Hi every one,
I have a question about the correctness of my analysis method and i appreciate your advice. I have measured the number of fixations on some target images and some distractor images. I expected to see a higher number of fixations on the targets compared with the distractors. At the same time, these targets and distractors were in the left visual field or in the right visual field. I knew that there was an inherent bias toward left in our visuospatial processing. Hence I compared the number of fixations on the targets and distractors in each visual field separately (e.g. targets in the left visual field compared with distractors in the left visual field, and I did the same for the right visual field). I used a paired t-test for this comparison in each visual field. I was wondering whether my method is correct. I didn’t use repeated measure ANOVA because it doesn't count for any prediction about where the effect should be with higher probability. Hence my hypothesis wasn't to find any change between the targets and distractors. For example, if I saw a significant difference between the targets and the distractors in the left visual field but I didn’t see any difference in the right visual field it would be acceptable according to the leftward bias. But the opposite of this would not be acceptable. It would not be acceptable if I saw a difference in these measures in the right visual field but not the left visual field. I appreciate your advice.
Thanks
Hello Haleh,
You could set this up as a two-factor repeated measures anova (factor A: image, either target or distractor; factor B: visual field, either left or right). From your description, one would expect to see:
1. A significant image effect (target > distractor)
2. A significant field effect (left > right)
3. Possibly a significant image x field interaction (e.g., target-distractor differences are unequal across visual fields).
The advantage of the RM set-up is that you'll have a more suitable error term for the interaction test (which appears to represent your principal research interest). As well, simple effects tests may be evaluated should the interaction prove to be noteworthy.
Good luck with your work.
Just adding to David's answer:
Unless your sample size is not large (say, >30 per group) and or the "number of fixations on the targets and distractors in each visual field" is low, you should you a model for counts (Quasi-Poisson of negative binomial, depending on the assumed variance/mean relationship).
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA