> n.j.cox@durham.ac.uk
> >>>>
> >>>> P.S. no SHOUTING of command names please.
> >>>>
> >>>> Erasmo Giambona
> >>>>
> >>>> I am estimating the panel regression model with XTREG 
> and XTIVREG2.
> >>>> When I use XTREG, I obtain an adj-R2 of around 0.21. The 
> within and
> >>>> between R2's have a similar size. Similarly, when I use 
> XTIVREG2, the
> >>>> centered and uncentered R2 are around 0.20. However, if 
> I do: . di
> >>>> "R2-adj: " e(r2_a), I get: R2-adj: .02060245. This sems 
> a huge drop
> >>>> compared to the centered/uncentered R2 or the adj-R2 from XTREG.
> >>>>
> >>>> I have explored the statalist archive finding a lot of 
> very useful
> >>>> information about the nuances with the R2 with IV 
> regressions, but I
> >>>> didn't find a good answer to my question.
> >>>>
> >>>> I would appreciate any hints on this issue.
> >>>
> >>> *
> >>> *   For searches and help try:
> >>> *   http://www.stata.com/help.cgi?search
> >>> *   http://www.stata.com/support/statalist/faq
> >>> *   http://www.ats.ucla.edu/stat/stata/
> >>>
> >>
> >
> 
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
> 


-- 
Heriot-Watt University is a Scottish charity
registered under charity number SC000278.


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index