Predict from a count regression model

Generic prediction method for the model classes produced by poissonGLM(), negbinGLM(), quasiPoissonGLM(), tweedieGLM(), zeroinflPoissonGLM(), zeroinflNegbinGLM(), zeroinflTweedieGLM(), and countGLM(). The call is forwarded to the backend's own predict() method (stats::predict.glm, glmmTMB:::predict.glmmTMB, or pscl:::predict.zeroinfl) so the full set of supported type values is available. Defaults to type = "response".

Usage

# S3 method for class 'countGLMfit'
predict(object, newdata = NULL, type = "response", ...)

# S3 method for class 'zeroinflGLMfit'
predict(object, newdata = NULL, type = "response", ...)

# S3 method for class 'countGLM'
predict(object, newdata = NULL, type = "response", ...)

Arguments

object

A fitted model returned by any of the package fitters, or a countGLM comparison object.

newdata

Optional data frame to predict on. When omitted, predictions on the training data are returned.

type

Backend-dependent prediction type. Common values: "response" (default; mean on the response scale), "link" (linear predictor), and for zero-inflated models "count", "zero", or "prob". See ?stats::predict.glm, ?glmmTMB::predict.glmmTMB, and ?pscl::predict.zeroinfl for the full list per backend.

...

Additional arguments forwarded to the backend predict method.

Value

A numeric vector (or matrix, for type = "prob") of predictions.

Details

For a countGLM() object the prediction comes from whichever model was selected by decide.

Examples

df <- data.frame(
  y  = c(0L, 1L, 2L, 3L, 5L, 0L, 2L, 4L, 1L, 3L),
  x1 = c(1.2, -0.4, 0.8, -1.1, 2.0, 0.3, -0.9, 1.5, -0.2, 0.7)
)
fit <- poissonGLM(y ~ x1, data = df)
#> Warning: Count component: 8 events (y > 0) for 1 predictor(s) (8.0 per predictor). At least 10 events per predictor is recommended.
predict(fit, newdata = data.frame(x1 = c(0, 1, 2)))
#>        1        2        3 
#> 1.798861 2.409767 3.228141