Political Advertising Research Paper

The second agenda concerns the extent of voter autonomy. Are votes hollow echoes of campaign messages, or does political discourse enable voters to reach reasonable decisions that reflect their own preferences and values? Electoral reform motivates much research on advertising and voter autonomy. If political advertising easily misleads voters, then restrictions on campaign advertising or some sort of ‘political consumer protection’ may be necessary in order to make democracy work. To the extent that there is a scholarly consensus on this matter it is that concerns about voter autonomy are overdrawn. Behavioral research in a wide variety of settings indicates that advertising has modest but important effects on election outcomes: advertising can make a difference in a close election, but it does not appear to persuade large numbers of voters to turn against a candidate or party that is well-known, well-liked, and represents the voters’ interests.

1. Advertising And Vote Production

Political advertising covers a wide range of media that candidates use to mobilize supporters to turn out or to persuade people to vote for them. Commentary on political advertising often equates it with television advertising, as, for instance, when European commentators descry the ‘Americanization’ of their politics. But even in the US there is, in fact, much less television in politics than is often supposed. Campaigns for the US House of Representatives, for instance, use as much direct mail as television advertising. And in local campaigns candidates use little television but a vast range of other media, including mail, billboards, lawn signs, telephones, and door-to-door canvassing (Key 1966).

Campaigns’ usage of different media appears to follow the same principles that apply to firms. Candidates choose what media to use based on the productivity of the media and their relative prices. Variation in the mix of media used at different levels of office suggests that there are economies of scale to different sorts of advertising. Personal contact is reputed to be the most effective mode of political communication and it is a staple of local elections everywhere. But it is impossible to spend 15 minutes with each of the millions voters in a national election.

Mass media are most cost-effective at reaching broad constituencies. Evidence of the rational use of media comes from US House races. The average household in the typical House district receives about two pieces of mail and sees at least two television advertisements. However, candidates choose a mix of advertising media based on the prices and productivity of those means of reaching potential voters (or consumers). Campaigns for the US House in districts with expensive broadcasting media (such as Los Angeles and New York) spend $2.50 on mail for every $1 broadcasting, or three pieces of mail and 15 seconds of advertising per person. House campaigns in districts with inexpensive broadcasting (such as Dayton, Ohio and El Paso, Texas) spend $10 on broadcasting for every $1 on mail, or one piece of mail and 8 minutes of advertising per person. (For details see Ansolabehere et al. 1999a.) To my knowledge, similar sorts of calculations have not been conducted for other countries, though it is evident from patterns of media usage that parties and candidates in Europe, Australia, and elsewhere are similarly responsive to media prices. (See, for example, Blumler 1983 and Australian Election Commission 1997.)

Social science research has focused on how voters respond to the overall volume and different sorts of campaign advertising that they receive. The estimated effects depend in part on the methods used.

Three different approaches have been taken to the study of advertising. First, laboratory and field experiments have been used to gain precise measures of the effects of advertising exposure in specific campaign settings. Such studies control what messages individuals receive and then measure differences in opinions or behaviors in response to different sorts of messages (e.g., Gosnell 1927, 1935, Ansolabehere and Iyengar 1996, Gerber and Green 1999, Norris 1999).

Second, studies of aggregate data correlate the observed vote to levels of campaign spending. Some of these studies design their inquiries around ‘quasiexperiments’ which exploit factors that change levels of spending (such as personal wealth) in order to measure the effect of spending on the vote. The seminal work in this vein, for both American and European researchers, is Jacobson (1978). Subsequent research has improved on this model; see, for example, Gerber (1998). Races involving three or more candidates pose statistical problems in estimation of vote production functions, for which widely accepted statistical models have not yet been developed. Third, surveys associate reported vote or political attitudes with reported or recalled exposure to messages. Almost all research outside of the US relies on survey methods. For a discussion of the problems with survey based measures of campaign effects see Ansolabehere et al. (1999b). Analogous with the aggregate studies, instrumental variables can solve such problems.

Experiments, which offer high internal validity, show a remarkably consistent set of results about the capacity of advertising to produce votes. Researchers observe exactly what messages citizens see and can design the study so that other factors do not confound the effect of the experimental manipulation. Perhaps the earliest experimental study of political advertising and campaigning took place in the 1920s in the US. To measure the mobilizing effects of direct campaigning, Gosnell (1927) sent a cadre of graduate students out to canvass a set of blocs of the city of Chicago in an effort to manipulate the levels of voter contact. They found higher in turnout ranging from 4 to 7 percent in the areas that the students canvassed over the areas that they did not canvass. Fast forward to the 1990s. Gerber and Green (1999) replicated Gosnell’s study in several towns in southern Connecticut. They found increases in participation similar to those reported in the Chicago studies 70 years earlier. They also found increases in vote shares of candidates in the neighbor- hood of 4 percentage points for two pieces of mail. In a series of laboratory experiments conducted in Los Angeles from 1990 to 1993, Ansolabehere and Iyengar (1996) found strikingly similar results for television advertising. Exposure to a 30-second television advertisement raised the percentage of viewers who intended to vote for a candidate by 4 to 10 points. These and many other experiments show substantial effects of political advertising on candidates vote shares. Those who see a broadcast message, receive a mailing, or talk to a campaign worker are between 4 and 10 percent more likely to support the candidate being promoted.

Aggregate studies of campaign spending yield similar, albeit lower estimates of the effectiveness of campaign advertising. Using fairly complicated, and controversial, statistical models, researchers have attempted to measure the marginal effect on the vote of spending an additional campaign dollar. Such techniques yield estimates ranging from miniscule effects of a dollar spend on the vote to estimates of approximately $10 per vote. The key papers in this subfield are Jacobson (1978), Green and Krasno (1988), and Gerber (1998).

Such estimates are in the right ballpark, but the wide range of estimates in the literature on campaign spending suggests that additional work finding valid statistical models is in order before we can put much confidence in the estimates emerging from this literature.

The problem is that it is difficult to sort out cause and effect in aggregate electoral data. Consider, for example, two similar legislators: one holds a highly competitive seat and the other holds a very safe seat. The legislator in the competitive seat must spend large amounts of money to win re-election; the legislator in the safe seat is unlikely to be challenged and if they are they do not need to spend much to retain the seat. In this example the correlation between spending and vote shares is negative! To fix this problem researchers have taken a ‘quasi-experimental’ approach, using factors such as candidate wealth, which affect directly how much candidates spend and in turn affect their vote shares. Improving such statistical models remains an important research challenge.

Survey research has presented the most modest estimates of the effectiveness of advertising. A long line of studies report minimal effects of campaigns in general and of advertising in particular on the vote (in the US, see Campbell et al. 1960; in the UK, see Norris 1999). Serious problems of question validity lead to serious doubts about these research findings. In the laboratory setting, Ansolabehere and Iyengar (1996) document that survey-based measures of advertising exposure miss one exposure for every one that they detect. This degree of misreporting is sufficient to explain why surveys often produce insignificant findings or even estimates in the ‘wrong’ direction. Quasi-experimental techniques analogous to those used in aggregate studies can correct measurement problems (Bartels 1993).

Beyond differences in media and methodologies, a further theme in the study of political advertising is the mechanism through which messages become votes. There are three. Advertising may solidify an uncertain vote (reinforcement), it may stimulate a supporter to vote (mobilization), and it may lead a voter to change their mind (conversion). This typology is due to Berelson et al. (1954). Although political observers often talk of ‘swing’ voters and movement in the polls, there appears to be relatively little evidence in experiments or surveys of conversion. Few Republicans decide to vote Democrat and few Democrats decide to vote Republican in response to political advertisements. Perhaps the strongest evidence against conversion, though, comes from nonpartisans. Non- partisans, especially low information independents, are the least responsive to political advertising.

Reinforcement appears to be the dominant mechanism in political communication. For instance, in the experimental research reinforcement produces the clearest effect on the vote. Nonpartisans and viewers of the opposite party as the sponsor of the ad showed no statistically significant response to a candidate’s ads. Low information partisans of the same party as the sponsor showed by far the strongest favorable response to political advertising. That effect shrank, though it remained statistically and substantively large, with the partisan’s level of information and interest in politics (Ansolabehere and Iyengar 1996). This pattern suggests that political advertising fits neatly into the more general model of information acquisition developed by Zaller, in which both pre- disposition to believe a piece of information and prior levels of knowledge determine responsiveness to new information (Zaller 1992).

Mobilization has proved the most controversial of the three mechanisms. Experimental research on political advertising has called into question the long-standing belief among political scientists that campaigns are inherently stimulating affairs. Berleson et al. (1954) describe the process in detail. This is essentially the model behind retrospective voting developed by Key (1966) and widely used to under- stand elections in the US, Europe, and elsewhere. A large number of experimental studies have demonstrated that not all viewers are stimulated to vote by political advertisements. For example, nonpartisan and low information viewers exposed to negative advertising appears express significantly lower intentions to vote than others. As a result, negative campaigning can produce lower turnout among the key group of ‘swing’ voters and an electorate that reflects two ideologically opposed groups.

The demobilizing effects of negative advertisements on non-partisans underscores the difference between reinforcement and mobilization. Negative advertising might win the highest vote share, mainly by reinforcing partisans, but it can lead to lower overall participation.

One important caveat about the state of social science research on political advertising is in order. The bulk of the research on the effectiveness of political advertising has been conducted in the United States. Many of the findings about mobilization and reinforcement likely depend on the electoral system, which influences the number of parties or candidates. The dynamics of campaigns and the consequences of advertising appear to be quite different in races involving more than two candidates. Johnston et al. (1992) provide excellent examples of political communications having complicated effects on competition in Canada’s multi-party system. Careful experimental research on multicandidate and multiparty races, though, is in its nascency.

2. Voter Autonomy

The central problem for democratic theory in modern societies is that people apparently know little about the choices they face and they have little incentive to seek new information. Candidates and parties, then, have very strong incentives to provide information to voters at little or no search costs: candidates try to reach us, rather than us, them. As a result we are highly reliant on what politicians choose to say in their advertisements.

This is the problem of voter autonomy. By one telling voters have abdicated to political elites the important jobs of choosing what issues will be discussed, what elections and therefore government will be about, and even what we will think as people walk into the voting booth. From this view, people are highly susceptible to false or deceptive advertising and will vote for the candidate whom we see the most rather than the one we think is best representative. Regulation of the volume and content of political advertising may, therefore, be necessary.

The counter argument is that voters, even though they do not know much by way of details, have a clear sense of what they themselves value or prefer, and, in order to win elections, politicians must adjust their messages to what the voters want. In crafting their advertisements, a candidate selects issues on which he or she can make electoral gains, because the voters care about it, because the candidate has a strong record, or because the candidate has have taken the more popular position. This leads to majoritarian biases in what is discussed, but that is the essence of elections.

Voter autonomy presents a much more difficult set of research questions. In order to say that people are or are not misled by political communications, we must establish what it is that they prefer or value a priori. The effects of advertisements must then be judged against that baseline. Two questions are central.

First, do advertisements lead people to choose a candidate who is less good for their interests or values? Research showing strong reinforcement and few conversion effects suggests that the answer to this question is probably no. In both survey and experimental data it appears that people who lean toward a party become more solidly in favor of that party as the campaigns progress (Ansolabehere and Iyengar 1996, Chap. 4, Gelman and King 1990). An important variation on this arises from tactical voting in Britain, Canada, and other countries. As campaigns progress, voters supporting a candidate or party in third place or lower might vote for the second place candidate if they prefer that candidate over the front-runner. This seems to have happened in the 2000 Mexican election as support for Cardenas fell over the course of the campaign and support for Fox, who was in second place most of the race, rose.

Nonpartisans tend to move very little, and tend not to vote. The immediate political implication is that campaign discourse is polarizing: it ultimately pits Democrats against Republicans, Conservatives against Labourites. The low amount of conversion suggests, however, that people fundamentally are not misled.

Second, can advertising set the public agenda, or are people mainly responsive to what they themselves care about? While agenda setting effects have been widely documented in other areas of study, they have received surprisingly little careful study in the field of political advertising. Early evidence of agenda setting is provided by Bereleson et al. (1954, Chap. 12). Two careful studies stand out. Jamieson’s (1992) Dirty Politics documents how the Republican and Democratic strategists use their advertising and the news media to magnify their own agendas, especially Republican’s usage of crime and race. Johnston et al. (1992) demonstrate that the debate over free trade completely transformed the Canadian voters’ priorities, without changing their preferences on free trade itself.

Reinforcement and agenda setting can hold conflicting implications for voter autonomy. Partisan reinforcement suggests that there is little evidence that people vote against their expressed preferences; agenda setting suggests that people are vulnerable to what the campaigns choose to discuss. There may, however, be no contradiction here. It may indeed be the case in each of the examples discussed that voters were simply very receptive to new agendas, and that elections gave voters an opportunity to change public policies on trade, affirmative action, and other issues from those set decades earlier.

3. New Directions

Political advertising poses both methodological and normative challenges for social scientists, and our understanding on both levels has advanced considerably over the last two decades. Advertising has important effects on election outcomes, but it does not appear to lead people to vote for candidates with whom they fundamentally disagree. It, therefore, does not on the whole undercut the normal workings of majority rule in democratic elections.

In terms of methods, a high priority for further inquiry is bridging the gaps between experimental, aggregate, and survey methods, and developing more powerful quasi-experimental approaches to correct measurement and simultaneity problems. To this end, research on vote production, which has relied mainly on quasi-experiments in aggregate data, should make greater use of experiments to map out features of the vote production function more fully, especially the effects of repeated exposures to messages and interactions between the messages and the viewers preferences.

In addition, there is considerable need for research on political advertising outside the US. The American focus of campaign research stems in part from existing regulations on elections and media. The US imposes few limits on how candidates and parties raise and spend campaign funds, on their access to media, and on what can be said on the various media. As a result, the US has proved a rich field to study advertising effects. Research on campaigning in European countries has placed a much greater emphasis on electoral laws, political parties, and tactical voting. There is, however, growing interest in campaign effects in European political science.

Beyond methodological differences, there is the matter of why. Reinforcement and mobilization are a sufficiently well established phenomenon that it is worth opening up the black box. Why are these the routes through which political advertising works, and why not conversion? The highly inductive approach of empirical researchers is likely to pay the biggest dividends, as political theory has yet to produce a powerful model of why people vote.

Bibliography:

1. Ansolabehere S, Gerber A, Snyder J M Jr 1999a How Campaigns Respond to Media Prices. MIT, Cambridge, MA
2. Ansolabehere S, Iyengar S 1996 Going Negative: How Political Advertising Shrinks and Polarizes the Electorate. Free Press, New York
3. Ansolabehere S, Iyengar S, Simon A 1999 Replicating experiments using surveys and aggregate data: The case of negative advertising. American Political Science Review 93: 901–10
4. Australian Election Commission 1997 Funding and Disclosure Report, 1996. AEC, Sydney, Australia
5. Bartels L 1993 Message received. American Political Science Review 87: 267–85
6. Berelson B, Lazarsfeld P, McPhee W 1954 Voting. University of Chicago Press, Chicago
7. Blumler J 1983 Communicating to Voters: Television in the First European Parliament Elections. Sage, London
8. Campbell A, Converse P, Miller W, Stokes D 1960 The American Voter. University of Chicago Press, Chicago
9. Fritz S, Morris D 1992 The Handbook of Campaign Spending, 1990. CQ Press, Washington, DC
10. Gelman A, King G 1990 Why are presidential election campaign polls so variable when votes are so predictable? British Journal of Political Science 23: 409–52
11. Gosnell H F 1927 Getting-out-the vote: An Experiment in the Stimulation of Voting. University of Chicago Press, Chicago
12. Gosnell H F 1935 Machine Politics: Chicago Model. University of Chicago Press, Chicago
13. Green D, Krasno J 1988 Salvation for the spendthrift incumbent. American Journal of Political Science 32: 884–907
14. Gerber A 1998 Estimating the effect of campaign spending on senate election outcomes using instrumental variables. American Political Science Review 92: 401–11
15. Gerber A, Green D 1999 Does canvassing increase voter turnout? A field experiment. Proceedings of the National Academy of Sciences 96: 10939–42
16. Jacobson G 1978 The effects of campaign spending in congressional elections. American Political Science Review 72: 469–91
17. Jamieson K H 1992 Dirty Politics. Oxford University Press, New York
18. Johnston R, Blais A, Brady H E, Crete J 1992 Letting the People Decide: Dynamics of a Canadian Election. Stanford University Press, Palo Alto, CA
19. Key V O 1966 The Responsible Electorate. Vintage, New York
20. Morris D, Gamache M 1994 The Handbook of Campaign Spending, 1992. CQ Press, Washington, DC
21. Norris P 1999 Political communications in British election campaigns: Reconsidering media eff British Elections and Parties Review 5: 190–211
22. Rosenstone S 1983 Forecasting Presidential Elections. Yale University Press, New Haven, CT
23. Zaller J 1992 The Nature and Origins of Mass Opinion. Cambridge University Press, New York