ref_transition

Usage

ref_transition(dynamic, data, id, time)

Distinguishes the longitudinal quantities of the velocity design memo: observed velocity, expected velocity, velocity centile, innovation Z, and change Z.

Arguments

Argument	Description
dynamic	A ref_dynamics object.
data	Subject visits.
id	Subject identifier.
time	Time variable. Every transition is conditioned on the subject’s full prior history.

Details

• innovation_z is the history-conditioned score $(Z_2 - m)/s$; with one previous visit it equals the Z-gain $(Z_2 - rZ_1)/\sqrt{1-r^2}$.
• velocity_centile is the conditional centile of the end value given the history, i.e. pnorm(innovation_z) after the copula map.
• change_z is the unconditional standardised difference $(Z_2 - Z_1)/\sqrt{2(1-r)}$ and change_centile = pnorm(change_z).

When the process for an outcome is not identified (dynamic$processes[[outcome]]$identified is FALSE), every history-conditioned quantity is NA and support is "unidentified". Duplicate visit times within a subject give a warning and NA velocities for the zero-length transition.

inferential_status is "heldout" only when the rows differ from the marginal and dynamics training data and the process and measurement components are identified. It is otherwise "in_sample", "measurement_not_separated", or "unidentified". This is provenance, not a claim that calibration has been empirically established.

Value

A ref_transition tibble, one row per consecutive visit pair and outcome (zero rows, full column set, when no subject has two visits).