Re: st: RE: problems with conducting OLS-PCSE analysis

[Nick Cox <njcoxstata@gmail.com>]

I'd add that interpolation would solve one problem only to create
another: you would end up with a messy mixture of the real data with
unknown dependence structure and interpolated data which will usually
be smoother than the original. The more you interpolate, the more you
would be obscuring the error properties you are trying to establish.

Nick

On Tue, Jan 17, 2012 at 9:45 AM, Gordon Hughes <G.A.Hughes@ed.ac.uk> wrote:
> There are two separate problems with what you are trying to do:
>
> A.  There are no time periods (years) for which data is available for all
> countries.  Hence, using casewise inclusion/deletion means that you cannot
> construct the cross-country error covariance matrix.  Specify the options
> "pairwise"  or "hetonly" to address this - see the manual for information on
> the different options.
>
> B.  As David Jacobs points out, autocorrelation routines don't like gaps in
> data.  His suggestion of imputing or interpolating missing years will work
> if there are only a few gaps, especially if there are strong time trends in
> the country time series.  Alternatively, suppress the calculation of the
> autocorrelation coefficient by specifying the option "corr(independent)".
>
> However, as always you should think carefully about the specification of
> your model.  It seems possible that you are trying to estimate a model that
> is too complex given the nature of your data.
>
> Gordon Hughes
> g.a.hughes@ed.ac.uk
>
> =======================================
>
> Date: Mon, 16 Jan 2012 20:54:01 +0000
> From: "Jacobs, David" <jacobs.184@sociology.osu.edu>
> Subject: st: RE: problems with conducting OLS-PCSE analysis
>
> As you probably already know, you have missing values or gaps within various
> series included in your models.
>
> Some obvious tricks to determine where these missing values are located
> include running the command -xtdescribe if e(sample)- immediately after an
> -xtpcse- run. That will show you what the exact pattern of gaps is. Another
> possibility that will provide more detail is to type ".browse if e(sample)"
> again immediately after a run. The last command works best if you have only
> a few explanatory variables in your model as it will show the entire data
> set the xtpcse routine could access.
>
> In general and as you probably already know AR(1) routines don't like gaps
> within a series and you seem to have that problem. One possibility is to
> employ Stata's commands that impute missing data to overcome these
> difficulties.
>
> D. Jacobs
>
> - -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Edward James
> Sent: Monday, January 16, 2012 9:33 AM
> To: statalist@hsphsun2.harvard.edu
> Subject: st: problems with conducting OLS-PCSE analysis
>
> Dear Statalist.
>
> I am currently conducting ols-pcse(pannel corrected standard error) with
> stata.
>
> Although the number of groups(countries) are 19, the result shows only
> 16 groups.
>
> In addition, when I conduct different model, following error message comes
> up:
>
> "Number of gaps in sample: 40
> (note: computations for rho restarted at each gap)
> (note: estimates of rho outside [-1,1] bounded to be in the range [-1,1])
> no time periods are common to all panels, cannot estimate disturbance
> covariance matrix using casewise inclusion"
>
> Do you have any idea for including all countries?
>
> Thanks.
> *
>
> *
> *   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/