Forecasting Elections

This is a page for students of PS 240. On October 29th and 31st we discussed election forecasting.

Possible Exam Questions

1. 1 What are the "keys"?

Readings

L. Arcuri, Castelli, L., Galdi, S., Zogmaister, C. and Amadori, A. (2008). Predicting the Vote: Implicit Attitudes as Predictors of the Future Behavior of Decided and Undecided Voters. Political Psychology, 29, 369-387.

Summary: This article focused on one method that can be used to predict how undecided voters will ultimately vote. Findings in the study report that despite what undecided voters say about their lack of preference to one candidate or another people have implicit attitudes towards these candidates. Such attitudes that undecided voters would not verbally express are good predictors of future voting behavior. The study used an Implicit Associations Test (IAT) to measure implicit attitudes. IATs measure how quickly people respond to associations between two things. For example, the test might pair a picture of one of the candidates with negative words (war, death, evil). A quicker reaction by the respondent indicates his or she already associates that candidate with 'bad' things even though the person never explicitly indicated this attitude.

M.A. Barreto, Streb, M. J., Marks, M. and Guerra, F. (2006). Do Absentee Voters Differ from Polling Place Voters?: New Evidence from California. Public Opinion Quarterly, 70, 224-234.

Summary: In 2003 California gubernatorial recall election 2,775,785 absentee ballots were cast, which represented roughly 30 percent of the state wide vote. Because the number of absentee voters is growing in California and across the nation, this article found out that the absentee voters, at least in California, are an older and a more educated demographic. The also set out to find out whether absentee voters, have different voting preferences. They found no significant difference between the two groups voting preferences. They found that Republicans were not more likely to vote absentee than Democrats; and that it is just easier, and more convenient.

The Goal of this article was to find out if absentee voters were different from those that voted on election day. Several days before the election, absentee voters were surveyed to see why they voted absentee, along with partisan, identification, ideology, demographics and other questions. The results of the experiment showed that absentee votes are not extremely different from the states voters as a whole. Absentee voters were also found to be older and better educated.

R.S. Erikson and Wlezien, C. (2008). Are Political Markets really Superior to Polls as Election Predictors? Public opinion quarterly, 72, 190-215.

Summary: This article argues that election markets are better predictors of elections because people have to put something at stake. The markets do not get effected by the "convention bounce," whereas polls do. This is because polls measure voter sentiment rather than the likelihood of someone actually getting elected. Polls are far more volatile than the markets as well--the markets typically hold stable above or below polls for an extended period of time. With the exception of the 1988 election, the markets preformed better than the polls in the predictions of elections and were closer to the outcome of the election 74% of the time. This can at least be partially attributed to people who invest in the markets wanting to find the most information possible to make the choice that will generate a profit.

Election markets are better able to predict elections than "trial-heat" polls. This article attempts to challenge this idea, using the Iowa Electronic Market. The IEM is specifically used for education and research purposes. The paper states that it is wrong for markets to be compared to polls because polls reflect public opinion on the day of the poll. WIns based on polls dominated those from winner take-all markets. When the leads in polls are discounted in the poll.

M.S. Lewis-Beck (2005). Election Forecasting: Principles and Practice. British Journal of Politics & International Relations, 7, 145-164.

Summary:

From the International Journal of Forecasting

Campbell and Lewis-Beck. An introduction.

Summary: First, this article talks about the history of election forecasting. It then goes on to summarize the predictions of the other articles predictions. all of which predict a democratic victory.

Abramowitz. Time-for-change model.

Summary:Basically, this article theorizes that the longer one party is in power, the less likely it is to get elected. As such, it predicted an Obama win in 2008.

Berg, Nelson, and Rietz. Prediction market accuracy.

Summary: This article argues that election markets are better predictors of elections because people have to put something at stake. The markets do not get effected by the "convention bounce," whereas polls do. This is because polls measure voter sentiment rather than the likelihood of someone actually getting elected. Polls are far more volatile than the markets as well--the markets typically hold stable above or below polls for an extended period of time. With the exception of the 1988 election, the markets preformed better than the polls in the predictions of elections and were closer to the outcome of the election 74% of the time. This can at least be partially attributed to people who invest in the markets wanting to find the most information possible to make the choice that will generate a profit.

The article discussed "prediction markets" which been conducted for nearly 20 years but have never been used in experiments or the efficiency of their predictors. When the market predictions were compared with polls on the Presidential elections since 1988, the market is close to the outcome 74% of the time. Markets are also better at predictions over 100 days before the election. The article brings up 6 different ways that elections are predicted- 1) naive forecasting, 2) polls, 3) prediction markets, 4) structural models, 5) time series models, and 6) methods using focus groups, interviews of knowledgeable parties and expert panels. The article brought up that selecting a truly random sample is different because of different factors but mainly because many voters do not have land lines. Evidence shows that election prediction is very accurate in the short term, very close to the election.

Erikson and Wlezien. Economic indicators.

Summary:This article talks about how the economy is indicative of who will win the election. However, elections are more responsive to short term economic shocks, and so to predict the election early, one must first predict the future economy. This article is an update of a previous model, just showing that this model was correct in 2004, making it correcrt 14 election in a row. A very interesting point of the model is weighing the quarters, such that the further away a quarter is from the election, the less influence it has on the election.

Lewis-Beck and Tien. Changing the model.

Summary:

Lichtman. The keys.

Summary: The keys is a binary forecasting model that does not use any polling data, instead it uses a wide range of indicators. They keys were created in 1981, and have correctly predicted every election from 1860-2008. They keys do not predict the winner of the election, but the popular vote winner. The keys are: Party Mandate, Party Contest, Incumbency, Third Party, Short-Term Economy, Long-Term Economy, Policy Change, Social Unrest, Scandal, Foreign or Military Failure, Foreign or Military Success, Incumbent Charisma/Hero, Challenger Charisma/Hero. The economic keys, particularly the short-term economy key, could be argued that it is the most important key for the incumbency. No incumbent party has won during a recession.

Sidman, Mak, and Lebo. Non-incumbent elections.

Summary: This article looks at the 2000 election and what caused the forecasting models to be incorrect and whether the election was simply atypical. The variables that were looked at were incumbent vote share, third term, approval, economic indicators, and economic evaluations. In unweighted these variables, the authors found that by doing so the outcome would be closer. However, in using the models that best predicted the 1956 and 1996 elections the models were found to do a poor job at predicting the 2000 election. The authors use this as evidence that the 2000 election was unusual, not that the models were made improperly. The findings conclude that, on average, when ignoring whether or not an incumbent is in the election, the model preforms better.