predict.RdThese methods return predicted values from unmarkedFit objects, fitLists,
or ranef output. Most object types use the same method, but some have
unique options; see usage below.
# S4 method for class 'unmarkedFit'
predict(object, type, newdata, backTransform = TRUE,
na.rm = TRUE, appendData = FALSE, level = 0.95, re.form = NULL, ...)
# S4 method for class 'unmarkedFitOccuMulti'
predict(object, type, newdata,
level = 0.95, species = NULL, cond = NULL, nsims = 100, ...)
# S4 method for class 'unmarkedFitOccuMS'
predict(object, type, newdata,
level = 0.95, nsims = 100, ...)
# S4 method for class 'unmarkedFitList'
predict(object, type, newdata = NULL,
backTransform = TRUE, appendData = FALSE, level = 0.95)
# S4 method for class 'unmarkedRanef'
predict(object, func, nsims = 100, ...)A unmarkedFit, unmarkedFitList, or unmarkedRanef object.
The submodel to predict values for, such as state or det.
The available types depend on the model.
Optional; provide a data.frame of new covariate values to
predict with. If not supplied, the original data are used.
Logical. If TRUE, returned values are on the
original (e.g. probability, abundance) scale.
Logical. Should missing values be removed when predicting from original data?
Logical. Should covariate data used for prediction be appended to the output data frame?
The confidence interval to calculate. For example 0.95
results in lower and upper bounds for a 95% confidence interval. If set
to NULL, no SE or confidence intervals will be returned.
For unmarkedFit types that support random effects,
should the random effects be included in the prediction? If NULL,
they will be, if NA, they will not be.
Which species (integer or species name as a string) should
predictions be calculated for? For unmarkedFitOccuMulti only. If
multiple species are provided, the co-occurence probability is returned.
See occuMulti.
Which species (integer or species name as a string) should
predictions be calculated conditional on? If conditional on species presence,
supply just the species name; if species absence, put a minus sign in front
of the species name, e.g. "-coyote". See occuMulti.
Number of bootstrap simulations to use. Relevant only for some predict types.
A function to apply to bootstrapped unmarkedRanef samples;
see details.
Other arguments, currently ignored
For most methods, a data.frame with four columns: the
predicted values, standard errors, and lower and upper bounds. If
appendData = TRUE covariate data are also in the output. For
unmarkedRanef, an array is returned. See details.
The predict method for unmarkedRanef objects generates
the empirical Bayes posterior predictive distribution for functions of the
random variables (latent abundance or occurrence).
In addition to the output object from ranef, you must also supply a
custom function to argument func. The function must take as input a matrix
with dimensions M x T, where M is the number of sites and T is the number of
primary periods (T=1 for single-season models). The output of this function should
be a vector or matrix containing the derived parameters of interest.
You may also manually set the number of draws from the posterior predictive
distribution with argument nsims; the default is 100.
The output of predict will be a vector or array with one more dimension
than the output of the function supplied func, corresponding to the number
of draws requested nsims. For example, if func
outputs a scalar, the output of predict will be a vector with length
equal to nsims. If func outputs a 3x2 matrix, the output of
predict will be an array with dimensions 3x2xnsims.
See ranef for an example.
Alternatively, you can use the posteriorSamples function on the
ranef output object to obtain the full posterior predictive distribution.
This is useful if you are having trouble designing your custom function or if
you want to obtain multiple different derived parameters from the same posterior
predictive distribution.