top of page


AutoPolitic’s experience with election management has given us various prediction models, to be deployed at different stages of an election cycle to help campaigns and candidates to measure effort and to improve strategy. Using Pulse Asia’s candidate list, we project the current public support in as ranges of votes.

The model’s projection should be used for ranking purposes only.

The range of votes represents the stability of a candidate’s public image and support. The smaller the range, the higher the confidence of the vote generation within the range, it also represents a stable view, which makes the adjustment to the public persona difficult. The larger the range, the lower the confidence of vote generation within that range, but it also represents a more malleable public persona that can be adjusted, given the 9 months to voting day.

Model Disclosure and Disclaimer

The current vote generation projection is a mixed modality model; candidate’s past vote history and year-long social media exposure and engagement. This model over-states the advantage of the recent election while linearly discount the votes of the past. This model also severely disadvantages Justice Carpio, who has not held an elected office. Instead of using Mayor Sara Duterte’s municipal vote, we are using her father’s voting recording, with an assumption a majority of the support will transfer from her father to her.

The voting history assumes 1) the previous vote generated is correlated with a campaign’s organizational capacity, candidate appeal, and coalition, 2) more recent vote generated is more correlated with an appeal to voters, 3) COVID plays no significant impact in historical support to 2022 support.

The social media model assumes 1) a candidate’s potential electoral performance is more correlated with long-term social reaction, 2) a candidate’s long-term appeal to media and channels will remain stable even during the campaigning period. The social media component is composed of the number of posts (Exposure), public reaction (A combination of like share and comments), and net approval (an adjusted calculation of positive and negative comments and sentiment markers)

The model assumes each candidate will generate votes independent of other candidates, and the model does not assume any interaction effect. This model has a logical flaw that is contradicted by reality. Based on the total votes generates under this model, it accounts for 128 million votes, which is larger than the entire population of the Philippines. We do not have a non-bayesian method to adjust for this fundamental model assumption. Rather than adjusting the model to introduce more biases, we represent the model even with the unrealistic total population which has the fewest heuristic biases.

Data provided by QSearch from 2020/8/9 to 2021/8/8. Total of 261,204,849 data points, inclusive of sentiment markers, comments, and internet UL addresses.

For questions, elaboration, or campaign application, please email


Featured Posts
Recent Posts
Search By Tags
Follow Us
  • Facebook Basic Square
  • Twitter Basic Square
  • Google+ Basic Square
bottom of page