- About us
- Research Review
- Latest Research
- Info Center
- Teaching Resources
Do campaign expenditures and advertising affect elections outcomes? Surprisingly, political scientists and economists alike have had trouble answering this seemingly obvious question in the affirmative. The existing empirical research that uses non-experimental data suffers from identification issues, and often finds an intriguing “minimal effects” result. Examples of studies that detect a small impact of money and advertising abound, the most well-known among economists being Levitt (1994). If money and advertising are irrelevant, how can we rationalize the large sums of money spent on political campaigns in general, and on TV advertising in particular? From a policy perspective, if money has no impact on elections, a large debate on public campaign finance is misplaced.
In a paper forthcoming in the Review of Economic Studies, Bernardo Silveira and I explore a feature of Brazil’s electoral legislation to provide credible causal estimates of the impact of TV advertising on electoral outcomes, which is tantamount to measuring the impact of campaign spending.1 Furthermore, we document where advertising has a stronger impact, thus shedding light on its transmission channel.
Difficulties in the existing empirical literature are due to limitations in dealing with reverse causality and omitted variables. Most research consists of regressing election outcomes on a measure of either campaign spending or voters’ exposure to advertising. Reverse causality occurs because donors expect their money to buy political influence, and thus contribute more generously to candidates with a high probability of winning. Unobserved heterogeneity in the quality of candidates causes omitted variable bias. Stronger candidates in electoral terms may also be more adept in raising funds. In fact, the same unobserved characteristics, say charisma, that convince voters to vote may also induce them to donate (think of Obama’s extremely successful fundraising online campaign). Another issue is unobserved electoral district preferences. For example, a right-wing candidate running in a leftist district has difficulties in both raising money and getting votes. Most theoretical mechanisms indicate that correlation-based methods, such as Ordinary Least Squares (OLS), produce upwardly biased estimates (an exception occurs when a candidate is strong enough to make money unnecessary).
The literature recognizes the difficulties, but has had only limited success in solving them. Some researchers resort to experiments. Ansolabehere and Iyengar (1996), among others, provide laboratory evidence that TV advertising influences voters’ behavior. Laboratory studies offer unparalleled internal validity, but have weak external validity. Randomization has also been employed. Gerber et al. (2007) run a field experiment during the 2006 Texas gubernatorial race. They randomize the date and volume of TV and radio advertisements for one of the candidates, and evaluate how they affect voters’ perception. They find a positive but transient advertising effect. Although Randomization is a powerful tool, Gerber et al. only randomize the advertisements for one of the candidates. More importantly, the experiment took place several months before the election. Voters’ attention at this stage is not comparable to right before the election.
Levitt (1994) compares the same pairs of contestants in multiple U.S. congressional races. Using only differences in voting and spending between races, he eliminates all unobserved heterogeneity that is constant over multi-year periods – thus mitigating problems associated with omitted variables. However, difference the data removes their main source of variation. If the identification strategy is successful, then candidates’ fundraising should be roughly constant over the time – except for small random variations. Starting from an equilibrium situation, small variations in campaign spending would have a small impact on electoral performance. Not surprisingly, this is precisely what Levitt finds.
We explore a quasi-natural experiment resulting from the Brazilian electoral legislation. Inn Brazil, gubernatorial elections work in a two-round system. A runoff takes place if no candidate reaches 50% of the votes in the first round the two top candidates advance to the second round.2 For a period of time before each round, TV and radio networks must air candidates’ advertising free of charge, and no additional paid TV or radio advertising is allowed.3 First-round TV time is distributed among candidates according to the following rule. A third of the available time is equally distributed among all registered candidates. The remaining two-thirds of the time are allocated according to their coalitions’ representation at the National Parliament. In the second round, candidates split TV time equally.
We explore the difference in TV time between rounds to estimate the impact of advertising on election outcomes. We consider only the pairs of candidates that make it to the second round, and take first-differences to control for time-invariant unobserved candidate and district heterogeneity.
Our strategy resembles Levitt’s (1994) in the sense that we compare the same pair of contestants in different races – two rounds of the same election in our case. There are two major differences, however. Levitt’s races are at least two years apart. In contrast, the runoff is no more than 28 days after the first round, increasing the confidence that most candidate and district characteristics are constant over races. More importantly, TV time is determined by law, and the rule produces large changes in the distribution of TV time between rounds.
In summary, we observe the same pair of candidates with different TV time shares in two races close to each other in time. In addition, the variation in TV time between rounds is largely outside the candidates’ control. Unlike Gerber et al. (2007), we have TV advertising for both candidates during the entire campaign. We find a large impact of TV advertising on election outcomes. Using our preferred estimate, a one percentage point (p.p.) increase in a candidate’s TV time share causes a 0.272 p.p. increase in her vote share. Using only first-round data, which is admittedly endogenous, we detect a stronger impact of advertising on election outcomes. This is precisely what theory predicts: OLS overestimates the impact of advertising.
We then investigate the determinants of advertising effects. The literature is silent about the circumstances under which media has an impact on electoral outcomes. We can address this issue because of the large demographical variation across cities in our sample. We find stronger media effects in poorly educated cities where TV penetration is high. We also find somewhat weaker evidence that advertising has a stronger impact in poorer and more unequal cities. The fact that media effects vary with education and TV penetration sheds light on the underlying mechanisms involved, and provides additional support for causal interpretation (Deaton, 2009).
Brazil’s electoral rules provide policy lessons. Public campaign finance is a much debated issue in various countries. Our results contribute to the policy debate in several ways. First, and somewhat trivially, the discussion on public campaign finance is only relevant if money does buy votes. Contrary to the literature, we show that it does. Furthermore, those without money are the most influenced, suggesting that the political preference of the poorest, ill-educated can be made to coincide with moneyed interests.
Second, centralized allocation of TV time may dissociate advertising capability from economic prowess. Small parties get a share of TV no matter what. One may buy TV time by convincing parties to join a coalition, but that is observable to voters, who may punish parties for doing so.
Finally, centralized allocation of TV time may counterbalance not only private economic prowess but also the governments’ advertising capacity. In Brazil, the government is a major advertiser. In a context in which the media has a progressively diminished ability to generate revenues through paid advertising, government advertising will become more and more important. Thus, governments will have an increasing ability not only to influence public opinion and slant the media. Preliminary evidence in the literature suggests so (Di Tella and Franceschelli, 2009).
The Brazilian experiment with centralized allocation of political advertising has many potential pitfalls. Guaranteeing some TV time to small parties may fragment the political system, which some may find undesirable. Additionally, players may try to game the system. Nevertheless, the Brazilian rules were set ex-ante, when the original political contract was written, making it harder (but certainly not impossible) for economic power to influence elections through the acquisition of TV advertising time.
ANSOLABEHERE, S. AND IYENGAR, S., Going Negative: How Political Advertisements Shrink and Polarize the Electorate, The Free Press, New York, 1996.
DEATON, A., (2009) “Instruments of Development: Randomization in the Tropics, and the Search for the Elusive Keys to Economic Development,” NBER Working Paper No 14690.
DI TELLA, R. AND FRANCESCHELLI, I., (2009) “Government Advertising and Media Coverage of Corruption Scandals,” NBER Working Paper No 15402.
GERBER A., GIMPEL J., GREEN D. AND SHAW D., (2007) “The Influence of Television and Radio Advertising on Candidate Evaluations: Results from a Large Scale Randomized Experiment,” Available at http://www.allacademic.com/meta/p197939_index.html.
GOLDSTEIN, K. AND RIDOUT T., (2004) "Measuring the Effects of Televised Political Advertising in the United States," Annual Review of Political Science 7, 205 – 226
LEVITT, S., (1994) “Using Repeat Challengers to Estimate the Effect of Campaign Spending on Election Outcomes in the U.S. House,” Journal of Political Economy, 102, 777 – 798.
PRAT, A., (2006) “Rational Voters and Political Advertising,” in B. Weingast and D. Wittman (eds.) The Oxford Handbook of Political Economy, Oxford: Oxford University Press, 50 – 63.
1 Most campaign spending goes to TV advertising. In the political science literature, campaign spending is commonly used as proxy for TV advertising (Goldstein and Ridout, 2004).
2 Voting is mandatory and turnout rates are similar in the first and second rounds.
3 Campaign advertising takes place for a period of 45 days preceding the first round. In case of a runoff, mandatory advertising starts 48 hours after the first-round results are announced, and ends Friday before the elections. In the first round, gubernatorial elections compete for airtime with both presidential and legislative races. In the second round, it shares time only with the presidential race.