Hello,

We are estimating a model with binary Y and binary X, using the CMP command. In our main analysis, everything works perfectly.

However, for running a placebo test, we then create a random dummy variable which we use in place of our X variable. On re-doing the CMP estimation, the model fails to converge. The iterations continue to be concave. When we use the "nonrtol" option, it outputs a result (although it stays concave). However, we only get the coefficient but the standard errors, p-values, confidence intervals are all missing. We are also able to estimate margins of this random X variable after that. But is this interpretable? What is this nonrtol option doing such that we get a result?

Any advice on this would be really helpful.

Thanks!