Bland-Altman Agreement Data

Computes the quantities required for a Bland-Altman agreement analysis.

Usage

blandAltmanData(x, ...)

# Default S3 method
blandAltmanData(x, y, conf.level = 0.95, na.rm = FALSE, ...)

# S3 method for class 'formula'
blandAltmanData(x, data = NULL, conf.level = 0.95, na.rm = FALSE, ...)

Arguments

x

numeric vector or formula

...

further arguments passed to or from other methods

y

numeric vector

conf.level

confidence level for the intervals reported for the bias and the limits of agreement

na.rm

logical; if TRUE, incomplete observation pairs are removed before computation

data

optional data frame used with the formula interface

Value

an object of class "BlandAltman" with components:

mean

pairwise means

diff

pairwise differences (y - x)

bias

mean difference

loaLower

lower limit of agreement

loaUpper

upper limit of agreement

biasCI

confidence interval for the bias

loaLowerCI

confidence interval for the lower LoA

loaUpperCI

confidence interval for the upper LoA

nObs

number of observations used

conf.level

the confidence level used

Details

For each observation pair, the arithmetic mean and the difference (y - x) are calculated. The function further computes the mean difference (bias), limits of agreement (LoA), and approximate confidence intervals for the bias and LoA according to Bland and Altman.

The returned object is of class "BlandAltman" and can be plotted using by a plot() routine residing in the pharos package.

The limits of agreement are the conventional bias +/- 1.96 * sd, independent of conf.level: the multiplier fixes the nominal coverage of the interval of differences, whereas conf.level governs the confidence intervals reported for the bias and for the two limits. The standard error of a limit uses the approximation \(\sqrt{3 s_d^2 / n}\) given by Bland and Altman.

References

Bland JM, Altman DG (1986). Statistical methods for assessing agreement between two methods of clinical measurement. Lancet, 327, 307-310.

See also

Other agreement: normalizeToConfusion(), raterFrame()

Examples

set.seed(1)
x <- rnorm(100)
y <- x + rnorm(100, sd = 0.5)

ba <- blandAltmanData(x, y)
ba
#> 
#> Bland-Altman Agreement Analysis
#> 
#> Bias      : -0.019 (-0.114, 0.076)
#> Lower LoA : -0.958 (-1.122, -0.793)
#> Upper LoA : 0.920 (0.755, 1.084)
#> 
#> n = 100, conf.level = 0.95