I am attempting to examine how over-time changes in romantic partners' goals during an argument are associated with their post-interaction assessments of conflict resolution.Using Karney and Bradbury's (1995) 2-step procedure, I ran a separate OLS regression for each individual to obtain person-specific intercepts, slopes and standard errors, then used these as predictors in a subsequent APIM in AMOS.

But in the above approach, the growth curve parameters could be biased by the fact that couples differed in argument length. Goals were reported at one minute intervals, and arguments ranged from 3 to 10 minutes (couples stopped when they decided they were finished). Thus, there were anywhere from 3 to 10 time points for a given individual on which to compute their growth curve parameters. My question is whether this could potentially bias the parameters (I am primarily interested in the effects of slopes and standard errors, i.e., "goal fluctuations," on conflict resolution).

Because of this, I'm wondering if MLM is a better approach, due to its greater flexibility with missing data. I know that I could run a random-intercepts, random-slopes model to test whether within-person intercepts and/or slopes influence the growth rate in resolution over time. But I'm not aware of any way to model the within-person standard errors as predictors with MLM using SPSS. And one of my primary questions is the extent to which the standard errors are themselves predictive of actor and partner resolution perceptions. Is there any way to run a "random standard errors" model?