Stories, Science, and Public Opinion about the Estate Tax. John Sides Department of Political Science George Washington University

Transcription

1 Stories, Science, and Public Opinion about the Estate Tax John Sides Department of Political Science George Washington University June 2010 Abstract The majority of Americans opposes the estate, even though it affects only a small number of wealthy people. Scholarly investigations of this puzzle have debated the role of knowledge about the. Do at least some people support the estate because they are ignorant of its limited scope? If people were better informed, would more of them support this? Via experiments, I demonstrate the causal effects of information about who pays the estate, as well as arguments for and against it. I find that this information does reduce opposition to the estate, and that it does so primarily among the poorer conservatives and Republicans. The combination of this information and various arguments also tend to shift opinion in favor of the. These findings contrast with previous research and suggest that knowledge (or lack thereof) is central to understanding public opinion about the estate.

2 1 Many facts about public policy elude citizens. Policies can be difficult to understand, politicians have little incentive to educate, and citizens have little incentive to learn. This much is well-understood. Less well-understood are the consequences of learning facts. Scholars debate whether citizens are open to new facts and willing to change their minds accordingly, whether citizens appreciate new facts only when those facts confirm existing opinions, or whether citizens even need factual information to make good decisions. If citizens know more about policy, one story goes, they will make better decisions better for themselves, because they can more readily align their preferences with their interests, and perhaps better for society, if they tend to prefer the policies most beneficial for everyone. To address this debate, I focus on public opinion about the estate. Although just one element of American fiscal policy, the estate has attracted considerable political and scholarly controversy. The political controversy continues apace because the most recent reforms to the estate, in 2001, were not permanent. Both the rate on inheritances and the amount exempt from the will revert to 2002 levels in The scholarly controversy revolves around the public s knowledge of the estate and the large disjuncture between people s apparent interests and their opinions. Why does a majority of the public want to repeal a that few of them pay? Is it because the public is misinformed about who actually pays the? Or are deeply held values driving opinion, values that would neutralize the effects of correct information, were it provided? In answering these questions, scholars have differed on nearly every conceivable issue: the malleability of opinion, whether the public s lack of information about the is consequential (and for whom), and ultimately the broader implications of public opinion for public policy. Despite these differences, extant studies have one similarity: they do not estimate the causal effect of information about who pays the estate. They rely instead on measures of general political knowledge which are related to, but distinct from, knowledge of the estate or on

3 2 observational measures of knowledge about the estate, which could themselves be endogenous to opinion. A central contribution of this study is to estimate a causal effect via experiments embedded in surveys, thereby helping to answer a crucial question: Would more people support the estate if they knew that only a few rich people have to pay it? This study also addresses a second question: do facts affect opinion about the estate, even when they are paired with arguments that appeal to underlying values rather than offering correct information? Scholars have often focused on the isolated consequences of learning new facts, but political debate about public policies mixes facts with other kinds of appeals, often ones that ignore or even misrepresent facts. The debate about the estate is no exception. Some scholars have argued that rhetorical arguments about the estate matter more than facts, that stories trump science (Graetz and Shapiro 2005). Experiments allow me to test this claim by including treatments that combine facts with arguments both for and against the estate. Finally, this study investigates this question: whose attitudes shift in response to correct factual information? The answer speaks first to the politics of the estate. If factual information polarizes people along partisan or ideological lines, then debate about the estate is likely to grow more intense and compromise more difficult. But if correct information leads people to positions counter to their ostensible predispositions, then a bipartisan consensus could result. Second, the answer speaks to theories of information-processing and especially the debate about the degree of bias or motivation in how people reason about politics. Do people tend to reject factual information that contradicts their predispositions or preexisting views? Or can countervailing information actually be persuasive? My findings constitute a significant departure from previous literature. Correct information about who pays the estate does reduce opposition to the. In fact, this correct information appears no less powerful than rhetorical arguments. Stories do not appear to trump science. I also

4 3 find that this correct information affects conservatives and Republicans more than liberals and Democrats suggesting that it not only enlarges support for the but makes that support more bipartisan. These findings not only speak to high-profile scholarly and political debates about why the estate has proven so unpopular, but also enhance our understanding of how political values, correct information, and rhetorical arguments affect policy attitudes. The Origins of Opposition to the Estate Tax Why does a majority of people oppose a that costs very few of them anything, and may provide them with important benefits when that revenue is spent on various programs? A starting point in the scholarly debate over this question is the public s knowledge of the estate. One perspective implicates a lack of knowledge about policy and the estate in particular. Obviously, low levels of knowledge about politics are a familiar finding (see, e.g., Neuman 1986; Delli Carpini and Keeter 1996), but policy may be particularly challenging. Bartels (2005: 21) argues that most ordinary citizens are remarkably ignorant and uncertain about the workings of the system and the policy options under consideration, or actually adopted. 1 The estate appears to be no exception. Slemrod (2006) finds that a majority of people either believe that the estate affects most families (49%) or do not know how many families it affects (20%). But other scholars argue that the public actually knows more than these studies would suggest. Krupnikov et al. (2006) asked respondents a multiple-choice question about the fraction 1 Using Delli Carpini and Keeter s list of 112 knowledge items related to domestic policy (pp.80-81), I calculated the mean percent who answered the 10 items about policy correctly, compared to the percent who answered the other items correctly. An average of only 27% answered the policy items correctly, compared to 42% for the other items.

5 4 that pays the estate and found that just over one-third (36%) of respondents answered correctly. 2 Moreover, this fraction increased to 47% in an experiment where subjects were offered $1 for a correct answer (see also Prior and Lupia 2008). These authors assert that traditional political knowledge questions on surveys underestimate the true extent of knowledge, including knowledge about the estate (for a rejoinder, see Bartels 2007). A more important question than what the public knows is whether knowledge matters. Some studies find that facts change opinions (e.g., Gilens 2001), while others find that facts do not (Berinsky 2007; Kuklinski et al. 2000). Does some support for the estate derive from misinformation, particularly about who has to pay the? Slemrod (2006) estimates that a person who believes that most families pay the estate is 10 points more likely to oppose the, compared to someone who believes that few families pay the estate. Although cautious about claiming that misconceptions cause policy preferences, he nevertheless concludes that betterinformed voters would be much less likely to support these reforms (72). Others, however, are skeptical. Bartels (2005) downplays the importance of (mis)information. First, he finds that general knowledge of politics, as measured by interviewers assessments, does not affect attitudes toward the estate : support for repealing the inheritance was virtually constant across information levels, controlling for partisanship and family income (25). Second, he suspects that a better informed populace one disabused of the misperception that most people have to pay the estate would still not embrace the estate : Would correcting this misperception produce widespread public support for the estate? That is much less clear. Americans have always found the juxtaposition of death and es peculiarly unsettling (Bartels 2 The correct answer in their question was less than 5% of all Americans. The other possible responses were 95%, 70%, 50%, and 25%.

6 5 2006: 411). This conclusion is shared by Krupnikov et al, despite their different views about the level of knowledge about the estate : Many supporters of progressive ation and liberal intellectuals believe that if only citizens know more about the operation of the estate, the masses would come to support the same forms of ation that they do such beliefs constitute wishful thinking (435). Graetz and Shapiro also argue that correct knowledge would not increase support for the estate : the opponents of estate repeal failed to grasp that in politics, science is never enough (226). Why might correct information have little impact on attitudes? Three possible answers emerge from extant literature, all of which emphasize how values and other political predispositions structure attitudes and guide information-processing. First, the direct effect of values on opinion about the estate is evident in every study (Slemrod 2006; Bartels 2006; Krupnikov et al. 2006). Party identification and ideology are strongly associated with attitudes, with Republicans and conservatives more likely to oppose the. Indeed, the strong impact of predispositions contrasts with the weak impact of economic interests: poor people are no more likely than rich people to favor the estate. If predispositions anchor opinions, then there may be little information can do to move opinions. Citizens may approach the estate as motivated reasoners (Lodge and Taber 2000). In the minds of its opponents, for example, the estate has a negative affective charge perhaps because of this juxtaposition of death and es and learning that they do not have to pay the would not convince them otherwise. A second view sees potential for correct information to matter, but only conditional on predispositions (Krupnikov et al. 2006; Lupia et al. 2007). In this view, predispositions affect how people process political information; specifically, more informed partisans are likely to adopt their party s point of view the pattern of polarization described by Zaller (1992). Knowledge of who pays the estate makes all partisans more likely to support it, although the effects are about twice

7 6 as large among Democrats as among Republicans (Krupnikov et al. 2006: ). These findings are of a piece of motivated reasoning: given their predispositions, such as the strength of their opposition to many forms of ation, Republicans may be more resistant than independents or Democrats to information that casts the estate in a favorable light. It would certainly be no surprise that partisans process new information in different ways (e.g., Bartels 2002; Campbell et al. 1960). Ultimately, polarization implies a small aggregate effect at best: the balance of Democrats and Republicans is close enough to parity that if they react to new information by moving in opposite directions, the overall distribution of opinion will not change much. Third, opinions about the estate may be more susceptible to arguments based on values than to factual information. Graetz and Shapiro (2005) argue that opponents of the estate made headway by arguing that it is unfair to people after death and interfere with a family s right to pass on accumulated wealth to its children. Graetz and Shapiro criticize estate supporters for the dubious assumption that Americans would respond more to facts (such as who actually has to pay the estate ) than values: the groups working against repeal assumed that if people could only be enlightened about scientific realities, their hearts and minds would follow (227). Some other research supports their view. Citizens often reason from moral values in evaluating public policy (Stoker 1992) and concerns about fairness emerge quickly when citizens discuss the estate (Hochshild 1980). Moreover, citizens seem to consider correct information about who pays the an unconvincing argument against the estate, according to a poll conducted for the liberal research group OMB Watch (Greenberg, Quinlan, and Rosner 2002). 3 3 When told that The estate only applies to the wealthiest few, less than 2 percent of the estates in this country, 61% of respondents ranked the argument at 5 or below on a 0-10 scale where 10 signified extremely convincing.

8 7 Expectations Taken together, these perspectives generate several testable propositions. The first involves the underpinnings of attitudes toward the estate. Opinions about the estate should derive from predispositions such as party identification and ideology, but less so from economic interests, as captured by income. This expectation dovetails with the venerable finding that political attitudes depend more on values than on self-interest (see Citrin and Green 1990). A second set of propositions concerns the effect of information on attitudes. On the one hand, Slemrod s findings suggest that correct information about the estate lowers opposition to it. On the other hand, most other studies of the estate suggest the information is unlikely to affect preferences especially if opposition to the estate is firmly anchored in political values. The empirical analysis will adjudicate between these competing expectations. Third, there are two expectations about the conditional effects of information. One concerns economic interests: although they often fail to influence opinion, they can do so when the benefits and costs of a policy are clear (Citrin and Green 1990). Thus, information about who pays the estate should strengthen the relationship between interests and attitudes, making poorer respondents more supportive of the estate once they learn that only a few wealthy people have to pay it. The second expectation concerns the conditioning effect of partisanship. Krupnikov et al. s (2006) findings suggest that specific information about the estate should have larger effects among Democrats than Republicans, thereby polarizing citizens along partisan lines. The fourth set of propositions concerns the impact of values-based arguments on attitudes toward the. Graetz and Shapiro suggest that these arguments matter more than information about who pays the estate, especially when the argument opposes the and centers on concerns about the fairness of the estate. When this information and values-based arguments

9 8 against the estate are presented by themselves, attitudes should therefore shift more in response to values than the information. Presumably, the greater impact of values would also be evident when this information and values are presented together. This would confirm that stories trump science. Data To evaluate these expectations, I draw on original data from the 2007 and 2008 Cooperative Congressional Election Surveys (CCES). These were surveys of voting-age Americans conducted in the fall of each year. Each survey had two waves, bracketing the November elections (naturally, few states or localities held elections in November 2007). 4 The survey was conducted on-line by YouGov/Polimetrix. YouGov selects respondents from its PollingPoint panel a large sample of people who have agreed to take periodic surveys by first drawing a random sample from the Census Bureau s American Community Survey (ACS). They then match each individual from the ACS to his or her most similar counterpart in the PollingPoint panel. This matching procedure, combined with traditional post-stratification weights, produces a sample that is nationally representative in terms of age, sex, race, and education. Studies comparing samples in on-line surveys to traditional probability samples have found some differences in the marginal distributions of variables and in their inter-relationships, although there is no consensus on the seriousness of these differences and whether they imply flaws in any particular on-line survey methodology (Malhotra and Krosnick 2007; Sanders et al. 2007). These concerns are less salient for this study, which focuses on randomized experiments within these 4 The response rate for the entire 2008 CCES, encompassing the interviews done for all participating universities, was 47% (using the AAPOR RR1 formula).

10 9 surveys. I have no reason to suspect that the CCES sample affects the generalizability of the experiment s effects, especially compared to the convenience samples (e.g., of college students) often employed in experiments. That said, because the American Association of Public Opinion Research recommends that researchers avoid using nonprobability on-line panels to make inferences about population values (Baker et al. 2010), I describe the results of these experiments in terms of these particular samples and respondents and avoid terminology ( Americans, public opinion ) that would suggest broader inferences. The first wave of the 2007 CCES (N=1000) and the second wave of the 2008 CCES (N=829) included separate experiments that provided respondents with different information or arguments before ascertaining their attitudes toward the estate. In 2007, attitudes were measured using this American National Election Study (ANES) item: There has been a lot of talk recently about doing away with the on large inheritances, the so-called estate. Do you favor or oppose doing away with the estate? 5 Below I discuss the information that preceded this 5 This way of posing the question has been criticized. Hacker and Pierson (2005) argue that public opinion about cuts needs to be measured by making explicit the trade-offs (e.g., reduced spending on favored programs, higher deficits). Indeed, support for the estate is lower when the trade-offs are made explicit (Birney, Graetz, and Shapiro 2006; Frank 2007). Moreover, significant fractions of people support reforming but not repealing the (Birney, Graetz, and Shapiro 2006; Penn, Schoen, and Berland 2005). I use this formulation of the question to be consistent with previous literature and because, as Bartels (2005) notes, politics often calls upon the public to vote yea or nay rather than considering trade-offs. Another issue involves the use of the term death in the survey question wording. I do not incorporate that term, as analysis of a survey experiment in the 2002 American National Election Study suggests that it makes little difference in the distribution

11 10 question in the various experimental conditions. In 2008, respondents were first asked about their knowledge of the estate : In the United States, there is a on large inheritances that is sometimes called the estate. What percent of Americans do you think have a large enough estate to be subject to this? If you don t know, just give your best guess. 6 Respondents gave their estimate as a percentage from This question is different from knowledge measures on previous surveys about the estate, which have employed less specific quantitative terminology e.g., whether most or a few families have to pay (see Slemrod 2006) or have provided specific percentages in a multiple-choice framework e.g., about 95% of all Americans, about 70% of all Americans, etc. (see Krupnikov et al. 2006). The CCES item more precisely measures respondents knowledge. Respondents were then asked about their preferences: There has been a lot of talk recently about doing away with the estate. Do you favor or oppose doing away with the estate? Again, there was prefatory text provided in various experimental conditions, which are discussed below. 7 of opinion (see Bartels 2008: 199). Schaffner and Atkinson (2009) report similar results from a 2003 Kaiser Family Foundation poll. Graetz and Shapiro (2005: 124) suggest that by 2003, the fusion of death, inheritance, and estate es was so complete in the public mind that it scarcely mattered which term was used. The term death does appear to increase the fraction who believe that most families must pay the estate, but only by a small amount (9%) and only among the 30% of the sample who believed the Republican Party is better at dealing with ation (Schaffner and Atkinson 2009). 6 The prompt for a best guess mirrors the practice in Prior and Lupia (2008). 7 The 2008 survey also included an open-ended follow-up question: Please tell us more about why you [favor oppose] doing away with the estate. Respondents typed an answer into a text box.

12 11 Knowledge of the Estate Tax and Its Consequences The literature on attitudes toward the estate concurs on this point: many citizens do not know that very few, and only very wealthy, people have to pay the estate. Data from the 2008 CCES buttress this point. Figure 1 presents a histogram of estimates of the percentage of people who must pay the ; each bar captures a five percentage-point interval. 8 [insert Figure 1 about here] While the distribution of responses is skewed toward lower estimates, a minority of respondents provided the correct answer. If we define correct strictly, as 1% or 2%, then about 20% of respondents answered correctly. (Only about 2% said that no one, or 0%, pays the.) If we adopt a more lenient definition of 1-5% which roughly corresponds to the correct answer in Krupnikov et al. ( less than 5% of all Americans ) then 36% answered correctly. This is exactly the same fraction that correctly answered Krupnikov et al. s multiple-choice question. Some other respondents were close to the truth. For example, about 15% of respondents gave an estimate between 6-10%. But many others were not accurate at all: approximately one in five respondents believed that 40% or more of Americans paid the estate. These figures are perhaps not as grim In the 2008 CCES, approximately 2% of respondents (19 of 881) indicated that they opposed estate repeal but then gave an open-ended response indicating their support for estate repeal (perhaps because they believed they were indicating opposition to the estate in the first question). I considered their open-ended response to indicate their true attitude, and recoded their closedended response accordingly. For reasons of space, I do not further analyze these open-ended responses here. 8 Of the 829 respondents interviewed in the post-election wave, 813 (98%) provided an estimate.

13 12 as those reported in Slemrod (2006) in particular, the fact that 49% believed that most families have to pay the and 20% did not know. Nevertheless, they confirm that accurate knowledge about the estate is the exception rather than the rule. Does knowledge affect opinions? To replicate extant studies, I regressed attitudes toward the estate on knowledge, controlling for personal income and two items that measure underlying predispositions: party identification and self-reported ideology on the liberal-conservative scale (see Table A-1 of the appendix for the results). 9 Other things equal, a respondent who believes that 40% of Americans pay the estate is 16 percentage points more likely to oppose the estate than a respondent who believes (correctly) that 1% of Americans pay this. A weak Republican respondent is 21 points more likely to oppose the than is a weak Democrat. A weak conservative respondent is 34 points more likely than a weak liberal to oppose it. In this sample, opinion about the estate appears to depend on information but, equally, if not more so, on underlying values and predispositions. By contrast, the effect of income is in the direction opposite to expectations and is not statistically significant. To test for the conditional effects of knowledge, I estimated additional models that interact the knowledge item with party identification or ideology (see Figure A-1 in the appendix). The marginal effect of knowledge is larger and statistically significant only among liberal and moderate respondents and only among respondents who are Democrats, independents, or independents who lean Republican. For example, its effect among strong Democrats (0.24) is much larger than its 9 Party identification is measured with the traditional seven-point scale, and ideology with a fivepoint scale. Although older respondents are in theory closer to a potential estate obligation, age has no statistically significant effect when included in this model. The full results are available on request.

14 13 effect among strong Republicans (0.05). These results confirm Krupnikov et al. (2006): knowledge of who pays the estate is associated with support for the estate among these respondents, but this relationship is largely confined to the left and center of the political spectrum. Among conservative and Republican respondents, the effects of information are weaker. An Experimental Test These observational data suggest that knowledge about who pays the estate matters, at least for certain groups within the population. But questions about causality lurk. Beliefs about the estate could be endogenous to preferences: those who favor the may then decide that it affects few people, and those who oppose the may decide that it affects many people. Perceptions of facts are then rationalizations, rather than sources, of preferences. A better way to measure the causal effect of information is via experimental manipulation that randomizes whether respondents receive correct information about who pays the as well as whether they receive other values-based arguments for and against the. Table 1 summarizes the experimental designs in the 2007 and 2008 CCES. [insert Table 1 about here] In the 2007 CCES, respondents were randomized into 5 different groups. The first a control was simply asked whether they supported or opposed repealing the estate. The second group was first given correct information about how many and who pay the, based on analysis by the Center on Budget and Policy Priorities: This is a that affects those who inherit estates of more than $2 million. The vast majority of Americans, about 99%, do not have an estate large enough to be ed. 10 I refer to this experimental condition with the shorthand Who pays. The 10 See

15 14 difference between these two groups in their support for the estate constitutes a better estimate of how much information affects attitudes. To capture the effects of this information when combined with rhetorical arguments, the third group ( Who pays + Aristocracy ) was provided the same factual information about who pays the alongside an argument for the estate : Supporters of the say that without it, we will have an aristocracy of wealth, where financial resources depend on heredity rather than merit. This paraphrases a statement by Warren Buffet (quoted in Johnston 2001), one of a number of wealthy Americans who publicly opposed estate repeal (see Graetz and Shapiro 2005, ch. 16). Graetz and Shapiro argue that this argument was ineffectual when it was wielded by Buffet and other corporate titans. One poll found that citizens also tended to consider it unconvincing. 11 The fourth group ( Who pays + Punishes success ) was provided the information alongside an argument against the estate : Opponents of the say that it punishes financial success and discourages people from saving and investing. This is a frequent assertion of estate opponents (see Janjigian 2008), although Birney et al. (2006: 440) conclude that this argument had limited appeal: Not all arguments supplied by the pro-repeal side proved to be effective Arguments that tried to trade on public support for what benefits the economy are one example [thus] proponents of the repeal de-emphasized arguments about economic effects The fifth group was provided the correct information and both the pro and con arguments. 11 Respondents were told that, America is founded on the notion of equal opportunity for all. Eliminating the estate creates a two-tired society where some individuals do better than others based on inherited wealth rather than hard work. Almost two-third of respondents (65%) rated this at 5 or below on the 0-10 scale, where 10 indicated extremely convincing (see Greenberg, Quinlan, and Rosner 2002).

16 15 The experiment in the 2008 CCES elaborates on the 2007 design in two important ways. First, the design is a 2 4 factorial. The 8 experimental conditions randomize the provision of correct information about who pays the, as well as the provision of a pro- argument, an anti argument, neither argument, or both arguments. Unlike in the 2007 experiment, there are experimental conditions for those who did not receive the correct information but received only a pro- argument, an anti- argument, or both arguments. Second, the 2008 experiment included a different anti- argument: Opponents of the say that it infringes on the right of families to pass along inheritance to their children. This argument ( Family rights ) portrays the estate not as penalizing prosperity but as violating a fairness norm i.e., that it is only fair for parents to provide for their children via inherited wealth. This norm is invoked by some ordinary Americans who oppose the estate, as demonstrated in Hochschild (1980), appears to increase opposition to the when referenced in a survey question (Birney, Graetz, and Shapiro 2006), and has been central to the repeal movement s strategy (see Graetz and Shapiro 2005). 12 Otherwise, the 2008 experiment mimics the 2007 experiment in the inclusion and wording of the Who pays and Aristocracy arguments. Both experiments examine the effects of information alone and of information combined with pro- and/or anti- arguments. Both experiments include a control 12 For example, in Hochschild s interviews, one person described the estate as awful because a family has a perfect right to hand it down to their children if they want to (see others quoted in Bartels 2006: 412). Similarly, one estate opponent, Rep. Sanford Bishop (D-GA) said on the House floor: The death represents all that is unfair and unjust about the structure in America because it undermines the life work and life savings of Americans who want only to pass along the fruits of their labors and the realization of the American dream (quoted in Graetz and Shapiro 2005: 69).

17 16 group received neither the correct information nor any of the arguments. These experiments have several advantages. First and most important, respondents are randomly assigned to receive correct information about who pays the estate : I can thus evaluate its effects without concerns about endogeneity. 13 Of course, experimental designs have their wellknown liabilities (see Gaines, Kuklinski, and Quirk 2007). For one, I do not measure the durability of any changes in attitudes. There is also the potential gap between the effects of experimental treatments and the (often weaker) effects of their real-world analogues (Barabas and Jerit 2010). But the advantages of experimentation in this case arguably outweigh the costs, especially because prior literature on public opinion about the estate has been observational. A second advantage is that I can compare the effects of factual information to those of rhetorical arguments commonly used by estate proponents and opponents to determine whether factual information matters less than appeals based on values. These rhetoric arguments provide the kind of moral interpretation that Bartels (2004) considers crucial. Third, by including conditions that combine both factual information and these arguments, I am better approximating the broader discourse about the estate, which includes factual claims as well as other rhetoric. Naturally, the 13 While no extant study provides an experimental test of the effects of correct information, two findings are worthy of note. First, Schaffner and Atkinson (2009) find that some estate opponents change their position when they learn that (as of 2003) only estates worth at least $1 million were subject to the. Second, Graetz and Shapiro (2005: 124), citing Greenberg, Quinlan, and Rosner (2002), state that support for repeal drops after respondents were exposed to information about who actually pays the. However, this study actually exposed respondents to a variety of arguments against the estate before querying them about their attitudes toward the. Thus, it is not a good test of information effects.

18 17 estate s proponents and opponents make multiple arguments, not merely one or two. But even if including two such arguments is a weak approximation of, say, congressional debate (see Mucciaroni and Quirk 2006), it is a better than ignoring these arguments and assuming that factual information is provided (or not) in a political vacuum. The Effects of Stories and Science To show the effects of these experimental treatments, I first present the distribution of attitudes toward the estate across the conditions within each experiment. Figure 2a presents the data from the 2007 CCES. In the control condition, 60% of respondents said that they opposed the and 40% supported it. However, respondents given correct information about who pays the estate were virtually evenly divided (49-51%). This information increased support for the by 11 points. Pairing this information with the pro- argument about a potential aristocracy of wealth increased support further, to 56%. Interestingly, the anti- argument how the punishes financial success had little impact when paired with the correct information. Opinion in this condition was virtually the same as when the correct information was presented alone. Finally, the combination of the correct information with both arguments produced the most lopsided distribution of opinion, with 62% favoring the. The anti- argument actually appeared to backfire (see Chong and Druckman 2007), leading respondents to be more supportive of the than even those exposed only to the correct information and the pro- argument. [insert Figure 2a about here] Figure 2b presents the data from the 2008 CCES. In this sample, few respondents in the control group (28%) supported the estate. The correct information once again increased support for the (to 40%) and by about the same amount as in the 2007 experiment. The pro- argument had a similar effect, increasing support to 39%. By contrast, the anti- argument about

19 18 family rights did not affect the distribution of opinion, relative to the control group. Neither did it neutralize the pro- argument: combining them produced virtually the same distribution of opinion (62-28) as did the pro- argument by itself (61-39). Other combinations also suggest the power of correct information and the aristocracy argument, relative to the family rights argument. As in 2007, combining the correct information and the aristocracy argument increased support for the more than did the correct information alone (to 47%). Combining the family rights argument with the correct information had only a small effect on the distribution of opinion, relative to the group receiving only the correct information (63-37 vs ). Combining the anti- argument with both the correct information and the aristocracy argument produced a distribution of opinion (62-38) that was less favorable to the estate than that in the condition including only the information and aristocracy argument (53-47) but still more favorable than that in the control group (72-28). [insert Figure 2b about here] These descriptive results provide initial evidence that both correct information and rhetorical arguments affected opinion toward the estate. A multivariate model of attitudes toward the estate allows for formal tests of the effects of the different treatments as well as tests comparing the magnitude of the effects of different treatments. Here, I take advantage of the commonalities across the 2007 and 2008 CCES and pool the two experiments to maximize sample size. The dependent variable is coded 1 for supporting the and 0 for opposing it. The model is estimated via logit and includes a set of dummy variables for the experimental conditions (with the control group the excluded category), the same set of covariates (party identification, ideology, and income), and a dummy variable for the year of the survey. The coefficients are presented in the first column of Table 2. To convey their substantive meaning, I present marginal effects and confidence intervals in Figure 3.

20 19 [insert Table 2 and Figure 3 about here] Political predispositions are powerful predictors, as expected: support for the estate declines as party identification becomes more strongly Republican and ideology becomes more strongly conservative. Income, however, has no statistically significant relationship to attitudes. Many experimental treatments produce a statistically and substantively significant increase in respondents support for the estate. For example, the marginal effect of the information about who pays the was 0.14 (s.e.=0.05). The effect of the combination of this information and the pro- aristocracy argument is 0.19 (s.e.=0.05). The effect of the combination of information, the pro- argument, and the anti- punishes success argument is 0.24 (s.e.=0.06). Experimental conditions involving the anti- family rights argument tend to have weaker effects. This anti- argument presented by itself has a negative effect and the combination of this argument and the correct information has a positive effect but neither effect is statistically significant. However, the treatment featuring this anti- argument, the pro- argument, and the correct information is associated with increased support for the (0.20; s.e.=0.08). Almost every treatment whether it involved information, the pro- argument, or combinations of these with each other or with the anti- arguments increased support for the estate. Opinion about the estate appears malleable, at least among these respondents. In terms of what affects opinion in these samples, stories do not clearly trump science. Opinion does respond to correct information about who pays the, which increased support for the by a similar amount in both the 2007 and 2008 experiments. 14 Moreover, this correct information is not 14 An interaction between the effect of the who pays treatment and the year of the survey produced a statistically insignificant coefficient in a multivariate model otherwise identical to Model 1 of Table 2.

21 20 necessarily less effective than values-based arguments against the estate. The family rights argument which was identified by Graetz and Shapiro as the most potent weapon employed by the anti- movement did not have a statistically significant effect on opinion. Neither did this argument neutralize the effect of the correct information. The coefficient for the who pays treatment is not statistically different than the coefficient for the treatment combining the correct information and the family rights argument (χ 2 =0.65; p=0.42). Indeed, there are no statistically significant differences among those treatment effects that were themselves positive in sign and statistically significant. Although hostility to the estate appears longstanding and linked strongly to political predispositions, this sentiment may be far from entrenched. Public opinion about the estate takes the shape of public opinion about many issues, with diehards on both sides and a mushier middle in between. Although none of these experimental treatments produce overwhelming majorities in favor of an estate, they produced shifts in the opinions of respondents that would arguably create challenges for the movement to repeal the, were those shifts realized in public opinion write large. The Conditional Effects of Factual Information If respondents opinions about the estate are affected by factual information and rhetorical arguments, the next question is: who is affected? Based on previous literature, I hypothesized that factual information about who pays the would have larger effects on poorer respondents, liberals, and Democrats. To test these hypotheses, I estimated models of estate preferences, identical in specification to the first model in Table 2, but including an interaction term between the information treatment and each of party identification, ideology, and income. The results of these models are also presented in Table 2. Figure 4 depicts the interactions between the

22 21 information treatment and these variables. 15 The graphs plot each variable s effect on the probability of supporting the versus opposing it; there are separate lines and confidence intervals for respondents who received the correct information (the solid line) and for respondents in the control group (the dashed line). [insert Figure 4 about here] Factual information does exacerbate the effect of income. In the control group, the poorest respondents are only 1 point more likely than the richest respondents to support the estate. But once told who pays this, this difference increases to 28 points. Among those making less than $10,000 annually, the probability of supporting the jumps from 0.34 to 0.63, but it remains essentially constant among the richest group, those making $150,000 or more (0.33 vs. 0.36). Information about who pays the estate pushes lower-income respondents toward an opinion arguably more in line with their economic self-interest. By contrast, the conditioning effects of party identification and ideology are precisely the opposite of those found in previous literature. Providing this factual information increases support for the estate among Republican and conservative respondents, not Democratic and liberal respondents. It thereby weakens the relationship between political predispositions and attitudes toward the estate. In the control group, strong Republicans have a 0.17 probability of supporting the, on average. Among strong Republicans told who pays the estate, that probability increases to Among strong conservatives, the results are nearly identical: their probability of supporting the is 0.15 in the control group, but 0.37 in the information group. This finding contradicts those based on observational measures of information about the estate 15 These graphs are based on the code developed by Brambor, Clark, and Golder (2006), which is available at:

23 22 (Krupnikov et al. 2006). Correct information tends to render opposing partisans in these samples more similar, not more polarized. Why are conservative and Republican respondents more affected by correct information about who pays the estate? One possibility is that the Republicans are less informed to begin with, and thus the correct information simply has more to correct in these groups. As it turns out, Republican and conservative respondents are less knowledgeable about the estate. Among Republicans interviewed in the 2008 CCES, the median estimate of the percentage that must pay the estate is 15%, compared to 10% among Democrats. Among conservatives, the median is 20%, compared to 5% among liberals. 16 Perhaps, then, the interactions between the information treatment and party identification and ideology are spurious. The truly important interaction could involve the information treatment and knowledge about the estate. Drawing on the 2008 CCES, which included the knowledge measure, I estimated models identical to Models 2 and 3 in Table 2, and then those same models including knowledge about the estate as well as an interaction between knowledge and the information treatment (see Table A-3 in the appendix). The results are unequivocal: including an interaction between the information treatment and knowledge barely affects the interactions between the information treatment and either measure of political predispositions, which remain substantively and statistically significant. Moreover, the interaction between the information treatment and knowledge, albeit in the 16 These descriptive findings are confirmed by multivariate models displayed in Table A-2 of the appendix. Controlling for income and education, a strong Republican is expected to have an estimate 7 points higher than that of a strong Democrat (s.e.=2.3). In a similar model including ideology instead of party identification, a strong conservative is expected to have an estimate 12 points higher than that of a strong liberal (s.e.=3.6).

24 23 expected direction, is not statistically significant. This seems counterintuitive on its face: why would information about how few pay the estate not matter more for those who believe that a large fraction must pay this? One answer derives from the information treatment s mention of not only how many pay the estate (about 1%) but who pays the ( those who inherit estates of more than $2 million ). If the who is more consequential to respondents than the how many, then the information treatment s effect should not depend on respondents estimates of how many. This also has an important implication for estate politics: supporters of the could actually rely on science to persuade voters, provided that their message emphasizes facts about the wealth of estate payers, not simply their small numbers. A second possibility is that correct information makes salient the cross-pressures that some Republicans and conservatives feel, while having more muted effects on cross-pressured Democrats and liberals. Cross-pressures arise when an individual s political predispositions and identities conflict with each other and make this individual more susceptible to persuasive information. For example, partisans who are out-of-step with their party on key issues are more likely to vote for the opposite party in presidential elections, especially when they live in battleground states and have experienced more campaign activity (see Hillygus and Shields 2007). In the case of the estate, cross-pressures could derive from conflicts between economic interests and political values. The conflict is likely to be particularly acute for lower-income Republicans and conservatives, once informed that only a few very wealthy people pay the estate. They have an ostensible economic interest in keeping the estate or at least no economic interest in opposing it but their political values push them toward opposing it. Once informed that the is paid by a few wealthy people, interests may take precedence over values, leading poorer Republicans and conservatives to support the. By contrast, wealthy Democrats who would appear similarly cross-pressured will not have the same reaction to this correct information. Because the treatment

25 24 states that 99% of Americans do not have to pay this, most wealthy Democrats will not be affected. Thus, the information treatment should strengthen the relationship between income and attitudes among Republicans and conservatives mainly by increasing support for the among poorer members of these groups but should not much affect the relationship between income and attitudes among Democrats and liberals. A test of this hypothesis elaborates the models presented in Table 2 by including three-way interactions between the factual information treatment, income, and party identification or ideology, as well as the constituent variables and two-way interactions. The results are presented in Table A-4. I focus on a graphical illustration of the findings in Figure The lefthand panel shows that, among Democratic respondents, income has almost no effect on attitudes in either the control group or among those provided the correct information. Among Republican respondents, however, the story is much different. The righthand panel displays results for a weak Republican respondent who identifies as somewhat conservative. In the control group, the predicted probability of supporting the estate declines from 0.38 to 0.23 as income increases from its lower to highest values. In the who pays group, the relationship between income and attitudes is much stronger: the comparable probabilities are 0.67 and Other simulations from the model generate similar conclusions. For example, among leaning Republicans who identify as moderate, the comparable probabilities among the lowest income category are 0.38 in the control group and 0.54 in the information group. Providing factual information appears to encourage some lower-income Republican and conservative respondents to abandon their opposition to the. To be sure, the evidence here is only tentative: the multiple interactions mean that the estimated coefficients have 17 The predicted probabilities in Figure 5 are calculated using the Clarify program (King, Tomz, and Wittenberg 2000).

26 25 large standard errors, and thus the differences across experimental conditions are estimated imprecisely. 18 Nevertheless, these results suggest that correct information intensifies cross-pressures among Republicans whose economic interests and political values may be at odds. Conclusion The animating puzzle is why some people oppose a that they are unlikely to pay. Clearly part of the reason is that they simply believe the estate is wrong. Attitudes toward the are grounded in bedrock values and identities, much as previous research has argued. But information about who pays the is a crucial part of the story. Simply telling respondents that only a few very large estates are subject to this reduced opposition to the and increased opposition. Science mattered. The question is how much. This fact alone did not persuade a vast majority of these respondents to support the. Those who are skeptical of the role of factual information in this policy domain including Bartels, Krupnikov et al., and Graetz and Shapiro are correct: it did not induce a sea change of opinion. However, its effect could still prove politically meaningful, that informed opinion opposes the less strongly could provide ammunition to pro- forces. As important as whether information matters is who information affects. Extant research suggests that those on the left of the ideological spectrum are more susceptible (Krupnikov et al. 2006). My replication using an observational measure of information confirmed this finding, but the experimental manipulation demonstrated the opposite: Republican and conservative respondents, and in particular those who have lower incomes, were most affected. This finding implies the potential for a pro- campaign to attract bipartisan support much as the anti- campaign has 18 For example, at the lowest level of income, the estimated difference between Republicans in the control and factual information groups is 0.29 (s.e.=0.20).