margins & marginsplot after mixed effect/multilevel modelling

[Amal Khanolkar]

Hi All

I've used linear spline multilevel modelling to analyses rate in change of multimorbidity in a group of individuals:

generate the spline variables (4 variables that correspond to 4 periods, wave1 = 1982 to 1989, wave2 = 1989 to 1999, wave3 = 1999 to 2009 & wave4 = 2009 to 2015:

mkspline wave1 1 wave2 2 wave3 3 wave4 = wavex

Run the multilevel model with the 4 spline variables (randome slope & random intercept):

mixed score wave1 wave2 wave3 wave4 sex || NSHD_ID: wavex, cov(unstr) mle stddev
margins sex, at(wavex=(0 1 2 3 4 ))

The multilevel model works just fine:

Mixed-effects ML regression Number of obs = 18,615
Group variable: NSHD_ID Number of groups = 3,723

Obs per group:
min = 5
avg = 5.0
max = 5

Wald chi2(5) = 2057.52
Log likelihood = -29275.269 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
score | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
wave1 | .4517862 .0221717 20.38 0.000 .4083305 .4952419
wave2 | .3639538 .0221717 16.42 0.000 .3204981 .4074095
wave3 | .2409347 .0221717 10.87 0.000 .197479 .2843904
wave4 | .4963739 .0221717 22.39 0.000 .4529182 .5398296
sex | .1654415 .0260374 6.35 0.000 .1144092 .2164739
_cons | .3395521 .0425618 7.98 0.000 .2561326 .4229716
------------------------------------------------------------------------------

------------------------------------------------------------------------------
Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------------------------
NSHD_ID: Unstructured |
sd(wavex) | .4243546 .0072879 .4103083 .4388818
sd(_cons) | .5270375 .0181214 .4926906 .5637788
corr(wavex,_cons) | -.1251645 .033656 -.1904833 -.0587422
-----------------------------+------------------------------------------------
sd(Residual) | .9083197 .0060774 .896486 .9203096
------------------------------------------------------------------------------
LR test vs. linear model: chi2(3) = 7825.50 Prob > chi2 = 0.0000

However, I get the error msg below when trying to run the margins command. I haven't had this issue before. I don't want to include the 'wavex' variable in the fixed effects part of the model as I have the spline variables. Is there a way around this?

variable 'wavex' not found in list of covariates
r(322);

end of do-file

r(322);

Thanks!
/Amal