2

I have set of 16 questions that have been answered on a numerical scale, from 1 to 7. They are answered by 200 couples, and I want to show there is a difference in the answers on gender. On a casual glance at the data, I see there is little difference between the genders for 6 of the questions, but is for the other.

If I run a series of t-tests for each, and show significant differences for 10 of them; I should be able to conclude that on the set as a whole, there are differences. Is there a better way to show that for this series, answers differ based on gender? Mu

user35534
  • 21
  • You will probably need to test each question separately by gender. Are the questions independent of each other? Or, could the answers to certain questions influence others? You can look at the correlation between questions separated by gender as well. – Drew75 Dec 01 '13 at 20:28

1 Answers1

1

You have a completely crossed three-way design, Gender x Question x Couple, with one observation in each cell. Do an analysis of variance. G and Q are fixed, C is random. The error term for the G main effect is the GC interaction. The error term for the GQ interaction is the GQC interaction. (Don't forget the correct for non-sphericity.) If the GQ interaction is not significant then, unless you make Bonferroni corrections, you may have trouble convincing people that apparent Gender differences on individual Questions are not just Type I errors.

Ray Koopman
  • 2,303