Merge pull request #4 from bschneidr/sample-based-calibration

Sample based calibration (for Issue #3)

.Rbuildignore

@@ -6,3 +6,5 @@
 ^codecov\.yml$
 ^cran-comments\.md$
 ^CRAN-SUBMISSION$
+^doc$
+^Meta$

.gitignore

@@ -3,3 +3,8 @@
 .RData
 .Ruserdata
 *.Rproj
+inst/doc
+/doc/
+/Meta/
+census-api-key
+successive-differences-replication.R

DESCRIPTION

@@ -8,8 +8,9 @@ Description: Provides tools for creating and working with survey replicate weigh
   extending functionality of the 'survey' package from Lumley (2004) <doi:10.18637/jss.v009.i08>.
   Methods are provided for applying nonresponse adjustments to
   both full-sample and replicate weights as suggested by 
-  Rust and Rao (1996) <doi:10.1177/096228029600500305> in order to account for
-  the impact of these adjustments on sampling variances.
+  Rust and Rao (1996) <doi:10.1177/096228029600500305>.
+  Implements methods for sample-based calibration described by Opsomer and Erciulescu (2021) 
+  <http://www.statcan.gc.ca/pub/12-001-x/2021002/article/00006-eng.htm>.
   Diagnostic functions are included to compare weights and weighted estimates
   from different sets of replicate weights.
 License: GPL (>= 3)
@@ -21,6 +22,13 @@ Imports:
     survey (>= 4.1),
     utils
 Suggests: 
+    knitr,
     covr,
-    testthat (>= 3.0.0)
+    testthat (>= 3.0.0),
+    rmarkdown,
+    tidycensus,
+    dplyr
 Config/testthat/edition: 3
+Depends: 
+    R (>= 2.10)
+VignetteBuilder: knitr

NAMESPACE

@@ -1,10 +1,18 @@
 # Generated by roxygen2: do not edit by hand
 
 S3method(redistribute_weights,svyrep.design)
+export(calibrate_to_estimate)
+export(calibrate_to_sample)
 export(redistribute_weights)
 export(stack_replicate_designs)
 export(summarize_rep_weights)
 export(svyby_repwts)
 importFrom(stats,as.formula)
+importFrom(stats,coef)
+importFrom(stats,model.frame)
+importFrom(stats,model.matrix)
+importFrom(stats,na.pass)
 importFrom(stats,setNames)
+importFrom(stats,terms)
+importFrom(stats,weights)
 importFrom(utils,packageVersion)

NEWS.md

@@ -1,5 +1,35 @@
 # svrep (development version)
 
+* Added functions `calibrate_to_estimate()` and `calibrate_to_sample()`
+for calibrating to estimated control totals with methods
+that account for the sampling variance of the control totals.
+For an overview of these functions, please see the new vignette
+"Calibrating to Estimated Control Totals".
+
+  * The function `calibrate_to_estimate()` requires the user
+  to supply a vector of control totals and its variance-covariance matrix.
+  The function applies Fuller's proposed adjustments to the replicate weights,
+  in which control totals are varied across replicates by perturbing the control
+  totals using a spectral decomposition of the control totals'
+  variance-covariance matrix.
+
+  * The function `calibrate_to_sample()` requires the user to supply
+  a replicate design for the primary survey of interest as well as a replicate
+  design for the control survey used to estimate control totals for calibration.
+  The function applies Opsomer & Erciulescu's method of varying 
+  the control totals across replicates of the primary survey by matching each 
+  primary survey replicate to a replicate from the control survey.
+
+* Added an example dataset, `lou_vax_survey`, which is a simulated survey 
+measuring Covid-19 vaccination status and a handful of demographic variables,
+based on a simple random sample of 1,000 residents of Louisville, Kentucky
+with an approximately 50% response rate.
+  * An accompanying dataset `lou_pums_microdata` provides person-level microdata
+  from the American Community Survey (ACS) 2015-2019 public-use microdata sample
+  (PUMS) data for Louisville, KY. The dataset `lou_pums_microdata` includes
+  replicate weights to use for variance estimation and can be used to generate
+  control totals for `lou_vax_survey`.
+
 # svrep 0.1.0
 
 * Initial release of the package.