Traj
uses a discrete mixture model to model longitudinal data. This model
allows for data grouping using different parameter values for each
group distribution. Groupings may identify distinct subpopulations.
Alternatively, groupings may represent components for
approximating an
unknown (possibly complex) data distribution.
Supported distributions are: censored (or
regular) normal, zero inflated (or regular) Poisson, and Bernoulli
distributions (logistic model). The censored normal model is useful for
psychometric scale data, the zero inflated Poisson model useful for count data
with more zeros than would be expected under the Poisson assumption, and the Bernoulli model useful for 0/1 data. The model is
appropriate for data with average values changing smoothly as a function of the
dependent variable (time, age, ...). Some sharp changes can be handled through
the inclusion of time dependent covariates.