topic Re: The best test in Statistical Procedures

The best test

aska_ujita — Fri, 05 Apr 2019 14:29:46 GMT

Hello there,

I would like to know which test is better to use in my data.

I am testing two treatments (trained and not trained). And after I collected the behaviour and scored it in 0 (absence of reactivity) and 1 (presence of reactivity). The data were collected in days of lactation (1, 3, 5, 7 and 9 ).

So... I would like to compare between treatments by day.

Example:

Day of lactation ---> Percentage of reactivity by group ---> Significance

Day 1 ---> 30% (trained group) 80% (not trained group) ---> P value

Day 3 ---> 20% (trained group) 60% (not trained group) ---> P value

Day 5 ---> 10% (trained group) 45% (not trained group) ---> P value

Day 7 ---> 0% (trained group) 30% (not trained group) ---> P value

Day 9 ---> 0% (trained group) 20% (not trained group) ---> P value

Mean ---> 12% (trained group) 47% (not trained group) ---> P value

Thank you for the help.

Re: The best test

StatDave — Sun, 07 Apr 2019 20:38:40 GMT

From your description, I'm guessing that you have subjects that are randomly assigned to the treatments and then are observed over time so that you do not have independent observations across the days. If correct, you could consider various longitudinal approaches including a random effects model (in PROC GLIMMIX), a Generalized Estimating Equations (GEE) model (in PROC GEE or GENMOD), or a stratified conditional logistic model (in PROC LOGISTIC). For example, a GEE model could be fit with code like the following which models the probability that REACTIVITY=1 and allows for the effect of treatment to vary by day. The SLICE statement provides tests of the treatment effect in each day.

proc gee;
  class subjID treatment day;
  model reactivity(event="1") = treatment day treatment*day / dist=bin;
  repeated subject=subjID;
  slice treatment*day / sliceby=day;
  run;

Re: The best test

aska_ujita — Wed, 10 Apr 2019 11:35:32 GMT

Hello StatDave_sas!

Thank you for your reply!

I tried your example, but I got errors because the final hessian.

My proc structure is this:

proc glimmix data=A;
class treatment day;
model reactivity=treatment day treatment*day / dist=binomial DDFM=KR;
random subject subjectID;
lsmeans treatment/ adjust=tukey lines;
by day;
run;

Is that correct?

Thank you very mych.

Aska.