Throughout his career, John Roemer’s work has been uniquely situated between the fields of microeconomics, game theory, philosophy, and political science. His research makes use of the tools of classical economics to analyze dynamics typically thought to be outside the scope of economics: from notions of fairness and morality, to the possibility of overcoming capitalist social relations. In doing so, it defends those tools against charges that they can’t describe the behaviors we see, at the same time as it renders vital social questions digestible for disciplines that rarely engage them.
Roemer is perhaps best known for his contributions to theories of distributive justice. Within the field of moral philosophy, he is one of a handful of scholars who have sought to formalize distributive theories in order to compare their merits. To moral philosophers, he argues that outright dismissal of consequentialist theories of justice, and their replacement by complicated deontological models, is a mistake. And to the world of economics, he posits that economic theory cannot be divorced from moral philosophy—that the emphasis on reaching equilibrium itself necessarily carries moral assumptions.
Roemer’s first major work was in Marxian economics. In A General Theory of Exploitation and Class, he identifies five classes with distinct material interests: pure capitalist, petty bourgeois, self-employed, semi-proletarian, and pure proletarian. According to his Class Exploitation Correspondence Principle (CECP), if individuals optimize according to their best interests, then those who optimize by hiring labor are necessarily exploiters, while those who optimize by selling their labor are necessarily exploited. Through the CECP, Roemer employs rational choice theory to prove as a theorem what Marxists had previously presumed from observation.
Some of his more recent work develops a theory of “Kantian optimization.” Here, he challenges the individualist definitions of strategic behavior propagated by standard rational choice theory, instead proposing a rational model for cooperative behavior. In this framework individuals reason according to the categorical imperative, asking: “If I were to deviate from my stipulated action, and all others were to deviate in like manner from their stipulated actions, would I prefer the consequences of the new action profile?” Roemer finds not only that there are a variety of games in which this kind of optimization leads to Pareto efficient outcomes, but that it can furthermore resolve both the tragedy of the commons and the free rider problem, both of which are Pareto inefficient consequences of Nash equilibrium.
Our discussion covers the range of his career and projects, and considers the broader import of theoretical work in the current political conjuncture.
An interview with John Roemer
jerome hodges: We want to start off with your background story, because, try as we might, we couldn’t find any detailed biographical information about you. Someone on Wikipedia included this very suggestive paragraph about you being dismissed from Berkeley for anti-war activism during Vietnam; I’d love to hear that story, and about the origins of the September Group.
john roemer: I grew up in a socialist household. My parents were first-generation Americans of Eastern European Jewish descent, and committed leftists. So I grew up always thinking I was a socialist; when I was a child I remember thinking the good guys were the Brooklyn Dodgers, the workers, and the Democrats and the bad guys were the Yankees, the bosses and the Republicans. Nevertheless, I wasn’t politically active until graduate school. I chose Berkeley in large part because it was such a politically active place, and once I got there I joined a Marxist-Leninist party. I could no longer justify studying pure mathematics and changed my major to economics. In the fall of 1968, I was one of about 100 students who occupied the administration building. We were eventually arrested by about 1000 highway patrolman whom Ronald Reagan had called from around California to extricate us. During the faculty student committee hearings we were given the option to apologize for our participation and be put on probation. I told the committee that I acted on principle, and they suspended me for a year.
Having been suspended I also lost my draft deferment, and so I began to teach math at a high school in San Francisco in order to obtain another deferment. I ended up staying for five years, past the age of 26 at which I could no longer be drafted, because I got involved in workplace organizing through the Progressive Labor Party. We organized a left-wing caucus of the teacher’s union, and facilitated a number of active strikes. It was only at the age of 29 that I went back to Berkeley to finish my dissertation.
When I got my first job at UC Davis, I had never read any Marxian economics. My colleagues suggested that I read Michio Morishima’s Marx’s Economics and teach a course on it. It blew me away. That mathematical approach to Marxian economics set my research agenda for the next 10 years, one product of which was my book, A General Theory of Exploitation and Class. In that book I propose that the Marxist theory of exploitation was really a special case of a more general theory, and I rephrase historical materialism in terms of the evolution of different modes of exploitation.
While writing that book, I read G.A. Cohen’s Karl Marx’s Theory of History. He was doing in philosophy what I was trying to do in economics, namely applying contemporary theories of our fields to Marx’s writings. Another person thinking about this was Jon Elster, and he and Cohen organized a meeting of like-minded Marxists in 1979. I sent Cohen, whom I hadn’t met, some chapters of the book I was working on, and he invited me to the second meeting of this group, which was in London in 1980. I attended and suggested that people like Eric Olin Wright, Pranab Bardhan, and Robert Brenner join as well. We called ourselves the No Bullshit Marxist Group. The group continues to meet every year, though most of us no longer identify as Marxists.
Cohen was an absolutely magnetic character; he had a mind sharp as a knife combined with a fantastic sense of humor. Under his influence, I became committed to understanding the ethics of exploitation: why was it ethically bad for people to be able to purchase goods embodying less labor than they had expended in production? Neoclassical economists say it’s not a bad thing, the worker has simply paid rent for access to capital (of course, the Marxian view is that the surplus in production after workers are paid is a genuine surplus, and it should be divided among the workers rather than going to the capitalist).
As I continued to think about that, I became increasingly convinced that what was ethically bad about exploitation was not the differential labor exchange. To the extent that the relationship between capitalists and workers is unjust or exploitative, I came to argue that it was due to the unjustness of property relations, and this unjustness had its origins in Marx’s theory of primitive accumulation. In 1974, Robert Nozick published Anarchy, State, and Utopia, in which he admitted that the historical development of capitalism depended on plunder, murder, and conquest. He argued, however, that one could conceive of capital accumulation acquired through savings decisions people have made, and in that instance there would be nothing morally wrong about capitalism as an economic system. Both Cohen and I were very taken with that; we thought the argument had some truth to it, particularly given that Marx thought workers were the just owners of their labor power. If you’re the just owner of your labor power, why shouldn’t you be able to save if you want to, and then maybe in the future, hire other workers with your savings? In 1985 I published a paper titled “Should Marxists Be Interested in Exploitation?” And my argument in that paper was that it was not exploitation as such that was ethically bad, but the injustice of property relations underlying it.
maya adereth: I remember reading an interview with Eric Olin Wright in which he said the most consistent element in his thought has been that “the purpose of understanding the class structure of capitalism is to understand the conditions of transforming it.” Which aspects of Marx’s writing continue to influence your thinking?
jr: Historical materialism remains very attractive to me. The notion that economic structures change to accommodate advances in the forces of production is a grand insight, and I remain very inspired by Marx’s theory of history.
I wrote an article this summer titled “What is Socialism Today?” which articulates three models of a cooperative economic society. Socialism is back on the agenda in the United States, which is very exciting. Nevertheless, what people like Bernie Sanders and Alexandria Ocasio-Cortez are calling socialism is in fact just a series of economic reforms that don’t alter property relations. In the paper, I begin by proposing models of socialism other than social democracy, all written out as general equilibrium models, and I prove that you can have efficient resource allocations with very different (socialist) property relations in society.
I am undecided on what the most attractive model of socialism is from among those I propose.
ma: One of the things I wanted to ask you is about the relationship of effort and responsibility to universalist social demands like publicly provided healthcare. In Theories of Distributive Justice, you weigh the comparative responsibility of a hypothetical smoker over their condition against the environmental factors which may have led them to adopt an unhealthy habit. How do you reconcile the introduction of responsibility into a socialist call for publicly available benefits?
jr: I’ve thought about this quite a lot, and my answer to it is following: There is both a right wing and left wing application of my theory of equality of opportunity. The right wing view is that people should be rewarded according to their effort. The left wing view is that people should be compensated according to their disadvantaged circumstances. By far, the primary aspect of the theory is compensation of people for disadvantageous circumstances. The other part is not particularly important to me. I have an example of applying equality of opportunity to health care. In the example, there are two classes of people: poor people and rich people. Poor people suffer from cancer and tuberculosis, and rich people suffer only from cancer. The probability of contracting cancer or tuberculosis is a function of your lifestyle quality, and the distribution of lifestyle qualities is better among the rich than among the poor. So there is a distribution of lifestyle quality according to circumstance, but where in the distribution you put yourself is considered to be a matter of choice.
If you’re a utilitarian, you want to maximize the average life expectancy. That will lead you to spending much more on treating cancer than tuberculosis. However, if your goal is to maximize the life expectancy of the more disadvantaged group, you’ll spend much more on tuberculosis, even though you hold people responsible for their lifestyle. In fact, one rule which I insist on as a postulate is that when a person comes into the clinic with cancer or tuberculosis, nobody asks them what their lifestyle quality was. So, in the model, the amount you spend depends only on the disease, not on people’s lifestyles.
jh: When I first read Theories of Distributive Justice, I was blown away. It was the first time I’d seen someone provide a general, systematic framework for thinking about how consequentialist theories of justice can be constructed. Of course this is incredibly valuable from a scholarly perspective. But I’m curious as to whether you think formal theory and metatheory of this sort has a more applied role.
jr: The purpose of my work is to conduct an argument within academic economics for what justice and socialism consist of, and what is ultimately desirable for human society. I am carrying out a debate of ideas, and getting the ideas right is the most important thing I can do. The hope is that if you get the ideas right, you will attract other people to agree with them, and that shapes society at a broader scale. I’m not particularly worried about the application of these ideas because I’m sure if the ideas are right, there are going to be people who figure out how to apply them.
jh: One reason the applied question interests me is because of work in digital ethics. Decisions from predictive policing to municipal zoning are made based on algorithms and machine learning, and there’s a growing community of people thinking about algorithmic justice. Have you been thinking about that?
jr: Actually, I attended a conference at Harvard in May to talk about equality of opportunity to an audience of computer scientists. Again, I know nothing about their algorithms, but I hope that understanding this approach to equality of opportunity will shape the way people write them. This is an excellent example of where theory is of paramount importance: If you don’t have the right theory, you’re not going to have the right practice.
jh: You might be interested to know there was a paper presented at this year’s ACM FAT* conference that included a Roemerian test for algorithmic fairness. It seems like the scale and consistency at which these models can now be applied really raises the stakes for theory. I wonder if this new environment wherein mathematical models are implemented by algorithms actually begins to break down the division of labor and place the theory closer to the action?
jr: Yes, it seems like it would. But given that the stakes are higher, it should only strengthen our commitment to finding and disseminating the right theories. At the Harvard conference I learned that one of the results that computer scientists are rediscovering is the impossibility theorem, which we’ve known about for 70 years in economics.
jh: One last question along these lines is about the role of idealization and the limitations of models. Inevitably, formal models abstract and simplify. And sometimes it’s not obvious that the complexities that are smoothed over are unimportant, or that the results of the idealized theory will continue to hold, at least approximately, if those bits are added back in. For instance, it seems to me that one of the lessons of feminist theory is that a lot of the goods essential to human thriving are socially constituted—they don’t make sense outside of particular social arrangements. Gender expression is one example. If this is right, there is cause for concern about how much we can generalize from theories that presuppose a fixed set of goods. Do you have thoughts about that, or about the role of modeling generally?
jr: In terms of modeling, what economists do is take a theory they have, and try to reduce it to a model. In doing that, you have to make huge simplifications about the world you’re describing so that the model is tractable. The disadvantage of proceeding to the modeling stage is that you cut off imagination, because once you’ve got the model, you stop questioning its premises. On the other hand, the advantage of getting to the modeling stage is that you can check whether you’ve isolated the correct parts of the system—does the model imply what we observe? I don’t think there’s a general rule for deciding when you should proceed to a model, but I think models are invaluable.
People who don’t know how to model often think that we are just using the mathematics to communicate our ideas. For me, the most important aspect of modeling is the learning process. I’ll give you an example from General Theory: Marx had always maintained that the embodied labor times of commodities were ‘logically prior’ to prices—leading to the so-called transformation problem, of how labor values are ‘transformed’ into prices. I certainly believed that when I began working on the book. Initially, I took the Marxist definition of exploitation, whereby an individual is exploited if the labor that she expends in production is greater than the labor embodied in the goods she can purchase with her earnings from work. I also defined class by considering two factors of production: labor power and capital. An individual in my model could either sell their labor, work their own capital without hiring others, hire other workers, or any combination of the three. Given reasonable preferences over income and labor, I showed that an individual would occupy one of five class positions: a pure capitalist or landlord, who only hires others to work on her capital stock; a petty bourgeois or rich peasant, who hires others and works her land herself; an artisan, who only works her own land; a poor peasant or semi proletarian, who both works her land and sells her labor power; or finally, a pure proletarian or landless laborer, who only sells her labor power. Each one of these class positions is the result of optimizing in the standard (neoclassical) way, subject to one’s constraints defined by the amount of capital and labor one owns.
The theorem I proved—called the Class Exploitation Correspondence Principle—was the following: any individual who optimizes by hiring labor will necessarily be an exploiter, in the sense that the goods that he can purchase with his revenues will contain more labor embodied than he himself expended in production. Conversely, any individual who optimizes by selling labor power must be exploited. In this way I proved as a theorem, from primitives, what Marx had taken as a tautology.
At this point I made the technological premise of the model more complex by assuming multiple production possibilities. It turned out that the theorem wasn’t true if you maintained the classical definition of exploitation. If you defined “labor embodied” as “the minimum labor needed to produce goods, given your technology,” the theorem was false. However, if you defined “labor embodied” as “the minimum way of producing goods using only processes which were maximally profitable at equilibrium prices,” then the theorem was true. That meant that in order to preserve the Class Exploitation Correspondence Principle, I had to define labor values as depending on equilibrium prices! I could no longer maintain the view that labor values were logically prior to prices. That is an example of how modeling can teach you something that you couldn’t see with the naked mind.
jh: It seems like you’re engaged in a similar method in your recent work on cooperation and Kantian optimization.
jr: Recently, I’ve argued that that it’s incorrect to define capitalism and socialism as simply differing in property relations. There’s an old tradition that says ethos matters: capitalism has an individualistic ethos, socialism has a cooperative ethos. But nobody has ever modeled that. Using Kantian optimization, I can embed a theory of cooperation into models of the economy. In fact, I found that there’s a real difference between what happens in the economy when people optimize in the cooperative manner, rather than the Nash manner. So I’ve enriched the model of socialism by embedding in the model a formal protocol of cooperation. For instance, in social democracy, where we still have private distribution of firm ownership, but workers optimize their labor supply decisions in the Kantian protocol as opposed to the Nash protocol, one can have redistribution through income taxation to any degree, and regardless of the tax rate, the equilibrium is Pareto efficient.
Using Kantian optimization, I’m therefore able to separate the issue of equity from the issue of efficiency. Under capitalism, if you tax people at a rate of one—so that everybody pays their entire wages in taxes, but receives back the average product as his income—nobody will work. In my model of social democracy, you can tax people at a rate up to and including one, and you still get a Pareto efficient allocation. Again, I made advances through the modeling process that I wasn’t able to see with the naked mind.
What’s an example of a model that is more socialist than that? Well, one model is what I call worker management. The key idea here is that socialism as we conceive of it in the United States today cannot be socialism as Marx conceived of it, because when Marx was writing, there was virtually no middle class. Today, people between the 50th and 99th percentile of the income distribution own 56% of the wealth in the United States. Two percent is owned by the bottom half of the wealth distribution. The other 42% is owned by the top 1%. If socialism is going to come about democratically, it can’t involve confiscating all of the middle class’s wealth. So one formula for socialism is that the entire product of firms should be distributed to investors and workers in proportion to the labor and capital they’ve invested, leaving no profits. More generally, in an arrangement like this, you have to design a way of paying people that: (i) can be decentralized, (ii) exhausts the whole product, and (iii) is Pareto efficient.
There’s also a worker-managed model in which investors are paid a competitive interest rate and the profits that remain after interest is paid are divided among workers in proportion to their labor expended. I have not concluded as yet the extent to which suppliers of capital (that is, middle class people who invest their savings) should share in the economic surplus.
Societies will have to choose how much to remunerate capital. As I said, if we confiscate savings, who is going to advocate socialism? Those who occupy the 50th to the 99th percentiles of the income distribution are not the enemy. This seems to be a big intellectual puzzle for socialists: today, we have a very large middle class which hasn’t accumulated its wealth through robbery, plunder, and murder. They have accumulated capital and savings through processes that are unjust because of unequal opportunity. Our message has to be that under socialism we want to eliminate unequal opportunity—in particular, inheritance of large amounts of wealth—but people should not be discouraged from saving, and hence investment will bring a return. The growth of significant middle class wealth since 1850 (what Piketty calls the “patrimonial middle class”) means that socialism can no longer be thought of as a system in which the economic product is distributed according to labor expended.
jh: I understand why forms of Kantian optimization are Kantian in spirit, in the sense that this universalization procedure involves thinking: what would happen if everyone changed their behavior in the same way I’m considering? But one feature of this principle is that the results are going to strongly depend upon what counts as “the same way.” That’s exactly what leads to the difference between, for instance, additive and multiplicative Kantian optimization; if I’m working two hours instead of one, should I be imagining a world in which everyone worked one more hour (additive) or one in which everyone worked twice as much (multiplicative)? Is there an easy intuition as to why a particular Kantian variation would emerge in a given context, or to what ethical and social norms the different variations capture?
jr: My view is that the most prevalent kind of Kantian optimization is what I call simple Kantian optimization, where an individual takes the action she would like all others (in the given situation) to take. In a strictly monotone game which is symmetric, where the producers are all similarly situated, simple Kantian optimization is always Pareto efficient. It ‘resolves’ the two negative consequences of the Nash equilibrium: the free rider problem and the tragedy of the commons.
The question next becomes whether the model is generalizable to asymmetric game. The two main generalizations are multiplicative and additive Kantian equilibrium. Here, instead of thinking about “the same action,” we should think about vectors of contributions. Global emissions are a good example of this; if no nation would like to rescale the whole vector, by any constant—increase everybody’s emissions by 7% or decrease everybody’s emissions by 3%—that produces a multiplicative Kantian equilibrium, and it’s always Pareto efficient. But re-scaling is not the only thing you can do. Another thing you can do is translate the vector by a constant: would everyone like to add one gigaton to the emissions of each country? If given the possibility of translating, everyone would rather stand pat, then that’s an additive Kantian equilibrium, and that turns out to always be Pareto efficient also.
jh: Have you looked at the stability of Kantian equilibria when people deviate from Kantian optimization?
jr: We can conceive of a situation with some Nash players and some Kantian players. If the Nash players optimize in the Nash way and the Kantian players apply Kantian optimization among the group of Kantian players, holding fixed the other group’s behavior, you reach an equilibrium. But the equilibrium won’t be Pareto efficient, and it will deviate from efficiency in proportion to the number of Nash players, roughly speaking.
ma: How do you think the nature of political coalitions has changed in the United States? And how are automation and new forms of labor relations feeding into this realignment?
jr: Given capitalist property relations, automation may hurt a lot of people. There’s no question that it’s economically feasible to design a way of using these technologies so that everybody’s life improves. Automation should be a good thing—it should liberate people to engage in activities other than work—in particular, other than dull work. Society must invest in more education, so people can learn how to use additional free time in stimulating ways. In Keynes’s 1930 essay “Economic Possibilities for our Grandchildren,” he wrote that by 2030 the work week was going to be 12 hours long, and the big problem would be how people would occupy themselves. He wasn’t worried about mass poverty—he thought that society would arrange to share the work reasonably and everybody would have a decent income. (Of course, Keynes underplayed the reality that serious class struggle would be required to implement those desirable social arrangements.)
Now, will a capitalist society learn how to use these new technologies in a socially fair way? Not unless there are mass social movements that demand that the gains of automation are shared. I think this will mean transforming what Marx called the relations of production—just as the industrial revolution transformed the relations of production from the era of household and artisan production. This is the theory of historical materialism in practice: advances in the technology (the forces of production) bring about changes in property relations. The midwife that facilitates the birth of the new property relations is class struggle.
To get all future Phenomenal World posts directly in your inbox, sign up here.