Predicted Counts for a Log-Link Generalised Linear Model

Teaching helper for count predictions from a log-link generalised linear model. It wraps predict.glm, constructs confidence intervals on the link scale, exponentiates the fitted values and limits, rounds the result, and optionally prints the returned table.

Usage

predictCount(object, newdata, cilevel = 0.95, digit = 3, print.out = TRUE, ...)

Arguments

object: a glm object, i.e. the output from glm.
newdata: prediction data frame.
cilevel: confidence level for the intervals.
digit: number of decimal places to print.
print.out: if TRUE, print the prediction table.
...: optional arguments that are passed to predict.glm.

Value

Invisibly returns a data frame with three columns:

Predicted: the predicted count on the response scale.
Conf.lower: the lower confidence limit on the response scale.
Conf.upper: the upper confidence limit on the response scale.

Details

New work should use predict20x, which handles lm and glm fits with one function. predictCount() always exponentiates, so for a logit-link model it returns odds, as predict20x(type = "odds") does.

Note: newdata must be a data frame with the same column order and data types as those used in fitting the model. This stricter interface is kept for compatibility with the original teaching wrapper.

Note

Deprecated in s20x 3.4.0. Use predict20x instead, with type = "odds" for the odds from a logit-link model.