Keywords: election,online polling,CATI,probability panel

Predicting the outcomes of election results via polling is often a complex and demanding process, which can easily result in providing inaccurate and biassed information to the general public, as we have witnessed in the past, using traditional polling methods. In countries where online access does not have full coverage, it can however lead to coverage bias that needs to be accounted for.

This study examines the accuracy and evolution of polling data leading up to the 2024 North Macedonian parliamentary elections, comparing results obtained via an online probability panel using email and SMS invitations with data from traditional CATI (Computer-Assisted Telephone Interviewing) surveys and the actual election outcomes. The research aims to evaluate the predictive validity of online surveys, explore temporal shifts in voter preferences over the campaign period, and identify key differences between methodologies.
SMS messages were used to invite participants from an online probability panel (Lifepanel) to the survey, ensuring demographic representativeness. CATI surveys were conducted by independent organisations over the same period. Polling results were analysed across three timeframes: the first three months, the following two months, and the final month before the election. Each method's results were compared to the actual election outcomes to assess accuracy.

The online probability panel surveyed 3,000 respondents, with new samples introduced monthly to account for potential attrition. CATI survey data were drawn from publicly available sources. The study also includes actual election results for benchmarking.
The key metrics that were examined include polling accuracy (measured as deviation from actual results), trends in voter preference shifts over time, demographic differences between methods, and the impact of methodological bias. The evolution of undecided voter proportions was also analysed. Finally, an analysis was conducted on the online panel survey data which was re-weighted after the election.

The results of this research suggest that the online survey predicted the distribution of votes between the first and second party candidates more accurately than the CATI survey, while the CATI survey captured the distribution of votes among the lesser voted candidates more precisely. Notable shifts in voter preferences occurred in the final two months, with undecided voters consolidating their preferences closer to election day. Both methods displayed significant deviations from the actual results in the early months, narrowing more closely in the final month.

This research highlights the strengths and limitations of Email and SMS-based online surveys compared to CATI in pre-election polling. By understanding these differences, researchers can better interpret polling data and adapt methodologies to improve accuracy in diverse electoral contexts.