logit: binary regression with robust standard errors

Model a zero/one outcome on the log-odds scale and keep coefficient interpretation separate from probability and causal claims.

Tested 2026-08-06 with the current stats.camp development build · View the do-file on GitHub

Open this example View supported commands

Independent software. Not affiliated with, sponsored by, or endorsed by StataCorp LLC.

Complete example

clear
import delimited "data/student_scores.csv"
logit female math read, robust

The outcome is a zero/one indicator. robust requests a sandwich variance estimate while leaving the maximum-likelihood coefficients unchanged.

Output from stats.camp

Logistic regression                                     Number of obs =     24
                                                        Wald chi2(2)    =   4.19
                                                        Prob > chi2   = 0.1232
Log likelihood =   -6.82215                             Pseudo R2     = 0.5899

--------------------------------------------------------------------------------
             |               Robust
      female | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+------------------------------------------------------------------
        math |   -1.7951443   1.236561    -1.45   0.147    -4.218758    .6284698
        read |    2.7685663   1.785122     1.55   0.121    -.7302075     6.26734
       _cons |   -52.723487   30.56276    -1.73   0.085    -112.6254    7.178413
--------------------------------------------------------------------------------

How to read the coefficients

A coefficient is a change in log odds for a one-unit predictor change while holding the other included predictor fixed. It is not a change in probability. Exponentiating a coefficient produces an odds ratio, but this tiny synthetic dataset yields imprecise and unstable estimates that should not receive substantive interpretation.

The overall robust Wald test has p=0.1232. A pseudo R-squared is not interpreted like the OLS R-squared.

Common options and modeling checks

robust and vce(robust) request robust standard errors; supported vce() forms should be chosen from the sampling design.
iterate() and tolerance() control convergence. Changing them does not repair separation or a poorly specified model.
Supported factor-variable notation such as i.group creates categorical indicators with an omitted reference group.
Inspect outcome coding, missing observations, convergence, separation, functional form, and sample size before interpreting results.
The related logistic command reports odds ratios; logit reports coefficients.

See the current logit entry and Stata's logit manual for an external reference.

Related guides

Run this exact example

Clone the example repository in stats.camp and open the exact do-file shown on this page. No stats.camp account is required.

Open this example