Re: st: interpreting margins results

["Michael N. Mitchell" <Michael.Norman.Mitchell@gmail.com>]

Dear Richard

  Understanding this kind of output can be very tricky, because of the presence of the 
interaction. In your case, the interaction is not significant and life would be simplified 
if you could remove that non-significant interaction from your model. Then, you could 
interpret the main effects of the binary factor and continuous variable rather easily.

  With respect to your current model, the -regress- command is telling us about the slope 
of -lpchgdef- when -d1982- is 0, and the difference in means betweeen -d1982- (1 vs 0) 
when lpchgdef is held constant at 0. The -margins- command is telling you about the means 
of -d1982-, and the z tests are testing whether each mean is significantly different from 
0 (e.g., is 5.590552 different from 0). These z-tests are rather boring.

I hope this helps,

Michael N. Mitchell
Data Management Using Stata      - http://www.stata.com/bookstore/dmus.html
A Visual Guide to Stata Graphics - http://www.stata.com/bookstore/vgsg.html
Stata tidbit of the week         - http://www.MichaelNormanMitchell.com



On 2010-08-09 9.01 PM, Richard J Stoll wrote:
> I am using Stata/SE 11.1 on a PC.
>
> I would appreciate help on interpreting the output from the margins
> command.
>
> I have read the documentation and looked around for examples on the web
> and cannot find an example in which the calculated margin is significant
> for one (or some) values of the variable, but not for others.  My question
> is how to interpret such a finding.
>
> Note: what follows is taken from a real dataset I am using, but the model
> shown below is simpler than what I am doing.
>
> Here is the regression command and results (note: I have chopped the right
> side of the results so they won't wrap across lines; I always find that
> hard to read).
>
> . reg dmnetsupp i.d1982##c.lpchgdef
>
>        Source |       SS       df       MS
> -------------+------------------------------
>         Model |  1148.21514     3   382.73838
>      Residual |  4337.83844    33   131.44965
> -------------+------------------------------
>         Total |  5486.05358    36  152.390377
>
> -----------------------------------------------------
>     dmnetsupp |      Coef.   Std. Err.      t    P>|t|
> -------------+---------------------------------------
>       1.d1982 |   -6.62355   4.204622    -1.58   0.125
>      lpchgdef |  -.2784506    .210933    -1.32   0.196
>               |
>         d1982#|
>    c.lpchgdef |
>            1  |   -.235843      .3527    -0.67   0.508
>               |
>         _cons |   6.204468   3.459617     1.79   0.082
> -----------------------------------------------------
>
> And then I use the following margins command, followed by the results
> (again I have chopped the right side to avoid wrapping lines):
>
> . margins i.d1982
>
> Predictive margins  Number of obs   =         37
> Model VCE    : OLS
>
> Expression   : Linear prediction, predict()
>
> ------------------------------------------------------
>               |            Delta-method
>               |     Margin   Std. Err.      z    P>|z|
> -------------+----------------------------------------
>         d1982 |
>            0  |   5.590552   3.472332     1.61   0.107
>            1  |  -1.552974   2.256145    -0.69   0.491
> ------------------------------------------------------
>
> How would you discuss the difference in the z-scores/probabilities for the
> two margins values of d1982?
>
> Thanks for your help.
> -------------------------------------------------------------------
> Richard J. Stoll * Albert Thomas Professor of Political Science *
> Rice University * Political Science MS 24 * P.O. Box 1892 * Houston
> TX 77251-1892 * stoll@rice.edu * 713-348-3362 * FAX: 713-348-5273
>
> *
> *   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/