cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Choose Language Hide Translation Bar
Fitting Repeated Measures Data with JMP® Pro 13: Bigger and Better

 Don McCormack, Technical Enablement Engineer, JMP

 

Accurate characterization of repeated measures mixed models depends on assumptions about the correlation of observations across measurement periods. Adjacent time periods are likely more similar than distant time periods, but how is that similarity captured? The correlation structure of a repeated measures model answers that question. From the perspective of model implementation, the more structures available, the better the chance of approximating the way data looks in reality. Prior to JMP Pro 13 only three commonly used time based repeated measures covariance structures were available: unstructured, autoregressive (with a period of 1), and residual. This made characterization challenging, leaving a large number of potential solutions unavailable to the modeler. JMP Pro 13 introduces seven new structures, greatly expanding the modeling opportunities. This talk will discuss modeling repeated measures data using any of the existing covariance structures including those traditionally reserved for spatial relationships. It will provide examples illustrating their relationship to each other, when to use which structure, and how to compare them to find the best fit. A brief overview of repeated measures mixed models will be given, as well as the SAS code corresponding to the examples.

 

Comments

Great overview of repeated measures analysis in JMP Pro

Excellent overiew of mixed models in JMP Pro. Thank you! I have two questions:

 

1. When baseline is used as a covariate - why is the time point 0 (baseline) excluded from the fixed factor "time period" (1, 2, 3, 4, 5, 6, 7, 8 hrs). Could time 0 hr be included in the fixed factor "time period" or is this wrong?

2. Should the covariate be tested for interactions befor inclusion in the model? If yes, how is this done?

1. You couldn't include time 0 as both a covariate and a time 0 observation for the example dataset. They are completely correlated (by the virtue that they are the same things). You get to do one or the other, but not both. If you had an independent measure of baseline FEV then both a covariate and time 0 point could be included in the model. In that case, though, you'd have two different sets of data.

2. You could. Add it to the model as an interaction (e.g., Baseline*Time Period)

Thank you for the explanation!