
Confidence intervals for misskappa coefficients
Source:R/estimate_object.R
confint.misskappa_estimate.RdWald 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.
Arguments
- object
A
misskappa_estimateobject.- 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.
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