topic Re: out of sample prediction standard error and 95% confidence interval in SAS Programming

out of sample prediction standard error and 95% confidence interval

cedric5 — Tue, 05 Jan 2021 21:21:01 GMT

I am analyzing growth data to determine if sibling competition affect growth using a GLIMMIX with different class and continuous variables and 2 random effects. I want to predict the out of sample data. I appended the dataset with missing values for the data I want to predict. It seems to be working but how do I get the standard error and the 95% confidence limits for the predicted data.

Thanks for your help!

proc GLIMMIX data=WORK.stdizedata namelen=200 method=MSPL scoring=3 IC=Q EXPHESSIAN PLOTS=all;
class hatch_year hatch_area rear_area rear_nest_trt home_or_away
hatch_nest_breed_ID rear_nest_breed_ID ;
model SMI_robust = hatch_year hatch_area rear_area rear_nest_trt home_or_away
d14_rear_nest_brood_size rear_Cs_at_start_of_rearing net_rearing_manipulation
Date_of_day14 Date_of_day14*hatch_year
hatch_year*hatch_area hatch_year*rear_area hatch_year*rear_nest_trt hatch_year*home_or_away
hatch_area*rear_area hatch_area*rear_nest_trt rear_area*rear_nest_trt
d14_rear_nest_brood_size*rear_Cs_at_start_of_rearing rear_Cs_at_start_of_rearing*net_rearing_manipulation
/ dist=gamma DDFM=KENWARDROGER solution;
nloptions maxiter=15000 technique=NRRIDG;
random intercept / subject=hatch_nest_breed_ID type=UN(1);
random intercept / subject=rear_nest_breed_ID type=UN(1);
output out=pred pred=xbeta pred(ilink)=predprob pred(ilink noblup)=fix_predprob;
run;

Re: out of sample prediction standard error and 95% confidence interval

Norman21 — Wed, 06 Jan 2021 03:48:22 GMT

If I understand your question correctly, the list of predicted values will be point estimates. If you want to summarise the predicted values data set, you can use the file in a PROC MEANS procedure (or similar).

A useful summary of PROC GLIMMIX can be found here: https://support.sas.com/resources/papers/proceedings/proceedings/sugi30/196-30.pdf

Re: out of sample prediction standard error and 95% confidence interval

cedric5 — Wed, 06 Jan 2021 13:06:38 GMT

Thank you for your answer. I actually want the standard error and 95% confidence interval for each point estimate for the new data. I have 3 scenario that are out of sample and I use the model to predict the value for these 3 scenario (3 new data) and for each I need the the standard error and 95% confidence interval.

Re: out of sample prediction standard error and 95% confidence interval

cedric5 — Thu, 14 Jan 2021 13:24:02 GMT

Just to clarify, I want to account for uncertainty arising from residual variance, from the fixed effect coefficients, and also in the variance parameters for the random effects. I want to compute standard errors and 95% confidence intervals around the predictions using bootstrapping methods.

Anyone can help?