I'm wondering what would be the best method to analyze a "cross-over" interaction between a factor and a continuous variable.

Here's my experimental set-up and hypotheses in a nutshell: 58 participants were randomly assigned to one of two conditions (each n=29) with different manipulations. It was hypothesized that a continuous individual difference variable would make participants more susceptible to the manipulations. The different manipulations were intended to have the opposite effects on the DV. Thus, it was hypothesized that in Condition 1 the DV and the IDV would have a negative correlation, and in Condition 2 they would have a positive correlation.

Indeed, correlation analyses (between the DV and the IDV) revealed that r = .-37, p < .05 in Condition 1, and r = .26, p = .18 in Condition 2.

I've tried the General and Generalized Linear Models on SPSS to properly test my hypothesis. In the models, I included main effects of Condition and the IDV (which I entered as a covariate), and an interaction effect, but the results don't appear to be very robust (i.e., even the smallest changes in the models seem to have a big effect on the results). I've used dummy coding (0s and 1s) for Condition, and the DV and the IDV appear to be normally distributed.

Does this seem like an appropriate way of testing my hypothesis, or can someone suggest a better approach?


If the data is as follows:

id  condition   IDV DV
1   1           5   9
2   1           6   8
3   1           2   7
1   2           4   6
2   2           5   8
3   2           6   9

Then you can use anova with error correction as follows:

aov(DV ~ IDV * condition +Error(id/(IDV*condition)), data=mydata)

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.