The proposer or the proposal? An experimental analysis of constitutional beliefs

Abstract In many countries, constitutional amendments require the direct approval of voters, but the consequences of fundamental changes to the powers and operations of the state are difficult to anticipate. The referendums literature suggests that citizens weigh their prior beliefs about the merits of proposals against the heuristic provided by the partisanship of the proposer, but the relative salience of these factors across constitutional issue areas remains underexplored. This paper examines the determinants of citizen preferences on 12 diverse constitutional issues, based on a novel survey experiment in Japan. We show that support for amendments is greater when its proposer is described as non-partisan. However, constitutional ideology moderates this effect. Those who prefer idealistic constitutions that elevate national traditions tend to value proposals that expand government powers, compared to those who prefer pragmatic constitutions that constrain government authority. These results highlight the significance of constitutional beliefs that are independent of partisanship.

However, constitutional preferences are not easy to divine. Amendment referenda are single-issue votes that do not influence the makeup of government directly. As such, the factors that drive ballot choice likely differ from those in regular legislative or executive elections, where multidimensional policy preferences are collapsed into one choice.
In this context, the referendums literature has long debated the relative salience of 'issue voting' and 'second-order voting' (Garry et al., 2005;Silagadze and Gherghina, 2018). The former posits that referendum vote choice depends primarily on people's prior views on that specific issue. The latter argues that vote choice mirrors partisan preferences about national politics and is used to reward or punish incumbent governments. Existing work points to the salience of both: while issue voting tends to dominate when voters have strong views (Elkink et al., 2020), partisan cues have a powerful impact on less-informed voters (Hobolt, 2007).
This paper extends this debate to constitutional amendments. Citizens in many countries have some passing knowledge of their constitution's history and contents via primary and secondary school civics courses. However, evaluating the consequences of constitutional change is trickier, since formal amendments often entail downstream changes to the interpretation and execution of a wide body of statutory law. This complexity is even greater when the amendment text is brief, granting governments and constitutional courts more discretion over its interpretation (Manfredi and Lusztig, 1998). As a result, people's choices are likely to be informed both by their prior views about the status quo constitution and by external cues that serve as informational shortcuts on the purpose and merits of constitutional change. The latter gives room for elites to influence voter choices through information provision (Lupia, 1994) and issue framing (Hobolt and Brouard, 2010).
With this in mind, this study evaluates how the identity of the proposer can alter citizens' perceptions of the merits of amendment proposals. The first dimension speaks to partisanship: people may support amendments because these are proposed or backed by their preferred parties (the proposer). This would be a reasonable heuristic, insofar as elected politicians are unlikely to push for reforms that are detrimental to their backers. The effects of partisanship and polarization on assessments of public policy is well-documented (Slothuus and de Vreese, 2010;Bullock, 2011;Druckman et al., 2013), particularly when the issue is less familiar to voters (Baum and Groeling, 2009).
The second dimension is constitutional philosophy: voters may have prior beliefs about what topics should be included in an ideal constitution (the proposal), independent of their affinity for the proposer. Some may prefer pragmatic constitutions that focus on institutions and rights, while others may favor idealistic constitutions that incorporate national principles and symbols. To our knowledge, this dimension is rarely asked in existing surveys, but the literature notes that constitutions with greater legitimacy should be more successful in constraining government overreach (Weingast, 1997) and endure over a longer period of time (Elkins et al., 2009). To the extent that citizens vary in their notions of an ideal constitution, these beliefs should influence assessments of different amendment proposals. This paper examines the relative salience of the 'proposer' across different amendment 'proposals' through a unique survey experiment on constitutional attitudes in Japan. The experiment tests how respondents' evaluations of 12 hypothetical amendments change when their proposer is described as the governing 'Liberal Democratic Party' (LDP) or as a 'neutral panel of constitutional experts'. We find that support for many amendment topics rises by 3−9 percentage points when respondents are told they are proposed by a neutral expert panel, as opposed to the LDP. The effects are stronger for proposals that are less obviously conservative, such as the addition of new human rights, where the informational value of the proposer label is greater.
We also find two important subgroup differences in treatment effects. First, self-identified progressives and independentswho generally oppose the LDPare more likely to be swayed by 'neutral' proposals than conservative respondents. Second, differences in constitutional ideology moderate the treatment effect, particularly on conservative topics such as amending the Japanese constitution's pacifist principle and expanding national emergency powers. Constitutional idealists who want to restore Japan's traditions respond more positively to the treatment than pragmatists who focus more on human rights and political institutions. These results speak to the nuanced role that partisanship and ideology play in shaping constitutional preferences and have ramifications for our broader understanding of constitutional politics.
We should note that inferences drawn from one country cannot be automatically transported elsewhere. People's familiarity with and opinions on amendments are inevitably colored by what is already in their constitution, as well as the nature of the partisan debate over the constitution's strengths and flaws. That said, we believe that Japan is a valuable case for three reasons. First, constitutional revision is intimately tied to the partisan competition. Established in 1947, the Constitution of Japan (COJ) is the oldest unamended constitution in the world today, despite its contentious post-WWII origins during the Allied Occupation. Conservatives have accused the COJ of being too antiquated to deal with emerging domestic and international problems. Progressives, by contrast, have decried most revision proposals as veiled attempts to revive the autocratic pre-war (Meiji) constitution.
Second, the salience of constitutional politics has risen since the late 1990s, largely in response to post-Cold War geopolitical challenges and public consternation with economic and political malaise (Winkler, 2011;Sakaiya, 2017). The LDP published new constitutional proposals in 2005 and 2012, and the issue has been given pride of place in its recent election manifestos. While some of the LDP's proposed amendments are specific to the Japanese status quo, most seek to expand the range of enumerated human rights and political institutions. This comports with the global trend towards greater constitutional detail over time, suggesting that this paper's analysis should be of relevance to other national contexts (Versteeg and Zackin, 2016;Abramson and Barber, 2019).
Third, amendments to the COJ require majority approval in a national referendum, and opinion polls suggest that public sentiment is both divided and volatile. Japanlike many developed democracieshas a large share of independent voters, making it a valuable case for testing how (non-) partisanship shapes public attitudes. At the same time, the origins and core principles of the COJ are taught to all children in civics and history classes, giving ample opportunity for preference formation on dimensions other than partisanship.
In sum, we believe that Japan is a reasonable case to test how constitutional beliefs and political partisanship interact in the formation of constitutional preferences. Later sections will review the COJ's history and contents in greater detail, and the conclusion includes a longer discussion of the transportability of our findings to other countries.

Visions of constitutional change
The motivations for constitutional change are diverse (cf. Rasch and Congleton, 2006). One is political crises, such as policy paralysis in the face of geopolitical threats or the revelation of systemic government corruption. Another is elite self-interest, such as the manipulation of electoral rules or presidential term limits for partisan gain. A third is gradual changes in social norms concerning gender equality, minority rights, or minimum standards of living (Elkins et al., 2013). Constitutions that fail to allocate political power effectively or that enumerate outmoded values are more likely to face amendment pressures, but the specifics of constitutional politics vary over time and across national contexts. They reflect country-specific debates about initial constitutional design, as well as global trends in constitutional best practices (Law and Versteeg, 2011). With that in mind, we will begin by describing the constitutional discourse in post-WWII Japan.
First, the fact that the COJ has never been amended should not be mistaken for universal acclamation. The General Headquarters (GHQ) of the Allied Occupation, which wrote the draft on which the COJ is based, sought to demilitarize Japan, replace prewar imperial sovereignty with popular democracy, and elevate the primacy of individual and social rights (Hellegers, 2001). Once Japan regained its sovereignty following the end of the Allied Occupation in 1952, conservative elites pushed to revise what they viewed as a radical and deplorable break from Japan's proud history and traditions (Winkler, 2011). This was one of the reasons for the 1955 formation of the LDP, whose early leaders criticized the COJ for 'wrongly suppressing respect for the state and patriotism and unduly fragmenting and weakening the power of the state' (LDP, 1955; authors' translation).
At its core, conservative criticism towards the COJ rests on two pillars: amending Article 9 and inserting more conservative values. The first, and by far the most politicized issue, has been the Peace Clause, which writes: Article 9.1 Aspiring sincerely to an international peace based on justice and order, the Japanese people forever renounce war as a sovereign right of the nation and the threat or use of force as a means of settling international disputes. Article 9.2 In order to accomplish the aim of the preceding paragraph, land, sea, and air forces, as well as other war potential, will never be maintained. The right of belligerency of the state will not be recognized. 2 Conservatives have long criticized Article 9 for its naïve assumptions about world peace, and for the limitations, it has imposed on Japan's Self-Defense Forces (SDF) to act like the military of a 'normal nation' that has the capability to respond to (post-) Cold War threats and meet its international obligations (Winkler, 2011). 3 While the debate surrounding Article 9 has remained the most controversial issue, it has been far from the only target for amendment. The second pillar of conservative criticism relates to the constitution's origins. Conservatives have long held that the COJ's version of liberal democracy amounted to a 'forced import' of American values on a defeated, occupied nation (Winkler, 2012). Despite propagating democracy and promulgating the COJ in the name of 'we, the people', Occupation officers had written the constitutional draft within a matter of a week (Hiranuma, 2005) and censored free debate (Etō, 1994). The result of this 'imposed constitution', critics claim, was a whitewashing of Japanese character. The COJ excluded references to Japan's history and traditions from the preamble, and its emphasis of negative rights made it out of sync with contemporary realities. The COJ had thus contributed to excessive individualism and the atomization of society (e.g., Yomiuri Shimbunsha, 2004;Winkler, 2011).
One proposed solution to these perceived ills was the strengthening of intermediary organizations such as the family and religious organizations. Currently, Article 24 does not include an explicit reference to the family; instead, it stresses the 'essential equality of the sexes' and based thereon the equal rights of married couples, which scholars generally understand as preventing the state from interfering with family matters (Tsujimura, 2008). In its 2012 constitutional proposal, the LDP has suggested adding to Article 24 that, 'The family has to be respected as the natural and fundamental unit of society. Family members have to support one another.' 4 On the religious dimension, critics claim that the 'too strict' separation between state and religion was unbefitting of a country worshipping eight million Shinto deities (Watanabe, 2002). In reaction to state supervision and preferential treatment of Shinto since the late nineteenth century, the COJ includes clauses separating the state from religious activities (Article 20) and prohibiting the public subsidization of religion (Article 89) (see, Hardacre, 2016). In reality, however, the court rulings 2 For English translations, see Cabinet Office (N/A). 3 Traditionally, the LDP had argued that the second paragraph impeded Japan's right to collective self-defense (Boyd and Samuels, 2005). Thus, the 2005 and 2012 drafts opted to delete the controversial paragraph. However, to some extent, this has been become a non-issue since 2014, when the Cabinet Legislation Bureauthe legal counsel to the governmentrevised its interpretation of Article 9 to permit limited collective self-defense, over objections from experts of constitutional law (Stockwin and Ampiah, 2017). To this end, the LDP has suggested a new template for Article 9.2 that would (1) state that Japan had the right to self-defense, (2) to this end, could maintain the SDF, and (3) name the Prime Minister commander-in-chief, while allowing for Diet oversight (Mainichi Shimbun 8 July 2019). The LDP's support for a more minimalist position owns itself to changing political realities and interpretations of Article 9. However, our experiment focuses on the earlier, 2012 version of Article 9 revision, which is the most recent 'official' proposal from the LDP and is the only one with concrete legislative language. The 2012 draft called for clear-cut changes centering on the formal enumeration of a defense army and the deletion of the second paragraph. have demarcated what amounts to religious activity and what to tradition. In 1977, for example, the Supreme Court ruled that it was constitutional for Shinto priests to pray for the safe construction of a local gymnasium. 5 The LDP argued that its amendment proposal of Article 20 would bring the COJ in line with these rulings, such that the separation of religion and state 'does not preclude activities that remain within the realm of social rites/etiquette or traditional activities/customs' (LDP, 2013).
Another amendment proposed to make the COJ 'more Japanese' is the restriction of political participation by foreigners. The Public Office Election Law (Article 9) grants suffrage only to Japanese nationals over the age of 18. The Supreme Court has ruled consistently that foreign residents also were entitled to enjoy fundamental human rights in principle, yet this was not necessarily the case for the right to suffrage. 6 That being said, the justices opened the door for a permanent residents' right to vote in subnational elections, arguing in 1995 that parliament was free to draft corresponding legislation (Ashibe and Takahashi, 2019). 7 This alarmed parts of the LDP concerned about the potentially threatening influences of foreign agents on Japanese politics (see, e.g., Yamasaki, 2001). The party's 2012 constitutional draft would amend Article 15.3 to explicitly restrict voting rights to adults possessing Japanese nationality, based on the rationale that local self-government was a vital part of state institutions (LDP, 2013).
It should be noted that the LDP's proposals on human rights are not limited to traditional, conservative issues. In fact, its 2012 draft includes relatively progressive changes that would expand individual and social rights, particularly rights to access government information (i.e., to transparency), to enjoy a favorable environment, and to privacy. At present, they are missing from the constitution, though many scholars contend that these rights can be deduced from existing COJ stipulations and Supreme Court rulings. For example, environmental rights are implied by the rights to pursue happiness (Article 13) and maintain wholesome and cultured living (Article 25) (Kondō, 2016). However, proponents of these amendments have argued that these rights had become a global standard and were too important to be left to statutory law, let alone interpretation (Nishi, 1995). Many progressives, who have traditionally opposed the constitutional change, seem open to adding those rights, although some caution that conservatives are merely using these stipulations to win support for amending Article 9 (Watanabe, 2002).
A final topic, whose prominence has risen in recent years, is the addition of an entirely new chapter on national emergency provisions. In the commentary accompanying its 2012 draft, the LDP (2013) explains that it was the state's most important responsibility to protect the people's life, physical security, and property, not only in times of peace but also in the case of a national emergency. To that end, it has called for new articles that detail the conditions for declaring a state of emergency, the expansion of the executive's authority during emergencies, and the obligation of people to follow government directives. These purportedly reflect lessons learned from the government's poor response to the Great Eastern Japan Earthquake of 11 March 2011. Their inclusion in the LDP's election manifesto of 2017 has increased their public salience further, and the topic has been popular with its coalition partner, Kōmeitō. That said, while the events of 11 March may have heightened the urgency of conservative authors, they were hardly the impetus, as the first drafts including such stipulations had appeared as early as the 1950s (Winkler, 2020). At the same time, administrative responses to natural disasters and international conflicts are already outlined in legislation, lessening their urgency. The Disaster Countermeasure Basic Act (Saigaitaisaku Kihon-hō) defines how national and local governments should utilize public resources to protect lives and property, and the various 'Contingency Laws' (Yuji Hōsei) describes when and how the SDF can be mobilized to protect Japanese citizens and territory. Taken collectively, these proposals from the LDP's (2012) draft, with their focus on ideologicalas opposed to institutionalissues, represent one major thrust of the amendment debate ever since the 1950s. However, this desire to make the COJ more 'Japanese' is not universally shared among political parties presently represented in the National Diet. Within the governing coalition, the LDP's junior partner Kōmeitō has argued that the 2014 reinterpretation of Article 9 has done away with the necessity to amend Article 9; instead, it has advocated for the addition of new environmental and privacy rights to the COJ. Kōmeitō's votes are crucial to the coalition's electoral competitiveness in urban districts, and the party has played a moderating role in shaping the LDP's amendment agenda (Liff and Maeda, 2019).
Views are more varied outside of government. The center-left Constitutional Democratic Party (CDP)the largest opposition partyhas also supported the addition of new rights. However, its priority in the 2017 election was to restrict the prime minister's power to dissolve the House of Representatives, which it believes the LDP has abused. The center-right Japan Innovation Party (JIP; Nippon Ishin no Kai), meanwhile, aims primarily for institutional changes that include the introduction of more decentralized governance and the establishment of a constitutional court, as well as the expansion of public education. Two minor left-wing parties, the Japan Communist Party (JCP) and the Social Democratic Party (SDP), oppose discussions about any potential amendments, particularly to Article 9.
The lack of agreement on amendment topics within the ruling coalition, much less within parliament, poses a challenge to the LDP. Article 96, which details the procedure for revising the COJ, requires the concurring vote of two-thirds of the House of Representatives (lower house) and the House of Councillors (upper house), followed by a simple majority in a national voter referendum. For an amendment to be successful, the LDP would need not only the votes of the Kōmeitō but also those of the JIP and other opposition forces. While the LDP has failed to gain the endorsement of other political parties in parliament in the past, should voters express strong enthusiasm for its proposals, then sympathetic parties, such as Kōmeitō and the JIP, may follow suit. What, then, drives grassroots support?

Two interpretations of public sentiment
Various media have conducted public opinion surveys on the Japanese Constitution over seven decades, but their data paints a complicated picture (cf. Sakaiya, 2017). On the one hand, most citizens appear to be reasonably happy with the historical performance of the COJ. In a 2017 poll by the Yomiuri Shimbun, the largest newspaper in Japan, 89% of respondents agreed that the COJ had played a positive role in society since its ratification. On the other hand, many people seem to believe that amendments would make the constitution even better. Figure 1 shows the net support rate (support minus opposition) for constitutional revision, taken from the Yomiuri Shimbun's annual surveys, dating back to 1991. 8 Support rose steadily through the 1990s and 2000s, hitting a peak of +42 percentage points in 2004, but it has declined quite dramatically since then. In the 2020 survey, 49% of respondents favored the amendment, while 48% were opposed.
Of particular interest is the striking fall in popular support after 2005 and 2012, when the LDP announced its proposals for a complete rewriting of the constitution. There are (at least) two ways to interpret this phenomenon. One is that many respondents who supported amendment in the abstract disagreed with the specific content of the LDP's drafts. 9 Prior research demonstrates that voters' 8 Many national newspapers run annual surveys on attitudes towards constitutional revision, but their estimates tend to vary because of differences in question wording. We rely on surveys by the conservative-leaning Yomiuri, because of the longer time coverage. 9 Sakaiya (2017: 190-193) argues that vague survey question wording, in conjunction with negative connotations attached to the term 'national defense army' and wide-spread criticism of collective-self-defense, was to blame for this decline in support. He concludes that voter sentiment towards Article 9 was conservative, in that it preferred retention of the status quo and reacted negatively to proposed changes of that status quo. amendment preferences can be issue-specific (Elkink et al., 2020), especially when they are more familiar with the matter at hand (Garry et al., 2005). Because the LDP's proposals covered a wide variety of topics, we cannot easily assess issue preferences on each item. However, we can look more broadly at people's underlying philosophies about what should be in a constitution, and hence whether different amendments are necessary.
Let us expand on this point further. On the one hand, modern conceptions of constitutional democracy, including those in standard textbooks, are based on the primacy of individual (negative) rights, which were augmented by socioeconomic (positive) rights from the early twentieth century onwards (Tsujimura, 2008;Hasebe, 2014). Governments are delegated taxation and law enforcement powers to maximize those rights, but their powers are also separated across distinct branches so that the state does not become predatory (cf. Montesquieu, [1748Montesquieu, [ ] 1989. While the constitutional menu of fundamental human rights has expanded over time (Law and Versteeg, 2011), this philosophy speaks to a more 'practical' guidebook for managing the powers and operations of the state. The progressive Constitutional Democratic Party, for instance, has stressed that 'the legitimacy of the state's power is derived from the Constitution and all state power is limited and restrained by the Constitution' (CDP, 2018).
On the other hand, some people also see the Constitution as the embodiment of national traditions and ideals. As discussed in the preceding section, conservative elites have faulted the COJ for not mentioning Japanese history and traditions in the preamble. To some extent, the desire to amend Article 9 or include state of emergency provisions speaks to the belief that improvements to the COJ can make Japan 'better', even if there is no pressing necessity. As Prime Minister Abe said to the national media, 'the constitution is the [document] that conveys the ideal of what country Japan should aim to be during the [new imperial] Reiwa era' (Jiji, 26 June, 2019).
A second interpretation for the dips in amendment favorability is that people mistrust the LDP. Relatively brief constitutions like the COJ grant significant latitude over the interpretation and enforcement of provisions to the government. As a result, the full consequences of amendments are not fully predictable. In this high uncertainty environment, voters may infer the purpose of the revisions from who is proposing them, thus making support for amendment conditional on affinity towards the LDP. Partisanship is a meaningful heuristic when parties have taken clear positions (Silagadze and Gherghina, 2018) on complex issues, on which voters' issue preferences are soft (Baum and Groeling, 2009). In addition, the government has significant agenda-setting powers, both on the content and timing of the amendment. This can increase the extent to which referendum votes are 'second-order' to party identification (Hobolt, 2007). As LeDuc (2003: 176) writes, 'In referendums, as in elections, the messenger often matters as much as the message.' The distinction between these two logicsthe merits of the proposal vs the identity of the proposeris of crucial importance to predicting the likelihood of constitutional amendment. More generally, it offers a new lens for conceptualizing public attitudes towards the national constitution, which is of relevance to any country where amendments require referenda. However, the relative salience of these two logics is difficult to test from existing surveys, especially when comparing attitudes towards a broad range of topics. Widely-asked questions in Japan, such as affinity for the constitution or support for the amendment, do not distinguish which aspects of the COJ or amendment are preferred. More importantly, there are no questions that allow us to assess the crucial counterfactual raised above: would support or opposition to amendment change if it was being proposed by somebody other than the LDP?

Survey experiment
In order to estimate the effects of partisan framing on attitudes towards constitutional revision, we examine the results from an original survey experiment of Japanese citizens. The survey asked respondents for their views on 12 amendment proposals, but the provenance of these proposals was varied: some respondents were told they were drafted by the LDP, while others were told they were proposed by a neutral panel of experts. Our main test looks at differences in amendment evaluation across these respondent groups, which we supplement with analyses of subgroup variation by respondents' own party identification and constitutional philosophy.

Setup
The survey experiment was conducted online on 1-6 June 2017. 10 Respondents were sampled nationally by Nikkei Research, balanced by census distributions on gender and region. 11 The total sample size used in our analysis, after excluding incomplete or irregular responses, is 1819. 12 We should note that constitutional amendment had higher-than-usual public salience at the time, as Prime Minister Abe had declared in a May interview that the matter would be prioritized in the LDP's (2017) election manifesto (Yomiuri Shimbun, 2017). However, instead of restricting ourselves to topics raised by Abe at the time, we inquire about a wide range of amendment proposals, which were taken word-for-word from the LDP's (2012) constitutional revision draft to add real-world validity.
The survey is structured as follows. In the pre-treatment section, respondents were asked about their interest in constitutional issues and their support for amendment generally. The 'treatment' component followed. In the post-treatment section, respondents were asked about their party identification, their beliefs about ideal constitutions, and their demographic characteristics.
The key innovation of this study is the experimental component, which varied the identity of the amendment proposer. Respondents were randomly assigned into two treatment groups and asked to evaluate the merits of different amendment proposals on a five-point Likert scale.
• Control group respondents were told that the amendment was being proposed by the 'Liberal Democratic Party'. • Treatment group respondents were told the amendment was being proposed by a 'panel of neutral experts'. 10 The survey experiment was approved by the institutional review board of the corresponding author's university on 22 May 2017. The survey was conducted on the Qualtrics platform.

11
In order to estimate heterogeneity in response patterns by age, we sampled each age decile equally. Given the rapid aging of the Japanese population, this was necessary to ensure sufficient responses from the under-30 age group. 12 The total number of respondents surveyed was 2,740. In our analysis, we exclude respondents who did not complete the survey, or whose response time was less than 300 s. The 300-s cutoff was roughly the time it took the authors to complete the survey when going through it as quickly as possible. We also re-estimated the models using different cutoffs, such as omitting the bottom 10% and top 5% response times, but these are also arbitrary choices, and there were no substantive differences in the results.
While the LDP labeling is straightforward, our choice for the alternate group, neutral experts, deserves greater elaboration. Given the specific wording of the constitutional amendment, we believed that respondents would find it unnatural if we introduced the proposal without identifying any actor. At the same time, given the postwar discourse on constitutional revision, it would be implausible to assign an opposition party as the proposer, particularly on high-salience issues like Article 9. 13 Instead, we chose the neutral expert option by drawing on previous research that links the heuristic value of elite cues to the credibility of the messenger. When elites are seen as sharing their values and having greater expertise, citizens may reasonably use the identity of the elite to inform their own decisions (Lupia, 1994;Druckman, 2001).
Credibility is obviously a nuanced, subjective attribute, but one pertinent element is neutrality. For example, in Ireland, which has held 41 amendment votes since the constitution's ratification in 1937, undecided voters appear to use bipartisanship as a signal that the amendment is not biased in favor of one region or social group and is more likely to improve public welfare (Marsh, 1996). In the Japanese context, there is no bipartisan consensus on constitutional revision. However, the government often calls on a 'panel of neutral experts' (yūshikisha-kaigi) to inform policymaking. While the term can connote collaboration with the LDP, constitutional law scholars have often taken positions contrary to the government (Yomiuri Shimbun, 2018). As such, we believe that the neutral experts designation best serves as the primary alternative to the LDP descriptor of the messenger.
In this paper, we analyze responses to 12 discrete questions about constitutional amendment. We have chosen these questions to maximize topical variety, practical vs ideological significance, and historical salience. Their contents were taken directly from the LDP's (2012) draft and was identical between the two groups, i.e., the only difference was the description of the proposer. Each respondent gave responses to all 12 questions, whose order of presentation was randomized. The historical context of these proposals was discussed in the preceding section, but we briefly summarize the main points below. The full wording of questions is listed in Appendix A: First, we have three questions on Article 9, the core issue driving amendment debates.
(2) Permit participation in overseas missions to preserve international peace and security.
(3) Require the defense of national territory and the preservation of natural resources.
References to a self-defense/national defense army have been mainstays of conservative amendment proposals since the 1950s; this has become particularly true for participation in international missions since the end of the Cold War.
Next, we ask about three provisions on establishing constitutional 'states of emergency', which has become more popular after the Great Eastern Japan Earthquake and Tsunami of 2011. The LDP's (2012) proposal seeks to expand the powers of the executive in the following ways: (4) The Cabinet can issue decrees with the force of law. (5) The dissolution of the House of Representatives is prohibited. (6) All persons must follow orders issued by government agencies.
We have three items that conservatives have raised as important symbols of national character and tradition. 13 While we deliberated whether to use an 'opposition party' label, we did not do so for two reasons. First, the Institutional Review Board of the University of Tokyo was concerned about deceiving respondents on a sensitive and important political issue, even with a clear explanation in the post-survey debriefing. While one alternative would have been to use amendment proposals other than those of the LDP, we prioritized using 'realistic' constitutional proposals to improve the real-world salience of our analysis. Other political parties have proposed amendment topics, but none to our knowledge used specific legal language.
(7) The state has the responsibility to protect the family. (8) Allow national and local governments to support religious organizations within the limits of social customs and traditions. (9) Restrict voting rights in national and local elections to Japanese citizens.
Adding protection of the family to Article 24 or weakening separation between state and religion through amendments to Articles 20 and 89 have been key elements of conservative amendment proposals dating back to the 1950s. By contrast, the explicit ban of foreign suffrage has been a recent addition, following a 1995 Supreme Court ruling on its permissibility in local elections.
Finally, we include three questions that expand the range of protected human rights, and which have received greater support from progressives.
(10) The state must maintain a favorable environment for the people (environmental rights). (11) No person's private information can be obtained, possessed, or used illegitimately (privacy rights). (12) The state is obligated to explain its actions to citizens (government transparency).
Our quantity of interest is the average treatment effect (ATE), or the difference in average outcomes between units assigned to the treatment (proposer = neutral panel of experts) and units assigned to the control (proposer = LDP). Since the outcome variable is measured in intervals (good, slightly good, cannot say, slightly bad, bad), we estimate the ATE using a Bayesian ordered logistic regression. 14 While treatment assignment is balanced on key demographic variables, we include covariates for gender, age (continuous), and education (binary: some college and above versus less than college), to adjust for the possible imbalance caused by dropping 'do not know' or 'no answer' responses. 15 In addition to the main effects, we use two variables as interaction terms to test whether the identity of the proposer has heterogeneous treatment effects on distinct subsets of voters. These correspond to the two types of framing used by proponents and critics of constitutional amendment, which we explored in Section 2. First, we compare voters with different constitutional philosophies that may guide their preferences about the overall shape of the constitution. Earlier work on referendums has similarly examined issue-specific preferences (Hobolt and Brouard, 2010;Elkink et al., 2020), but given the wide variety of topics included in our experiment, we ask a broader question in the posttreatment segment of our survey.
There are various opinions on what constitutes 'an ideal constitution', but between statements (A) and (B), which is closer to your own thoughts about the constitution? We expect 'idealistic' respondents who agree with (A) to express greater support for adding new ideational topics to the constitution, such as restricting foreign suffrage or requiring family reverence. By contrast, we expect 'pragmatic' respondents who agree with (B) to be more supportive of adding new rights to the constitution, such as government transparency and environmental rights. A plurality of our sample believe that the main function of constitutions is to enumerate human rights and political institutions (43%), but a third believe that they should embody national ideals and history (34%). 14 We employ RStan to estimate parameters in the ordered logistic regression model. The estimation algorithm, Hamiltonian Monte Carlo, requires checking that sampled parameters are converged. We used Gelman-Rubin convergence statistics, the default summary statistics in RStan, to confirm that all chains have converged. We run each model with 1,500 iterations for each of five chains, discarding the first 750 iterations as a burn-in period. We discarded the unconverged chains based on Geweke Diagnostics. 15 We have also estimated each model without any covariates, other than the treatment status, but the results do not vary substantively.
In the models including this regressor, we drop those who answered this question neutrally or did not provide an answer. 16 Second, we compare voters with different partisan affinities. We separate respondents into LDP supporters, left-wing party supporters, and independents. 17 To the extent that partisanship guides constitutional preferences, we expect LDP supporters to respond negatively to the treatment, i.e., they are more likely to support an amendment when it is phrased as an LDP initiative. By contrast, we expect left-wing party supporters to respond positively to the treatment, i.e., they are more likely to support amendments that are proposed by a neutral panel of experts. Regarding independents, to the extent that they are not attracted to any given party, we expect them to behave like progressives and respond positively to the treatment. The expected differential impact of partisan information on preferences mirrors Baum and Groeling's (2009) survey experiment on partisanship and foreign policy. In our analysis, we dichotomize party affinity (LDP vs not-LDP, Left vs not-Left, Independent vs non-Independent) and run three separate models for each affinity type, in order to simplify the interpretation of the results relative to the omitted category. A majority of our respondents identify as 'independents' (54%), while LDP supporters (22%) greatly outnumber those of left-wing parties (9%). Table 1 lists descriptive statistics for the constitutional proposals. Under Amendment Topic, we list the average responses to our 12 questions as percentages, separated for the control (LDP proposal) and treatment (neutral proposal) groups. Six proposals received majority approval in our survey (shaded in gray): amending Article 9 to allow for SDF deployment for international cooperation and the defense of Japanese territory and resources; denying non-nationals the right to vote; adding new rights to a favorable environment, privacy, and government transparency. Only one, weakening the separation of religion and state, was opposed by a majority of respondents.
An important caveat is warranted here: the conditional ATEs, based on the interaction between treatment assignment and partisanship or constitutional ideology, must be interpreted with caution. First, these were asked in the post-treatment section of the survey, and thus we cannot exclude the possibility that the treatment itself influenced their responses (Montgomery et al., 2018). We made this choice because we did not want to 'prime' respondents into thinking about constitutional revision in partisan or ideological terms. However, this comes with some costs. If respondents have strong constitutional preferences that trump partisan affinity, then seeing that the LDP proposed a particular amendment may nudge respondents to change their partisanship or constitutional beliefs. While we cannot assuage all concerns about this possibility, a balance test shows that there is no statistically significant relationship between treatment assignment and responses to the post-treatment questions. In addition, we find that partisanship and constitutional beliefs are not strongly correlated. 18 This suggests that the treatmentwhich is about partisanshipmay have subtle effects on party identification but is less likely to affect responses to the constitutional ideology item.
Second, because partisanship or constitutional beliefs are not randomly assigned to respondents, we need to take care in interpreting causal relationships. For example, while differences within subgroups, 16 We use the 'idealist' vs 'pragmatist' label, in part because these dimensions do not overlap neatly with conventional ideological labels. For example, 'conservatives' may be expected to defend the status quo, while 'liberals' or 'progressives' may be expected to prefer adaptation to changing times. However, in the Japanese context, the conservative LDP prefers constitutional change while the progressive JCP or DPJ prefer to maintain the COJ as it is. At the same time, former-PM Abe has changed his priorities for amendment with an eye towards public opinion and saliency, but this has been criticized by Japanese conservative intellectuals for lacking principle (e.g., Nakanishi and Saeki, 2018). 17 We assessed party affinity through two questions. We first asked respondents whether they support a particular party. If they answered 'no', we asked a follow-up question, pressing them for a choice. Respondents were categorized as 'independent' if they answered 'no' to both questions. They were categorized as LDP or left-wing party supporters if they answered 'yes' to either the first or follow-up question. During our survey period, the clear anti-amendment/left-wing parties were the Democratic Party of Japan (DPJ), the JCP, and SDPJ. 18 Party identification and constitutional ideology are not strongly correlated. For example, the share of respondents who believe that constitutions should be 'idealistic' by party identification are: LDP=52.4%, Ishin=47.9%, JCP=46.7%, DPJ=37.9%, Komeito=37.5%, Independents=36.4%. The JCParguably the party that is most opposed to constitutional changeis closer to the LDP and Ishin than to the DPJ. The most 'pragmatic' are political independents who lack party identification. e.g., treated-LDP partisans vs non-treated LDP partisans, are causal, those between subgroups, e.g., treated-LDP vs treated-Left partisans, are not. We will make note of these distinctions in the following discussion.

Results
Our analysis finds that framing the provenance of constitutional change as 'neutral' vs 'LDP' influences support for the amendment. Figure 2 arrays the ATE with 90% Bayesian credible intervals, showing the marginal change in support for each amendment when the proposer is described as a neutral expert panel, rather than the LDP. For ease of presentation, Figure 2 collapses the ATE for the five response categories into binary 'support' and 'oppose' estimates; the original ordered response estimates are show in Appendix B. 19 We can see that the ATE is never negative, meaning that the neutral framing is seen at least as positively as the LDP framing. However, the statistical and substantive significance of the treatment varies across issues.
On Article 9the most hotly contested topic in Japanthe treatment does not have a statistically significant effect on any item, although the effect is in a positive direction. By contrast, the ATE is statistically significant for two sets of topics: the addition of national emergency procedures, and the addition of new human rights. In these categories, respondents are more likely to support the proposal when they believe it is suggested by a neutral panel of experts, as opposed to the LDP. On national emergency procedures, the addition of Cabinet decree powers is supported by an additional 7.6 percentage points, and limiting the dissolution of the lower house by 4.7 points. There is a 2.1-point increase in the requirement that people must obey government directives, but it is not significant at the 90% level. On the addition of new rights, that to a favorable environment is supported by an additional 4.5 percentage points, to privacy by 4.6 points, and to government transparency by 2.9 points. On balance, our experiment confirms that respondents are more likely to prefer amendments when they are given a neutral political cue, rather than when it is described as coming from We also estimated these models using a frequentist framework with linear regressions. In one model, we used the same five-choice outcomes; in a second model, we dichotomized the outcome into support/opposition to amendment, dropping the neutral choice. We find similar patterns to our main analysis. For example, the treatment does not have a statistically significant effect on the Article 9 items, but it improves assessments of national emergency procedures. the LDP. To reiterate, there is no revision for which the neutral framing reduces support for amendment. We will explore why levels of statistical significance vary later in this section.
We do not expect people's responses to the treatment to be homogenous, given differences among respondents in prior issue preferences and partisanship (Hobolt and Brouard, 2010). We estimate heterogenous treatment effects by interacting treatment assignment with responses to questions assessing these two dimensions. Table 2 lists the posterior means for the ATEs (identical to Figure 2), supplemented with the heterogenous treatment effects whose 90% credible intervals do not include zero. We only note the estimated change in positive answer categories: the percentage point change in respondents who answer 'good' (G) or 'lean good' (LG) to the amendment proposal, for space purposes. Figures depicting the heterogeneous treatment effects on all response categories are included in Appendix C.
First, on constitutional philosophy, idealists who believe that constitutions should focus on national symbols and history are more likely to favor the establishment of a national state of emergency than those who prefer a more pragmatic, rights-and-institutions-focused constitution. Among this group, the neutral framing increases support for establishing limits on lower house dissolution (+8.5 percentage points) and the mandate to follow government orders (+5.3) during national emergencies. This belief also increases support for changing Article 9 to require the government to defend Japanese territory and resources (+5.9). There is no specific issue that only pragmatists, who believe that constitutions should prioritize individual rights, rate more highly under treatment. Both idealists and pragmatists support giving decree powers to the cabinet during states of emergency (+10.9 and +6.4, respectively) and adding rights to a favorable environment (+6.5 and +4.1, respectively). Speaking generally, idealists seem to support those issues that conservatives have asserted are necessary Blue lines indicate that the credible intervals do not cover 0%. We find positive treatment effects for the addition of national emergency provisions and of new rights, meaning that the neutral framing is seen more favorably than the LDP framing on these issues.
to restore Japan's might and world status, while there is a less obvious pattern to the preferences of pragmatists.
Partisanship, by contrast, has a more consistent relationship with the framing of constitutional proposals. LDP supporters are not swayed by the treatment on any issue: their support is consistent across all question categories, regardless of the framing of the proposer. This suggests fairly strong, deeply held views about the merits of constitutional amendment. One possible explanation is that constitutional beliefs precede and determine partisanship, i.e., they support the LDP because of the party's stance on the amendment. However, Horiuchi et al. (2018) find that the LDP's manifesto position on constitutionalism, which calls for a new constitution written by the Japanese people, is actually less popular among voters than the progressive proposal to add new rights. With this in mind, a more likely interpretation is that, by virtue of these proposals having been in the public limelight or by their nationalistic content, voters implicitly see them as 'LDP-ish', regardless of the treatment frame. The LDP has been the dominant party of the postwar period, and the content of its proposals, particularly high-salience issues like Article 9, may already be well known to LDP supporters, who are more attuned to the party's past and present messages.
That said, saliency or implicit partisanship is not the full story, as we find a positive ATE among progressives and independents on the establishment of national emergency provisions, an issue that the LDP has pushed strongly in the last decade. This highlights the importance of contextualizing the meaning of constitutional amendment within real world events and concerns. As McElwain and Winkler (2015) note, the salience of specific amendment issues has shifted in line with the political agenda of the day. In the annual Yomiuri survey, the percentage of respondents supporting the addition of the state of emergency stipulations increased in the wake of the Great Eastern Japan Earthquake and Tsunami in 2011. While in the years before the disaster, 11% supported their addition, that number reached 22% by 2012, only to drop back to 11% in 2019. It is quite possible that many people see national emergency issues not as a conservative agenda, but as a more neutral issue designed to shore up the government's rapid response capabilities. This would be consistent with the positive heterogeneous treatment effect among non-partisans and supporters of left-wing parties.
Similarly, we see different treatment effects on the addition of new human rights. Left-wing voters are +11.7 net percentage points more likely to favor environmental rights, and +8.1 points more likely to favor government transparency. Independents are more likely to judge environmental rights (+6.7 points), privacy (+8.6), and transparency (+5.3) positively. Both also like the addition of government decree powers during emergencies (+9.5, +8.0), when these are described as neutral rather than LDP proposals. In sum, constitutional preferences are a function of both content and provenance or put differently, these are malleable attitudes. The perception of neutral or nonpartisan backing for amendments can drive those who otherwise oppose the government to support constitutional change. However, these effects are not consistent, as voters who believe in pragmatic, rights-based constitutions are less likely to be swayed, even by neutral framings.
These findings have two practical implications for the LDP's quest to amend the COJ. First, it will be difficult to change voters' minds on Article 9, the LDP's historical priority, at least through political communication alone. The party has better odds of convincing citizens to accept new national emergency provisions, but only on certain matters, such as granting decree powers to the government. Second, if the LDP's goal is to maximize the probability of any amendment, then its best target is the addition of new rights. Even those who oppose the LDP are more likely to support these changes when the proposer is described as neutral. If the LDP can obtain the endorsement of opposition parties, many of whom have supported these amendments in the past, then it may be able to acquire the patina of neutrality necessary to nudge a majority of voters to accept constitutional revision.

Discussion
There are potential concerns about the empirical and theoretical implications of our analysis. One issue is how prior knowledge about the COJ and historical debates over amendment may influence the treatment effect. Partisan cuessuch as our treatment designare more likely to impact less-informed voter with weak pre-existing preferences. In this context, constitutional politics is a tricky domain. On the one hand, constitutional amendments are not an obscure topic. The historical background and core principles of the COJ are taught to all school children, and the need for revisions has been highlighted in LDP election manifestos since 2000. On the other hand, given the partisan nature of amendment debates, LDP supporters may be more knowledgeable about and sympathetic to revision proposals from the LDP.
As a robustness test, we examine the conditional relationship between treatment status and constitutional knowledge. We included three factual questions about the COJ in the pre-treatment section of the survey: (1) the 'three core principles' of the COJ; (2) enumerated 'duties' of citizens; (3) enumerated political institutions. 20 11% of respondents answered all three questions correctly; 45% answered two correctly; 34% answered one correctly; 10% answered none correctly. For analytical purposes, we classify as 'knowledgeble' those respondents who answered two or more questions correctly.
There are two empirical patterns worth noting. First, constitutional knowledge does not vary significantly by party identification or constitutional ideology. 56% of LDP supporters were 'knowledgeable', compared to 57% of Left-wing partisans and 56% of Independents. 56% of constitutional pragmatists were knowledgeable, compared to 57% of idealists. We, therefore, do not consider the conditional effects of partisanship and constitutional ideology, detailed in Table 2, to be mediated by knowledge about the COJ.
Second, we examine the conditional effect of the treatment by constitutional knowledge. In general, the treatment effect was greater among less knowledgeable respondents, as expected. On Article 9: Territory and Resources, foreigners' voting rights, environmental rights, and government transparency, the treatment was statistically significant among unknowledgeable respondents but not among knowledgeable respondents. On emergency: decree, emergency: HR dissolution, and privacy rights, the treatment effect was significant on both groups. On other items, the treatment was not significant in either group. More detailed results are available in Appendix C.
A separate potential concern about our analysis is the strong treatment effect on national emergency provisions. These have long been a staple of conservative amendment proposals, particularly 20 The three knowledge questions are as follows. (1) Identify the 'Three Core Principles' of the COJ. (2) The COJ specifies several citizens' duties. Select all that apply. (3) The COJ specifies several political institutions. Select all that apply. In each question, respondents were given six options. An answer was recorded as 'correct' if the response matched the correct answer options exactly. on the issue of granting decree powers to the prime minister (Winkler, 2020). While not as wellknown as Article 9, it is reasonable that many voters find this issue to be of importance, regardless of prior constitutional knowledge.
While speculative, we believe that the positive treatment effects on national emergency issues is a product of the tension between the desire for concrete constitutional provisions and fear of their abuse. On the one hand, pre-WWII history points to the risk that the State may abuse its powers in the name of the public good. On the other hand, voters may lack technical knowledge about optimal policies during emergency governance and may underestimate the range of powers already available to the central government. This is compounded by the fact that natural disasters are very real, concrete phenomena for most Japanese. Put together, many respondents may want the government to be better at emergency responses but may also be worried about potential abuses. This tension may be alleviated by the 'neutral' framing implemented in our information treatment. We hope that future studies can explore this point further, due to its importance to contemporary society given the long-term effects of COVID-19 in Japan and the world today.

Conclusion
Constitutions serve two mutually reinforcing functions. First, they are pragmatic instruction manuals that define the powers and operations of the state. By enumerating individual and social rights and erecting institutional checks-and-balances, they clarify the limits of state intrusion into the private domain and provide citizens with a focal point for judging (in)appropriate state behavior (Weingast, 1997). Second, they serve as idealistic symbols that describe the origins and legitimacy of the state (Elazar, 1985). In providing a moral basis for the polity, they encourage elites and citizens alike to abide by their principles, even in the face of short-term incentives to defect. By both logics, the smooth functioning of constitutions is enabled by public support, if not veneration. However, existing surveys in Japan have shown significant popular disagreement: while most people think that the constitution has performed well so far, opinion is split on the need for amendments to improve its contents.
This paper seeks to understand the determinants of public opinion towards constitutional change using metrics that extend beyond the Japanese case. Our survey experiment tests whether attitudes are shaped by the contents of the proposal or the nature of the proposer. In general, we find that the ATE is positive: the neutrality of proposals increases support for the amendment. The ATE is stronger for the addition of new human rights and the enumeration of new national emergency provisions, whose provenance is less obviously partisan, but weaker for changes to Article 9 and the inclusion of symbolic, nationalistic principles, which are easier to trace back to a conservative agenda.
Given that the politics of constitutional change is heavily influenced by national context, one possible concern may be the external validity of our findings. The COJ was ratified during the post-WWII Allied Occupation, and it has suffered historically from accusations of illegitimacy. While recent surveys from the Yomiuri show that only 6% of respondents support the constitutional amendment because of the COJ's origin story, this rationale is regularly reinforced by the LDP, and thus may influence preferences through party identification. At the same time, because Japan has never held a constitutional referendum, the lack of prior experience may make respondents more susceptible to elite framing that in countries with frequent referenda (Lupia, 1994;Hobolt, 2007).
That said, constitutional referenda are rare events in most countries, and our overall findingthat both the proposer and the proposal matterhas implications for a wide variety of national contexts. The percentage of constitutions that give citizens direct input in amendments has increased rapidly, from only 20.4% in 1950 to 37.9% in 1990 and 56.8% in 2010 (Elkins et al., 2009). 21 Among established democracies, prominent cases include Australia, Ireland, and Switzerland. At the same time, 21 Calculations are based on the authors' analysis of data from the Comparative Constitutions Project. This is the percentage of constitutions whose value for 'amndappr_8' equals '1'. the substantive range of the amendment proposals examined in this paper comports with global trends of increasing constitutional specificity (Versteeg and Zackin, 2016;Abramson and Barber, 2019).
In conclusion, to divine voter behavior, we must first understand how voters conceptualize the merits of constitutional change, and how that resembles or differs from their behavior in legislative or executive elections. We find that partisanship tends to affect public preferences, but that it is not consistent. Progressives, who largely oppose the LDP, do not automatically prefer proposals that are non-partisan, suggesting that they have independent beliefs about the merits of constitutional change. One separate dimension is a preference for more ideational vs pragmatic constitutions, but this too does not seem to apply to all topics. We believe that this paper can be a stepping stone for further, comparative studies about the nature of constitutional preferences and the intersection of constitutional ideology and partisanship.
Average treatment effect of neutral vs partisan proposals, conditional on supporting the LDP or not. Marginal effects, based on Bayesian ordinal logistic regression models. Marker denotes posterior means; lines denote 90% credible intervals.  Average treatment effect of neutral vs partisan proposals, conditional on not supporting any party. Marginal effects, based on Bayesian ordinal logistic regression models. Marker denotes posterior means; lines denote 90% credible intervals. Note that 'Emergency: Follow orders' does not pass the Geweke diagnosis but Rhat is less than 1.1.