This chapter considers analysis of longitudinal data from two methods. It extends the methodology for linked repeated measurements data in Chapter 5 to allow systematic effects of time on the measurements. The within-subject errors of the methods may be correlated over time. Time may be treated as a discrete or continuous quantity. Its effect is captured primarily by letting the means of the methods depend on it. This in turn implies that the measures of similarity and agreement also depend on time. The methodology provides confidence bands for simultaneous evaluation of similarity and agreement over time. It does not require the measurements from both methods to be always observed together. It, however, assumes a common scale for the methods. A case study illustrates the methodology.


Longitudinal data typically arise in method comparison studies when a cohort of subjects is followed over a period of time and the true value for a subject changes over time. Suppose there are n subjects in the study, labeled as i = 1, ..., n. The two methods being compared are labeled as j = 1, 2. Suppose also that the measurements are to be taken at prespecified time points 1, ..., . We think of these time points as the values of a discrete quantity called measurement occasion. The study design usually warrants that, for every subject, paired ...

Get Measuring Agreement now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.