Politicians, whose job it is to learn about public opinion, should know more about public opinion than does the general public. In this study, they did. The politicians made better predictions of the outcomes of the votes for two controversial referendum issues in Oregon in 1982. (The issues dealt with land use planning and property tax limitations.) This result is not surprising. The general public is best surveyed for their intentions on issues, not for their opinions as to which position will win.
The paper provides support for the ‘looking-glass effect.’ This states that when one looks out to see the world, a reflection of one’s own beliefs is seen. Others are seen to have the same beliefs. I am not sure where this term originated, but evidence on the effect goes back to at least Hayes (1936); thus, of those who voted ‘yes’ on one of the issues examined by Lemert, 74% predicted that it would pass. Of those who voted ‘no,’ only 19% predicted that it would pass. (The record of voting was obtained from interviews with voters as they left the voting booths.) These results are almost as extreme as those obtained by Hayes. The looking-glass effect is not surprising either, but it does indicate the dangers in trying to use participants in a process as experts about that process.
Another interesting aspect of the study is that people often violate the ‘law of conservatism for forecasting.’ This states that the greater the uncertainty you face, the more conservative you should be when making a forecast. For example, if you were asked to make a forecast about a vote when there are two possible outcomes, you should use the minimum criterion and select a 50-50 split if uncertainty is extremely high. (This minimizes the largest error that you could make.) But Lemert found that voters, who face much uncertainty about the intentions of other voters, made much more extreme prediction than did the politicians on the tax issue, none of the politicians fell outside the 40-60% range, while 30% of the voters’ prediction fell outside.