Introduction

Recent scholarship on morality politics has challenged the long-standing view that public policies can be typologized according to their intrinsic moral content. These scholars claim the use of morality is strictly a framing strategy by advocates, and any policy could be “moralized” or “demoralized” depending on the context of current public opinion. Mucciaroni (2011, 191) notes that scholars should “probably speak of morality frames rather than morality policy.” That viewpoint rejects the long-standing idea that some policies may have intrinsic characteristics that make them more identifiably moral than other policies. These views call for a paradigm shift for morality policy research by asking scholars to abandon the morality policy typologies that have defined the literature thus far in favor of focusing analysis on moral rhetoric framing strategies.

We disagree with those assertions and offer theoretical arguments and empirical evidence demonstrating that morality cannot be solely a matter of rhetoric strategy. We offer four contributions for consideration. First, we begin by describing how the classical literature on morality politics has always viewed framing as a part—but not the whole—of the morality policy process. Morality policy has more intrinsic moral content than other types of public policies. Second, we describe how this criticism of US-based scholarship fails to draw important lessons from comparative morality policy research done abroad, which has improved scholars’ understanding of the traditional morality policy typology. Third, we utilize new insights from political psychology—specifically Moral Foundations Theory—to conduct a quantitative content analysis designed to measure the amount of moral language present in these statements from eleven public policy issues (death penalty, same-sex marriage, physician-assisted suicide, tribal gaming, Official English, abortion funding bans, medical marijuana, recreational marijuana, Right-to-Work legislation, minimum wage increases, and limits on property taxes). Our analysis demonstrates detectable patterns of moral rhetoric in a manner that is strongly consistent with the traditional morality policy typology, thus supporting the paradigm that continues to serve well the literature. Fourth, we compare a typical subjective framing analysis that could be found in this literature against our quantitative content analysis to demonstrate the utility of using more objective, more empirical criteria for the moral categorization of policies.

Our results support the argument that morality policies have intrinsic moral content and counter the assertion that morality policy is nothing more than the strategic use of morality frames in policy debates. Our application of Moral Foundations Theory to a quantitative text analysis allows us to detect and quantify moral language in written statements, and this analysis supports the conclusion that future research should differentiate between “pure,” “mixed,” and “nonmoral” policy categories.

We begin our study with an overview of the theoretical origins of morality policy and then discuss a trichotomy that classifies public policies according to their moral content. To guide our empirical analysis, we draw from Moral Foundations Theory, arguably the best-known use of moral psychology in political science (Graham et al. 2013). It is based on five foundations of morality (Care, Fairness, Loyalty, Authority, and Sanctity) that have application both historically and cross-culturally.

We then proceed with our two-pronged quantitative content analysis. Study 1 is a quasi-replication of the Mucciaroni (2011) study of same-sex marriage which shows a higher level of moral content in sixteen referenda written statements than what Mucciaroni had reported with respect to state legislative speeches. Study 2 employs statistical tests to categorize the eleven public policies based on their intrinsic moral content. On a priori grounds, we had categorized all of our issues as morality policies except for the non-morality issues of minimum wage,  Right-to-Work and property tax limits. But the empirical analysis in Study 2 further differentiated a subgroup of “mixed” morality policies: abortion funding, Official English, and tribal gaming. No doubt there are many other highly politicized issues today that blend morality and instrumentalist attributes, such as the debates over immigration and environmentalism. Given that our empirical analysis offered a needed correction to our theoretical assumptions, we conclude by statistically verifying why future research should employ a tripartite classification of policies, not the dichotomy as originally ensconced in the morality policy scholarship.

The theoretical origins of morality policy

Research on morality policy was inaugurated by Meier (1994). From those beginnings a corpus of scholarship has developed in the USA (Mooney 2001; Tatalovich and Daynes 2011) and across Europe (Engeli et al. 2012; Knill et al. 2015). The consensus among morality policy scholars is that the core attribute differentiating moral from nonmoral policies is that contested values or religious beliefs (Permoser 2019) and not economic interests divide the opposing advocacy coalitions (Meier 1994; Mooney 2001). When Mooney (2001) speaks about “first principles,” he is not limiting morality policy to only matters of personal immorality, but embraces such lofty values as freedom, equality, justice, and dignity or personal autonomy. Moral Foundations Theory embraces the breadth and depth of that observation by allowing us to measure and detail the multiple dimensions of morality.

The concept of framing has a long scholarly pedigree (Goffman 1974; Entman 1993), and researchers in morality policy have been acutely aware of how the strategic use of framing could influence policy debates over, for example physician-assisted suicide (Haider-Markel and Joslyn 2004), late-term abortions (Schonhardt-Bailey 2008), and capital punishment (Baumgartner et al. 2008). But even earlier Meier (1994, 247) drew the distinction between one-sided or consensual morality policy, which he labeled the politics of sin, and two-sided or contentious morality policy that redistributes values. Sinful policies are universally condemned because one dominant frame has mobilized a powerful supportive “policy monopoly” (Baumgartner and Jones 1993), whereas two-sided morality policy provokes a legitimate debate between competing advocacy coalitions. To break the stranglehold of morality as sin, Meier (2001, 25) argued that “[t]he only possible option is to change the social construction of the debate from sin to some other dimension, that is, to frame the issue in such a manner that opposition becomes legitimate and the redistributive nature of the policy becomes open and acknowledged.”

Probably the best-known example of how re-framing fundamentally changed a morality policy debate was the adoption of state lotteries (Pierce and Miller 2001). From the late 1800s until the mid-1900s, the general prohibition on new lotteries was due to growing opposition from religious leaders and social reformers who viewed them as morally corrupting, dishonest, and the cause of numerous social problems (Clotfelter and Cook 1989, 37). So long as lottery revenues were budgeted through a general fund, opponents enjoyed the one-sided advantage of moralizing that gambling was a sin without fear of contradiction. But re-framing lotteries in terms of dedicating new gambling revenues for only education effectively transformed a one-sided issue that galvanized Protestant (though not Catholic) opposition to gambling into a two-sided issue that weakened that religious source of opposition (Pierce and Miller 2001). Earmarking lottery revenues for education became the touchstone for the post-1993 lottery enactments, which included Southern states with large populations of conservative Protestants (Nelson 2013, 82). Nonetheless, this re-framing did not change the character of lotteries as gambling. Thus, Pierce and Miller (2001, 167) believe that “we need to appreciate the variation within the general category of morality policy” because “different morality policies can involve different constellations of values. Identification of those values yields important insights into the nature of the politics of that policy.”

In other words, nothing in that body of scholarship sought to deny that morality policies were fundamentally different than non-morality policies. The denial that morality policy has any “intrinsic” moral content can be dated from a Mucciaroni (2011) publication that prompted later studies alleging the same claim (Ferraiolo 2013; Mucciaroni et al. 2019). The Mucciaroni (2011, 191) argument that “we should probably speak of morality frames rather than morality policy” essentially dismisses any substantive content in favor of strategic purpose. After a content analysis of state and congressional legislative debates over gay rights showed that instrumental or consequentialist frames were employed and not simply a polar debate over moral absolutes, Mucciaroni (2011, 211) concluded that “the morality politics literature not only exaggerates the importance of private morality frames, but it offers an overly simplified view of framing strategies by reducing most politics on these issues to the morality of private conduct.” Thus, he alleged that “Meier’s emphasis on sin and perversity also overly restricts the range” of morality policies, whereupon Mucciaroni (2011, 189) differentiated between private behavior morality and the morality of social or governmental actions.

A policy trichotomy of moral content

Mooney (2001, 8) spoke for many of us morality policy scholars when he downplayed the relevance of instrumental aspects: “All morality policies have certain technical and instrumental questions associated with them, but the distinction is that nontechnical, controversial moral questions are far more prominent and primary in the debate over them than they are in the debate over nonmorality policy.” But that conventional wisdom of a dichotomy that separates morality policy and non-morality policies, embraced in much US-based scholarship (Tatalovich and Daynes 2011; Mooney 2001; Meier 1994), may need refinement.

Unlike the Mooney (2001) formulation, scholars based in the University of Munich (Knill et al. 2015, 6–8) have conceptualized gradations of moral content that range between “manifest” or pure and “latent” or mixed morality policies, which have instrumental qualities as well. From their perspective, gambling, illicit drugs and tobacco regulation, as well as gun control illustrate “latent” morality policies. Mixed morality policies are clearly more amenable for asserting divergent frames since different stakeholders would have differing perspectives, some moral, and others not. According to Knill et al. 2015, 7): “Latent morality policies might therefore either be framed as morality or non-morality issues, depending on the underlying constellation of interests. What distinguishes both manifest and latent morality policies from non-morality policies is their connectivity to value issues.”

Across nineteen European nations, there has been a movement during 1960-2010 toward permissiveness in restrictions on eight morality policies (one exception are the greater restrictions on access to child pornography [Person et al. 2016]). As Knill et al. (2015, 27) explains, “[punitive] authority has practically vanished as the dominant style of moral regulation by 2010; all countries have moved toward the ideal types [of] permissiveness or punitive permissiveness.” While we can presume that those 1960 s restrictive regimes resulted from one-sided policy monopolies, unfortunately Knill et al. (2015) did not identify strategic framing as a pivotal variable in the fifty-year trajectory of why those policies changed. Similarly, a study of policy change in the aftermath of seventeen rampage shootings also raised doubts about the impact of any framing effects. According to Hurka (2017, 172; also see Hurka and Nebel 2013), “the empirical evidence broadly suggests that the [framing] factor is not decisive for the way focusing events are processed politically and their outcomes in terms of policy change. Both events of high and low causal complexity led to high and low levels of policy change, which implies that we must shift our focus to other factors.” In sum, this European research on policy change does not even imply that framing altered the fundamental character of a morality policy. The morality policy status quo was ruptured by punctuation, but only to the extent that a window of opportunity opened that allowed a new frame to influence public opinion.

Detecting moral content with Moral Foundations Theory

In trying to untangle whether moral content is intrinsic to a policy or merely strategic, we confront a chicken-and-egg dilemma. Public policies are not cells that can put under a microscope or geological formations from which soil samples can be extracted. They are social constructions, so the only way policies can be analyzed is through the language used to describe, justify, and criticize them. But how can we differentiate between language as rhetoric frames or as indicators of substantive content?

A measure of reliability is gained when many scholars agree in identifying the same morality policies, abortion and same-sex marriage are notable examples (Studlar 2001), or when samples of respondents are asked to differentiate between moral and nonmoral policies (Mooney and Schuldt 2008; Smith 2002). But both those approaches also rely on perceptions, not empiricism. What further compounds our chicken-and-egg dilemma is that rhetoric frames which, at first glance, look like consequentialist or instrumental arguments may actually have implicit moral meaning. But policy researchers are not trained in the philosophical traditions of linguistic analysis, so we likely would not appreciate the depth of moral content in consequentialist statements even if recognized. In sum, to delineate policy substance from strategic framing we need a more objective, empirical, and value-free methodology to analyze the language of public policy. That is the promise and potential of Moral Foundations Theory for policy scholars (Tatalovich and Wendell 2018).

Moral Foundations Theory was developed by social psychologists trying to understand why morality varied across countries, and yet still showed fundamental agreement on many moral themes. Empirical research has revealed five “foundations” of moral intuitions that appear to be human universals. The Care foundation, which pertains to helping others and avoiding danger, is indicated by such virtues as compassion, peace, and security and such vices as cruel, violence, and war. The Fairness foundation, which pertains to treating others the same way, is indicated by such virtues as rights, equality, and reasonable and such vices as bigot, favoritism, and prejudice. The Loyalty foundation, which pertains to supporting one’s group or community, is indicated by such virtues as family, nation, and member and such vices as foreign, betray, and deceive. The Authority foundation, which pertains to giving deference to social hierarchies, is indicated by such virtues as duty, law, and leader and such vices as dissent, illegal, and sedition. The Sanctity foundation, which pertains to the bodily and spiritual integrity of the individual that should not be desecrated, is indicated by such virtues as purity, innocence, wholesomeness and such vices as wicked, obscene, and lewd.

Social scientists immediately recognized that Moral Foundations Theory had great relevance to politics. Liberals tend to emphasize Care and Fairness foundations, which comprise the individualizing foundations. Care is measured by items such as “compassion for those who are suffering is the most crucial virtue” or “one of the worst things a person could do is hurt a defenseless animal” or “it can never be right to kill a human being.” Examples of Fairness items include: “when the government makes laws, the number one principle should be ensuring that everyone is treated fairly” or “justice is the most important requirement for a society” or “I think it’s morally wrong that rich children inherit a lot of money, while poor children inherit nothing.” In contrast, conservatives are sensitive to all five foundations in roughly equal amounts, especially the other three foundations (Loyalty, Authority, and, particularly, Sanctity) that comprise the binding foundations of Moral Foundations Theory (Haidt and Graham 2007; see also Koleva et al. 2012). The Sanctity foundation is measured with questions such as: “People should not do things that are disgusting, even if no one is harmed” or “I would call some acts wrong on the grounds that they are unnatural” or “chastity is an important and valuable virtue.”

Moral Foundations Theory gives scholars a powerful empirical tool for codifying moral content with textual analysis. Graham et al. (2009, 1045–1046) developed a Moral Foundations Dictionary (or MFD, at MoralFoundations.org) of keywords that designate the positive (virtue) or negative (vice) expressions within each of the five psychological intuitions of the Moral Foundations Theory (Haidt and Joseph 2004; Haidt and Graham 2007). This dictionary allows scholars to examine texts to quantify the amount of moral language present in those texts, and to differentiate between rhetoric by Moral Foundations Values. Frimer et al. (2019) are currently developing a Moral Foundations Dictionaries 2.0, and we used this version of the dictionary in our analysis below (available at http://www.jeremyfrimer.com/research-downloads.html).

An important advantage of Moral Foundations Theory is that it recognizes “moral pluralism” and is able to detect morality beyond religious traditions or beliefs. For example, the moral content of consequentialist rhetoric is largely grounded in the utilitarian principle of “the greatest good for the greatest number” (Kahane et al. 2018), and this precept would engage the Care foundation. In contrast, religious values tend to strongly engage the Sanctity foundation. In Europe morality policy is often viewed through the prism of religious values (Minkenberg 2003; Permoser 2019), and one prominent party-based theory argues that morality policies are often politicized in those nations where secular political parties are opposed by a religion-based party (Engeli et al. 2012). However, subsequent research finds that moral conflicts also can occur in countries like Great Britain or France that are considered part of the “secular world” of party politics (Knill et al. 2015). In fact, that distinction between a “religious world” and a “secular world” of morality politics may be an artifact which can be reconciled by using Moral Foundations Theory, a universal and multifaceted accounting of morality that is unbounded by time or space.

Data and methodology

We employ quantitative content analysis to detect the moral content of arguments in favor and against specific public policies in Voter Guide written statements that accompany each referendum. The use of Voter Guides has several advantages. First, they are addressed to the general public, whereas previous framing studies looked at state legislative speeches (Mucciaroni 2011; Mucciaroni et al. 2019; Ferraiolo 2013) or letters to the editor (Burlone and Richmond 2018). Voter Guides are a superior venue for targeting frames directly to the electorate as Ferraiolo (2009, 360) explained, “materials, such as voter guides, advertising, and position papers, ought to be seen not only as information for voters, but as strategic resources for ballot petitioners and rich data sources for scholars interested in the content of policy advocacy and possible reasons for its success or failure” (emphasis added). Second, complete lists of referenda by state and topic are readily accessible at Ballotpedia.com, sometimes with links to Voter Guides.

Third, most often a single written statement makes the salient points in favor or against each proposition. Voter Guides are typically authorized by the Secretary of State. To illustrate, in Washington State (with the largest number of referenda analyzed; see footnote #2) the Voter Guides are prepared by the Secretary of State who chooses an advocate (either a political/civic leader or an organization) to write the supporting and opposing statements. This approach is also followed by California (with the second largest number of referenda). The major exceptions to this common practice are Arizona and Oregon which publish as many pro and con statements as there are individuals or organizations willing to pay the required fee.

Case selection

Voter Guides provide great opportunities for scholars to examine written statements from across many issue domains, but case selection for quantitative analysis can be difficult because of challenges in data availability. For example, a relatively new morality policy is funding stem cell research, which California voters approved by passing Proposition 71 in 2004, but there had been only three other plebiscites on this topic in Missouri, New Jersey, and Michigan. (A fifth, the California Stem Cell Research Institute Bond Initiative, is scheduled for the November 3, 2020, ballot.) However, Amendment 2 in Missouri (2006) only regulated stem cell research, but provided no funding. On the other hand, most issues did not have an unlimited number of referenda from which to choose. The largest number (39) was held on banning same-sex marriage, but the other issues had many fewer. In total, for example, there were nine plebiscites on Official English, eleven on banning abortion funding, nine on physician-assisted suicide, eight on legalizing recreational marijuana, and twelve on tribal gaming. To keep the number of cases manageable, but allow for some variety in our analysis, we included no fewer than five referenda on each issue, but all five had to be identical or virtually the same. The Official English propositions were identical insofar as they simply codified the official “status” of English. The minimum wage propositions were virtually the same by giving the voters a binary choice of raising or not raising the lowest wage level, although the percentage increase and the rate of pay varied across the five referenda.

A geographical limitation is that very few states in the South or along the Midwest and Northeast corridors are represented among the 23 states that allow citizen initiatives to place constitutional amendments or statutes directly on the ballot for voter approval.Footnote 1 Even among those 23 states that allow direct democracy, a further complication is that Voter Guides are not readily available for every referendum on every issue. Just as content analysis of legislative speeches (Ferraiolo 2013; Mucciaroni 2011) was limited to a select number of states which archived verbatim transcripts or oral tape recordings, few states (notably California, Colorado, Oregon, and Washington) have complete historical inventories of Voter Guides. For most states with direct democracy, easy access to Voter Guides is fragmentary and may be limited to recent and not historical referenda. Thus, a listing of the referenda subjected to content analysis shows reliance on a relatively small number of states.Footnote 2 In sum, our case selection had to include both morality and non-morality policies, where no fewer than five referenda were held on the identical or virtually the same issue, and where Voter Guides were readily available.

Selecting eleven public policies

From arguably the most exhaustive listing of morality policies (Studlar 2001), these eight were included: same-sex marriage, capital punishment, abortion funding bans, physician-assisted suicide, Official English, tribal gaming, medical marijuana, and recreational marijuana. All were listed or implied by Studlar (2001), though some have instrumental features. Abortion is conventionally viewed to be the ideal type morality policy, but is here a “mixed” case because the ban on public funding adds economic and equity issues. Gambling is also listed, but legislation allowing tribal gaming was primarily designed to spur economic development. The historic opposition to illicit drugs clearly includes recreational marijuana, but legalization of medical marijuana is another “mixed” case because advocates promoted the therapeutic benefits of the drug to undercut its stigma as being a “Killer Weed.” And while the policy debate over language as a symbol of national unity or of identity politics would fit under the ethnic/racial category, Official English may well be considered a final “mixed” case given the strong anti-immigrant sentiments that are unleashed. To draw comparisons with putatively non-morality policies, we also include minimum wage, Right-to-Work legislation, and property tax limits.

Capital punishment

There were 34 referenda on capital punishment from 1912 through 2016, and voters supported capital punishment in all but three of those plebiscites. From an earlier period, these five referenda were selected for content analysis: Oregon (1920 and 1978), California (1972 and 1978), and Washington (1975).

Same-sex marriage

From 1998 through 2018, thirty-nine plebiscites were held on same-sex marriage (see list in Simon et al. 2018). To provide a more robust test for Study 1 (see below), all sixteen referenda with accessible Voter Guides are included in this analysis (see Appendix Table 3).

Physician-assisted suicide

Nine attempts to legalize physician-assisted suicide (PAS) through referenda were held in seven states (four were approved). Included in our analysis are the plebiscites held in California (1992), Massachusetts (2012), Oregon (1994), and Washington (1991 and 2008).

Official English

Although a majority of states have enacted laws establishing English as the “official” language of the state, only nine referenda have been held on this issue. All were approved (but the 1988 Arizona plebiscite was struck down by the state high court). Included in this analysis were those held in Alaska (1998), Arizona (2006), California (1986), Colorado (1988), and Utah (2000).

Legalized marijuana

From 1996 through early 2018, thirty states had authorized the medical use of marijuana, and nine of those states subsequently legalized its recreational use. All but one (Vermont) of the recreational marijuana laws were enacted by referenda, and fourteen states also employed plebiscites to enact their medical marijuana laws. Since eight states with recreational marijuana had also previously enacted medical marijuana by referenda, five of these states were included in both analyses: California (1996 and 2016), Colorado (2000 and 2012), Massachusetts (2012 and 2016), Nevada (2000 and 2016), and Washington (1998 and 2012).

Illicit drugs illustrate morality policy (Meier 1994), but these issues are different. First, opinion polling shows greater support for medical marijuana than for its recreational use. The 73% in favor of legalizing medical marijuana included broad support from every demographic group (Pew Research Center 2010), whereas on recreational marijuana the Pew Research Center (2015) reported lower overall support (53%). Reformers had deliberately re-framed the debate over medical marijuana to emphasize its therapeutic value in order to challenge the dominant view of the “Killer Weed” as a dangerous drug (Ferraiolo 2007).

Abortion funding ban

From 1972 through 2018, there have been forty-eight plebiscites on abortion, most designed to restrict access. But only eleven referenda seeking to ban abortion funding, except to save the mother’s life or where medically necessary, were considered. The five included in this analysis are: Colorado (1984 and 1988Footnote 3), Oregon (1978 and 1986), and Washington (1984).

Indian tribal gaming

The Indian Gaming Regulatory Act of 1988 (IGRA) allowed Indian tribes to negotiate with the states to establish casinos within their geographical boundaries. The primary objective of IGRA was to promote economic development, but there is a legacy of moral opposition to gambling in the USA (Clotfelter and Cook (1989, 47). Included in this analysis are the referenda held in California (1998 and 2008), Washington (1995 and 1995), and Idaho (2002).

Right-to-work

Section 14(b) of the Taft-Hartley Act of 1947, which labor unions termed the “slave labor” act, allowed states to prohibit compulsory union (“shop”) membership as a condition of employment.  Right-to-Work (RTW) is an economic regulatory policy, and pursuant to the 1947 Act twenty-four plebiscites were held on this issue. Included in our analysis were the referenda in California (1944 and 1958), Colorado (2008), and Washington (1956 and 1958).

Minimum wage

Minimum wage has all the earmarks of what Lowi (1964) called regulatory policy. Historically there have been 29 referenda during 1912-2016 when voters approved minimum wage increases in all but three of those plebiscites. Included in our content analysis were five referenda clustered over a fifteen year period: California (1996), Montana (1996), Nevada (2004), and Washington (1988 and 1998).

Property tax limits

The “tax revolt” was launched by Proposition 13 in California (1978), which mandated constitutional limits on property tax increases throughout the state. That anti-tax agitation spread to neighboring states of Oregon (1978 and 1984) and Washington (1997 and 2001), which comprise the universe of cases for this analysis. Limits on property taxes would seem to be the perfect non-morality policy.

Quantitative text analysis methods

Our quantitative content analysis uses dictionary-based text analysis techniques to quantify the percentage of moral language in the selected Voter Guide written statements described above. The content analysis method is to compare these supporting and opposing verbatim texts with a categorized dictionary of Moral Foundations Theory values. We employ the Moral Foundations Dictionary 2.0 (Frimer et al. 2019), downloaded on June 25th, 2019, which is an updated version of the original Moral Foundations Dictionary created by Jesse Graham and Jonathan Haidt (https://www.moralfoundations.org/othermaterials). This methodology has been commonly used in linguistic and sentiment analysis studies, in which a programming script analyzes natural language to determine whether the emotional sentiment of the text is negative, neutral, or positive (for a review, see: Iliev et al. 2015). The technique allows researchers to see how the words in these texts match to predetermined (theoretically driven) categories. In our case, we use the Moral Foundations Dictionary to categorize the Voter Guide text.

Our full-text corpus included 132 pro/con Voter Guide statements across eleven policy debates. In Study 1, we focus on 16 pro and 16 con statements in same-sex marriage referendum. In Study 2, we expand to also include five pro and five con statements for ten other policy debates. This corpus of text was analyzed in R Statistics relying heavily on several key open-source R packages (“tidyverse” by Wickham et al. 2019; “tidytext” by Silge and Robinson 2016; “quanteda” by Benoit et al. 2018). The full-text corpus began with 71,399 words, which was then filtered to remove common English “stop words,” punctuation, and numbers with the quanteda package in R Statistics. Stop words such as “an,” “the,” and “because” are typically filtered prior to processing natural language data because they are high-frequency words that are important for syntax, but do not carry meaning. This reduced the total word count across all the documents to 35,523 words.

We then conducted the dictionary-based text analysis with the quanteda package in R Statistics to compute the total number of words within each document of our corpus that matched in one or more of the five moral foundations. Descriptive summary data are included in Appendix Table 3. Percentages were calculated by dividing the total number of matched words with the total word count of each document. Although Moral Foundations Theory is known for its ability to allow comparison between all five moral foundations, the analysis that follows simply aggregates all of the five foundations together into a single measure of total moral language. This allows us to examine the overall picture of “moral talk” in these Voter Guide statements, which is the focus of this paper.

Results

Study 1 begins with an extensive content analysis of same-sex marriage debate that was targeted to the electorate of sixteen states by referendum. This study responds to the research on state legislative speeches by Mucciaroni (2011) which claimed that strategic framing of debate rhetoric (and not substantive content) comprises the essence of morality policy. A direct replication of that study would require that his legislative speeches be subjected to our same methodology, which is beyond the reach of this paper, so we propose to address the same issue of same-sex marriage using state-level texts, but which are targeted specifically to the public. The results of our content analysis (Fig. 1) show the total percentage of moral content from both supporting and opposing statements in each of sixteen Voter Guides. Given his objective of maximizing the likelihood that legislators will employ moral arguments, Mucciaroni (2011) selected states with large populations of fundamentalist Protestants (TX, TN, UT, MO, OK, and KY) or Catholics (NJ, CT, and MA). We could not replicate his selection of states because only two of the nine (Utah and Texas) held referenda on this issue.

Fig. 1
figure 1

Moral Language in Referenda Voter Guides

Full size image

Across all sixteen referenda the moral content ranged from a low of 10.5% to a high of 23.1% with an average of 16.1%. This quantity of “morality talk” is higher than what Mucciaroni (2011, 204-205) found in three Catholic (4%) and six fundamentalist Protestant (11%) states. But Mucciaroni (2011, 209) also claimed that by “reducing the opposition to gay rights to a single dimension – moral disapproval of homosexuality – the literature greatly oversimplifies and obfuscates a much richer and more complex set of arguments.” However, his definition limiting morality policy to acts of personal immorality likely understated the moral content in those legislative speeches, whereas the Moral Foundations Dictionary is more expansive and better accommodates the breadth of morality policy as a debate over (any) contested values. Thus, this method using Moral Foundations Theory detected much more “moral content” in the policy debate over same-sex marriage than what Mucciaroni (2011) found based on his subjective framing analysis.

We can see some of this underlying complexity of moral language being employed in the same-sex marriage debates by looking at how the language loads on specific moral foundations. Figure 2 shows the percentage of moral language under each moral foundation detected across the 16 pro and 16 con Voter Guide statements on banning same-sex marriage. Past research using Moral Foundations Theory has found that liberals prefer the individualizing foundations (Care and Fairness), while conservatives tend to be more balanced across all five foundations, but particularly emphasize the binding foundations (Loyalty, Authority, and Sanctity). As expected, then, statements opposed to banning same sex marriage have modestly larger amounts of Fairness language (p < 0.01). These liberal-aligned statements also had modestly larger amounts of Care language, but this difference was not statistically significant (p = 0.2). Loyalty moral language, unexpectedly, is higher for liberal-aligned statements (p = 0.01), but these differences are extremely modest. Finally, as expected, conservative-aligned statements have more Authority language (p < 0.04) and have greatly increased Sanctity language (p < 0.01). Overall, what most differentiates the supporters and opponents is the Sanctity foundation, which should be expected given that the topic is same-sex marriage.

Fig. 2
figure 2

Moral Foundations Language in Voter Guides for banning same-sex marriage

Full size image

Our Study 2 expands the scope of investigation from same-sex marriage by adding ten additional public policy issues. Employing the dictionary-based text analysis techniques described above, Fig. 3 orders the eleven issues according to the percentage of total moral content in each. (Additional descriptive data are available in Appendix Table 3.) To determine whether particular issues are statistically different from each other in Fig. 3, we employed one-way ANOVA and Tukey post hoc testing to calculate pairwise comparisons (Table 1). The one-way ANOVA data met the assumption of homogeneity of variance [Levene’s F(10,121) = 1.47, p = .16]; indicating that each issue type had acceptably equal variance. The overall ANOVA model was significant; F(10,121) = 17.5, p < .001, indicating that there was at least one significant difference in terms of morality between the issues.

Fig. 3
figure 3

Percentage of moral language across public policy domains

Full size image
Table 1 Turkey pairwise comparisons from ANOVA of eleven public policiesa
Full size table

Of the fifty-five possible Tukey pairwise comparisons, twenty-eight showed statistically significant differences, and the pattern of these results shows the issues to cluster in their total amount of moral content. First, minimum wage and property tax limits were statistically different from most (six) of the other issues, including the morality policies of physician-assisted suicide (PAS), death penalty, same-sex marriage, and medical marijuana. Second, physician-assisted suicide, death penalty, same-sex marriage, and medical marijuana are not statistically different from each other, but they are all statistically different from recreational marijuana, Official English, abortion funding, as well as minimum wage and property tax limits. Also PAS and same-sex marriage are statistically different from tribal gaming, though capital punishment and medical marijuana were not. This clustering pattern roughly forms three categories, which will be discussed in more depth below.

Third, the analysis also suggests, interestingly, that recreational marijuana and medical marijuana are decidedly different, as Table 1 shows a statistically significant difference in the moral content of these two issues. Medical marijuana is statistically similar to PAS, death penalty, and same-sex marriage, but statistically different from Official English, abortion funding as well as minimum wage and property tax limits. In sharp contrast, recreational marijuana shows a statistically significant difference with the moral content exhibited by our morality policies, but not with Official English, abortion funding, tribal gaming, Right-to-Work, or minimum wage and property tax limits. Thus, although we originally theorized that medical marijuana would be a “mixed” morality policy (due to the emphasis on its medicinal qualities), this empirical analysis shows that medical marijuana is quite analogous to the other “pure” morality policies of PAS, same-sex marriage, and capital punishment.

Fourth, as originally hypothesized, Official English, abortion funding and tribal gaming are all “mixed” morality policies. The moral content of Official English and abortion funding shows significant statistical differences with PAS, death penalty, same-sex marriage, and medical marijuana although the moral content of tribal gaming is statistically more similar to capital punishment. Finally, although we theorized that Right-to-Work should be a non-morality policy, these statistical relationships show once again how apparently nonmoral issues can exhibit moral content. Right-to-Work shows a significant statistical difference from only physician-assisted suicide and same-sex marriage and none of the other issues except minimum wage and property tax limits.

Our empirical work demonstrates clustering patterns in the moral language across these eleven issues, and this is presented graphically in Fig. 4. This figure groups the 132 pro/con Voter Guide statements from all eleven policy debates into the three clusters suggested from the above analysis: nonmoral (property tax limits and minimum wage); moral (same-sex marriage, death penalty, medical marijuana, and physician-assisted suicide); and mixed (the remaining five issues). Figure 4, thus, shows the eleven issues as a morality policy trichotomy, and the pattern is striking. ANOVA and post hoc tests reveal that all three morality policy categories are statistically different from each other (p < 0.01). This supports the conclusion that these eleven policies formed three clusters, which correspond to three morality policy categories: nonmoral, mixed, and moral. In sum, scholars might profit by following the lead of Knill et al. (2015) and differentiate between “manifest” (pure) or “latent” (mixed) morality policies and other non-morality policies based primarily on economic or instrumentalist considerations.

Fig. 4
figure 4

Three-category model of the morality policy paradigm

Full size image

Postscript: comparing moral language with morality frames

Is there any relationship between the use of morality frames in these written statements and the moral content of the words employed? Here, we conclude by employing a traditional qualitative framing analysis of our Voter Guide cases to categorize the frames employed in the written statements by each side of these eleven issues (see Appendix Table 4). The frames were identified by the authors who read the supporting and opposing statements and came into agreement if they differed. Both authors also coded each frame in the conventional manner of policy scholars as making expressive, instrumentalist, or morality arguments. In this analysis, we did not apply Moral Foundations Theory to identify “morality frames” because analysis with Moral Foundations Dictionary codes words, not full statements or arguments. This exercise is subjective and follows the approach used by those critics who assert that morality policy is simply the strategic manipulation of morality frames in building a policy argument.

As such, the extant research on framing employed very precise, issue-specific concepts of morality. In coding anti-gay rhetoric Mucciaroni (2011, 197) stated that “[f]or a statement to be coded as ‘immorality of homosexual status or conduct,’ the speaker had to make clear that the speaker him or herself (or those that they represent) believe that the status or conduct violates a moral code or religious teaching, text, or belief, or that extending civil rights or policy benefits to gays and lesbians would implicitly or explicitly sanction the status or conduct.” In the policy debate over state lotteries, whenever “a speaker explicitly put forward an anti-lottery argument grounded in the notion that participating in gambling constituted a violation of a ‘moral code or religious teaching, text, or belief’” such statements were categorized as “‘immorality of gambling’” (Ferraiolo 2013, 225). In analyzing letters to the editor on end-of-life policies (Burlone and Richmond 2018, 322), the two frames that made moral arguments were “no dignity without autonomy” for the proponents and “legality is not morality” for the opponents. In a recent study of state legislative debates on abortion restrictions, Mucciaroni et al. (2019, 175) presumed that “both sides should be expected to frame the issue on a single dimension that reflects their profound moral disagreement over whether public policy should either protect fetal life [the “right to life” frame], or, permit women to make their own choices about whether to terminate a pregnancy” [the “women’s choice” frame].

Consistent with their methodology, we subjectively coded as “morality frames” only those that expressed the core normative argument in each policy debate. Also, the previous studies of morality policy rhetoric simply calculated the percentage of articles, speeches, debates, or frames that made arguments about morality, which led those scholars to conclude that morality was not the dominant frame in those policy debates over same-sex marriage or abortion. We report similar results here, after making the same calculation based on the frames identified in Appendix Table 4. Overall there were only 12 explicit morality frames, 35 expressive frames, and 101 instrumentalist frames. Beyond that, we also compared those percentages against the percentage of moral content that each of our eleven issues had based on Moral Foundations Theory quantitative analysis. These comparisons are shown in Table 2. While there is some overlap between these indices, the correlation coefficient of 0.534 is not statistically significant at the conventional .05 level. That said, three of the top four issues which are “pure” morality policies in terms of their moral content are also among the top four morality frame issues as well (same-sex marriage, physician-assisted suicide, and death penalty). On the other hand, the worst disparities involve medical marijuana and tribal gaming, both having over ten percent moral content, but no explicit morality frames.

Table 2 Comparing morality frames with moral content in eleven policy debates
Full size table

Discussion

Our objective in this paper was to demonstrate that the deployment of morally laden language, when quantified, strikingly fits with what scholars would expect to find according to morality policy theory. Our data suggest that policies viewed by past morality scholars as “more moral” actually did have quantitatively more moral language. Like Ryan (2014), we find that any issue may contain some moral content, but we also find detectable patterns of some issues having less “morality talk” and others having more. The long-standing observation in the literature that certain issues may have intrinsically more moral content appears to be valid in our empirical investigation. Our view is that political advocates trying to moralize and demoralize rhetoric for strategic purposes are confronted by real limitations that certain policies are intrinsically more suitable (or less suitable) to moral appeals. Some issues will intrinsically synergize better with biblical appeals, while others synergize better with utilitarian appeals.

We also agree with the critics of morality policy theory that the strategic considerations of rhetoric framing are important and strongly influence the policy process. Yet, ultimately, we find ourselves concluding that this argument could be pushed too far with the claim that morality policy has no intrinsic moral content, but is simply a matter of strategic framing. In point of fact, it is both. In today’s secular world, it makes little sense for advocates to enter a policy debate by asserting religious tenets or moral truths when the context of the debate will determine what kind of rhetoric would work best (Schonhardt-Bailey 2008; Mos 2018; Knutson 2011). That said, however, no policy can be re-framed without regard to the generally perceived content of that issue, and some policies obviously engage contested values rather than conflicting economic interests. As Permoser (2019, 5) noted in her critical review of another scholar challenging the morality policy paradigm, “are we to conclude that the issue of gay rights is actually not a morality policy issue? This would go against all common-sense perceptions of the issue, be it in politics, public opinion, or academia.” As we have shown, it also goes against empirical evidence.

Looking ahead, the next step in this research agenda should follow Ferraiolo’s (2014, 370) advice and determine whether there is “more morality talk, fewer instrumental arguments, and more expressive arguments” whenever “debates are dominated by interest group activists, nonprofit leaders, or citizens, compared to when political parties and elected officials control the discussion.” That logic would extend beyond “morality talk” to include the moral content of policy advocacy as well. Beyond that, and by extrapolating from Meier’s (1994) distinction between one-sided and two-sided morality policies, presumably rhetoric intended to persuade the public would have less moral content than rhetoric targeted to a membership or constituency.

A compelling reason why Voter Guides were employed for this analysis is because referenda have been held on a huge number of policy questions, indicative of both morality and non-morality policies, even though there are acknowledged geographical limits on their availability. The use of popular plebiscites also limits the generalizability of our findings to rhetoric targeted to the general electorate, but this issue of external validity would apply with equal force to those previous studies that relied on state legislative debates. In order to design a comprehensive and comparative analysis of policy debates within a legislative assembly that are targeted to public opinion and also addressed to a membership is no easy task. One possibility might be gun control. Here, the objective would be to compare the moral content of magazine text targeted to the National Rifle Association membership against the moral content of presidential speeches on gun control addressed to the general public, but also parliamentary debates on gun control in Congress or state legislatures during roughly the same time frame. Whether the same (one) issue would be debated in all three venues, however, seems more problematic. But in theory if not in practice, this kind of research design offers a measure of external validity and thus allows us to evaluate the use of “morality talk” targeted to different audiences.