I. INTRODUCTION

"And then I see the disinfectant, where it knocks it out in a minute," President Donald Trump said about the possibility of using bleach to treat COVID-19 during a press conference. "One minute," he exclaimed. "And is there a way we can do something like that, by injection inside or almost a cleaning? [Trump asks of CDC administrators.] Because you see it gets in the lungs and it does a tremendous number on the lungs." (WSLS 10 Local News 2020). When presidents make specious justifications like Trump's bleach idea, how does the public react?

This study explores why elites, and in particular presidential candidates, are seemingly unconstrained by the justifications delivered to the public. This study uses the term "justifications", which can be described as messages delivered by politicians to the public about policy (Broockman and Butler 2017; redacted). Justifications provide the reasoning for why the public should support or oppose a policy. Examples of political justifications can include slogans that are only a few words, a detailed press release, a letter to constituents, a statement made to the press, or a speech. The term broadly describes communications to the public by officeholders.

Justifications can be thought of as a type of emphasis frame in the sense that a politician can select one consideration (from a number of possible options) to explain their policy position (Tversky and Kahneman 1981; Druckman 2001). Emphasis frames are typically used to describe how the media covers an issue or event, and they are similar to justifications because both provide context. Likewise, justifications and emphasis frames can influence how the public views events, policies, and politicians (Broockman and Butler 2017; Iyengar and Kinder, 2010; Druckman 2001).

The analysis to follow varies the content of the justifications from reasonable, regularly used justifications to specious and uncommon. In other words, does the content of the justification matter? Or do politicians simply have to provide some reasoning for their position regardless of how cogent it is?

Testing specious justifications provides leverage on evaluating the persuasiveness of reasonable justifications. Thinking about justifications as a continuum from reasonable to specious, a study might have different conclusions about persuasion if treatments included justifications that were incoherent or ridiculous compared to reasonable ones. Outside of the experimental setting, most successful politicians rarely make a mistake of justifying a position with a specious justification that is wholly unreasonable or utterly ridiculous because they know it could jeopardize their presidency or candidacy. Most successful politicians use "message tested" justifications that they believe will persuade the public of their worldviews (Druckman and Jacobs 2016). The motivation behind this test of specious justifications is precisely because they are used infrequently. It is difficult to evaluate the persuasiveness of a reasonable justification without aspecious comparison. In other words, to understand the power of reasonable justifications from politicians, it might be useful to compare them to specious ones. Therefore, specious justifications are used in this study for causal inference reasons, not because every politician uses them.

Some persuasion research downplays the role of justifications regardless of their content; instead, elites shape public opinion by using cues or heuristics (i.e., Lenz 2009; Achen and Bartels 2017). Moreover, the most efficient political cue that provides a shortcut to opinion formation is party identification (i.e., Zaller 1992). The most striking evidence comes from studies of opinion change over time that show an individual will switch their position on an issue after a public debate or campaign in order to match their representative instead of reconsidering their support for the representative (Broockman and Butler 2017). To that end, the experiments to follow will also measure the effect of partisan cues on support for policies separated from justifications.

In the experiments to follow, I also vary the messenger and the policy. I randomly assigned issue positions and justifications to actual politicians while they were in office (e.g., Barack Obama, Joe Biden, and John McCain). Candidates are associated with positions that range from policies that are regularly occurring during campaigns (Social Security or education reform) to policies that are almost never observed during campaigns (colonizing the Moon). This variation is especially important because previous research suggests that a connection between the public and an agency is dependent on the saliency of an issue (Wlezien and Soroka 2012), and the policies used in this study are highly salient (taxes for Social Security), modestly salient (education reform), and not at all salient (colonizing the moon).

Results show that partisan cues and justifications can influence the support for policies, but the effect of specious justifications is not overwhelmingly different from reasonable ones except in the case of Social Security (a highly salient issue). In addition, the views of agencies are not affected by specious or reasonable justifications regardless of the issue. This suggests that views of public administration are somewhat detached from the rhetoric and policy positions of presidents and presidential candidates.

II. THEORETICAL EXPECTATIONS: JUSTIFICATIONS, CUES, AND THE THERMOSTATIC PUBLIC

Decades of political science research demonstrates thatelites can shape public opinion by using partisancues to generate support for issues (Converse 1964; Zaller 1992; Achen and

Bartels 2017). Voters only need to know the party of the elite in order to cue support or opposition to a policy (Zaller 1992). It could be the case that poorly constructed justifications are secondary to partisan cues. These observations led to the conclusion that voters are "following" elites for various reasons, such as trusting the party, but none of which have to do with the content of policies or justifications (Lenz, 2013). Therefore, elites might have a lot of leeway when making public statements – if the public simply follows along with cueing, then the content of the justification or policy will be irrelevant. As robust as the findings on cues might be, elite discourse tends to contain more than just cues – elites do not just announce their issue positions and party identification in isolation. Elites will give reasons for their positions.

Justifications include different frames meant to persuade the public. Framing effects are characterized as either a semantic difference of the same object or an emphasis of one relevant consideration about an object while ignoring others (Tversky and Kahneman 1981; Druckman 2004; Chong and Druckman 2007). For example, an individual might support a hate rally in her town if a justification for it is framed in terms of free speech, but the same individual might oppose a hate rally if it is framed in terms of public safety (Iyengar 1987; Druckman 2004). Therefore, emphasis frames from elites are justifications that provide individuals with reasons why they should support or oppose an issue by emphasizing "a subset of potentially relevant considerations" about a policy (Druckman and Nelson 2003, 730). These justifications can also be transmitted from peer-to-peer, not just top-down from elites (Druckman and Nelson 2003). And since the knowledge of citizens is limited, a justification can help citizens make a more informed opinion about policy. For example, those at the lower level of the political knowledge spectrum benefit the most from competing arguments about policies (Sniderman and Theriault 2004).

The present study goes a step beyond weak or minimal justification. The experiments to follow randomized justifications and policies that are wholly specious or ridiculous. Aside from Donald Trump, a common occurrence in American politics involves politicians using similar policies and justifications as their co-partisans to promote policies. These positions (and the messages used to justify them) are not randomly determined and are rarely haphazardly decided upon. In fact, elite justifications are usually predictable -- parties tend to coalesce around one policy and only a handful of justifications areused to support that policy. While justifications have become more specious in recent years, including the bleach example, we know comparatively little about how voters evaluate government leaders or agencies who use specious justifications.

Politicians and parties try to limit the policy choices and justifications that are supported publicly. Elites strategically pick issues that can benefit their party (Cox and McCubbins 2005; Sniderman and Bullock 2004). Elites only present the "menu" of policy choices that benefit them, which only gives the public a handful of polices to choose from (Sniderman and Bullock 2004). Moreover, policies that receive roll call votes are strategically decided by parties while undesirable polices are purposely kept off the agenda (Cox and McCubbins 2005). I extend this logic from issue position to include justifications. From this perspective of menu dependence(Sniderman and

Bullock 2004), when voters switch issue positions to match their preferred candidate, they are attempting to maintain consistency given the available menu of issues and justifications. And that menu is usually very limited, sometimes only one policy options with competing justifications. This reality makes it difficult for researchers to evaluate the causal effects of a range of different justifications, most of which might be strategically left off the 'menu'. The experiments to follow expand the menu of options to include specious policies and justifications as a way of providing a fuller picture of the influence of elite messaging.

This study extends previous research on elite cues and justifications to include views of administrating agencies and their spending priorities. Important studies exist showing a dynamic between the public and agencies. Agencies react to public opinion, and governing institutions can change the public's preferences for policies by changing spending priorities (Wlezien, 1995; Wlezien and Soroka, 2012). Governing agencies can spend more money on programs to match policy demand from citizens, and as a result, the public becomes less demanding for a given policy. These results are consistent in the United States, United Kingdom, and Canada. This relationship between institutions and the public demonstrates a "thermostatic model" where demand and policy move in opposite directions (Wlezien,1995). The public needs "clarity" on what agency is responsible for which policy, and therefore, the public still must rely on elite cues and communication to evaluate agencies. The experiments that follow evaluate how policy preferences and spending priorities of agencies change from partisan cues and justifications.

III. EXPERIMENTAL METHODOLOGY

3.1 Policies Selected

Two outcome measures for each experiment tap into policy or agency support. Respondents are asked about support for colonizing the moon and increasing funding to NASA (experiment one and two), support for lengthening the school day and increasing funding to the Department of

Education (experiment three), and support for increasing taxes to pay for Social Security benefits (experiment four). The reasoning behind picking these policies is explained below.

I varied the policies and agencies in terms of familiarity. By familiarity, I mean the level of exposure to an issue from elite messaging (Zaller, 1992). Unfamiliar issues are not debated regularly by elites, and therefore, most individuals have not been exposed to any messages about unfamiliar issues (Zaller 1992). If the public is less familiar with an issue, they might be less likely to reject a specious justification associated with it. This design provides leverage on the effect of cues and justifications with different baseline levels of familiarity for policies and agencies.

Colonizing the moon is an issue that is not only unaligned with party identification, but it is also an issue that most citizens have never thought about (although space travel has received increase interest in recent years, fielding of this study in 2014 pre-dates those trends). Perhaps as a last-ditched effort to win the Republican presidential primary elections in 2012, Newt Gingrich declared that America "will have the first permanent base on the moon" if elected, but this policy understandably was not taken seriously by other candidates or the media (Sneed, 2012). In fact, colonizing the moon is such an unfamiliar issue that I cannot find any reliable polling data on it before fielding this experiment. But just to reiterate, picking an unfamiliar and possibly specious policy provides leverage on the limits of partisan cues and justifications.

Increasing the length of the school day is a more familiar issue than colonizing the moon. Voters might have heard the debate about lengthening the school day at the state or local level, have children in public school, or simply attended public school themselves, and therefore might be familiar enough with the issue to provide an opinion when asked. Increasing the length of the school day became law in five U.S. states by 2014 (Colorado, Connecticut, New York, Tennessee, and Massachusetts). Former New Jersey Governor Chris Christie argued in his 2014 State of the State address, "I believe we need to take bigger and broader steps to adjust our approach to K-12 education to address the new competitive world we live in.... It is time to lengthen both the school day and school year... student achievement is lagging at the exact moment when we need improvement more than ever in order to compete in the world economy..." (CBS Philadelphia 2014). Public opinion data about increasing the length of the school day indicates that the issue was not aligned by party. A poll conducted by The Times Union and Siena College in New York in June of 2014 found that 36 percent of Republicans and 37 percent of Democrats favor increasing the length of the school day. Moreover, a poll conducted in Virginia in 2008 found that 36 percent of Republicans and 39 percent of Democrats favor increasing the length of the school day.

In my fourth experiment, I tested one very familiar policy proposal, Social Security reform, which elites debate regularly. For example, Social Security was the most salient issue in the 2000 presidential election. George W. Bush also picked Social Security reform as his major policy goal after the 2004 election. In my experiment, I tested a proposal to raise the Social Security contribution rate for high income earners, and my justification for this reform mirrors an argument used by the Obama Administration. Although there is disagreement at the elite level, Democratic and Republican voters agree on the importance of Social Security. For example, Pew Research reports that 74% of American do not want Social Security benefits cut.

3.2 Randomized Conditions and Outcome Measures

The design of all four experiments is similar. Outcome measures asked respondents their support for a policy and support for the agency that administers the policy. The control groups simply asked these questions with no additional information. Each treatment arm builds on the control groups by adding additional information for the respondent to consider. Partisan cue groups added an endorsement of the policy from a well-known partisan. From experiment one to four, my partisan cues are President Barack Obama, Presidential Candidate John McCain,

Vice President Joe Biden, and President Barack Obama. The remaining treatment arms build on the partisan cue groups by adding justifications for why the partisan supports the policy. By including a partisan cue in all treatment groups, I can measure the additional support that is generated by the justification.

Justification treatments include variation on the quality of the message delivered by elites. The text of each can be found in Table 1. Note that the colonizing the move justification is considered specious because no lifesaving minerals exist on the Moon. Each column summarizes a given experiment and should be read vertically. Each subsequent row starting from the top includes only new information that is added in the treatment. See appendix for full questionnaires. If specious justifications advocated by elites are effective in increasing support for a policy or agency, then the content of the justification is less important, and therefore, voters are simply taking cues from elites.

Table 1: Summary Each Randomized Treatment Arm

	Social Security Reform	Lengthening the School Day	Colonizing the Moon
Control	Would you favor or oppose a proposal to raise the Social Security contribution rate for high income individuals, or haven't you thought much about this?	Would you favor or oppose increasing the length of the school day by an hour for Kindergarten through 12th grade, or haven't you thought much about this?	Would you favor or oppose colonizing the moon, or haven't you thought much about this?
Partisan cue	Would you favor or oppose a proposal by President Barack Obama...	Would you favor or oppose a proposal by Vice President Joe Biden	Would you favor or oppose a proposal by President Barack Obama (experiment 1) / former presidential candidate John McCain (experiment 2)...
Cue plus reasonable justification	... because all seniors should be able to retire with dignity, not just a privileged few...	... in order to keep American students competitive in the global economy in science and math...	(not applicable)
Cue plus specious justification	... because all seniors should be able to go on more vacations...	... in order to teach vocational skills in areas America is still competitive like cosmetology and automotive repair...	... in order to discover potentially lifesaving minerals not available to doctors on earth...

3.3 Sample

This study uses data from four survey experiments administered online in 2014 by the survey research firm Penn, Schoen and Berland (PSB). Total sample size across all experiments is 3,923. PSB recruits respondents for surveys with pre-treatment characteristics to mirror a likely voter sample. They create their samples using marginal distributions by age, gender, and race. These respondents are then sent emails to take the survey. For this study, subjects must have said that they voted in the 2012 presidential election at the beginning of the survey. Those who did not self-identify as a voter were removed from the survey. This makes my sample more interested in, and knowledgeable about, politics than a general population sample. Lastly, I did not weight the data to match a voting sample. Miratrix, Sekhon, Theodoridis, and Campos (2018) argued that unweighted analysis of experimental data is preferred because weighting does not improve estimates and would require a larger sample size to compensate for the loss in statistical power. I am most interested in differences between treatment and control, and the analysis to follow focuses on these differences. See the online appendix for sample demographics and balance tests demonstrating that the randomization was successful

I estimated ordinary least squares (OLS) regressions, although I present the results graphically in the results section:

All dependent variables are scaled from zero to one, with higher values indicating more positive evaluations for the policy or agency. All independent variables, including controls, are indicator variables (1=yes, o=no). The excluded treatment arm is the control group, such that , , and are all differences from the control and therefore measure treatment effects. I present these results graphically to show differences from the control group. The vector Controls included a number of variables meant to account for expected differences in my dependent variables including party identification, age, race, gender, education, and region of the country that the respondent resides. I use controls even though the data comes from random assignment in order to reduce error that arises from sampling variability (Gerber and Green, 2012). Therefore, my treatment results are covariate adjusted. Full regression results appear in the appendix on the author's website. In the appendix, I also run each model stratified by five-point party identification to show heterogeneous effects, and these results are noted in-text when they differ from the main result. Sample size is limited when stratifying by party, so those results are estimated with more error.

IV. RESULTS

All figures are formatted identically. The dotted lines in each plot represent evaluations in the control group, and each point is the difference from the control with 95 percent confidence intervals. Confidence intervals that cross the dotted line, therefore, indicate no difference compared to the control. The top panel of each plot are policy support and the bottom panel is support for the agency (when applicable). Please note that I report in-text treatment effects with p-values regardless if they reach the traditional threshold of significance – this is done for completeness so that all results are reported regardless of their significance.

In Figure 1, I start with results from experiment one. I find a 6.4 percentage point drop (p <.001) in support for colonizing the Moon when only Barack Obama is associated with the policy. This suggests that a cue alone reduces support for an unfamiliar and potentially specious policy. These results are consistent with past experimental research that shows a partisan cue alone can only serve to reduce (reasonable) policy support (Nicholson, 2012). When looking at these results by party, I find that this reduction is concentrated among Republicans (the out-party relative to the partisan cue) and Independents. The cue plus justification group, however, improves on the cue group but only reaches levels that are indistinguishable from the control group. That is, I find a 1.1 percentage point drop (p =.54) in support for colonizing the Moon when Obama justifies the policy with a justification about finding lifesaving minerals on the Moon. These results mask heterogeneous effects. Strong Democrats are significantly more likely to favor the policy with a justification, and strong Republicans are significantly more likely to oppose. In the bottom panel of Figure 1, I display the second outcome measure of funding for NASA. This question considers whether or not the specious policy and justification can influence views of the agency that is known for space exploration. I find no spillover to the governing agency even when I find reduction in support for the policy. I find a 2.9 percentage point drop (p =.17) in support for NASA with an Obama cue and a 2.6 percentage point drop (p =.24) in support with a cue and justification. Taken together, policy support can change from cues and justifications, but views of the agency remain unmoved even with a specious policy and justification.

{"image_source":{"path":"images/5f40112d87a47938b18f4a18be0e39275ec980561356f5a6634047559bd8661b.jpg"},"content":"","chart_caption":[{"type":"text","content":"Treatment Effects for NASA Funding with Obama Cue"},{"type":"text","content":"Treatment Effects for Colonizing the Moon with Obama Cue"},{"type":"text","content":"n=1,002. Plotted quantities are change in policy support relative to the control group with 95% CI. Lincom estimates from regression with controls. See Appendix for full results."}],"chart_footnote":[]} {"image_source":{"path":"images/3aa9e4306fda5d41b56adfd194b979af4321c4cf00302707139e2b68b0b3754f.jpg"},"content":"","chart_caption":[{"type":"text","content":"n=1,002. Plotted quantities are change in agency support relative to the control group with 95% CI. Lincom estimates from regression with controls. See Appendix for full results."}],"chart_footnote":[]}

Figure 1: Experiment 1, Support for Colonizing the Moon and NASA Funding

Experiment two replicates experiment one but with a Republican partisan cue, former presidential candidate John McCain. The findings in Figure 2 are strikingly similar to experiment one. I found a significant reduction in support for colonizing the Moon with a partisan cue only, and I find no difference between the control and the cue plus justification group. Like experiment one, the justification seems to compensate for the reduction in support that is found with a partisan cue only. Specifically, I find a 5.2 percentage point drop (p =.006) in support for colonizing the Moon with a McCain cue and a 0.2 percentage point drop (p =.91) in support with a cue and justification. Likewise, results for NASA funding mirror my results in experiment one, which show no significant differences between either treatment groups and the control group. I find a 2.7 percentage point drop (p =.24) in support for NASA with a McCain cue and a 0.8 percentage point drop (p =.73) in support with a cue and justification. Taking experiment one and two together, my results suggest that public perceptions of NASA are durable when connected to a specious issue, and this result is consistent across partisan cues. In addition, the justification does not improve support for the policy compared to the control group, suggesting that justifications do not move respondents toward a potentially specious policy.

{"image_source":{"path":"images/c6df95c46fc93a6db6f92253f4c6207972a38f31248370552a03f75f4861f51f.jpg"},"content":"","chart_caption":[{"type":"text","content":"n=983. Plotted quantities are change in policy support relative to the control group with 95% CI. Lincom estimates from regression with controls. See Appendix for full results."}],"chart_footnote":[]} {"image_source":{"path":"images/6d5a91a3c52521a5be0c74a3665289350bc0f0bda04394407cb28bae48c57b8f.jpg"},"content":"","chart_caption":[{"type":"text","content":"n=983. Plotted quantities are change in agency support relative to the control group with 95% CI. Lincom estimates from regression with controls. See Appendix for full results."}],"chart_footnote":[]}

Figure 2: Experiment 2, Support for Colonizing the Moon and NASA Funding

What about when the policy is not specious but one of the justification is? Experiment three tests lengthening the school day with a partisan cue, a reasonable justification, and a specious justification. These results can be found in Figure 3. The pattern of policy support is different than colonizing the Moon. Instead, the partisan cue shows almost no difference from the control group, but cues with justifications increase support relative to the control. I find a 1.1 percentage point increase (p =.67) in support for lengthening school with a Biden cue, a 4.8 percentage point increase (p =.08) in support with a cue and justification about increasing competitiveness in the global economy, and a 7.8 percentage point increase (p =.004) in support with a cue and justification about remaining competitive in more populist industries. Interestingly, the more reasonable justification is less persuasive as the more populist message that contain specious elements. This result might also be consistent with the rise in populist candidates in the U.S. soon after these experiments were run. For example, the specious justification primes working class identities (automotive and cosmetology). It is consistent with populist messages that argue working class people are the in-group and should be represented (Mudd 2004). Moreover, populist messages might be cognitively easier to evaluate.

However, for views of the Department of Education, I find similar non-significant results for all of the treatment conditions. Specifically, I find a 2.0 percentage point drop (p =.44) in support for the Department of Education with a Biden cue, a 4.0 percentage point drop (p =.13) in support with a cue and justification about increasing competitiveness in the global economy, and a 1.8 percentage point drop (p =.46) in support with a cue and justification about remaining competitive in more populist industries. This suggests that party leaders have control over public opinion with regard to policies, but those effects do not extend to administrative agencies.

{"image_source":{"path":"images/e9cde07f38f8c57e82218457b349b6f811041805fd06b7ae3bec9efab63e74cb.jpg"},"content":"","chart_caption":[{"type":"text","content":"n=961. Plotted quantities are change in policy support relative to the control group with 95% CI. Lincom estimates from regression with controls. See Appendix for full results."}],"chart_footnote":[]} {"image_source":{"path":"images/a353148dcae73667bf503eb4dc21393edc8b6ef0598b60aba6127c41480428c9.jpg"},"content":"","chart_caption":[{"type":"text","content":"Treatment Effects for Department of Education Funding with Biden Cue"},{"type":"text","content":"n=961. Plotted quantities are change in policy support relative to the control group with 95% CI. Lincom estimates from regression with controls. See Appendix for full results."}],"chart_footnote":[]}

Figure 3: Experiment 3, Support for Lengthening School and Department of Education Funding

Finally, Figure 4 displays results from experiment 4, which evaluates funding for Social Security benefits. As a reminder, Social Security is a very familiar government program that enjoys broad support. My results demonstrate that support can change with cues and justifications. With a cue alone, I find a 4.7 percentage point drop (p =.053) in support for taxing the rich to fund Social Security. With a justification about retiring with dignity from Obama, I find a 3.3 percentage point increase (p =.163). Note that the increase from the cue treatment group to this justification group exceeds 7 percentage points. The biggest treatment effects, however, includes the justification about retirees going on vacation to the beach. I find a 14.4 percentage point drop (p<.001) relative to the control group. These results suggest that a president can slightly improve support for a well-known policy with a reasonable justification, but a seemingly ridiculous justification can make an otherwise popular policy less popular. But compared to the previous experiments, this pattern is only present for a very salient issue like Social Security. Strong Republicans, who were most critical of Obama's tenure in-office, are not receptive to any treatment condition – all show roughly the same reduction in support relative to the control group.

{"image_source":{"path":"images/6a7c8c7a521e77bd6f97bd3db071b7a409624f948b9afeff8af0a34498ff3e39.jpg"},"content":"","chart_caption":[{"type":"text","content":"n=997. Plotted quantities are change in policy support relative to the control group with 95% CI. Lincom estimates from regression with controls. See Appendix for full results."}],"chart_footnote":[]}

Figure 4: Experiment 4, Support for Social Security Administration

V. CONCLUSION AND DISCUSSION

These results demonstrate that respondents evaluate policies and agencies differently, where respondents were most sensitive to treatment conditions when evaluating policies. Partisan cues decrease support for policies on average. These results are consistent with past research that shows cues alone reduce support for policies (Nicholson 2012). Justifications are also most effective at changing support for familiar issues, which is consistent with past work that shows familiarity of a policy can change the importance of elite messages (Zaller 1992). For Social Security, a reasonable justification was essential for increasing support, and a specious justification caused backlash, and this might be due to how familiar the issue is to citizens. However, I find null results for all conditions when asking about agencies, which suggests that views of agencies are somewhat divorced from views of presidents and aspiring presidents who might oversee them.

Aside from Social Security reform (a very familiar policy), specious justifications and issues had modest effects on opinions. These results suggest that presidents have a substantial amount of leeway when discussing their policy objectives. They are not punished when justifying a specious policy (colonizing the moon), and they can actually improve support for a policy (lengthening the school day) from a specious, populist justifications. Aalbeit the reasonable justifications about global competitiveness was roughly as effective as the populist justifications, but the point is that both justifications were effective regardless of the content.

These results have implications for politics and policy. On one hand, results demonstrate that politicians do not have to be precisely careful about their justifications when discussing policy. Moreover, unfamiliar issues provide even greater opportunity for specious justifications – only the well-known issue of Social Security yielded a large backlash. As a result, when a new issue takes center stage, like the Trump/COVID-19 example from the introduction, specious justifications might be more acceptable to the public compared to long-standing issues like Social Security. This poses a problem for democratic accountability because politicians are less likely to be punished for speciousness in times of uncertainty. On the other hand, specious justifications do not influence perceptions of the agencies responsible for policy implementation. The legitimacy of an agency is needed for effective policymaking, and results show that specious rhetoric from partisan leaders does not undermine an agency's authority. Taken together, these results suggest that specious justifications can modestly influence policy support, but it does not spillover to agency support.

APPENDIX

Table of Contents

1. Experimental Questionnaires, page 2

2. Balance Tests, page 14

3. Regression Tables of Main Results, page 21

4. Regression Tables by Party Identification, page 29

1. Experimental Questionnaires

The experiments in this study were survey questions added to an "omnibus" poll that was fielded once a month by the survey research firm Penn, Schoen, and Berland. Omnibus polls include questions on a variety of political and non-political topics submitted by a variety of researcher. I was able to submit a limited number of questions about party identification, randomized policy questions, and questions about agency support. I did not have control over which demographic questions were and were not asked before my experiments. For example, I was only able to ask level of education for experiments 3 and 4. Also see regression tables for specific demographics that were included.

Experiment 1 Which of the following best describes you...

randomly reverse answer choices

1) Strong Democrat

2) Not Strong Democrat

3) Independent

4) Not Strong Republican

5) Strong Republican

6) Other ##OPEN END

ASK IF Q1=C3 OR Q1=C6 # But if you had to choose, which party would you say you are closer to?

randomly reverse answer choices

1) Closer to Democrats

2) Neither party

3) Closer to Republicans

Here is the one question experiment: respondents are randomized into 1 of 3 Qs

randomly select one question from Q4, Q5, OR Q6 TO ASK

3. Hidden Question, Randomly Select One Choice:

1) Control

2) Candidate Cue

3) Justification

control

ASK IF Q3=C1# Now we have a few more questions about issues that may become important in the future. Would you favor or oppose colonizing the moon, or haven't you thought much about this?

randomly reverse answer choices

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

#candidate cue

5. ASK IF Q3=C2# Now we have a few more questions about issues that may become important in the future. Would you favor or oppose a proposal by President Barack Obama to colonize the moon, or haven't you thought much about this?

randomly reverse answer choices

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

justification treatment

6. ASK IF Q3=C3#Now we have a few more questions about issues that may become important in the future. Would you favor or oppose a proposal by President Barack Obama to colonize the moon in order to discover potentially lifesaving minerals not available to doctors on earth, or haven't you thought much about this?

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

everyone sees this question

7. And what about space travel more generally? Would you favor a proposal to increase the budget of the National Aeronautics and Space Administration (NASA)?

randomly reverse answer choices

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

Experiment 2

Find the direction they will follow

1. Which of the following best describes you...

randomly reverse answer choices

1. Strong Democrat
2. Not Strong Democrat
3. Independent
4. Not Strong Republican
5. Strong Republican
6. Other ##OPEN END##

ASK IF Q1=C3 OR Q1=C6 # But if you had to choose, which party would you say you are closer to?

randomly reverse answer choices

1. Closer to Democrats
2. Neither party
3. Closer to Republicans

Here is the one question experiment: respondents are randomized into 1 of 3 Qs

randomly select one question from Q4, Q5, OR Q6 TO ASK

3. Hidden Question, Randomly Select One Choice

1) Control

2) Candidate Cue

3) Argumentation

ASK IF Q3=C1# Now we have a few more questions about issues that may become important in the future. Would you favor or oppose colonizing the moon, or haven't you thought much about this?

randomly reverse answer choices

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

#candidate cue

ASK IF Q3=C2# Now we have a few more questions about issues that may become important in the future. Would you favor or oppose a proposal by former Presidential candidate John McCain to colonize the moon, or haven't you thought much about this?

randomly reverse answer choices

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this #do not reverse

#justification treatment

6. ASK IF Now we have a few more questions about issues that may become important in the future. Would you favor or oppose a proposal by former Presidential candidate John McCain to colonize the moon in order to discover potentially lifesaving minerals not available to doctors on earth, or haven't you thought much about this?

randomly reverse answer choices

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

7. And what about space travel more generally? Would you favor a proposal to increase the budget of the National Aeronautics and Space Administration (NASA)?

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

Experiment 3

Which of the following best describes you...

randomly reverse answer choices

1) Strong Democrat

2) Not Strong Democrat

3) Independent

4) Not Strong Republican

5) Strong Republican

6) Other ##OPEN END

ASK IF Q1=C3 OR Q1=C6 # But if you had to choose, which party would you say you are closer to?

randomly reverse answer choices

1) Closer to Democrats

2) Neither party

3) Closer to Republicans

#Demographic: Education level

8. What is the highest degree or level of school you have completed?

randomly reverse answer choices

1) No high school diploma

2) High school graduate

3) Some college or 2-year college graduate

4) 4-year college graduate

5) Post-graduate degree

randomly select one question from Q4, Q5, Q6, OR Q7 TO ASK

3. Hidden Question, Randomly Select One Choice

1) Control

2) Obama Cue

3) Reasonable argumentation

4) Unreasonable argumentation

#control

4. ASK IF Now we have a few more questions about hypothetical issues that are not on the agenda now but may become important in the future. Would you favor or oppose increasing the length of the school day by an hour for Kindergarten through grade, or haven't you thought much about this?

randomly reverse answer choices

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

#Biden cue

5. ASK IF Now we have a few more questions about hypothetical issues that are not on the agenda now but may become important in the future. Would you favor or oppose a proposal by Vice President Joe Biden to increase the length of the school day by an hour for Kindergarten through grade, or haven't you thought much about this?

randomly reverse answer choices

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

#Reasonable treatment

6. ASK IF #Now we have a few more questions about hypothetical issues that are not on the agenda now but may become important in the future. Would you favor or oppose a proposal by Vice President Joe Biden to increase the length of the school day by an hour for Kindergarten through grade in order to keep American students competitive in the global economy in science and math, or haven't you thought much about this?

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

#Unreasonable/populist treatment

7. ASK IF Now we have a few more questions about hypothetical issues that are not on the agenda now but may become important in the future. Would you favor or oppose a proposal by Vice President Joe Biden to increase the length of the school day by an hour for Kindergarten through grade in order to teach vocational skills in areas America is still competitive like cosmetology and automotive repair, or haven't you thought much about this?

randomly reverse answer choices

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

Ask all: Favorability, must be asked after Q4/Q5/Q6/Q7

7. And what about education more generally? Would you favor a proposal to increase the budget of the Department of Education so the United States can provide better education? #randomly reverse answer choices# 1) Very favorable 2) Somewhat favorable 3) Neither favorable nor unfavorable 4) Somewhat unfavorable 5) Very unfavorable 6) Haven't heard enough to say#do not reverse

Experiment 4

Demographic: Education level

What is the highest degree or level of school you have completed?

randomly reverse answer choices

1. No high school diploma
2. High school graduate

3) Some college or 2-year college graduate

4. 4-year college graduate
5. Post-graduate degree

Which of the following best describes you...

ASK IF Q1=C3 OR Q1=C6 # But if you had to choose, which party would you say you are closer to?

randomly reverse answer choices

1. Closer to Democrats
2. Neither party
3. Closer to Republicans

Here is the experiment: respondents are randomized into 1 of 4 Qs

in other words, randomly select one question from G2, G3, G4, OR G5 TO ASK

G1 HIDDEN QUESTION, RANDOMLY SELECT ONE CHOICE:

1) G2 Control

2. G3 Partisan Cue
3. G4 Reasonable argumentation
4. G5 Specious argumentation

#control

G2. #ASK IF Now we have a few more questions about hypothetical issues that are not on the agenda now but may become important in the future. Would you favor or oppose a proposal to raise the Social Security contribution rate for high income individuals, or haven't you thought much about this?

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

#Obama cue

G3. #ASK IF G1=C2# Now we have a few more questions about hypothetical issues that are not on the agenda now but may become important in the future. Would you favor or oppose a proposal by President Barack Obama to raise the Social Security contribution rate for high income individuals, or haven't you thought much about this?

randomly reverse answer choices

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

Dignity argumentation treatment

G4. #ASK IF G1=C3#Now we have a few more questions about hypothetical issues that are not on the agenda now but may become important in the future. Would you favor or oppose a proposal by President Barack Obama to raise the Social Security contribution rate for high income individuals because all seniors should be able to retire with dignity, not just a privileged few, or haven't you thought much about this?

randomly reverse answer choices

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

Specious argumentation treatment

G5. #ASK IF Now we have a few more questions about hypothetical issues that are not on the agenda now but may become important in the future. Would you favor or oppose a proposal by President Barack Obama to raise the Social Security contribution rate for high income individuals because all seniors want to go on more vacations by the beach, or haven't you thought much about this?

1) Strongly favor

2) Favor

3) Neither favor nor oppose

4) Oppose

5) Strongly oppose

6) Haven't thought much about this#do not reverse

In this section, I present means and standard deviations in parenthesis for each covariate by treatment assignment (balance table). I also use a multinomial logit regression where the dependent variable is treatment assignment to show that demographic covariates do not predict treatment assignment. For each mlogit, I conducted an F-test, which rejects that the covariates jointly predict treatment assignment. F-test results are listed at the bottom of each balance table. These analyses show that covariates are balanced, and therefore, the randomizations were successful. Note that education level was only asked in experiments 3 and 4.

Table a1: Balance Table with F-test p-value for Experiment 1

Variable	Control	Cue	Justification
Weak Democrat	0.1889	0.1978	0.1812
	[.392]	[.3711]	[.3858]
Weak Democrat	0.1548	0.1643	0.1469
	[.3623]	[.3989]	[.3545]
Leaning Democrat	0.0774	0.0613	0.0781
	[.2676]	[.2402]	[.2688]
Pure Independents	0.2291	0.2312	0.2594
	[.4209]	[.4222]	[.439]
Leaning Republican	0.0867	0.0724	0.0563
	[.2818]	[.2595]	[.2308]
Weak Republican	0.1053	0.1448	0.1563
	[.3074]	[.3524]	[.3637]
Region: Northwest	0.1734	0.1671	0.1938
	[.3792]	[.3736]	[.3959]
Region: Midwest	0.2477	0.2117	0.2219
	[.4323]	[.4091]	[.4162]
Region: West	0.2198	0.2451	0.2313
	[.4148]	[.4308]	[.4223]
Age: 18 to 24	0.0991	0.0891	0.1094
	[.2992]	[.2853]	[.3126]
Age: 25 to 34	0.2105	0.1588	0.1938
	[.4083]	[.366]	[.3959]
Age: 35 to 49	0.2972	0.3203	0.2844
	[.4577]	[.4673]	[.4518]
Age: 50 to 64	0.2693	0.2758	0.2844
	[.4443]	[.4475]	[.4518]
Gender: Female	0.4954	0.4986	0.5
	[.5008]	[.5007]	[.5008]
African American	0.0743	0.0501	0.05
	[.2627]	[.2185]	[.2183]
Hispanic	0.0805	0.0585	0.0969
	[.2725]	[.235]	[.2963]
Other race	0.0805	0.0641	0.0531
	[.2725]	[.2452]	[.2246]
Observations	323	359	320
F test p value: 0.85

Table a2: Balance Table with F-test p-value for Experiment 2

Variable	Control	Cue	Justification
Weak Democrat	0.2378	0.1774	0.2215
	[.4264]	[.3428]	[.3809]
Weak Democrat	0.1524	0.1355	0.1754
	[.36]	[.3826]	[.4159]
Leaning Democrat	0.0671	0.0903	0.0646
	[.2505]	[.2871]	[.2462]
Pure Independents	0.1677	0.2258	0.2277
	[.3742]	[.4188]	[.42]
Leaning Republican	0.0915	0.1	0.0646
	[.2887]	[.3005]	[.2462]
Weak Republican	0.1189	0.1516	0.12
	[.3242]	[.3592]	[.3255]
Region: Northwest	0.1341	0.1742	0.2092
	[.3413]	[.3799]	[.4074]
Region: Midwest	0.247	0.2387	0.2185
	[.4319]	[.427]	[.4138]
Region: West	0.2195	0.2161	0.2677
	[.4145]	[.4123]	[.4434]
Age: 18 to 24	0.1189	0.0903	0.0985
	[.3242]	[.2871]	[.2984]
Age: 25 to 34	0.1799	0.1677	0.1877
	[.3847]	[.3742]	[.3911]
Age: 35 to 49	0.2835	0.2968	0.28
	[.4514]	[.4576]	[.4497]
Age: 50 to 64	0.2683	0.2871	0.2892
	[.4437]	[.4531]	[.4541]
Gender: Female	0.4665	0.5065	0.4923
	[.4996]	[.5008]	[.5007]
African American	0.0945	0.0613	0.1015
	[.293]	[.2402]	[.3025]
Hispanic	0.1128	0.0774	0.0862
	[.3168]	[.2677]	[.281]
Other race	0.0518	0.0742	0.0462
	[.222]	[.2625]	[.2101]
Observations	328	310	325
F test p value: 0.23

Table a3: Balance Table with F-test p-value for Experiment 3

Variable	Control	Cue	Reasonable	Specious
Weak Democrat	0.2467	0.1595	0.2127	0.1719
	[.432]	[.3841]	[.4101]	[.378]
Weak Democrat	0.1366	0.179	0.1222	0.1523
	[.3441]	[.3669]	[.3282]	[.3601]
Leaning Democrat	0.0969	0.1051	0.1086	0.1172
	[.2965]	[.3072]	[.3118]	[.3223]
Pure Independents	0.1762	0.1868	0.1855	0.1563
	[.3818]	[.3905]	[.3896]	[.3638]
Leaning Republican	0.0925	0.1012	0.1403	0.1484
	[.2904]	[.3021]	[.3481]	[.3562]
Weak Republican	0.1145	0.1479	0.1357	0.1289
	[.3192]	[.3557]	[.3433]	[.3358]
Region: Northwest	0.1454	0.179	0.1719	0.2344
	[.3533]	[.3841]	[.3782]	[.4244]
Region: Midwest	0.2203	0.2218	0.2262	0.2031
	[.4153]	[.4163]	[.4193]	[.4031]
Region: West	0.2467	0.2062	0.2081	0.2148
	[.432]	[.4054]	[.4069]	[.4115]
Age: 18 to 24	0.1101	0.0778	0.0814	0.1016
	[.3137]	[.2684]	[.2741]	[.3027]
Age: 25 to 34	0.185	0.2101	0.19	0.1563
	[.3892]	[.4082]	[.3932]	[.3638]
Age: 35 to 49	0.2687	0.3074	0.2624	0.2813
	[.4443]	[.4623]	[.441]	[.4505]
Age: 50 to 64	0.304	0.284	0.2896	0.2617
	[.461]	[.4518]	[.4546]	[.4404]
Gender: Female	0.489	0.5136	0.4842	0.5391
	[.501]	[.5008]	[.5009]	[.4994]
High School Diploma	0.2511	0.2296	0.2036	0.2188
	[.4346]	[.4214]	[.4036]	[.4142]
Some College	0.3524	0.358	0.3937	0.3906
	[.4788]	[.4803]	[.4897]	[.4888]
4 year College	0.2819	0.3035	0.2986	0.2734
	[.4509]	[.4607]	[.4587]	[.4466]
Post Graduate	0.0925	0.0973	0.1041	0.1055
	[.2904]	[.2969]	[.306]	[.3078]
African American	0.0881	0.0973	0.095	0.0742
	[.2841]	[.2969]	[.2939]	[.2626]
Hispanic	0.0441	0.0584	0.0814	0.0742
	[.2057]	[.2349]	[.2741]	[.2626]
Other race	0.0661	0.0506	0.0407	0.0586
	[.249]	[.2196]	[.1981]	[.2353]
Observations	227	257	221	256
F test p value: 0.95

Table a4: Balance Table with F-test p-value for Experiment 4

Variable	Control	Cue	Reasonable	Specious
Weak Democrat	0.1297	0.2273	0.1866	0.1653
	[.3367]	[.4199]	[.3606]	[.3722]
Weak Democrat	0.1674	0.1777	0.153	0.1492
	[.3741]	[.383]	[.3903]	[.357]
Leaning Democrat	0.0921	0.0909	0.1045	0.1331
	[.2897]	[.2881]	[.3065]	[.3403]
Pure Independents	0.1506	0.1322	0.153	0.1855
	[.3584]	[.3394]	[.3606]	[.3895]
Leaning Republican	0.113	0.0826	0.0896	0.0927
	[.3172]	[.2759]	[.2861]	[.2907]
Weak Republican	0.2008	0.157	0.1604	0.1331
	[.4015]	[.3646]	[.3677]	[.3403]
Region: Northwest	0.1381	0.2025	0.1866	0.1976
	[.3457]	[.4027]	[.3903]	[.399]
Region: Midwest	0.2301	0.2397	0.2463	0.2218
	[.4218]	[.4278]	[.4316]	[.4163]
Region: West	0.2343	0.1983	0.2127	0.2419
	[.4245]	[.3996]	[.41]	[.4291]
Age: 18 to 24	0.0669	0.0826	0.0858	0.1008
	[.2505]	[.2759]	[.2806]	[.3017]
Age: 25 to 34	0.1883	0.1942	0.1642	0.2137
	[.3918]	[.3964]	[.3711]	[.4108]
Age: 35 to 49	0.2803	0.2603	0.3545	0.2702
	[.4501]	[.4397]	[.4792]	[.4449]
Age: 50 to 64	0.2845	0.2975	0.2649	0.2661
	[.4521]	[.4581]	[.4421]	[.4428]
Gender: Female	0.4812	0.5289	0.5149	0.5282
	[.5007]	[.5002]	[.5007]	[.5002]
High School Diploma	0.2218	0.1529	0.2052	0.2419
	[.4163]	[.3606]	[.4046]	[.4291]
Some College	0.3808	0.3595	0.3769	0.3347
	[.4866]	[.4808]	[.4855]	[.4728]
4 year College	0.2887	0.3512	0.2948	0.2742
	[.4541]	[.4783]	[.4568]	[.447]
Post Graduate	0.1046	0.1116	0.1119	0.125
	[.3067]	[.3155]	[.3159]	[.3314]

Full Regression Results for Main Results

African American	0.1088	0.0826	0.0485	0.0806
	[.312]	[.2759]	[.2152]	[.2728]
Hispanic	0.0628	0.0785	0.056	0.0927
	[.243]	[.2695]	[.2303]	[.2907]
Other race	0.0628	0.0537	0.0634	0.0605
	[.243]	[.2259]	[.2442]	[.2389]
Observations	239	242	268	248
F test p value: 0.44

Table A5: Experiment 1 OLS Regression Tables for Policy and Agency Support

	(1)	(2)
VARIABLES	Policy Support	Agency Support
Partisan Cue Treatment	-0.0632***	-0.0279
	[0.0175]	[0.0213]
Justification Treatment	-0.0116	-0.0267
	[0.0180]	[0.0219]
Strong Democrat	0.135***	0.0664**
	[0.0260]	[0.0317]
Weak Democrat	0.102***	0.0245
	[0.0270]	[0.0329]
Leaning Democrat	0.0934***	-0.00814
	[0.0335]	[0.0407]
Pure Independent	0.0857***	-0.0318
	[0.0247]	[0.0300]
Lean Republican	0.0658**	0.0295
	[0.0332]	[0.0405]
Weak Republican	-0.000143	-0.0277
	[0.0277]	[0.0337]
Region: Northwest	-0.0394*	-0.0175
	[0.0210]	[0.0255]
Region: Midwest	-0.0195	-0.0227
	[0.0196]	[0.0238]
Region: West	-0.0284	0.0141
	[0.0192]	[0.0234]
Age: 18 to 24	0.0656**	0.0628*
	[0.0310]	[0.0378]
Age: 25 to 34	0.0759***	0.103***
	[0.0265]	[0.0322]
Age: 35 to 49	0.0876***	0.0402
	[0.0238]	[0.0290]
Age: 50 to 64	0.0385	0.0259
	[0.0239]	[0.0291]
Gender: Female	-0.0396***	-0.108***
	[0.0145]	[0.0176]
Race: African American	-0.00194	-0.000753
	[0.0322]	[0.0392]
Race: Hispanic	0.0332	0.0595*
	[0.0283]	[0.0344]
Race: Other	-0.00382	0.0126
	[0.0301]	[0.0366]
Constant	0.402***	0.573***
	[0.0301]	[0.0366]
Observations	1,002	1,002
R-squared	0.090	0.075
Standard errors in brackets
*** p<0.01, ** p<0.05, * p<0.1

Table a6: Experiment 2 OLS Regression Tables for Policy and Agency Support

	(1)	(2)
VARIABLES	Policy Support	Agency Support
Partisan Cue Treatment	-0.0521***	-0.0283
	[0.0186]	[0.0232]
Justification Treatment	-0.00138	-0.00802
	[0.0184]	[0.0230]
Strong Democrat	-0.0228	0.0497
	[0.0270]	[0.0336]
Weak Democrat	0.0155	0.0496
	[0.0283]	[0.0352]
Leaning Democrat	-0.0783**	0.00764
	[0.0347]	[0.0432]
Pure Independent	-0.0285	-0.00715
	[0.0267]	[0.0333]
Lean Republican	-0.0241	0.00819
	[0.0332]	[0.0414]
Weak Republican	-0.0441	-0.00700
	[0.0292]	[0.0364]
Region: Northwest	0.00503	-0.0684**
	[0.0223]	[0.0277]
Region: Midwest	-0.0319	-0.0351
	[0.0202]	[0.0252]
Region: West	-0.0219	-0.00619
	[0.0203]	[0.0253]
Age: 18 to 24	0.0441	0.0933**
	[0.0317]	[0.0395]
Age: 25 to 34	0.0903***	0.135***
	[0.0272]	[0.0339]
Age: 35 to 49	0.0417*	0.0827***
	[0.0245]	[0.0306]
Age: 50 to 64	-0.0157	0.0444
	[0.0242]	[0.0302]
Gender: Female	-0.0297**	-0.0690***
	[0.0151]	[0.0188]
Race: African American	-0.0553*	-0.0521
	[0.0288]	[0.0359]
Race: Hispanic	-0.000674	-0.0186
	[0.0275]	[0.0342]
Race: Other	-0.0260	-0.0232
	[0.0336]	[0.0419]
Constant	0.532***	0.562***
	[0.0299]	[0.0372]
Observations	963	963
R-squared	0.058	0.051
Standard errors in brackets
*** p<0.01, ** p<0.05, * p<0.1

Table a7: Experiment 3 OLS Regression Tables for Policy and Agency Support

	(1)	(2)
VARIABLES	Policy Support	Agency Support
Partisan Cue Treatment	0.0110	-0.0235
	[0.0267]	[0.0250]
Justification Treatment	0.0469*	-0.0468*
	[0.0278]	[0.0261]
Strong Democrat	0.0760***	-0.0257
	[0.0269]	[0.0252]
Weak Democrat	0.0759**	0.517***
	[0.0353]	[0.0332]
Leaning Democrat	0.0637*	0.410***
	[0.0370]	[0.0348]
Pure Independent	0.0205	0.421***

	[0.0398]	[0.0374]
Lean Republican	0.0156	0.156***
	[0.0355]	[0.0334]
Weak Republican	-0.0148	-0.0400
	[0.0386]	[0.0363]
Region: Northwest	-0.0759**	-0.0802**
	[0.0378]	[0.0355]
Region: Midwest	0.00561	0.00866
	[0.0273]	[0.0256]
Region: West	0.0178	0.0296
	[0.0257]	[0.0241]
Age: 18 to 24	0.00441	-0.0120
	[0.0261]	[0.0245]
Age: 25 to 34	-0.161***	0.0340
	[0.0404]	[0.0379]
Age: 35 to 49	-0.123***	-0.00409
	[0.0333]	[0.0313]
Age: 50 to 64	-0.0836***	0.0168
	[0.0302]	[0.0284]
Gender: Female	-0.0173	0.0304
	[0.0300]	[0.0282]
Race: African American	0.0515	0.140***
	[0.0360]	[0.0338]
Race: Hispanic	0.0497	0.144***
	[0.0399]	[0.0374]
Race: Other	0.0969**	0.134***
	[0.0432]	[0.0406]
Education: High School	0.0908	-0.0296
	[0.0912]	[0.0857]
Education: Some College	0.132	-0.0156
	[0.0904]	[0.0850]
Education: 4-year College	0.131	0.00713
	[0.0910]	[0.0855]
Education: Post-graduate	0.149	0.0367
	[0.0943]	[0.0885]
Constant	0.390***	0.160*
	[0.0961]	[0.0903]
Observations	961	961
R-squared	0.082	0.467
Standard errors in brackets
*** p<0.01, ** p<0.05, * p<0.1

Table a8: Experiment 4 OLS Regression Tables for Policy

	(1)
VARIABLES	Policy Support
Partisan Cue Treatment	-0.0506**
	[0.0243]
Justification Treatment	0.0266
	[0.0237]
Strong Democrat	-0.144***
	[0.0241]
Weak Democrat	0.164***
	[0.0311]
Leaning Democrat	0.0813***
	[0.0309]
Pure Independent	0.158***
	[0.0341]
Lean Republican	0.00451
	[0.0309]
Weak Republican	-0.0548
	[0.0352]
Region: Northwest	-0.116***
	[0.0303]
Region: Midwest	-0.0117
	[0.0244]
Region: West	-0.00359
	[0.0223]
Age: 18 to 24	-0.0344
	[0.0230]
Age: 25 to 34	-0.159***
	[0.0367]
Age: 35 to 49	-0.0284
	[0.0293]
Age: 50 to 64	-0.0676**
	[0.0268]
Gender: Female	0.00892
	[0.0266]
Race: African American	-0.0818**
	[0.0334]
Race: Hispanic	-0.0245
	[0.0335]
Race: Other	-0.00479
	[0.0364]
Education: High School	-0.0643
	[0.0690]
Education: Some College	-0.0153
	[0.0678]
Education: 4-year College	-0.0140
	[0.0683]
Education: Post-graduate	-0.0782
	[0.0712]
Constant	0.704***
	[0.0751]
Observations	997
R-squared	0.193
Standard errors in brackets
*** p<0.01, ** p<0.05, * p<0.1

Treatment Effect Regression Models by Party Identification

Table a9: Experiment 1 OLS Regression Tables for Policy Support by Party

	(1)	(2)	(3)	(4)	(5)
VARIABLES	Strong Dems	Weak Dems	Ind	Weak Rep	Strong Rep
Partisan Cue Treatment	0.0448	-0.0206	-0.0805***	-0.125**	-0.137***
	[0.0407]	[0.0372]	[0.0296]	[0.0538]	[0.0476]
Justification Treatment	0.109**	0.00261	-0.0222	-0.00218	-0.114**
	[0.0431]	[0.0404]	[0.0298]	[0.0540]	[0.0505]
Region: Northwest	-0.0701	-0.0125	-0.0378	-0.0807	-0.0686
	[0.0521]	[0.0407]	[0.0367]	[0.0636]	[0.0567]
Region: Midwest	-0.0800*	-0.0182	-0.0334	0.0914*	-0.0427
	[0.0456]	[0.0468]	[0.0327]	[0.0534]	[0.0556]
Region: West	-0.0974**	0.0287	-0.0212	-0.102*	0.0404
	[0.0458]	[0.0417]	[0.0322]	[0.0557]	[0.0531]
Age: 18 to 24	0.0650	-0.0600	0.0621	0.153	0.149
	[0.0804]	[0.0676]	[0.0486]	[0.0938]	[0.109]
Age: 25 to 34	0.0941	-0.0855	0.132***	0.0623	0.0803
	[0.0661]	[0.0558]	[0.0438]	[0.0757]	[0.0754]
Age: 35 to 49	0.0726	-0.00221	0.0819**	0.180***	0.100
	[0.0597]	[0.0520]	[0.0399]	[0.0655]	[0.0629]
Age: 50 to 64	0.0359	-0.0123	1.45e-05	0.0940	0.0611
	[0.0593]	[0.0489]	[0.0421]	[0.0666]	[0.0630]
Gender: Female	0.0115	-0.0701**	-0.0670***	-0.0322	-0.0357
	[0.0348]	[0.0309]	[0.0243]	[0.0442]	[0.0406]
Race: African American	-0.0402	0.0343	0.0587	-	-0.116
	[0.0520]	[0.0597]	[0.0589]	-	[0.241]
Race: Hispanic	0.0662	0.00803	0.0524	-0.0480	0.136
	[0.0581]	[0.0504]	[0.0466]	[0.103]	[0.136]
Race: Other	0.0553	0.0270	-0.0182	0.126	-0.184**
	[0.0896]	[0.0583]	[0.0466]	[0.0893]	[0.0880]
Constant	0.473***	0.558***	0.508***	0.359***	0.442***
	[0.0619]	[0.0541]	[0.0427]	[0.0749]	[0.0623]
Observations	190	156	364	136	136
R-squared	0.096	0.071	0.104	0.197	0.158
Standard errors in brackets
*** p<0.01, ** p<0.05, * p<0.1

Table A10: Experiment 1 OLS Regression Tables for Agency Support by Party

	(1)	(2)	(3)	(4)	(5)
VARIABLES	Strong Dems	Weak Dems	Ind	Weak Rep	Strong Rep
Partisan Cue Treatment	0.00137	-0.0360	-0.0187	-0.130*	0.0162
	[0.0490]	[0.0558]	[0.0331]	[0.0667]	[0.0650]
Justification Treatment	0.0511	0.0350	-0.0152	-0.0713	-0.104
	[0.0519]	[0.0606]	[0.0333]	[0.0669]	[0.0689]
Region: Northwest	-0.0297	-0.0315	-0.0249	-0.0415	0.00541
	[0.0627]	[0.0611]	[0.0410]	[0.0789]	[0.0773]
Region: Midwest	-0.0720	-0.116	-0.0138	0.0800	-0.0111
	[0.0549]	[0.0701]	[0.0366]	[0.0662]	[0.0759]
Region: West	0.0268	-0.0509	-0.0151	0.0545	0.0605
	[0.0552]	[0.0625]	[0.0360]	[0.0690]	[0.0725]
Age: 18 to 24	0.0857	0.0408	0.0481	0.0837	0.0703
	[0.0968]	[0.101]	[0.0544]	[0.116]	[0.149]
Age: 25 to 34	0.171**	0.167**	0.0881*	-0.00617	-0.000642
	[0.0796]	[0.0837]	[0.0490]	[0.0939]	[0.103]
Age: 35 to 49	0.0561	0.0535	0.0317	0.0602	-0.00851
	[0.0719]	[0.0779]	[0.0446]	[0.0812]	[0.0859]
Age: 50 to 64	0.124*	-0.00888	-0.00873	0.0245	0.00388
	[0.0714]	[0.0734]	[0.0471]	[0.0825]	[0.0860]
Gender: Female	-0.0997**	-0.0694	-0.148***	-0.0609	-0.0978*
	[0.0419]	[0.0463]	[0.0272]	[0.0549]	[0.0555]
Race: African American	-0.0920	-0.0296	0.0636	-	0.264
	[0.0626]	[0.0896]	[0.0659]	-	[0.328]
Race: Hispanic	0.0901	0.0723	0.0361	0.0105	0.111
	[0.0700]	[0.0756]	[0.0522]	[0.128]	[0.185]
Race: Other	0.0227	0.0741	0.0517	-0.0527	-0.106
	[0.108]	[0.0874]	[0.0521]	[0.111]	[0.120]
Constant	0.570***	0.588***	0.591***	0.561***	0.601***
	[0.0745]	[0.0811]	[0.0478]	[0.0928]	[0.0849]
Observations	190	156	364	136	136
R-squared	0.124	0.127	0.105	0.071	0.076
Standard errors in brackets
*** p<0.01, ** p<0.05, * p<0.1

Table a11: Experiment 2 OLS Regression Tables for Policy Support by Party

	(1)	(2)	(3)	(4)	(5)
VARIABLES	Strong Dems	Weak Dems	Ind	Weak Rep	Strong Rep
Partisan Cue Treatment	-0.103**	-0.0603	-0.0560*	-0.00581	-0.0165
	[0.0433]	[0.0453]	[0.0318]	[0.0517]	[0.0506]
Justification Treatment	0.0335	-0.0218	-0.0253	0.0319	0.0531
	[0.0396]	[0.0422]	[0.0326]	[0.0551]	[0.0529]
Region: Northwest	0.0968*	0.00943	-0.0476	-0.0502	-0.0935
	[0.0523]	[0.0528]	[0.0376]	[0.0660]	[0.0661]
Region: Midwest	0.00403	-0.0190	-0.0353	-0.0430	-0.0997*
	[0.0455]	[0.0489]	[0.0353]	[0.0535]	[0.0536]
Region: West	-0.00533	-0.0363	0.000355	-0.118*	-0.0229
	[0.0475]	[0.0497]	[0.0339]	[0.0626]	[0.0532]
Age: 18 to 24	0.0740	0.0955	-0.00833	0.0548	0.145
	[0.0713]	[0.0765]	[0.0553]	[0.107]	[0.0893]
Age: 25 to 34	0.126**	0.0992	0.102**	0.0675	0.0176
	[0.0625]	[0.0695]	[0.0475]	[0.0755]	[0.0760]
Age: 35 to 49	-0.00638	0.123*	0.0770*	-0.00708	0.0824
	[0.0572]	[0.0628]	[0.0462]	[0.0590]	[0.0566]
Age: 50 to 64	-0.111**	-0.00872	0.0566	-0.0210	0.0391
	[0.0561]	[0.0627]	[0.0463]	[0.0575]	[0.0554]
Gender: Female	-0.0616*	-0.0470	-0.00207	-0.0490	-0.0167
	[0.0340]	[0.0363]	[0.0258]	[0.0425]	[0.0418]
Race: African American	0.00830	-0.0482	-0.160***	0.107	-0.231
	[0.0432]	[0.0625]	[0.0603]	[0.173]	[0.173]
Race: Hispanic	0.0521	0.0186	-0.0332	-0.0593	-0.0177
	[0.0548]	[0.0574]	[0.0477]	[0.104]	[0.0923]
Race: Other	0.0236	-0.00882	-0.0657	0.0251	0.0754
	[0.0968]	[0.0802]	[0.0489]	[0.102]	[0.121]
Constant	0.510***	0.527***	0.482***	0.518***	0.506***
	[0.0594]	[0.0694]	[0.0489]	[0.0615]	[0.0547]
Observations	205	149	338	125	132
R-squared	0.218	0.109	0.059	0.075	0.083
Standard errors in brackets
*** p<0.01, ** p<0.05, * p<0.1

Table a12: Experiment 2 OLS Regression Tables for Agency Support by Party

	(1)	(2)	(3)	(4)	(5)
VARIABLES	Strong Dems	Weak Dems	Ind	Weak Rep	Strong Rep
Partisan Cue Treatment	-0.0924*	-0.0454	-0.00897	-0.133**	0.0820
	[0.0550]	[0.0596]	[0.0379]	[0.0645]	[0.0689]
Justification Treatment	0.0399	0.000615	-0.0309	-0.0134	0.0110
	[0.0502]	[0.0555]	[0.0389]	[0.0688]	[0.0720]
Region: Northwest	-0.122*	0.0637	-0.0763*	-0.213**	-0.0741
	[0.0664]	[0.0694]	[0.0449]	[0.0823]	[0.0899]
Region: Midwest	-0.0907	-0.00343	0.00804	-0.117*	-0.0444
	[0.0578]	[0.0643]	[0.0421]	[0.0668]	[0.0730]
Region: West	0.0308	0.0128	-0.0164	-0.0843	0.0218
	[0.0603]	[0.0653]	[0.0405]	[0.0781]	[0.0724]
Age: 18 to 24	0.175*	0.250**	0.00625	0.178	0.120
	[0.0906]	[0.101]	[0.0660]	[0.134]	[0.122]
Age: 25 to 34	0.183**	0.272***	0.0705	0.223**	0.0868
	[0.0793]	[0.0913]	[0.0568]	[0.0942]	[0.103]
Age: 35 to 49	0.139*	0.232***	0.0278	0.0671	0.0542
	[0.0727]	[0.0826]	[0.0551]	[0.0736]	[0.0771]
Age: 50 to 64	-0.00678	0.217***	0.0246	0.0480	0.0509
	[0.0713]	[0.0824]	[0.0553]	[0.0717]	[0.0754]
Gender: Female	-0.0923**	-0.0206	-0.0663**	-0.0872	-0.0354
	[0.0432]	[0.0477]	[0.0307]	[0.0530]	[0.0568]
Race: African American	-0.0596	-0.0398	-0.0623	0.243	-0.311
	[0.0549]	[0.0821]	[0.0720]	[0.216]	[0.236]
Race: Hispanic	0.0587	-0.0491	-0.0393	-0.0193	-0.0494
	[0.0696]	[0.0755]	[0.0570]	[0.129]	[0.126]
Race: Other	-0.109	-0.0793	-0.0168	-0.0642	0.0729
	[0.123]	[0.105]	[0.0583]	[0.128]	[0.164]
Constant	0.615***	0.424***	0.602***	0.653***	0.520***
	[0.0755]	[0.0912]	[0.0583]	[0.0768]	[0.0745]
Observations	205	149	338	125	132
R-squared	0.150	0.089	0.038	0.193	0.060
Standard errors in brackets
*** p<0 01, ** p<0.05, *p<0.1

Table a13: Experiment 3 OLS Regression Tables for Policy Support by Party

	(1)	(2)	(3)	(4)	(5)
VARIABLES	Strong Dems	Weak Dems	Ind	Weak Rep	Strong Rep
Partisan Cue Treatment	0.0993[0.0633]	-0.0329[0.0692]	-0.0272[0.0431]	-0.00171[0.0758]	-0.00394[0.0714]
Justification Treatment	0.136**[0.0653]	0.0194[0.0777]	0.0662[0.0439]	-0.111[0.0768]	0.0735[0.0770]
Region: Northwest	0.203***[0.0662]	-0.00543[0.0720]	0.0789*[0.0428]	-0.0162[0.0748]	0.101[0.0708]
Region: Midwest	0.0115[0.0712]	0.0530[0.0689]	0.00317[0.0417]	0.0176[0.0832]	-0.0198[0.0737]
Region: West	0.0605[0.0626]	0.0505[0.0727]	0.0118[0.0399]	-0.00209[0.0650]	0.0416[0.0689]
Age: 18 to 24	0.0608[0.0644]	0.0820[0.0683]	-0.0194[0.0413]	0.0393[0.0789]	-0.136*[0.0695]
Age: 25 to 34	-0.0675[0.107]	-0.438***[0.112]	-0.113*[0.0594]	-0.0225[0.139]	-0.220*[0.124]
Age: 35 to 49	-0.233***[0.0802]	-0.224**[0.109]	-0.0493[0.0527]	-0.106[0.0823]	-0.0742[0.0916]
Age: 50 to 64	-0.00435[0.0774]	-0.240**[0.0937]	-0.0308[0.0475]	-0.125[0.0757]	-0.184**[0.0778]
Gender: Female	-0.0964[0.0724]	-0.134[0.0898]	0.0218[0.0486]	0.0748[0.0699]	0.0336[0.0827]
Race: African American	0.0220[0.0470]	-0.0160[0.0539]	-0.0507*[0.0302]	0.0375[0.0536]	-0.0562[0.0526]
Race: Hispanic	0.150[0.176]	0.286[0.223]	0.0227[0.149]	-0.0802[0.296]
Race: Other	0.207[0.172]	0.350[0.221]	-0.0301[0.149]	0.0124[0.294]	0.226***[0.0707]
Education: High School	0.208	0.367	-0.0347	-0.00728	0.239***
	[0.174]	[0.226]	[0.150]	[0.293]	[0.0724]
Education: Some College	0.225	0.314	0.0502	-0.0236	0.322***
	[0.181]	[0.234]	[0.154]	[0.304]	[0.112]
Education: 4-year College	0.0702	0.226	0.0486	-0.0789	0.0236
	[0.0615]	[0.142]	[0.0582]	[0.164]	[0.159]
Education: Post-graduate	-0.00413	0.0802	0.125*	-0.165	0.0929
	[0.0832]	[0.0976]	[0.0646]	[0.154]	[0.129]
Constant	0.111	0.0556	0.0551	0.118	0.324***
	[0.122]	[0.105]	[0.0648]	[0.147]	[0.119]
Observations	193	138	388	127	115
R-squared	0.148	0.202	0.066	0.150	0.324
Standard errors in brackets
*** p<0.01, ** p<0.05, * p<0.1

Table a14: Experiment 3 OLS Regression Tables for Agency Support by Party

	(1)	(2)	(3)	(4)	(5)
VARIABLES	Strong Dems	Weak Dems	Ind	Weak Rep	Strong Rep
Partisan Cue Treatment	-0.0225	-0.0112	-0.0262	0.0167	-0.120*
	[0.0608]	[0.0727]	[0.0469]	[0.0447]	[0.0687]
Justification Treatment	-0.0773	-0.157*	-0.000243	-0.0501	-0.0707
	[0.0626]	[0.0817]	[0.0477]	[0.0453]	[0.0740]
Region: Northwest	0.0789	-0.0818	-0.0297	-0.0361	-0.0341
	[0.0636]	[0.0757]	[0.0465]	[0.0441]	[0.0681]
Region: Midwest	-0.157**	0.0114	0.0919**	0.0287	0.0132
	[0.0683]	[0.0725]	[0.0453]	[0.0491]	[0.0708]
Region: West	-0.0711	0.169**	0.0627	0.0302	0.00858
	[0.0601]	[0.0764]	[0.0433]	[0.0384]	[0.0662]
Age: 18 to 24	-0.0769	0.0240	-0.00866	0.0308	-0.0335
	[0.0618]	[0.0718]	[0.0449]	[0.0466]	[0.0668]
Age: 25 to 34	-0.0754	0.0857	0.138**	0.0882	-0.0948
	[0.103]	[0.117]	[0.0645]	[0.0819]	[0.119]
Age: 35 to 49	-0.0769	-0.0427	0.0192	0.177***	0.0246
	[0.0770]	[0.114]	[0.0572]	[0.0486]	[0.0880]
Age: 50 to 64	-0.00781	0.0280	0.0288	0.0923**	0.0342
	[0.0743]	[0.0986]	[0.0516]	[0.0447]	[0.0748]
Gender: Female	-0.0375	0.0886	0.125**	0.0560	-0.0324
	[0.0694]	[0.0944]	[0.0528]	[0.0412]	[0.0795]
Race: African American	0.0680	-0.122**	0.0588*	-0.0260	-0.0130
	[0.0451]	[0.0567]	[0.0328]	[0.0317]	[0.0505]
Race: Hispanic	0.322*	-0.0933	-0.163	-0.220	-
	[0.169]	[0.235]	[0.162]	[0.175]	-
Race: Other	0.361**	-0.0410	-0.163	-0.266	0.0969
	[0.165]	[0.232]	[0.161]	[0.173]	[0.0680]
Education: High School	0.352**	0.115	-0.166	-0.199	0.0209
	[0.167]	[0.238]	[0.163]	[0.173]	[0.0697]
Education: Some College	0.444**	-0.132	-0.0696	-0.206	0.230**
	[0.174]	[0.246]	[0.168]	[0.179]	[0.108]
Education: 4-year College	-0.0109	0.250*	0.296***	0.479***	0.453***
	[0.0590]	[0.149]	[0.0632]	[0.0967]	[0.153]
Education: Post-graduate	0.181**	0.103	0.106	0.371***	0.0755
	[0.0799]	[0.103]	[0.0702]	[0.0908]	[0.124]
Constant	0.117	0.209*	0.181**	0.171*	0.0787
	[0.117]	[0.110]	[0.0704]	[0.0867]	[0.114]
Observations	193	138	388	127	115
R-squared	0.135	0.231	0.128	0.446	0.189
Standard errors in brackets
*** p<0.01, ** p<0.05, *p<0.1

Table A16: Experiment 4 OLS Regression Tables for Messenger Support by Party

	(1)	(2)	(3)	(4)	(5)
VARIABLES	Strong Dems	Weak Dems	Ind	Weak Rep	Strong Rep
Partisan Cue Treatment	0.0180	0.0732	0.181***	0.186***	-0.0279
	[0.0582]	[0.0598]	[0.0484]	[0.0598]	[0.0712]
Justification Treatment	0.0668	0.0524	0.121***	0.0976*	0.115*
	[0.0590]	[0.0609]	[0.0462]	[0.0563]	[0.0689]
Region: Northwest	0.0544	0.0368	0.0969**	0.0619	-0.0417
	[0.0658]	[0.0609]	[0.0442]	[0.0610]	[0.0700]
Region: Midwest	0.0451[0.0594]	0.00700[0.0670]	0.0230[0.0457]	0.0489[0.0661]	0.0132[0.0704]
Region: West	-0.0326	0.0569	0.0243	0.0270	-0.0210
	[0.0585]	[0.0556]	[0.0425]	[0.0531]	[0.0755]
Age: 18 to 24	0.107**	0.000934	-0.0504	0.00225	0.00843
	[0.0505]	[0.0619]	[0.0454]	[0.0546]	[0.0768]
Age: 25 to 34	-0.0663	0.0751	-0.0382	-0.0724	-0.0695
	[0.0852]	[0.0997]	[0.0697]	[0.112]	[0.142]
Age: 35 to 49	-0.0967	0.0996	0.0347	0.228***	0.0718
	[0.0677]	[0.0848]	[0.0567]	[0.0766]	[0.0843]
Age: 50 to 64	-0.160**	-0.0207	0.0473	-0.00214	0.105
	[0.0617]	[0.0857]	[0.0523]	[0.0587]	[0.0720]
Gender: Female	-0.0995	0.0565	0.0198	0.0235	0.0455
	[0.0616]	[0.0841]	[0.0496]	[0.0638]	[0.0682]
Race: African American	-0.0662*	-0.0371	0.0121	0.0901**	0.0633
	[0.0400]	[0.0428]	[0.0333]	[0.0440]	[0.0543]
Race: Hispanic	-0.0671	-0.0353	0.0540	0.273	0.0208
	[0.121]	[0.0802]	[0.148]	[0.266]	[0.207]
Race: Other	0.0442	0.0243	0.0846	0.212	-0.0685
	[0.114]	[0.0774]	[0.147]	[0.269]	[0.209]
Education: High School	0.0340	0.0775	0.0724	0.266	-0.0719
	[0.115]	[0.0781]	[0.148]	[0.271]	[0.207]
Education: Some College	-0.0305	-	0.00787	0.288	-0.0265
	[0.122]	-	[0.153]	[0.273]	[0.215]
Education: 4-year College	0.0339	0.146*	0.205***	0.665**	0.410*
	[0.0487]	[0.0753]	[0.0759]	[0.257]	[0.210]
Education: Post-graduate	-0.0753	-0.0188	0.182***	-0.0987	-0.230
	[0.0643]	[0.0766]	[0.0625]	[0.0959]	[0.283]
Constant	0.135	-0.0229	0.0810	-0.118	0.341**
	[0.116]	[0.0714]	[0.0625]	[0.152]	[0.170]
Observations	159	140	316	148	123
R-squared	0.165	0.130	0.115	0.260	0.176
Standard errors in brackets
*** p<0.01, ** p<0.05, * p<0.1