Re: st: margins after xtlogit

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

Dear Traci

  I will start with a possible answer, and then work back to why this is so. After the 
-xtlogit- command, try this...

margins race1, predict(pu0)

  This should then express the results in terms of predicted probabilities (as the 
-logit- model did). The reason is that the -predict- command defaults to predicting 
probabilities in after the -logit- command. This is described in
  -help logit postestimation- under the section on predict that will say
"pr            probability of a positive outcome; the default"

  Contrast this with -help xtlogit postestimation-, in which the section about predict 
says that the default prediction is "xb           linear prediction; the default".

I hope that 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 2010-09-10 4.59 PM, Traci Schlesinger wrote:
> hi all:
>
> i am analyzing racial disparities in pretrial diversions (a yes no,
> i.e. 0/1, criminal justice outcome) using individual level data from
> the SCPS, which is clustered by county--an observation for every
> individual charged with a felony in sampled counties is included.  to
> account for the county level sampling, i'm using xtlogit with county
> level random effects.
>
> however, i'm having difficulty interpreting the results from margins
> after xtlogit.
>
> if i estimate a model with logistic and then ask for margins on race i get:
>
> . margins race1, post
>
> Predictive margins                                Number of obs   =      46019
> Model VCE    : OIM
>
> Expression   : Pr(diversion), predict()
>
> ------------------------------------------------------------------------------
>               |            Delta-method
>               |     Margin   Std. Err.      z    P>|z|     [95% Conf. Interval]
> -------------+----------------------------------------------------------------
>         race1 |
>            1  |   .1025184   .0023145    44.29   0.000      .097982    .1070548
>            2  |   .0848741   .0020596    41.21   0.000     .0808374    .0889109
>            3  |   .0858849   .0023203    37.01   0.000     .0813372    .0904327
>
> ------------------------------------------------------------------------------
>
> which i interpret as meaning that if everyone in my sample were white
> (race1 = 1), 10% of defendants would be offered pretrial diversions.
> if everyone were black (race1=2), only 8% of defendants would be
> offered pretrial diversions.  (race1=3 are Latinos, with 8.5% of
> people getting diversions).
>
> however, if i estimate xtlogit --either getting my results as
> coefficients or odds-ratios-- and then margins, i get the following
> table.
>
> . margins race1, post
>
> Predictive margins                                Number of obs   =      46019
> Model VCE    : OIM
>
> Expression   : Linear prediction, predict()
>
> ------------------------------------------------------------------------------
>               |            Delta-method
>               |     Margin   Std. Err.      z    P>|z|     [95% Conf. Interval]
> -------------+----------------------------------------------------------------
>         race1 |
>            1  |  -3.580741   .2277247   -15.72   0.000    -4.027073   -3.134409
>            2  |  -3.919428   .2274633   -17.23   0.000    -4.365248   -3.473608
>            3  |   -3.67982   .2301685   -15.99   0.000    -4.130942   -3.228698
> ------------------------------------------------------------------------------
>
> i am at a loss as to how to interpret this.  for starters, it seems
> strange that all three racial groups have negative margins.  also, i'm
> clearly not looking at the percent of defendants who get a pretiral
> diversion any more.  i've looked through the manual, but have not been
> able to figure this out.  i would appreciate any help.
>
> cheers,
> traci
> *
> *   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/