Mann–Whitney U test

[aska_ujita]

Hello there, 

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 ).

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

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

My datas have non-normal distribution and are binomial.

I used PROC GLIMMIX:

           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;

But I read some papers that used "Mann–Whitney U test", is there somebody who can explain how to do it?

or just PROC GLIMMIX is already enough?

Thank you for the help.

[Norman21]

Hi,

If you look near the top of this page:

https://stats.idre.ucla.edu/sas/whatstat/what-statistical-analysis-should-i-usestatistical-analyses-...

(or search for "Wilcoxon-Mann-Whitney", which is another name for the Mann-Whitney U test), you will see an example using SAS.

[aska_ujita]

Hello Norman21, thank you for replying.

I tried the proc, but I have no idea how to interpretate...

Thank you for your help!!

[Doc_Duke]

Aska,

You could use the Mann-Whitney-Wilcoxon (MWW) within a day to assess whether the trained or untrained had different reactivity.  As each subject is assessed on each day, it would be inappropriate to simply lump all of the observations into one MWW analysis (violates the independence assumption).

As far as interpretation is concerned, this Wiki is reasonable

https://en.wikipedia.org/wiki/Mann%E2%80%93Whitney_U_test

and you can also get a lot from the SAS documentation.

I generally use the two-sided t-approximation for the test for the MWW.

I've not used GLIMMIX in years, so I can't comment on it's appropriateness.

Doc

[bikashten]

   Aska,

i think you can use this code for binomial distribution with repeated measures:

proc glimmix data=A;
	class subjectID treatment day;
	model reactivity=treatment day treatment*day / dist=binomial DDFM=KR;
	random subjectID;
	lsmeans treatment day treatment*day/ pdiff lines;

	 /* if there is a significant interaction then you can compare the treatment effects for each day*/;
	lsmeans treatment*day/sliceby=day pdiff lines;
run;

Thank you,

Bikash