Corinna posted on Wednesday, August 05, 2015 - 2:20 am
Dear Linda & Bengt, I am wondering if it is possible to get plots or "results in probability scale" when calculating a conditional model (LCGA) with a binary outcome? With an unconditional model and the same outcome I get "results in probability scale" and can use the bottom "graphs". However, when I am including predictors, I am not able do this anymore. Is there a way to still get graphs or the needed data to create the graphs myself?
This is available for only the unconditional model.
Corinna posted on Wednesday, August 05, 2015 - 7:07 am
Dear Linda, thank you very much for your answer. Does this mean that there is no possiblity to describe the exact trajectory of the classes in the conditional model? Because I do not know the starting point, i.e. the probability of headache at the starting level and I do not know the course of the probability of headache across the assessment points?! I only know if the slope is positive or negative... so I can only make a statement about the influence of the predictors on i and s and on the class membership but not on the exact course of the trajectories of the different classes?
TECH4 contains the model estimated means and variances of the growth factors.
Corinna posted on Tuesday, August 11, 2015 - 5:49 am
Dear Linda, I do get means for intercept and slope when I use tech4 but since I am having a binary outcome I cannot interpret these means. On the y-axis I got the probability (between 0 and 1) and if I am correct the intercept represents the intersection with the y-axis. But since I have a binary outcome I cannot interpret the mean of my intercept directly (in my case it is 2.67 and this cannot be a probability). So can you tell me if there is a possibility to convert these scores into a score that represents the probability at first assessment (that means the intercection with the y-axis)? Thank you very much, best regards
Corinna posted on Tuesday, August 11, 2015 - 5:52 am
...moreover I only get one mean for the intercept (for each class) but I need to have the probability of headache (my outcome variable) for every time point. Is there any possibility to get this?
For an LCGA, which has zero growth factor variances, it is easy to express the probabilities of the binary outcomes a functions of the estimated model parameters and the covariate values using Model Constraint. First you express the growth factor means as functions of the covariates and then you express the outcome probabilities as functions of the growth factor means. For the former you use the regular growth formulas we give in our Topic 1 handout and for the latter you use the regular logit expression
P = 1/(1+exp(-L))
where L is the logit computed using the intercept mean plus time score times the slope mean.