Re: st: Low varability in dependent as a reason for period differing regression results?

[Nick Cox <njcoxstata@gmail.com>]

On Tue, Jan 15, 2013 at 12:22 PM, Birk Teuchert <b.teuchert@gmx.de> wrote:
> Thanks for the quick reply Nick!
>
> could you explain more precisely how i do the plots

You should type -help regress postestimation-.

> and what you with a different scale (i.e. the commands -glm or link(log)-?

What is the question here? -glm- is a command and -link(log)- is one
of its options. That is what -glm, link(log)- indicates. Again, there
is help at -help glm-, which is linked to the manual entry, which
gives literature references.

> Regarding your third point: i dont think that the predictors should play different roles in the two regressions..

Nature and society often are indifferent to what we think they should
do. You asked: Is it enough to look at the distribution of y, and my
answer is definitely No.

>
> Thanks a lot
> -------- Original-Nachricht --------
>> Datum: Tue, 15 Jan 2013 12:09:00 +0000
>> Von: Nick Cox <njcoxstata@gmail.com>
>> An: statalist@hsphsun2.harvard.edu
>> Betreff: Re: st: Low varability in dependent as a reason for period differing regression results?
>
>> Plot observed vs predicted and residual vs predicted for each
>> regression. You can do this regardless of how complicated each
>> regression is. R-square is the square of the correlation between
>> observed and predicted for plain regression, so such graphs make
>> explicit what underlies each R-square.
>>
>> With a difference of means that is an order of magnitude, you may be
>> better off modelling on some different scale, e.g. -glm, link(log)-.
>>
>> You do also need to check whether predictors play similar roles in each
>> case.
>>
>> Nick
>>
>> On Tue, Jan 15, 2013 at 12:00 PM, Birk Teuchert <b.teuchert@gmx.de> wrote:
>>
>> > I have problems identifying the reasons for the differing results for 2
>> different time periods (the general regression equation is the same for
>> both periods) i would like to do my analysis seperatly with.
>> > How can I identify possible reasons for the differing results?
>> >
>> > I had a look at the descriptives and indeed I noticed that mean (10
>> times higher) and SD (4 times hihger) of my dependent is a lot higher for one
>> time period which results in an R2 of 30% compared to 4% in the other time
>> period.
>> >
>> > Is this difference in the distribution of the dependent variable enough
>> to explain the different results or how do I identify possible other
>> explanations??
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/