Skip to contents

Wald or exact-delete-one jackknife confidence intervals for a misskappa_estimate. Wald intervals use the estimator covariance and a normal critical value. Jackknife intervals transform every complete-data leave-one-subject-out estimate directly, use the ordinary delete-one jackknife standard error, and use a Student t critical value with n - 1 degrees of freedom.

Besides the natural and Fisher scales, transform = "log_complement" uses the upper-bound-only transform -log(1 - theta), which stays defined for coefficients below -1. At an exact boundary (perfect agreement), a degenerate transformed jackknife is replaced by a one-sided hybrid: a Clopper–Pearson upper limit for the probability of any item disagreement is divided by a delete-one t lower limit for chance disagreement. When that fallback is unavailable the interval is returned as NA with a warning.

Usage

# S3 method for class 'misskappa_estimate'
confint(
  object,
  parm = NULL,
  level = 0.95,
  method = c("wald", "jackknife"),
  transform = c("none", "fisher", "log_complement"),
  ...
)

Arguments

object

A misskappa_estimate object.

parm

Optional subset of coefficients (names or indices), defaults to all.

level

Confidence level.

method

Either "wald" or "jackknife". See the section on jackknife availability.

transform

One of "none", "fisher", or "log_complement". Wald uses a delta-method standard error on the selected scale; jackknife transforms the exact delete-one estimates directly.

...

Unused, present for S3 generic conformance.

Value

A two-column numeric matrix of lower and upper limits, one row per coefficient.

Jackknife availability

The exact delete-one summaries behind method = "jackknife" are attached automatically to fits with at least three subjects in these cases:

  • agreement() on complete scalar ratings: "ipw" and "cat_fiml" (g = 2), and "pairwise".

  • agreement_counts(): "pairwise" for any row totals, and "cat_fiml" when every row has the same total.

Fits on incomplete data, g > 2, vector-valued ratings, the compound-symmetry estimators, and categorical FIML with flatten > 0 support only method = "wald".

References

Efron, B., and Tibshirani, R. J. (1993). An Introduction to the Bootstrap. Chapman and Hall, Chapter 11.

Clopper, C. J., and Pearson, E. S. (1934). The use of confidence or fiducial limits illustrated in the case of the binomial. Biometrika, 26(4), 404–413. doi:10.1093/biomet/26.4.404 .

Examples

fit <- agreement(dat.gwet2014, estimator = "ipw")
#> Warning: rater pair(s) rater4-rater5 co-observed by only one subject; the corresponding pairwise covariance is degenerate and the standard error unreliable.
confint(fit)                       # natural-scale Wald interval
#>                      2.5 %    97.5 %
#> Conger           0.2040066 0.6426036
#> Fleiss           0.1966366 0.6399174
#> Brennan-Prediger 0.2045079 0.6797335
#> attr(,"method")
#> [1] "wald"
#> attr(,"transform")
#> [1] "none"
#> attr(,"boundary_fallback")
#>           Conger           Fleiss Brennan-Prediger 
#>            FALSE            FALSE            FALSE 
confint(fit, transform = "fisher") # Fisher z interval, always within (-1, 1)
#>                      2.5 %    97.5 %
#> Conger           0.1824728 0.6162177
#> Fleiss           0.1751369 0.6133302
#> Brennan-Prediger 0.1776159 0.6470501
#> attr(,"method")
#> [1] "wald"
#> attr(,"transform")
#> [1] "fisher"
#> attr(,"boundary_fallback")
#>           Conger           Fleiss Brennan-Prediger 
#>            FALSE            FALSE            FALSE 
complete_fit <- agreement_counts(dat.fleiss1971)
confint(complete_fit, method = "jackknife", transform = "fisher")
#>                      2.5 %    97.5 %
#> Fleiss           0.3117814 0.5355365
#> Brennan-Prediger 0.3255388 0.5495116
#> attr(,"method")
#> [1] "jackknife"
#> attr(,"transform")
#> [1] "fisher"
#> attr(,"boundary_fallback")
#>           Fleiss Brennan-Prediger 
#>            FALSE            FALSE