Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


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

Re: st: Does xtreg (or xtivreg) assume equally spaced time points?


From   Lisa Marie Yarnell <lisayarnell@yahoo.com>
To   "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>, David Greenberg <dg4@nyu.edu>
Subject   Re: st: Does xtreg (or xtivreg) assume equally spaced time points?
Date   Tue, 7 Aug 2012 16:41:54 -0700 (PDT)

Actually, I have one last question on this.

I am generally intrested in comparing the results found in this RE or FE xtivreg model estimated in Stata with results from a structural equation model (SEM) where the unequal spacing of time points IS accounted for.

(In latent growth models in SEM, it is typical to account for unequal spacing of time points by changing the loadings of variables onto a latent slope factor.)

So is it true that in Stata's xtivreg model, it is not a *problem* that the time points are unequal, but there is also no way to really *account for* the unequal time points--unless I get into dynamic panel modeling or introduce the FD (first differences) option, where the spacing between time points is a more important consideration.  Is that right?

In other words, an xtivreg model estimated in Stata without dynamic or FD options--whether I use dummies for wave or not--will not estimate the parameters taking the unequal time points into account.

Am I right?  So, this may not make for a clean comparison with the SEM model I mentioned, where we *are* accounting for the unequal time points.

Thanks,
Lisa




----- Original Message -----
From: Lisa Marie Yarnell <lisayarnell@yahoo.com>
To: David Greenberg <dg4@nyu.edu>; "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Cc: 
Sent: Tuesday, August 7, 2012 2:15 PM
Subject: Re: st: Does xtreg (or xtivreg) assume equally spaced time points?

Thanks, everyone.  I am going to read a bit about fixed and random effects models and try to determine which is best for this data.

Thanks again,
Lisa

 
----- Original Message -----
From: David Greenberg <dg4@nyu.edu>
To: Lisa Marie Yarnell <lisayarnell@yahoo.com>
Cc: 
Sent: Tuesday, August 7, 2012 2:03 PM
Subject: Re: st: Does xtreg (or xtivreg) assume equally spaced time points?

A random effects model is viable in the sense that you can run it, but
in doing so, you are tacitly assuming that all the predictors in your
model are uncorrelated with the error terms. If there are omitted
variables contributing to those error terms, they could well produce
such non-vanishing correlations. You can test for these non-vanishing
correlations by conducting a Hausman test. The test will tell you
whether to use an RE or FE model. Ignoring the results of that test in
order to get an estimate for the effect of a time-invariant predictor
is making a bargain with the devil. You get your estimate, but you
will be getting an inconsistent estimate. What good is that? David
Greenberg, SOciology Department, NYU

On Tue, Aug 7, 2012 at 4:55 PM, Lisa Marie Yarnell
<lisayarnell@yahoo.com> wrote:
> Thank you all so much--I believe that I have gained some clarity
>
> I suppose that my last question is whether I can specify the model as a Random Effects rather then Fixed Effects panel model, perhaps with the dummies and perhaps without, i.e.,
>
> xtivreg depper wave2 wave3 wave4 (bmi = momobes), i(id) re
>
> I don't want to specify it as FE because I don't want to assume that all tme-invariant characteristics are differenced out, which I know is an aspect of the fixed effects model. So would the RE take into account that the DV "depper" and the BMI predictor are measured multiple times per person across the four waves in panel form?  So time points are nested within persons.  Here, Wave I would be the omitted category for the wave dummies.  And it would not matter that the time points are unequally spaced?
>
> Is this a viable model and understanding?
>
> Many thanks,
> Lisa
>
>
> ----- Original Message -----
> From: David Greenberg <dg4@nyu.edu>
> To: statalist@hsphsun2.harvard.edu; lisayarnell@yahoo.com
> Cc:
> Sent: Tuesday, August 7, 2012 12:19 PM
> Subject: Re: st: Does xtreg (or xtivreg) assume equally spaced time points?
>
> In a purely static model, which assumes that all causal effects are
> contemporaneous, not lagged, the waves need not be evenly spaced. If
> you start putting in lagged effects, they would be. It is not
> mathematically necessary to put in fixed effects for the waves, but it
> is often a good idea, in order to take into account temporal variation
> that effects all of your units of analysis in the same way. You can
> try putting them in; creating them and adding them your model is easy
> enough. If, collectively, they do not make a significant difference,
> then you can legitimately drop them. David Greenberg, Sociology
> Department, New York University
>
> On Tue, Aug 7, 2012 at 10:31 AM, Lisa Marie Yarnell
> <lisayarnell@yahoo.com> wrote:
>> Hi Statalisters,
>>
>> Does xtreg (or really, what we are using, xtivreg--the extension of xtreg that allows for an instrument) asusme equally-spaced time points in the panel model?
>>
>> We have four waves of data.
>> Wave I and Wave II are 1.5 years apart.
>> Wave II and Wave III are 5.5 years apart.
>> Wave III and Wave IV are 6 years apart.
>>
>> Is it wrong to put data from these time points into the xtivreg model? Is there a way to adjust the model for unequally spaced time points?
>>
>> Our model is: xtivreg depper (bmi = momobes)
>>
>> How can we alter the model to account for the unequal spacing?
>>
>> Thanks,
>> Lisa

*
*   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   |   Site index