predict.opsr: Predict Method for OPSR Model Fits

Description

Obtains predictions for the selection process (probabilities), the outcome process, or returns the inverse mills ratio. Handles also log-transformed outcomes.

Usage

# S3 method for opsr
predict(
  object,
  newdata,
  group,
  counterfact = NULL,
  type = c("response", "unlog-response", "prob", "mills", "correction", "Xb"),
  delta = 1,
  ...
)

Value

a vector of length nrow(newdata) (or data used during estimation).

Arguments

object: an object of class "opsr".
newdata: an optional data frame in which to look for variables used in object$formula. See also model.matrix.opsr. Ensure that at least one selection outcome matches the 'group' argument (see 'Examples').
group: predict outcome of this group (regime).
counterfact: counterfactual group.
type: type of prediction. Can be abbreviated. See 'Details' section for more information.
delta: constant that was added during the continuity correction ($log(Y_j + \delta)$). Only applies for type = "unlog-response".
...: further arguments passed to or from other methods.

Details

Elements are NA_real_ if the group does not correspond to the observed regime (selection outcome). This ensures consistent output length.

If the type argument is "response" then the continuous outcome is predicted. Use "unlog-response" if the outcome response was log-transformed (i.e., either in the formula specification or during data pre-processing). "prob" returns the probability vector of belonging to group, "mills" returns the inverse mills ratio, "correction" the heckman correction (i.e., $\rho_j * \sigma_j * \text{mills}$) and "Xb" returns $X \beta$.

Examples

Run this code

set.seed(123)
sim_dat <- opsr_simulate()
dat <- sim_dat$data
model <- ys | yo ~ xs1 + xs2 | xo1 + xo2
fit <- opsr(model, dat)
p <- predict(fit, group = 1, type = "response")

fit_log <- update(fit, . | log(yo) ~ .)
p_unlog <- predict(fit, group = 1, type = "unlog-response")

## newdata
newdat <- dat[1:4, ]
unique(newdat$ys)  # available selection outcomes
predict(fit, newdata = newdat, group = 2)