The Mixed Street

In a poor part of Kampala, streets where neighbours share a background keep their drains clear. Mixed streets often don't. Why?

About five minutes. You play the researcher. Every result is from a 2007 study in which 300 neighbours played money games.

The five suspects

Pick a test

Each test is one of the study's games, played with people from the same group and from different groups. Which tests would tell these suspects apart?

the player's own groupanother group

Your hunch first. Then the study's result.

Your calls

For each suspect: is it part of why mixed streets fail? A call counts as proven only if it matches the study and one of your tests backs it.

Suspect by suspect

How to play

The setting. Mulago-Kyebando is a poor, mixed neighbourhood of Kampala, Uganda. Drains, rubbish collection and night patrols there depend on neighbours chipping in. Streets where most people come from one ethnic group do better than mixed ones. In 2005 a team of researchers set out to find why.

The suspects. Five possible reasons, each one a serious idea in the research: groups want different things; people are kinder to their own; same-group teams work better; people can find their own group; people are watched by their own group and punished if they don't help.

The tests. Seven games that the study ran with 300 randomly chosen neighbours. Sometimes the players came from the same group, sometimes not. Teal always means the player's own group, ochre another group. You can afford five tests. Before each result you give your hunch. Hunches don't count towards your score; they just make you commit.

Reading a result. Two bars: what happened with the player's own group, and with another group. A tag says whether the study found a clear difference, a weak one, or one that could be chance. "The stingiest 40%" are the players who kept everything they could in a separate game. The study looked at them separately because they are the ones a street most needs to win over.

The calls. When you're done, call each suspect: guilty, not guilty, or can't tell. You can go back to the lab while tests are left.

The yardstick. Proven calls, out of 5. A call is proven if it matches the study's conclusion and a test you ran backs it. A right guess with no test behind it doesn't count.

What it shows. How a well-designed comparison separates explanations that all predict the same thing: mixed streets doing worse.

About the paper

Habyarimana, J., Humphreys, M., Posner, D. N. & Weinstein, J. M. (2007). Why Does Ethnic Diversity Undermine Public Goods Provision? American Political Science Review 101(4): 709–725. doi:10.1017/S0003055407070499

Fidelity: statistical, from reported estimates only

Nothing in this game is simulated and nothing is random. Every bar shows a number the paper reports, or one computed directly from its tables. Where the paper reports only the gap between same-group and mixed play (the dictator and prisoners'-dilemma games), the two bars are placed symmetrically around the reported average; the gap is the paper's, the exact heights are ours.

It demonstrates the paper's pattern of results and the logic of its design: which game separates which explanation from which. It assumes our plain-words versions of the five mechanisms, our five-test budget and our rule for which tests "back" which verdict (taken from the paper's own use of each game). The paper establishes how 300 randomly sampled residents of Mulago-Kyebando played these games with co-ethnics and non-co-ethnics, and argues from that to why diverse communities provide fewer public goods. The games do not measure drains or patrols; that link is the paper's argument.

What the paper hands over

Estimates from a field lab in Kampala, and a taxonomy. Three families of mechanisms, five in all: preferences (commonality of tastes; other-regarding preferences, i.e. altruism towards one's own), technology (efficacy, i.e. working well together; findability, i.e. being reachable through social networks) and strategy selection (a norm of reciprocity among co-ethnics, enforced by sanctioning). Each game isolates one of them. Players were told only their partners' photos; co-ethnicity here is the paper's "benchmark" coding, from self-reported group.

The discovery

Identification. All five mechanisms predict the same thing: mixed communities do worse. The paper's design tells them apart. When the giver in a dictator game is anonymous, co-ethnics get no more (Tables 2–3): no in-group altruism. When the giver is seen, the "egoists" (the 40% who kept the larger share in every round of a separate game) give co-ethnics 36 USh more of a 1,000 USh pot (Table 6); a third-party enforcer erases the co-ethnic edge in cooperation and punishes co-ethnic defectors more often (Table 9). Co-ethnics find each other more often (Table 5). Tastes do not differ by group (Table 1), and co-ethnic teams are not clearly more productive (Table 4).

The game

You are the researcher with a budget of five of the paper's seven games. Each result is the paper's. At the end you call each mechanism; a call counts as proven only if it matches the paper's conclusion and you ran a game that backs it. The scoring is ours. One plan proves all five; it is the paper's own set of tests. The prisoners'-dilemma pair is a second route to the sanctioning verdict, but it cannot clear altruism on its own, because the paper uses the anonymous dictator game for that.

What is cut

Region-based and subjective (perceived) co-ethnicity, which give similar or stronger results (for example, with perceived co-ethnicity seen egoists give 52 USh more instead of 36); the 500 USh "discrimination" variant of the dictator game (anonymous: no favouritism; seen: the co-ethnic is favoured 12 points more often); the lockbox game (mentioned in the puzzle result); group-by-group survey estimates; all regression controls and fixed effects.

Divergences

Per-type averages in the coin games are derived from the reported overall average and the egoist coefficient. The prisoners'-dilemma gaps use the paper's subjective co-ethnicity, the only coding it reports for that game. The paper's verdict on teamwork is "only weak evidence", so "not guilty" and "can't tell" both count as right. "Watched by their own" and "Easier to find" are both called guilty: the paper names the first as the main mechanism and reads the second as what makes sanctioning possible.

For the author

See the author note in the repository. This game has had the builder's review only; it has not been tested with real players.

Game built from this paper; the paper is the authority where the two disagree.