predict.spdur {spduration} | R Documentation |
Predict methods for spdur Objects
Description
predict
and related methods for class “spdur
”.
Usage
## S3 method for class 'spdur'
predict(
object,
newdata = NULL,
type = "response",
truncate = TRUE,
na.action = na.exclude,
...
)
## S3 method for class 'spdur'
fitted(object, ...)
## S3 method for class 'spdur'
residuals(object, type = c("response"), ...)
Arguments
object |
Object of class “ |
newdata |
Optional data for which to calculate fitted values, defaults to training data. |
type |
Quantity of interest to calculate. Default conditional hazard,
i.e. conditioned on observed survival up to time |
truncate |
For conditional hazard, truncate values greater than 1. |
na.action |
Function determining what should be done with missing values
in newdata. The default is to predict NA ( |
... |
not used, for compatibility with generic function. |
Details
Calculates various types of probabilities, where “conditional” is used in
reference to conditioning on the observed survival time of a spell up to
time t
, in addition to conditioning on any variables included in the
model (which is always done). Valid values for the type
option
include:
“conditional risk”:
Pr(Cure=0|Z\gamma, T>t)
“conditional cure”:
Pr(Cure=1|Z\gamma, T>t)
“hazard”:
Pr(T=t|T>t, C=0, X\beta) * Pr(Cure=0|Z\gamma)
“failure”:
Pr(T=t|T>t-1, C=0, X\beta) * Pr(Cure=0|Z\gamma)
“unconditional risk”:
Pr(Cure=0|Z\gamma)
“unconditional cure”:
Pr(Cure=1|Z\gamma)
“conditional hazard” or “response”:
Pr(T=t|T>t, C=0, X\beta) * Pr(Cure=0|Z\gamma, T>t)
“conditional failure”:
Pr(T=t|T>t-1, C=0, X\beta) * Pr(Cure=0|Z\gamma, T>t)
The vector Z\gamma
indicates the cure/at risk equation
covariate vector, while X\beta
indicates the duration equation
covariate vector.
Value
Returns a data frame with 1 column corresponding to type
, in the same
order as the data frame used to estimate object
.
Note
See forecast.spdur
for producing forecasts when future
covariate values are unknown.
Examples
# get model estimates
data(model.coups)
ch <- predict(model.coups)
head(fitted(model.coups))
head(residuals(model.coups))