An outfit called Terror Free Tomorrow claims in an op-ed in today’s Washington Post that the contested Iranian elections likely were not fraudulent, since a pre-election poll it sponsored showed the declared winner, Mahmoud Ahmadinejad, with a big lead.
TFT’s own data, though, tell a different story – as, oddly, did its own previous polling analysis.
The poll, done by telephone last month, found 34 percent support for Ahmadinejad vs. 14 percent for Mir Hossein Mousavi. The incumbent led by “a more than 2 to 1 margin – greater than his actual margin of apparent victory in Friday’s election,” today’s op-ed says. “Our scientific sampling from across all 30 of Iran’s provinces showed Ahmadinejad well ahead.”
Strange, then, that TFT’s analysis of these same data last month predicted a runoff.
The problem with both analyses is the vast number of respondents who declined to answer the vote preference question at all. Fifty percent either said they had no opinion (27 percent), refused to answer (15 percent) or favored “none” of the candidates (8 percent) – higher levels of non-response than on any other question in the survey.
To declare Ahmadinejad comfortably ahead based on these data is to assume that the people who did not express a preference divided precisely the same as those who did answer the question. This theoretical calculation produces a majority for the incumbent. The question is whether such a calculation is justified – and the reality is that even TFT did not make this leap in its pre-election analysis.
Rather it leaped in another direction, noting that “the race may actually be closer than a first look at the numbers would indicate,” because more than six in 10 respondents who expressed no opinion “reflect individuals who favor political reform and change in the current system.” It went on to predict “that none of the candidates will likely pass the 50 percent threshold.”
As today’s claim that this poll reliably indicated a large lead for the incumbent is ill-supported, so is the previous analysis. We’re not told precisely what informed the judgment that “more than 60 percent” of undecideds looked like reformers, whether this included refusals as well as no-opinion respondents, or how robust the conclusion may have been. Nor does today’s piece explain the turnaround from the earlier evaluation.
(4 p.m. update: Note also that TFT's 34-14 percent figure isn't even limited to the 89 percent of respondents who said they intended to vote. Including non-voters in a vote estimate is as inexplicable as the rest of its approach.)
TFT’s op-ed today defends its data by noting that respondents gave “politically risky” responses to other questions, including that many “said they wanted to change the political system to give them the right to elect Iran’s supreme leader.” The phrase “change the political system" in fact was not part of that question, leaving open the possibility of misinterpretation. More to the point, though, is the level of non-response specifically in the vote-preference question.
A critical factor in the accuracy of pre-election polls is that they actually measure voter preferences. There are many ways to encourage likely voters to state a choice – simple steps such as encouraging respondents that their answers are confidential and asking them which candidate they lean towards. The TFT poll’s flaws are evident in the fact that half its respondents didn't express a vote preference at all – a result that makes evaluation of its data in hindsight a highly fraught exercise.