See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/262188489

US Presidential Election Forecasting

Article  in  Political Science and Politics · April 2014

DOI: 10.1017/S104909651400002X

CITATIONS READS
27 554

2 authors:

Michael Lewis-Beck Mary Stegmaier

University of Iowa University of Missouri
292 PUBLICATIONS   10,918 CITATIONS    55 PUBLICATIONS   2,033 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

MAPLE - Mapping and Analysing the Politicisation of Europe before and after the Eurozone Crisis View project

Forecasting elections in democraties View project

All content following this page was uploaded by Mary Stegmaier on 01 January 2016.

The user has requested enhancement of the downloaded file.

 SYMPOSIUM

US Presidential Election
Forecasting
(e.g., Abramowitz, Campbell, Lewis-Beck and Stegmaier,
Introduction Norpoth) estimate, via standard regression techniques, sin-
gle-equation explanatory voting models at the national level
Michael S. Lewis-Beck, University of Iowa of analysis. Commonly, these models begin with a core politi-
Mary Stegmaier, University of Missouri cal economy explanation, something like vote = f (presidential
popularity, economic growth). Generally these models offer a
unique, final preelection forecast. Aggregators (e.g., Berg and

T
he 2012 United States presidential contest Rietz, Blumenthal, Jackman, Traugott ) examine vote inten-
ushered in a revolution in election forecasting. tion directly (or indirectly) through national opinion data.
While serious efforts to forecast American elec- A leading example, that of Real Clear Politics, summarizes the
tions have been around for more than 30 years, preferences from likely voters, over multiple polls. While
suddenly things have changed. Competing news these poll results are intended by the polling houses as snap-
agencies and election prediction websites proliferated to satisfy shots of opinion at the moment, they are frequently used by
the public’s appetite for forecasts during the campaign. And in election watchers to aid in election prediction, as Blumenthal
terms of forecasting approaches, a new generation of dynamic discusses. Jackman’s model-based poll aggregation approach
modeling has emerged. exemplifies this innovation.2 Taking a different slant from
The elevated profile of election forecasting offers us the the polls themselves, the Iowa Electronic Markets summarize
opportunity to consider what this means for the credibility, the election predictions of market traders. These Aggregators
theory, and ultimately the future of election forecasting. Early offer repeated forecasts during the campaign. Both of these
election prediction models were met with the criticism that approaches—Structuralist or Aggregator—base their infer-
such forecasts were simply fun and games, not “real” politi- ences on quantitative methods.
cal science, although these models were based on established Synthesizers combine properties of Structuralists and
election theory, public opinion polling techniques, and econo- Aggregators. That is, they begin with an explanation in politi-
metric estimation (Fair 1978; Lewis-Beck and Rice 1982, 1984; cal economy form, and embed aggregated and updated polling
Rosenstone 1983; Sigelman 1979). Since the publication of preferences. The data, analyzed either at the national level
these seminal works, model modifications have put our estab- (e.g., Erikson and Wlezien) or the state level (e.g., Linzer),
lished election theories to the test. Through this process, we are subjected to rigorous quantitative modeling. These
have learned much.1 With these advances, and the increased models bring together election theory and the powers of
demand for forecasts from campaigns and news consum- aggregation and dynamic updating. A similar approach
ers, election forecasting finally is gaining the respect that was widely followed in the run-up to the 2012 presidential
it deserves. election in the media examples from Nate Silver at the New
In this symposium, we offer 16 articles that tackle the task York Times.
of election prediction. These pieces, written by leaders in the The foregoing forecasting approaches are distilled by
fields of election forecasting and commentary, are accessible thoughtful campaign observers (e.g., Cook and Wasserman,
presentations that examine a particular method or problem. The Rothenberg), who effectively act as Judges. This judging does not
approaches to forecasting represented here can be grouped into necessarily remain inside a positivist quantitative framework.
four types: Structuralists, Aggregators, Synthesizers, and Judges. These experts go further, weighing the sometimes conflicting
Next, we look at these forecasting types in practice. Then, we claims of the polls, models, and markets, putting in their own
explore advances and obstacles in forecasting theory, and end admittedly qualitative assessment of the horse-race and follow-
with how that bears on election theory. ing their own rules of thumb. In this way, they promise added
value like the local weather forecasters who use their exper-
APPROACHES tise of local conditions and patterns to adjust their forecasts
The four forecasting types drawn on here can be distinguished against those of the Numerical Weather Prediction models
by their uses of theory, data, time, and inference. Structuralists (Novak et al. 2011).

284 PS • April 2014 © American Political Science Association, 2014 doi:10.1017/S104909651400002X

ELECTION FORECASTING THEORY: ADVANCES AND and other forms of combining (such as poll averaging) mask
OBSTACLES the problem of the differential quality among the models or
The newest election forecasting models, exemplified by Linzer’s polls. Rothenberg and Traugott, in their commentaries, raise
work, look more like theoretically and technically sophisticat- the particular issue of poll quality, Rothenberg with respect
ed physical science forecasting models, such as those used in to partisan polls, Traugott with respect to interactive voice
meteorology (Lewis-Beck and Stegmaier). They are based response (IVR) polls.
on a political economy theory of election outcomes, a theory The idea of combining either models or polls raises the other
tested against massive amounts of geographically appropriate evaluation issues—parsimony and replication (i.e., transpar-
observations (i.e., on the states), with these data and their pre- ency). Take parsimony first. The meaning of a parsimonious
dictions updated until the election occurs. In the 2012 presi- model becomes opaque when the predictions of many models
dential contest, such models correctly forecasted the Electoral or polls are averaged, especially if the unit of analysis is the
College winner in all but one or two states. state. When the unit of analysis is the nation, as used to be
How should these models, and others, be evaluated as fore- routine, the parsimony question had an easier answer. For
casting instruments? In the literature, we earlier offered the one, these earlier models were based on such a small sample
following evaluation criteria: accuracy, lead, parsimony, and that parsimony was a practical necessity. One encouraging
replication (Lewis-Beck 2005). The work of the Aggregators technique, which may allow more clarity and parsimony at
and Synthesizers, with their frequent updating, make clear both state and national levels of analysis, is the uniform swing
that the word “dynamic” should be added to the list of crite- idea, as applied by Jackman.
ria for several reasons. For one, as Sides remarks, updating Turning to the problem of replication, the issue of opacity
the forecast “engages the campaign narrative.” For another, a becomes greater. In particular, it is impossible for an interested
forecasting instrument works better, to the extent that it can investigator to replicate the results of a proprietary (i.e., clas-
be updated on a regular (even daily) basis up to Election Day. sified) poll or model. This lack of transparency undercuts a
This more or less continuous release of forecasts from one canon of scientific research. As Linzer remarks, statistical

The idea of combining either models or polls raises the other evaluation issues—parsimony
and replication (i.e., transparency). Take parsimony first. The meaning of a parsimonious
model becomes opaque when the predictions of many models or polls are averaged, especially
if the unit of analysis is the state.

overarching model has been labeled by some as nowcasting models are based on assumptions whose validity can only
(Lewis-Beck and Stegmaier). be evaluated if the model and its operations are made
What about the first four evaluation criteria? In the popular known.
mind, accuracy looms as most important. Updating, combining Besides these difficulties, other issues relate to replica-
polls, using state-level measures, are all techniques that have tion, and, in particular, data. Accuracy may heavily rest on
helped improve accuracy. But, as the Campbell article suggests, the availability of a sufficient number of reliable polls at the
accuracy alone is not enough. To take the extreme case, while state or national level. But, as Blumenthal observes, the num-
a poll of voters exiting the voting booth might be highly accu- ber of available state polls decreased from 2008 to 2012, and
rate, it can only tell us something we will know in a few hours. many forecasters fear that the number might further decrease
The intrinsic attraction of forecasting comes from its ability to as polling aggregation increases (for it is a much less costly
see into the future, when the future stands far away. Blumen- forecasting strategy). If polls remain plentiful, the problem
thal, in his article, argues that more focus on the accuracy of of their representativeness as voter samples persists, accord-
early polls is needed. With respect to a specific time horizon, ing to many of the articles in this symposium. In particular,
Linzer emphasizes the need to generate early forecasts, per- Blumenthal asks why different polls may converge on the
haps three to four months before Election Day. In this regard, “right prediction.” Is aggregate voter opinion more stable, or
Erikson and Wlezien, and Lewis-Beck and Stegmaier, tout the are the polling houses adjusting their final forecasts toward
forecasting ability of early campaign perceptions of national central values?
economic conditions.
In the 2012 US presidential election, all the leading ELECTION THEORY: LESSONS FROM PRESIDENTIAL
approaches to forecasting generally “got it right,” at least ELECTION FORECASTING
in the rough sense that, collectively, they forecast an Obama Sometimes, election forecasting can appear to be a limited
win. Part of that collective accuracy was due to the rising enterprise. For example, as Abramowitz notes, if interested
practice of ensemble forecasting, wherein the forecasts from citizens simply predicted that each state in 2012 would vote for
different models are averaged, as was done in the pre-2012 its party choice in 2008, they would have been correct for 48
election forecasting symposium published in PS: Political Sci- out of 50 states. In other words, no fancy equations, surveys,
ence and Politics (Campbell 2012). But ensemble forecasting or models were needed to pick Obama as the presidential

PS • April 2014 285

 Sy m p o s i u m : U S P r e s i d e n t i a l E l e c t i o n Fo r e c a s t i n g

winner. But election forecasting is not always so easy for sophisticated modeling providing accurate, long-range work.
many reasons, as Campbell discusses. For one, we may be The accuracy level, while high, is not perfect and never can
interested in point forecasts of the popular vote margin (in be. Error will always remain, and some contests will be fore-
the state or the nation). For another, presidential elections cast incorrectly. However, this error may be reduced by careful
recently have become very close, making them harder to attention to the more qualitative elements in the race, elements
forecast. Therefore, in the long run, theory becomes more that go beyond the usual quantitative strictures. Finally, con-
important. Indeed, Sides argues that the forecasting exercise siderable accuracy can generally be achieved at some tempo-
itself tests election theory. ral distance from Election Day. A trade-off exists between

Can sufficient accuracy be obtained weeks, even months before voting day? This
question—that of the optimal lead—stands as an important next question to be solved
in this burgeoning field.

It seems valuable, then, to ask what forecasting has taught accuracy and lead. At some point the gains in accuracy may
us, as political scientists, about election theory. What have we not offset the costs in lead. Can sufficient accuracy be obtained
learned about the behavior of American voters in presidential weeks, even months before voting day? This question—that of
elections? Here we list five propositions: the optimal lead—stands as an important next question to be
solved in this burgeoning field.
I. Electoral cycles exist. As Norpoth shows, the incumbent
party will generally only hold the White House for two, ACKNOWLEDGMENTS
maybe three terms. Further, first-term incumbent parties We would like to express our gratitude to Bill Jacoby and the
are most advantaged, as Campbell and Norpoth observe. ICPSR Summer School at the University of Michigan for
After that, the costs of ruling increase dramatically. hosting the roundtable “Presidential Election Forecasting:
II. Campaigns influence the electoral outcome. This influ- Frontiers and Controversies,” where a group of the contributors
ence comes in obvious and less-obvious ways. In partic- shared and discussed ideas on the future of election forecast-
ular, it is conditioned by how candidates use economic ing. We also appreciate Drew Linzer’s thoughtful comments
information (strategically or not) to win votes, as Vavreck on this introduction. Q
demonstrates.
III. The economy matters a great deal in the voter’s elector-
NOTES
al calculus. Further, with respect to national econom-
1. A number of excellent sources exist for readers who are interested in learning
ic performance, trends matter more than absolutes more about the development of the election forecasting field, the methodol-
(Vavreck). Also, economic effects manifest themselves ogy of forecasting, and election prediction in other democracies: Lewis-Beck
and Rice 1992; Lewis-Beck and Tien 2011; Jones 2002; Stegmaier and Norpoth
with a time lag (Erikson and Wlezien, Lewis-Beck and 2013.
Stegmaier). Finally, economic perceptions count and can 2. For a synopsis of Jackman’s Model-Based Poll Averaging approach, see:
count even more than the economic facts (Lewis-Beck and http://www.huffingtonpost.com/simon-jackman/modelbased-poll-
averaging_b_1883525.html.
Stegmaier, Vavreck).
IV. Voters are retrospective, and myopic. As Mayer points
out, voters base their incumbent assessments largely on
REFERENCES
past performance, and they form that assessment roughly
Campbell, James E. 2012. “Forecasting the 2012 American National Elections.”
from events of the last year. PS: Political Science & Politics 45: 610–13.
V. Voter opinion cannot be easily swayed. The forecasts tend
Fair, Ray C. 1978. “The Effect of Economic Events on Votes for President.”
to show considerable inertia in candidate preference (day The Review of Economics and Statistics 60: 159–73.
after day, month after month), contrary to the expectation Jones, Randall J. 2002. Who Will Be in the White House: Predicting Presidential
of many journalists, as Mayer observes. Moreover, accord- Elections. New York: Longman.
ing to Dickinson, the media tend to exaggerate the impact Lewis-Beck, Michael S. 2005. “Election Forecasting: Principles and Practice.”
British Journal of Politics and International Relations 7: 145–64.
of candidate personality and campaign tactics.
Lewis-Beck, Michael S., and Tom W. Rice. 1982. “Presidential Popularity and
Presidential Vote.” Public Opinion Quarterly” 46: 534–37.
While these propositions are not incontrovertible, they appear
———. 1984. “Forecasting Presidential Elections: A Comparison of Naive Models.”
to rest on a solid empirical base developed from the repeated Political Behavior 6:9–21.
ex-ante forecasting by different research teams on United States ———. 1992. Forecasting Elections. Washington, DC: Congressional Quarterly
presidential elections since 1980. Press.
Lewis-Beck, Michael S., and Charles Tien. 2011. “Election Forecasting.”
CONCLUSION In The Oxford Handbook of Economic Forecasting, ed. Michael Clements
and David Hendry, 655–71. Oxford: Oxford University Press.
US presidential election forecasting has firmly established Novak, David R., Chris Bailey, Keith Brill, Michael Eckert, Dan Petersen,
itself as a scientific forecasting enterprise that is capable of Robert Rausch, and Michael Schichtel. 2011. “Human Improvement to

286 PS • April 2014

 Numerical Weather Prediction at the Hydrometeorological Prediction Sigelman, Lee. 1979. “Presidential Popularity and Presidential Elections.” Public
Center.” 91th Annual AMS meeting, AMS 24th conference on weather Opinion Quarterly 43: 532–34.
and forecasting/20th conference on numerical weather prediction,
Stegmaier, Mary, and Helmut Norpoth. 2013. “Election Forecasting.” In Oxford
Paper No. 440.
Bibliographies in Political Science, ed. Rick Valelly. New York: Oxford
Rosenstone, Steven J. 1983. Forecasting Presidential Elections. New Haven, CT: University Press. http://www.oxfordbibliographies.com/view/document/
Yale University Press. obo-9780199756223/obo-9780199756223-0023.xml

SYMPOSIUM CONTRIBUTORS

Alan I. Abramowitz is the Alben W. Barkley Professor Congress, and the executive branch. His current book and many other articles on public opinion, voting
of Political Science at Emory University. He has manuscript, The President and the White House Staff: and elections, media and politics, and the presidential
authored or coauthored six books, dozens of contribu- People, Positions, and Processes, 1945–2012, examines nomination process. He can be reached at w.mayer@
tions to edited volumes, and more than 50 articles in the growth of presidential staff in the post–World War neu.edu.
political science journals dealing with political parties, II era. He can be reached at [email protected].
Helmut Norpoth is a professor of political science at
elections, and voting behavior in the United States. He
Robert S. Erikson is professor of political science at Stony Brook University. He is coauthor of The American
is also one of the nation’s leading election forecasters.
Columbia University. His research on American politics Voter Revisited and has published widely on topics
Abramowitz’s most recent book, The Polarized Public:
and elections has been published in a wide range of of electoral behavior. His current research focuses
Why American Government Is So Dysfunctional
examines the causes and consequences of growing par-
scholarly journals, and he is coauthor of The Timeline of on public opinion in wartime. He can be reached at
tisan polarization among political leaders and ordinary
Presidential Elections (Chicago), The Macro Polity [email protected].
(Cambridge), Statehouse Democracy (Cambridge), and
Americans. He can be reached at [email protected]. Thomas A. Rietz is a professor in the department of
American Public Opinion (Pearson). He is a former
Joyce E. Berg is a professor in the Department of editor of the American Journal of Political Science finance, Henry B. Tippie College of Business, University
Accounting, Henry B. Tippie College of Business, and Political Analysis. He can be reached at rse14@ of Iowa. He has been working with the Iowa Electronic
University of Iowa. She has been working with the Iowa columbia.edu. Markets project since 1993 and is currently a member
Electronic Markets project since 1992 and is currently of the steering committee. His recent research has
Simon Jackman is a professor in the department of appeared in Quantitative Economics, Games and
the director. Her recent research has appeared in
political science, Stanford University. He has worked Economic Behavior, Management Science, and The
Neuropsychologia, Quantitative Economics, Games
on state-level, poll-averaging models since the 2000 US Proceedings of the National Academy of Science. He
and Economic Behavior, and Management Science.
presidential election. In 2012 his poll-averaging models can be reached at [email protected].
She can be reached at [email protected].
were used by the pollster section HuffingtonPost.com.
Mark Blumenthal is the senior polling editor of the He is a former president of the Society for Political Stuart Rothenberg is editor and publisher of the
Huffington Post and the founding editor of the site for- Methodology, the author of Bayesian Analysis for the Rothenberg Political Report, a nonpartisan newsletter
merly known as Pollster.com, now HuffPost Pollster. He Social Sciences, and one of the principal investigators that reports on and handicaps US House, Senate, and
has been writing about polls and their methodology since of the American National Election Study. He can be gubernatorial campaigns and elections. A former political
launching the MysteryPollster blog in 2004. Blumenthal contacted at [email protected]. analyst for CNN, CBS News, and the News Hour on
also worked in the political polling business for more PBS, he is a columnist for Roll Call. He holds an under-
Michael S. Lewis-Beck is F. Wendell Miller Distin- graduate degree from Colby College and a PhD in politi-
than 20 years, conducting surveys on behalf of Demo-
guished Professor of Political Science at the University cal science from the University of Connecticut. He can
cratic candidates and market research for major corpora-
of Iowa. His interests are comparative elections, election be reached at [email protected].
tions. He can be reached at mark@huffingtonpost.com.
forecasting, political economy, and quantitative
James E. Campbell is a UB Distinguished Professor methodology. Professor Lewis-Beck has authored or John Sides is an associate professor in the department
of Political Science at the University at Buffalo, SUNY. coauthored more than 225 articles and books, including of political science at George Washington University.
He is the author of three books and more than 80 journal Forecasting Elections, Economics and Elections and His work focuses on political behavior in American
articles and book chapters. He previously served as Applied Regression. He has served as editor of the and comparative politics. He is the author, with Lynn
Chair of the Political Forecasting Group, President of American Journal of Political Science and of the Sage Vavreck, of The Gamble: Choice and Chance in the
Pi Sigma Alpha, an APSA Congressional Fellow, and QASS series (the green monographs) in quantitative 2012 Election. He can be reached at [email protected].
an NSF program director. He has edited forecasting methods. Currently he is associate editor of International
Mary Stegmaier is teaching assistant professor in the
symposia in each of the last five presidential elections. Journal of Forecasting and data editor of French
Truman School of Public Affairs at the University of
He can be reached at jcampbel@buffalo.edu. Politics. In spring 2013, Professor Lewis-Beck was visit-
Missouri. Her recent research on voting behavior,
ing scholar, Centennial Center, American Political Science
Charles E. Cook, Jr. has been editor and publisher of elections, and political representation in the United
Association, Washington, DC. He can be reached at
the Cook Political Report since its founding in 1984. States and abroad has appeared in Electoral Studies,
[email protected].
He has worked as a columnist for Roll Call (1986– Political Science Research and Methods, Public
1988) and for National Journal (1998–present). He Drew A. Linzer, in 2012, published the forecasting Choice, The Journal of Elections, Public Opinion,
has been a political analyst for NBC News since 2002. website votamatic.org, which offered state-by-state poll and Parties, and Parliamentary Affairs. She can be
Prior to 2002, Charlie had been a consultant and/or tracking and predictions of the US presidential election. reached at [email protected].
analyst for CNN and CBS News. He can be reached at His research has appeared in the Journal of the
Michael W. Traugott is professor of communication
[email protected]. American Statistical Association, Political Analysis,
studies and political science and a senior research
American Political Science Review, Journal of Politics,
Matthew J. Dickinson is professor of political sci- scientist in the Center for Political Studies at the Insti-
World Politics, and the Journal of Statistical Software.
ence at Middlebury College. His blog on presidential tute for Social Research at the University of Michigan.
From 2008 to 2013, Linzer was assistant professor of
power can be found at http://blogs.middlebury.edu/ He is an editor of the International Journal of Public
political science at Emory University. He can be reached
presidentialpower. He is author of Bitter Harvest: Opinion Research, the Poll Review Section of Public
at [email protected].
FDR, Presidential Power, and the Growth of the Opinion Quarterly, and the invited editor of the Public
Presidential Branch (1999), the coeditor of Guardian of William G. Mayer is a professor of political science at Opinion Quarterly special issue on the 2012 election
the Presidency: The Legacy of Richard E. Neustadt, Northeastern University. He is the author of the first in the United States. He has been the president of the
and has published numerous articles on the presidency, published forecasting model for presidential nominations American Association for Public Opinion Research

PS • April 2014 287

 Sy m p o s i u m : U S P r e s i d e n t i a l E l e c t i o n Fo r e c a s t i n g

(AAPOR), the World Association for Public Opinion Gamble: Choice and Chance in the 2012 Presidential 2012, 2010, and 2008. He can be reached at dwasserman@
Research (WAPOR), and the Midwest Association Election, portions of which she and John Sides wrote cookpolitical.com.
for Public Opinion Research (MAPOR). In 2010 he and released in real time during the 2012 election. She
Christopher Wlezien is Hogg Professor of Govern-
received the AAPOR Award for Distinguished Lifetime is the originator of the Cooperative Campaign Analysis
ment at the University of Texas at Austin. His research
Achievement. He is currently consulting with the Gal- Project and cofounder of the Model Politics blog. She
on American and comparative politics has appeared in
lup Organization on a review of their 2012 preelection can be reached at [email protected].
numerous journals, and he is coauthor of Degrees of
polling methodology. He can be reached at mtrau@
David Wasserman is House editor of the Cook Democracy (Cambridge) and The Timeline of Presi-
umich.edu.
Political Report, where he is responsible for handi- dential Elections (Chicago) and coeditor of a number
Lynn Vavreck is associate professor of political science capping and analyzing US House races. He also of other books, including Who Gets Represented?
and communication studies at University of California, serves as an associate editor of National Journal (Russell Sage). He was founding coeditor of The Journal
Los Angeles. She has published three books on presi- magazine and a contributor to the Almanac of of Elections, Public Opinion and Parties and currently
dential campaigns including The Message Matters: American Politics 2014, and has served as an analyst is associate editor of Public Opinion Quarterly. He
The Economy and Presidential Campaigns and The for the NBC News Election Night Decision Desk in can be reached at [email protected].

288 PS • April 2014

View publication stats