using within interaction, or run t-tests?

[jamie10]

Hi everyone,

I have a lot of within factors for data I am analysing - it was a 2x2x2x2 within subjects ANOVA.

I am interested in one condition (called condition (repeated vs long) - that all participants had) and another condition (response: hits vs. correct rejections).

Not I can see from the output in SPSS that I am given within effects for each and also an interaction.

However, I am not sure if this interaction is near being equivalent to running two seperate t-tests (dependent t-tests) for each condition (1 t-test for repeated and 1 t-test for long) with the responses as the DVs?

Any thoughts on which to go for?

Thanks!

[Dason]

If the assumptions of the anova are met then anything you could do with t-tests you could do with the anova and you'd have a better estimate of the MSE in the anova (assuming the assumptions are correct).

My problem is I don't really understand what you're asking. Can you rephrase what it is you're interested in?

[jamie10]

Thanks Dason,

Sorry if what I was asking was unclear, let me try to rephrase...

So with the output from SPSS I will be given the main effects of each within factor and also the interaction between the factors. I am particularly interested in the conditions: 'condition' (two levels: repeated vs. long) and 'response' (two levels: hit vs. correct rejection).

Rather than looking at the interaction between these two factors (which are part of a 2x2x2x2 repeated measures ANOVA) can I not just do two t-tests: the IV will be condition for each t-test, with the DV being the response (hit for 1 t-test, and correct rejection for another t-test).

Does this make sense?

Perhaps this is not really possible because by doing this I may not fully account for the other within factors that contribute to the values in the column (ERP magnitude)...

[Dason]

Well you could but I would recommend against that. You should probably do some post-hoc comparisons using the ANOVA. You should be able to get the exact tests you're looking for but you'll have a better estimate of the standard deviation if you do it using post-hoc tests with the ANOVA.

[jamie10]

Thanks Dason, that's fair enough, I will probably avoid doing it...