For many years political scientists have argued that campaigns have minimal effects on election outcomes. When campaign-related information flows activate latent predispositions, given balanced resources, election results are largely predetermined. (1) This perspective appears to be reinforced by the finding that election outcomes can be accurately forecast well before a campaign has run its course (Campbell 2000; Fair 1978; Lewis-Beck and Rice 1992; Rosenstone 1983). However, a consensus has emerged over the past decade among political scientists that campaigns have substantive persuasion effects (Franz and Ridout 2010; Hillygus and Jackman 2003; Huber and Arceneaux 2007; Shaw 1999a, 1999b; Vavreck 2009).
Though there has been a great deal of work on campaigns' persuasion effects over the past decade, there has been relatively little on campaigns' activation effects (see Andersen, Tilley, and Heath  for an important exception, and Huber and Arceneaux  for a recent analysis of a campaign's reinforcement effect). A few of the more recent studies of campaign effects do distinguish between types of activities and their associated persuasion effects. For example, Holbrook (1996) attempts to parse the effects of presidential conventions and debates, and Shaw (1999a) distinguishes between the persuasion effect of presidential candidate TV advertising and presidential candidate visits by state. But much of the recent work on campaigns' persuasion effects often does not attempt to distinguish between persuasion effects and activation effects. For example, Franz and Ridout (2007, 467n.1) write: "We do not distinguish here between the different processes by which advertising might influence candidate preferences; that is, we do not distinguish between attitude change brought by conversion, the activation of predisposition or the reinforcement of prior preferences." In a similar manner, Hillygus and Jackman (2003) also categorize any movement toward a candidate as a manifestation of a persuasion effect.
Here we consider only presidential general election campaigns, which are characterized by long lead times, high media exposure, only two major candidates (in most states in most years), and generally clear partisan and ideological separation between the candidates. These conditions combine to increase the predictability of votes and the stability of opinions, and to minimize feedback effects arising from polling and other sources of information that can affect expectations. We would expect multicandidate elections, primaries, low-salience elections, and nonpartisan contests to show much less stability and predictability.
In a long campaign, such as a presidential election, the performance of these persuasion and activation models should depend on the time scale being considered, and it is these changes as a campaign progresses that we discuss.
This article contributes to the existing literature conceptually and empirically. We offer clean definitions of campaigns' persuasion and activation effects, which clearly distinguish between the two types of campaign effects. These definitions permit us to estimate the extent of campaigns' activation effects. To do so, we deploy a relatively straightforward but new methodology to estimate how the effects of the "fundamentals" change over the course of the campaign. Our empirical analysis estimates the magnitude of the variance in vote choice that can be ascribed to campaigns' activation effects. Finally, based upon existing theory, we argue that activation effects should vary by partisanship and political context, and we deploy our methodology to parse the relative activation effects of the 2000, 2004, and 2008 campaigns by region and partisanship. In the end, we find that the 2000, 2004, and 2008 campaigns have substantive activation effects.
In addition to our contribution to the political science literature, this article addresses a persistent confusion in the news media, where even the most sophisticated journalistic analysts tend to assume a random walk model as a matter of course (perhaps via a mistaken analogy to the efficient markets hypothesis in finance). By explicitly separating the random walk and mean reversion models, we are able to situate journalistic and political science models on common ground.
Campaign Effects: Persuasion and Activation
Studies of persuasion and activation effects tend to use one of four methodologies: (1) individual level panel studies to determine changes in vote intention over the course of a campaign, (2) experimental studies to isolate the causal effect of specific stimuli, (3) individual level cross-sectional studies that take advantage of measures of respondents' ad exposure, (4) aggregate level studies based upon rolling cross-sectional data. A small number of recent studies that fall into the latter two categories also integrate natural experiments into their analyses.
The earliest work on campaign effects used multiwave (panel) survey data in a particular locale to assess changes in vote attention at the individual level (Berelson, Lazarsfeld, and McPhee 1954; Katz and Lazarsfeld 1955; Lazarsfeld, Berelson, and Gaudet 1944) to conclude that political "campaigns are important primarily because they activate latent predispositions" (Lazarsfeld, Berelson, and Gaudet 1944, 74, italics in the original). More recent analyses based upon individual level panel data do not attempt to distinguish the direction of change and hence do not distinguish between persuasion effects and activation effects. For example, Hillygus and Jackman (2003) categorize any change in vote attention as a manifestation of persuasion, even if such change was toward the party that the voter identified with (or leaned towards) in the initial wave of the study. (2)
Experimental and aggregate level studies often attempt to assess the magnitude of effect of television advertising on respondents' candidate preferences. Experimental studies often find a statistically significant and substantive effect of television advertising on respondents' candidate preferences (Kahn and Geer 1994; Valentino, Hutchings, and
Williams 2004). However, as has been commonly noted, experimental studies often suffer from a variety of shortcomings, such as poor external validity, problematic samples, and unrealistic settings and time frames.
A number of studies have matched data on candidate television advertising with individual cross-sectional or panel survey data to assess the extent to which individual-level exposure to such advertising affects the probability of supporting a candidate (Franz and Ridout 2007; Goldstein and Freedman 2000). However, Huber and Arceneaux (2007) note several potential confounding factors and, taking advantage of natural experiments to more finely gauge the effects of candidate television ads aired on vote choice, find larger persuasion effects than almost all previous nonexperimental studies. To their credit, Huber and Arceneaux (2007) also parse the data in search of reinforcement effects but find that candidate television advertising has almost no statistically significant reinforcement effects. (3)
Gelman and King (1993), Romer et al. (2006), Romer et al. (2004), and Kenski, Hardy, and Jamieson (2010) each use multiple polls or rolling cross-sectional polls over the course of a campaign to examine campaign effects. Gelman and King (1993) argue voters become more informed as the campaign progresses, resulting in their final vote choice on election day closely matching their "enlightened preferences" (433). The decrease in variance in the polls over the course of the campaign and the increasing ability of the fundamental factors to predict vote choice over the course of the campaign suggest that campaigns function to activate voters' latent predispositions. The inherent limitation of Gelman and King (1993) is that the change in the relative weight given by respondents to the various fundamentals over the course of the campaign could be due to campaigns successfully persuading voters to focus on those specific fundamental factors, thus persuading voters via priming. Indeed, it is commonly suggested that campaigns are about determining the factors upon which the voters make their vote choice (Simon 2002). Thus, the changes in weights given to the various "fundamentals" identified by Gelman and King (1993) could be due to campaign effects other than activation.
The methodology deployed in the present article ensures that any changes in weight given to the fundamentals over the course of a particular campaign cannot be due to persuasion or other campaign effects specific to the particular campaign.
Journalists' and Political Scientists' Perspectives on Persuasion and Activation
As noted above, scholars have long argued over the effect of campaigns on vote choice in presidential elections (see Bennett and Iyengar 2008; Hillygus 2010; Iyengar and Simon 2000 for systematic reviews of the literature). Despite extensive research in the area, no one has offered a complete explanation of the meaning of the question "Do campaigns matter?" To this end, we consider aggregate preferences as a time series leading up to an election, following the example of Wlezien and Erikson (2002). This time series may be stationary, so that preferences have a mean reversion quality where final votes are predicted by predispositions. Or preferences may be better characterized as the result of a random walk where campaigns change or create preferences apart from how people would be predisposed to vote.
The popularity of the random walk model for polls may be partially explained via analogy to the widespread idea that stock prices reflect all available information, as popularized in the book, A Random Walk Down Wall Street (Malkiel 1996). Once the idea has sunk in that short-term changes in the stock market are inherently unpredictable, it is natural to...