Case Study 19-2: Meridian's Probability Report — The Garza-Whitfield Race

Case Study 19-2: Meridian's Probability Report — The Garza-Whitfield Race

Background

This case study follows Vivian Park, Carlos Mendez, and the Meridian Research Group as they produce a full probabilistic assessment of the Garza-Whitfield Senate race — specifically examining how Meridian communicates uncertainty to two very different clients: the Garza campaign (their primary client) and a nonpartisan civic organization (the Westfield Democracy Project) that wanted to use the assessment for voter education.

The case illustrates how the same underlying data can be framed very differently for different audiences, and what Vivian considered the ethical limits of that framing.

The Underlying Analysis

Carlos had spent four days building Meridian's probabilistic model for the Garza-Whitfield race. The inputs:

Polling average (quality-weighted): Garza +3.8 Fundamentals-adjusted baseline: Garza +4.0 Historical SD for comparable Senate races: 3.2 points Correlated error adjustment: 20% probability of a ±2-point national environment shift in the final two weeks

Monte Carlo results (50,000 simulations): - Garza win probability: 79.4% - Whitfield win probability: 20.6% - Median simulated margin: Garza +4.0 - 80% confidence interval: Garza +0.1 to Garza +7.9 - 95% confidence interval: Whitfield +2.2 to Garza +10.2

Scenario decomposition: - Strong Garza (>+6): 28% - Moderate Garza (+3 to +6): 35% - Narrow Garza (0 to +3): 16% - Narrow Whitfield (0 to -3): 13% - Strong Whitfield (>+3 for Whitfield): 8%

This was Meridian's most honest assessment. The question was how to communicate it.

The Garza Campaign Report

Vivian drafted the campaign-facing report herself, with Carlos editing for clarity. She applied each component of the Vivian Park method:

Opening summary: "Our integrated model gives Senator Garza a 79% win probability as of October 14. This is a meaningful advantage — comparable to the probability of drawing a face card (J, Q, K) from a full deck. But it is not a certainty: roughly one in five scenarios produces a Whitfield win, and those scenarios are specific and addressable."

The range emphasis: "Our 80% confidence interval on the election margin runs from essentially tied (Garza +0.1) to a comfortable Garza victory (+7.9). We expect the result most likely to fall in the +2 to +6 range, but a closer or wider margin is genuinely possible."

Making uncertainty actionable: "The 20% Whitfield scenarios cluster around two conditions: (1) national Republican wave materializing in the final two weeks (something we'd see in generic ballot shifts and in our tracking polls), and (2) Latino turnout falling below 2020 levels. The campaign can monitor both of these directly:

Track generic ballot weekly (a shift >2 points Republican should trigger resource reallocation)
Monitor Latino early voting through the secretary of state's office — if Latino early vote share in [key counties] falls below 2020 pacing by more than 15%, accelerate GOTV in those areas"

Transparency about uncertainty drivers: "Our confidence in Garza's lead is high — all methods (structural, polling, integrated) agree on direction and approximate magnitude. Our uncertainty about the final margin is driven primarily by (a) likely voter screen variability across our polls, (b) uncertainty about whether 2020 or 2018 better predicts Latino turnout this cycle, and (c) the inherent unpredictability of the final two weeks."

Vivian showed the draft to Carlos. "Does this give them what they need to make decisions?"

"Yes," Carlos said. "It also gives them the truth."

"Those things should be the same," Vivian said.

The Westfield Democracy Project Report

The Westfield Democracy Project had a very different brief: they wanted to use Meridian's assessment for a voter education publication that would be distributed widely. Their editor was concerned about one thing: would publishing a 79% win probability for Garza suppress voter turnout?

This was precisely the influence problem from Chapter 17, applied now at the level of a commissioned probabilistic report.

Vivian's response was careful:

"I understand the concern," she told the Westfield director. "But I don't think the solution is to distort or suppress the probability. The solution is to communicate the uncertainty honestly — and to contextualize the 79% correctly."

The Westfield report was drafted with different emphases:

Opening: "Based on current polling and structural conditions, this Senate race appears to be a moderate Democratic-leaning race, with genuine uncertainty about the outcome. The current evidence suggests roughly four out of five scenarios in which Senator Garza wins, and roughly one in five in which Mr. Whitfield wins."

The uncertainty emphasis: "Neither outcome is a certainty. Races with this profile can and do go either way — the one-in-five Whitfield scenario is not a remote possibility. Every vote matters in determining which scenario materializes."

No strategic information: Unlike the campaign report, the Westfield document included no strategic framing about resource allocation or county-level GOTV. That information would be inappropriate in a voter education context.

Call to action: "With a meaningful chance of either candidate winning, voter turnout is likely decisive. Voters who want to affect the outcome should participate."

Vivian believed this framing was honest and appropriate for the audience. It didn't suppress the probability assessment, but it contextualized it in a way that emphasized rather than dampened the importance of participation.

The Ethical Negotiation

The Westfield director pushed back on one thing: "Can't you just say it's a 'competitive race' without giving the specific probability?"

Vivian's answer was no, with explanation.

"Calling it 'competitive' without a probability lets every reader fill in their own number. Some will hear 'competitive' and think 50/50. Some will hear it and think 90/10. The specific probability — communicated with its uncertainty — is more informative and more honest than a vague label. If we're in the business of civic information, we should be as specific as we honestly can be."

Carlos, sitting in on this conversation, thought about what Vivian had told him when he first joined Meridian: "The probability is the most honest thing we have. It's more honest than a headline, more honest than a quote, more honest than a 'likely.' If we start softening probabilities for audiences we think can't handle them, we're not in the information business anymore."

Discussion Questions

1. Vivian used different framings for the same underlying 79% win probability: "face card from a deck" for the campaign, "four out of five scenarios" for the civic organization. Is this legitimate, or is it manipulative? What's the ethical line between appropriate audience adaptation and misleading framing?

2. The Westfield Democracy Project was concerned about turnout suppression. Vivian's response was to communicate uncertainty more prominently rather than suppress the probability. Is this a satisfying answer? Are there circumstances where a forecaster should decline to publish a probability because of expected social consequences?

3. The campaign report includes strategic information (which counties to target for GOTV) that the civic organization report doesn't. What are the appropriate uses of probabilistic forecasting information, and who should have access to which levels of detail?

4. Carlos remembers Vivian saying "the probability is the most honest thing we have." Evaluate this claim. In what ways is a probability more honest than other representations of uncertainty? In what ways might it be less honest or less informative?

5. If you were the Westfield Democracy Project director, would you publish the 79% win probability, even with Vivian's contextualization? What factors would you weigh in making this editorial decision?

Quantitative Extension

The following exercise asks you to apply the Vivian Park method to a different race.

Imagine you are briefing a campaign on a gubernatorial race. Your model shows: - Polling average: Republican candidate up 1.8 points - Historical SD: 3.5 points - Monte Carlo result: Republican wins in 69% of simulations, Democrat in 31% - 80% CI: Democrat +2.7 to Republican +6.3 - Scenario decomposition: Strong R (>+5): 22%, Moderate R (+2 to +5): 28%, Narrow R (0 to +2): 19%, Narrow D (0 to -2): 17%, Strong D (>+2 for D): 14%

Write a 300-400 word briefing for the Republican campaign, applying all four components of the Vivian Park method. Then write a separate 200-word version for a general public voter education publication. Identify the key differences between the two versions and explain why they differ.