2016-08-10-Follow-up V

[andkov]

Current results

for the MAP study

• model-results-1, model specified by Q matrix:
• model-results-2, model specified by Q matrix:

[andkov]

Discussion topics

1. age is always treated as time-dependent predictor. What does it mean when we have to enter it into the b.covariates = statement?
2. What the practical implication of playing with the h value in the elect() call? (grid parameter) (p.8 of the ELECT vignette)
3. Interpreting marginal LE. Where did 1990 come from in Fig. 4 of the ELECT vignette? (p.8)
4. What is the logic of specifying the ematrix for misclassification? Must always specify in pairs of counterpart transitions?
5. WHat is a good strategy for interpreting the provided output? What should we look at first? what is a tell-tell signs of model problems in the output?

[andkov]

Minutes

Judith

• problem with classificaiton

Ardo

• (misclassification) means that some individuals were classified to have dimenatia,

  but later observation proved this incorrect

• if age scale is "year", h=.5 means that a half-year is used as approximation (the size of the rectangle). The smaller value - the better, but it means that it will take to estimate. ( try a few values and see if results change)

Judith

- "how many people in your network"

Emiel

• data is ready
• but some data problems preventing estimation, will check in the next few days
• first models should be here soon
• covars: age
• no LE yet
• bottle neck: R fluency

Johan

• ran model with misclassification
• excluded subject that were responsible for illegal transition (8 subject?)
• models works after excluding these subjects
• social engagement is missing from the data
• covar: age + sex
• no LE yet

Annie

• social engagement
• covar: age + sex + ses
• no LE yet

How to ensure that we are doing the same thing?