# Re: st: margeff

 From "Clive Nicholas" <[email protected]> To [email protected] Subject Re: st: margeff Date Thu, 10 Mar 2005 04:05:30 -0000 (GMT)

```Peter Harper wrote:

> I have run -margeff-, on the three responders
> categories of price being 'very' 'fairly' and 'not'
> important. But the z-statistic of one of the variables
> in the 'fairly important' category is 998.42. For the
> other two categories they were around 2.00. Can anyone
> tell me how to correctly interpret the z-statistic of

Looks a bit suss to me, but it's difficult to advise you further without
seeing some actual output. Also, you may find it more useful to use -mfx,
predict(p outcome()) at(mean)-. You should find that your results are
slightly different using -mfx-. An example:

. use http://www.gseis.ucla.edu/courses/data/hsb2, clear
(highschool and beyond (200 cases))

. ologit ses female race read write math

Iteration 0:   log likelihood = -210.58254
Iteration 1:   log likelihood = -197.51869
Iteration 2:   log likelihood = -197.36092
Iteration 3:   log likelihood = -197.36061

Ordered logit estimates                         Number of obs   =        200
LR chi2(5)      =      26.44
Prob > chi2     =     0.0001
Log likelihood = -197.36061                     Pseudo R2       =     0.0628
----------------------------------------------------------------------------
ses |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-----------+----------------------------------------------------------------
female |  -.4909873   .2952412    -1.66   0.096    -1.069649    .0876748
race |   .2350773    .137125     1.71   0.086    -.0336828    .5038373
read |   .0311981   .0193404     1.61   0.107    -.0067085    .0691046
write |   .0118174   .0209855     0.56   0.573    -.0293134    .0529481
math |   .0228895   .0209798     1.09   0.275    -.0182301    .0640091
-----------+----------------------------------------------------------------
_cut1 |   2.692794   .9248859          (Ancillary parameters)
_cut2 |    4.99739   .9796558
----------------------------------------------------------------------------

. margeff

Marginal effects on Prob(ses) after ologit
----------------------------------------------------------------------------
ses |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-----------+----------------------------------------------------------------
low        |
female |   .0803274    .119315     0.67   0.501    -.1535258    .3141805
race |  -.0388374   .0480915    -0.81   0.419    -.1330949    .0554202
read |  -.0051543   .0068777    -0.75   0.454    -.0186343    .0083258
write |  -.0019524   .0037731    -0.52   0.605    -.0093475    .0054428
math |  -.0037816   .0049511    -0.76   0.445    -.0134856    .0059224
-----------+----------------------------------------------------------------
middle     |
female |  -.0802519    .196229    -0.41   0.683    -.4648537      .30435
race |   .1212574   .0938469     1.29   0.196    -.0626791     .305194
read |   .1593779   .0124249    12.83   0.000     .1350254    .1837303
write |   .1630016   .0048255    33.78   0.000     .1535437    .1724595
math |   .1609314     .00923    17.44   0.000     .1428409    .1790218
-----------+----------------------------------------------------------------
high       |
female |  -.0930442   .1030793    -0.90   0.367    -.2950758    .1089874
race |   .0439537   .0564441     0.78   0.436    -.0666746    .1545821
read |   .0058333   .0072293     0.81   0.420    -.0083359    .0200025
write |   .0022096   .0049231     0.45   0.654    -.0074394    .0118586
math |   .0042798    .006634     0.65   0.519    -.0087226    .0172821
----------------------------------------------------------------------------

. mfx, predict(p outcome(1)) at(mean)

Marginal effects after ologit
y  = Pr(ses==1) (predict, p outcome(1))
=  .21347808
----------------------------------------------------------------------------
variab |      dy/dx    Std. Err.     z    P>|z|  [    95% C.I.   ]      X
-------+--------------------------------------------------------------------
female*|   .0813982      .04854    1.68   0.094  -.013747  .176543      .545
race |  -.0394707      .02316   -1.70   0.088  -.084865  .005923      3.43
read |  -.0052383      .00326   -1.61   0.108  -.011619  .001143     52.23
write |  -.0019842      .00352   -0.56   0.573  -.008889  .004921    52.775
math |  -.0038433      .00352   -1.09   0.275  -.010748  .003061    52.645
----------------------------------------------------------------------------
(*) dy/dx is for discrete change of dummy variable from 0 to 1

. mfx, predict(p outcome(2)) at(mean)

Marginal effects after ologit
y  = Pr(ses==2) (predict, p outcome(2))
=  .51768066
----------------------------------------------------------------------------
variab |      dy/dx    Std. Err.     z    P>|z|  [    95% C.I.   ]      X
-------+--------------------------------------------------------------------
female*|   .0159065      .01619    0.98   0.326  -.015826  .047638      .545
race |  -.0067374       .0071   -0.95   0.343  -.020658  .007183      3.43
read |  -.0008942      .00097   -0.92   0.358  -.002799  .001011     52.23
write |  -.0003387      .00068   -0.50   0.617  -.001665  .000988    52.775
math |   -.000656      .00084   -0.78   0.436  -.002307  .000995    52.645
----------------------------------------------------------------------------
(*) dy/dx is for discrete change of dummy variable from 0 to 1

. mfx, predict(p outcome(3)) at(mean)

Marginal effects after ologit
y  = Pr(ses==3) (predict, p outcome(3))
=  .26884126
----------------------------------------------------------------------------
variab |      dy/dx    Std. Err.     z    P>|z|  [    95% C.I.   ]      X
-------+--------------------------------------------------------------------
female*|  -.0973046      .05883   -1.65   0.098  -.212612  .018002      .545
race |   .0462081      .02689    1.72   0.086  -.006496  .098913      3.43
read |   .0061325       .0038    1.61   0.107  -.001323  .013588     52.23
write |   .0023229      .00413    0.56   0.574  -.005767  .010413    52.775
math |   .0044993      .00413    1.09   0.275  -.003586  .012584    52.645
----------------------------------------------------------------------------
(*) dy/dx is for discrete change of dummy variable from 0 to 1

In this example, we can see that the elasticities (dy/dx) are very
similiar to those given in -margeff- for low and high values of
socioeconomic status (and, indeed, the pattern is identical), but are
completely different at 'middle' values of SES.

Note that -mfx- provides you with some extra information here: the
(overall) predicted probability of Y = 1, 2 or 3 given the X-variables set
at their mean values. In this example, respondents with 'average' social
and educational characteristics are much more likely to be in the 'middle'
SES category than in the other two.

I hope that helps.

CLIVE NICHOLAS        |t: 0(044)7903 397793
Politics              |e: [email protected]
Newcastle University  |http://www.ncl.ac.uk/geps

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```