Beyond the Numbers - Statistical Lens of the 2025 Congressional Elections in Marinduque using the Monte Carlo Method(10,000 simulated elections)

2025 June 5

By Eduardo P. Carabeo Jr.

(Certified Google Advanced Data Analytics Professional)

As a data analyst deeply invested in understanding the dynamics of Marinduque’s politics, I was struck by the 2025 congressional election’s unexpected outcome. I simply just had to make sense of it all in the lens of Statistics and Data Analytics.

This experience didn’t just spark my curiosity—it underscored for me the power and limitations of data. It reminded me that surveys are snapshots, not predictions set in stone. And it reaffirmed my belief that timely, relevant scientific surveys are crucial to truly understanding the voice of the electorate as it changes and evolves.

The 2025 congressional elections in Marinduque was expected to be a reaffirmation of popular support for the incumbent governor, who was running for the province’s lone congressional district seat. He had commanded an overwhelming 70% of voter preference in a January 2025 scientific survey. However, the actual result—a 17,000-vote loss out of 129,000 total votes cast—has left many Marinduqueños wondering how such a dramatic reversal was possible. This article provides a comprehensive, data-driven analysis of the election outcome, using multiple statistical tools to assess whether the result can be explained by normal variations in voter behavior.

Z-Score Test - Uncovering Statistical Impossibility

The Z-score test is a way of measuring how far away a number is from what you’d expect, using the idea of “standard steps.” Imagine you’re on a hike and you know the average step length is 2 feet. If someone suddenly takes a 20-foot step, you’d say, “Whoa! That’s 10 times bigger than normal.”

A Z-score does the same thing with election results (or any other data). It says, “This result is this many ‘standard steps’ away from the usual.” If the number of steps is really big, it means the result is way outside what’s normal—so it’s probably not just a random accident.

A Z-score test compares the actual result to what would be expected if the incumbent's 70% support level was accurate. The observed share of 43.4% translates to a Z-score of -208.75—more than 208 standard deviations below expectation. The corresponding probability (p-value) is virtually zero.

Z = -208.75
P-value ≈ 0

Monte Carlo Simulation - Testing the Odds 10,000 Times

(10,000 Simulated Elections)

The Monte Carlo Method is like rolling dice thousands of times to see how likely a certain result is. Instead of making a single prediction, it runs a huge number of “what if” scenarios. By seeing how often each possible outcome pops up, we can better understand what’s likely—and what’s not.

In the case of elections, the Monte Carlo Method helps us see how often a certain vote count might happen if people voted the way surveys say they will. If a surprising result never shows up in these thousands of “pretend elections,” it suggests that the real result was very unlikely—or that something changed in how people really felt.

A Monte Carlo simulation with 10,000 trials was conducted, each simulating an election with 70% support for the incumbent governor running for congress. In none of these trials did the governor receive as few as 56,000 votes—the actual total that led to the 17,000-vote loss.

Sensitivity Analysis - Finding the Breaking Point

A sensitivity analysis is like testing “what if” scenarios to see how much things change when you tweak a starting point.

Imagine adjusting the volume on a radio - if you turn it just a little, does the music stay the same, or does it get a lot louder?

In elections, a sensitivity analysis asks - “How much would voter support have to change for a surprising result to start making sense?” By adjusting the expected support up or down in small steps, we find the point where a big loss (or gain) becomes statistically believable.

In simple terms - it’s a way of checking if the final result is just a fluke, or if it could actually happen if things had shifted a lot behind the scenes.

A sensitivity analysis was performed to pinpoint the expected support level at which a 17,000-vote loss would become statistically plausible (defined as having at least a 5% chance of occurring). The analysis found that such a loss is only plausible if the governor’s true support was below 44%.

Reverse Modeling - Working Backward to Reality

Reverse modelling is like solving a puzzle backward. Instead of starting with what you know and predicting what might happen, you start with the final result and work backward to figure out what must have caused it.

In elections, if we know how many votes someone got, reverse modelling asks: “What level of voter support would typically produce that number of votes?” It’s like asking, “What speed would I have to be driving to cover this distance in 10 minutes?”—you’re using the final answer to figure out the starting conditions.

Reverse modeling was also performed, asking what actual support level would produce 56,000 votes in a typical election? The analysis showed that this would require the governor's true support to be about 43.4%—again, drastically lower than the January survey’s 70%.

Statistics, The Power of Surveys and Their Limits

Surveys are a snapshot in time. In January 2025, the governor’s 70% support appeared to guarantee victory. However, surveys are only as accurate as the moment they’re conducted—and as relevant as the environment they reflect.

As months passed, new issues, events, rehashed and unanswered social media downright malicious and false accusations, disinformation spread on the ground and misinformation could have caused dramatic shifts in public sentiment which have gone undetected without updated scientific surveys. Relying solely on older data can create a dangerous blind spot for candidates and the public alike.

For the people of Marinduque, these findings have profound implications. The statistical analyses—Z-score, Monte Carlo simulation, sensitivity analysis, and reverse modeling—consistently show that a 17,000-vote loss is not plausible if the governor's true support was anywhere near 70%. This suggests either an unprecedented collapse in voter sentiment that went undetected because of lacking more recent scientific survey data, a failure of polling methods, or factors that may warrant further examination to ensure the integrity of future elections.

Understanding these statistical truths empowers us to ask the right questions. Elections are not just about numbers; they are about the trust of the people and the integrity of our democratic process.

As I reflect on the insights from this analysis, one message resonates most clearly – Statistics are not just about numbers; they are about capturing our hopes, frustrations, and aspirations. It tells stories in its own unfiltered lens.

Moving forward, I hope we can all support and value scientific, transparent surveys—not as political tools, but as bridges between voters and those who seek to serve. Lastly, this article is not meant to denigrate the actual results, nor the people’s will – the Marinduquenos have spoken.

In this spirit, let’s keep asking questions, sharing our stories, and staying engaged—because a democracy that listens is a democracy that grows stronger every day. /epcj050625