Sociology
multinomial logit
testing hypotheses
The data for this exercise again comes from the 1991 General Social Survey. The
categorical dependent variable occ is coded as follows:
occ=0 if a workers occupation is laborer, operative or craft;
occ=1 if occupation is clerical, sales, or service;
occ=2 if occupation is managerial, technical, or professional.
The independent variables are: educ is years of schooling; age is age in years; sexx
is coded 1 male, 0 female; rural is coded 1 if grew up in rural area, 0 otherwise; mid
and wst are dummy variables for region, with other parts of the country omitted.
Let’s fit what we’ll treat for most of this exercise as the null model.
3. mlogit occ educ age sexx rural mid wst,base(0)
Multinomial regression Number of obs = 633
LR chi2(12) = 353.13
Prob > chi2 = 0.0000
Log likelihood = -511.92941 Pseudo R2 = 0.2564
------------------------------------------------------------------------------
occ | Coef. Std. Err. z P>|z| [95% Conf. Interval]
---------+--------------------------------------------------------------------
1 |
educ | .2490034 .056606 4.399 0.000 .1380577 .3599492
age | .0156041 .0099216 1.573 0.116 -.0038418 .0350501
sexx | -2.028054 .2392113 -8.478 0.000 -2.4969 -1.559209
rural | -.7635868 .2619814 -2.915 0.004 -1.277061 -.2501126
mid | .4081406 .2761675 1.478 0.139 -.1331378 .9494189
wst | .4151271 .3078639 1.348 0.178 -.188275 1.018529
_cons | -2.253103 .853224 -2.641 0.008 -3.925391 -.5808147
---------+--------------------------------------------------------------------
2 |
educ | .7840261 .0684775 11.449 0.000 .6498126 .9182395
age | .01764 .011552 1.527 0.127 -.0050015 .0402816
sexx | -1.680553 .2778157 -6.049 0.000 -2.225062 -1.136044
rural | -.128399 .2965349 -0.433 0.665 -.7095968 .4527988
mid | .144635 .3137103 0.461 0.645 -.4702258 .7594958
wst | .3873871 .3445527 1.124 0.261 -.2879237 1.062698
_cons | -10.27188 1.063177 -9.661 0.000 -12.35567 -8.188092
------------------------------------------------------------------------------
Tests of individual coefficients
You can use the z-scores to test for individual coefficients in separate equations.
To do a test of ALL the coefficients of a given variable, say educ, in all the
equations, you need to impose the constraint of the null hypothesis,
and then estimate the restricted model:
docsity.com
