You asked and we answered: How the Free Press polls work

A lot of readers seem to have questions about how the Free Press' Michigan polls — and polling in general — works, so we wanted to clear the air somewhat headed into the Nov. 5 general election.

The first thing to know is that polls are based on the statistical likelihood that a survey of a sample population, such as likely voters, will predict an outcome consistent with circumstances at the time they were surveyed. It doesn't mean circumstances won't change and the outcome with it and it doesn't mean it's always right. In fact, a polling error — when looking across averages of political polls compared to the Election Day outcomes — of 3 or 4 percentage points is not at all unusual (and can seem like a great deal in a closely contested race).

Still, independent polls are more reliable than anecdotal evidence, campaign trail predictions by politicians or media punditry independent of data and are generally accurate up to a point: Polls always come with what's known as a margin of error, which is based on factors like the size of the sample. The margin of error says that, for the full sample, the result's accuracy is really expected within a range (such as, for a Free Press poll where 600 likely voters were sampled, a margin of error of plus or minus 4 percentage points).

What that means is that if the poll says Candidate X is ahead of Candidate Y, 52% to 48%, it could be a tie or X could be ahead 56%-44% or Y could even be ahead by a couple of points. The range of those possibilities certainly favors X but that's not meant to be a guarantee. That margin of error and the poll as a whole could be off, too, though statistically speaking that should happen rarely in good polls (that's what's known as the confidence level).

More: Kamala Harris leads Donald Trump in Michigan by 3 points in final Detroit Free Press poll

More: Slotkin stakes out 5-point lead on Rogers in US Senate race, Detroit Free Press poll says

But let's take a look at the Free Press' and EPIC-MRA's final polls of the top-of-the-ticket races from 2016 to 2022.

In 2016, the final poll had Democratic presidential nominee Hillary Clinton winning by 4 points. She lost, to Donald Trump, on his way to becoming president, in effectively what was a tie, by two-tenths of a percentage point. Two years later, the final poll had Democrat Gretchen Whitmer winning the gubernatorial election by 5 points. She beat Republican nominee Bill Schuette by 9.

In 2020, the final EPIC-MRA poll had Democrat Joe Biden winning Michigan over Trump by 7 points. He did indeed win on his way to becoming president but by about 3 points. And in 2022, the firm had Whitmer winning reelection over Republican Tudor Dixon by 11 points ? which she did.

In each of those cases, the final result may not have hit the exact percentages of support that were predicted for each of the top two candidates. But in terms of the margin of victory or defeat they did come within (or pretty close to) that 4-percentage-point margin of error of being correct — yes, even when Trump won Michigan in 2016.

Again, that's not to say a poll can't be wrong. It can be. Sometimes, it is.

It's also important to remember that reputable pollsters, like EPIC-MRA of Lansing, which does polling for the Free Press, don't pick respondents to its surveys based on their political preferences or the answers they might give. They're picked randomly. And every effort is taken so that the questions themselves don't seem to be leading to a particular answer that might favor one political position or another.

We talked this week with EPIC-MRA pollster Bernie Porn about how that random selection of voters is made and how the Free Press' Michigan polls come together. He said that the firm maintains a database of Michigan voters by matching phone numbers with the state's Qualified Voter File. Generally speaking, the universe of voters selected for possible contact by the pollsters for a survey can include newly registered voters but it's also going to zero in on voters who have been active in recent elections, voting, say in presidential or gubernatorial contests.

As part of the survey, if it's of likely voters, people who are contacted are asked how sure they are that they're going to vote and if they're unsure, the survey ends.

Porn said that EPIC-MRA's statewide surveys are arranged geographically, meaning the firm looks at how much of the statewide vote each region of Michigan and the state's largest cities accounted for in a given set of previous elections (such as the previous two presidential elections), then that's used to divvy up the target sample proportionally between those geographic subdivisions. Then the firm's callers — EPIC-MRA uses live callers, not computerized calls, and these days 80% of those calls are made to cell phones, not landlines — start making the calls with a set script to ask each person who agrees to take part and responds to all of the questions. If they don't get enough responses to hit the set geographic proportion, they keep calling until they do.

Porn said the only other forced part of the survey is the gender breakdown, which is 53% women and 47% men and approximates the statewide electorate. (In Detroit, he said, that breakdown is 57% or 58% women because of the difference in the electorate there and its size relative to the rest of the state.)

"Very seldom do we have to do any weighting," said Porn, referring to a practice where the responses from a subgroup — say self-identified Democrats or men without a college degree or union households or voters over the age of 65 — are balanced within the poll in an attempt to represent their perceived voting strength. One caveat, he said, is Black voters, whose responses are weighted only if they represent less than 11% of the statewide survey. (The 2020 Census showed Black people made up about 14% of the state's population and exit polls indicated Black people were 12% of the electorate in the 2020 election.)

Porn also said that EPIC-MRA tries to make sure more younger voters are represented in the random voter pool in the hope they are adequately represented in the final sample as they tend to be harder to reach.

And, yes, he said, getting responses from voters can be difficult. Across all regions of the state taken together, Porn said, it can take from 15 to 20 calls in order to get one respondent that completes a survey. That means 9,000 to 12,000 calls need to be made for every survey.

One final thing: We get a lot of emails from readers who wonder why, after living in the state for so long, they haven't been polled. The answer is it's a simple matter of math. There are about 7.2 million registered voters in Michigan, meaning the chance of being selected for a 600-person sample is about 1-in-12,000.

That's just a little less than the odds of being struck by lightning at least once in your lifetime.

Contact Todd Spangler: [email protected]. Follow him on Twitter@tsspangler.

This article originally appeared on Detroit Free Press: You asked and we answered: How the Free Press polls work