In mcconvil/mase: Model-Assisted Survey Estimators

mase

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "README-"
)

Overview

mase contains a collection of model-assisted generalized regression estimators for finite population estimation of a total or mean from a single stage, unequal probability without replacement design. It also contains several variance estimators.

The available estimators are currently:

generalized regression: greg()
hotvitz-thompson: horvitzThompson()
post-stratification: postStrat()
elastic net generalized regression: gregElasticNet()
regression tree: gregTree()
modified generalized regression: modifiedGreg()
ratio estimator: ratioEstimator()
estimate of a ratio estimator: ratio()

The available variance estimation techniques are:

LinHB
LinHH
LinHTSRS
LinHT
bootstrapSRS

See mase/inst/REFERENCES.bib for sources related to each variance estimator.

Installation

Install the latest CRAN release with:

install.packages("mase")

You can also install the development version of mase from GitHub as follows:

# install.packages("pak")
pak::pkg_install("mcconvil/mase")

Usage

Horvitz-Thompson

Here's an example of fitting the Horvitz-Thompson estimator using Forestry data in Idaho. The data is publicly available and comes from the Forestry Inventory & Analysis (FIA) program.

library(mase)
library(dplyr)

data(IdahoSamp)
data(IdahoPop)

samp <- filter(IdahoSamp, COUNTYFIPS == 16055) 
pop <- filter(IdahoPop, COUNTYFIPS == 16055) 

horvitzThompson(y = samp$BA_TPA_ADJ,
                N = pop$npixels,
                var_est = TRUE,
                var_method = "LinHTSRS")

Linear Regression Estimator

We can also fit a linear regression estimator using that same data:

xsample <- select(samp, c(tcc, elev, ppt, tmean))

xpop <- select(pop, names(xsample))

greg_est <- greg(y = samp$BA_TPA_ADJ,
                 N = pop$npixels,
                 xsample = xsample,
                 xpop = xpop,
                 var_est = TRUE,
                 var_method = "LinHB",
                 datatype = "means")

We still get the population total and mean estimates along with their variance estimates:

greg_est[c('pop_total','pop_mean', 'pop_total_var', 'pop_mean_var')]

But with this estimator we also get the weights

greg_est["weights"]

and the coefficients for the model

greg_est["coefficients"]

Variable Selection

All of the mase regression estimators can also perform variable selection internally using the parameter modelselect

greg_select <- greg(y = samp$BA_TPA_ADJ,
                    N = pop$npixels,
                    xsample = xsample,
                    xpop = xpop,
                    modelselect = TRUE,
                    var_est = TRUE,
                    var_method = "LinHB",
                    datatype = "means")