Justin Wolfers and David Rothschild are coming out with a paper called “Forecasting Elections: Voter Intentions versus Expectations.” In it, they explain that voters predict election outcomes more accurately when they are asked who they think will win, as opposed to who they intend to vote for. The authors’ takeaway?
The answers we receive from the expectation question are about as informative as if they were themselves based on a personal poll of approximately twenty friends, family, and coworkers. This “turbocharging” of the effective sample size makes the expectation question remarkably valuable with small sample sizes. Moreover, because our model gives insight into the correlation between voting expectations and intentions, even samples with a strong partisan bias can be used to generate useful forecasts.
The key insight from our study—that analysts pay greater attention to polls of voter forecasts—in fact represents a return to historical practice. In the decades prior to the advent of scientific polling, the standard approach to election forecasting involved both newspapers and business associations writing to correspondents around the country, asking who they thought would win. We are in many respects, recommending a similar practice. Having shown the usefulness of this approach for forecasting elections, we hope that future work will explore how similar questions can be used to provide better forecasts in a variety of market research contexts from forecasting product demand to predicting electoral outcomes, to better measuring consumer confidence.
I’ve noticed before that polling analysis usually fails to take into account the wording used in the question. Assuming Wolfers’ and Rothschild’s reasoning bears out, this will be a very useful start to rethinking exactly how pollsters should frame their questions if they want to achieve the most accurate results (which, quite frankly, is not the objective of many organizations that conduct polls).
- The right question to ask: Who do you think will win the election? (aei-ideas.org)
- ‘For Whom Will You Vote?’ May Be Wrong Question (wnyc.org)
- Eric Zuesse: Revolutionary Study Shows Romney’s Chances Are Overestimated In The Polls (huffingtonpost.com)
- A Better Poll Question: Who Do You Think Will Win? (nytimes.com)
- Useful poll results depend on good questions. Which are the right questions? (prairieweather.typepad.com)
- Obama Will Win, Say Expectation Polls – Bloomberg (bloomberg.com)
- Study: Whom voters want to win is less predictive than whom they expect to win (washingtonpost.com)
- Who We Expect To Win Is A Better Predictor Than Who We Want To Win (underpaidgenius.com)