Default random-effects structure of a single-case data set

Returns the random-effects part of the default piecewise model, the same string the package adds to the fixed terms when no formula is supplied. Raw data (with a phase column but no coded columns) are first coded with process_scd_phases() for design.

Usage

default_random_effects(data, design = detect_design_type(data))

Arguments

data

Raw or coded single-case data.

design

Design passed to process_scd_phases() when data are not coded yet; defaults to detect_design_type().

Value

Character string such as "(1 | study) + (1 + level_AB | study:case)", or NULL when the data hold a single series.

Details

Level-2 units are the tiers of tier_vars(): the combination of study, cluster, case, and within-case nest (behavior, setting) columns, so a case identifier repeated in two studies is two units. With two or more tiers every level-change coefficient of the coded design (reversal steps, criterion steps, alternating-treatment dummies, or level_AB/level_BM/level_BG) varies across tiers; omitting a random slope that exists inflates the Type I error rate with few cases. Each outer grouping (study, study:cluster, study:cluster:case) with more levels than the grouping above it and fewer than the tiers gets a random intercept, so behaviors or settings form a level between measurements and participants and cases sit within clusters and studies. Study, cluster, and nest columns constant in the data are dropped; the case column is kept, so the tier term always names cases.

In phase designs the onset step is correlated with the tier intercept and later steps enter as uncorrelated (||) terms, a reduction of the full random-effects covariance matrix that keeps the number of correlations small when units are few. Alternating-treatment dummies have no onset, so all of them enter as uncorrelated terms next to an intercept-only tier term; no condition is singled out.

Examples

default_random_effects(simulate_scd_data(n_cases = 3))
#> [1] "(1 + level_AB | case)"