Predicting individual-level slopes and intercepts after running mixed-effects models

[Amal Khanolkar]

Hi All

I'm running two linear spline mixed-effects models (growth curve analysis) looking at mental health and BMI development over time (models include 4 splines for 4 time periods):

for mental health:

 
 eststo: mi estimate, post: mixed mh age10_16 age16_23 age23_33 age33_42 || id: agec, cov(unstr) mle stddev

for BMI:

 
 eststo: mi estimate, post: mixed bmi age10_16 age16_23 age23_33 age33_42 || id: agec, cov(unstr) mle stddev

The above models run well.

I would now like to predict the individual-level intercepts and slopes for mental health & BMI generated by the above two models as I plan to run further linear regressions between them.

So, I would have mental heath slope and intercepts variables and BMI slope and intercepts variables (generated from the spline models above) to run a regular linear regression like:

regress bmislope mentalhealthslope bmi-intercept mh-intercept

Running, for mental health:

 
 predict mh

only gives me the average slopes for the four splines (i.e. all individuals have the same set of four slopes) and not the individual level slopes that I would like to have. I would I obtain these?

Thanks!
/Amal

[Clyde Schechter]

Following -mixed-, -predict- has an -reffects- option that will get you the random intercepts and slopes. -help mixed postestimation- and click the -predict- link for more information.

[Amal Khanolkar]

Hi Clyde

Thanks - this is my mixed effects model:

eststo: mi estimate, post: mixed bmi age10_16 age16_23 age23_33 age33_42 || id: agec, cov(unstr) mle stddev

I then tried:

predict bmifull, fitted relevel(id)

and:

predict bmifull, reffects relevel(id)

and:

predict u1 u0, reffects reses(s1 s0)

but in all cases, I get error msgs:

option fitted not allowed or
option reffects not allowed

Am I misspecified something?

Thanks
/Amal

[Clyde Schechter]

Oh, I overlooked the fact that you are using multiple-imputation. -predict- is very restricted after multiple imputation. I'm sorry, I don't know how to get the random component estimates in this situation. If somebody else does, I hope he or she will chime in.