meeting with @GracielaMuniz, @annierobi, @casslbrown, @smhofer
Quick points
last fall we became interested in the general question "What happens to the statistical model of growth as the longitudinal study matures?"
the interest stemmed from a practical problem deciding the number of waves that should be included into estimation of the model of change.
we focused bivariate linear model used during the Portland workshop, which is also the key model in Rast and Hofer (2014) and (see model specification)
in October 2015 we gave this presentation, in which we presented two informational displays designed to investigate this question: (A) summary view of a single model across unfolding waves, and (B) dynamic view of unfolding model predictions
(A) Kb_profiles (google slides or github show what happens to the values of the intercept and slope as we increase the number of waves (y-dimension) over which a bivariate linear model fits. Each graph examines a unique pair of outcomes estimated over all possible(reasonable) number of waves for males and females separately.
(B) Kb_fans ( on example of grip - number comparison pair: google slides or see github for a better labelling) focuses on a given pair of outcomes (as in this case grip and numbercomp) and graphs (1) the smooths of the observed trajectories (2) trajectories reconstructed from the estimated fixed effect and (3) trajectories reconstructed from the factor scores reported by MPlus. These three form the columns. The rows in the kb_fans displays is the two time metrics: time (top) and age (bottom). They are animated, each frame showing a graph for a time point in a life of a longitudinal studies. So the animation has an interesting interpretation: Imaging that you start fitting a bivariate linear model on grip and number comparison when this study had only 4 waves. With each consecutive year you refit your models with the new waved added to the dataset. So you proceed until the end of the study. The animation shows how your model will be changing with the life of the study.
NEXT STEPS
(1) In short, we would like to use these informational displays to examine the effects particular variation in temporal designs of longitudinal studies.
meeting with @GracielaMuniz, @annierobi, @casslbrown, @smhofer
Quick points
grip - number comparison
pair: google slides or see github for a better labelling) focuses on a given pair of outcomes (as in this case grip and numbercomp) and graphs (1) the smooths of the observed trajectories (2) trajectories reconstructed from the estimated fixed effect and (3) trajectories reconstructed from the factor scores reported by MPlus. These three form the columns. The rows in the kb_fans displays is the two time metrics: time (top) and age (bottom). They are animated, each frame showing a graph for a time point in a life of a longitudinal studies. So the animation has an interesting interpretation: Imaging that you start fitting a bivariate linear model on grip and number comparison when this study had only 4 waves. With each consecutive year you refit your models with the new waved added to the dataset. So you proceed until the end of the study. The animation shows how your model will be changing with the life of the study.NEXT STEPS