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

Re: st: AW: save estimated coefficients and use in other regressions

From   Jia Li <[email protected]>
To   "[email protected]" <[email protected]>
Subject   Re: st: AW: save estimated coefficients and use in other regressions
Date   Mon, 9 Nov 2009 09:33:49 -0500

Dear Martin and Maarten: 

Thanks for your suggestions. I ran separate regressions using completely different data sets: one uses consumer expenditure survey to derive relationships between income and expenditure; and I then use the estimated coefficients in my dataset to establish estimates of expenditure based on income.  I could of course just input the estimated coefficients, but I was wondering if there are more efficient ways to do this. I will experiment with the commands you suggested and let you know what i find out.


Date: Fri, 6 Nov 2009 01:46:30 -0800 (PST)
From: Maarten buis <[email protected]>
Subject: Re: st: AW: save estimated coefficients and use in other regressions

- --- Jia Li wrote:
> I have a simple question which I couldn’t find a
> solution:  I want to use estimated coefficients from one
> regression analysis in other analyses. What would be the
> best way to save the estimation results and call them in
> other do-files?

- --- On Fri, 6/11/09, Martin Weiss wrote:
> Depends very much on the context. -estimates store- and
> -estimates save- are obvious candidates. Also look at
> -postfile-

This is not a simple question as it could mean multiple
things: It could mean that you want to estimate a
regression for each region/school/company separately, and
than study how the resulting coefficients change as a
result of some characteristic of the region/school/company.
Basically, you are then trying to do some form of
multi-level analysis, and the first recommendation would
be not to do that and use the appropriate -xt- commands
instead. The second recommendation would be to use either
- -statsby- or -postfile- to collect the results, but now
you'll have to think about how to weight the different
estimates: An estimate which was measured with very low
precision should not get the same weight as an estimate
measured with very high precision. Reasoning by analogy from
the meta-analysis literature I would use 1/se^2, but you'll
need to find your own justification for such a weight (and
which weight type to use).

Alternatively, your question could mean that you want to
combine that you have multiple models and want to compare
coefficients. This is usually a mistake. Most of the times
that this is asked on the statalist, the answer is that
you should estimate one model and make proper use of
dummies and interactions. If this is not the case you
could consider looking at -suest-.

Bottom line: you'll need to tell us more about your problem,
even if you found the answer in the previous posts; the
polite thing is to close a thread that you started.

- -- Maarten

- --------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
- --------------------------

*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index