topic Re: t-values for PROC MIXED in Statistical Procedures

t-values for PROC MIXED

stats2554 — Mon, 03 Apr 2023 21:31:03 GMT

I've performed a repeated measures ANCOVA with PROC MIXED (effects: Treatment, Time, TreatmentxTime) and have the results I need from the pairwise comparisons with Dunnett's. I now need to determine minimum detectable difference using a formula I was given. I have the standard error of the difference between LSMeans but I need to compute the following t-values for the equation:

[t(.05,ddf) + t(.80,ddf)]

What is the best way of determining this value from my results output? ddf=denominator degrees from Treatment.

Thanks!

Re: t-values for PROC MIXED

ballardw — Mon, 03 Apr 2023 23:13:46 GMT

There are multiple statements that will result in tvalues of one sort or another in Proc Mixed output.

If you have one or more ESTIMATE statements then that is one of the more common statements.

It might help to start by showing the code you are currently using as well.

Re: t-values for PROC MIXED

stats2554 — Mon, 03 Apr 2023 23:47:27 GMT

We're mostly concerned with the pairwise comparisons following a significant treatment or treatment by time interaction, at which point each timepoint is analyzed with the following code:

proc mixed data=Result;
class subject treatment;
model result = baseline subject treatment;
lsmeans treatment / adjust=dunnett;
run; quit;

Finding the standard error of the difference in lsmeans is easy enough, but I'm not sure how to find the 2 t-values [t(.05,ddf) + t(.80,ddf)], which will be multiplied by the stderr of the difference in lsmeans. Here is an example output:

Thanks!

Re: t-values for PROC MIXED

ballardw — Tue, 04 Apr 2023 16:25:22 GMT

I am not sure what you mean by t(.05,ddf).

SAS has several functions related to returning values from distributions. I suspect you may want one of these but insufficient information for sure.

data example;
   a1 = cdf('T',0.05,2);
   a2 = pdf('T',0.05,2);
run;

Return values for the T distribution cumulative density function and probability density functions. The first parameter is the distribution name. The other parameters depend on the distribution. There are also SDF (and LOGC/P/Sdf) and Quantile functions for different pieces.

2 above is the Degrees of Freedom.