RMSEP {analogue}R Documentation

Root mean square error of prediction


Calculates or extracts the RMSEP from transfer function models.


RMSEP(object, ...)

## S3 method for class 'mat'
RMSEP(object, k, weighted = FALSE,

## S3 method for class 'bootstrap.mat'
RMSEP(object, type = c("birks1990", "standard"),

## S3 method for class 'bootstrap.wa'
RMSEP(object, type = c("birks1990", "standard"),



An R object.


numeric; the number of analogues to use in calculating the RMSEP. May be missing. If missing, k is extracted from the model using getK.


logical; Return the RMSEP for the weighted or unweighted model? The default is for an unweighted model.


The type of RMSEP to return/calculate. See Details, below.


Arguments passed to other methods.


There are two forms of RMSEP in common usage. Within palaeoecology, the RMSEP of Birks et al. (1990) is most familiar:

\mathrm{RMSEP} = \sqrt{s_1^2 + s_2^2}

where where s_1 is the standard deviation of the out-of-bag (OOB) residuals and s_2 is the mean bias or the mean of the OOB residuals.

In the wider statistical literature, the following form of RMSEP is more commonly used:

\mathrm{RMSEP} = \sqrt{\frac{\sum_{i=1}^n (y_i - \hat{y}_i)^2}{n}}

where y_i are the observed values and \hat{y}_i the transfer function predictions/fitted values.

The first form of RMSEP is returned by default or if type = "birks1990" is supplied. The latter form is returned if type = "standard" is supplied.

The RMSEP for objects of class "mat" is a leave-one-out cross-validated RMSEP, and is calculated as for type = "standard".


A numeric vector of length 1 that is the RMSEP of object.


Gavin L. Simpson


Birks, H.J.B., Line, J.M., Juggins, S., Stevenson, A.C. and ter Braak, C.J.F. (1990). Diatoms and pH reconstruction. Philosophical Transactions of the Royal Society of London; Series B, 327; 263–278.

See Also

mat, bootstrap, wa, bootstrap.wa.


## Imbrie and Kipp example
## load the example data

## merge training and test set on columns
dat <- join(ImbrieKipp, V12.122, verbose = TRUE)

## extract the merged data sets and convert to proportions
ImbrieKipp <- dat[[1]] / 100
V12.122 <- dat[[2]] / 100

## fit the MAT model using the squared chord distance measure
(ik.mat <- mat(ImbrieKipp, SumSST, method = "chord"))

## Leave-one-out RMSEP for the MAT model

## bootstrap training set
(ik.boot <- bootstrap(ik.mat, n.boot = 100))

## extract the Birks et al (1990) RMSEP

## Calculate the alternative formulation
RMSEP(ik.boot, type = "standard")

[Package analogue version 0.17-6 Index]