How Hard Can It Be to Predict An Election?
November 6, 2012
Bailey Hall 207
Refreshments will be served in Bailey Hall 204 at 4:30
All of us have been bombarded recently by advertisements for political candidates (or more accurately, ads against their opponents). In addition, we have no doubt heard of various polls showing President Obama up or down a few points relative to Governor Romney, with perhaps some confidence level or “margin of error” mentioned in the fine print. Yet, the major independent polling agencies, such as Gallup and Rasmussen, haven’t called a winner at the time of writing this abstract. Why not? Thousands of polls have been conducted, and millions of opinions obtained. How difficult can it be to predict a presidential election? Very difficult, it turns out! In addition to the standard statistical question of how much uncertainty there might be in a poll of a thousand people, it turns out that there are also complex data quality issues. For example, who should Gallup poll? Should they poll registered voters, or registered voters who have actually voted in the past few elections (“likely voters”)? How should they find such people, especially those with unlisted telephone numbers? How should they count people who refuse to reveal their voting intentions? Could some people even lie about who they are voting for? And of course, we haven’t even gotten to that wonderful invention called the Electoral College. During this seminar I will highlight some key points in the history of election polling, beginning with a few disastrous miscalculations, not all of which are in the distant past (does anyone remember the 2000 election being called for Al Gore by the major TV networks?). I will briefly explain the theory of how a poll of 1,000 people can produce a reasonably accurate prediction of how 150,000 million people will vote, and then explain why the theory has so many practical challenges, primarily due to data quality.
|Union College Math Department Home Page|
Comments to: email@example.com
Created automatically on: Sun Jan 21 17:39:29 EST 2018