Deliberate Public Goods Games
An Opportunity to Play "Public Goods Games" in a Deliberate and Measurable Fashion
A Casual Introduction to Public Goods and Collective Action Problems
Across many domains of life, we're all constantly entangled in a variety of "public goods games". In a Stanford Encyclopedia of Philosophy article on Free Riding, Hardin and Cullity highlight quite a few examples of public goods that you may be familiar with, and perhaps even a big fan of, including clean air, public safety, concessions won by labor unions, and highways [HC03].
As another simple example, consider a local public park. There might be some rules ("Closed after dusk!"), and occasionally the locals might try to throw out a rowdy group of teenagers, but for many cases the park is effectively open to everyone. For the sake of a running example, let us imagine the park is also big enough that one person's "consumption" (i.e. spending time in the park) does not ruin the experience for others.
Looking to a technical definition from economics, a public good is a good that is available to everyone ("non-excludable") and for which one person's consumption does not affect another person's ("non-rival") [C08]. In practice, many goods, including our imaginary local park, roughly meet these criteria, and thus are subject to "collective action problems". It is important to note that some goods that seem like they might be "public goods" are in fact provided privately through markets (e.g. firefighting services in someplace [C08]). Whether the choice to provide certain goods privately actually creates better outcomes remains a case-by-case debate. Nonetheless, the point remains: we're surrounded by public goods games!
Early theorizing about public goods and collective action suggested a bit of puzzle: when goods are both non-excludable and non-rival, there tends to be a problem of incentives. If nobody can prevent me from enjoying the local public park, what's to stop me from littering in the park but continuing to visit it? Taken to the extreme, why doesn't everybody litter in the park, ruining its utility?
Despite this "dismal logic of collective action" [HC03], we enjoy a surprising amount of public goods, including some of the examples above (with other examples like clean air being an ongoing struggle). And with the advent of new opportunities to use technology to facilitate collective action, there may be more opportunities to draw on a large body of knowledge about the production of public goods.
Below, I'll briefly introduce how some scholars have examined public goods with the lens of game theory, and then focus on two particularly influential papers from sociology. Ultimately, I hope to convince you that Govrn's Outcome-Based Donations (OBDs) [G21] and similar types of online collective action provide a very real possibility that we might be able to participate in public goods in a way that is deliberate (you know you're playing this game) and measurable (you can actually reason quantitatively about how you want to play the game). This means we can all (collectively) benefit from the immense body of work that has sought to understand and improve the production of public goods.
The Game Theory of Public Goods
Many authors have explored questions about public goods using the mathematics of game theory. Such explorations generally start with some assumptions — "Let's assume people are rational; let's assume people measure 'utility' in this particular manner" — and follow these assumptions to their logical conclusions about whether or not a particular public good will be produced. So we might model the "convenience utility" somebody gets when littering in the park (instead of using a bin) relative to the "enjoyment utility" people receive when they visit a litter-less park. This might create a game, in which each individual makes a careful calculation to choose if they will litter.
There can even exist "second-order games" (or in other words, a game on top of a game) around the choice to "sanction" other people [H96]. If you saw someone littering in your favorite public park, would you glare at them? If so, you'd be participating in a second-order game to produce the "secondary public good" of a "sanctioning system". In other words, by helping to punish people who defect, you're helping to create a set of incentives that result in more public goods (in this case, less litter).
“Looking to a technical definition from economics, a public good is a good that is available to everyone ("non-excludable") and for which one person's consumption does not affect another person's ("non-rival") [C08].”
Littering in the park may seem like a bit low stakes compared to keeping the air clean or improving global labor conditions. But even in this simple example, things are already getting complicated. You might be wondering: is it really reasonable to model people's littering decisions as some kind of rational optimization problem, with potentially many layers of games being played?
In other cases, the mechanics of a public goods game can be quite opaque, to say the least. It seems far-fetched to say we might make political donation decisions conditional on the expectation that a particular donation will have a precise marginal effect on some politician's chances of success. For instance, you might be shocked if I claimed "I will donate $100 to this cause if and only if my donation will improve the likelihood of success by 0.1%". To use the words of Kollock, "No one will be fired and no program will go off the air if I do not send in a $30 contribution to public television, and even if I do conserve water in a drought, it will have no measurable impact on the overall situation" [K98].
If you try to start calculating the likely behaviors of other people, the math to estimate these kinds of marginal effects becomes very complicated very quickly! Despite this, we generally don't find ourselves in computational gridlock; we identify public goods we're interested in (e.g. a litter-free park, cleaner air, a particular politician's agenda) and make a decision using heuristics and norms (for instance, we might not litter because someone told us it's the "right" thing to do, an instance of "moral suasion" [K98]). Indeed, later work suggested it might be better to model real-life decision-making as some kind of "social learning", not as a swarm of bot-like optimizing people unleashed upon the world (or the park) [M91].
Ok, so maybe it's not worthwhile to try to create a hyper-accurate model of park-littering agents (validating such a model would likely require surveilling the park, to boot). When a park does have a littering problem, locals (or city officials) might be able to solve it with much simpler interventions (e.g. re-allocating trash cans to lower the relative "convenience reward" for littering, or cultivating a sense of local community so visitors feel more invested in the park's cleanliness).
We might not want to model our local park; but what about Modeling Outcome-Based Donations?
Still, a body of highly influential sociological research suggests that trying to understand the specific nature of a public goods problem can be enormously helpful in figuring out strategies for coordination, thus making it more likely the public good will be produced. This means, while the costs of trying to carefully build a computational model of park limiting outweigh the benefits, for other collective action problems such modeling may be very worthwhile.
In the rest of this post, I want to briefly introduce two influential papers on collective action, and their exciting implications for Outcome-Based Donations and the Govrn platform [G21] (TLDR: constituents, community leaders, and experts come together to propose specific outcomes for politicians to achieve, and constituents make donations that are conditional on these specific outcomes).
Oliver, Marwell, and Teixeira’s Three Independent Variables for Public Goods Production
In a 1985 paper, Oliver, Marwell, and Teixeira suggest three important "independent variables" that influence the likelihood that a public good will be produced [OMT85]. If we have knowledge about these variables for a particular case of collective action, it's possible to intervene to make success more likely.
The three variables are:
The shape of the "production function". Specifically, how does the likelihood of success relate to the amount of resources contributed?
The distribution of interest in producing the good. Does each member of the group interested in the public good desire the good with equal intensity, or are there some people with high or low levels of interest?
The distribution of resources needed to produce the good. Does everyone have an equal amount of resources (e.g. money, time, etc.) to contribute?
On the left: an “accelerating” curve that begins flat, but ends nearly vertical. On the right, a “decelerating” curve, also known as “diminishing returns”.
The production function is particularly important to account for. Oliver, Marwell, and Teixeira focus on the difference between an "accelerating" function (the likelihood of success doesn't start to ramp up until many people have already contributed) and a "decelerating" function (as people contribute the additional likelihood of success gained from new contributors starts to "flatten off"). A decelerating production function occurs in any case in which there are diminishing returns from contributions. For a quick visual example, see the two very simple illustrations above: the left is an accelerating curve, and the right is a decelerating curve. Another simple way to think about the difference is that an accelerating curve is flat early on (low participation) and a decelerating curve is flat towards the end of the curve (high participation).
When the production function is decelerating, the main problem is the temptation to be a free-rider. If other people will contribute anyway, why bother? Your (late-stage) contribution barely has an impact.
When a production function is accelerating, the main problem collective action faces is not free-riding, but instead the problem of gaining critical mass. As Oliver, Marwell, and Teixeira note, "The usual outcome of the accelerative collective dilemma is that nobody rides free because nobody contributes and there is no ride" [OMT85]. In other words, the perceived reward for early contributors is so low that it's never rational for anyone to start contributing at all, unless there are a group of activists who "get the ball rolling" by providing a critical mass.
Armed with knowledge about these variables (and ideally, a few other things, such as the nature of any social networks that connect individual contributors), we can use the lessons from the many scholars that have built on this work to provide suggestions that make collective action easier.
Kollock's "Anatomy" of Cooperation and Three Types of Solutions to Public Goods Problems
Later, in a 1998 paper, Kollock introduced three kinds of solutions to public goods problems [K98].
Motivational solutions try to change the way people see a game. For instance, framing a problem in terms of group identity could cause an individual to become more altruistic ("I don't want to litter in the park because members of my in-group also use this park, and I care about their utility").
Strategic solutions try to help actors shape outcomes without changing the structure of the game. A famous example is the use of "tit-for-tat" to play the "Prisoner's Dilemma". One strategic solution that's particularly relevant to OBD is that of "grim triggers" — each person agrees to cooperate if and only if everybody else does. While implementing such grim triggers is now easier with the advent of new tools, older work suggests this rarely works out in practice [OGW+94] . Thus, this suggests a powerful but unproven strategic solution.
Finally, structural solutions try to change the nature of the game. For instance, perhaps the park requires visitors to sign in — it's now more likely that litterers will be identified. Alternatively, perhaps the city officials can introduce a step function reward: if less than 10 pieces of litter are found on Sunday, we will provide a free picnic.
Connecting this all to OBDs
Above, I made a (brief) case that it is more or less totally fine that nobody is trying to create careful quantitative models to describe the collective action that goes in our local parks. This is in part because we would impose potentially serious costs: the need to build surveillance systems, the need to hire teams to maintain and analyze data, etc.
However, when it comes to platforms like Govrn, there exist many exciting new opportunities to put into practice the concepts and findings from these highly influential social science papers (and the subsequent body of work). Why is Govrn different than our theoretical park? By nature of being an online technology, Govrn (or similar platforms) is naturally amenable to measurement. Every donation is, by definition, recorded! More pessimistically, we might say that online technologies often enable the elements of surveillance which are a prerequisite for measurement and modeling baked in. This is certainly cause for concern!
However, Govrn has a relatively focused scope: civic engagement. At a park, you may want to have a picnic with your family or a deep conversation with a friend. The ability to do so is part of the "good" the park provides! This isn't to say that a platform like Govrn is entirely free from concerns about privacy or that people's contributions to public goods are being "surveilled", but it does take a huge step in the right direction. Quantifying public goods games makes much more sense in a context that's deliberately aimed at producing public goods. Of course, there may be cases in which people participating in platforms like Govrn want to push back on choices about measurement and surveillance, and navigating this tension will likely require an ongoing dialog between participants.
Focusing on the benefit of online collective action, the low cost of measuring contributions makes it possible to model production quite accurately. Critically, we really can try to figure out what the "production function" looks like for certain outcomes. Part of the reason this is so valuable is to identify "accelerating" vs. "decelerating" cases. As more people contribute, does the likelihood of success start to ramp up even faster, or does it start level off? This knowledge could help experts and organizers design targeted interventions to increase the likelihood of success, like implementing threshold-based rewards in the accelerating case (to help get over the early "low rewards" period).
Furthermore, it may be possible to measure and communicate the degree to which interest in a particular public good is distributed amongst Govrn users. Oliver and Marwell provide one example of why this is useful: if there are distinct groups of "very interested" people and "uninterested" people, a tactic like fund matching ("if you donate $5, I'll donate $5") can be a good way for the interested people to motivate less interested people (this would likely fall into Kollock's "strategic" bucket).
Looking to Kollock's work, Govrn actually provides opportunities to implement all three types of "solutions" to public goods problems. Govrn exists as a platform with very deliberate design choices (just see the community-written white papers!), which means design interventions could be made within the express intention of restructuring the "game". On top of this, individual actors within Govrn can engage in motivational solutions.
All that goes to say, Outcome-Based Donations offer an exciting opportunity to participate in the production of public goods in a very deliberate and transparent way. Of course, this doesn't mean we can just look to prior work to gain all the answers — I expect future work on Govrn will continue to produce new knowledge about collective action — but it does mean there is a wealth of lessons to draw on.
Ultimately, this post is more speculative than even the discussion section of a research paper, and I don't intend to suggest that one particular strategy from one of these papers represents a final word on the matter of public goods. Rather, I hope I may have convinced you that Govrn and the OBD concept provide a very exciting chance to engage in public goods game in a very deliberate way, and that this could be a very good thing for the production of more public goods! Looking forward, Govrn presents an exciting opportunity to generate new knowledge about collective action.
[C08] - Cowen, T. (2008). Public goods. The concise encyclopedia of economics, 197-9. https://www.econlib.org/library/Enc/PublicGoods.html
[G21] - Govrn team. https://www.govrn.io/how-govrn-works
[HC03] - Hardin, Russell and Garrett Cullity, "The Free Rider Problem", The Stanford Encyclopedia of Philosophy (Winter 2020 Edition), Edward N. Zalta (ed.), URL = https://plato.stanford.edu/archives/win2020/entries/free-rider/
[H96] Heckathorn, D. D. (1996). The dynamics and dilemmas of collective action. American sociological review, 250-277. https://doi.org/10.2307/2096334
[K98] Kollock, P. (1998). Social dilemmas: The anatomy of cooperation. Annual review of sociology, 24(1), 183-214. https://doi.org/10.1146/annurev.soc.24.1.183
[M91] Macy, M. W. (1991). Chains of cooperation: Threshold effects in collective action. American Sociological Review, 730-747.
[OMT85] Oliver, P., Marwell, G., & Teixeira, R. (1985). A theory of the critical mass. I. Interdependence, group heterogeneity, and the production of collective action. American journal of Sociology, 91(3), 522-556. https://www.journals.uchicago.edu/doi/abs/10.1086/228313
[OGW+94] - Ostrom, E., Gardner, R., Walker, J., Walker, J. M., & Walker, J. (1994). Rules, games, and common-pool resources. University of Michigan Press.