The R `sampling' package
Alina Matei and Yves Tillé
University of Neuchâtel
Euskal Estatistika Erakundea
XXIII Seminario Internacional de Estadística
November 2010
Alina Matei and Yves Tillé () The R `sampling' package November 2010 1 / 13
Introduction and aim The R language
The R language
Shareware available on [Link]
The Comprehensive R Archive Network
Installation: 10 minutes
All the manuels are available in pdf
Everyone can write an additional package (600 packages are available)
Packages are loaded directly from R.
The manual of the package is available online and in pdf.
Package `sampling' written by Matei and Tillé.
Alina Matei and Yves Tillé () The R `sampling' package November 2010 2 / 13
Introduction and aim History of the package
Continuous distributions
EFTA course for public statisticians (April 2005).
Objective : to apply directly the theory with the R language.
Theory + Exercices with a laptop and R.
Writing of a large set of procedures.
Finally, decision of submitting the package to the CRAN.
Alina Matei and Yves Tillé () The R `sampling' package November 2010 3 / 13
Content of the package Topics
Content of the package
Strati cation, two-stage, unequal probabilities, balanced sampling
Estimation: calibration and regression estimator
Tools : computation of inclusion probabilities, crossing strata
Data bases, Swiss municipalities, Belgian municipalities.
Alina Matei and Yves Tillé () The R `sampling' package November 2010 4 / 13
Content of the package Tools
Tools
writesample: return the list of all the samples of xed sample size
cleanstrata: renumbering of the strata
disjonctive return a matrix wit 0 and 1 that is the disjonctive
representation of the stratum.
inclusionprobabilities: compute unequal inclusion probabilities
from an auxiliary variable variable.
Alina Matei and Yves Tillé () The R `sampling' package November 2010 5 / 13
Content of the package Data bases
Data bases
MU284 A data frame with 284 municipalities on the following 11
variables : populations, political results.
swissmunicipalities: 2896 Swiss municipalities. Surfaces and
population.
belgianmunicipalities: 589 Belgian municipalities 11 variables,
population and taxes.
Alina Matei and Yves Tillé () The R `sampling' package November 2010 6 / 13
Content of the package Simple random sampling
Simple random sampling
srswor: Simple random sampling with replacement.
srswor1: Simple random sampling without replacement (sequential
method).
srswr: Simple random sampling with replacement.
Alina Matei and Yves Tillé () The R `sampling' package November 2010 7 / 13
Content of the package Unequal probability sampling
Unequal probability sampling
UPbrewer,
UPmaxentropy, (set of function)
UPmidzuno, UPmidzunopi2,
UPmultinomial,
UPpivotal, UPrandompivotal,
UPpoisson,
UPsampford,
UPsystematic, UPrandomsystematic, UPsystematicpi2,
UPtille, UPtillepi2,
Alina Matei and Yves Tillé () The R `sampling' package November 2010 8 / 13
Content of the package Balanced sampling
Balanced sampling
Design that satis es the balancing equations
X xk X
= xk ,
πk
k ∈S k ∈U
where xk is a vector of auxiliary variables.
Cube algorithm: ight phase and landing phase.
samplecube, fastflightcube, landingcube
Complex survey balancedstratification balancedcluster
balancedtwostage
Alina Matei and Yves Tillé () The R `sampling' package November 2010 9 / 13
Content of the package Balanced sampling
Exercises
Exercise
Compute inclusion probabilities 200 Belgian municipalities with inclusion
probabilities proportional to the population in 2004.
Alina Matei and Yves Tillé () The R `sampling' package November 2010 10 / 13
Content of the package Balanced sampling
Exercises
Exercise
Use the Belgian database. Select a sample of 200 municipalities with
unequal probabilities proportional to the number of inhabitants in 2004.
with Poisson sampling
with a method of unequal probabilities and xed sample size
with simple random sampling.
Compute the Horvitz-Thompson estimators of the taxable income for 50
samples and draw a boxplot of the estimators for each method.
Alina Matei and Yves Tillé () The R `sampling' package November 2010 11 / 13
Content of the package Balanced sampling
Exercises
Exercise
Use the database of Swiss municipalities, and select a strati ed balanced
sample. A balanced sample is rst selected in each strata. Next the results
of the ight phase are merged and a ight phase is applied again on the
whole population. Finally, a landing phase is applied on all the population.
Use the following balancing variables: HApoly, Surfacesbois, P00BMTOT,
P00BWTOT, POPTOT, Pop020, Pop2040, Pop4065, Pop65P, H00PTOT.
The sample size is 400 and the municipalities must be selected with
inclusion probabilities proportional to POPTOT. The strati cation variable
is REG (swiss regions). Next, print the names of the selected municipalities.
Alina Matei and Yves Tillé () The R `sampling' package November 2010 12 / 13
Content of the package Balanced sampling
Exercises
Exercise
Use the Belgian database. Select a sample of 200 municipalities with
unequal probabilities proportional to the number of inhabitants in 2004
with Poisson sampling design. Next calibrate the sample by means of the
raking ratio estimator on the variables:
mean(Men03), mean(Women03), Dimen, Diwom, TaxableIncome,
Totaltaxation, averageincome, medianincome.
The division by the means is necessary to avoid too large numbers.
Compute the Horvitz-Thompson estimators and the calibrated estimators
for the calibration variables. Limit the variation of the g-weights between
0.5 and 1.5.
Alina Matei and Yves Tillé () The R `sampling' package November 2010 13 / 13