Motivation

Over the past twenty years, political campaign expenditures in the United States have skyrocketed. In 1998, these expenditures amounted to $1.6 billion. In 2018, this number is projected to hit $5.2 billion (Center for Responsive Politics). This effect is often credited to the controversial 2010 Citizens United decision, which effectively prohibited the United States government from restricting independent expenditures for political communications. However, this upward trend in expenditures began long before 2010. Importantly, we can infer from this pattern that political professionals see a direct causal relationship between campaign expenditures and electoral outcomes.

Many prior studies of campaign spending are limited in scope, and often concern elections in the 1970s and 1980s. Additionally, the preponderance of existing scholarship focuses mainly on the effects of incumbency. In this study, I examine the shape and the form of the relationship between campaign expenditures and campaign efficacy with several sets of controls in place. I use a comprehensive set of data encompassing all House of Representatives elections between 2006 and 2016.

I look to develop a comprehensive picture of campaign spending in the 21st century through two main avenues. First, I analyze the overall relationship between spending and voting as denominated in dollars and votes. Second, I examine competitive dynamics within districts through the analysis of candidate spending share, which is defined as the candidate’s spending divided by all spending by candidates in the district. In this branch of my analysis, I compare spending share to vote share.

Literature Review

The first body of scholarship consulted was the literature surrounding the relationship between campaign spending and electoral outcomes. This work has mainly focused on the effects of incumbency. In 1978, Gary Jacobson established that the marginal utility of campaign spending is different for incumbents than it is for challengers. Challengers often begin their campaigns in relative obscurity and can quickly boost their public profiles through high levels of campaign spending. Incumbents, on the other hand, can leverage their established political networks and extensive communication resources to develop name recognition and support outside of the formal campaign spending process. In 1985, Jacobson argues that “no matter how the data are analyzed, no matter what view one takes of the simultaneity problem, one finding remains undisturbed: incumbents gain nothing in the way of votes by spending money in campaigns.”

This conclusion has been debated extensively. In their 1988 paper, Green and Krasno argue that Jacobson failed to control for challenger quality, pertinent interaction effects, and reciprocal causation. Green and Krasno incorporate prior incumbent spending as an exogenous variable to identify the incumbent’s general propensity to spend. Green and Krasno arrive at the conclusion that incumbents can readily affect their vote share by spending on the margin. In 1990, Jacobson responded, claiming that Green and Krasno’s model incorporates a high degree of multicollinearity, as their prior incumbent spending instrument correlates highly with the challenger political quality and expenditure variables in the model. Jacobson argues that the instrument for incumbent spending “is, in part, a proxy for party strength and challenger quality.” One goal of my study is to re-examine this debate in the context of the modern campaign finance landscape.

I also rely on scholarship surrounding election forecasting. In 2014, Patrick Hummel and David Rothschild explored the usage of econometric modeling to forecast campaign outcomes. Specifically, they focus on presidential, gubernatorial, and senatorial races. Hummel and Rothschild use presidential approval ratings, incumbency, past election results, economic factors, and state ideology to predict electoral outcomes. They find that past election results, incumbency, and state-level economic factors can be used to predict electoral outcomes with a high degree of statistical significance. Consequently, I use both past election results and incumbency in my models. I exclude economic factors for two reasons. First, I was unable to acquire district-level personal income data in the timeframe of this project. Second, I follow the example of FiveThirtyEight’s 2018 House forecast, which did not include economic factors. Indirectly, voters’ economic attitudes and opinions are incorporated into the model through generic ballot polling. In my model, cycle-level fixed effects also likely pick up some of this noise.

Based on FiveThirtyEight’s usage of generic ballot polling, I decided to include a generic ballot measure in my models. I use a variation of the technique described by Joseph Bafumi, Robert S. Erikson, and Christopher Wlezien in their 2008 paper, in which they average all generic ballot polls over the last 30 days of a given campaign, converting all polls to a percent Democratic of the two-party vote. For instance, 50-42 converts to 54.3% Democratic. I interact this variable with a factor variable for political party to capture the obvious differences in the ways that Democrat-favorable polling affects candidates of different parties.

Data, Hypotheses, and Methods

Results

Discussion

I report two main findings. First, on a high level, I find a severely diminishing marginal relationship between campaign spending and votes won, with the curve flattening out just before the $1 million mark. For incumbents, the marginal utility of spending is negative. Second, I observe a strong, positive, and highly linear relationship between candidates’ share of spending within their district and their respective share of the vote. This effect is highly significant. I control for incumbency, generic ballot polling, party affiliation, opponent spending, and district partisanship, and I generate fixed effects for district and cycle.

Framed in terms of my initial hypotheses, we see that all three hypotheses are supported by the data. The first hypothesis posits that candidates will face a diminishing marginal utility of campaign spending. When denominated in dollars spent and votes won, the marginal utility of campaign spending indeed appears to decline rapidly, approaching the zero point at around $1 million. The second hypothesis is also supported by the data. In my regressions, party affiliation, Democratic share in the prior election, and incumbent status are all statistically significant. Democratic share of generic ballot is only significant when interacted with party. Importantly, we observe that incumbents face a negative marginal utility of campaign spending. Finally, the third hypothesis asserts that the share of spending will correlate with vote share, but that the relationship will look different than in the first hypothesis. This is also supported. We observe a highly linear relationship between a candidate’s share of spending and their ultimate share of the vote.

There is an important conflict between the results of the dollars spent-votes won analysis and the results of the spending share-vote share analysis. The dollars spent-votes won regression shows a positive marginal utility of spending for non-incumbents and a negative marginal utility of spending for incumbents. This supports Jacobson’s view. However, the spending share-vote share regression shows a non-significant interaction between incumbent status and share of spending. This supports Green and Krasno’s finding that, with the proper effects in place, incumbents can significantly affect their chances of winning by spending more.

The dollars spent-votes won examines candidates on an individual basis, with opponent spending included as a covariate. In contrast, the spending share variable captures a candidate’s spending in terms of total activity by candidates in the district, just as the vote share variable captures a candidate’s results in terms of all results in the race. When results are expressed as vote share, a vote for Candidate A is mathematically largely offset by a vote for their opponent. Importantly, vote share is what ultimately matters for election results. When spending is expressed as spending share, a dollar spent by Candidate A is mathematically largely offset by a dollar spent by Candidate A’s opponent. As such, I argue that the spending share-vote share comparison better captures the competitive dynamics at play. Consequently, I find a strong and linear relationship between spending share and vote share.

The findings yielded by the two analyses can help explain political professionals’ heavy focus on campaign fundraising and spending. The dollars spent-votes won comparison indicates that there is nearly zero marginal utility past roughly the first $1 million spent. This would indicate that the existing high levels of fundraising and spending are irrational. However, the spending share-vote share comparison suggests that a given candidate’s marginal dollar spent is always spent rationally, so long that the candidate is increasing their share of district spending. By observing campaign spending through the lens of spending share and vote share, we can empirically examine and evaluate modern trends in campaign spending. While dollars spent may offer a diminishing marginal utility, candidates can always stand to benefit by outspending their opponent to the greatest extent possible.

Limitations and Next Steps

I plan on continuing to analyze this data, pursuing both extensions of the present research inquiry and entirely new projects. One clear limitation in my present analysis is my lack of attention to the simultaneity problem. As Gary Jacobson explains in his 1990 paper, campaign donations are often made in response to electoral expectations. At the same time, campaign spending often translates to a higher probability of victory for challengers and a lower probability of victory for incumbents. This is because endangered incumbents often engage in more vigorous fundraising, and promising challengers are provided with funds by hopeful supporters. This simultaneous causation leads to some estimation problems, as it is difficult to parse out the direction of causality. In their 1994 paper, Goidel and Gross attempt to deal with this issue through a three-stage least squares design, modeling incumbent and challenger expenditures as endogenous variables through a simultaneous system of equations. My design does not directly address the simultaneity problem. As such, my estimates may be prone to systematic bias driven by simultaneous causality. However, in future extensions of this design, I look to incorporate an adapted version of Goidel and Gross’ three-stage least squares design.

Another extension that I may pursue is the incorporation of logistic regression into my analyses. In these regressions, the dependent variable would be the logged odds of victory. By directly analyzing the impact of campaign spending on probability of victory, I may be able to generate more intuitive and useful results than simply looking at votes and vote share.

Finally, I look to dig more into the potential impact of Citizens United on campaign spending. While I’ve explored this in a roundabout way through my disaggregation of the dollars spent-votes won trend into the different cycles, I do not focus specifically on the impact of this decision. While I considered including a supplemental analysis of the impact of Citizens United into this project, I ultimately decided that such an important topic would be ill-suited as an analytical afterthought. Consequently, I will look to explore this in a future project.

References