Going Postal
2019 Sri Lankan Presidential Election
Postal Voting for the 2019 Sri Lankan Presidential Election commenced on October 31st. Various state employees including members of the armed forces and police are eligible to vote by post.
In 2015, postal votes made up about 4.3% of the valid votes polled. Of the 12,123,452 valid votes, 522,663 were voted-in by post. In 2019, 655,915 voters will be eligible to vote by post.
Postal votes are interesting because they are among the earliest results announced. Hence, many people try to gauge the direction and swing of an election based on postal ballots. But is this actually possible?
In this article, I try to answer: Can we use postal votes alone, to accurately predict the final result of an election?
Brief History of Postal Votes
First, let’s look at previous presidential elections, and how postal votes compared to the final result. For simplicity, I analyse the two main parties. Greens (UNP and NDF), and Blues (SLFP, PA, and UPFA).

In all elections but 1988, the “winner” of the total postal votes, went on to win the election. In 1994, 1999 and 2005, the proportion of postal votes a party received was very close to the final result. 1999 was also unusual in that both main parties got relatively fewer postal votes, with the JVP candidate (Nandana Gunatilleke) receiving disproportionately more postal ballots.
In 1982, 2005 and 2010, the winning party received a disproportionately larger number of postal votes. The 2005 postal votes were particularly significant. Without these, the UPFA would not have reached the 50% limit needed to win a presidential election outright. A second/third preference count might have been necessary.

Across electoral districts, there is some variation in the proportion of postal votes relative to total votes. In 2015, the Colombo electoral district had the smallest portion of postal votes (1.9%), while Anuradhapura (8.2%) had the largest. This probably reflects the portion of the population employed by the state and armed forces.

Can we use postal votes to predict the final result of an election?
For 2005, 2010 and 2015, I built a regression model to predict the final vote proportion for Green and Blue, based on only the postal votes the parties received from the 22 electoral districts.
This chart represents the “Best Case Predictions” for Blues and the Greens. In all three elections, the regression predictions were wide enough for both parties to have some possibility of winning the election.
Even in 2010 (an extremely one-sided election), postal votes, alone, were not enough to predict a win for Blue.

I anticipated this “non-result” in Election Projections. In 2010, while the projection was able to call the final result of the election, this was only after about 30 results, and after all the postal votes were announced.

Conclusion
So, can we use postal votes alone, to accurately predict the final result of an election? The answer is, no.
While postal votes are quite interesting, and are strongly correlated with the final election result, alone, they are insufficient to predict the outcome of the election with statistical confidence.
Hence, as the results of November 16th are announced, and if you see your team “winning” the postal votes, don’t get too excited. Things could change.
Happy Voting!

