proc MI fcs discrim

Yang2018 — Fri, 11 May 2018 15:42:09 GMT

Hello,

I have a question related to the topic of multiple imputation. I am hoping that you can help me out.

I am trying to impute missing values in a dataset that has mostly categorical variables (binary, ordinal or nominal) and one continuous variable (age). I am using the discriminant FCS method for missing categorical variables (total 😎 and regpmm for one missing age (1). The other variables have no missing value.

I am doing this in SAS 9.4 (SAS/Stat 14.1) and only set imputation number as 1, I tried to get the imputed data set quickly. This below code can be run without error information, however, it can't give me the final imputed data set, even after 1 day run. Did I include too many missing variables (total 9) in this model? Why it runs but no output after long time waiting, even I just set nimpute=1? What further model modification I can do to get final imputed data set? Thanks.

Here is the example code I used:

proc mi data=OUTPT.analysis MU0=35 minimum=0 maximum=108 round=1 nimpute=1 out=OUTPT.imputed;

class SEX_C RACETH_C D_C P_C I_C BD_C DI_C LO_C TY_C IN_C FM_C PS_C ST_C SP_C SC_C PR_C HO_C;

fcs regpmm(AGEYR_N/details);

fcs discrim(RACETH_C/details classeffects=include);

fcs discrim(SEX_C/details classeffects=include);

fcs discrim(P_C/details classeffects=include);

fcs discrim(I_C/details classeffects=include);

fcs discrim(BD_C/details classeffects=include);

fcs discrim(DI_C/details classeffects=include);

fcs discrim(LO_C/details classeffects=include);

fcs discrim(TY_C/details classeffects=include);

var AGEYR_N SEX_C RACETH_C D_C P_C I_C BD_C DI_C LO_C TY_C IN_C FM_C PS_C ST_C SP_C SC_C PR_C HO_C TRE_N;

run;

Re: proc MI fcs discrim

Yang2018 — Mon, 21 May 2018 13:37:49 GMT

Is there anybody who can help to solve this issue? Why the code can be run but no final imputed data set is available? Thanks.

Re: proc MI fcs discrim

SAS_Rob — Tue, 22 May 2018 14:01:34 GMT

Are there any messages in the LOG? How large is the input data set? How many levels are there to the categorical variables?

I would probably start by reducing the size of the models, especially if there are lot of levels to the categorical variables.

Instead of using all the categorical variables for the DISCRIM models, what happens when you use only a few predictors in each model.

topic Re: proc MI fcs discrim in Statistical Procedures

proc MI fcs discrim

Re: proc MI fcs discrim

Re: proc MI fcs discrim