Coefplot: Displaying each level of a factor variable in one plot

[Ellesse Akre]

Greetings:

I am trying to visualize the data for logistic regressions. My independent variable of interest is a categorical variable with 18 levels. I am trying to use coefplot (or anything is stata) to make a chart of the different AORs and the confidence intervals but I can't seem to find a way. The closest I've seen is that I can use coefplot to graph interaction variables and it'll include each level of the interaction and I can drop the covariates to isolate the results of my IVs on interest. I cannot seem to figure out how to do it for one variable with multiple levels.

First problem that occurs. I use the following code:

coefplot poorhealth, drop(agecat_v3 education hh_child partnered hh_income nonenglish uninsured no_usualcare state year) xline(1)

And I get this :



I can't seem to get past the first step. Any help would be greatly appreciated.

Thank you

[Andrew Musau]

coefplot is from the Stata Journal (FAQ Advice #12).

webuse nlswork
set seed 02262021
gen catvar= runiformint(1,18)
logit union age i.race i.catvar grade collgrad
coefplot, keep(*.catvar) baselevels sort scheme(s1mono)

Res.:

[Weiwen Ng]

Greetings:

I am trying to visualize the data for logistic regressions. My independent variable of interest is a categorical variable with 18 levels. I am trying to use coefplot (or anything is stata) to make a chart of the different AORs and the confidence intervals but I can't seem to find a way. The closest I've seen is that I can use coefplot to graph interaction variables and it'll include each level of the interaction and I can drop the covariates to isolate the results of my IVs on interest. I cannot seem to figure out how to do it for one variable with multiple levels.

...

A point of information because you mentioned adjusted odds ratios: even if you used the logistic command (which shows odds ratios), coefplot plots the coefficients in their raw scale (i.e. in log-odds units). You can verify this by comparing your chart to the odds ratio outputs. Or, you drew a line through x = 1, but a whole bunch of your coefficients are around x = 0, i.e. an odds ratio of 1.

This is easy to solve, you just add the eform option as detailed here. One thing to remember is that the confidence intervals will be asymmetric - they're symmetric on the log-odds scale, but they asymmetric once you exponentiate everything. Some people may get nervous about this, but that is the nature of logistic regression.

[Ellesse Akre]

Thank you both! this is extremely helpful and I was able to get the correct chart.