A methodology to perform multivariate measurement error adjustment using external validation data. Allows users to remove the attenuating effect of measurement error by incorporating a distribution of external validation data, and allows for plotting of all resultant adjustments. Sensitivity analyses can also be run through this package to test how different ranges of validity coefficients can impact the effect of the measurement error adjustment. The methods implemented in this package are based on the work by Muoka, A., Agogo, G., Ngesa, O., Mwambi, H. (2020):
Adjust for Measurement Error in Multiple Exposures from External Validation
How do you adjust for multi-variate measurement error in the absence of
an internal validation study? With rACME-MEEV of course! Based off of
the work of Muoka et al., 2020 (1), this library takes the proposed
methodology and exposes it as a clean R package.
You can install the development version of rACME-MEEV from GitHub with:
# install.packages("pak")
pak::pak("westford14/rACMEMEEV")
This is a basic example running through a standard workflow of adjusting survey data in the absenece of a validation study.
library(rACMEMEEV)
# Read the data
data <- read_csv("path/to/your/data")
columns <- c("fruit", "vegetable", "tobacco")
# Generate the coefficients
fruit_v_coef <- generate_coefficient(1000, fruit_low, fruit_upper, 0.95)
veg_v_coef <- generate_coefficient(1000, veg_low, veg_upper, 0.95)
tob_v_coef <- generate_coefficient(1000, tobacco_low, tobacco_upper, 0.95)
# Compute the pre-model
output <- acme_model(data, columns)
# Solve the attenuation-contamination matrix
validity_coefficients <- c(fruit_v_coef, veg_v_coef, tob_v_coef)
lambda <- attenuation_matrix(output, columns, validity_coefficients)
# Fit the multivariate model
model_output <- multivariate_model(
"target ~ fruit + veg + tobacco",
data = data,
columns = c("fruit", "veg", "tobacco"),
jags_model = output$model,
a_c_matrix = lambda$matrix,
sds = lambda$sds,
variances =lambda$variances,
univariate = TRUE
)
# Plot covariate adjustments
plots <- plot_covariates(model_output, columns)
# Create Traceplots Etc.
acf_plots(model_output$naive, c("fruit", "veg", "tobacco"))
traceplots(output$samples, c("fruit", "veg", "tobacco"))
acf_plots(model_output$naive, c("fruit", "veg", "tobacco"), pre_model = True)
traceplots(output$samples, c("fruit", "veg", "tobacco"), pre_model = True)
The model has been most thoroughly validated with the JAGS backend, but there is experimental support for the Stan backend. Please do note that it is experimental and liable to change. Here is an example of the same above workflow using the Stan backend.
library(rACMEMEEV)
# Read the data
data <- read_csv("path/to/your/data")
columns <- c("fruit", "vegetable", "tobacco")
# Generate the coefficients
fruit_v_coef <- generate_coefficient(1000, fruit_low, fruit_upper, 0.95)
veg_v_coef <- generate_coefficient(1000, veg_low, veg_upper, 0.95)
tob_v_coef <- generate_coefficient(1000, tobacco_low, tobacco_upper, 0.95)
# Compute the pre-model
output <- acme_model(data, columns, stan = TRUE)
# Solve the attenuation-contamination matrix
validity_coefficients <- c(fruit_v_coef, veg_v_coef, tob_v_coef)
lambda <- attenuation_matrix(output, columns, validity_coefficients, stan = TRUE)
# Fit the multivariate model
model_output <- multivariate_model(
"target ~ fruit + veg + tobacco",
data = data,
columns = c("fruit", "veg", "tobacco"),
jags_model = output$model,
a_c_matrix = lambda$matrix,
sds = lambda$sds,
variances =lambda$variances,
univariate = TRUE
)
# Plot covariate adjustments
plots <- plot_covariates(model_output, columns)
# Create Traceplots Etc.
acf_plots(model_output$naive, c("fruit", "veg", "tobacco"), pre_model = TRUE, stan = TRUE)
traceplots(output$samples, c("fruit", "veg", "tobacco"), pre_model = TRUE, stan = TRUE)
acf_plots(model_output$naive, c("fruit", "veg", "tobacco"), stan = TRUE)
traceplots(output$samples, c("fruit", "veg", "tobacco"), stan = TRUE)
If you would like continue development of this library or open up feature requests, you can fork the repo then create a descriptive branch name while continuing devleopment.
git checkout -b your/descriptive/branch/name
Once development is complete, please open a pull request making sure to note the fields provided in the pull request template.
There are a couple of key commands that you should keep in mind when
developing this library. First, make sure that devtools and knitr
are installed into your R workspace. Then typically the workflow will
follow as:
devtools::check()devtools::install()devtools::test()devtools::document()knitr::build_readme()make lint from the command line in the root of the
repositorySometimes R will have trouble reading the .db that has all of the
namespace exports and will throw an error. So, the easiest way to
rectify this is to uninstall and then reinstall the package.
detach("package:rACMEMEEV", unload=TRUE)devtools::install()library(rACMEMEEV)If you are building this package from source there are a few platform
specific issues that may arise. This package was developed principally
on R version 4.5.1 on an OSX operation system, so
RTools may need to be
installed on a Windows machine to get proper builds. If there are any
issues running devtools::check() that specifically cite RTools, you
can try to set the options(buildtools.check = function(action) TRUE )
option so that it skips the explicit local build. This is not
recommended, but is a way to speed up the development loop if there are
build issues arising.
Please note that the rACMEMEEV project is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.