2020 Sri Lankan Parliamentary Election

Real-Time Projections (#GenElecSL2020)

“Predicting” the Election in Real-Time

With the hardships of CoViD19, many seem to have lost interest in #GenElecSL2020. It would be interesting to see if turnout exceeds the previous lows of 61.26% (in 2010) and 63.60% (1989). But as with many things with little data, and no precedent, predictions are almost impossible. Hence, I’m not going to attempt predicting turnout, nor the final result.

Instead, this article is like the Election Projections article I wrote around #PresPollSL2019. Where I ask the question:

“Can we use early results to project and predict the final result? And if so, how accurate would such a projection be?”

The Experiment

To find out, I simulated #GenElecSL2015. As with my previous article,

• I assumed that election results report in order of result size. Starting with smaller results like postal votes. Ending with larger polling divisions like Homagama, Kaduwela, and Nuwara Eliya-Maskeliya.
• For each set of reported results, I built a regression model. That predicted the outcome based on these reported results. The model trained on the previous six general elections (between 1989 and 2010).
• I validated the results by computing error bounds.