ROC.MRC: ROC.MRC - ROC areas for matched control data
In BushmanLab/hotROCs: Interactive Heatmaps for Areas under ROC Curves

Description Usage Arguments Details Value Author(s) Examples

View source: R/ROC.MRC.R

ROC curve areas, variances, and p-values for datasets in which each case has a collection of matching controls.

1 2	ROC.MRC(response, stratum, variables, origin = NULL, origin.levels = NULL, ragged.OK = TRUE)

`response`	logical vector identifying cases or character vector or factor vector with “insertion” marking the cases. Alternatively, a factor or character vector with elements matching `"insertion"` denoting the cases.
`stratum`	a vector of `length(response)` elements whose unique values correspond to cases and their matched controls. There must be exactly one case and `k` controls in every stratum.
`variables`	numeric a `matrix` or `data.frame` with `length(response)` rows and two or more columns.
`origin`	`NULL` or a vector of `length(response)` elements to identify different data sources.
`origin.levels`	optional character vector of origin levels
`ragged.OK`	logical - `TRUE`, if differing numbers of MRCs of any one origin are acceptable.

When data are collected under a scheme in which matched controls are used to allow for biased selection and every case has k controls, the ROC area is estimated as the average fraction of controls whose variable value is less than that of its reference case. Variances are computed from the usual sample variance divided by the number of cases. It is the responsiibility of the caller to remove NA values from the arguments.

list with elements ‘ROC’ giving a matrix of ROC curve areas, ‘var’ giving a list of variance matrices, and ‘pvalues’ giving a list of matrices containing pvalues for various contrasts.

Charles Berry

case <- rep(rep(c(TRUE,FALSE),c(1,3)),1000)
group <- rep(1:1000,each=4)
var1 <- case + rnorm(4000)
var2 <- case + rexp(4000)
var3 <- case + runif(4000,-1,1)
origin <- rep(factor(letters[1:4]),each=1000)
ROC.MRC(case,group,cbind(var1,var2,var3),origin)