Estimates the variance of d' (and criterion) from summary SDT counts using
the Gourevitch & Galanter (1967) delta method, the Miller (1996)
exact-binomial method, or a parametric bootstrap.
Usage
sdt_summary_variance(
hits,
fas,
misses,
cr,
method = c("miller", "gourevitch", "bootstrap"),
correction = c("loglinear", "half", "none"),
nboot = 1000,
seed = NULL
)
Arguments
- hits
Number of hits.
- fas
Number of false alarms.
- misses
Number of misses.
- cr
Number of correct rejections.
- method
Variance method: "gourevitch", "miller", or "bootstrap".
- correction
Correction for extreme rates ("loglinear", "half", or "none").
- nboot
Number of bootstrap samples when method = "bootstrap".
- seed
Optional random seed for bootstrap variance.
Value
A list with variance estimates for zH, zF, d', and c.
Details
Miller enumeration is exact for a binomial distribution at its
supplied probability. Here that probability is estimated from the observed
counts. The resulting variance and downstream Wald test are plug-in
approximations, not exact finite-sample inference.
Examples
v <- sdt_summary_variance(hits = 70, fas = 20, misses = 30, cr = 80,
method = "miller")
v$var_dprime
#> [1] 0.03758609