Companion to Functional Regression with Intensively Measured Longitudinal Outcomes: A New Lens through Data Partitioning
This contains both general code to implement the method in a sequential method and code to mimic the simulations included in the original paper. Functions require data to have labels following the below mapping:
| Input | Label |
|---|---|
| Response | Y |
X corresponding to &beta(t) |
x1,...,xq |
Z corresponding to &eta |
z1,...,zp |
| Time | time |
| Id |
Sourcing of files assumes that the working directory is the main folder.
File structure
- Rsource all code to fit method described in paper
- combine.R computes matrices and vectors for the one-step estimator using output from distribute.R
- distribute.R implements QIF on data by block
- functions.R includes functions related to making the design matrix for basis functions and derivatives of basius functions
- gcv.R wraps combine and distribute functions with implementation of generalized cross validation statistic
- matrix_inverse.cpp allows implementation of matrix inverse in C++ rather than R
- qif.cpp implementation of quadratic inference functions to return necessary summary statistics for combine step
- Examples code to implement and run examples similar those included in the paper
- nhanes contains two scripts to convert data from Leroux's package to usable format for SCM estimator, and implementation of SCM estimator
- Rsource_pqif _our implementation of penalized quadratic inference functions following the method of Qu and Li (2006)
- gcvpqif.R wraps the penalized quadratic influence function code to estimate lambda
- pqif.cpp our implementation in C++ of the penalized quadratic influence functions
- datasets.R used to generate data sets as found in paper
- psim1.R run a single iteration of the broken stick simulation
- psim2_single_onebeta.R run a single iteration of the second simulation with known gammas
- psim3_single.R run a single iteration of the third simulation with known functional form
- psimPoissonSingle.R run a single iteration of the simulation with Poisson link function