samplingNR is an R package that allows for computing optimal allocations under anticipated nonresponse. The underlying theory is provided in Mendelson & Elliott (in press).
Installation
You can install the latest development version of samplingNR from GitHub with:
# install.packages("devtools")
devtools::install_github("jmendelson256/samplingNR", build_vignettes = TRUE)Vignette
You can learn about how to use the package in vignette("samplingNR"), which shows how to replicate two of the tables from our paper in an application to a post-election survey of military personnel.
Basic usage
The main allocation function is opt_nh_nonresp(), which provides the exact version of our proposed optimal allocation under anticipated nonresponse. The current version assumes that the goal is to minimize the (expected) variance subject to a constraint on the total (expected) costs or invited sample size.
Fixed total sample size
If the aim is to allocate a fixed total (invited) sample size, the basic usage is:
where vectors N_h and phibar_hdenote the strata population sizes, anticipated response rates, and (optionally) strata variances, respectively, and where scalar n_total denotes the total sample size. If S_h is omitted, strata variances are assumed constant across strata.
Fixed total costs
If the aim is to allocate sample subject to a constraint on total costs, the basic usage is:
Here, cost_total denotes the total allowable costs, c_NR_h denotes the unit costs per nonrespondent (by strata), and tau_h denotes the ratio of the unit costs per respondent to those of nonrespondents (by strata). The arguments c_NR_h and tau_h can be specified as vectors of dimension H if these quantities vary by strata; alternatively, if assumed constant across strata, they can be specified as scalars.