Yves here. I suspect some readers will try invoking Thomas Schelling to debunk this post. We addressed his famous “tipping point” theory in ECONNED:
Indeed, appealing or impressive theories are often accepted without being validated. In 1971, highly respected economist, game theorist, and future Nobelist Thomas Schelling published an article, “Models of Segregation,” which set forth the concept called the “tipping point.” Schelling developed an elegant analysis using coins of two kinds placed on a game board to simulate mixed-race neighborhoods. He demonstrated that it would take only a small proportion who preferred living with people of the same race to lead to a series of moves that would produce racial segregation. Each set of departures would leave some of the people who remained uncomfortable with the new neighborhood mix, precipitating more departures.
Aside from being a clever and novel approach, Schelling’s explanation may have become popular for darker reasons: after a period of protracted white flight from decaying inner cities, it suggested that perhaps most people weren’t really all that prejudiced; it took only a few bigots to produce ghettos.
Although the theory seemed obviously true in 1971, more recent work by New York University professor of economics William Easterly has found that Schelling’s predictions were for the most part not borne out. Easterly tabulated census tract data from 1970 to 2000 for metropolitan areas and found that whites had departed neighborhoods that were mainly white to a greater degree than they had mixed-race neighborhoods. Easterly did stress that his findings were only a single check of the theory over a particular time frame. But his analysis still serves to illustrate how appealing stories are too often accepted as received wisdom.
By Klaus Desmet, Altshuler Professor of Cities, Regions and Globalisation, Southern Methodist University. Joseph Flavian Gomes, Assistant Professor, Navarra Center for International Development, University of Navarra, and Ignacio Ortuño-Ortin, Professor of Economics, Universidad Carlos III de Madrid. Originally published at
Diverse countries tend to have more conflict, lower development, and worse public goods, possibly due to antagonism between groups. Based on recent research mapping local linguistic diversity across the entire globe, this column argues that local interaction with people of other ethnolinguistic groups can mitigate the negative effect of overall diversity on a country’s outcomes in health, education and public goods. This finding lends support to policies that influence the local mixing of ethnolinguistic groups.
Ethnocentrism is ‘in’, and multiculturalism is ‘out’, in many Western democracies. The result of the 2016 presidential election in the US and the victory of Brexit in the UK partly reflect a growing unease among the electorate about living in societies that are increasingly diverse. Likewise, in continental Europe, the refugee situation in Germany is bound to become a central theme in the 2017 national elections. The National Front in France is also likely to benefit in the 2017 presidential election by relating home-grown terrorism to diversity.
Although diversity might be increasingly looked upon with suspicion, paradoxically it is often viewed more negatively in relatively homogenous places than in areas that are already highly diverse. In the Brexit referendum, the perception that there were too many immigrants was especially strong in areas with few foreign residents, and much less so in cosmopolitan London: 85% of UK districts with a lower-than-average share of foreign-born residents voted in favour of leaving the EU, compared to only 44% of the other districts (Lawton and Ackrill 2016). Past US presidential elections showed the same pattern. President Trump’s America-first rhetoric found little echo in the regions of the US with most undocumented immigrants.
These observations are consistent with theory. Although individuals may feel antagonism towards other groups in society, that prejudice is less strong if they interact with these groups in their daily lives (Allport 1954). At face value, this suggests that antagonism between groups in the UK would be minimised if every town mirrored the country’s overall diversity. It is not clear, however, that we can say this relationship is causal, or if it generalises to the whole world.
In fact, not everyone agrees with theory. The proponents of conflict theory argue the exact opposite: interaction with individuals of other groups is costly and generates greater antagonism. Empirical evidence is inconclusive on which theory prevails.
One reason we should care whether local interaction mitigates or reinforces antagonism is that diverse countries tend to have more conflict, lower development, and worse public goods, and this antagonism would be an explanation. In our recent work, we develop a global database of local language use to investigate how local interaction changes the impact of a country’s overall ethnolinguistic diversity on a country’s public goods outcomes in health, education and infrastructure (Desmet et al. 2016). If it were to mitigate the negative effect of overall diversity, we would interpret this as evidence in favour of theory.
The Theory of Local Interaction, Local Learning and Antagonism
Our starting point is a simple framework to measure a country’s antagonism. Suppose that an individual feels antagonism towards another randomly chosen individual in his country if they belong to different ethnolinguistic groups. Averaging across all possible random matches yields an antagonism measure that corresponds to the standard ethnolinguistic fractionalisation index – the probability that two random individuals of a country speak a different language. For example, if we take two Belgians at random, there is a 54% chance that they have a different mother tongue.
So far we have not taken into account local interaction. Now assume that the antagonism an individual feels towards someone from another group in his country is affected by the amount of local interaction he has with people from that group. For example, the antagonism of a Dutch-speaking Belgian towards all French-speaking Belgians in his country depends on how much he locally interacts with French-speaking Belgians. We refer to this additional effect as local learning.1 Depending on whether theory or conflict theory dominates, local learning can either mitigate or reinforce the existing antagonism as measured by fractionalisation.
Note that not all local interaction leads to the same amount of local learning. For example, if a Dutch-speaking Belgian interacts with an Italian-speaking Belgian locally, this will not affect overall antagonism much if there are few Italian-speaking Belgians in the rest of the country. In this example, there might be a lot of local interaction, but not much local learning. Hence, what matters for antagonism is the amount of effective learning that occurs because of local interaction.
The Geography of Diversity
Next we measure fractionalisation and local learning in the data. Combining detailed maps from Ethnologue on 6,905 unique languages spoken, and population counts at a fine geographic resolution from Landscan, we created a global database on local language use for each 5km-by-5km grid cell in the world.
This allows us to compute fractionalisation (Figure 1) and average local learning (Figure 2) for all countries. There are many interesting differences. For example, fractionalisation is much higher in Chad than in the Central African Republic, but the reverse is true for average local learning. This is a result of more local mixing in the Central African Republic, as can be seen in Figure 3, which shows local learning for each 5km-by-5km grid cell.
Another example may be useful to clarify the difference between overall fractionalisation and local learning. While fractionalisation is virtually identical in Guatemala (0.53) and Mauritius (0.52), local learning is much lower in Guatemala. In Guatemala indigenous language speakers are concentrated in the central and northwestern highlands, and have limited with Spanish speakers. In contrast, Mauritians “switch languages according to the occasion in the way other people change clothes” (Chiba 2006). As a result, our index of local learning was much higher in Mauritius (0.20) than in Guatemala (0.06).
Of course, how local learning affects a country’s overall antagonism depends on whether theory or conflict theory is a better explanation. If theory is correct, then local learning mitigates the antagonism that comes from fractionalisation. In that case, antagonism would be lower in Mauritius than in Guatemala. The reverse would be true if conflict theory were the dominant force.
Figure 1 Ethnolinguistic fractionalisation
Figure 2 Average local learning
Figure 3 Local learning
Health, Education and Infrastructure
Empirical research has found that ethnolinguistic fractionalisation tends to worsen public goods outcomes (La Porta et al. 1999, Alesina et al. 2003, Desmet et al. 2012). One interpretation is that antagonism makes it hard for a diverse society to agree on public goods. Different groups may fight over which language to use in education, the shape of the road network, or where to put the nation’s main hospitals.
Our discussion suggests that local learning may mitigate or reinforce the overall antagonism coming from a fractionalised society. Hence, we should take the degree of local learning into account when exploring the link between diversity and public goods outcomes. This is what we do in Desmet et al. (2016). Holding overall fractionalisation constant, we find that local learning improved a wide variety of public goods outcomes in health, education and infrastructure.
These results lend support to theory. The effects are large. For example, a one-standard-deviation increase in local learning lowers child mortality by 7.4 per thousand live births. To put this figure into perspective, in its effect on child mortality, a one-standard deviation increase in local learning is equivalent to a 61% increase in GDP per capita.
Before jumping to policy conclusions about local mixing, reverse causality is potentially a concern. In societies with poor public goods, individuals from the same linguistic group may prefer to cluster geographically to support each other. If so, this would lead to reverse causality, with public goods outcomes affecting the spatial sorting of individuals of different linguistic groups. To address this concern, we use an instrumental variable approach following Alesina and Zhuravskaya (2011). This allows us to conclude that there is a causal positive effect of local learning on the quality of public goods. Overall, theory trumps conflict theory.
Going back to our earlier discussion, we can conclude that making each town mirror a country’s overall diversity would improve public goods outcomes. Although in most countries governments do not tell people where to live, there are many policies that would influence the local mixing of ethnolinguistic groups. European governments commonly use social housing to geographically spread ethnic minorities, making neighbourhoods and cities more equal in their diversity. Singapore, where more than 80% of the population lives in public housing, has a quota system ensuring that each housing block resembles the nation’s ethnic make-up. In a different setting, theory was also an important argument in the US Supreme Court case Brown v. Board of Education, which led to the racial desegregation of public schools (Putnam 2007). Of course, these policies would be controversial, because they trade off individual freedom of choice with desirable social outcomes.
See for references