mis: A '"link-glm"' object for misclassified responses in binomial...

In brglm2: Bias Reduction in Generalized Linear Models

A "link-glm" object for misclassified responses in binomial regression models

Description

mis() is a "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.

Usage

mis(link = "logit", sensitivity = 1, specificity = 1)

Arguments

link	the baseline link to be used.
sensitivity	the probability of observing a success given that a success actually took place given any covariate values.
specificity	the probability of observing a failure given that a failure actually took place given any covariate values.

Details

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.

References

Neuhaus J M (1999). Bias and efficiency loss due to misclassified responses in binary regression. Biometrika, 86, 843-855. https://www.jstor.org/stable/2673589.

See Also

glm(), brglm_fit()

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))

brglm2 documentation built on Oct. 12, 2023, 1:07 a.m.