1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
|
% Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mis-link.R
\name{mis}
\alias{mis}
\title{A \code{\link[=make.link]{"link-glm"}} object for misclassified responses in binomial regression models}
\usage{
mis(link = "logit", sensitivity = 1, specificity = 1)
}
\arguments{
\item{link}{the baseline link to be used.}
\item{sensitivity}{the probability of observing a success given that a success actually took place given any covariate values.}
\item{specificity}{the probability of observing a failure given that a failure actually took place given any covariate values.}
}
\description{
\code{\link[=mis]{mis()}} is a \code{\link[=make.link]{"link-glm"}} object that specifies the link function in Neuhaus (1999, expression (8)) for handling misclassified responses in binomial regression models using maximum likelihood. A prior specification of the sensitivity and specificity is required.
}
\details{
\code{sensitivity + specificity} should be greater or equal to 1,
otherwise it is implied that the procedure producing the responses
performs worse than chance in terms of misclassification.
}
\examples{
## Define a few links with some misclassification
logit_mis <- mis(link = "logit", sensitivity = 0.9, specificity = 0.9)
lizards_f <- cbind(grahami, opalinus) ~ height + diameter + light + time
lizardsML <- glm(lizards_f, family = binomial(logit), data = lizards)
lizardsML_mis <- update(lizardsML, family = binomial(logit_mis),
start = coef(lizardsML))
## A notable change is coefficients is noted here compared to when
## specificity and sensitity are 1
coef(lizardsML)
coef(lizardsML_mis)
## Bias reduction is also possible
update(lizardsML_mis, method = "brglmFit", type = "AS_mean",
start = coef(lizardsML))
update(lizardsML_mis, method = "brglmFit", type = "AS_median",
start = coef(lizardsML))
}
\references{
Neuhaus J M (1999). Bias and efficiency loss due to misclassified
responses in binary regression. Biometrika, \strong{86}, 843-855.
\url{https://www.jstor.org/stable/2673589}.
}
\seealso{
\code{\link[=glm]{glm()}}, \code{\link[=brglm_fit]{brglm_fit()}}
}
|
