predict.secr {secr}

SECR Model Predictions

Description

Evaluate a spatially explicit capture–recapture model. That is, compute the ‘real’ parameters corresponding to the ‘beta’ parameters of a fitted model for arbitrary levels of any variables in the linear predictor.

Usage

## S3 method for class 'secr'
predict(object, newdata = NULL, realnames = NULL, type = c("response", "link"), 
    se.fit = TRUE, alpha = 0.05, savenew = FALSE, ...)

## S3 method for class 'secrlist'
predict(object, newdata = NULL, realnames = NULL, type = c("response", "link"),
    se.fit = TRUE, alpha = 0.05, savenew = FALSE, ...)

## S3 method for class 'secr'
detectpar(object, ..., byclass = FALSE) 

Arguments

object

secr object output from secr.fit, or list of secr objects (secrlist)

newdata

optional dataframe of values at which to evaluate model

realnames

character vector of real parameter names

type

character; type of prediction required. The default ("response") provides estimates of the ‘real’ parameters.

se.fit

logical for whether output should include SE and confidence intervals

alpha

alpha level for confidence intervals

savenew

logical for whether newdata should be saved

...

other arguments passed to makeNewData

byclass

logical; if TRUE values are returned for each latent class in a mixture model, or class in a hybrid mixture (hcov) model

Details

The variables in the various linear predictors are described in secr-models.pdf and listed for the particular model in the vars component of object.

Optional newdata should be a dataframe with a column for each of the variables in the model (see ‘vars’ component of object). If newdata is missing then a dataframe is constructed automatically.

Default newdata are for a naive animal on the first occasion; numeric covariates are set to zero and factor covariates to their base (first) level. From secr 3.1.4 the argument ‘all.levels’ may be passed to makeNewData; if TRUE then the default newdata includes all factor levels.

realnames may be used to select a subset of parameters.

Standard errors for parameters on the response (real) scale are by the delta method (Lebreton et al. 1992), and confidence intervals are backtransformed from the link scale.

The value of newdata is optionally saved as an attribute.

detectpar is used to extract the detection parameter estimates from a simple model to pass to functions such as esa.plot. detectpar calls predict.secr. Parameters will be evaluated by default at base levels of the covariates, although this may be overcome by passing a one-line newdata to predict via the ... argument. Groups and mixtures are a headache for detectpar: it merely returns the estimated detection parameters of the first group or mixture.

If the ‘a0’ parameterization has been used in secr.fit (i.e., object$details$param == 3) then detectpar automatically backtransforms (a0, sigma) to (g0, sigma) or (lambda0, sigma) depending on the value of object$detectfn.

Value

When se.fit = FALSE, a dataframe identical to newdata except for the addition of one column for each ‘real’ parameter. Otherwise, a list with one component for each row in newdata. Each component is a dataframe with one row for each ‘real’ parameter (density, g0, sigma, b) and columns as below

link	link function
estimate	estimate of real parameter
SE.estimate	standard error of the estimate
lcl	lower 100(1--alpha)% confidence limit
ucl	upper 100(1--alpha)% confidence limit

When newdata has only one row, the structure of the list is ‘dissolved’ and the return value is one data frame.

For detectpar, a list with the estimated values of detection parameters (e.g., g0 and sigma if detectfn = "halfnormal"). In the case of multi-session data the result is a list of lists (one list per session).

Note

predictDsurface should be used for predicting density at many points from a model with spatial variation. This deals automatically with scaling of x- and y-coordinates, and is much is faster than predict.secr. The resulting Dsurface object has its own plot method.

The argument ‘scaled’ was removed from both predict methods in version 2.10 as the scaleg0 and scalesigma features had been superceded by other parameterisations.

Overdispersion results in confidence intervals that are too narrow. See adjustVarD for a partial solution.

References

Lebreton, J.-D., Burnham, K. P., Clobert, J. and Anderson, D. R. (1992) Modeling survival and testing biological hypotheses using marked animals: a unified approach with case studies. Ecological Monographs 62, 67–118.

See Also

secr.fit, predictDsurface, adjustVarD, makeNewData

Examples

## load previously fitted secr model with trap response
## and extract estimates of `real' parameters for both
## naive (b = 0) and previously captured (b = 1) animals

predict (secrdemo.b, newdata = data.frame(b = 0:1))

## OR from secr 3.1.4 
predict (secrdemo.b, all.levels = TRUE)

temp <- predict (secrdemo.b, all.levels = TRUE, save = TRUE)
attr(temp, "newdata")

detectpar(secrdemo.0)