# st: cmp-module and changing results

 From "U.G. Narloch" To statalist@hsphsun2.harvard.edu Subject st: cmp-module and changing results Date 28 Oct 2008 12:44:08 +0000

```Dear Statalist-users,

I have a question concerning the use of the the cmp-command in STATA 9.2.

I have the following mixed model:
y1 = x1 + y2 + y3 + y4 + y5
y2 = x2 + y6
y3 = x3 + y7
y4 = x4 + y8
y5 = x5
y6 = x6
y7 = x7
y8 = x8

Some exogenous explaining variables are used in several equations while
others are only included in one equation (i.e. x1 = x11 + x12 and x2 = x21
+ x22 while x11 = x21 ). y1 is a dummy and should be estimated within a
probit model, whereas y2-y8 are estimated by OLS.

I have problems to make the model converge, so that I run the following
command: cmp (y1 = x1 + y2 + y3 + y4 + y5)(y2 = x2 + y6 )(y3 = x3 + y7) (y4
= x4 + y8) (y5 = x5) (y6 = x6) (y7 = x7) (y8 = x8), ind(4 1 1 1 1 1 1 1)
nonrtol tech(dfp nr).

Doing so I obtain results that vary significantly from each other when I
only change one of the exogenous explaining variables (in total there are
more than 20 different exogenous explaining variables included in this
model). So it happens that in one specification I get very reasonable
results (with a couple of variables having the expected significant signs),
while in another (slightly changed) specification most of the explaining
variables turn to be insignificant.

Do anyone has any idea, what is going on? May it be linked to a relatively
small sample size (445 households)?

Thanks a lot,
Ulf Narloch

--
PhD Candidate
Department of Land Economy
19 Silver Street
Cambridge
CB3 9EP
UK

*
*   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/
```