I have some data with a continuous outcome variable, several time-varying predictors, and two conditions (Group A and B). The variables were collected across several waves.
I want to take a look at whether these time varying predictors are associated with the outcome variable over time (e.g., Is quality of life increased in times of high social support)?)
My assumption is that these predictors will have a greater association with the outcome in one group than the other (e.g., social support will be a greater predictor of QOL for people in Group A than Group B).
Is there a good way to go about this? I've considered and ran a fixed effects regression which reveals these predictors are significant for the whole sample, but I am unsure how to compare predictors between the two groups.
Any help would be much appreciated!!!!
One approach would be to include the grouping variable as a moderator using interaction terms (moderated regression). Another, more flexible approach is multigroup analysis that allows you to directly compare the regression coefficients and other parameters across groups.
Also, since you have longitudinal data, autoregressive/crosslagged path analysis may be a good choice for you because you can then include multiple outcome (dependent) variables at multiple time points in a single model/analysis. Such a path analysis model can also be specified as a multigroup analysis.
Multigroup regression and path analysis require software for structural equation modeling such as lavaan (free R package), Mplus, AMOS, or Open MX.
Christian Geiser, as I understood Justin S. Bell's question, there is only one outcome variable (quality of life). In that case, a multilevel random intercept (or possibly random coefficients) model would seem to be sufficient, would it not? Or did I miss something? Thanks.
See the attachment for an excellent set of course notes by a leader in the field.. Best wishes David Booth
Bruce Weaver You may be right. My reading of Justin S. Bell 's question was that the outcome (dependent) variable may have been measured several times as well (Justin S. Bell wrote: "The variables were collected across several waves."). In that situation, a crosslagged SEM/path analysis may be more flexible than a random effects model, particularly also for testing group differences.
Hi Christian Geiser. Okay, I see now that we did read the original question the same way, i.e., one DV (QOL) measured multiple times (i.e., over multiple waves). Thanks for clarifying.
- Badges
- Science topic
- Similar topics
- Correction