Anonymous posted on Wednesday, May 25, 2005 - 7:18 am
Hello Dr. Muthen,
I have a rather basic question. I have a data-set with three time-points and I used cross-lagged analysis to capture direction of effects and LGCM to capture correlated change. For the cross-lagged analyses, I used SEM controlling for all stability and within-time correlations. I found that variable X at times 1 and 2 did not predict variable Y at times 2 and 3, respectively. Similarly, variable Y at times 1 and 2 did not predict variable X at times 2 and 3, respectively. So the cross-lagged analyses could not substantiate any causal links between both variables in either direction. The LGCM analyses, however, indicated that the slopes of both variables were significantly related. How exactly must I explain this combination of findings because a reviewer stated that these findings are rather incompatible? I know that both analyses capture different phenomena but I have difficulty puting this into words. Also, do you know any articles or chapters that clearly explain the difference between both sets of analyses and how they complement one another?
bmuthen posted on Wednesday, May 25, 2005 - 8:37 am
I can see a couple of reasons for the contradictory results. First, your growth model for the 2 processes may be misfitting - for example if the 2 outcomes are allowed to correlate only via their growth factors and not through direct residual correlation at each time point, too much correlation is channeled through the growth factors. Second, the growth model is concerned with slopes, while the cross-lagged model is concerned with each time point.
Somewhat related literature included writings on hybrids of growth and auto-regressive modeling by McArdle, and by Bollen and Curran.
So, if I understand it correctly, you would change the default of m-plus and let the residuals of both variables correlate at each time-point and then see if the same significant correlations among the slopes would still pop up?