paper/paper.md

In adibender/coalitions: Bayesian "Now-Cast" Estimation of Event Probabilities in Multi-Party Democracies

title: 'coalitions: Coalition probabilities in multi-party democracies' tags: - coalitions - rstats - coalition probabilities - political science authors: - name: Andreas Bender orcid: 0000-0001-5628-8611 affiliation: "1" - name: Alexander Bauer orcid: 0000-0003-3495-5131 affiliation: 1 affiliations: - name: Statistical Consulting Unit, StaBLab, LMU Munich index: 1 date: 03 March 2018 bibliography: paper.bib

Summary

In multi-party democracies, election coverage usually focuses on raw results from polls on questions like

Who would you vote for if the election was tomorrow?

Whether a coalition (union of multiple parties) will obtain enough votes to form a governing coalition is discussed by adding up votes obtained by the parties in question, while ignoring sample uncertainty and redistribution of votes for parties beneath a specific threshold (e.g., 5% in Germany).

The R [@R] package coalitions [@coalitions_zenodo] implements methods that overcome those shortcomings and quantifies sample uncertainty in terms of probabilities for events of interest. Specifically, it contains functions to

• Obtain survey results from different polling agencies,
• Aggregate (pool) multiple surveys (from different pollsters) within a pre-specified time-window, taking into account the correlation between different polling agencies
• Perform Monte Carlo simulations of election outcomes based on the (pooled) survey results
• Redistribute votes based on the method specific to the election of interest (e.g., Saint-Lague-Scheppers for German Bundestag election)
• Calculate Bayesian posterior probabilities for specific events, e.g., to obtain enough votes (> 50%) to form a governing coalition

To get started

• the workflow vignette describes the usual steps during the analysis
• the pooling vignette gives details on the aggregation of multiple surveys.

An example for the (backend) application of the package can be found at

• http://koala.stat.uni-muenchen.de,

where it is applied to German (federal and state wide) elections/surveys.

Currently, the functionality focuses on German federal and state-wide elections. However, it can be easily extended to other multi-party democracies, given the user can obtain the necessary data and transform it to a suitable format. For example, the methods have been successfully applied to calculate coalition probabilities for the 2017 elections in Austria. Contributions are welcome at: https://github.com/adibender/coalitions

References

adibender/coalitions documentation built on July 23, 2024, 4:28 p.m.