Concurrent Elections and Voter Attention: How voters search for political information in crowded campaign environments

Transcription

1 Concurrent Elections and Voter Attention: How voters search for political information in crowded campaign environments David J. Andersen PhD Candidate Rutgers, The State University of New Jersey New Brunswick, NJ

2 Concurrent Elections and Voter Attention: How voters search for political information in crowded campaign environments In this paper, I use an experiment run in the dynamic process-tracing environment (DPTE) to examine how voters change their search for information about candidates in different offices based simply upon the electoral environment. I find that voters tend to not change their overall information search for political information, but alter the rate at which they search for each office they encounter. They do gravitate towards certain offices more than others, so that the more offices appear on the ballot, the less information is encountered about each set of candidates. This does, however, hurt lower offices to a greater extent than higher offices. Representative governments rely upon voters to voluntarily expend time and energy learning about candidates and issues, evaluate the various options against their own personal preferences, and then cast a ballot registering all of their candidate preferences. The simple fact that some voters meet these expectations has captivated political scientists for decades. The costs of voting - the effort to learn, the cognitive energy to evaluate, and the physical effort to register to vote, locate a polling center or obtain an absentee ballot, and then physically vote - seem to vastly outweigh the benefits of participating (Riker and Ordeshook 1968). It also is extremely unlikely that in any large-scale democracy a single voter could expect to cast the pivotal vote in any contest, further narrowing the enticement to participate. Even if voters are able to surmount the obstacles to participation and cast a vote, there is an additional challenge to the elections process how many votes can people really make at once? Each vote decision relies upon unique information about the candidates for a particular office. Voters must invest the time and energy to gather information about each office, develop preferences and then cast votes. Completing this task for a single office may be possible, but what about for multiple offices all at once? 1

3 The curious fact that people do participate and vote, even for a single contest, seems unusual when voters are confronted and asked to explain their understanding of the political system and the candidates they vote on. Citizens and voters consistently fail to demonstrate the most rudimentary knowledge about politics, either in its structure, functioning or even the names of people currently holding power (Delli Carpini and Keeter 1996). How, given a lack of incentives to participate, and clear evidence that voters are uninformed, can a democratic electoral system function? And even if people demonstrate that they are able to make a vote decision, can they make multiple simultaneous vote decisions all at once? Research has already demonstrated that concurrent campaigns decrease what voters can recall about congressional candidates after an election (Wolak 2009, Andersen 2009), so is there a point at which voters simply can no longer keep all the candidates straight, and must simply cast arbitrary votes? Low Information, Rational Voting The widespread solution to the participatory dilemma in voting where voters apparently know very little about politics, yet still vote has been that voters rely upon cues and heuristics to streamline the effort required to vote competently into a manageable size. By relying upon shortcuts, voters are able to assess candidates by simple cues, such as their political party membership (Campbell et al 1960), personal likeability (Sniderman, Brody and Tetlock 1991), outside endorsements (Lupia 1994) or by relying on retrospective evaluations of incumbents and the current state of the world (Key 1966, Fiorina 1981). A small cornucopia of shortcuts are recognized as being available to voters, all permitting decisions to be made regarding candidates with lessthan-full sets of information. 2

4 Among the very first identified was from the iconic American Voter, pointing to the clear cue of political identification when making vote decisions (Campbell et al 1960). Voters aware of which party they preferred, or identified with, are spared the burden of learning the details of policy space and candidate issue stances, because the simple information of which party a candidate represented would provide general indications for them. This alone can account for the vast majority of low information political decision-making witnessed in society. So long as people understand the political parties, can recognize a preferred party and identify the party a candidate represents, decision-making can be reduced to the simplest of functions. Party identification provides a prospective cue, allowing voters to select a candidate aligned with their own preferences. Another easy way to select a candidate without even having to understand parties is to simply evaluate the state of the world and reward/punish the party in power for that state (Key 1966, Fiorina 1981). This retrospective evaluation requires less work for a voter, because evaluating the state of the world requires no special knowledge on their part and can be done with simple pocketbook evaluations. If you feel that your life is better today than it was four years ago, you should probably retain the party in power. If not, you should probably replace them. The key information necessary, aside from recognizing the state of the world and whether it is preferable today compared to during the previous election, is the simple ability to recognize which candidates are the incumbents. Such retrospective voting may minimize the role of political campaigns too completely though, leaving little to no room for candidates themselves to influence their election. Clearly, campaigns play a role and candidates clearly are important figures, 3

5 drawing support in part by conveying a positive image among the electorate. Without relying upon issue stances, party cues or past evaluations, some voters seem willing and able to select preferred candidates based upon their personal likeability, or based upon their individual issue stances or by who else is supporting them (Sniderman, Brody and Tetlock 1991, Lupia 1994, Popkin 1994). By all of these measures, voters can learn a little about a candidate allowing them to surmise how they will behave once in office. Such shortcuts permit voters to use a bare minimum of knowledge to guess at future behavior of candidates, allowing them to select that candidate who will hopefully represent their preferences best. Weaknesses to Cognitive Shortcuts These shortcuts, while efficient, are not perfect, nor are they necessarily easy to use. Party knowledge, for example, is not a perfect proxy for candidate issue stances and likely becomes less accurate when used farther down the hierarchy of offices. Current party platforms are generated primarily by presidential candidates, forcing their political beliefs upon the rest of the party. On a national level, knowing the party platform or general stance of the party may simplify presidential selection, because of the close association of the two. For lower ballot candidates, even those proximal such as Governors, Senators and House members, the clear association may be lacking. A simple look at popular ratings of members of Congress, such as NOMINATE scores, reveals great variance within parties on the voting habits of their members. Party membership may connote closeness among its members, but only represents a general indication of individual candidate stances. This association, presumably, becomes weaker in lower offices that have less impact in the development of party stances. Within any specific 4

6 congressional district, knowing a candidates party identification may not reveal a great deal about that candidate s political beliefs or priorities. Particularly when voting for representatives that are expected to focus on local concerns, being able to associate a candidate with a national party (which by definition focuses on national concerns) may not provide sufficient insight into what the candidate represents. Using party identification in a less precise way, by simply identifying the party in power with the general tone of the country, rewarding that party and its members when things are going well, punishing them when things are poor, also suffers some weaknesses (Key 1966, Fiorina 1981, Rahn 2000). While simplifying voter decisionmaking, the rationality of such a heuristic for all levels of office is questionable. Selecting, for example, which party members are to blame for a faulty economy or disagreeable public policy may be far more complex than a specific candidates incumbency provides (Popkin et al 1976, Rudolph 2003). Nevertheless, voters do appear to use incumbency as a tool for selecting candidates, often punishing members of the President s party during midterm elections, and rewarding Presidential candidates and their Congressional running mates during positive economic times (MacKuen et al 1989). Rational or not, party identification and incumbency together provide two powerful cues facilitating vote choice. For more candidate specific cues, such as issue stances, personal likeability or other candidate-specific information, voters would require a much greater store of information when making a selection. In light of the recurrent finding that American voters typically lack much political information, this would seem beyond most voters, but this may not be the case. When dealing with candidate specific information, voters may 5

7 rely upon running tallies about candidates, aggregating likes and dislikes about the candidate to help them form a general impression, while discarding the individual bits of information after they are processed (Lodge 1989). In this way, voters form educated judgments about candidates, even while being unable to recall from where exactly those judgments arise, explaining the general low levels of political knowledge found on surveys. For on-line processing to be effective however, voters must first encounter information in sufficient degree to form these preferences between the candidates. There remains a minimum level of information that is required to make the shortcut effective. For all of these cognitive shortcuts, the assumption is that voters have encountered a sufficient level of information to form distinctions between the candidates in each particular elective office. Whether relying upon partisan cues, retrospective evaluations or some more particular candidate evaluations, voters must have at least encountered enough information to know something about candidates for each office. That information, in a candidate-centered election system, must be unique to each candidate and also applicable to the particular office they are competing for if it is to be truly relevant to the vote decision. The Electoral Environment When discussing the public s ability to make vote decisions, and especially vote decisions concurrently, one more factor is extremely relevant the distribution of political knowledge across offices. In campaigning and the spread of information, all offices are clearly not equal. Higher offices generally have much greater finances at hand that enable them to campaign. Gaining additional information about candidates and offices requires an effort upon the part of the voter to search out information, or at least 6

8 be receptive to that which is brought directly to them. This creates two interacting efforts during an election cycle that contribute to political knowledge: 1) campaigns attempts to educate voters about a candidate and 2) an individual s attempt to learn about all of the races they will be asked to vote on. Both of these processes differ by office, or at least are likely to. Campaigns widely differ based upon the office in question, in large part because candidates for higher offices typically have larger financial backing with which to advertise themselves. Higher offices spend many times the money spent by lower races, allowing them to get their information out to more voters, more often. In 2008, spending by all 27 Presidential candidates totaled $1.8 billion, while all 435 House races combined, totaling 1374 candidates, remained a mere fraction of that amount nationwide, $977 million. The average Senate candidate had $2.4 million to spend in 2008, while the average House candidate had less than a third of that, $700, Higher races also benefit from the relative ease of reaching their constituent base efficiently via mass media, whereas lower races must struggle to reach their audiences via media markets that are not contiguous with their districts. Every advertising dollar spent in a Presidential race is nearly assured to reach potential voters, whereas House candidates risk spending campaign dollars advertising in markets that overlap outside their constituency, wasting their efforts on citizens who cannot vote for them. Finally, Presidential candidates and Senate candidates benefit from notoriety and prestige, granting them access to free press in the media that lower candidates can only aspire to. The information environment simply is not as rich farther down the ballot. 1 According to Opensecrets.org, 7

9 A sufficiently motivated citizenry may overcome this. If voters are completely passive when living through campaigns, they will receive only those messages that they encounter. The great bulk of these will be from higher offices, that have greater finances and capability to get their messages out. Only if voters are willing to put in the effort to seek out information about lower offices will they obtain a comparable level of information about all of the offices they face. A major unanswered question is how voters attempt to learn about candidates, especially in regards to when multiple offices appear on the ballot. Motivated citizens who want to make informed vote decisions must attempt to learn something about the candidates in each race they will cast a vote in, overcoming differences in the information environment. Since the information environment tends to be skewed in favor of higher office candidates, voters would have to work harder to learn a comparable level of information about lower office candidates, because less information is available about them. If voters fail to do so, they will encounter less information about candidates for lower office, making even cognitive shortcuts even harder to use effectively. It is this search for information that I am concerned with. Little to no research exists on how voters seek out information about different types of offices, or what those search strategies are affected by. Voter decision-making on all levels requires a minimal base of knowledge about each office that is voted on. On the presidential level, political psychology has demonstrated that voters often rely upon cognitive shortcuts to make their vote choice manageable amid low levels of political information. For such strategies to also work for other offices, at a minimum we would have to see that voters seek out information about those offices in a similar manner as for presidential candidates. Based 8

10 upon the availability of information that is typical in most real-world campaigns, voters would actually have to search more actively for lower office information if they were to assemble a comparable level of information as encountered for higher offices. Even if they were to use the same cognitive short cuts for all races, voters have to put in a greater effort in gathering information in lower races to make such strategies effective. Methods In order to assess how voters search for information about candidates, I rely on the Dynamic Process Tracing Environment (DPTE), a web-based software application developed by Richard Lau and David Redlawsk with support from the National Science Foundation. DPTE permits researchers to create experiments using dynamic information boards that mimic a campaign and monitor how subjects choose to learn about the various candidates. Lau and Redlawsk have used similar designs to study Correct Voting (Lau and Redlawsk 1997), and voter information searches (Lau and Redlawsk 2006, Redlawsk 2004). Those studies looked solely at presidential elections, however. The DPTE system allows researchers to create fictional political candidates, complete with demographic information, political identities, issue stances and realistic backgrounds. Each distinct piece of information about the candidates is located within an information box. During the campaign portion of the experiment, these boxes scroll down the screen at a steady pace. If the subject decides to learn the information within the box, they click on it using the mouse and the box opens, filling the screen. When the subject has read the information, they then click the box again, closing it and returning them to the view of scrolling boxes. This is the standard method Lau and Redlawsk have 9

11 used over the past decade in exploring both Correct Voting and voter information searches. The primary contribution I make to Lau and Redlawsk s experimental design is the inclusion of additional contests into the electoral scenario. Their work has focused exclusively on candidates at the presidential level, running in isolation from other contests. In my scenario, I created a 2 x 2 x 2 research design creating 8 possible electoral scenarios that subjects could experience, but all of which included at least 2 offices simultaneously. Thus, subjects were forced to divide their information search among candidates and among offices as well, adding external validity to the scenario and allowing me to monitor what strategies subjects adopted to accomplish this, and how effective those strategies were. Sample In the spring of 2010, 279 research subjects were recruited from various political science courses at Rutgers University to participate in this study. Subjects participated either as a course requirement or to receive extra credit, depending on the course they were enrolled in. After arriving at the experimental lab, each subject was brought to a workstation, given a description of the experiment and an informed consent form, and then started on the program. Participation typically took about 1 hour. Of those original 279 subjects, data from 12 subjects were eliminated from the final data set; 4 because they experienced a flawed election 2 and 8 because they apparently discovered the ability to access the internet from the computers and did not actually participate in the 2 Due to a design error on my part, several election scenarios permitted subjects to view information about candidates whom did not actually appear on the ballot. The ability to access this outside information makes the data on how they structured their information search unusable. 10

12 campaign scenario, opening fewer than 10 information boxes over the course of the 20 minute campaign. The use of college students as subjects is always a risky proposition in experimental research and has its positives and negatives (Sears 1986). When looking at strictly cognitive or neurological functions college students can often be considered representative of the population as a whole, because such functions are typically outside of conscious control and thus factors such as age and experience become irrelevant. When asking subjects to make conscious choices however, including choosing how to search for information about political candidates, factors such as age and experience can make a difference. With age may come better understanding of how to cope with concurrent elections, or perhaps just different strategies. Even in recognition of this fact, with this research the ready availability and low cost of using college students was the prime motivation in their selection. As a first look, college students were selected as the natural choice. Manipulations and Innovation to Previous Research Each of the 8 scenarios to which subjects could be assigned presented multiple offices for subjects to learn about and then vote for. In each version, a race for the House of Representatives was contested, as well as an election for an executive office (either the President or Governor). Additionally, some subjects viewed a Senate race, as well. These four offices the House, Senate, President and Governor were chosen because they are typically the most visible races on a given ballot, providing voters with the most available information. In addition to the information boxes for the candidates for each political office, boxes for Current Events also appeared, filling in space during the campaign 11

13 and acting as a distraction (for more on Current Events, see below). The 3 major manipulations embedded in this experiment were: Senate Race 1. Presence of a Senate Race 2. Executive at top of the ballot President or Governor 3. Distribution of Information Equal or Realistic The presence of a Senate race on the ballot manipulates the number of offices subjects are asked to learn about. With no Senate race on the ballot subjects witness contests for two political offices, while in the Senate race manipulation they witness three. The difference between two and three races is not huge, but provides leverage for seeing how search patterns for information about one set of candidates differs based simply upon the presence of another set of candidates. Larger affects are likely to occur as greater numbers of offices, and less familiar offices are added to the ballot, but this design first examines these affects among the more prominent offices. President or Governor Manipulating what appears at the top of the ballot is achieved by manipulating the executive office being contested subjects vote for either the president or the governor. Presidential contests have been the norm in previous dynamic information board research and are what we know most about. In midterm election cycles however, 36 states typically contest their governor s election, adding this to the biannual federal races. Thus, in most areas of the country, voters must balance learning about congressional candidates with either presidential campaigns during one cycle, and gubernatorial campaigns during the midterms. This provides leverage in seeing whether the type of office that appears on the top of the ballot influences how people search for information about the candidates farther down the ballot does the presence of a presidential race 12

14 alter information searches about House candidates in different ways the presence of a gubernatorial race? Equal vs. Realistic A final manipulation embedded within this experiment altered the availability of information about the candidates. In every manipulation, each unique piece of information about each candidate appears at least once. In most campaigns however, information is available more than just once, but its availability often hinges upon campaign resources. Indeed, the current understanding of why voters can recall more about higher-office candidates than lower-office candidates is that higher offices, particularly the presidency, receive and can purchase much greater media attention than lower offices, particularly the House. To explore this question, some subjects received an equal amount of information about all of the candidates, with each box of information for each candidate appearing twice. For other subjects, a more realistic distribution of information was presented. For House candidates, each piece of information was only presented a single time, for senate and gubernatorial candidates they appeared twice, and presidential candidates had their information boxes appear four times each. While this does not precisely match the distribution of information typically available in campaigns, is does approximate it. 3 Current Events 3 Determining what would be a realistic distribution of information is exceedingly difficult, because the information environment is highly responsive to incumbency, candidate-specific attributes, region of the country and state and the competitiveness of the electoral environment. In general though, senate candidates typically at least double the spending of house candidates in any given area, while presidential candidates at least double the spending of senate candidates within a state. 13

15 An added feature of this experiment that has not been used before is the inclusion of Current Events boxes. 50 current events were created that had as little bearing on the political world as possible. They were mainly entertainment news ( MTV plays first music video in 10 years ), sports news ( Rutgers joins the Big 10 Conference ), or technology news ( Apple unveils new ithing ). They operate in an identical manner as the political information boxes, and opened when clicked, but rather than provide information on a candidate would explain the current event listed on the label. These boxes serve two functions: 1. Filler 2. Distractions The first function is to allow each manipulation to have the exact same number of boxes flow through the campaign and keep their length identical. Regardless of the offices on the ballot, this keeps each campaign at 20-minutes in length, giving all subjects identical time constraints to work under. The time left in a campaign until the election appeared on the screen in the upper right hand corner, counting down the minutes and seconds subjects had left to learn about the candidates. As a result of the inclusion of different numbers of candidates (senate manipulation) and from showing different amounts of repeat boxes (distribution of information manipulation) the campaign times could potentially be longer or shorter than one another based upon the manipulations a subject was assigned. In the real world, campaigns do not grow longer or shorter as a result of the length of the ballot, so in order to compensate for this current event boxes are used to fill in the gaps, making each campaign 20 minutes long. These boxes were designed to be as politically neutral as possible, so that reading them would have no conceivable impact on how the viewer perceived the ongoing campaigns. 14

16 The other role of the current events was to allow subjects to ignore and escape the campaigns. Most political science experiments lock subjects into a political task, forcing them to focus upon the objective at hand. This clearly weakens the external validity of an experiment in important ways. It forces the subject to perform at their maximum capabilities, focusing upon political stimuli in a manner that simply does not exist in the real world. This biases results, often in the researcher s favor, because it only provides feedback on what subjects are capable of when they experience politics in isolation and solely focus on the matter before them. In the real world however, people opt to focus on politics as a matter of choice. Most people endure a long workday and then come home to enjoy relaxation and calm, following politics alongside other more enjoyable activities. Thus, the current events are here designed to mimic the distractions of life they are designed to be funny, lighthearted and amusing. They are time-wasters. Just like surfing the internet or watching TV, they are present as a more enjoyable and enticing distraction than learning political information about candidates. Subjects are free to view as many or as few as they wish, but doing so prevents them for learning more about the candidates they will be asked to vote for. Summary In total, the 279 subjects were randomly assigned into one of eight conditions formed by the 2x2x2 presence of the manipulations. In each scenario, an equal number of total information boxes were available (400), and each attribute about every candidate was viewable at least once. It was completely up to the subjects to choose which and how many attributes about which candidate they wished to view. This design permits 15

17 information searches to be compared between matching scenarios, where similar number of candidates (though of different types), similar sets of candidates (though sometimes with an additional contest as well) and similar complete ballot structures (though with different rates of information availability for the offices) compete against each other in a total information environment that presents an identical set of total information boxes. Hypotheses I expect that voters approaching concurrent elections will react rationally towards the information environment they encounter, seeking out as much information as they can about campaigns before they have to make a vote decision. However, I believe that voters are also limited in how much political information they are willing to consume at once, creating a cap on the total level of information search they will conduct. Further, I feel that these limitations will lead voters to gravitate to focus a greater level of attention on those offices that are perceived as more important, or more powerful, than others. From these basic precepts, I derive several hypotheses about how voters will react to my experimental manipulations listed in table 1, below. <<<<<< Insert Table 1 about here >>>>>> Results Three-way Analysis of Variance In order to control for the three major manipulations present in this experiment - the presence or absence of a senate race, the presence of either a governor or president as the lead executive office, and the distribution of information between the offices - I employ three-way analyses of variance. Three-way ANOVA s provide both statistical tests of the effects of the manipulations as well as descriptive results by controlling for 16

18 the presence of the various manipulations and testing for both main and interaction effects. Table 2, below, shows the descriptive statistics for the number of attributes viewed in total, and for each office, in each of the manipulation conditions. <<<<<< Insert Table 2 about here >>>>>> The descriptive statistics reveal a great deal of movement between the average number of attributes subjects viewed for each set of candidates, depending upon which set of manipulations they experienced. The total information encountered ranged from a high of attributes in the fourth column when the presidency, a senate and congressional contest were fought with equal information - to a minimum of attributes when only the presidency and congress were contested, but under with realistic information availability. This is a vast difference in the average information encountered between these conditions, amounting to more than a 20% shift in information from the minimum to maximum. In a real-world environment, this implies that under different conditions voters overall level of political information may fluctuate by a similarly large degree under different electoral scenarios. The differences withinoffices are also dramatic. Views of the executives attributes range from a minimum of to a high of a greater than 50% increase in information access depending on the manipulation conditions. Senate searches are more stable, which is expected considering its very presence is a manipulation and that its information availability does not change. It experiences only a slight shift from to views at its extremes. Congressional information searches also show wide variability, gaining the deepest search when contested only alongside the presidency in the equal information condition (

19 attributes viewed) and falling below half of that rate when contested alongside both the gubernatorial and senatorial candidates in the realistic condition (14.04 attributes viewed). <<<<<< Insert Figure 1 about here >>>>>> These findings are also viewable graphically. Figure 1 presents a bar graph separating the average total attributes viewed by subjects into their component parts, by manipulation. The bar chart allows quick comparison between the depth of information search for each office, and with total information search in general, and appears to show large differences. In order to determine if these differences are actually statistically significant, I also use the significance tests generated by the three-way ANOVA procedure, presented below in table 3. <<<<<< Insert Table 3 about here >>>>>> Subjects responded to the manipulations in many ways. One of the interesting findings here is that not all of the manipulations affected all of the various information searches. Each manipulation had effects on particular types of searches, and each search responded to different manipulations. The one exception to this was the depth of search for Senate candidates, which produced no substantive findings here. Both the lack of true manipulations in this condition, alongside the low number of subjects who experienced a Senate contest, made it unlikely that statistically significant results will appear here. The other offices provide much more information about how changes in the electoral environment can alter information searches. Table 3 shows that subjects altered their total search for information in response to only one of the manipulations, the distribution of information. Looking at both table 2 18

20 and table 3 together can explain both the significance and the direction of changes, but it is much easier to understand how information searches responded to the manipulations by plotting these results graphically. In order to see how subjects altered each search, I plotted the estimated marginal means, taking into account the three manipulations present. The results for the overall information search are presented below in Figure 2. <<<<<<Insert Figure 2 about here >>>>>> As Figure 2 shows, the information distribution manipulations had significant effects upon how many attributes subjects viewed in total, decreasing the number of attributes viewed when there was a more realistic, rather than equal, availability of information about the various offices. While it made no statistical difference whether the Governor or President was on the ballot, or whether the Senate also was contested, subjects did view significantly fewer items when there was a realistic distribution of information rather than an equal distribution. The direction of all four lines in figure 2 also suggests that when the senatorial contest appeared, subjects did view more information, though table 2 indicates that it is not a significantly greater amount. This is surprising, because with the senate race on the ballot, the number of total attributes available to view increases by 25%. Subjects did not respond to this by altering their overall information search however, but instead average a statistically identical number of attributes viewed. Hypothesis H1 proposed that when an additional contest was added to the ballot subjects would seek out more information overall, but this is not what this analysis reveals. Hypothesis H2 further suggested that when the presidency was contested subjects would also seek out greater amounts of overall information. Both of these hypotheses are rejected here. Only hypothesis H3 is 19

21 confirmed, finding that in the realistic distribution of information, subjects viewed fewer attributes overall. These results suggest that subjects and, via them, voters, may indeed have a set upper limit for information search during a campaign. When more offices and candidates appear, voters respond by dividing their search to a greater extent between the various campaigns, but not searching for more information, overall. 4 <<<<<<Insert Figure 3 about here>>>>>> Looking at how subjects searched for information about each office is also instructive, and points to some clear differences in the level of information subjects seek out each type of office. Figure 3 clearly shows a difference in the number of attributes viewed for congressional candidates depending upon which conditions subjects faced. Those subjects in the equal information distribution condition viewed significantly more attributes than did those in the realistic-distribution condition. By reducing the availability of information about congressional candidates, subjects do lose the ability to become aware of what those candidates stand for. Similarly, candidates who are unable to put sufficient advertising into the campaign environment must similarly suffer, with voters unable to learn what they stand for, and forced to either guess or draw inferences from other information. The inclusion of a Senate contest also significantly decreases the depth of congressional search, again suggesting that when additional offices appear on the ballot, overall information search for each is divided up between the offices, lowering the depth of information search about each one. 4 This could also be a result of subjects hitting a time limit in the experiment, where they viewed as many items as possible, but simply ran out of time to view more. This is highly unlikely, because subjects on average only viewed about 25% of all the boxes that scrolled before them, spending little more than half of the available time doing so. Most subjects had a great deal of extra time left to open more information, but simply chose not to. 20

22 <<<<<< Insert Figure 4 about here >>>>>> In contrast, none of the manipulations altered the search for information about Senate candidates. Though only half the sample viewed a Senate contest, making it more difficult to achieve statistical significance, the line graph in figure 4 demonstrates that there was also no substantive effect of the manipulations, and subjects tended to view the same amount of information about Senate candidates regardless of the election environment they were in. This is not very surprising, because the Senate race was not subject to the manipulations per se. When it appeared on the ballot, there were always two other contests, so it should have received a constant depth of search, which it did. When appearing, the senatorial contests also produced a constant availability of information, so the only effects that can be expected there are in response to changes in congressional or presidential information differences. Nothing is found in these results, which indicates that the depth of Senate information search was largely immune to outside campaign spending in this experiment. <<<<<< Insert Figure 5 about here >>>>>> The final office to examine, the Executive, is affected by two manipulations, but in this case it is the presence of the Senate contest and which executive appears that make a significant difference. Both of those manipulations are statistically significant, while the distribution of information manipulation is just below the standard threshold for significance. Figure 5, below, shows that the presence of a Senate contest drastically decreases the depth of information search for presidential candidates. While the difference in the level of search based upon which executive office appears is difficult to determine in this set of charts, there is a clear difference in the effect that the distribution 21

23 of information had upon the executive when they were a governor compared to a president. For the gubernatorial candidates, the two lines overlap, indicating that the information search for gubernatorial candidates was unresponsive to the information availability condition. For the presidential candidates however, the lines are separate and parallel, indicating that the increased availability of information about the presidential candidates did encourage subjects to view more information about them, though not significantly so. The three-way ANOVA for the executive answers hypothesis 4, which proposed that, controlling for everything else, subjects would search for more information about the president than the governor. This statistical test in the analysis of variance shows that there is a statistically significant difference in the depth of search for information about these two offices, and the descriptive statistics indicate that it is in favor of the presidency. This hypothesis is thus confirmed. While the addition of another contest, the Senate, decreased the depth of information searches for both congressional and executive candidates, the campaign spending manipulation did not produce strong results. The distribution of information creates variance both in how offices were able to campaign, with the realistic condition increasing campaigning by the presidential candidates, while decreasing campaigning by the congressional candidates. Conclusions This analysis presents a first rough look into how voters learn about multiple candidates campaigning for different offices all at once. In it, I demonstrate that voters do appear to have an upper limit on their total search for political information during 22

24 campaigns. This limit results in voters learning a rather stable amount of information about politics, regardless of how many offices are contested at once. As a consequence, when more offices are contested at once, voters learn less about each set of candidates on the ballot. Voters are also responsive to the information environment around them, choosing to seek out more information about campaigns that have more active campaigns, while being unable to learn a great deal about less active campaigns. Additionally, voters search out different levels of information about different offices on the ballot, prioritizing information search about executives over legislative offices, and Senate candidates over House candidates. Subjects have two options when presented with additional contests during a campaign cycle. First, they could increase the level of information they view in relation to the number of offices being contested, adding on to the number of views in an additive fashion with no indentified limits. This would imply that voters always attempt to learn a certain level of information about each contest they encounter, altering their attention to politics proportionally to the number of offices being contested, and increasing their information search to allow them to learn a constant level of information about each office on the ballot. The other alternative is less ambitious, suggesting instead that subjects generally operate under an identifiable upper boundary of information search. This boundary prevents voters from searching for a constant set of information about all of the candidates, but allows them to distribute their information search equivalently (or according to some set ratio) between the offices. If voters have a set maximum attention level that they are willing and able to expend when monitoring campaigns, they must 23

25 shift their allocation of search somehow among the available offices, decreasing the actual information they learn about the individual candidates, whether that means implementing equal decreases in search across the board, or systematically choosing which candidates get more attention than others. The consequence of the first style of information search is that, while voters may be able to make well informed decisions, they would require a great deal of effort to do so. The consequence of the second is that, while maintaining a comfortable level of information search, voters faced with many contests may not be able to learn enough about all of the office-candidates Between these two options, the latter receives greater support at this stage. Simply looking at the line graphs of the changes in the total information search along with the related significance levels suggest that subjects did not respond to the offices on the ballot when searching for information overall. There is no evidence that the addition of 54 new unique information items for the senate caused subjects to view more information boxes in a sufficient quantity to support the first option. The only manipulation that affected total information search was that for the distribution of information between candidates. Subjects exposed themselves to greater levels of information when all of the offices campaigned to an equal degree, presenting each piece of information about the candidates twice, rather than when the presidential candidates (when appearing) presented their attributes four times to the House s single appearance. Subjects may have simply recognized the overwhelming presence of repeat information, and had a limit to what they were interested in learning about each candidate. In response, they chose to forego the opportunity to seek out additional information about the president, because they have already learned enough, while being 24

26 unable to open more boxes about the House because of their unavailability. This would lead subjects to open boxes for the president at roughly the same rate between conditions, while being able to open fewer for the House, leading to a significant difference overall (the Senate and governor did not change between conditions, so no difference would be expected). That is what the results point to. In real world elections, a scenario in which only two or three offices is contested is exceedingly rare. States currently attempt to minimize the costs of running an election by placing as many offices as possible on the ballot at once. If the results found here generalize further down the ballot, this poses great challenges to our current understanding of democracy and voting. If voters respond to concurrent elections by learning less and less about the offices they face, and particularly less about those offices with the least prestige, power and ability to raise campaign finances, at some point they will pass below a minimum threshold of information necessary to use even the most basic heuristic. Unable to identify any information about a candidate, voters must cast votes based upon the information they view on the ballot typically the name, incumbency and party of the candidates. Without any other information about the candidates, voters must guess, using heuristics that are better tailored for higher-level offices. This guesswork makes voting at lower levels of the political process into a blunt tool, basing state and local decisions upon the understanding that voters develop of national level politics. Local political matters become susceptible to national level forces that have little relevance to the powers and responsibilities of state and local government. In a world where dozens of states face structural deficits developed over years of poor governmental decision- 25

27 making, and millions of voters wonder who was responsible, this may provide a clue as to the source of current troubles. When elections are clustered together, voters pay little attention to the bottom of the ballot, learning very little about the actions and proposals of state and local officials. This leaves these offices largely unmonitored, and puts the occupants of those offices in a peculiar position. If they do little to attract attention, they are likely to win reelection and keep a position they enjoy. If they do make bold moves however perhaps attempting to resolve societal problems through tough choices, such as tax increases or spending cuts those actions will be noticed by the public. As the only bits of information that voters receive, because their attention remains largely at the top of the ballot, such bold actions result in negative consequences on Election Day, turning lower-ballot elected officials out of office when they make bold moves. The only rational strategy that remains is to do little to attract negative attention, even when the consequences for society are dire. Concurrent elections, by exceeding the ability of the average voter to gather enough information about all the offices and candidates, make lower office-holders less likely to act in ways that may necessary for the public good. Ironically, under the design of the constitution, these are the very offices that are supposed to be most responsive to public will. 26

28 References Andersen, David J The Electoral Environment and House Races: Political Knowledge, Political Interest and Government Evaluations as a Function of Electoral Density. Paper presented at the 2009 annual meeting of the Midwest Political Science Association. Campbell, Angus, Philip E. Converse, Warren E. Miller, and Donald E. Stokes The American Voter. Chicago: University of Chicago Press. Conover, Pamela Johnston and Stanley Feldman. Candidate Perception in an Ambiguous World: Campaigns, Cues, and Inference Processes. American Journal of Political Science, Vol. 33, No. 4 (Nov., 1989), pp Delli Carpini, Michael X. and Scott Keeter What Americans Know About Politics and Why it Matters. Yale University Press, New Haven, CT. Downs, Anthony An Economic Theory of Democracy. New York: Harper & Row. Fiorina, Morris P Retrospective Voting in American national Elections. New Haven, CT: Yale University Press. Funk, Carolyn L. Bringing the Candidate into Models of Candidate Evaluation. The Journal of Politics, Vol. 61, No. 3. (Aug., 1999), pp Green, Donald P., Bradley Palmquist, and Eric Schickler Macropartisanship: A Replication and Critique. American Political Science Review, 92(December): Key, V.O The Responsible Electorate. Cambridge, MA. Harvard University Press. Lau, Richard R. and David P. Redlawsk "Voting Correctly." American Political Science Review, 91(September): Lau, Richard R. and David P. Redlawsk "Advantages and Disadvantages of Cognitive Heuristics in Political Decision Making." American Journal of Political Science, 45(October): Lau, Richard R. and David P. Redlawsk How Voters Decide: Information Processing During Election Campaigns. New York: Cambridge University Press. Lodge, Milton, Kathleen M. McGraw, and Patrick Stroh An Impression Driven Model of Candidate Evaluation. American Political Science Review 83(June):

29 Arthur Lupia Shortcuts versus Encyclopedias: Information and Voting Behavior in California Insurance Reform Elections. American Political Science Review 88: Lupia, Arthur, Mathew D. McCubbins, and Samuel L. Popkin Beyond Rationality: Reason and the Study of Politics. In Arthur Lupia, Mathew D. McCubbins, and Samuel L. Popkin (eds.), Elements of Reason: Cognition, Choice, and the Bounds of Rationality. New York: Cambridge University Press. MacKuen, Michael B., Robert S. Erikson, and James A. Stimson Macropartisanship. American Political Science Review, 83(December): Popkin, Samuel The Reasoning Voter: Communication and Persuasion in Presidential Campaigns. Chicago, IL. University of Chicago Press. Rahn, Wendy Affect as information: The Role of Public Mood in Poltitical Reasoning. In Arthur Lupia, Mathew D. McCubbins, and Samuel L. Popkin (eds.), Elements of Reason: Cognition, Choice, and the Bounds of Rationality. New York: Cambridge University Press Redlawsk, David P What Voters Do: Information Search during Election Campaigns. Political Psychology 25(August): Riker, William H. and Peter C. Ordeshook A Theory of the Calculus of Voting. American Political Science Review 62(1): Rudolph, T. J Who's Responsible for the Economy? The Formation and Consequences of Responsibility Attributions. American Journal of Political Science, 47: Sears, D. O College sophomores in the laboratory: Influences of a narrow database on social psychology's view of human nature. Journal of Personality and Social Psychology, (51) Sniderman, Paul M., Richard A. Brody, and Philip E. Tetlock Reasoning and Choice: Explorations in Political Psychology. New York: Cambridge University Press Wolak, Jennifer The Consequences of Concurrent Campaigns for Citizen Knowledge of Congressional Candidates. Political Behavior, 31(2),

30 Table 1. List of Hypotheses Tables and Figures Hypotheses about total information search H1 When the Senate contest appears, subjects will search for a greater total level of information. H2 When the presidential contest appears, subjects will search for a greater total level of information. H3 When the realistic distribution of information appears, subjects will search for a lesser level of overall information. Hypotheses about information search for each office H4 When the President appears on the ballot, subjects will seek out more information about the executive office. H5 When the Senate appears on the ballot, subjects will seek out less information about each office on the ballot. H6 When there is a realistic distribution of information, subjects will seek out more information about the executive. H6a When there is a realistic distribution of information and the President is on the ballot, subjects will seek out more information about the executive. H6b When there is a realistic distribution of information, subjects will seek out less information about the House. Table 2. Three-way ANOVA descriptive statistics for total number of attributes viewed by subjects Equal Distribution of Information Realistic Distribution of Information Pres and House Only Gov and House Only Pres, Senate and House Gov, Senate and House Pres and House Only Gov and House Only Pres, Senate and House Gov Senate and House Executive (11.59) (10.87) (8.89) (7.51) (10.87) (13.98) (8.59) (11.70) Senate (8.00) (8.70) (9.83) (11.97) House (11.15) (11.68) (8.52) (8.06) (8.94) (12.31) (7.02) (9.11) Current Events (11.67) (11.37) (11.75) (9.65) (10.07) (10.11) 5.14 (5.02) (10.33) Total Items (22.66) (24.13) (26.11) (20.97) (19.36) (24.85) (20.53) (26.63) 29

31 Table 3. Three-way ANOVA of number of attributes viewed, by office Corrected Model Intercept Presidential Race Senate Race Realistic Pres*Sen Pres*Real Sen*Real Pres* Sen* Real Adjusted R Squared *** - >.001 ** - >.010 * - >.050 Total House Senate Executive Current Events *** *** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** * 0.003** *** 0.000*** 0.003** 0.015* 0.000*** *

32 Figure 1. Total information search, by manipulation 31

33 Figure 2. Three-way ANOVA of total attributes viewed 32

34 Figure 3. Three-way ANOVA of House attributes viewed 33

35 Figure 4. Three-way ANOVA of Senate attributes viewed 34

36 Figure 5. Three-way ANOVA of Executive attributes viewed 35