Paired samples t-test for an interaction


New Member
Hi all,

I'm currently writing up my final project for university.
The general idea is testing the role of language on emotion perception in faces, with the hypotheses being that emotion words will facilitate more accurate perception of emotional faces than do neutral words (i.e. more correct 'yes, this is an emotional face' responses). Word type and relevance have also been manipulated.

So in all, there are 3 variables with two levels each:
Word Valence (Emotion / Neutral)
Word Type (Label / Verb)
Relevance (Relevant / Irrelevant word to the emotion depicted on the face)

Example: the Emotion Label Relevant word is 'sad', the Emotion Label Irrelevant word is 'surprise'. The corresponding Verbs are 'sob' and 'gasp', respectively. Neutral words were simply matched for letter length to their emotion word counterparts.

It is a repeated measures design.

I have already conducted 3 separate ANOVAs for 'HITS' (correct yes responses), FALSE ALARMS (incorrect yes responses) and D PRIME (proportion data indicating how well participants actually discriminated between emotion and neutral faces); I found a statistically significant interaction between Word Valence and Word Relevance from the HITS ANOVA.

This is where I get stuck. I need to conduct paired samples t-tests to see where the differences lie - I plan on doing tests to compare:
Emotion Relevant and Emotion Irrelevant scores
Neutral Relevant and Neutral Irrelevant scores
Emotion Relevant and Neutral Relevant scores
Emotion Irrelevant and Neutral Irrelevant scores.

However, the data I have is also split in to Word Type - i.e. I have two scores for every participant for Emotion Relevant, Emotion Irrelevant, Neutral Relevant, etc.

My question is (and I'm hoping you guys can help!) what do I need to do to the data so that I can conduct a paired samples t-test correctly? I have tried simply adding the Label and Verb data in to one column together (taking N from 44 to 88), but that doesn't seem right to me.
Or, I'm thinking my other option is to go through each participant's scores and average their Label and Verb scores together, so that N will still equal 44.

Any suggestions will be greatly appreciated, thank you in advance and thanks for taking the time to read this if you got this far!