I developed the “Issues and Leaders” model in an effort to aid decision making in political campaigns. The model, which was used to predict the national popular two-party vote in the US presidential elections of 2012 (Graefe Reference Graefe2013) and 2016 (Graefe Reference Graefe2016), is consistent with standard voting theories in assuming that both short-term (issues and candidate evaluations) and long-term (party identification) forces affect people’s vote choice (Campbell et al. Reference Campbell, Converse, Miller and Stokes1980; Clarke et al. Reference Clarke, Kornberg, Scotto, Reifler, Sanders, Stewart and Whiteley2011; Lewis-Beck et al. Reference Lewis-Beck, Norpoth, Jacoby and Weisberg2008). In particular, the model assumes that voters prefer a candidate who they (1) expect to do a better job in handling the issues, and (2) perceive as a stronger leader.
ISSUE-HANDLING COMPETENCE
For an issue to affect vote choice, a voter has to (1) be aware of the issue, (2) consider the issue to be important, and (3) expect one party to do a better job in handling the issue than other parties (Campbell et al. Reference Campbell, Converse, Miller and Stokes1980).
I operationalize the first two conditions using answers to Gallup’s Most Important Problem question (i.e., “What do you think is the most important problem facing the country today?”). After assigning each issue to one of three categories (i.e., economic, foreign, and other), I calculate the normalized percentage of mentions per category to determine their relative importance. As of August 2020, 12% of mentioned issues were economic; 89% were other issues, largely driven by “Coronavirus/Diseases” (35%); “the Government/Poor Leadership” (22%); and “Race Relations/Racism” (10%). None (<1%) of the issues that voters named as most important referred to foreign policy.
The third condition is operationalized using answers to issues-handling questions. These questions ask respondents which candidate they expect to do a better job in handling a particular issue.Footnote 1 For each issue, I calculate the incumbent-party candidate’s two-party support, exponentially smoothing over time by assigning a weight of 70% to the most recent observation.Footnote 2 Then I average incumbent support across all issues in each category (i.e., economic, foreign, and other). At the time of writing, voters favored Trump on economic issues (55% to 45%), whereas Biden was favored on foreign (55% to 45%) and other (59% to 41%) issues.
The final issues-handling score is the incumbent’s average voter support across the three categories, weighted by the categories’ relative importance. I update the issues score whenever new polls on either issue-salience or issues-handling competence become available. As of August 19, Trump’s issue score was 42% versus 58% for Biden.
LEADERSHIP
Elections are choices between candidates, and leadership plays an important role when voters make their choice. To measure candidates’ perceived leadership skills, the Issues and Leaders model draws on survey questions that ask respondents which candidate they think would be the stronger leader.Footnote 3 The leadership score is calculated as the incumbent-party’s candidate two-party support, exponentially smoothed (70%) over time. At the time of writing, Trump’s and Biden’s leadership scores were similar (50%).
MODEL ESTIMATION
I estimate the model’s vote equation using multiple linear regression, with the issues (I) and leadership (L) scores as independent variables and the incumbent-party’s national two-party vote (V) as the dependent variable. Historical data are available across the 12 elections from 1972 to 2016 (Graefe Reference Graefe2020). Using data across the last week before Election Day, the vote equation was as follows:
$$ \quad {\displaystyle \begin{array}{l}\mathrm{V}1=9.3\quad +\quad 0.50\mathrm{I}1\quad +\quad 0.31\mathrm{L}1\quad \mathrm{R}2=0.96;\quad \mathrm{SSE}=1.0\\ {}\quad (1.00)\quad (0.02)\quad (0.02)\quad \left(\mathrm{standard}\ \mathrm{errors}\ \mathrm{in}\quad \mathrm{parentheses}\right)\end{array}} $$
This equation predicts that the incumbent-party’s candidate (i.e., Trump) will gain 9.3% of the two-party vote, plus a variable share depending on his issues and leadership scores. For example, if Trump were able to increase his issues-handling score by 10 points, his vote share would increase by 5 points.
One way to interpret this equation is that the issues-handling and leadership scores capture voters who either (1) cast their vote based on how they view the candidates on issues and leadership, (2) align their views of the candidates with party identification, or (3) use a mix of both strategies. The intercept captures voters who pay no attention to the campaign and whose vote choice therefore is driven solely by long-term forces (e.g., socioeconomics and party identification).
Nowcasting
Because the previous vote equation is estimated from Election Eve data, it can be used for nowcasting. That is, the equation allows for calculating what would happen if the election were held today, given the latest issues-handling and leadership scores. For example, assuming Trump’s latest issues-handling (42%) and leadership (50%) scores were to remain unchanged until Election Eve, the model would predict that he would receive roughly 46% of the two-party vote.
Forecasting
As shown by Graefe (Reference Graefe2013), the relative importance of issues-handling and leadership scores changes during the course of the campaign, which must be considered when making long-term forecasts. For example, the vote equation estimated 12 weeks prior to the election was as follows:
$$ {\displaystyle \begin{array}{l}\mathrm{V}12=22.2+0.32\mathrm{I}12+0.26\mathrm{L}12\quad \mathrm{R}2=0.35;\quad \mathrm{SSE}=4.0\\ {}\quad (4.2)\quad (0.07)\quad (0.05)\quad \left(\mathrm{standard}\quad \mathrm{errors}\quad \mathrm{in}\quad \mathrm{parentheses}\right)\end{array}} $$
That is, about three months before Election Day, perceived issues-handling competence and leadership are of roughly equal importance, albeit less influential compared to the week before Election Day. In comparison, the intercept is much larger. One possible interpretation for this is that fewer voters pay attention to the campaign when the election is still months away. As Election Day nears, short-term forces (i.e., leadership and particularly issues-handling competence) gain importance, whereas the influence of long-term forces (i.e., the intercept) decreases (cf. equation 1).Footnote 4
2020 Forecast
When entering the latest issues and leadership scores into the vote-equation coefficients at the time of writing (i.e., 76 days prior to Election Day), the Issues and Leaders model predicted that Trump would gain 48.6% of the two-party vote, with an 80% empirical prediction interval of ±3.4 percentage points. The chance for Trump to win the popular vote was estimated to be 31%.
DISCUSSION
The Issues and Leaders model predicts the election outcome by trying to capture the short-term forces that affect voter choice—namely, perceptions of a candidate’s issues-handling competence and leadership skills. Needless to say, much can happen during the course of a campaign, and candidates can affect how voters perceive them. For example, Benoit, Hansen, and Verser (Reference Benoit, Hansen and Verser2003) found that presidential debates can affect which issues voters regard as important and which candidate they regard as best suited to addressing certain issues.
The Issues and Leaders model predicts the election outcome by trying to capture the short-term forces that affect voter choice—namely, perceptions of a candidate’s issues-handling competence and leadership skills.
Campaign strategists can use the Issues and Leaders model as a decision aid. For example, concerning agenda setting, the candidates should try to increase the salience of issues that favor them. Voters favor Trump on economic issues; assuming this does not change, the president would be well advised to make the election about the economy. Biden, conversely, is clearly favored on foreign issues—which currently play no role in the public’s mind—as well as on other issues that already are perceived as important.
Candidates also could try to gain “issue ownership” by convincing voters that they are better in solving important problems than their opponent—or at least by closing the gap for issues where their opponent is ahead. That said, issue ownership is heavily intertwined with party constituency and therefore quite stable (Petrocik Reference Petrocik1996). An exception is performance issues, for which issue ownership can change depending on the context of the election. In 2020, the coronavirus appears to be the high-stake performance issue on which candidates should focus.
At the time of writing, Trump had an advantage concerning whom voters regard as better at handling the economy. At the same time, however, voters do not regard the economy as an important issue. With the continuing coronavirus pandemic, it remains to be seen how voters’ perceptions of issue salience and Trump’s issues-handling and leadership skills evolve in the next two months. If Trump manages to keep his advantage in handling economic issues despite the economic downturn—which likely will put those issues back on voters’ minds—the Issues and Leaders model might give him a reasonable chance at winning the popular vote. If, however, voters lose trust in Trump’s ability to handle the economy—perhaps as a result of the downturn—and/or the economy remains of little importance to voters, his chances to win reelection will decrease.
The Issues and Leaders model provides a useful tool to track voter perceptions of issues-handling and leadership skills. Journalists could use the model to assess how issue salience as well as the candidates’ perceived issues-handling competence and leadership skills affect the election outcome. In this way, journalists could move away from simply focusing on the “horse race” (i.e., who is ahead in the polls and by how much) and instead provide answers about why a candidate is ahead.
DATA AVAILABILITY STATEMENT
Replication materials are available on Dataverse at https://doi.org/10.7910/DVN/T6FKLQ.
