Change scores: t-test or mann-whitney U test ?

#1
Hi there

Nice forum btw.

Question:
I have change-scores on pain in 37 patients pre/post intervention. which test do you use? t-test or mann-whitney U test?

i was told that with change-scores they are always normally distributed and i can use a t-test... any reference for such a statement?
p.s. a kolmogorov-smirnof test proved normal distibution, but still, anybody knows why change scores are supposed to be normally distributed ???

kind regrads and thx for help (and reference :D )
 

Karabiner

TS Contributor
#2
I have change-scores on pain in 37 patients pre/post intervention.
On what scale(s) were they measured?
i was told that with change-scores they are always normally distributed
No, that is not true.
and i can use a t-test
With n=37 and no outliers you could consider t-test.
By the way, it's the residuals, not the unconditional
distribution what matters.
kolmogorov-smirnof test proved normal distibution
No, it just failed to reject the H0 that the sample
data aree prom a normal population. Since K-S
withoput Liliefors correction has very low power, this
doesn't tell you much. You should perhaps take a look
at the Q-Q plot (of the residuals) to gain better insight.

With kind regards

K.