Bayesian Estimation of Presidential Elections in Ghana: A Validation Approach

Abstract

Elections are one of the barometers through which electorates measure the performance of governments and decide whether to renew their mandate or not. The success of every election goes a long way to strengthen the frontiers of a country's democracy and provide legitimacy for those who hold political power. However, the electoral process of many African countries has been challenged in courts or allegations of fraud and vote rigging are leveled against the winning party or candidate. Therefore, there is the need for a statistical method for checking and validating election results to ascertain fraud and vote rigging claims. Existing validation methods include the Parallel Vote Tabulation methodology. However, some significant disadvantages of this approach are issues of cost, sampling techniques and sample size determination. To overcome these, this study resorts to using the Dirichlet multinomial Bayesian model to compute posterior probabilities of valid votes cast and Bayesian credible intervals to ascertain the legitimacy of the votes cast. Using the Ghana general elections in 2020, the fitted Bayesian model accurately predicted approximately 99% of the proportion of votes obtained by New Patriotic Party, National Democratic Congress and all Other Political Parties. Also, the valid votes received by all the political parties fall within the Bayesian credible intervals indicating that the credibility of the 2020 presidential elections held in Ghana may not be in doubt.