Xtmelogit, reffects, forvalues and predicted probabilities

[James Gray]

Dear Stata users,

I'm trying to generate and graph predicted probabilities for multilevel models with random components, and I'm having some trouble. Can you help?

To do so, I follow the advice offered on this page: http://www.ats.ucla.edu/stat/stata/f..._xtmelogit.htm

I am using xtemelogit and predict, with the reffects option to generate best linear unbiased predictors of the random effects. I then manually code the formulae to express the predicted probabilities. So far, so good.

The problem comes about three fifths down the page. To generate predicted probabilities for a variety of values of one variable and graph the results, the instructions suggest that I use a local macro and a forvalues loop. I have been using the code below, which has been carefully reproduced from the page above.

Something is wrong with the code, and I can't figure out what. When I run it, the screen reads 'varlist required'. Can anyone spot what mistake I've made?


use http://www.ats.ucla.edu/stat/data/hsbdemo, clear
xtmelogit honors read female socst || cid: read, var covar(unstr)
predict re*, reffects

tempvar rxb0 rxb1 p0 p1
generate `rxb0'=.
generate `rxb1'=.
generate `p0'=.
generate `p1'=.
matrix m=J(9,1,0)
matrix f=J(9,1,0)
matrix d=J(9,1,0)
local i = 1

forvalues j=30(5)70 {
quietly replace `rxb0'=_b[female]*0+(_b[read]+re1)*`j'+_b[socst]*socst+_b[_cons]+re2
quietly replace `rxb1'=_b[female]*1+(_b[read]+re1)*`j'+_b[socst]*socst+_b[_cons]+re2
quietly replace `p0'=exp(`rxb0')/(1+exp(`rbx0'))
quietly sum `p0'
matrix m[`i',1]=r(mean)
quietly replace `p1'=exp(`rxb1')/(1+exp(`rbx10'))
quietly sum `p1'
matrix f[`i',1]=r(mean)
matrix d[`i',1]=f[`i',1]-m[`i',1]
local i = `i' + 1
}

[Lei Xuan]

I am dealing with the same issue. I also want to add the confidence interval to the graph. Do you know how to estimate the standard error for the probabilities that include random error and fixed error. -Thanks!

[eric_a_booth]

The tempvars in the bolded lines below are incorrectly spelled:

use http://www.ats.ucla.edu/stat/data/hsbdemo, clear
xtmelogit honors read female socst || cid: read, var covar(unstr)
predict re*, reffects

tempvar rxb0 rxb1 p0 p1
generate `rxb0'=. 
generate `rxb1'=. 
generate `p0'=.
generate `p1'=.
matrix m=J(9,1,0)
matrix f=J(9,1,0)
matrix d=J(9,1,0)
local i = 1

forvalues j=30(5)70 {
quietly replace `rxb0'=_b[female]*0+(_b[read]+re1)*`j'+_b[socst]*socst+_b[_cons]+re2
quietly replace `rxb1'=_b[female]*1+(_b[read]+re1)*`j'+_b[socst]*socst+_b[_cons]+re2
quietly replace `p0'=exp(`rxb0')/(1+exp(`rbx0'))
quietly sum `p0'
matrix m[`i',1]=r(mean)
quietly replace `p1'=exp(`rxb1')/(1+exp(`rbx10'))
quietly sum `p1'
matrix f[`i',1]=r(mean)
matrix d[`i',1]=f[`i',1]-m[`i',1]
local i = `i' + 1
}

also for your first line of code reading in the data,this location is not working for me (using Stata 15), this works though:

use https://stats.idre.ucla.edu/stat/data/hsbdemo, clear