topic PROC SURVEYSELECT FOR CROSSOVERS DESIGN: in Statistical Procedures

PROC SURVEYSELECT FOR CROSSOVERS DESIGN:

Weidsoncs — Tue, 13 Jul 2021 00:06:38 GMT

Dear sir,

An adhesiveness dataset was generated for 63 subjects (64 total -1 dropout = 63 final) from a Crossover Design with 2 periods and 2 treatments (Crossover 2x2 ). I intend to estimate a parametric boostraping confidence interval for the difference of the two treatments (T, R) in order to estimate the size of the bias. A more experienced SAS user suggested using PROC SURVEYSELECT to resampling from the final dataset (63 survey subjects). The code suggested by him can be verified below.

However, when I run this code I notice that in many replicates generated the treatment sequences (one of the cluster levels) were quite unbalanced (with very different sample sizes) when ideally they would have similar sample sizes. Does anyone have any suggestions on how to modify the code to get less unbalanced treatment sequences?

 Title Bootstrap analysis;

PROC SURVEYSELECT data=work.ADHESIVENESS method=BALBOOTSTRAP reps=100;
SAMPLINGUNIT Subject;
run;
quit;

Re: PROC SURVEYSELECT FOR CROSSOVERS DESIGN:

FreelanceReinh — Tue, 13 Jul 2021 20:11:34 GMT

Hello @Weidsoncs and welcome to the SAS Support Communities!

A simple way to improve the balance of the treatment sequences in the bootstrap samples would be to use stratified sampling, i.e., add the STRATA statement

strata Sequence;

to your PROC SURVEYSELECT step after sorting the input dataset (work.ADHESIVENESS) by variable Sequence. Then all 100 replicates will have 31 subjects (62 observations) with Sequence='RT' and 32 subjects (64 observations) with Sequence='TR'.

Not sure how this will affect your intended parametric CI estimation, though.

Re: PROC SURVEYSELECT FOR CROSSOVERS DESIGN:

Weidsoncs — Tue, 13 Jul 2021 22:13:09 GMT

The solution that I found was exataly this !!! Thank you for help...

see the complete code:

proc sort data=work.maydataset1;
   by Sequence;
run;

PROC SURVEYSELECT data=work.mydataset method=BALBOOTSTRAP reps=100 out=maydataset2;
SAMPLINGUNIT Subject;
Strata Sequence;
run;
quit;