WASHINGTON -- Two recent polls in Ohio suggest an overwhelming defeat in the offing for Issue 2, the ballot measure that would ratify Republican Gov. John Kasich's controversial limits on collective bargaining by state public employees. But past misfires in polling on statewide referendums provide good reason for caution until all votes are counted on Tuesday night.
An automated, recorded-voice telephone survey conducted over the weekend by the Democratic firm Public Policy Polling (PPP) found voter sentiment trending against Issue 2, by a margin of 59 to 36 percent.
A live interview poll conducted a week earlier by Quinnipiac University found a similar result: By a 56 to 33 percent margin, Ohio's registered voters said they oppose limiting collective bargaining for public employees, as Kasich's Senate Bill 5 does.
Despite the apparent consistency of these results, the pro-union group Progress Ohio put out a memo in late October bluntly warning that polling on "complicated issues" like Ohio's Issue 2 "is unreliable" and that the "blowout" predicted in the PPP and Quinnipiac polls is based "on flawed public opinion samples."
While unions have an incentive to warn their Ohio supporters against complacency, they are nonetheless right to highlight the unreliability of public polling on ballot measures. History shows it has been frequently wrong or misleading.
Georgetown University political scientist Daniel Hopkins recently examined 438 public surveys that asked about support for state-level referendums between 2003 and 2010. He found that the average error -- the difference between the margin forecast by the poll and the actual vote -- was 7.8 percentage points. More important, roughly a quarter of these polls (26.5 percent) incorrectly forecast the outcome.
Ohio has a particularly checkered recent history in polls on issue referendums, including the spectacular failure of a Columbus Dispatch mail-in poll in October 2005. That survey forecast that two election reform questions would pass with roughly 60 percent of the vote; they lost by margins of better than 2-to-1.
Why is issue referendum polling more error-prone? One reason is that voter turnout in off-year elections is significantly lower than in even-year general elections, so pollsters have a more difficult time identifying the true likely electorate. Just 3.1 million voters turned out for Ohio's special election in 2005, for example, compared to the 5.8 million who voted in the 2008 presidential election.
PPP's automated surveys use a combination of two methods to identify likely voters: They randomly select households from a list of registered voters with some history of past voting and then begin each interview with an instruction that those unlikely to vote should hang up and not complete the survey. It's a somewhat unorthodox technique, but PPP's pollsters believe their record of accuracy owes much to their use of the automated method to simulate the secret ballot.
The poll conducted by Quinnipiac University made no effort to identify likely voters. It simply reported results among all registered voters.
A second big challenge for issue referendum polling is replicating the actual ballot language. Referendum questions are often long and confusing, and some voters will not make up their minds until they read the text on the ballot. Both supporters and opponents of Issue 2 have complained that neither the Quinnipiac nor earlier PPP survey questions used the exact language that will appear on the Ohio ballot.
On its most recent survey, however, PPP reproduced the actual ballot language, asking respondents how they would vote on "Senate Bill 5, which is a new law relative to government union contracts and other government employment contracts and policies." Later in the same interview, PPP repeated a question asked on previous surveys, which describes Issue 2 as "a referendum on whether to approve or reject Senate Bill 5, which was passed earlier this year, and limits collective bargaining rights for public employees." The results on the two questions are virtually identical.
Despite the past miscues, this latest round of Issue 2 polls may be accurate. As The Huffington Post's Sam Stein reports, the referendum campaign has been "heated and expensive" and even made an entrance into the presidential primary when Mitt Romney first declined to take a position and then later endorsed the anti-collective bargaining measure. By this time, Ohio's voters may have well-formed preferences on Issue 2 that are not easily distorted by the vagaries of poll wording or likely-voter methodology.
Hopkins, the Georgetown professor, noted that while polling on ballot issues can be highly volatile, predictable errors emerge only on issues like same-sex marriage, immigration and marijuana legalization. Pre-election polls on these issues typically overstate one particular side, presumably because respondents don't want "to seem homophobic, anti-immigrant, or pro-marijuana" to a stranger on the phone, he wrote. But issues "like education and tax reduction" -- in other words, questions similar to Issue 2 -- tend to produce random errors in both directions.
So the opponents of Ohio's Senate Bill 5 have good reason to be optimistic about the outcome of this week's election, but the history of polling on ballot issues suggests they should take nothing for granted.