Listers,
This is really more of a stat than an spss question. It’s kind of involved, but any advice would be much appreciated.
I want to use the slopes and intercepts from a longitudinal growth model to predict a dichotomous outcome variable that only occurs at the last time point. There are three time points.
It is couples data, with individuals nested in couples, nested within time points. PersDist is my dv, time (1,2,3), Male_dum & female_dum (0,1, 1 when the line is male’s or female’s respectively)
mixed
PersDist with Time male_dum female_dum
/criteria = mxiter(1000000000)
/fixed = male_dum female_dum male_dum*Time female_dum*Time | NOINT
/method = ml
/print = solution testcov
/random = male_dum female_dum male_dum*time female_dum*time | subject(dyad) covtype(un)
/repeated = sex | subject(dyad*time) covtype(un)
/save = pred(persdist_pred).
This results in the predicted scores for men and women on this dependent variable. To get the slope and intercept for each person a split file on individual is run. With split file on a individual regressions were run for each person, using time to predict the predicted scores (persdist_pred). The coefficients from each regression were then read into a new file. That file then contained the intercepts and slopes for each person across the three time points.
The plan is to then use those intercepts and slopes to predict the outcome variable that happens at time 3.
On top of that I’m then going to use the actor partner interdependence model to estimate actor and partner effects for the intercepts and slopes.
Meaning that for each slope and intercept they would be entered like so:
Slope Intercept
Actor Partner Actor Partner
Man 1.2 3 8 6
Women 3 1.2 6 8
This is a dyad design. You can see that the individual’s actual data is under actor and their partner’s data is under partner. I will then use these intercepts and slopes to run another multilevel model with individuals nested within couples predicting a dichotomous outcome.
Is this kosher? I’m worried that I may be violating some assumption by using slopes and intercepts like this. And was my method for deriving the slopes and intercepts acceptable?
Thanks,
Matt
Locked
