Sample Design, Drawing & Data Analysis Using Data Frames.

sampler R package

R Package for Sample Design, Drawing, & Data Analysis Using Data Frames

The sampler R package is designed to enable data scientists to design, draw, and analyze simple or complex samples using data frames. It enables you to load machine-readable files (e.g. .csv, .tsv, etc.) in R containing a sampling frame or collected data, store them as objects, and perform sampling techniques and analysis using clear and concise methods.

Specifically, a data scientist can use the sampler R package to:

• determine simple random sample sizes, stratified sample sizes, and complex stratified sample sizes using a secondary variable such as population
• draw simple random samples and stratified random samples from sampling data frames
• determine which observations are missing from a random sample, missing by strata, duplicated within a dataset
• perform data analysis, including proportions, margins of error and upper and lower bounds for simple, stratified and cluster sample designs

The sampler R package builds a bridge for survey administrators between the free and open-source R environment and no-to-low cost Open Data Kit (ODK)-based toolkits such as Ona and ELMO. The sampler package is available via CRAN or GitHub for use in R and R Studio.

To install in R from CRAN:

install.packages("sampler")
library(sampler)

To install in R from GitHub:

install.packages("devtools"); library(devtools)
devtools::install_github("mbaldassaro/sampler"); library(sampler)

The sampler R package includes the following datasets:

• albania: dataset containing 2017 Albania election results by polling station published by the Central Election Commission and opened by the Coalition of Domestic Observers & Democracy International
• opening: dataset containing 2017 Albania election observation findings on polling station opening process by the Coalition of Domestic Observers (CDO) CDO conducted a statistically-based observation (SBO) exercise, deploying observers to a random sample of polling stations for the 25 June 2017 Albanian elections. This is a subset of observation data collected by CDO observers that includes data that was used to perform statistical analysis

Full documentation of datasets and functions can be found on RDocumentation

Determine random sample size

rsampcalc(N, e, ci=95,p=0.5, over=0)

Where:

• N is population universe (e.g. 10000, nrow(df))
• e is tolerable margin of error (integer or float, e.g. 5, 2.5)
• ci (optional) is confidence level for establishing a confidence interval using z-score (defaults to 95; restricted to 80, 85, 90, 95 or 99 as input)
• p (optional) is anticipated response distribution (defaults to 0.5; takes value between 0 and 1 as input)
• over (optional) is desired oversampling proportion (defaults to 0; takes value between 0 and 1 as input)

Returns appropriate sample size (rounded up to nearest integer)

Example:

rsampcalc(N=5361, e=3, ci=95, p=0.5, over=0.1)

Source: Sampling Design & Analysis, S. Lohr, 1999, equation 2.17

Draw a simple random sample

rsamp(df, n, over=0, rep=FALSE)

Where:

• df is object containing full sampling data frame
• n is sample size (integer or object containing sample size)
• over (optional) is desired oversampling proportion (defaults to 0; takes value between 0 and 1 as input)
• rep (optional) is boolean for a sample with repalcement (TRUE) or without replacement (defaults to FALSE)

Returns a simple random sample of size n

Example:

rsamp(albania, n=360, over=0.1, rep=FALSE)

or

size <- rsampcalc(nrow(albania), 3, 95, 0.5)
rsamp(albania, size)

Determine sample size by strata using proportional allocation

ssampcalc(df, n, strata, over=0)

Where:

• df is object containing sampling data frame
• n is sample size (integer) or object containing sample size
• strata is variable in sampling data frame by which to stratify
• over (optional) is desired oversampling proportion (defaults to 0; takes value between 0 and 1 as input)

Returns proportional sample size per strata (rounded up to nearest integer)

Example:

ssampcalc(df=albania, n=544, strata=qarku, over=0.05)

or

size <- rsampcalc(nrow(albania), 3, 95, 0.5)
ssampcalc(albania, size, qarku)

Source: Sampling Design & Analysis, S. Lohr, 1999, 4.4

Draw stratified sample (proportional allocation)

ssamp(df, n, strata, over=1)

Where:

• df is object containing full sampling data frame
• n is sample size (integer, or object containing sample size)
• strata is variable in sampling data frame by which to stratify (e.g. region)
• over (optional) is desired oversampling proportion (defaults to 0; takes value between 0 and 1 as input)

Returns stratified sample using proportional allocation without replacement

Example:

ssamp(df=albania, n=360, strata=qarku, over=0.1)

or

size <- rsampcalc(nrow(albania), 3, 95, 0.5)
ssamp(albania, size, qarku)

Determine sample size by strata using sub-units

psampcalc(df, n, strata, unit, over=0)

Where:

• df is object containing full sampling data frame
• n is sample size (integer) or object containing sample size
• strata is variable in sampling data frame by which to stratify
• unit is variable in sampling data frame containing sub-units (e.g. population)
• over (optional) is desired oversampling proportion (defaults to 0; takes value between 0 and 1 as input)

Returns sample size per strata based on sub-units (rounded up to nearest integer)

Example

psampcalc(df=albania, n=544, strata=qarku, unit=zgjedhes, over=0.1)

Source: Sampling Design & Analysis, S. Lohr, 1999, 4.4

Calculate proportion and margin of error (simple random sample)

rpro(df, col_name, ci=95, na="", N=0)

Where:

• df is object containing data frame on which to perform analysis (e.g. data)
• col_name is variable in data frame for which you want to calculate proportion and margin of error
• ci (optional) is confidence level for establishing a confidence interval using z-score (defaults to 95; restricted to 80, 85, 90, 95 or 99 as input)
• na (optional) is value that you want to filter and exclude (defaults to include everything)
• N (optional) is population universe (e.g. 10000, nrow(df)); if N value is passed as an argument, margin of error will be calculated using fpc

Returns table of responses (n), proportions (midpoint), margins of error, lower and upper bounds by factor for a given variable

Example:

rpro(df=opening, col_name=openTime, ci=95, na="n/a", N=5361)

Source: Sampling Design & Analysis, S. Lohr, 1999, Equation 2.15

Calculate proportion and margin of error (stratified sample)

spro(fulldf, sampdf, strata, col_name, ci=95, na="")

Where:

• fulldf is object containing original data frame used to draw sample
• sampdf is object containing data frame on which to perform analysis
• strata is variable in both data frames by which to stratify
• col_name is variable in data frame for which you want to calculate proportion and margin of error
• ci (optional) is confidence level for establishing a confidence interval using z-score (defaults to 95; restricted to 80, 85, 90, 95 or 99 as input)
• na (optional) is value that you want to filter and exclude (defaults to include everything)

Returns table of responses (n), proportions (midpoint), margins of error, lower and upper bounds by factor for a given variable in a stratified sample

Example:

spro(fulldf=albania, sampdf=opening, strata=qarku, col_name=openTime, ci=95, na="n/a")

Source: Sampling Design & Analysis, S. Lohr, 1999, 4.6 & 4.7

Calculate proportion and margin of error (unequal-sized cluster sample)

cpro(df, numerator, denominator, ci=95, na="", N=0)

Where:

• df is object containing data frame on which to perform analysis
• numerator is variable in data frame for which you want to calculate proportion and margin of error
• denominator is variable in data frame containing population of unequal cluster sizes
• ci (optional) is confidence level for establishing a confidence interval using z-score (defaults to 95; restricted to 80, 85, 90, 95 or 99 as input)
• na (optional) is value that you want to filter and exclude (defaults to include everything)
• N (optional) is population universe (e.g. 10000, nrow(df)); if N value is passed as an argument, margin of error will be calculated using fpc

Returns table of responses (n), proportions (midpoint), margins of error, lower and upper bounds by factor for a given variable in an unequal-sized cluster sample

Example:

alresults <- ssamp(albania, 890, qarku)
cpro(df=alresults, numerator=totalVoters, denominator=zgjedhes, ci=95)
cpro(df=alresults, numerator=pd, denominator=validVotes, ci=95, N=5361)

Source 1: Survey Sampling, L. Kish, 1965, Equation 6.3.4

Source 2: Sampling Techniques, W.G. Cochran, 1977, Equation 3.34

Identify missing points between sample and collected data

rmissing(sampdf, colldf, col_name)

Where:

• sampdf is object containing data frame of sample points
• colldf is object containing data frame of collected data
• col_name is common variable (i.e. key) in data frames by which to check for missing points

Returns table of sample points missing from collected data

Example:

alsample <- rsamp(df=albania, 544)
alreceived <- rsamp(df=alsample, 390)
rmissing(sampdf=alsample, colldf=alreceived, col_name=qvKod)

Identify number of missing points by strata between sample and collected data

smissing(sampdf, colldf, strata, col_name)

Where:

• sampdf is object containing data frame of sample points
• colldf is object containing data frame of collected data
• strata is variable in both data frames by which to stratify
• col_name is common variable (i.e. key) in data frames by which to check for missing points

Returns table of number of sample points by strata missing from collected data

Example:

alsample <- rsamp(df=albania, 544)
alreceived <- rsamp(df=alsample, 390)
smissing(sampdf=alsample, colldf=alreceived, strata=qarku, col_name=qvKod)

Identify duplicate values within collected data

dupe(df, col_name)

Where:

• df is object containing data frame of collected data
• col_name is variable within data frame by which to filter for duplicate values

Returns table of duplicate values within collected data

Example:

aldupe <- rsamp(df=albania, n=390, rep=TRUE)
dupe(df=aldupe, col_name=qvKod)

Remove observations based on duplicate values within collected data

dedupe(df, col_name)

Where:

• df is object containing data frame of collected data
• col_name is variable within data frame by which to filter for duplicate values

Returns table of observations based on unique values within collected data

Example:

aldupe <- rsamp(df=albania, n=390, rep=TRUE)
dedupe(df=aldupe, col_name=qvKod)