Confidence Regions on ROC space

Share:

Description

Plot individual confidence regions for the estimate from each primary study on ROC space or add such regions to an existing plot.

Usage

1
2
3
4
5
## Default S3 method:
ROCellipse(x, correction = 0.5, level = 0.95, 
           xlim = c(0, 1), ylim = c(0, 1), method = "wilson", 
           pch = 1, add = FALSE, corr = 0, suppress = TRUE, 
           ellipsecol = "grey", ...)

Arguments

x

a data frame with variables including TP, FN, FP, TN, alternatively a matrix with column names including these.

correction

numeric, continuity correction applied to zero cells.

level

numeric, confidence level for the calculations of confidence intervals.

xlim

numeric of length 2, which portion of ROC space should be plotted? All reasonable values should be within (0,1).

ylim

numeric of length 2, which portion of ROC space should be plotted? All reasonable values should be within (0,1).

method

character, method used to calculate the confidence intervals for sensitivities, specificities and false positive rates. One of "wald", "wilson", "agresti-coull", "jeffreys", "modified wilson", "modified jeffreys", "clopper-pearson", "arcsine", "logit", "witting"

pch

Symbol used to plot point estimates. Use pch = "" to suppress plotting point estimates.

add

logical, should the plot be added to the current plot?

corr

numeric or character, the correlation assumed in the calculation of the confidence ellipsoids on logit-ROC space. If set to "logit", the correlation of the logit-transformed sensitivities and false positive rates will be used in the correlations. See details for further explanation.

suppress

logical, should the warnings produced by the internal call to madad be suppressed? Defaults to TRUE, since only the diagnostic accuracies and their confidence intervals are used in subsequent calculations.

ellipsecol

The color used for plotting the ellipses.

...

further arguments passed on to plot.

Details

The confindence regions are ellipses on logit-ROC space, hence the name of the function. The standard deviations underlying confidence intervals for the sensitivities and false positive rates are used to determine the scale of the ellipses on logit-ROC space. These ellipses get backtransformed to ROC space and plotted. As a default no correlation is assumed on logit-ROC space.

The objects of class reitsma have their own ROCellipse method to add a confidence region for the pooled estimate, see reitsma-class.

Value

Besides plotting an invisble NULL is returned.

Author(s)

Philipp Doebler <philipp.doebler@googlemail.com>

See Also

crosshair, reitsma-class

Examples

1
2