Yesterday, I got to give an International Society for Quantitative Ethnography (ISQE) Webinar on “Moral and Political Discussion and Epistemic Networks.” I really enjoyed the questions and conversation. This is the video of the whole event:
Abstract: An individual holds linked beliefs about political or moral issues, which we can model as a network. How these ideas are linked together influences the person’s actions and opinions. When individuals discuss, they share some portions of their respective networks of which they are conscious at the time. Some network structures are better than others for discussion: overly centralized or scattered networks are problematic. Individuals tend to demonstrate similar network structures on different issues. Thus, relying on certain kinds of networks is a character trait. People, with their respective networks of ideas, are also embedded in social networks. An idea is more likely to spread depending on features of both the social network and the idea networks of the people who interact. As a whole, a population may develop a shared network structure. An idea that is widely shared and frequently central in individuals’ networks becomes a norm. Institutions are partly composed of such norms. A community or a culture is a single network with disagreement. Ultimately, all such networks interconnect. This theory has implications for politics, ethics, and research methodologies.
In 2015, students at my university, Tufts, and at Kansas State University discussed the same topic–the social determinants of health–in online forums. Colleagues and I analyzed the text in a novel way. The underlying theory is that conversations can be modeled as networks, where the nodes are specific ideas and the links are reasons and other connections that people assert. Specifically, we sought to use Epistemic Network Analysis (ENA) to model and compare the conversations. ENA is now very easy to use online and I have been playing with it for other purposes. One of its distinctive features is its ability to locate specific ideas at meaningful locations on a two-dimensional graph so that you can see dimensions of agreement and disagreement.
Our results are now published as Peter Levine, Brendan Eagan & David Williamson Shaffer, “Deliberation as an Epistemic Network: A Method for Analyzing Discussion,” in Barbara Wasson and Szilvia Zörgo (eds.), Advances in Quantitative Ethnography, proceedings of the Third International Conference, ICQE 2021 Virtual Event, November 6–11, 2021 (Springer Switzerland, 2022), pp. 17-33.
The image I reproduce with this post shows one particular ENA visualization of the discussions. The one at Tufts is in blue; the one at KSU is in red; and the one in the middle shows the difference between them (literally, the Tufts network subtracted from the KSU network).
You would have to read our paper to get a full explanation, but here is a glimpse. Basically, the Tufts students tended to connect inequities in health with race and class. Some of the KSU students also made those connections, but some of them drew connections between bad health outcomes and personal behavior, for which they blamed individuals’ upbringings. Thus the KSU discussion roughly looks like a triangle with three corners (race and class, personal choices, and health outcomes), while the Tufts discussion omits one of those corners.
The same result might have been clear enough from a conventional approach–reading, interpreting, and (possibly) coding the transcripts. However, we argue that the plausibility of the ENA findings validates the method, which can then be used to model other discussions.
In linguistics, a language is a whole system of communication used by a group, encompassing its semantics, grammar, and pragmatics. A dialect is a particular form of a language typical of a specific region. A sociolect is like a dialect, except that it is used by a dispersed social group, such as a profession or a class. A register is a form of a language used in a particular situation or context, such as by lawyers in court. And an idiolect is the distinctive way that an individual speaks: that person’s active vocabulary, grammar, accent, and preferred forms and style of speech.
Previous authors have used these distinctions from linguistics to inform models of culture. Language offers a useful model or example for understanding culture because we are familiar with explicit efforts to learn second languages, to list components of a given language in a dictionary, and to invent a writing system to represent a language (Goodenough 1981, p. 3; Goodenough 2003). I am especially interested in the aspects of culture that involve value-judgments: ethics, politics, aesthetics and religion.
I believe the following ten points from linguistics are relevant to thinking about culture (in general) and specifically about values:
Registers, dialects, sociolects, and languages all encompass variation. Each person talks and understands differently from others and evolves as a speaker over time. Therefore, the same body of communication by real human beings can be carved up in many ways. For instance, we can draw a dialect map of the USA that shows more or fewer regions, and we can declare the same material to be a language or a dialect depending on whether we prefer to accentuate differences or similarities.
Idiolects do not nest neatly inside dialects and sociolects, which do not not nest neatly inside languages. They are more like complex Venn diagrams. An individual may draw on several dialects or languages. In cases of code-switching, the individual sometimes uses one register or sociolect and sometimes switches to a different one. Or people may consistently use a mix of influences from multiple sources. For instance, Spanglish is characterized by Spanish influences on English–and there are several different regional Spanglish dialects within the United States today. On the other hand, imagine an elderly American whose speech reflects some influence of Yiddish from the old country. Her idiolect is then not part of any dialect but is a unique mix of two languages with which she may successfully communicate even though no one else nearby sounds like her.
Nobody knows all of any given language, dialect, or sociolect. An effort to catalogue an entire language describes a body of material that none of the users know in full. For instance, nobody knows all the meanings of all the words in an impressive dictionary (which is, itself, an incomplete catalogue of the entire language). Instead, people share sufficient, overlapping portions of the whole that they can communicate, to varying degrees.
Each of us knows our own entire idiolect. (That is what the word means.) However, we could not fully describe it. For instance, I would not be able to sit down and list all the words I know and all the definitions I employ of those words, let alone the grammatical rules I employ. If I were presented with a dictionary and asked which words I already know, I would check too many of them. The dictionary would prompt me to recall words and meanings that are normally buried too deep in my memory for me to use them. That brings up the next point:
Language is dynamic, constantly changing as a result of interaction. A language constantly borrows from other languages. A person’s idiolect is subject to change depending on whom the person talks to. The best way to characterize an idiolect or a language (or anything in between) is to collect a corpus of material and investigate its vocabulary and grammar. That is a worthwhile empirical exercise, but it requires a caveat. The corpus is a sample taken within a specific timeframe. The idiolect or the language will change.
It is possible to make a language (or a smaller unit, like a sociolect) relatively consistent among individuals and sharply distinguish it from other languages. For instance, a government can set rules for one national language and encourage or even compel compliance through schooling, media, religious observance, and even criminal penalties. Law schools teach students to talk like lawyers in court. However, there are also language continua, in which local people speak alike, people a little further away speak a little differently, and so on until they become mutually unintelligible. Before the rise of the nation-state and modern mass media, language continua were much more common than sharply distinguishable languages. Linguistic boundaries require exercises of power that are costly and never fully successful. Meanwhile, at the level of an idiolect, individuals may strive to make their own speech consistent and distinctive or else let it change dynamically in relation with others. People vary in this respect.
Language is–in some way or degree–holistic. That is to say, the components of a language depend on other components. For instance, a definition of a word uses other words that need definitions. I leave aside interesting and complicated debates about holism in the philosophy of language and presume that some degree of holism is inescapable.
Therefore, it can be helpful to diagram an idiolect, a dialect, or a language as a network of connected components rather than a mere list. I am not saying that language is a network; language is language. However, semantic network diagrams are useful models of idiolects, dialects, and languages because they identify important components (e.g., words) and connections among them. A semantic network diagram for a group of people will capture only a small proportion of each individual’s language but may illuminate what they share, showing how they communicate effectively.
While words and other components of language are linked in ways that can be modeled as networks, people also belong to social networks, linked together by relationships of influence. A celebrity influences many others because many people receive her communications. The celebrity is the hub of a large social network. At the opposite extreme, a hermit would not influence, or be influenced by, anyone (except perhaps by way of memories).
Whole populations change their languages surprisingly quickly, and sometimes without mass physical migration. The same population that spoke a Celtic language (Common Brittonic), transitioned partly to Latin, then fully to Germanic Anglo-Saxon, and then to a mixture of Anglo-Saxon and French without very many people ever moving across the sea to England. Today about 33 million people in South Asia also speak dialects of it. A few people can strongly influence a whole population due to their network position–which, in turn, often reflects power.
The point of this list is to suggest some similarities with other aspects of culture. Like a language, another part of a culture can be modeled as a shared network of components (e.g., beliefs, values, or practices) as they are used by people who are organized in social networks, which reflect power.
Such a model is a radical simplification, because each individual holds a distinctive and evolving set of components and connections (e.g., linked moral values); but simplifications are useful. And we can model culture usefully at multiple levels, from the individual to a vast nation.
There are precedents for this kind of analysis. The anthropologist Ward Hunt Goodenough encountered the concept of idiolects in the 1940s, while studying for a doctorate in anthropology (Goodenough 2003). By 1962, he had postulated the idea of “each individual’s private culture, if we may call it that, [which] includes his conception of several wholly or partially distinct cultures (some well elaborated and others only crudely developed in his mind) which he attributes to others individually and collectively, both within and without his community. A person’s private culture is likely to include knowledge of more than one language, more than one system of etiquette, more than one set of beliefs, more than one hierarchy of choices, and more than one set of principles for getting things done” (Goodenough, 1963, 261)
Later, Goodenough named a private culture a “propriospect” from the Latin words for “private” and “view” (Goodenough 1981, p. 98). Harry Wolcott summarizes: “Propriospects … are networks of sense-making connections created and constantly being reformulated by each of us out of direct experience. Aswe develop and refine our competencies, simultaneously we ‘construct” (and continually ‘re-model’) our individual propriospects” (Wolcott 1991, p. 267).
Individuals may actually share fundamental characteristics of a group, they may perceive themselves to share those characteristics, and they may be viewed by others as sharing them–but these perceptions do not necessarily align, because every group encompasses diverse propriospects. For instance, you might think that believing in God is essential to a culture; and since you believe in God, you are part of that culture; but someone else may define the culture differently and view you as an outsider.
In a review of Goodenough, Mac Marshall (1982) wrote, “While some may disagree with the finer points of his argument, his position on these matters represents the dominant orientation in American anthropology today.”
A different precedent derives from the Polish tradition of humanistic sociology, founded by Florian Znaniecki in the early 1900s. In the 1980s, a Polish sociologist named Marek Ziolkowski (later a distinguished diplomat) wrote interestingly about “idio-epistemes” (presumably from the Greek words for “private” and “knowledge”) meaning “not only … the cognitive content of an individual’s consciousness at a given moment, but … the whole potential content that can be activated by the individual at any moment, used for definite action, and reproduced introspectively.”
Ziolkowski called on sociologists to “enquire into the social regularities in the formation of idio-epistemes, seek relationships among them, establish basic intragroup similarities an intergroup differences.” He acknowledge that any individual will have a unique mentality, “yet every individual shares an overwhelming majority of opinions and items of information with other definite individuals and/or groups.” Those similarities arise because of shared environments and deliberate mutual influence. Ziolkowski coined the word “socio-episteme” for those aspects of an idio-episteme that are shared with other individuals. As he noted, these distinctions were inspired by concepts from linguistics.
Even earlier, in 1951, the psychologist Saul Rosenzweig had coined the word “idioverse” for an individual’s universe of events” (Rosenzweig 1951, p. 301).
An analogous move is Wilfred Cantwell Smith’s influential redefinition of “religion” from a bounded system of beliefs (each of which contradicts all other religions in some key respects) to a “cumulative tradition” of thought and behavior that varies internally and overlaps with other traditions (Smith 1962). According to this model, everyone has a unique religion, although shared traditions are important.
These coinages–idioverse, propriospect, idio-episteme, and cumulative tradition–capture ideas that I would endorse, and they reflect research, respectively, in psychology, anthropology, sociology, and religious studies. However, none of the vocabulary has really caught on. There may still be room for a new entrant, and I would like to emphasize the network structure of culture more than the previous words have done. Thus I tentatively suggest idiodictuon, from the classical Greek words for “private” and “net” (as in a fishing net–but modern Greek uses a derivation to mean a network). An individual has an idiodictuon, a group has a phylodictuon, and a whole people shares a demodictuon.
These words are unlikely to stick any better than their predecessors, but if they did, it would reflect their diffusion through a social network plus their utility when added to people’s existing networks of ideas. That is how all ideas propagate, or so I would argue.
A final point: “culture” encompasses an enormous range of components, including words, values, beliefs, habits, desires, and many more. Given this range, it is useful to carve out narrower domains for study. Language is one. I am especially interested in the domain of values, so I would focus on the interconnections among the values that people hold.
Note, however, that there is no neat way to distinguish values from other aspects of culture, such as desires and urges or beliefs about nature. In his influential empirical theory of moral psychology, Jonathan Haidt identifies at least five “Moral Foundations,” one of which is “sanctity/degradation.” I would have treated this category as a powerful human motivation, akin to sexual desire or violence, but not as a moral category, parallel to “care/harm.” Maybe Haidt is right, or maybe I am, but the evidence isn’t empirical. People actually see all kinds of things as the basis for action and make all kinds of connections among the things they believe. They may link a moral judgment to a metaphysical belief, or a personal aversion, or an aspect of their identity, or a belief about prevailing norms. When we carve out an area for study and call it something like “morality” or “ethics” (or “religion,” or “politics”) we are making our own claims about how that domain should be defined. Such claims require philosophical arguments, not data.
So the steps are: (1) define a domain, a priori, (2) collect a corpus of material relevant to that domain, (3) map it as a network of ideas, and (4) map the human relationships among people who hold different idiodictuons.
For me, the ultimate point is to try to have better values, and the study of what people actually value is preparatory for that inquiry. Clifford Geertz concludes his famous “Thick Description” essay: “To look at the symbolic dimensions of social action–art, religion, ideology, science, law, morality, common sense–is not to turn away from the existential dilemmas of life for some empyrean realm of de-emotionalized forms; it is to plunge into the midst of them. The essential vocation of interpretive anthropology is not to answer our deepest questions, but to make available to us answers that others, guarding other sheep in other valleys, have given, and thus to include them in the consultable record of what man has said.”
Geertz, Clifford, “Thick description: Toward an interpretive theory of culture.” In Geertz, The interpretation of cultures. Basic books, 1973. (pp. 41-51).
Goodenough, Ward, Cooperation in change : an anthropological approach to community development. New York, Russel Sage 1963
Goodenough, Ward H. In Pursuit of Culture, Annual Review of Anthropology 2003 32:1, 1-12
Goodenough, Ward H. Culture, Language and Society (Menlo Park: Benjamin Cummings, 1981)
Mashall, Mac, “Culture, Language, and Society by Ward H. Goodenough” (review), American Anthropologist, 84: 936-937.
Rosenzweig, Saul. “Idiodynamics in Personality Theory with Special Reference to Projective Methods 1.” Dialectica 5, no. 3-4 (1951): 293-311.
Smith, William Cantwell, The Meaning and End of Religion (1962), Fortress Press Edition, Minneapolis, 1991
Wolcott, Harry F. “Propriospect and the acquisition of culture.” Anthropology & Education Quarterly 22, no. 3 (1991): 251-273.
Ziolkowski, Marek, “How to Make the Sociology of Knowledge Sociological?.” The Polish Sociological Bulletin 57/60 (1982): 85-105.
Hypothesis: People not only hold opinions about parties and specific issues; they also explicitly connect their various beliefs together to create more-or-less coherent logical structures. Understanding these structures yields insights about individuals that we would miss if we only knew a list of their opinions.
This thesis challenges a common assumption in political and moral psychology. Although scholars meticulously analyze the structure of the arguments in books, essays, and speeches, several streams of research suggest that ordinary people have no such structures–their political preferences are random, or chosen by elites, or driven by latent variables of which they are unconscious. I think that we won’t find the conscious structure in average people’s thinking unless we develop tools that can detect it.
Looking at some individuals
To that end, last summer I collected data from 100 Amazon Turk participants. I restricted the sample to liberal residents of the USA, in order to control the range of issues that would appeal to them so that I could analyze the small dataset effectively.
I showed each person a list of 30 ideas they might favor, about half of which leaned left. I simply asked, “Which of these would you like to see happen?” and encouraged them to be selective. Consistent with my effort to recruit liberals, they chose almost exclusively liberal items. For instance, 80% chose “health insurance for all” and zero chose “a return to traditional values.”*
After subjecting respondents to a bit of explanation and training, I showed them their own chosen items in pairs (in random order). For each pair, I asked, “Are you in favor of X because you are in favor of Y?” Each time they said yes, that created a link between two nodes (or an edge between two vertices, if you prefer the technical language). From these responses, I was able to generate network maps for each person and for groups.
Here is one such map. This person is a white man, age 32. He places himself as far as possible to the left on a unidimensional ideology scale and he says that he agrees with AOC most often, followed by Bernie, then Warren, then Obama, and finally, Biden. He believes that eight things that he would like to see happen (e.g., health insurance for all) provide reasons to reduce corporate influence on government. He also sees an indirect link between reducing corporate influence and racial justice (by way of reducing incarceration) and an indirect link to the environment (by way of enabling a tax on carbon). However, he doesn’t connect political reform to free college or LGBTQ+ issues, although he cares about those issues and connects them.
In short, his network is highly centralized around reducing corporate influence. This seems like an important fact about how he thinks.
Below is another map, this time for a white woman, age 30, who ranks the Democratic politicians in precisely the opposite order as the man did (Biden first; OAC last). However, she ranks herself as a one on a 0-10 ideology scale, i.e., very liberal.
Her graph is more complex than the man’s. She is 3.6 times more likely than he was to see any given pair of ideas as a connection. Two-thirds of her arrows are double-headed. (She favors less crime because she favors more trust in government, and vice-versa.) You can move around her graph from one item to the next, and then onward. It is not centralized around any single node.
As noted, I first asked respondents whether each of their goals provided a reason for another goal. It was up to them what counted as a “reason.” I believe, however, that we can put ideas together in numerous ways, with a range of logical connectors. A justification is one link; a cause is another.
Therefore, I next showed respondents pairs of outcomes and asked whether each one would cause the other one. I included outcomes that they had not selected in the first place, because you might not select something as a high value even though it causes an outcome that you do value highly.
Here is the first respondent’s map with the causal connections added in:
It becomes complicated–perhaps too complicated to understand visually. But it’s clear that he believes higher voter turnout and a more responsible electorate are “upstream” factors–they cause many desirable outcomes–even though he did not initially choose them as his priorities. It makes sense to me that they are means to the ends he favors.
Finally, I showed respondents random pairs of items and asked, “Imagine x happened. Would that change your opinion by making you more supportive of y?” This produced another kind of connection (hypotheticals), which I can add to the first respondent’s map along with the causes and reasons:
Metrics suggest that this addition doesn’t actually change this person’s map very much, although it adds a few ideas at the end of arrows.
An example: the role of reparations in the mentality of the left
Months ago, I hypothesized that reparations for slavery play an important role in the thinking of left-liberal Americans. Progressives oscillate between hostility to the US government and desire to expand federal economic action. If the government paid reparations, progressives would trust it more and would therefore tilt to democratic socialism.
I can identify some individuals who may exhibit this logic. For instance, here is a 40-year-old white woman who favors a set of progressive outcomes and sees all of them as reasons to reduce corporate power over government (rather like the first man discussed earlier):
She did not select reparations as one of her original priorities, but she was asked whether actually seeing reparations happen in the world would make her more favorable toward other outcomes. She said it would make her more supportive of 10 things, including trust in government and civility. One can imagine that she thinks: if the government paid reparations, I would want people to trust it and would want citizens to be civil to each other, because the regime would be more legitimate. However, this is largely my inference about her responses, and it’s not a common pattern in the data.
These are three examples from the 100 cases that I collected. I have many more to look at, and I have IRB permission to interview selected respondents about their reasoning.
I am also interested in examining the aggregate data. One evident finding is that people’s top priorities for how they would like to see the world change are not the same as the factors they see as most influential.
In my sample, the most frequently chosen goals are:
health insurance for all
a solution to climate change
a higher minimum wage
more school choice
The items that are seen as having the most direct effects on other items are:
health insurance for all
a less corrupt government
less corporate influence on government
a more responsible electorate
more equitable education
And the items that have the most links of any kind to other items are:
health insurance for all
a less corrupt government
more trust in government
Healthcare hangs in there as the top priority however you slice it. However, better government and economic growth–which are not chosen as high priorities from a simple list–emerge when people think about various kinds of premises or causes for the things they do value.
Another method would be to measure the relationships between people’s networks of ideas and a different variable of interest. I was planning to look at self-placement on an ideological scale, but my sample clusters together too closely. Seventy-four percent rate themselves between 0 and 2 on a 0-10 scale.
Instead, I am using as the dependent variable how people rank AOC compared to other Democratic politicians. I chose that outcome because her ranking correlates most strongly with ideological placement, and also because the biggest negative correlation in the whole matrix is between her rankings and Joe Biden’s (-0.67). Plus, her rankings have a nice distribution: the median is 3 on a 1-5 scale. So I take her strong supporters and opponents to mark the ends of a meaningful spectrum within the US left.
If we simply correlate the priorities that people choose (or don’t choose) with their rankings of AOC, then the correlations that are significant are:
more school choice and racial justice (which were chosen by identical people*)
equity for LGBTQ+ people
(not choosing) economic growth
(not choosing) less crime
(not choosing) a reduction in government debt
These correlations make intuitive sense to me. Among liberals, those who care less about crime, deficits, and growth would like OAC best.
If you put all the choices together with the demographics and ideology in one linear regression model to predict support for OAC, it does a decent job (r-squared = 0.488).** However, none of the individual items (including ideology) are statistically significant. One could conclude that knowing which items people pick helps you to predict their opinion of AOC, but you need to know most of the items.
If, instead, you look at which items people thought were reasons for other items, the model is more predictive (r-squared = .608 with the same number of variables) and ten items become stat. sig. on their own (p <.05). Knowing that an individual sees one of these items as a reason for other items gives a basis to predict that this same individual likes (or dislikes) AOC.
That seems like confirmation of my original hypothesis. The structure of people’s beliefs–more than the things they support–predicts a consequential opinion. However, I am still working on the regression and other aggregate methods; suggestions are welcome.
About one in four Americans supports reparations for slavery. There is a racial split on that question, with up to three in four African Americans–but only 15% of whites–in favor.
If you think that justice demands reparations, you should support them. You might not make reparations your main criterion for choosing candidates in a given political contest, because you might vote on other grounds, but you should endorse proposals that you believe are just.
Here I want to address a different issue. I’ll offer an explanation (not a justification or a critique) of the importance of reparations in the mentality of left-leaning Americans.
I think that many Americans on the left are torn between two political positions, each coherent on its own but in tension with the other:
1. A strong version of New Deal/Great Society liberalism and/or social democracy, in which the nation-state intervenes assertively in the economy to promote equity and environmental sustainability. This stance is compatible with enthusiastic support for voting and democratic processes. It requires a lot of trust in the state and a willingness to entrust state actors with the ability to, for example, investigate how much wealth (not just annual income) you have, which schools your kids will attend, and which health treatments will be paid for, given data about your body.
Martin Luther King, Jr., provides a classic statement of this view when he recalls the launch of the Great Society: “A few years ago there was a shining moment in that struggle. It seemed as if there was a real promise of hope for the poor — both black and white — through the poverty program. There were experiments, hopes, new beginnings.”
2. A deep suspicion of the United States government as white-supremacist, patriarchal, and colonialist: as a continuous entity that has played a leading role in genocide, enslavement, and apartheid, in part because those policies have sometimes been popular among the white majority of the country.
It’s debatable what positive program follows from the second position, but in practice, it can mean support for local initiatives, nonprofits, women- and minority-owned businesses, and autonomy at the neighborhood level. Malcolm X provides a classic text for this view:
The white man, the white man is too intelligent to let someone else come and gain control of the economy of his community. But you will let anybody come in and control the economy of your community, control the housing, control the education, control the jobs, control the businesses, under the pretext that you want to integrate.
… we haven’t had sense enough to set up stores and control the businesses of our community. … But the political and economic philosophy of black nationalism…the economic philosophy of black nationalism shows our people the importance of setting up these little stores, and developing them and expanding them into larger operations. Woolworth didn’t start out big like they are today; they started out with a dime store, and expanded, and expanded, and expanded until today they are all over the country and all over the world and they getting some of everybody’s money. …
So our people not only have to be reeducated to the importance of supporting black business, but the black man himself has to be made aware of the importance of going into business. And once you and I go into business, we own and operate at least the businesses in our community.
Note that this position is compatible with certain forms of libertarian thought but not with social democracy.
It is not embarrassing to be drawn to two incompatible views. The social world is complicated, and there are good reasons in favor of many positions. However, when you feel the pull of two incompatible ideas, a deciding factor becomes very important.
Reparations play that role for the American left. If the United States government were to pay reparations, that would tilt many left-leaning people from the second position to the first: from Malcolm to Martin, if those labels are helpful. The impact would be especially strong if Congress and the president decided to pay reparations of their own volition–not by grudgingly negotiating with a social movement–and if the payment were substantial.
The underlying theory here is similar to Homer-Dixon et al (2020). An ideology is a complex system that consists of numerous ideas with logical links among them. It cannot be described adequately by placing it on one left/right spectrum, nor even several such continua at once. It is not a point in logical space but a structure of ideas.
In complex systems, we frequently see multiple equilibria, and specific nodes have surprisingly large impact because of their location. A single node can tilt the system from one equilibrium to another.
My conjecture is that reparations plays such a role in the system of the ideology of the American left. Left-leaning people may not rate it as the most important issue. They may not even endorse it whole-heartedly. But it (perhaps uniquely) can tilt them from a libertarian equilibrium to a social-democratic equilibrium.
This is an empirical conjecture for which I do not have data. To test it, we would have to explore the epistemic network of left-leaning Americans, either by analyzing large bodies of text or by surveying individuals about their ideas and perceived connections among their ideas.