Guest column: Traditional polling worthless in predicting ballot-proposal votes

By Mark Grebner/Practical Political Consulting Inc.

Some polls prove to be right on the money, while a few miss by five or even 10 points. But some polls that aren’t worth anything at all: The ones that try to predict how a ballot proposal will fare with the voters.

When the final pre-election poll published by the Detroit Free Press showed Rick Snyder ahead of Virg Bernero by a 55-37 margin in the 2010 gubernatorial election, we could expect (after giving each candidate half the undecided vote) the Republican would take about 59 percent of the votes cast for major parties.

And that’s exactly what happened.

If that prediction had missed by two points, nobody would have blinked. If the prediction had missed by five points, the pundits would have been called upon to murmur about “late deciders” or a “strong ground game” or some such blarney.

But no one with any sense doubted that “The Nerd” would be sworn in as governor, and Virg would continue to occupy his perch across the street at Lansing City Hall. People on each side made plans for what they'd be doing after the election, confident that they knew who would win and who would lose.

In contrast, if somebody sold their house, or quit their job, based upon the predicted result of a ballot question, he’d be a fool. Predictions concerning ballot proposals often miss by 10 points, sometimes by 20, and even occasionally by 30 points. Ballot proposal polls aren't “inaccurate." They are "useless."

For example, there were five proposals on the Michigan statewide ballot in November 2006, four of which were polled in the week before the election. In order to compare them with the election results, I assume the undecided vote should be divided equally between yes and no. (Other rules for handling the undecideds can be proposed, but they don’t work any better -- except when selected after the fact because they work with a particular result.)

The most attention was paid to Proposal 2, ending affirmative action programs. Selzer, one of the most highly rated polling firms in the United States, showed Prop 2 losing, 39 percent to 49 percent. Allocating the undecided, we would expect an election result of 45 percent in favor.

But Prop 2 actually passed overwhelmingly, with 58 percent of the vote.

That’s a 13-point error, almost three times the supposed “margin of error” reported by the Detroit Free Press.

Proposal 3, which would have legalized the hunting of mourning doves, was polled by EPIC-MRA. The predicted result: A loss of 25 percent to 66 percent. This suggests Prop 3 would receive 29.5 percent on Election Day. Since the actual result was 31 percent in favor, you could say the poll was accurate -- in the same sense that a stopped clock is accurate, as long as you are careful when you look at it.

Proposal 4, which tightened the restrictions on taking property via eminent domain, was the most interesting. Selzer reported a very close contest, 43-44, which equates to a razor-thin 49.5 percent loss (with the undecideds). But, in fact, Prop 4 won with 80 percent of the vote. That 30-point error should occur roughly one time in 10 billion, according to statistical theory.

Finally, Proposal 5, which would have guaranteed a level of funding for public schools and colleges, was polled by Selzer at 43-45, predicting 49 percent support.  The actual result -- 38 percent in favor -- was “only” 11 points off. This counts as a success.

Notice that of four proposals, polling was right once, too low twice, and too high on the fourth. That's typical. After each election, the pollsters, instead of being embarrassed into silence, explain that their results were actually quite accurate, once they’re “adjusted” using a method that was selected only after the election was held.

I don’t have space to explain why phone-based polling fails to predict how ballot proposals actually fare at the polls, but the problems are deep-seated and cannot be fixed.

If you want to know how a ballot proposal will do, you have to use a different method than telephone interviews.

In any event, don’t make the mistake of believing the polls that claim to tell you how the voters would treat proposals on Right to Work, or the emergency manager law, if they're on the ballot.

Your gut instinct is probably a better guide than any telephone-based poll.