Using postestimation margins and returned results to generate new variable

[Itai Magodoro]

Dear Forum,

I will appreciate guidance on how I can create a new variable from stored results.

Specifically, I ran a logit model (#1, below) with postestimation margins (#2) to calculate probability of positive outcome (returned as "Margin" in #2). I would like to generate a new variable equal to the returned "Margin".

I thought if I used the "return list" function (#3, below) the "margin" will be returned under the scalars, and I could then use "generate newvar = r(margin)". However, the returned list of scalars does not have "margin".

Please kindly help with suggestions.

Regards,
Itai

__________


#1. logit hpt i.hiv i.sex i.race [pweight = wt_hiv], vce(robust) cformat(%9.2f)

#2. margins sex race, predict(pr) over(hiv) cformat(%9.2f)

#3. return list

[Andrew Musau]

If you use the -post- option of margins, you can extract these from the matrix -r(table)-.

logit hpt i.hiv i.sex i.race [pweight = wt_hiv], vce(robust) cformat(%9.2f)
margins sex race, predict(pr) over(hiv) cformat(%9.2f) post
mat Margin = (r(table)[1,1...])'
svmat Margin
list Margin in 1/15

[William Lisowski]

The results you seek are returned in the matrix r(table). Try

matrix list r(table)

following the margins command to see that it contains all the results from the table.

Added in edit: crossed with Andrew's post, which leads me to believe that I don't fully understand what you were seeking.

[Itai Magodoro]

Thank you very much Andrew and William. The suggestion by Andrew worked very well.

[Itai Magodoro]

A further question, for which guidance will be much appreciated. I would like to calculate prevalence ratio and prevalence difference between 2 levels of a dichotomous exposure (HIV) as shown in table below. I thought I might use margins with expression command along the lines shown below but obviously I am out of depth.

NOTE - the text in red is not part of the STATA code but illustrative of my coding intent.

logit hpt i.hiv i.sex i.agegrp i.race [pweight = wt_hiv], vce(robust) cformat(%9.2f) margins sex agegrp race, expression((predict(pr) [for HIV pos] - (predict(pr) [for HIV neg])) cformat(%9.2f) | for prevalence difference margins sex agegrp race, expression((predict(pr) [for HIV pos] / (predict(pr) [for HIV neg])) cformat(%9.2f) | for prevalence ratio


Please kindly help with suggestions.

____________