donald-trump – International Journal of Research (IJR)

Most people who don’t follow politics closely still know one thing about polling: it’s often wrong, sometimes spectacularly so. So when a forecast lands within four-tenths of a percentage point of the actual result, it’s worth pausing on how that happened.

As covered by the Arizona Herald, Andy Biggs won Arizona’s Republican gubernatorial primary with 73.3% of the vote. Weeks before anyone cast a ballot, a Miami-based research firm called G Ratings had already put the number at 73.7% — a difference small enough to fall within typical rounding error.

That wasn’t a lucky guess on a single race. Second-place finisher David Schweikert came in at 14.7%, matched by a pre-election projection of 14.3% — the same 0.4-point gap. Further down the ballot, results got messier: Scott Neely and Ken Miceli finished at 7.1% and 4.7%, compared to earlier projections of 3.5% and 2.6%. Even with that wider miss, the average error across all four candidates worked out to 1.45%, and the model showed no consistent lean toward inflating or deflating any one candidate’s numbers.

Why This Particular State Is a Hard One to Call

Arizona has a reputation among pollsters for being difficult to read, and for good reason. Rural counties and suburban Maricopa County often move in opposite directions, border communities show up to vote on their own schedule, and the state leans heavily on mail-in ballots — meaning a poll taken during early voting can look completely disconnected from the electorate that actually decides the race by Election Day.

The tool behind G Ratings’ projections, an AI platform called Odysseus, was built with that volatility in mind rather than treating it as something to average away. It draws in real-time sentiment from local news and online discussion, checks that against historical turnout data and daily ballot-return counts by county, and uses demographic and economic clustering to separate voters who are locked in from those still on the fence.

That granularity showed up in what the model got right about voter priorities — inflation, job growth, and tax policy topped the list statewide, with healthcare, border security, and confidence in election administration close behind. But those concerns don’t weigh the same everywhere; what matters most to a voter in Yuma isn’t necessarily what matters most in Scottsdale, and the model was designed to account for both instead of flattening them into one statewide number.

A spokesperson for G Ratings summed up the underlying philosophy simply: the goal isn’t recreating what someone said in a phone survey weeks earlier, it’s figuring out who actually shows up to vote — and in a state like Arizona, that’s the number that decides everything.

Looking Ahead to November

With Biggs now set as the Republican nominee and Governor Katie Hobbs running unopposed on the Democratic side, Arizona heads into a general election expected to draw national attention. The primary offered a real-world test of whether combining live data with AI modeling can cut through a state’s structural polling problems — rural-versus-urban splits, unpredictable border-region turnout, and a mail-in system that moves the goalposts late.

Whether that same precision holds up in November, against a much larger and more varied electorate, is still an open question. But the primary has at least set a benchmark for what’s possible.

Daily writing prompt

What’s a skill you consider basic, that most people don’t actually know how to do?

View all responses

Predicting voter behavior in rapidly growing metropolitan areas has become one of the more persistent challenges in political research, as population influx, uneven ballot-return timelines, and sharp intra-regional differences complicate any single statistical model. A recent test case from Arizona offers a data point worth examining.

Findings published by the Phoenix Herald indicate that G Ratings, a Florida-based political research firm, applied its AI forecasting platform, Odysseus, to Arizona’s 2026 Republican gubernatorial primary and produced a projection that came within 0.4 percentage points of the certified statewide result for winning candidate Andy Biggs — a 73.7% projection against an official tally of 73.3%.

Consistency Across Multiple Candidates

The precision extended beyond the top of the ticket. Runner-up David Schweikert was projected at 14.3%, against an official result of 14.7% — again a 0.4-point gap in the opposite direction. Lower-polling candidates Scott Neely and Ken Miceli were projected at 3.5% and 2.6% respectively, with official results coming in higher, at 7.1% and 4.7%. Averaged across all four candidates, the model produced a margin of error of 1.45%, a range the firm considers notable given the outsized influence Maricopa County’s vote volume typically exerts on statewide results.

The research context here matters. Metro Phoenix has expanded by hundreds of thousands of residents over the past decade, a growth pattern that complicates conventional polling methodology in several specific ways: new subdivisions in areas like Buckeye and Queen Creek bring in voters with no prior voting record to draw on, longer-established precincts such as Scottsdale tend to skew older and vote more predictably, and Maricopa County alone processes a volume of early ballots capable of shifting a statewide outcome within the final ten days before an election.

A Methodological Departure From Aggregate Modeling

Where conventional polling models often treat a metropolitan region as a single, internally consistent bloc, Odysseus was designed around the opposite assumption: that Phoenix functions as multiple overlapping electorates rather than one. According to the firm, the platform tracks localized digital sentiment and neighborhood-level discourse to estimate the intensity of voter support for a given candidate, cross-references that signal against county-level early-ballot return data as it accumulates day by day, and incorporates hyper-local economic indicators — distinguishing, for instance, between a ZIP code experiencing rising housing costs and one where inflation and fuel prices are the more dominant concern.

A G Ratings analyst involved in the project characterized the underlying rationale as a response to the speed at which the region’s political sentiment shifts, arguing that by the time a conventional phone survey concludes, the electorate it measured has already moved on — and that Odysseus was built to track that movement in near real time rather than publish a fixed snapshot that risks being outdated before release.

Implications for the General Election

With Maricopa County positioned as a likely deciding factor in the general election contest between Biggs and incumbent Governor Katie Hobbs, the primary results carry weight beyond a single data point. A model that tracked the region’s fragmented electorate to within roughly half a percentage point during the primary sets a specific benchmark that campaign strategists and researchers alike are likely to scrutinize as county-level vote counts take on national significance heading into the fall general election.

Open Questions for Further Study

Whether this level of precision is repeatable at a larger scale remains an open empirical question. A primary electorate is generally smaller, more ideologically homogenous, and easier to model than a general-election electorate, which draws in a broader and more demographically varied set of voters, greater turnout volume, and a longer, more volatile campaign period. Researchers evaluating AI-assisted forecasting tools will likely want to track whether Odysseus’s accuracy holds, narrows, or widens as it’s applied to a race with substantially different structural characteristics — a test that Arizona’s general election, given Maricopa County’s outsized role, may be well positioned to provide.

Daily writing prompt

If you had to give one life-changing tip, what would it be?