messy data: proc glm with random effects?

pbwn — Tue, 29 Aug 2017 01:48:27 GMT

i have a measurement on patients at 3 fixed timepoints, but at each timepoint the 'treatment' can change between two types (this is not a 'designed' experiment such as incomplete blocks crossover that would randomise treatment sequence to patients, and it is not literally a 'treatment', it is just easier to explain it this way). Hypothetical data would appear as follows

patient timepoint treatment Y

1 1 a #

1 2 a #

1 3 b .

2 1 b #

2 2 a #

2 3 b #

3 1 a .

3 2 a #

3 3 a #

thus a patient does not necessary receive both 'treatments' (a and b), the data are messy eg unbalanced + missing data (Y=outcome). It seems to me patients should be included as random effects. Maybe it is analogous to a split plot design, with patients as a blocking factor, although i read that "with PROC GLM, you must use a TEST statement to obtain the correct F test for A" (https://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/viewer.htm#statug_mixed_sect033.htm). Thus is the following code insufficient?:

proc glm;

class patient timepoint treatment;

model y = patient timepoiint treatment treatment*timepoint / ss3;

random patient;

run;

although it is repeated measures, the repeated statement doesn't seem useful here because the 'treatment' changes over time.

Thanks for any advice

Re: messy data: proc glm with random effects?

Ksharp — Tue, 29 Aug 2017 13:11:48 GMT

Why not use PROC MIXED ? since PROC GLM only fit dummy mixed model .

topic Re: messy data: proc glm with random effects? in Statistical Procedures

messy data: proc glm with random effects?

Re: messy data: proc glm with random effects?