Outcome Models
In CausalGPS: Matching on Generalized Propensity Scores with Continuous Exposures

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

After the pseudo population dataset was generated, we apply outcome models on the pseudo population as-if the dataset is from a randomized experiment.

We propose three types of outcome models using parametric, semi-parametric and non-parametric approaches, respectively.

estimate_pmetric_erf estimates the hazard ratios using a parametric regression model. By default, call gnm library to implement generalized nonlinear models.

estimate_semipmetric_erf estimates the smoothed exposure-response function using a generalized additive model with splines. By default, call gam library to implement generalized additive models.

estimate_npmetric_erf estimates the smoothed exposure-response function using a kernel smoothing approach. By default, call KernSmooth library to implement local polynomial fitting with a kernel weight. We use a data-driven bandwidth selection.