nlcom on imputed data + GEE model

[Kristen McArthur]

Hi all,

I would like to compute a non-linear combination of coefficients for my exposure (airborne lead concentration) and exposure squared for a GEE model using multiply-imputed data. I am able to use nlcom following the GEE model using complete case data, but when I try to use the MI data, I get "last estimates not found." Is there an mi version of nlcom, or some sort of workaround anyone is aware of? Thanks in advance!

[Rich Goldstein]

it is not clear to me exactly what you are trying to do, but see

help mi_postestimation

[Kristen McArthur]

Hi Rich,

for example, I am able to run the following:

xtgee lead lead_squared covariates, family(gau) link(i) robust
nlcom _b[lead] + _b[lead_squared]

but would prefer:

mi est: xtgee lead lead_squared covariates, family(gau) link(i) robust

followed by an equivalent to nlcom that works for MI data

thanks!

[Rich Goldstein]

sorry, but what you say you are "able to run" makes no sense to me - maybe someone else can help

added: note that your -xtgee- command calls for "lead" to be your outcome (dependent) variable and thus there will be no such thing as "_b[lead]"; this may just be a typo of course; also, I see no reason why one would want the sum of the coefficients of x and x-squared; maybe if you fix your estimation command and explain what you want to obtain from your -nlcom- command, someone can help

[Kristen McArthur]

Apologies - it was a typo, executive function is the outcome:

Hi Rich,

for example, I am able to run the following:

xtgee executive_function lead lead_squared covariates, family(gau) link(i) robust
nlcom _b[lead] + _b[lead_squared]

but would prefer:

mi est: xtgee executive_function lead lead_squared covariates, family(gau) link(i) robust

followed by an equivalent to nlcom that works for MI data

thanks!