Stealth Democracy: Authoritarianism and democratic deliberation Stealth Democracy: Authoritarianism and democratic deliberation

In Stealth Democracy, Hibbing and Theiss-Morse seek to show that much of the American public desires "stealth democracy"--a democracy run like a business with little deliberation or public input. The authors maintain that stealth democracy beliefs are largely reasonable preferences, and the public does not want and would react negatively to a more deliberative democracy. This paper introduces an opposing "authoritarian stealth democrats thesis" that suggests that stealth democracy beliefs may be driven by authoritarianism and a variety of related orientations including poor political perspective taking and low cognitive engagement. These orientations may be ameliorated through democratic deliberation. Hypotheses are tested with survey and experimental data from deliberations with a RDD sample of 568 Pittsburgh residents and of 99 Canadian young adults. Using confirmatory factor analysis and OLS regression with cluster-robust standard errors, the paper finds that authoritarianism and related orientations strongly explain stealth democracy beliefs among deliberation participants and that deliberation significantly reduces stealth democracy beliefs and factors behind these beliefs.

In their widely-read book, Stealth Democracy, John Hibbing and Elizabeth Theiss-Morse (2002) critique the view that the public should be more involved in public policy decisionmaking. Reviewing the literature, they point out that much of the American public is unknowledgeable about policies and shifts its policy positions in response to content-free counterarguments. In response to such findings, many good government advocates and social scientists have urged efforts to counteract public apathy and ignorance. Some have recommend efforts to instill a more deliberative democracy (Barber, 1984;Chambers, 1996;Fishkin & Laslett, 2003;Gutmann & Thompson, 1996). In contrast, Hibbing and Theiss-Morse suggest that the foregoing scholars are pursuing a dangerous illusion. They contend the public is not usually interested in policy and has good reasons not to be.
Hibbing and Theiss-Morse suggest the public cares less about particular policies than the impeachability of the governmental processes that put in place those policies. Much of the public, they say, has a preference for a certain type of "stealth democratic" process-one in which non-self-interested, well-intentioned elites implement a presumed broad public consensus on the issues while minimizing unnecessary internal debate and conflicting input from self-interested "publics." The authors find that 93.5% of respondents in a representative sample of Americans agree with one or more of three statements describing what they call "stealth democracy" beliefs. Seventy-six percent agreed with two or more of these statements. These are statements that express a desire to end debate and compromise and that favor government by elites. Hibbing and Theiss-Morse present evidence that these stealth democracy beliefs matter-they affect confidence in government, desire to reform the government, voting for nontraditional candidates, and compliance with the law. In short, they matter for the political system's stability, which is the authors' primary concern.
In addition, Hibbing and Theiss-Morse shape their various findings into a booklength argument against prescriptions to engage the public more deeply in politics, particularly deliberative engagement. Their "stealth democracy" thesis holds that the public prefers stealth democracy because it has good reason to find politics uninteresting, dislikes conflict, and believes there is wide consensus on political goals. Because of its belief in a consensus, the public interprets disagreement and conflict in politics as evidence that special interests have overtaken the public interest. In addition, much of the public has doubts about the ability of ordinary people to handle the details of governance and therefore prefer to delegate decisions. The authors maintain that efforts to politically engage such a public is a prescription for frustration, distrust, and, therefore, delegitimization of the political system. They do, however, call for educating the public against the one feature of stealth democracy beliefs they find concerning-the public's false belief in a political consensus. The authors fear false consensus beliefs create unrealistic and destabilizing expectations.
The stealth democracy thesis has been well received. The book received a favorable review by political science luminary Robert Shapiro (2003). And it became widely used in college courses in political science. The concept has also come into use outside academia, with 8,810 Google references to "stealth democracy" in a recent search.
If Hibbing and Theiss-Morse have discovered the basic, unalterable preferences of the public (the kind of preferences in neo-classical economic theory), then elites have no choice but to respect these preferences. However, much depends on whether the identified preferences for stealth processes are basic and unalterable or instead derivative and changeable, perhaps by better informing people and seeking to make them more reasonable, coherent, and competent as citizens, as may be essential for good government. In the former scenario, political disengagement must simply be accepted, while in the second, seeking to engage the public may be a necessity. This paper theorizes that stealth beliefs may in substantial part be a manifestation of authoritarianism. Authoritarians are characterized by a constellation of views and tendencies that include low cognitive engagement, inconsistent views of political conflict, and inability or unwillingness to take alternative political perspectives. Contrary to the Hibbing and Theiss-Morse thesis, the authoritarian stealth democrats thesis suggests deliberation could be beneficial-ameliorating stealth democracy beliefs by calling on people to refine their political thinking. This paper will examine whether stealth democracy beliefs are grounded in authoritarian and related views and dispositions and whether political deliberation helps to ameliorate stealth democracy beliefs and some of the sources of these beliefs. This paper examines these hypotheses with data from studies of democratic deliberations involving 568 Pittsburgh residents and 99 Canadian young adults selected by random digit dialing. The findings are consistent with the authoritarian stealth democrats thesis, and suggest that efforts to encourage citizen participation are beneficial, not harmful.

An Alternative Theory: Authoritarian Stealth Democrats
Hibbing and Theiss-Morse impose an interpretation of reasonableness on stealth democrats without rigorous evidence.
The authors' focus group based interpretations do not involve a rigorous content analysis. Nothing in the authors' national survey questions insures that stealth democrats wish to retain oversight of leaders or delegate complicated and uninteresting policy choices. Nor do their analyses of their survey data consider alternative explanations. This paper proposes and tests an alternative interpretation: that stealth democracy beliefs may be driven by authoritarianism. This hypothesis echoes a long tradition of research that raises concerns about the democratic commitment of the public (McClosky & Brill, 1983;Prothro & Grigg, 1960;Sullivan & Transue, 1999).

Understanding Authoritarianism.
Before presenting a theory of authoritarian stealth democrats, an analysis of what constitutes authoritarianism is needed. Decades of research have provided many and many-layered definitions of authoritarianism, but the core of the concept is a problematic belief in the need for deference to authority. Of course, people may defer to authority for good reasons-namely, they may have reasons to trust the authority such as perceived competency and past behavior, and they may find themselves needing to delegate decisions. Authoritarians, however, defer for less direct and less compelling or at least less liberal democratic reasons.
In childhood, survival and flourishing typically depend on obeying authority, often without questioning. The non-authoritarian view that grants of authority should be specific and contingent on reasons and rules represents a sophisticated additional layer of beliefs about authority that need not accompany child-rearing. People who emerge from childhood with a belief in the need for unquestioning obedience to authority are authoritarian. Such people may develop various reasons for the need for obedience, but the reasons may be less direct and compelling than the narrow grounds for justified authority described above.
A prominent justification for authoritarianism is the view that unquestioning obedience to authority is somehow crucial to maintaining the social order and preventing chaos and great harm. Authoritarianism in North America is perhaps most directly captured by the items of the right-wing authoritarianism (RWA) scale (Altemeyer, 1981). This scale measures such beliefs as the crucial importance of instilling obedience and respect for traditional authorities in children and a desire to punish people who seem deviant. Both of these follow from a view of authorities as crucial to maintaining the social order and the good that comes from it. Deviants must be harshly punished because they bring chaos and harm to all.
Authoritarianism may take forms other than RWA. One possible form is social dominance orientation or SDO (Sidanius & Pratto, 1999). SDO is the belief that society is better off when one or a few groups dominate other groups. Because society is better off with one or a few groups in dominance, persons high in SDO should also demand that dominated groups acquiesce and obey dominating groups -domination implies obedience. Obedience extends to the person high in SDO beliefs when that person faces a group higher on the hierarchy. SDO justifies a belief in the need for obedience to authority likely through a belief in a vicious, zero-sum competition between social groups and therefore the need for domination and obedience to avoid chaos.
A third possible form of authoritarian impulses is vertical collectivism or VC (Triandis, 1996). VC involves a willingness to sacrifice self-interest in favor of family or social group, in part because of a belief in a duty to do so. Family or other social groups command obedience. Here, obedience is apparently justified through a conceptualization of personal identity that sees personal desires as selfish and unimportant while elevating powerholders in social groups into authorities that define what is important and meaningful. This is likely supported by beliefs in the inefficacy of the self and the naturalness of conflict between selfinterest and society.
While academic acceptance of the authoritarianism concept has fluctuated over past decades, more recent research is strongly supportive. Altemeyer's (1981) research on RWA responds nicely to criticisms of psychometric shortcomings and political bias in Adorno et al.'s (1950) original research on authoritarianism. Altemeyer and Jost and his colleagues (Jost, 2006;Jost, Glaser, Kruglanski, & Sulloway, 2003) reinvigorated authoritarianism scholarship.

Authoritarianism and Stealth Democrats.
Authoritarians should be attracted to stealth democracy beliefs. These beliefs include the view that unelected experts and business people should run the government and that government would be better without debate or compromise. Experts and business people are social authorities typically acceptable to rightwing authoritarians, those high in SDO, and likely those high in VC (deference to one's social group implies deference to those groups recognized as authorities by that group). Note that the desire for government by unelected experts raises questions about whether stealth democrats are democratic. Finally, authoritarians want an authoritative will guiding action in any given domain, not squabbling elites who must debate and compromise.
Squabbling undermines the authoritativeness of the social order. Also, debate and compromise imply that authority should be circumscribed by reasons rather than unquestioningly obeyed. The authoritativeness of the social order and unquestioning obedience are special concerns for those who subscribe to RWA and SDO, but are also consistent with VC.
Stealth democracy beliefs and related attitudes may in fact constitute an authoritarian theory of democracy. Authoritarianism is fairly prevalent in the U.S. (see below), which raises the question of how authoritarians reconcile their beliefs about authority with living in a democracy. The desire for a single authoritative voice could be met by a dictator, but might also be satisfied by belief in a false consensus in the public-a key component of stealth democracy according to Hibbing and Theiss-Morse. The belief that the public constitutes one common will is plainly false, even by casual observations of an election. Nevertheless, authoritarians, who already feel quite negatively about those who are different, could redefine the "real" public to include only "true," like-minded Americans. Such a mythic public consensus likely needs to have an interpreter and agent in the form of a strong leader facing few checks and balances and consequently little debate or compromise. Stenner's (2005) findings support a relationship between authoritarianism and desire for a monolithic public will. She presents a number of experimental results showing that authoritarians are driven by, "…a fundamental and overwhelming desire to establish and defend some collective order of oneness and sameness." (p. 277) Stenner implies at one point that stealth democrats may be authoritarians.
One concern is whether there are sufficiently many authoritarians to account for the many stealth democrats Hibbing and Theiss-Morse find. Stenner's (2005) findings suggest there are. She shows that authoritarianism powerfully explains intolerance in representative U.S. surveys-accounting for 45% of the variance in general intolerance (racial, religious, and political) in General Social Survey data from 1990-2000.
The authoritarian stealth democrats interpretation yields importantly different empirical predictions than the Hibbing and Theiss-Morse delegating stealth democrats interpretation. Authoritarian stealth democrats realize that the consensus they attribute to the public depends on who they define as a member of the "real" public. Thus, while they may attribute high levels of consensus to "the public" as an abstraction, they should be concerned about conflict among random samples of the actual public brought together to discuss a political issue. For authoritarian stealth democrats, both belief in a false abstract consensus and fear of conflict among actual citizens should help predict stealth democracy beliefs. This combination would be contradictory under the delegating stealth democrats interpretation. In addition, for reasons elaborated below, authoritarian stealth democrats may give up some of their stealth democratic and authoritarian orientations in the course of citizen deliberations, contrary to Hibbing and Theiss-Morses's warnings.

Authoritarianism and Its Effects
Research on RWA shows that RWA is related to a number of problematic tendencies and orientations. This research finds that authoritarians prefer low cognitive effort in a wide range of contexts. A substantial literature, reviewed by Jost et al. (2003), connects RWA to low attributional complexity, high need for structure, and low openness to experience (Butler, 2000;Duriez & Soenens, 2006). Low openness to experience involves limited intellectual curiosity and introspectiveness. High RWAs hold beliefs about social and political facts that were less accurate than those of low RWAs in 13 of 16 facts in one study (Mirels & Dean, 2006). RWA interacts with SDO in predicting a false belief in a social consensus favoring racial stereotypes (Strube & Rahimi, 2006). This relates RWA and SDO to poor socio-political perspective taking and false consensus beliefs. Authoritarians reminded of their own mortality, which arouses a sense of threat, are more likely to selectively expose themselves to pro-attitudinal information on a policy issue and show inhibited attitude change (Lavine, Lodge, & Freitas, 2005). In a study that hints at a relation to stealth beliefs, Peterson et al. (2002) find that authoritarians are less knowledgeable about and interested in politics. Furthermore, the current paper presents evidence that RWA, SDO, and VC are associated with poor socio-political perspective taking and aversion to conflict. Recent research, however, raises questions about ideology and cognitive biases. Kahan (2013) finds that both liberals and conservatives are equally subject to motivated reasoning in certain experimental contexts. This research, however, does not examine authoritarianism, which is not equivalent to political ideology.
Consistent authoritarians should have low cognitive engagement and poor sociopolitical perspective taking. People thinking hard about issues potentially threaten the social order by replacing the proper authorities with their own reasoning. Taking the perspective of less mainstream members of society undermines the notion of a monolithic, correct social order and potentially takes the side of 'deviants' or undeserving social groups.
Because of such connections, authoritarianism may consist in a constellation of related tendencies.

Stealth Democracy and Deliberation
Public deliberation should mitigate authoritarianism-inspired stealth democracy and related beliefs. These effects stem from the implications of the deliberation context. A typical public deliberation, like the ones examined here, brings together a random selection of the public to discuss issues of serious concern in a setting that invokes the public, such as a university or the nation's capital. Participants are given detailed policy information and are informed that their choices could have real consequences-community leaders and newspapers are watching. These circumstances encourage a context with a number of features: publicness, salience, rationality, and equality.
Discussing an important issue in a highly public forum with potentially real policy consequences makes participants accountable. Participants will likely feel that they must give publicly acceptable reasons that stand up to broad scrutiny. Accountability-awareness that others will review reasons given-to unknown others increases cognitive effort and reduces certain biases (Lerner & Tetlock, 1999;Tetlock, 1992).
Importantly, engaging in such cognitive effort in a context that encourages independent decision making should create cognitive dissonance for authoritarians, because it places them in the role of an autonomous decision maker. The public deliberation context encourages independent choices in several ways. Accountability plays a role. Also, social influence should play less of a role in public deliberation because final choices are confidential and people do not know each other. In drawing their own conclusions, participants will be acting as authorities-contrary to their belief in relying on the judgments of authorities. Deliberation participants typically feel they have made good decisions. The resulting cognitive dissonance may cause authoritarians to question whether the judgment of authorities is necessarily better.
Public deliberation could also help clarify for participants that reasonable people hold a diversity of views and that, despite this, discussion can be productive. With such deliberative methods as the National Issues Forums and Deliberative Polls, it is commonplace for participants to engage in respectful and thoughtful discussions of the issues as well as their differences (Fishkin & Laslett, 2003;Price & Cappella, 2002). Participants reassured that deliberation can be productive should be less apt to fear conflict. Also, if diverse people can come together and productively make a decision, then perhaps rigid obedience to authority is not the only basis for social or personal order.

Hypotheses
The above theoretical speculations suggest that authoritarian tendencies should help explain stealth democracy beliefs. Hypothesis Set 1 is: stealth democracy beliefs should be appreciably explained by RWA, VC, and SDO. Also, as explained above, Hypothesis Set 2: both belief in a false consensus and fear of concrete conflict will help explain stealth democracy beliefs. Fear of conflict will be captured with the variable Expect Conflict.
Authoritarianism comes in a constellation of related attitudes, particularly low cognitive engagement and poor socio-political perspective taking. These variables may have direct effects on stealth democracy beliefs, even with authoritarianism controlled. For example, wanting experts to decide policy issues may be an effort to avoid personal cognitive engagement. Also, thinking decisions can be made without debate or compromise might result from a failure to take the perspective of diverse social groups.
Cognitive engagement will be measured here with the need for cognition scale or NFC (Cacioppo, Petty, Feinstein, & Jarvis, 1996) and the need for structure-order scale or NFS (Neuberg & Newson, 1993). Hypothesis Set 3: Stealth democracy beliefs will be negatively related to NFC and positively related to NFS. Sociopolitical perspective taking will be measured here in two ways: a scale of political empathy (M. H. Davis, 1980) and naive realism (Ross & Ward, 1996). Hypothesis Set 4: Both low political empathy and high naive realism will strengthen stealth democracy beliefs.
Given the relative strengths of the data presented here, important hypotheses in this paper are Hypothesis Set 5: Formal public deliberations, for reasons given above, will reduce stealth democracy beliefs and some of the inputs into stealth democracy beliefs such as authoritarian attitudes, expectations of conflict during deliberation, and low cognitive engagement.

Participants
Knowledge Networks (KN) conducted the recruitment for this study, named the Virtual Agora Project or "VAProject." Of a sample of 6,935 Pittsburgh city residents (defined by zip code area) who could be reached via random digit dialing (RDD), 22% agreed to participate in this research and took a phone survey. Sampling differed from KN's typical methodology on other deliberation projects in that it did not utilize quota sampling to make demographic statistics more apparently representative of the population as a whole. Thus, the sample represents who would come to deliberations without demographic oversampling. Because of legal requirements, deliberations could not be based on quota sampling were they more widely used instruments of government. Without quota sampling, there is less need for concern that those oversampled will be atypical for their demographic.
Of recruits who agreed to participate, 37% or 568 people showed for the Phase 1 on-campus deliberation. A modest response rate was expected because recruits were asked to participate in a series of online deliberations that would take most participants eight-months to complete and which they could join only by coming to the initial on-campus, all-day deliberation. The final participation percentages are not, however, incomparable to that of many current nationally representative opinion surveys that depend on survey panels, such as those produced by Pew Research. They are also similar to that of another substantial long-term deliberation study, Vincent Price's Electronic Dialogue Project at the Annenberg School of Communication (Price & Cappella, 2002).
Ultimately, the response rates here are modest. Comfort can be drawn from several considerations: good similarity to population demographics (see below), the fact that the sample represents people who might be expected to participate in public engagement to influence policy, and the objective of this research which is partly experimental and focused on psychological processes that should be universal.
Despite a non-quota based RDD sample and modest response rate, the participants in this project reasonably matched the Pittsburgh city population. The sample was 77% Caucasian and 18% African-American, compared with CPS population benchmarks for the relevant zip codes of 75% and 20%, respectively. Fifty-six percent of the sample was female, compared with 53% for the population. Twenty-five percent of the sample was 18-29 years old, 21% 30-44 years old, 39% 45-59, and 15% 60+. This compares with population values of 26%, 20%, 26%, and 27%. The young and thirty-somethings are accurately represented, the elderly are underrepresented, as might be expected, while mid-life adults are overrepresented. Average age, however, is the same as for the population. The greatest departure from population values is for education, which, unsurprisingly, is greater than for the population. Median education is "Some College" for both the sample and the population.
Lower educational categories are underrepresented, with 10% of the sample having less than a high school education and 14% having just a high school education, compared with 16% and 31% for the population.
Phase 2 of the project, the eight-month at-home online deliberations, was intended to include 410 of the original 568 participants who were selected to receive a computer. Substantial participant drop-off occurred by Phase 2 of the project, with response rates to questionnaires in the early part of Phase 2 dropping to about 230. Drop-out was perhaps driven in part by participant frustration with software and hardware problems. Fortunately, the post-experiment measurement of only one variable examined here occurred in Phase 2.
Pittsburgh is an ethnically and class diverse community with a city population of about 335,000 and over one million in surrounding areas at the time of the study. Neighborhoods range from suburb-like residential areas to areas of urban poverty. People intimately involved with public life in the city believe that the city is average in terms of political involvement and contentious public dialogue. The topic of the research was selected because of its interest to community organizations, who helped provide the information materials for the project.

Materials and Procedures
Knowledge Networks obtained phone numbers for households in the City of Pittsburgh from a RDD sample. Where numbers appeared in a reverse directory, the household was sent an advance letter on Carnegie Mellon University stationery. A Knowledge Networks phone center called households in the RDD sample and requested the household member with the most recent birth date. Both the letter and the call center indicated that in exchange for participation in the study, participants would have a four out of five chance of receiving a Windows computer and eight months of ISP service. The remainder would receive $100. Those who received a computer would be expected to participate in a longer-term online deliberation from home that would require six hours of discussion over eight months. People who agreed to participate were scheduled for a one-day, eight hour on-campus deliberation. Participants were asked to come to a randomly-chosen day from a three week span.
Deliberations were held with up to 60 participants daily. After informed consent and a brief training session, participants took a web-based pre-survey. Next, they were given a 40 minute "library session" to learn more about the four policy topics, a break, 90 minutes for "deliberation" (face-to-face, online, or individual contemplation, depending on condition), and lunch. The library session and so forth were repeated in the afternoon, and this was followed by the second survey. One experimental condition involved either receiving or not receiving reminders of citizenship. In the citizenship condition, participants were reminded to think as citizens, their rooms had an American flag, and they were given name tags with American flags and their names preceded by "Citizen."

Measures
The independent variables were each measured with multiple questions. A question for Conflict Averse (Phase 1 post-deliberation survey), measured on a 7point Likert scale, was: "When people argue about politics, I feel uneasy and uncomfortable." This wording is a minimal rewrite of the Hibbing and Theiss-Morse question so it would fit better into a set of Likert questions. It was joined by a companion reversed question.
The questions for false consensus beliefs occurred in the pre-deliberation questionnaire. False Consensus-"Thinking about the American people, what portion of Americans do you believe think <MostImpProblem> is the single biggest problem facing the country today?" and "What portion of Americans do you believe basically agree with you on what should be done about <MostImpProblem>?". The survey system replaced <MostImpProblem> with the most important problem facing America that the participant had earlier identified.
The zero to 10 response scale had labels: No Americans, Half of All Americans, All Americans. Another pre-deliberation measure of expected unproductive conflict was: Expect Conflict-"Overall, what portion of discussion in your discussion group do you anticipate will involve unproductive conflict?" (11-pt. scale anchors: None of the Discussion / Half of the Discussion / All of the Discussion).
Most authority attitudes and cognitive dispositions were measured using short versions (4-6 items) of scales widely used and accepted by political and personality psychologists. This includes social dominance orientation, SDO (Sidanius & Pratto, 1999); right-wing authoritarianism, RWA (Altemeyer, 1981(Altemeyer, , 1996; vertical collectivism, VC (Triandis, 1996); need for cognition, NFC (Cacioppo et al., 1996); and need for structure-order, NFS (Neuberg & Newson, 1993). One novel measure is naive realism, the idea for which was suggested by Ross and Ward (1996). It involves such questions as: "I can understand why people who disagree with me politically believe what they believe." and "People who disagree with me politically seem to have an agenda." The second novel measure is political empathy. The measure involved rewriting Interpersonal Reactivity Index (IRI) perspective taking questions (M. Davis, 1996) so that they focused on politically-relevant rather than interpersonal perspective taking. These include questions such as: "If I'm sure I'm right about a political issue, I don't waste much time listening to other people's arguments."

Factor Analyses
The surveys conducted for this Virtual Agora Project contain multiple questions for each of several conceptual factors, including the novel scales of political empathy and naive realism. This raises the issue of whether the questions that presumably tap a unique latent factor in fact each tap this factor and this factor alone. Another matter is whether the factors are all different from each other. Confirmatory factor analyses were conducted to address these issues, and provide evidence for the validity of the question scales. More details of these factor analyses, which help verify the scales used, can be found in Appendix A. Interpretation utilized the Goodness of Fit Index (GFI), the Adjusted Goodness of Fit Index (AGFI), the Root Mean Squared Error of the Approximation (RMSEA), the Bayes Information Criteria (BIC), Hoelter's N, and z-scores of loadings. Analyses were conducted in R with the sem package.
A single factor explanation of stealth beliefs does not adequately fit the data (RMSEA=.138 Confidence Interval .09 to .19; BIC=8.14; Hoelter's N=144), as elaborated in Appendix A. Hibbing and Theiss-Morse appear to simply assume the unidimensionality of stealth beliefs. A two-factor model, however, fits the data (RMSEA=.057 CI 0 to .14; BIC=-4.86; Hoelter's=759). One factor captures a dislike of discussion and compromise (henceforth "stealth beliefs #1-no debate"), and the other captures a belief that elites should rule (henceforth "stealth beliefs #2-elite rule"). The correlation of these two factors is .68. Because of the moderately high correlation, the two components of stealth beliefs may be variations of a common second-order factor. The two dimensions can be collapsed where their impact on other variables is not substantially different.
One confirmatory analysis conducted entertained the possibility that a single, underlying second-order factor, the "authoritarian constellation factor," explains all the observed covariances between RWA, VC, SDO, political empathy, NFC, NFS, and false consensus beliefs. This is not meant to suggest that these variables all measure the same factor. Rather, it suggests that there is some single factor, some underlying commonality of authoritarianism, behind the covariations of these factors. Indeed, this model fits the data quite well (RMSEA=.046; BIC= -6327.90; and Hoelter's N=274; GFI and AGFI are .86 and .84, but this is likely due to the large number of variables under consideration). BIC gives the model a strong vote of confidence, indicating that it is far more likely to be the correct model than a model in which there was no second-order factor and instead each factor is allowed to covary directly with every other. These calculations along with all other Bayes factors calculations in this paper were conducted using equation 20 in Raftery (1995).
The second-order factor that explains all these authoritarianism-related factors is appreciably captured by the RWA factor, which has a correlation of .90 with the second-order factor. This indicates that RWA comes close to capturing the unifying factor that explains the covariances between low cognitive engagement, poor political perspective taking, and other measures of authoritarian tendencies.
For subsequent analyses, observed variables were combined into scales to measure the first-order factors identified. Variables were averaged to create the scales, with the averages weighted by estimates of each variable's contribution to each factor, as determined via factor scores regressions.
In summary, confirmatory factor analyses show that the "authoritarian constellation" scales of interest here (RWA, VC, SDO, empathy, NFS, NFS, and false consensus) have variables that load on the factor defined by each scale and only that factor-helping to establish the validity of these scales. Also, a single second-order "authoritarian constellation factor" explains the covariations among all these first order factors, suggesting it may be worthwhile to examine the combination of all the scales. In addition, confirmatory analyses show that stealth democracy beliefs are best explained by two moderately strongly correlated factors. The variables can be collapsed into a single scale where findings do not differ appreciably with a single or two factors.

Convergent and Divergent Validity
Correlations between the scales formed from weighted averages of the variables were examined to determine convergent and divergent validity of the scales (for details, see Appendix A). The "authoritarian constellation" scale variable always has highly significant relationships with all the other scale variables, in the expected direction. Between the remaining first-order variables there are 91 correlations, 80% are significant and in the expected direction. An additional 8% are in the expected direction and show a statistical trend. Of the remaining correlations, only two are significant and in an unexpected direction. These are the correlations between Vertical Collectivism and two subscales of Political Empathy, which are significantly positive. Though not in the expected direction, these relationships make sense-people who wish to fulfill duties to their group (VC) may need to be high in political empathy, at least with respect to their group. The actual cross-group sensitivity of duty-oriented vertical collectivists can be questioned given that they are significantly more likely to believe in a false consensus, have a strong need for structure, and have low need for cognition.
An examination of a large number of correlations among the variable scales here reveals strong evidence for the convergent and divergent validity of the scales. The significance and strength of the preponderance of these correlations and the strength of at least some correlations for each scale helps establish the reliability of the scales. Table 1 shows regressions of stealth democracy on three models. For now, in order to replicate the Hibbing and Theiss-Morse results and simplify findings, stealth democracy will be analyzed as a single combined indicator rather than split into its two somewhat different factors. All analyses are conducted with 'cluster robust' errors that account for possible error covariance among people who deliberate in the same group. The model in the first column after the variable names (henceforth Column 2) seeks to reproduce Hibbing and Theiss-Morse's regression, with one difference. Hibbing and Theiss-Morse create a single indicator called "negative view of disagreement" that averages false perceptions of a public consensus, aversion to conflict, and political interest. Averaging these questions runs contrary to the authors' theoretical discussion and obscures important differences in the effects of the variables. The analysis here separates these variables. Column 2 shows that political (dis)interest plays no significant role in explaining stealth democracy beliefs in the current study, while false  Table 1 is VC with RWA a closesecond. As predicted by the authoritarian stealth democrats thesis, both false beliefs in a public consensus and expectations of unproductive conflict in the actual deliberations contribute to stealth democracy beliefs. This poses a paradox for the stealth democracy thesis. Naïve realism also proves to have a significant direct effect, though political empathy and the cognitive variables do not. Nevertheless, these variables significantly affect stealth democracy in bivariate correlations, which may mean their effects are perhaps mediated by other variables.

Explaining Stealth Democracy
Column 4 of Table 1 tests the possibility that a single composite indicator of the "authoritarian constellation" might do well in explaining stealth democracy beliefs. The composite quite potently explains stealth democracy beliefs, with a coefficient 2.5 times more powerful than any coefficient in Column 3. While the amount of explained variance is lower than for Column 3, this may be due to less overfitting. A Bayes factors analysis indicates that the Column 4 model is about 4,000 times more probable than the full model in Column 3, supporting the view that there is an underlying "authoritarian constellation" that explains stealth democracy beliefs. Bayes factors, however, tends to favor simple models.
The analyses here replicate the Hibbing and Theiss-Morse finding that a single "negative view of disagreement" variable strongly and significantly predicts stealth beliefs-showing that the population under study here is not different in this respect from the population studied by the authors. However, when the three variables constituting negative view of disagreement are regressed separately, an analysis Hibbing and Theiss-Morse do not present, one of the variables (political interest) proves insignificant and another (conflict aversion) of marginal strength -despite the fact that all three are important in the authors' theory. When the authoritarian constellation variables are added to the regression, only false consensus beliefs proves significant. This variable is important in both the stealth democracy and the authoritarian stealth democrats theses. In addition, both expectations of conflict in a real discussion and false belief in a public consensus on policy issues prove significant and positive. This is paradoxical in the stealth democracy hypothesis, but is explained by the authoritarian stealth democrats hypothesis. The authoritarian stealth democracy model proves to appreciably better explain stealth democracy beliefs, with authoritarian attitudes and authoritarian-related orientations playing substantial roles. A model with a single authoritarian constellation scale, combining all the variables in the authoritarian stealth democrats hypothesis, does well.  Table 2 presents results indicating that deliberation helps ameliorate stealth democracy beliefs and some of the variables feeding into these beliefs. Only two of the nine variables underlying the authoritarian stealth democrats thesis were available for consideration with immediately post-deliberation measures. A decision was made to not include comprehensive pre-and post-deliberation measures in Phase 1 of the study, with the expectation they would be included in Phase 2-a series of follow-up surveys. Regrettably, Phase 2 experienced considerable and growing respondent drop-out and began later than desired. The first follow-up survey in Phase 2 fortunately contained a post-measure for stealth democracy beliefs, for which there was no immediate post-discussion measure.

Effects of Deliberation on Stealth Democracy Beliefs and Related Variables
Other relevant questions appeared on later Phase 2 surveys, but with even more respondent attrition and at a greater time remove from the intervention. Therefore, only the crucial stealth democracy beliefs variable between Phases 1 and 2 will be considered. In short, all the variables analyzed in Tables 1 and 2, with one exception, were collected during the one day deliberation. Only the stealth democracy beliefs variable was collected in Phase 2. Table 2 includes controls for discussion media (f2f, online, or no discussion) and citizenship reminders, a 3X2 experiment built into this study, as described in the Materials and Procedures section above. These conditions are not the focus here, but they must be statistically addressed. Note: All F-values<.0001. *** is p<.001; ** is p<.01; * is p<.05; † is p<.10 All pvalues are robust and account for non-independence of errors by discussion group. P-values reported are two-sided.
One of the nine authoritarian stealth democrat related variables was collected only post-discussion in Phase 1: vertical collectivism (VC). Column 2 of Table 2 (the first column of results) shows an ANOVA-equivalent regression with a betweengroups comparison. The constant is the mean for VC in the excluded condition: the control condition with no discussion and no reminders of citizenship. Coefficients for the other conditions indicate deviation from the constant, which captures the control condition. Thus, for example, the mean level for Online is .73-.32 or .41. Column 2 shows that online discussion significantly lowers levels of post-discussion VC relative to the control condition. A post-hoc test reveals that VC is also significantly reduced in the online discussion with citizen reminders condition (p=.009). Thus, online deliberation reduces vertical collectivism, which is a major contributor to stealth democracy beliefs.
Pre-deliberation expectations of unproductive conflict can be compared with a post-deliberation Phase 1 measure of perceptions of actual conflict during discussion, a within-subject comparison. Column 3 of Table 2 shows a regression of the change in perceived conflict (post-deliberation perceived conflict minus pre-deliberation expected conflict) on the experimental conditions. Coefficients of the experimental conditions indicate the amount by which each condition reduces post-deliberation perceived conflict from pre-deliberation expected conflict, relative to the excluded control condition (no deliberation and no citizen reminders). Changes are quite substantial and negative, indicating large declines in perceived conflict, with significant effects for both online and f2f deliberation.
Column 4 of Table 2 shows significant reductions in the No Debate component of stealth democracy beliefs in the online deliberation condition. A post-hoc test also shows a significant effect for the f2f X citizen condition (p=.04, one-sided).
No effects were found for the Elite Rule component. This may be because few people agreed with Elite Rule to begin with-only 27% saw elite rule as desirable.
In short, the data allow analysis of change from pre-to post-deliberation for three of the authoritarian constellation variables. Online deliberation significantly reduces VC. Both online and face-to-face deliberation substantially reduce expectations of conflict. VC and expectations of conflict appreciably explain stealth democracy beliefs, as Table 1 shows. Finally, online deliberation and f2f deliberation with citizenship reminders both significantly reduce the No Debate component of stealth democracy beliefs measured months after the discussion. The Elite Rule component may show no significant effects because of a 'floor' effect-low initial levels of agreement with this component.

Discussion Study One
Leaving most discussion to the conclusion, this post-study discussion will focus on some peculiarities of the current study. Deliberation in the current study affects only the "no debate or compromise" component of stealth beliefs. This is unsurprising given that the value of debate and compromise are directly implicated by the deliberation experience. In contrast, the value of business leaders and experts in government are not. There may also be a 'floor' effect that prevents registering any significant decrease.
An interesting finding with respect to the ameliorative effects of deliberation is that online deliberation reduces all three stealth and stealth-related attitudes tested, while face-to-face deliberation directly affects only perceptions of conflict (though the f2f X citizenship reminders condition significantly reduces stealth beliefs as well). Perhaps, as the social identity and deindividuation model suggests (Postmes, Spears, & Lea, 1998), the absence of social cues in an online setting can give rise to an individualistic orientation. Such an orientation might contribute toward reducing authoritarian reasoning. This would explain why online discussion has a significant effect on vertical collectivism and on stealth beliefs in the VAProject findings. On the other hand, the CPRN findings below show that face-to-face deliberations lasting a few days can also reduce stealth beliefs in the short run, without special reminders of citizenship.

Study Two-Canadian Policy Research Networks (CPRN)
While the Virtual Agora Project data (Study 1) has the virtue of a broad depth of variables, its modest response rate and one-city focus raise the question of whether findings from that study would replicate in a broader population. Study 2, the CPRN Inc. National Dialogue and Summit on Engaging Young Canadians, involves a deliberation drawn from a nationally representative sample of Canada with a good response rate. The study, however, has a more limited scope of available variables.

Participants
EKOS Research Associates, a polling firm, conducted the recruitment for this CPRN study. Two-hundred and fifty-five young Canadian adults were selected using RDD. Some of these 255 did not meet demographic criteria (age, gender, education, ethnicity, and language), while others declined to participate. Onehundred and seventy-eight participants who met the demographic criteria and did not immediately decline to participate where invited to participate in this nationwide deliberation. The number who immediately declined was not recorded. One hundred and forty-four actually came to the deliberation. Of these 144, 99 completed both the pre-and post-surveys of this study, which were piggybacked onto this deliberation. The true response rate, depending on whether the 178 or 255 base figure is used, is between 39% and 56%-an appreciable percentage of the sample by both standard survey and deliberative standards.
The targeted participant population was Canadian young adults, ages 18-25. The sample of 99 participants who took the research surveys differ hardly at all demographically from the 144 participants in the event. Time constraints and mistakes with respondent identifiers account for the lack of full survey completion by 45 participants. The research survey participants' education levels were: 5% some high school, 24% high school graduates, 37% some college or university, 11% community / CEGEP / private college (technical schools), 18% Bachelors, and 5.6% graduate degrees. By comparison, Statistics Canada's Labour Force Survey for the previous year showed educational levels for young adults 20-24 of 12% some high school, 24% high school graduates, 28% some college / university, 24% community / CEGEP / private college, 11% Bachelors, and 1% graduate degrees. The study participants are somewhat better educated than the general public, but do represent all educational levels.
Gender was 54% female, compared with 50% for the 18-25 age group (Statistics Canada, prior census). Ten percent of the sample was visible minorities, 3% aboriginal, and 2% disability. This compares with 16%, 4%, and 4% for the population (Statistics Canada, prior census, 15-24 age group). Seventeen percent of the sample was Francophone, compared with 22% for the population (Statistics Canada, census, 15-24 age group). Overall, the sample is similar to the Canadian population for approximately this age cohort, though somewhat more educated.

Materials and Procedures
Those invited were offered an expenses-paid three night stay in Ottawa (the capital of Canada), travel included, but no other compensation. All participants arrived on a given Thurs., began their day Friday with questionnaires, including the questionnaire for this study, deliberated about the "future of Canada" Friday, Saturday, and part of Sunday, and were asked to do post-discussion questionnaires, including for this study, before leaving on Sunday. Time constraints limited the pre-and post-discussion research surveys each to ten minutes. Forty community, public and private sector leaders helped brief participants and answered questions. Participants were divided, initially, into four groups, each of which received briefings and held discussions on the topics of education, work, health, and the environment, respectively. Each of the four groups were further divided into four groups of about 10 persons each for more detailed discussions. Participants were encouraged to cross group boundaries to learn about other issues, and many did.

Measures
Measures are modeled after those in the Virtual Agora Project (VAProject), though, because of the limited survey space and differing aims of the study, fewer scales were given and fewer items were given for some scales. The pre-and postdiscussion design means that the same questions were asked both before and after discussion. False Consensus, expectations of unproductive conflict, political interest, and demographics are measured as in the VAProject (see above). Of the main authoritarianism measures, only RWA was collected. It was measured with two items, rather than the four items in the VAProject. Of the stealth beliefs, only the two items of the stealth beliefs #1 factor, no debate, were included, because these items changed mean value in response to the VAProject deliberation.

Factor Analyses
Both the pre-and post-discussion questions can be analyzed together, as separate estimates of the underlying factor-provided that the covariance relationships among variables remains the same both before and after. In addition, weighted average variables that include both pre-and post-discussion questions can be included in regression analyses, provided the covariance relationships remain the same. Testing, using cordif in R, indicates no significant differences among covariances or correlations from pre-to post-discussion. One other concern is missing data, which is inevitable with paper surveys. Only 1.4% of data values were missing. These were filled in with the R package "mice," using Bayesian imputation methods.
A confirmatory analysis for the variables making up scales for RWA, false consensus beliefs, and stealth democracy beliefs overall support the stipulated model.

Explaining Stealth Democracy
Before examining regressions, a prior issue is whether robust standard errors should be used in analyzing the CPRN data, given that participants were split into two groups and encouraged to meet with other groups. This suggests that groups may not matter much, a proposition that can be examined with multi-level models that allow groupings and nested groupings. Replicating the regression in Table 3, Column 2, no evidence was found for cluster error covaration. Consequently, cluster-robust standard errors will not be used. Table 3, Column 2 shows a partial replication of the analysis in Table 1, Column 3, with the current data. The replication is partial because only the RWA scale, of the main authoritarianism scales, was available in the current study. As the discussion will clarify, the RWA scale and its relationship with stealth beliefs is centrally important. Table 3, Column 2 indicates that both RWA and false consensus beliefs significantly increase stealth democracy beliefs, while education significantly decreases these beliefs, consistent with Table 1. Expectations of unproductive conflict have no effect on stealth beliefs. Column 3, for comparison purposes, shows an analysis of the same limited set of variables using the VAProject data. Importantly, there is no significant difference between the coefficients of RWA for the Column 2 and Column 3 models (bias-corrected accelerated, BCa, bootstrapped confidence intervals for the difference of the two coefficients indicates a p-value above .20, N=10,000 for bootstrapping). The difference of the education coefficients is also non-significant (BCa p>.15). The effects of false consensus beliefs is clearly indistinguishable between the models.

Effect of Deliberation on Stealth Democracy Beliefs
A t-test of the difference between stealth democracy beliefs measured from preand post-discussion finds a highly significant decrease in stealth beliefs (-.21, t=-3.33, p=.0006 one-sided). Because the before and after differences were measured within individual, it is unnecessary to control for other variables that remain constant within individual, such as demographics or stable attitudes. Note: All F-values<.0001. *** is p<.001; ** is p<.01; * is p<.05; † is p<.10 All pvalues are robust and account for non-independence by discussion group. Pvalues reported are one-sided for all non-demographic variables with coefficients in the expected direction.

Discussion Study Two
This section presents findings from a nationally representative sample of young adults with a good response rate, the CPRN study. With respect to the impact of RWA and other variables on stealth beliefs, the CPRN and the prior VAProject studies could have been drawn from the same population. Unfortunately, the remaining authoritarianism-related variables could not be included in the CPRN study due to space limitations. Nevertheless, as Study 1 shows, RWA largely explains all the covariances among the authoritarian constellation variables. Given RWA's centrality and the similar effects of RWA on stealth beliefs in the two studies, it is plausible that other authoritarian variables might have influenced stealth beliefs in the CPRN study.
Expectations of unproductive conflict did not have a significant effect on stealth beliefs in the CPRN study, though they did in the VAProject. This may due to the different contexts of the study. The VAProject, which focused on a highly contentious local issue, could have given rise to different expectations than the CPRN deliberations, which involved a general discussion of policies affecting the future of Canada.
Importantly, the CPRN analyses find that participation in the deliberation quite significantly reduced stealth democracy beliefs from immediately before the two and a half day deliberation to immediately after. This replicates the VAProject finding of significant change in stealth beliefs from before to months after the deliberation.

General Discussion
The research presented here has both strengths and limitations with respect to what it can reveal about deliberation and stealth democracy. The first study encountered data collection issues, and the second study had limited survey space. One result is that pre-and post-comparisons to determine whether deliberation significantly changed variables in the authoritarian stealth democracy model are limited to a few. Nevertheless, the data in both studies crucially establish that deliberation reduces a component of stealth democracy beliefs-the component less likely to experience a 'floor' effect. And the first study finds that deliberation significantly reduces two important inputs of stealth democracy beliefs.
One of the post-discussion measures examined in Table 2 (study 1), stealth democracy beliefs, was collected several months after the deliberation and after respondent attrition. Nonetheless, one of the two components of these stealth beliefs shows a significant decrease among online deliberators in this long run. Attrition of respondents raises the concern that only those who remained in the study showed this reduction. Then again, this means that 56% of the original sample randomly selected to participate in later months showed the reduction. Study 2 also provides a reassuring replication that finds immediate postdeliberation reductions in this component of stealth beliefs among participants.
With the exception of the stealth beliefs variable in study 1, all others were collected immediately pre-and post-deliberation in both studies. The research therefore does not establish whether vertical collectivism, expectations of conflict, or other variables show long-term changes. However, it may be unrealistic to expect one deliberation experience to permanently solve social ills. Regarding external validity, the research here includes one professionally-conducted deliberation and one deliberation that generally follows the Deliberative Poll methodology. The finding of reduced stealth beliefs also replicates in a third dialogue conducted by AmericaSpeaks. This creates the expectation that findings would generalize to common deliberation practices.
A related external validity issue is the capacity to generalize to the broader public.
Because of a relatively low response rate, Study 1 does not generalize well to the broad population, though it may say something about the more engaged public that comes to deliberations and that may be of special interest to policymakers. The likely greater engagement of participants may also mean that the attitudes and orientations of the general public would not be as consistent as found in the confirmatory factor analyses here. Study 2, however, is nationally representative and has a substantial response rate. Its findings appear to replicate the results of Study 1, though more limited in the variables considered. Overall, more research is needed to determine the full spectrum of authoritarian stealth democracy variables affected by deliberation, the duration of the effects, and what might prolong the effects.

Summary and Conclusion
The authoritarian stealth democrat thesis, which this paper supports, suggests that stealth democracy beliefs are driven by authoritarianism and authoritarianismrelated beliefs that are not basic, unchanging preferences. Instead, they are both changeable and problematic from the point of view of a democratic society. The widespread presence of stealth democracy beliefs raises the need for interventions to engage the public in ways that help that public embrace democracy rather than expect a non-existent consensus of experts and business people to choose policy. Findings here suggest that deliberative engagement helps reduce stealth democracy beliefs and the authoritarian inclinations that feed these beliefs.
In contrast, the stealth democracy thesis contends that most of the public has a basic preference for a democracy with little debate or compromise run by experts and business people. Hibbing and Theiss-Morse find the U.S. public generally agrees with questions tapping these stealth democracy beliefs. Their findings and theoretical assumptions lead these authors to conclude that encouraging political participation, particularly in the form of democratic deliberation, would delegitimize the political system. This paper proposes an alternative "authoritarian stealth democrats" thesis. It stipulates that stealth democracy beliefs are rooted in a problematic constellation of authoritarian orientations which may be ameliorated by involving people in political deliberation. The views and dispositions in this constellation include false consensus beliefs, fear of conflict, strong pro-authority attitudes, incapacity for socio-political perspective taking, and dispositions to cognitive lethargy.
This paper's findings are consistent with the authoritarian stealth democrats thesis. In Study 1 (Table 1), the authoritarian views and dispositions prove to be a much better explanatory model than Hibbing and Theiss-Morse's original model that focuses on false consensus beliefs, political disinterest, and aversion to conflict. Indeed, the latter two variables prove non-significant, challenging Hibbing and Theiss-Morse's interpretation of stealth democracy beliefs as rooted in understandable political disinterest and aversion to conflict. Authoritarian beliefs, associated in the literature with low cognitive engagement, intolerance, and punitive attitudes, are the most potent explanation of stealth democracy beliefs. Also, the Hibbing and Theiss-Morse interpretation does not explain why participants in the VAProject embraced stealth democracy both out of a belief that the public agrees on all the important issues and a fear of conflict in actual public discussions. The authoritarian stealth democrats thesis explains how this is possible.
The data here come from people who chose to participate in deliberations, which raises questions about how well findings generalize to the population. If there were any direct conflict between Hibbing and Theiss-Morse's survey findings and the Study 1 findings, then those authors' nationally representative dataset would be more persuasive. There is, however, no conflict of findings. The data for this paper replicates the key Hibbing and Theiss-Morse finding that a single "negative view of disagreement" variable, combining false consensus beliefs, aversion to conflict, and low political interest, strongly explains stealth beliefs. This replication suggests study participants are not atypical with respect to the stealth democracy thesis. Where the current paper departs from the authors' findings is in separately analyzing the components of "negative view of disagreement" and in adding the "authoritarian constellation" variables, which do better explaining stealth beliefs than the authors' model.
Also, the CPRN findings (Table 3) confirm that right-wing authoritarianism (RWA) false consensus beliefs, and education affect stealth beliefs to a statistically indistinguishable degree from their effects in the VAProject data (other variables were not included due to limited space). This provides some support that the findings here may generalize to a broader population, because the CPRN data involves a nationally-representative Canadian RDD sample with a more robust response rate. The authoritarian stealth democrats hypothesis not only clarifies why authoritarianism-related variables explain stealth beliefs among study participants, but also provides an elegant explanation of why deliberation reduces stealth beliefs and variables influencing these beliefs.
The Virtual Agora Project findings reveal that democratic deliberation mitigates two of the key factors behind stealth democracy beliefs, vertical collectivism and perceptions of conflict, as well as one of the two components of stealth democracy beliefs (other variables could not be tested with the available data). Importantly, the decline in stealth beliefs was found months after the deliberation. The CPRN data strongly replicate the result that the no-debate component of stealth democracy beliefs declines as a consequence of deliberation. Data from a third dialogue by AmericaSpeaks, not analyzed here, also shows significant pre-to post-discussion declines in stealth democracy beliefs (Muhlberger, 2007). Data from four deliberations, including the VAProject and CPRN, show that deliberations improved confidence in some governmental actors while not reducing confidence in others (Muhlberger, 2007). These findings conflict with the Hibbing and Theiss-Morse contention that efforts to engage the public may delegitimize the political system. Findings in the current paper suggest that, at least for people who choose to attend deliberations in the U.S. and a broader population sample in Canada, Hibbing and Theiss-Morse's concerns for system stability are incorrect.
Hibbing and Theiss-Morse's recommendations for improving system stability would likely not work if the authoritarian stealth democrats interpretation is correct. Authoritarian stealth democrats would not respond to the authors' proposed educational efforts meant to challenge the factual accuracy of their belief in a false consensus. The theory presented here suggests that authoritarians embrace a belief in a consensus not because they believe it is accurate but because they believe it is necessary to have a consensus for democracy to be possible at all. They likely restrict who constitutes the 'real' public to make belief in a consensus possible.
The authors' emphasis on system stability and legitimacy constitute a value commitment that can be challenged by other values, such as democracy. If political apathy were a basic, unchangeable preference, then academics should give up calls for a more engaged democracy. The authors have not, however, proven that preferences for apathy are basic and unchangeable. The current paper suggests that the stealth beliefs presumably emerging from apathy are driven by problematic authoritarian orientations, important components of which are ameliorated by democratic deliberation.
The authoritarian constellation of views and dispositions might be explained by cognitive development. Rosenberg (1988Rosenberg ( , 2002 finds that adults reason differently about causality, with some adults able to understand systems and others only able to understand simple linear causal relationships. People who understand systems are likely to appreciate a government involving a system of checks and balances, while linear reasoners would have difficulty understanding how social order could be possible without a powerful authority or public consensus. Authoritarianism is correlated with developmental measures of moral reasoning (Altemeyer, 1981;Boyes & Allen, 1993;McFarland & Mathews, 2005) and ego development (Browning, 1987). Additional research on the potential developmental basis of authoritarianism would be desirable.

APPENDIX A: CONFIRMATORY FACTOR ANALYSES
The surveys conducted for this study contain multiple questions for each of several conceptual factors, including the novel scales of social empathy and naive realism. This raises the issue of whether the questions that presumably tap a unique latent factor in fact each tap this factor and this factor alone. Another matter is whether the factors are all different from each other. Confirmatory factor analyses were conducted to address these issues, providing evidence for the validity of the question scales.
This dataset presents a challenge for confirmatory factor analysis in that balanced question sets typically split into two strong and separate factors-one for positively worded questions and one for negatively worded questions, even for such well-known question scales as RWA and SDO. This outcome is likely the consequence of participants being asked to answer a lengthy online survey in a limited time period. In order to remain consistent while rapidly answering questions, participants appear to have clustered answers to similarly valenced questions together. Coordinating such clustering across negatively and positively worded questions would be cognitively challenging, so instead two separate clusters appear to have emerged-one for each type of wording. Reassuringly, averaging questions across negative and positive wordings generally creates indicators that correlate better with other variables. Nevertheless, standard confirmatory factor analyses detect a meaningless factor for each question valence. To address this problem, an adjustment was made to the confirmatory model that allows for meaningless clustering of positive and of negative question wordings. Negatively worded questions were allowed to load on a negative wording factor, and similarly for positively worded questions. The two factors were given identical variances but not allowed to covary. This solution worked well. The adjusted model is identified.
Individual confirmatory factor analyses were conducted on each of the question scales to determine if a single factor could adequately explain the proposed set of questions. Table I shows the results of these analyses. A Goodness of Fit Index (GFI) of .90 and above is considered very good. It roughly corresponds to an R2 in regression analysis. An Adjusted Goodness of Fit Index (AGFI) above .80 is considered acceptable by some, above .90 by others. Both GFI and AGFI can be misleading in certain cases, as will be evident shortly. A Root Mean Squared Error of the Approximation (RMSEA) of .05 or below is considered good and above .10 is considered bad. Table I also reports a 90% confidence interval for RMSEA. Lower bounds below .05 are considered desirable. The Bayes Information Criteria (BIC) is a statistical-theory based measure that can be used to determine how much more probable one model is than another to be the true model behind a body of data. A BIC of zero indicates that a model is as probable as the saturation model, which is a model that simply stipulates that every measure is correlated with every other measure. Such a model perfectly fits the covariance matrix of the data but requires many parameters. A BIC below zero indicates a model that is more probable than the theory-free saturation model. Hoelter's N indicates the sample size below which the researcher could accept the hypothesis that the theoretical model adequately explains all the observed covariances of the variables. Beyond a certain size sample, this hypothesis is almost always rejected because of small and usually meaningless divergences from the theoretical model. A Hoelter's N of 200 or above is considered good. Table I shows that a single factor explanation of stealth beliefs does not adequately fit the data according to the more theoretically-driven indicators, while a two factor model fits well. Hibbing and Theiss-Morse (2002) simply assume the unidimensionality of stealth beliefs. One factor captures a dislike of discussion and compromise (henceforth "stealth beliefs #1-no debate"), and the other captures a belief that elites should rule (henceforth "stealth beliefs #2-elite rule"). Confirmatory analysis shows the correlation of these two factors is a substantial .68. Because of the high correlation, the two components of stealth beliefs may simply be variations of a common second-order factor. If so, the subscales of stealth beliefs need to be examined for differences in their relationships to other variables. Where differences do not emerge, they can be collapsed.
Rows 4 to 9 of Table I show that a single factor and, in the cases of vertical collectivism and social empathy, two factors adequately explain the relationships among indicators for each of these question scales, analyzed individually. Each of these models meets all criteria for GFI through Hoelter's N. Vertical collectivism loads on two factors that appear to capture willingness to make sacrifices for the group (henceforth "vertical collectivism #1-sacrifice"), while the other captures feelings of duty to the group ("vertical collectivism #2-duty"). Social empathy consists of two related factors, one of which involves respondents expressing the belief that they generally take into account many points of view in thinking about political issues ("social empathy #1-many views"), while the other involves questions that specifically ask if the respondent takes into account the views of people in different classes and racial groups ("social empathy #2-class and race). RWA is not included in the single factor models because its confirmatory model would not converge. False Consensus is also not included because it has too few variables for a standalone model. Order is the same comprehensive model but with a second-order factor, the "authoritarian constellation," that explains the covariances of all other factors. A standalone RWA model does not converge, and False Consensus has only two variables-insufficient for a standalone model. Analyses were conducted in the R statistical package "sem." a Two-factor model.
The "All Factors" row tests a comprehensive model in which all factors other than stealth beliefs are included. All Factors includes False Consensus and RWA. The GFI and AGFI for this model appear low, but this is an artifact of the modelparticularly an unusually large number of variables. As the number of variables increases, GFI and AGFI decrease because of an accumulation of differences between actual and model-predicted variances and covariances. RMSEA, BIC, and Hoelter's N indicate the model fits the data exceedingly well. The BIC score suggests the model is vastly more probable, under Bayesian theory, than the saturation model. The proposed factor structure thus adequately explains the data. Additional factors are unneeded and variables group together in coherent factors. All factor loadings, of which there are 43, are highly significant, with median absolute z-score of 11.43 and a minimum absolute z-score of 2.8 (p=.005) and the next lowest absolute z-score of 3.48 (p<.001). In addition, the model shows that all the factors are statistically different from each other. Likelihood ratio tests checking whether the factor correlations equal one reject this possibility with χ 2 values all exceeding 11.6 for 1 d.f. (p=.0007).

Existence of an Authoritarian Constellation Second-Order Factor
The final row of Table I helps establish the existence of a second-order authoritarian constellation factor. In theory, the authoritarian constellation is the common factor that explains the covariations of all the other factors. The "Plus 2 nd Order" model tests precisely this. It is the All Factors model plus a secondorder factor that explains all covariations among the eleven first-order factors. This model replaces 54 parameters that permit free covariation among the factors with 11 parameters that only allow covariation between the second-level authoritarian constellation factor and each first-order factor. Adequate model fit would indicate that these 11 covariations with an underlying authoritarian constellation factor adequately explain the 'observed' 54 covariations of first-order factors. That is, the factors covary with each other by virtue of their common relationship with the authoritarian constellation factor. The model could very well turn out not to fit properly.
Though the model fits mildly less well than the All Factors model, it readily passes the RMSEA, BIC, and Hoelter's N criteria. Indeed, BIC gives the model a strong vote of confidence, indicating that it is much more likely (7.9*10 30 times more probable according to Bayes factors) to be the correct model than the All Factors model. The results support the view that all the observed covariances of variables across factors can be explained by the factors' relationship to a single underlying factor. The hypothesis of an underlying authoritarian constellation factor proves very helpful in explaining the data. All factor loadings between first-and second-order factors have z-scores of at least 4.38 (p<.00001). The loadings, given how the model is designed, are roughly similar to correlations between each first-order factor and the second-order factor. The mean loading is .56 and median is .5, with an observed range of .31 to .87.
For subsequent analyses, observed variables were combined into scales to measure the factors identified using factor scores regressions. The All Factors model was used to determine the weights of most variables. To avoid positivity bias (response acquiescence), positively and negatively worded questions were each given the same cumulative weight-each set affected the derived scales equally. Plus 2 nd Order determined the weights for a authoritarian constellation scale that is formed from an average of all variables for the eleven scales.
Weights for Stealth Beliefs were determined by a standalone model to insure no transfer of information between the dependent and independent variables.

Convergent and Divergent Validity
The analyses in the last section confirm the construct validity of the measures by showing that they cluster together in the expected factors and that just these factors are needed to explain all observed covariations of the constituent variables. Table 1, however, does not clarify whether the factors covary with each other in the expected direction and whether they covary significantly, as would be expected from the authoritarian stealth democracy thesis. Table II addresses these issues by showing the observed correlations between the scales formed from weighted averages of the variables, as discussed above. If correlations prove significant and in the correct direction, this would help affirm the convergent and divergent validity of the scales. Table II shows a pattern of significant correlations that is remarkably consistent with expectations from the authoritarian stealth democracy thesis.
The correlations between the authoritarian constellation scale and all the other scales is always highly significant and in the expected direction. Between the remaining first-order variables there are 91 correlations, 80% are significant and in the expected direction. An additional 8% are in the expected direction and show a statistical trend. Of the remaining correlations, only two are significant and in an unexpected direction. These are the correlations between Vertical Collectivism #2 (Duty) and both Social Empathy #1 and #2 (many views, class and race), which are significantly positive.