I investigated Ann Selzer's botched Iowa poll showing Trump losing. This is why she got it all so shockingly wrong, writes top White House pollster

On the eve of the 2024 election, a shock poll out of Republican-red Iowa set the political world on fire.

The so-called 'gold standard' Des Moines Register survey, conducted by veteran pollster Ann Selzer, showed Vice President Kamala Harris leading Donald Trump 47 percent to 44 percent.

Trump carried the state in 2020 and 2016 by nearly 10-percentage points. 

Harris's sudden surge in the state was supposedly driven by support from women and especially female voters aged over 65 years old, who backed the Democrat by a whopping 35 points.

The survey was seized upon by Harris's surrogates and a left-leaning media that suggested it was evidence of a 'hidden Harris' voter that all the other national and swing state polls had somehow missed.

In the end, of course, this 'hidden Harris' voter never materialized – in Iowa or elsewhere.

Trump won the state by 13 percentage points, the largest margin ever recorded there since 1972, and a 16-point differential from Selzer's prediction.

Days after the election, the Des Moines Register announced they had launched a review of Selzer's polling methods to determine what had gone so dramatically wrong.

It wasn't only that the poll was extraordinarily off, but Selzer's erroneous numbers tarnished the reputation of the Register, the polling industry as a whole and may even have impacted the outcome of the election.

Could her results, broadcast around the country in the campaign's waning days, have driven more Harris voters to the polls? Could they have had the opposite effect and motivated Trump supporters to turn out?

Both scenarios are conceivable. We'll never know the full impact.

On Sunday, Selzer announced her pre-planned retirement from the Register to pursue 'other ventures and opportunities.'

Hours later, President-elect Trump made his own declaration.

'A totally Fake poll that caused great distrust and uncertainty at a very critical time. [Selzer] knew exactly what she was doing,' Trump claimed on Truth Social, adding, 'An investigation is fully called for!'

Well, I've conducted my own investigation into Selzer's polling. And my conclusion is this survey may be the shoddiest pieces of work I've ever seen in my five decades in the polling industry. 

First, I will say that I don't know Ann Selzer.

Our paths have never crossed. I know only her reputation as a reliable professional, dubbed the 'Queen of Polling' after she correctly predicted that a virtually unknown senator would defeat Hillary Clinton in the Iowa Caucuses in 2008.

But after having examined the methodology of her November poll, I am stunned. Its glaring problem isn't difficult to spot.

In the survey, Selzer asked 808 likely voters in Iowa: 'Who did you vote for in 2020, Biden or Trump?'

Selzer must have expected to receive results that resembled the actual vote in the 2020 general election – Trump won 53.1 percent to Biden's 44.9 percent, an eight-point differential.

Indeed, those were the results that she's gotten in the past.

In a February 2024 Register poll in Iowa, respondents said they voted for Trump over Biden in 2020: 45 percent to 38 percent – a seven-point margin.

In June 2024, the sample went for Trump 49 percent to Biden's 35 percent – a 14-point margin.

And in September, that gap narrowed to four points.

But in the November poll, the numbers flipped. Then, the respondents said they voted for Biden over Trump, 41 percent to 40 percent. That's a 9-point skew from the actual 2020 election outcome in the Democrats's favor.

Now, in some cases, people answering polls have been known to lie about their past voting behavior and claim to have backed the eventual winner.  But the magnitude of the variance here is too large to ignore.

Selzer acknowledged this discrepancy last weekend when she released her analysis of the polling blunder. Though she seemed to dismiss the finding, writing, 'maybe there is some merit in this' but ultimately concluded, 'I found nothing to illuminate the [November poll's] miss.'

As a polling analyst and a former project director on President Ronald Reagan's White House public opinion team, I know that a discrepancy of that magnitude should have sent a chill down the Register team's proverbial spine.

For it indicates that Selzer may have surveyed a group of Iowans who do not accurately reflect the sentiment of the overall electorate in the state.

At this point, to my mind, Selzer would have had two options: She could either have sent her polling team back into the field to collect a more representative sample or she could have not released the poll.

This is the industry standard when data doesn't smell right. However, Selzer did neither.

Instead, she 'weighted' or artificially increasing the influence of Trump 2020 voters in her poll.

'Weighting' is an accepted practice that enables a pollster to control for under or over-representation by inflating or deflating the relative percentage of a certain groups of voters in a sample.

But, when 'weighting' is applied to a fundamentally flawed sample, the results may only get worse.

In other words, it's like putting lipstick on a pig. It'll dress the animal up a bit, but it's still livestock.

I cannot presume to know exactly what went wrong with Selzer's poll, but the over-representation of Biden 2020 voters raises the possibility that she was over-sampling in predominately liberal regions of the state (like the areas around liberal Grinnell and Coe colleges).

If Selzer did survey a disproportionate number of Trump 2020 voters who were living in Democratic-enclaves than it is reasonable to assume that these Iowans may not have been representative of Trump 2020 voters statewide.

This type of sample can never be 'corrected' by weighting.

Beyond the curious potential internals of Selzer's poll methodology, the final results  themselves should have been the biggest red-flashing warning sign.

Iowa, save parts of Des Moines and its small college towns, is about as red as Judy Garland's slippers in The Wizard of Oz. In fact, Iowa is so reliably red that most pollsters now don't bother polling the state during a general election.

Additionally, Selzer has polled in the state for 30 years. She should know its voters like the back of her head and, frankly, Iowa is not the most difficult place to conduct a survey. It's mostly rural with only three significant urban centers.

If there was a massive voter surge toward Harris, a native daughter like Selzer would surely have also recognized supporting anecdotal evidence, like 'Harris 2024' signs in front yards in Republican areas, or major GOP influencers pulling their support from Trump.

But none of that evidence was present.

Selzer's poll might have been a shock to America. But it should not have been a shock to Selzer that her last survey was a bust.