I am new to statistics and R and want to run some analyses, I wonder if you can help?
I am looking at a moderated mediation effect where perceived stress predicts sleeping problems through rumination, and how this is moderated by biosex.
It was suggested I first run a t-test to see how these variables relate to each other (ie does stress predict sleeping problems and so on), but i cannot figure out why I would be comparing mean scores of all of these constructs or really how this would work in R (all the examples I have seen are very different).
If anyone could take the time to explain what a t-test can tell me about this relationship I would be really, really grateful
A t-test requires a continuous dependent variable and a discrete predictor with two levels. Is sleeping problem a continuous variable? Do any of your predictor variables have only two levels? If you could explain more about the structure of your data, it would be easier to give suggestions. Which variables are continuous, which are discrete and how many levels do they have?
My moderator is biological sex and only contains bio males and females so it is discrete with 2 levels..
stress, sleeping problems and rumination are all scales which produce a continuous score.
I eventuallly wanted to see if stress predicts sleeping problems through rumination and whether biological sex moderates this mediation (ie to females have a greater stress -> rumination relationship comared to males)
So you could do separate t-tests of stress vs. biosex, rumination vs. biosex, and sleeping problems vs. biosex. Imagine that there is not a difference in the stress but there a difference in the rumination and sleep problems. That could mean that stress is unrelated to the other two or that the same stress produces different results depending on biosex.
Keep in mind that I have no experience with the kind of data you are looking at and I don't know the goal of your work. I'm assuming that the guidance you are getting is reliable.
This gets us into the multiple t-test issue. I have forgotten most of the stats I ever learned but it looks like a one-way ANOVA might be a bit more appropriate but we probably need more substantive information about the study.
On the other hand
"I eventually wanted to see if stress predicts sleeping problems through rumination and whether biological sex moderates this mediation" looks like it complicates the problem.
I'm looking in to one-way ANOVA as an alternative to the t-tests, to initially test if there was any significant difference between sexes in rumination scores, sleeping scores and stress scores.
regarding the second part, my plan was then run a moderated mediation model using sem, to see the pathway through which stress might lead to worse sleep through rumination, and if this pathway is moderated by sex. Does this complicate the t-test/anova problem?
I tink the Anova idea is sensible . I really know nothing about moderated mediation model so I am of no help here. If you are in an academic institution does it have a statistical consulting service or, perhaps your department may have a methodology expert who could give you a little time?
