SL.glm.manning: Adaptive generalized linear model of Manning (2001)

In wuziyueemory/twostageSL: Two-Stage Super Learner Prediction Function

Description

This function implements the estimator of Manning and Mullahy (2001). The estimator adaptively selects a link function and family based on the skewness and kurtosis of the data. Some of the family/link functions recommended by Manning are numerically unstable and so this function returns Duan's smearing estimate if the GLM fitting step breaks down.

Usage

SL.glm.manning(
  Y,
  X,
  newX,
  family = gaussian(),
  obsWeights = rep(1, length(Y)),
  kCut = 3,
  lambdaCut = c(0.5, 1.5, 2.5),
  startNLS = 0,
  ...
)

Arguments

Y	A numeric outcome variable
X	A data.frame of covariates constituting the training sample
newX	A data.frame with the same column names and format as X constituting the validation sample.
family	Gaussian only
obsWeights	Observation-level weights (not currently used)
kCut	Cut point for kurtosis
lambdaCut	Cut points for skewness
startNLS	Starting values for the non-linear least squares
...	Other arguments (not currently used)

Value

pred

Predicted outcomes based on predictors in newX

fit

A list with named entries object (the fitted regression model object)

References

Manning WG, Mullahy J (2001). “Estimating log models: to transform or not to transform?” Journal of Health Economics, 20(4), 461–494.

Examples

# load cost data
data(cost_data)
# fit manning model
fit_glm.manning <- SL.glm.manning(Y = cost_data$totalcost, X = cost_data[, c("female", "race")],
                              newX = cost_data[, c("female", "race")])
# get back predictions
pred_glm.manning <- predict(fit_glm.manning$fit, newdata = cost_data[,c("female", "race")])

wuziyueemory/twostageSL documentation built on Oct. 19, 2020, 3:45 p.m.