Independent sample t test using the diff variable

[Ashwini_uci]

Hello,

I am trying to use the independent sample t test to see if the change in a variable is significantly different in 2 groups.

The variables are platelet inhibition before and after a intervention of a drug.Ideally the platelet inhibition should increase after the intervention. Now I want to see if the difference in the platelet inhibition (after minus before drug intervention) is significantly different in diabetic and non-diabetic patients.

But when i create this new variable of "difference" (after minus before), there are some values that are negative... it seems the platelet inhibition didnot increase in some patients after using the drug.

My question is if it is alright to still use this variable (with a few negative values) while pefomring the independent sample t test or non parametric the Man-whittney test?

Logically it sounds right to me..but i would like to get some experts opinion on this.

Response appreciated!

Thanks

Ashwini

[PGStats]

Well, look what happens when you simulate 80 normal values with a mean of 1.5 and a standard deviation of one. You get some negative values. That's perfectly normal.

data test;
call streaminit(56798);
do id = 1 to 80;
     x = rand("NORMAL",1.5,1);
     output;
     end;
run;

proc univariate data=test normal;
var x;
histogram / nmidpoints=8 href=0;
run;

PG

[Ashwini_uci]

Thanks very much for this illustration! that makes perfect sense.

I am wondering after I perform the Independent t test or the man-whittney test using this 'difference"
(after-before) variable, "what exactly' in the results tells me that the "difference" is positive (after values >than before values; which is expected) and not negative.

I know the results will tell us if this "difference/change in platelet inhibition after intervention " in the 2 groups (diabetic and nondiabetic) is significantly different or not. But I am curious to know how I would know if the overall difference/change is positive or negative for the whole data.

Appreicate your response!

Ashwini

[PGStats]

Whatever procedure you use to do the test will give you the means or some measure of location, so you will know which group is higher. Also, given that your hypothesis is one-sided, you should be doing unilateral testing. - PG