Polling vs. Prediction: Understanding the Difference

The distinction between polling and predicting election outcomes is crucial. Polls provide a snapshot of public opinion at a specific moment, capturing voters' preferences during the polling period. In contrast, predictions are broader estimates, often based on various techniques, models, and sometimes even gut feelings. While anyone with access to sufficient information can attempt a prediction, polling relies on scientifically gathering data from representative samples of the electorate.

In the case of Numbers.lk’s final poll, the results diverged slightly from the actual outcome. This is not uncommon, as polls capture voter sentiment during the polling window, while actual voting behavior can shift, particularly in the final days leading up to the election.

The Last-Minute Shift in Voter Behavior

One well-documented trend is the phenomenon of voters shifting their support from third-party or independent candidates to the top contenders in the final days before the election. This behavior is seen globally, where voters who initially support a third-party candidate may ultimately switch to one of the top two candidates in an effort to prevent a candidate they oppose from winning.

This pattern may have played a role in the 2024 Sri Lankan presidential election as well. In our final poll, Ranil Wickremesinghe was projected to have 25% support, but he ultimately received only 17.27% of the vote. The fact that both Anura Kumara Dissanayake (AKD) and Sajith Premadasa outperformed their polling numbers by 2-4 percentage points suggests that a portion of voters may have shifted their support from Ranil Wickremesinghe to one of the top two candidates in the final days.

This discrepancy between polling and actual results might not indicate a polling error, but rather the possibility of voters changing their minds at the last moment. Our poll was conducted between the 9th and 16th of September, while the election took place on the 21st.

This dynamic can be seen in numerous elections worldwide, including U.S. presidential races, as highlighted in research by the Pew Research Center.

This last-minute shift was also evident in the 2019 Sri Lankan presidential election, where the final poll showed 5% support for AKD. However, on Election Day, AKD only garnered 3% of the vote, with much of the missing support seemingly going to Premadasa.

Here we are, the Election Day!. This is our final prediction of the Results.

Gotabaya Rajapaksha 🌷- 52.74%
Sajith Premadasa 🦢- 38.12%
Anura Kumara Dissanayake 🧭- 5.2%
Mahesh Senanayake 💡- 1.73%
Others - 0.8%#PressPollSL #lka pic.twitter.com/pimbcicRCW

— Numbers.lk (@numberslka) November 15, 2019

Comparing Polling and Predictions from Other Sources

In addition to Numbers.lk, other polling agencies, such as the Institute of Health Policy (IHP), conducted their own surveys. The IHP’s final poll, conducted through phone interviews, also sought to capture voter sentiment. However, IHP revised its initial model due to what it identified as a "social desirability effect," where respondents over-reported their support for AKD. This adjustment may have introduced additional error into their numbers, underscoring the difficulty in capturing voter intent, especially when societal factors influence responses.

Moreover, other independent analysts, like Pradeep Chanaka, released predictions for each polling division. His predictions were mostly accurate in 2019; however, this year they tended to overestimate support for AKD while underestimating votes for Premadasa and Ranil Wickremesinghe. These deviations highlight the inherent challenges of predicting tight, multi-way races.

The Keys to Numbers.lk’s Success

The key to Numbers.lk’s success lies in its comprehensive approach. Instead of merely asking respondents whom they plan to vote for, the team gathers multiple data points, including perceptions of other candidates, favorability ratings, and rejection rates (i.e., candidates voters would never support). By also considering family voting patterns—such as how respondents’ partners or parents intend to vote—the team gains valuable insights that help gauge public support more effectively.

As an online poll, Numbers.lk also takes extensive measures to guard against manipulation by campaigns attempting to skew the results. Through creative and technical methods, the team ensures the dataset remains clean and representative.