Early Polls in Party Nomination Contests: Worth Less, Except When Worth More

• In Event: Political Parties, Perceptions, and Policies

Abstract

I) Introduction
This paper examines the role of early polling in party contests to nominate US presidential candidates. Previous research has established that early nomination contests in Iowa and New Hampshire have had disproportionate effects on results in subsequent states, and on which candidate may win a party's nomination (Bartels 1988; Adkins and Dowdle 2001; Collingwood et al 2012).
Early public opinion polling in these nomination contests has also been shown to predict (indirectly) how well candidates do in early contests, by affecting how much media attention candidates receive prior to, and after, Iowa and New Hampshire. That is, poll results may drive media attention, which, in turn, predicts candidate performance (Donovan and Hunsaker 2009; Redlawsk et al 2011; Donovan et al 2014 ). Our preliminary results in this paper suggest that new DNC rules that qualified candidates for presidential debates based on early poll standing, and the heavy volume of qualifying polling in Iowa and New Hampshire, disadvantaged candidates of color.

II) Early Iowa Polling in Perspective Across States
The substantial role of early polls from Iowa and nationally in presidential nomination contests, and those early contests, may have been enhanced when the Democratic Party opted to use early polls as a criteria to qualify for debates in 2019 and 2020. Democrats gave special status to early 2019 polls from their first four nomination contests (IA, NH, SC, and NV).
This section documents the volume of early polling done in Iowa and New Hampshire, assesses the construct validity of these early polls as a debate qualifier, and assesses the potential bias of early state polls (versus early national polls, and polls from states other than IA, NH, SC and NV). We also discuss the challenges of producing quality estimates from small, idiosyncratic populations (e.g. likely caucus attendees) that are over sampled.

III) Early Iowa Polling in Perspective Across Time
The second section examines a time series of early polls from Iowa (original data collected since 2007 by the authors) as one step in evaluating the predictive validity of early polls, and compares how likely Iowa caucus attendees rated prospective 2020 candidates compared to how they rated candidates in years past.

IV) What Do These Early Polls Predict in Iowa?
The DNC's use of early polls as a qualification standard for multiple presidential debates suggests they expected these polls served as a valid measure of a construct, namely, the depth and / or breadth of public support for a candidate at a given moment in time, and / or as a measure of the candidate's future viability. We report models that estimate the predictive ability of early Iowa polls, compared to early national polls. Early polls from Iowa predict better, but with a substantial amount of variance left unexplained. Early polls, however, may be a better predictor of media attention then how well a candidate actually does in Iowa.

V) Early Polls Predicting Eventual Nominees
This section addresses the question of whether or not leaders in early polls ended up as nominees. We model the predictive utility of early polling in presidential nomination contests by estimating how often early national polling accurately predicted eventual nominees from competitive nomination contests from 1972 to 2016, using data from national polls conducted in the second half of the year prior to the Iowa caucuses. This evaluation of early polling provides criteria for when early national polling has predictive validity. This evaluation of early polling tells us that in a crowded field, where no candidate is over 30%, early polls are worth less, predictively, than when polls show a candidate has solid support and a strong lead.

VI) The Silliness of 'Serious', 'Startling', 'Significant Setbacks' and 'Surprising Surges'
Although the predictive value of these early polls is relatively low, historically they appear to condition how news media reported on results from early contests. We illustrate this qualitatively by examining print headlines and reporting that appeared immediately after various Iowa caucuses since 1976. Day-after-Iowa headlines and reporting regularly discuss "surprising" results that reflect news media expectations set by early polls that had limited predictive validity.


VII) Discussion

We discuss the problems and paradoxes of the media use of early polls in nomination contests, and problems and paradoxes associated with Democrats granting special status to very early polls from Iowa and New Hampshire.

Select References (deleted, over word count)

Authors

• Todd Donovan, Western Washington University

• David P. Redlawsk, University of Delaware

• Caroline J. Tolbert, University of Iowa