[R-sig-ME] Model for irregular design
paul
pwschmitt at gmail.com
Mon Aug 6 22:08:16 CEST 2012
Dear All,
I am interested in using random effects/ mixed effects models for
estimation purposes. Suppose I study 7 treatments in each of 3 clinical
studies with treatments allocated among studies as indicated; study 1:
trt_1, trt_2, trt_3; study 2: trt_1, trt_4, trt_5; study 3: trt_1,
trt_6, trt_7. I recognize the fact that this design set up is
asymmetric and probably has little to recommend it. I will assume
complete pooling is inappropriate. Nevertheless, I want to analyze this
data using lmer/lme or Winbugs. I believe this study design imposes
certain conditional exchangeability constraints. For example, given
study 1 it would seem to follow that trt2 and trt3 are exchangeable.
For this reason it would seem the model formulation below is reasonable:
trt_i ~ N(s1, sigma^2); i = 2,3
trt_i ~ N(s2, sigma^2); i = 4,5
trt_i ~ N(s3, sigma^2); i = 6,7
Above s1, s2, and s3 denote random effects associated with treatment
pairs which may be considered exchangeable.
In addition, we have:
trt_1 ~ N(s0, sigma0^2)
where s0 = 1/3*(s1 + s2 + s3)
If I went ahead and imposed vague priors on the appropriate parameters
above, I believe I could set up a WINBUGS program to obtain posterior
estimates for each trt_i; i = 1, 2, ..., 7. My question: is it possible
to set up and solve this problem using lmer/lme? If so how would I do it?
Thank you for any assistance on this question.
Regards,
Paul
More information about the R-sig-mixed-models
mailing list