ceopinio / escons

An R package to simulate seat allocations using electoral surveys - Paquet d'R per simular la distribució d'escons a partir d'enquestes

The escons R package

escons is an R package from the Centre d'Estudis d'Opinió to simulate seat allocations using data from electoral surveys.

The package provides two main functions. simulate is the main tool for simulation. It consumes the vote shares and an associated margin of error for each party in each district and outputs an array of seat allocations in each district using the Catalan electoral system. The package also provides a method of as.data.frame to transform the output of simulate into a structure that is easier to analyze.

An example

Consider, for instance, a dataset with the results of the February 2021 elections to fill the 135 seats of the Parliament of Catalonia.

res2021 <- data.frame("Barcelona"=c(25.0, 20.4, 17.9, 7.8, 7.8, 6.3, 6.1, 4.0),
                      "Girona"=c(15.2, 21.8, 32.7, 6.1, 4.0, 9.0, 3.3, 2.0),
                      "Lleida"=c(15.0, 26.6, 28.0, 5.5, 3.2, 7.4, 3.2, 3.5),
                      "Tarragona"=c(20.0, 24.5, 19.4, 9.4, 4.9, 6.8, 5.2, 4.3))/100

Note that the package will consume data in proportions, which is why the dataset is divided by 100.

The package provides a utility function moe to calculate the margin of error for each vote share. The function takes two relevant arguments. One defines the confidence level associated to the margin of error. The other defines the sample size for each district.

sig2021 <- moe(res2021, N=3000, level=.95)

Users have the option to declare the name of the parties -- otherwise, parties will be identified using the row.names attribute of the data.

parties <- c("PSC", "ERC", "JxCat", "Vox", "ECP–PEC", "CUP–G", "Cs", "PP")

The main function of the package is simulate. It will make 1,000 (by default) extractions from a normal distributino centered at the vote share with standard deviation defined by the margin of error. The function also provides arguments to set the electoral threshold and the size of the district. The defaults values for these arguments use the current values in the Catalan electoral system -- an electoral threshold of 3% in each district and allocation of 85 seats for Barclona, 17 for Girona, 18 for Tarragona, and 15 for Lleida.

res <- simulate(res2021, sig2021, parties)

The resulting object from the simulation is a 3-dimensional array. To facilitate the use of the simulations for data analysis, the package exports a method as.data.frame that creates a "long" representation of the data

res <- as.data.frame(res)

The as.data.frame method will also attach the names of the parties and calculate party totals across districts to the resulting object.

Warning

The (unexported) function dhondt that is invoked during the simulation resolves ties in a district at random. This is, of course, different from how the electoral system resolves ties. Therefore, there is a possibility of small differences between the seats allocations produced by the package and the official ones -- a difference that disappears across simulations.

Installation

The package is not available on CRAN and must be installed from the GitHub repository directly.

install.packages("remotes")
remotes::install_github("ceopinio/escons", subdir="escons")