Fig 1.
The process in three parts (1) A representative survey was completed by 3050 randomly selected people living in Denmark, providing information on standard sociodemographic qualities, political values, and present-day voter intention toward parties eligible in the general election. As shown in S3 Table, the sample is somewhat demographically representative of the country’s entire population. Respondents were subsequently asked to log in with their Facebook account, and if willing to accept the same, respondents’ public Facebook ID was stored. (2) Post-likes were independently collected from all public profiles of Danish parties and politicians on Facebook. (3) After completion of steps 1 and 2, we linked each respondent to the collected Facebook data and applied a LASSO-based multinomial logistic regression model to predict voter intention based on Facebook data.
Fig 2.
Multinominal logistic regression models predicting present-day voting intention.
Table 1.
Prediction rates and sample sizes at different party-like caps with min likes = 7 (p < 0.001).
Fig 3.
The x-axis shows party-like cap (PLC), which denotes how many likes in percentages that at the least go toward only a single party, meaning that at PLC = 0.8, only users who have at least 80% likes toward a single party are included. The y-axis shows the percentage of users who are accurately labeled. Each colored line shows accuracy for samples where all respondents have a minimum of total likes. Because the two criteria, party-like cap and minimum likes, involve filtering out respondents and thus effectively cutting down the sample size, it is unfeasible to rely on the training of machine learning algorithms for classification. Consequently, we made a simple algorithm that derives predictions based on the party a respondent has liked the most at different intersections of the two criteria.
Fig 4.
Combining trending news topics with users’ voter intention.
Example showing trending topics for a group of users aligned with a specific party through their likes.