Stata Technical Bulletin
17
Options
robust specifies that the Huber/White/sandwich estimate of variance should be calculated and robust standard errors reported.
cluster(cluster.vαrnαme) specifies that the robust standard errors should be adjusted for clustering on the variable specified
by cluster.varname.
score Sscore± score?,) specifies that the scores from the two probit equations should be saved in the variables specified by score ∖
and score?. The scores have mean zero and are uncorrelated with the independent variable in their respective equations.
nochi specifies that the constant-only model should not be fit (as this can take a long time for many models). Specifying this
option means that there will be no statistics associated with the test of significance of the full model.
level (#) specifies the confidence level, in percent, for confidence intervals. The default is level (95) or as set by set level.
nested specifies for the suprob command that a nested probit should be fit. The score, robust, and cluster options are
not available with this model.
Example: Bivariate probit regression
Using a subset of data given in Pindyck and Rubinfeld (1981), we wish to estimate a model the decision to send at least
one child to private school and whether to vote yes on a new property tax on the number of years lived in the community, the
log-income, and the log of property taxes paid. We will use this data again in a subsequent example (how it was originally used
in their paper).
. biprob priv vote yrs inc ptax
Fitting constant only model
Iteration 0: Log Likelihood = -82.529057
Iteration 1: Log Likelihood = -82.077824
Iteration 2: Log Likelihood = -82.077005
Iteration 3: Log Likelihood = -82.076978
Fitting full model
Iteration 0: Log Likelihood = -75.148912
(output omitted )
Iteration 4: Log Likelihood = -74.171253
Bivariate probit regression Number of obs = 80
Model chi2(6) = 15.81
Prob > chi2 = 0.0148
Log Likelihood = -74.1712526 Pseudo R2 = 0.0963
|
— |
I |
Coef. |
Std. Err. |
z |
P>∣z∣ |
[9Б7. Conf. |
— Interval] |
|
— priv yrs inc pt ax „cons |
I I I |
-.0146627 .3644543 -.0923143 -4.040363 |
.0264237 .5588125 .6922492 4.872901 |
-0.555 0.6Б2 -0.133 -0.829 |
0.Б79 0.Б14 0.894 0.407 |
-.0664622 -.7307982 -1.449098 -13.Б9107 |
— .0371268 1.4Б9707 1.264469 Б.Б10349 |
|
— vote yrs inc pt ax „cons |
I I I |
-.008866 1.574388 -2.054462 -.9732723 |
.0159737 .5638389 .7310163 4.486987 |
-О.БББ 2.792 -2.810 -0.217 |
0.Б79 0.00Б 0.00Б 0.828 |
-.0401739 .4692842 -3.487228 -9.767606 |
— .0224419 2.679492 -.6216967 7.821061 |
|
— rho „cons |
I |
-.3297288 |
.2252396 |
-1.464 |
0.143 |
-.7711903 |
— .1117327 |
Note that we could have obtained these results using suprob:
. eq priv yrs inc ptax
. eq vote yrs inc ptax
. suprob priv vote
Example: Seemingly unrelated two-equation probit regression
In this example, we duplicate the original analysis using the previous data. Here there are two probit equations. In the first,
whether a family places at least one child in private school depends on the log of the family income and the number of years
that the family has resided in the neighborhood. Whether the family votes on a new property tax depends on the log of the
family income and the log of the property tax currently paid.