If you are wanting to make sure that your data is normal - before comparing male and female. Would you..

A) test for normality across the WHOLE sample?

B) test for normality for the male and female samples separately?

I'd test for normality for males and females spearately.
Though depending on sample size t-test and ANOVA are fairly robust and can stand up to some violations of normality (I think)

based on something I've just been doing I think you can test for the normality of the difference between male and female... and as long as that is normally distributed then its OK to compare even if the samples themselves are not normally distributed... I think (hope)!

Thanks - am going to test for normality separately in the groups. ANOVA is robust to deviations in normality IF the sample sizes are equal. T-tests should have normal distribution - HOWEVER, I'm doing bootstrapped versions, which helps.

======= Date Modified 01 Jun 2010 12:58:42 =======
Ribenagirl. i think you might be thinking of the repeated measures t-test (this term doesn't quite seem right...what's it called again?!), where you are testing the difference between two scores against 0 and so check the difference between score 1 and 2 are normally distributed.

B :)