By Dave Kopel
Rocky Mountain News. November 15, 2008
How accurate were the pre-election polls in the Rocky Mountain News and The Denver Post? Pretty good on the presidential and Senate contests, but inconsistent on the statewide ballot issues.
The final pre-election Rocky polls were a joint project with CBS4 News, and run by Republican pollster Lori Weigel of Public Opinion Strategies. Democratic opinion strategist Rick Ridder helped with the sample questionnaire and analysis. Samples were taken Oct. 21-23, and published in the Rocky in a series of articles Oct. 27-29.
The Post's final poll, by Mason-Dixon Polling & Research, was conducted a week closer to the election, on Oct. 27-28. Accordingly, it would be expected to be more accurate.
When the real votes were counted, Barack Obama won Colorado by 9 percent. The Post had put Obama +5, and the Rocky at +12.
This was within the 4 point margin of the Post poll, and the 4.38 margin in Rocky. What the papers call the "margin of error" is technically known as a "confidence interval." It means that if the poll is done right, then 95 percent of the time a true count of all Colorado voters would yield results within 4 (or 4.38) points of the poll result.
Incidentally, an Associated Press poll, published in the Post on Oct. 29, was right on the money, showing Obama ahead by 9.
Mark Udall defeated Bob Schaffer by 10 points in the U.S. Senate race. The Rocky put Udall +13, and thePost +14. Again, the polls were right, within the margin of error.
Amendment 46 would have prohibited race and sex discrimination by government. It lost by less than 2 percent. The Rocky showed the initiative ahead 53 percent to 40 percent, with 7 percent undecided. But the Rocky article reporting the poll did warn that any initiative that isn't well past 50 percent in a pre-election poll could lose.
The Post got much closer to the actual result, showing the anti-discrimination measure losing by 4 percent.
Amendment 47 said that employees who do not want to join a union may not be forced to pay fees to the union. It lost by 11 percent. The Rocky poll was dead-on, showing the amendment behind by 11 percent. The Post poll picked the winning side, but was far outside the margin error, showing the right-to-work amendment losing by 28 percent.
The voters crushed Amendment 48, which would have said that legal personhood begins from the moment of conception. The Amendment lost by 47 percent. The Rocky showed it down by 41 percent, and the Post by 38 percent. Both results were outside the margin of error, although the polls could still be accurate if we assume that the large majority of voters who were undecided in the poll eventually voted no.
Amendment 50 raised the limit on legal gambling bets in Colorado's three casino towns from $5 to $100. It passed with 58 percent, slightly outside the Rocky margin of error, which showed 64 percent support. ThePost did not poll this one.
Amendment 52 would have used some oil and gas severance tax revenue for highway improvements. It was demolished 64-36. The Rocky didn't poll this, while the Post showed 43 percent opposed, 34 in favor, and 23 percent undecided. Accurate if we assume that about 9 of 10 undecided voters eventually voted no.
Amendment 58 would have raised taxes on oil and gas producers. Fifty-eight percent of the public voted against Amendment 58. The Rocky showed the measure trailing by a mere 4 points, while the Postshowed -9.
Amendment 59 would have eliminated the Taxpayer's Bill of Rights limits on growth of government spending. It lost by 9 percent. The Rocky poll was wildly off, showing the Amendment ahead by 8 points. But the Rocky article about the poll did warn that much of the amendment's support was soft, whereas opponents were firmer in their convictions.
The Post was also on the wrong side of this one, showing a 3 percent lead, with 21 percent undecided.
Altogether, there were 12 Rocky and Post polls regarding ballot issues published shortly before the election. If we look at the yes/no margin in the polls, and compare those with the actual results, we find that of those 12 polls, only two were accurate (within the margin of error).
Arguably, this wasn't the pollsters' fault, since some voters changed their minds, and sometimes the undecideds broke very heavily in one direction. Apparently voters this year were much more volatile about the amendments than about the candidates, since the candidate polls were accurate. Next election, readers should remember that the gap between polls and results can be quite large, especially on ballot issues.