Semiparametric Likelihood Estimation with Errors in Variables

Efficient regression analysis under general two-phase sampling, where Phase I includes error-prone data and Phase II contains validated data on a subset.

SLEEV: Semiparametric Likelihood Estimation with Errors in Variables

License: file /inst/LICENSE_sleev.txt

Markdown:

HTML:

reStructuredText: .. image:: https://joss.theoj.org/papers/10.21105/joss.07320/status.svg :target: https://doi.org/10.21105/joss.07320

The complete R package logreg2ph and code for the simulation settings included in this paper.

Part of the R package TwoPhaseReg as described in this paper.

If you wish to make suggestions, report a problem, or need more support with package use, please email Dr. Ran Tao (r DOT tao AT vumc DOT org) or Dr. Sarah Lotspeich (lotspes AT wfu DOT edu).

Below is a simple example for running a linear regression with sleev. The complete vignette detailing all uses of example in sleev can be found as an article.

Install

To install the package for GitHub, run the following in your R console:

  devtools::install_github("dragontaoran/sleev")

or install from CRAN:

install.packages("sleev")

Load package:

  library(sleev)

mock.vccc data

This data is a simulated two-phase sampling dataset based on VCCC data. To load this dataset, run:

  data(mock.vccc)

Preprocessing

Because of skewness, we often transform both CD4 and VL. In our analysis, CD4 was divided by 10 and square-root transformed and VL was log_10 transformed:

  mock.vccc$CD4_val_sq10 <- sqrt(mock.vccc$CD4_val/10)
  mock.vccc$CD4_unval_sq10 <- sqrt(mock.vccc$CD4_unval/10)
  mock.vccc$VL_val_l10 <- log10(mock.vccc$VL_val)
  mock.vccc$VL_unval_l10 <- log10(mock.vccc$VL_unval)

Load splines package

  library(splines)

Construct B-spline Basis

sn=20
data.linear <- spline2ph(x = "VL_unval_l10", data = mock.vccc, size = sn,
                         degree = 3,  group = "Sex")

Model fitting

The SMLEs can be obtained by running

res_linear <- linear2ph(y_unval = "CD4_unval_sq10", y = "CD4_val_sq10", 
                        x_unval = "VL_unval_l10",x = "VL_val_l10",z = "Sex", 
                        b_spline = paste0("bs", 1:sn), data = data.linear,
                        hn_scale = 1, se = TRUE, tol = 1e-04, 
                        max_iter = 1000, verbose = FALSE)

Check convergence:

c(res_linear$converge, res_linear$converge_cov)