Here are some more details on the Counterfactual modeling of Philippine Presidential Elections and why only the 2024 has a counterfactual to FPJ (hawak ko Ang flip! 😂) based on Sensitivity Analysis of the @yudapearl Pearl-Halpern SCM (Structural Causal Model):

The reason Fernando Poe Jr. (FPJ) emerges as the only robust counterfactual winner across the entire modeling stack is that the 2004 election is the sole historical case in which every major component of the structural causal framework points in the same direction. The Pearl-Halpern Structural Causal Model (SCM) begins with observed first-preference votes (FP), latent second-preference and intensity structures (SP/Score), and a set of exogenous variables representing region, social class, historical alliances, dynastic relationships, campaign effects, and candidate characteristics. The key intervention is do(alternative electoral system), which severs the direct mapping from plurality first preferences to the final winner and instead routes the election outcome through latent transfer behavior. Under this intervention, the question becomes whether the winner changes when voters are allowed to express preferences beyond a single first-choice vote. In every election except 2004, the plurality winner retains sufficient latent coalition support to survive this intervention. In 2004, however, the intervention consistently redirects enough support toward FPJ to alter the final outcome.

The internal mechanics are important. The model does not assume arbitrary second preferences. Instead, transfer probabilities are calibrated using regional vote matrices, alliance structures, linguistic blocs, historical political behavior, SWS and Pulse Asia polling patterns, NAMFREL observations, and documented relationships among candidates. This produces Fermi priors in which eliminated candidates' supporters transfer to one of the remaining candidates at rates generally between 55% and 70%, depending on ideological and regional proximity. The model then evaluates runoff voting, ranked-choice voting (RCV), and score/range voting using these calibrated latent preference structures. Because all three systems expose information hidden by plurality voting, they function as independent tests of whether the plurality winner truly possessed the broadest coalition.

The 2004 election satisfies all conditions necessary for a flip. First, the observed plurality margin between Gloria Macapagal Arroyo and Fernando Poe Jr. was extremely small, approximately 3.5 percentage points. This means the amount of latent support required to overturn the plurality result is correspondingly modest. Second, the third-place vote pool was unusually large and unusually coherent. Lacson, Roco, and Villanueva collectively represented nearly one-quarter of the electorate. More importantly, these blocs were not politically neutral. Contemporary analyses consistently describe them as disproportionately anti-incumbent, reformist, outsider-oriented, or opposition-adjacent. Within the SCM, these characteristics create strong causal pathways from latent preference variables to FPJ rather than Arroyo. Once plurality is removed as the aggregation mechanism, these pathways become visible.

1/n