Get quantiles from prediction intervals — glm

This helper function runs the trending::predict.trending_fit method on a fitted model at specified values of "alpha" in order to create a range of prediction intervals. The processing also includes steps to convert the alpha to corresponding quantile values at upper and lower bounds. See "Details" for more information on the translation of "alpha" to quantile values. This function is used internally in glm_forecast.

Usage

glm_quibble(
  fit,
  new_data,
  alpha = c(0.01, 0.025, seq(0.05, 0.45, by = 0.05)) * 2
)

Arguments

fit: Fitted model object from glm_fit; note must be accessed from first element in "fit" column
new_data: A tibble with new data on which the trending::predict.trending_fit method should run
alpha: Vector specifying the threshold(s) to be used for prediction intervals (PI); alpha of 0.05 would correspond to 95% PI; default is c(0.01, 0.025, seq(0.05, 0.45, by = 0.05)) * 2 to create a range of intervals

Value

A tibble with forecasted data including the following columns:

epiweek: The epidemiological week for the forecasted horizon
epiyear: The epidemiological year for the forecasted horizon
quantile: The quantile for the forecasted value; NA for point estimate
value: The forecasted value

Details

The "alpha" parameter defines the width of prediction interval (PI). For example, an alpha = 0.05 would correspond to a 95% PI. This function uses the PI(s) (per the alpha value(s) specified) to construct a range of quantiles that fall at lower and upper bound of each PI. Continuing from the example of alpha = 0.05, the quantile estimates returned would fall at 0.025 (lower bound of PI) and 0.975 (upper bound of PI).