Estimating A Predicted Probability for A Sub-Population of a Survey

[Roman Johnson]

Hi, All:

I would like to estimate a predicted probability for my sub-population in my survey. I have already used my svyset specifications (e.g. strata, cluster, psu) using the design weights of the 2016 National Health Interview Survey.

How would I do this?

I have ran this logistic regression.

svy, subpop (hispanicorigin): logistic mammogramadherent born

However, I want to know how to estimate the predicted probability of the variable "born". (Sorry, for not using the dataex command. I'm not sure how to use it with sub-population commands.)

[Richard Williams]

Do you mean mammogramadherent? born is an independent variable so you wouldn't predict its probability.

My guess is you want

predict hispanicprob if hispanicborn

[Roman Johnson]

Yes, I meant mammogramadherent. What would be the correct syntax for that?

I am bit confused by the predict syntax you shared. I am not sure if hispanicprob and hispanicborn are generated variables.

[Richard Williams]

I should have said

predict hispanicprob if hispanicorigin

hispanicprob will be a generated variable. It will be generated for those who are nonzero and nonmissing on hispanicorigin and will be missing for everyone else,

[Phil Bromiley]

You can do Richard's comment more verbosely
predict yhat if born==2

This predicts using the sample values restricting the predictions to observations where born is 2. Is this what you want? Depending on your purpose, margins will also do predicted values.

[Roman Johnson]

Yes, Phil. Thanks.