Responder analysis of pre-to-post change on the Rasch interval scale

Classifies each person's change between two occasions using interval (logit) measures from raschAnalyze() / PhysioAppKit::pcm_measure(). The Rasch model's own standard errors give a distribution-free standard error of measurement and hence the Minimal Detectable Change; the pre/post change is then run through mdcResponder() (reliable change) and, when an MCID on the logit scale is supplied, classifyResponder() (the MDC x MCID framework).

Usage

raschResponder(
  pre,
  post,
  mcid = NULL,
  confidence = 0.95,
  direction = c("increase", "decrease")
)

Arguments

pre, post: Baseline and follow-up measures: a poly_rasch fit, a list with theta/theta_se, or a numeric measure vector. se is required (from the fit) to derive the MDC.
mcid: Optional MCID on the logit scale; enables the four-level classifyResponder() output.
confidence: Confidence level for the MDC (default 0.95).
direction: "increase" (default; higher measure = better) or "decrease".

Value

a data.frame, one row per person: pre, post, change, improvement (direction-aware), sem, mdc, reliable_change (from mdcResponder()) and, if mcid given, classification (from classifyResponder()). Attributes sem, mdc, mcid record the values used.

Details

The two occasions must be on a common metric – calibrate the items once and anchor them, or co-calibrate – and the persons must align by row.

Examples

pre  <- list(theta = c(0, 0, 0, 0), theta_se = rep(0.3, 4))
post <- list(theta = c(1.5, 0.1, -1.2, 0.05), theta_se = rep(0.3, 4))
raschResponder(pre, post, mcid = 1.0)
#>   pre  post change improvement sem       mdc reliable_change classification
#> 1   0  1.50   1.50        1.50 0.3 0.8315423        improved true_responder
#> 2   0  0.10   0.10        0.10 0.3 0.8315423          stable  non_responder
#> 3   0 -1.20  -1.20       -1.20 0.3 0.8315423        declined  non_responder
#> 4   0  0.05   0.05        0.05 0.3 0.8315423          stable  non_responder