Micromotives and Macrobehavior T.C. Schelling (1978)


T.C. Schelling won the Nobel Prize in 2005 for his contribution to the understanding of conflict and cooperation via game-theory analysis. His ideas on these topics are discussed in The Strategy of Conflict (1960) and Arms and Influence (1966). In 1969 and 1971 he published articles devoted to racial dynamics and segregation originating from people’s preferences for neighborhoods. In Micromotives and Macrobehavior T.C. Schelling provides generalized models describing how people making individual choices could come to the particular equilibrium distribution in place of living, school attendance, etc.

The first chapter starts with an example of people choosing seats in the auditorium skipping the first rows. Here authors discuss that such behavior could be driven by various individuals' motives. Then, the chapter provides a description of the market and non-market social interaction and introduces the concept of equilibrium analysis that will be used throughout the book.

In the second chapter, T.C. Schelling elaborates on patterns that do not depend on the individual choices, such patterns are realized due to the characteristics of the aggregate as a whole: there always will be people who will be standing in the musical chairs game.

The third chapter discusses the usage of models in social science and comprises several models that can be applied to various topics: self-fulfilling prophecy, critical mass, the commons, the market of lemons, acceleration principle.

— Self-fulfilling prophecy stands for situations when expectation induces behavior that cause the expectation to be fulfilled. The basic examples are run on banks lending to banks’ insolvency or inflation. There are several types of self-fulfilling prophecy: self-equilibrating expectation (if we believe that everyone will bring food, we might turn around and bring drinks), self-confirming expectation (if smokers believe that mentholated cigarettes are in green packages, manufacturers may use such package), self-displacing prophecy (if everyone expects same average behavior, they will displace the average from where they thought it would be), self-negating prophecy (if everyone believes that an event is overcrowded and stays home, it would not be overcrowded), etc.

— Critical mass could be exemplified by the contrast between a research seminar that fades after several meetings and a volleyball game gathering enough people to maintain its existence. T.C. Schelling defines two subclasses of critical mass phenomena: tipping and lemons. Tipping is a concept that manifests itself in migration processes, occupation, colleges, public beaches, etc. It describes how people react to new entrants, for instance, the coming of a few members of a minority can cause the departure of a formerly homogeneous population, which frees new space for more minority members. Thus, we could obtain a new allocation of households. In the lemons market described by Akerlof, the market could disappear without the introduction of institutional arrangements like guarantees.

— Commons is a paradigm referring to the situation in which people impinge on each other in pursuing their interests, but they could be better off collectively if they could restrain. Nevertheless, no one gains individually by self-restraint. One example is water-saving in summer.


In the next two chapters, the author turns to the issues of mixing and segregation both on discrete characteristics like race or language and continuous ones - age, income, etc. Here we have a model explaining how people organize themselves depending on the proportions of different groups and preferences on the neighborhood's composition.

The sixth chapter looks at the hypothetical situation of the possibility to choose genes for kids. Parents' individual choices lead to the shift in demographics for the society on a macro level. The possible government intervention aimed at the return to balance is discussed.

The concluding chapter discussed uniform multi-person prisoner’s dilemma showing how different equilibrium could be achieved under different preferences conditional on the behavior of others.

To sum up, the book covers some fundamental models that describe social equilibrium derived from individual choices. The latter depends on the actual behavior of others or the beliefs about their behavior since people comprise the social aggregate. Having different size groups and individual preferences we can obtain various equilibria, but not all of them will be efficient. In such cases, the introduction of institutional mechanisms could be socially beneficial.

Governmental initiatives are considered important for solving environmental problems, therefore one can be interested in the assessment of such measures' environmental impact and economic efficiency. D. A. Keiser and J. S. Shapiro analyzes the Clean Water Act showing its impact on water quality, allocation of funds, and Act’s impact on housing values.

Clean Water Act is a federal law in the USA regulating water pollution. It introduces wastewater standards for industries, water quality criteria, etc. The paper is aimed to solve two controversies around the Clean Water Act: its impact on pollution levels and the relationship between its cost and benefits. To do this the authors use a comprehensive set of data: atlas mapping all U.S. surface waters, panel description of the country’s wastewater treatment plants, historic extract of the Grants Information and Control System describing each Clean Water Act grants the federal government gave to cities; the Survey of Water Use in Manufacturing, a confidential plant-level data set of large industrial water users; and around 50 million water pollution readings at over 240,000 pollution monitoring sites during the years 1962–2001

From econometric analysis including difference-in-difference and triple-difference models, the following results are obtained:


1. Most variables that show the level of water pollution declined over the period 1962–2001 with a slowing rate.

2. Using triple difference regression (before vs after investments, upstream vs downstream of recipient plants), the authors show that receiving the grant decreases the probability of violating the standards for being fishable by half a percentage point. Concerning the flow from federal to municipal budgets on fighting pollution, it was found that $1 of a federal grant led to about $1 more of municipal sewerage capital spending.

3. In the analysis of house values around rivers and streams, D. A. Keiser and J. S. Shapiro estimates the effects of grants on home values are about 25% of their costs. Nevertheless, for absolute values, the analysis of effects on housing demand leads to the conclusion that the value of homes increases, but the effect is modest and not significant in most specifications.

Possible explanations are the following: people may have incomplete information about changes in water pollution and their welfare; these numbers exclude nonuse values; grants may increase sewer fees; these estimates abstract from general equilibrium effects; and they exclude the 5% of most distant recreational trips.


David A Keiser, Joseph S Shapiro, Consequences of the Clean Water Act and the Demand for Water Quality, The Quarterly Journal of Economics, Volume 134, Issue 1, February 2019, Pages 349–396, https://doi.org/10.1093/qje/qjy019

The Church was a highly powerful institution in the Middle Ages and later, and one of the colonization’s rationale was proselytizing mission. Priests that came to Latin America patronized the local population and kept records that helped to get information about the colonization process. Also, it is worth remembering that religion at those times played first fiddle in the educational process. Thus, religious missions influenced the further development of areas where they were located. F. V. Caicedo analyzed whether it is a case, and what impact the Jesuit order’s mission had on the current human capital in some regions of Argentina, Brazil, and Paraguay.

The author uses archival records, census data, and household surveys to analyze this question. The explanatory variable is the distance to the nearest mission with controls for geographic features and weather, and the dependent variables are schooling years/literacy level and income, if available, or multidimensional poverty index as a substitute. The strong and significant effect is shown for both education and income: there is a 10% reduction in illiteracy when moving 100km closer to a mission, poverty is also lower for locations close to missions. PCA combining income and education variables provide the same results.

To prove the causation, F. V. Caicedo performs two tests. First, the author uses abandoned missions to address the potential endogeneity of missionary locations. She conducts placebo-type tests and finds no consistent effects on modern education and income. Second, F. V. Caicedo studies the Franciscan Guaranı missions, which was aimed to maximize the number of souls converted to Christianity and did not stress the formation of human capital in contrast to Jesuits. The results also have no significant influence on Franciscan Guaranı missions on modern education.

Then, the author investigates the persistence of human capital. Using the earlier censuses, F. V. Caicedo shows that the effect is larger for intermediate historical periods. Also, the literacy gap is higher for women that corresponds to historical evidence of female instructions.

The last part of the article provides and tests mechanisms of persistence of differences in human capital and incomes. The author focuses on the following mechanisms:

— Occupational specialization: individuals that attended religious missions, receiving instruction and technical training, moved away from agriculture to start a proto-artisan class. The empirical tests suggest that structural transformation is a mechanism of persistence for missionary effect.

For this part of the analysis, I have the following concern: it seems like there is no control for natural resources ( soil, minerals), distance to big cities which are a sales market, and other variables which can potentially explain the specialization. In my opinion, it could be a limitation for the results.

-- Technology adoption: the author uses the adoption of GE soy seeds in Brazil and tests whether areas with higher human capital—close to Jesuit missions—adopted this new agricultural technology faster. Empirical evidence supports the explanation.

-- Alternative mechanisms: population density, investments in infrastructure, and migration. It was found that places close to Jesuit missions are less dense today. There is a significant effect of missionary distance on modern road network density and the opposite effect on railroad density, reflecting the few tracks on the Brazilian coast. Analyzing health as a human capital investment, the author shows negative and significant coefficients on distance from a mission for the Brazilian health index. Concerning migration, it does not seem that people are sorting themselves into former missionary locations during modern times.


Felipe Valencia Caicedo, 2019. "The Mission: Human Capital Transmission, Economic Persistence, and Culture in South America," The Quarterly Journal of Economics, Oxford University Press, vol. 134(1), pages 507-556.

Labor unions and shifts in inequality

David Card, Thomas Lemieux, and W. Craig Riddell studied the impact of unionization on wage inequality exploiting the survey data from the US, UK, and Canada. The choice of these countries is explained by similar organization of union systems and the large share of the non-unionized labor force which is needed for comparisons. They measure the inequality as a variance of wages within a skill group that was approximated by the observed characteristics such as age and education from the data. Also, they provide a comparison of unions’ effect between male and female workers. Here I would summarize some findings of the paper:

1. Unions tend to “flatten” wage differentials across skill groups. One explanation of this effect is related to unobserved skill differences between union and nonunion workers in different age-education groups. Union workers, as was shown in previous literature, are more skilled on average. Union wage structures do not fully compensate high-productivity individuals. Then, to reduce the burden of above-market wages, employers tend to be more selective. This induces a selectivity bias that manifests as the true union wage premium. Also, it leads to overestimation in “between-group” equalizing the effect of the union in the proposed theoretical framework.

2. Considering the effect on female workers, the authors show that the union wage gaps for women are roughly constant. Since the rate of unionization of women is rising across wages distribution, the absence of a “flattening” effect of unions on female wages implies that covariance between the nonunion wage and the union wage gain for a particular skill group is either zero or positive. Thus, the effect of unions on female inequality is limited.

3. Discussing the trends of unionization, the authors mention that for males the level of unionization has declined sharply while it is not the case for women. The reason is the shift of unionization from the private to the public sector where women, and unionized women, in particular, are concentrated.

4. The union wage gap is typically larger for women than for men. That can be explained by the fact that unionized women are more highly concentrated in the upper end of the skill distribution than unionized men. Thus, controlling for the skills reduces the union wage gap far more for women than for men.

5. Looking at the inequality trends, Card et. al. conclude that unions reduce the variance of wages for men. By contrast, the effect for women is pretty small and slightly positive, which means that unions raise inequality. Considering inequality in a broader context, the authors provide evidence that union wage compression effects help explain a reasonable fraction of the growth in male wage inequality and cross-country differences in male wage inequality. Nevertheless, the effect is not observed in Canada. They do not consider female workers since unions do not affect inequality for them much as was shown.

Card, David, Thomas Lemieux and W. Craig Riddell. “Unions and Wage Inequality.” Journal of Labor Research 25 (Fall 2004)

The origin of state: bandits in Congo

Literature in political science suggests several theories on the states’ origin. The main issue of these theories is their testability, we could rely on historical evidence but it is not enough. Nevertheless, we can test some of them in the context of states which emerge now. Raúl Sánchez de la Sierra studies the situation in Eastern Congo providing support that the state’s origin is associated with the emergence of essential function, taxation in this case. [1] Also, his work serves as an illustration of Mancur Olson’s stationary bandit model of government. [3]

De la Sierra shows that the fiscal and collective capacities are associated with the mines of gold rather than mines of coltan where we observe the distraction of the institutions. In Eastern Congo there is no strong state power, therefore its functions are performed by armed forces which control mines. The difference between coltan and gold is the ability to hide, this fact, according to the author, leads to the difference in institutions. Since bandits can easily observe the coltan which was mined, they can directly tax citizens on the way to the village, therefore they just extract income. Gold could be easily hidden, therefore bandits should create indirect taxation to derive income from the villagers. Thus, as De la Sierra suggests, bandits introduce lump-sum taxes to enter the mine and create taxes on transactions in the village, which require the investment in fiscal capacity for accounting as well as the collective capacity to take those taxes. Using the changes in prices of coltan and gold, De la Sierra provides a causal relationship between the easiness to hide the resource and the emergence of the state’s essential functions.

As a continuation of this work, Soeren J. Henn and coauthors examine the incentives to restrain from violence and arbitrary theft by an armed group in the Eastern Democratic Republic of the Congo, the Front de Liberation du Rwanda (FDLR). [2] They use this setting because of two reasons. First, the weakness of the central state enables the FDLR to perform state functions. Second, the FDLR descends from Rwanda and has weak ties with the Congolese population that help isolate the role of the time horizon from simple benevolence. Henn et. al. show that controlling most villages the FDLR used their power to collect taxes, provide protection, and run fiscal and judicial administrations. Also, they almost never attacked the villages they controlled.

To support the causality of their findings, authors conduct interviews that suggest that in response to Kimia II [4], the FDLR lost the ability to permanently tax in their villages. Also, they hid in the neighboring forest of Itombwe from which they performed violent attacks aimed at stealing wealth and food. It could be explained by the fact that they do not internalize the effect of these operations on village growth anymore, since it does not affect them via taxation. To isolate the effect of a long stealing time horizon, the authors use the timing and targeting of Kimia II. Implementing both an event study and a differences-in-differences framework, Henn et. al. compare the FDLR state villages to the rest of the sample before and after Kimia II. They conclude that compared to before Kimia II, FDLR’s violent expropriation operations increased by 350% in FDLR state villages. This finding supports the notion that a long stealing horizon before incentivized taxation over arbitrary expropriations.


[1] Raúl Sánchez de la Sierra. (2020). On the Origins of the State: Stationary Bandits and Taxation in Eastern Congo. Journal of Political Economy. Volume 128, Number 1

[2] Soeren J. Henn, Christian Mastaki Mugaruka, Miguel Ortiz, Raul Sanchez de la Sierra, David Qihang W. (2021). On the Ends of the State: Stationary Bandits and the Time Horizon in Eastern Congo. BFI WORKING PAPER

[3] Mancur Olson, (2000). Power And Prosperity: Outgrowing Communist And Capitalist Dictatorships

[4] Kimia II is a military mission of the government and UN against FDLR. For references, see this or this.

Long-term impact of institutions on culture

Institutional economics literature argues that both institutions and culture matter for economic development but this argument is built on separate research of the influence of each of these factors. Lowes et. al. study the interplay between culture and institutions, particularly, they investigate the impact of formal institutions on the development of social norms. [1] The paper suggests that centralized institutions are associated with weaker norms of rule following and a greater propensity to cheat for material gain.

The authors use the natural experiment stemming from the origin of The Kuba Kingdom. Approximately in the 15th century, there was an influx of migrants in an area near the confluence of the Kasai and Sankuru rivers. Historical evidence states that migrated groups are homogeneous in terms of culture. This fact leaves us with no pre-trend. The second fact is that the boundaries of the Kingdom were determined by an external ruler using the geographical features of the area. Also, these boundaries which coincide with the rivers’ valleys were stable during history, therefore authors exploit the emergence of the Kuba Kingdom as a natural experiment.

Historical evidence suggests that this state was relatively developed: it had a professional bureaucracy, a system of taxation, extensive public goods provision, an unwritten constitution, etc. To examine the long-term impact of those institutional advances on culture, Lowes et. al. compare individuals whose ancestors lived within the Kuba Kingdom to individuals whose ancestors lived just outside the Kingdom. They test for differences in the propensity to follow rules in cases when there is a monetary incentive to not do so. The measure of rule following comes from two sets of behavioral experiments. The first one is the resource allocation game (RAG): a player rolls the dice with different colors on sides and divides the money between her and the other player according to her color identification which she chooses in her head. So, there is no control for cheating, but it could be identified via the deviation from the theoretical distribution. The second experiment is a version of the standard ultimatum game, in which participants physically allocate money in a private setting that allows them to steal money before offering a division.

Lowes et. al. consider three samples. The first sample is the largest and includes all individuals whose ancestors lived inside and just outside the Kuba Kingdom (499 individuals in total). The second sample comprises only the population that was culturally homogeneous before the creation of the Kuba Kingdom, it includes the central Kuba and the Lele - descendants with the same origin (and culture) but leaving outside Kuba. The third sample focuses specifically on the core people of the Kuba Kingdom and compares them to the Lele.

The authors show that Kuba ancestry is associated with more rule breaking and more theft. This is true for both experiments and all three samples of interest. This finding could be interpreted as follows: culture can be shaped by state institutions, and this setting is in line with the predictions of the Tabellini model. [2] Thus, their results show that Kuba state is crowding out internal norms of rule following. This finding contradicts the hypothesis of Elias, Weber, and Foucault, who argues that culture and formal institutions are complements. [3] [4] [5]

Addressing the positive correlation which was found in previous studies, the authors claim that restricted samples produce negative estimates of the Kuba Kingdom on rule following with the larger magnitude compared to the full sample. Then, the restricted sample estimates, which exploit the cultural regression discontinuity, might be better identified than the full-sample estimates, this suggests a positive reverse effect of culture on institutions: groups with stronger norms of rule following are more likely to establish more centralized and formal state structures. This fact leads to the upward bias of the estimates of the effect of state centralization on rule-following norms.

[1] Lowes S, Nunn N, Robinson JA, Weigel J. (2017). The Evolution of Culture and Institutions: Evidence from the Kuba Kingdom. Econometrica. 85 (4) : 1065-1091.
[2] Tabellini, G. (2008). The Scope of Cooperation: Values and Incentives. The Quarterly Journal of Economics, 123(3), 905–950. http://www.jstor.org/stable/25098921
[3] Elias, N. (1994): The Civilizing Process. Oxford: Blackwell.
[4] Foucault, M. (1995): Discipline and Punish: The Birth of the Prison. New York: Vintage Books.
[5] Weber, E. (1976): Peasants Into Frenchmen: The Modernization of Rural France, 1870–1914. Stanford: Stanford University Press.

Online media and ideological segregation

In his book, Sunstein argues that the Internet facilitates peoples’ convergence to sources that correspond to their initial views that could be harmful to democracy. [2] The urgency of this issue is rising due to the Internet's fast expansion and its probable substitution of traditional media and face-to-face interaction. Nevertheless, M. Gentzkow and J. Shapiro provide evidence against this point. They conclude that online media is less ideologically segregated than national newspapers or face-to-face interaction. [1]

The authors collect data on the consumption of newspapers, magazines, broadcast television, and cable. To measure in-person interaction they use the political views of individuals’ acquaintances and political discussants as reported in the 2006 General Social Survey and the 1992 Cross-National Election Study. For each outlet, they measure the share of conservatives (those who report their political outlook as “conservative” among those who report being either “conservative” or “liberal”). Then, they calculate the individual’s conservative exposure to be the average share conservative on the outlets she visits. As a measure of segregation, the authors use the “isolation index”, which is a standard metric in the literature on racial segregation. It equals the difference between the average conservative exposure of conservatives and the average conservative exposure of liberals. For instance, if conservatives only visit foxnews.com and liberals only visit nytimes.com, the isolation index will be equal to 100 p.p., while if they both get all their news from one source, the isolation index will be equal to 0.

According to this approach, the estimated isolation index for the Internet is higher than that of broadcast television news (1.8), cable television news (3.3), magazines (4.7), and local newspapers (4.8) and lower than that of national newspapers (10.4). Online segregation is higher than that of a social network where individuals are matched randomly within counties (5.9) and lower than that of a network where individuals matched randomly within ZIP codes (9.4). Also, it is significantly lower than the segregation of actual networks formed through voluntary associations (14.5), work (16.8), neighborhoods (18.7), or family (24.3). The Internet is also far less segregated than networks of trusted friends (30.3) and political discussants (39.4).

M. Gentzkow and J. Shapiro discuss two mechanisms for this phenomenon:

— Online news consumption is concentrated in a small number of relatively centrist sites. More ideologically extreme sites account for a very small share of online consumption.

— The majority of individuals consume content from multiple sources. Especially those who visit more extreme blogs overall who spend time online and use more centrist sources or even sites with the opposite ideological slant as well.

One of the caveats of this study is that it does not explain how people translate the content they encounter into beliefs. People with different ideologies see similar content, but they might interpret it differently. Nevertheless, this paper provides a strong argument against previous results suggesting the higher level of segregation caused by internet media. [3]


[1] Gentzkow, M., and J. Shapiro (2011): “Ideological Segregation Online and Offline,” Quarterly Journal of Economics, 126 (4), 1799– 1839.

[2] Sunstein, Cass R. Republic.com (Princeton, NJ: Princeton University Press, 2001).

[3] Theoretical models of media markets predicting in which increasing the number of outlets may lead consumers to become more segregated ideologically:
- Mullainathan, Sendhil, and Andrei Shleifer. “The Market for News,” American Economics Review, 95 (2005), 1031–1053.
- Sobbrio, Francesco. “A Citizen-editors Model of News Media,” SSRN Working Paper, 2009.
- Stone, Daniel F. “Ideological Media Bias.” SSRN Working Paper, 2010.

Also, there are plenty of comments of public officials, commentators, and thinkers suggesting the increasing polarization. For instance, see this or this article from NYT.

Polarization vs antipolarization

Social media is often accused of dividing society apart. Sustain (2017) in his book argues that social media facilitates the creation of echo chambers that expose people to those who are similar to them. This effect makes their initial slant more strong. However, Gentzkow, M., and J. Shapiro (2011) show that the Internet is not as polarized as face-to-face contacts though they do not provide analysis for social media. C. Bail provides experimental evidence on the polarizing effect of exposure. However, these findings are not supported by R. Levy (2021). Indeed, the answer to this question is not obvious since one could come up with alternative mechanisms which justified the decrease in polarization, for instance, lower cost of searching for alternative explanations. I will briefly describe the findings of both papers and provide some arguments to explain this two-sided effect of social media.

Bail et. al. (2018) experimented on Twitter by suggesting users follow a bot that retweets posts opposed to their views. R. They select accounts of elected officials, opinion leaders, media organizations, and nonprofit groups, and regularly randomly select an account to make a retweet. Comparing the surveys before, within, and after the experiment authors make the following conclusions:

— Republicans who were exposed to a liberal Twitter bot became substantially more conservative post-treatment.
— Democrats exhibited slight increases in liberal attitudes after following a conservative Twitter bot. However, these results are not statistically significant.

In his book, C. Bail talks about his research based on in-depth interviews. Here he explained the channel of polarization. First, the polarization after exposure to the opposite opinion if initially, people have more extreme views. Second, he shows that there are discrepancies between online and offline behavior, particularly, people could be restricted to share their opinion online because offline or, on the contrary, be freer on social media compensating for the lack of social approval and admiration in real life. Third, he provides suggestions to improve communications online by creating a new discussion platform where people of different slants were incentivized to discuss various topics.

R. Levy conducted an online experiment suggesting users subscribe to opposing news outlets. He managed to gather data on subscriptions to outlets, exposure to news on Facebook, visits to online news sites, and sharing of posts, as well as changes in political opinions and attitudes from the surveys. This study has four main findings:

1. News sites visited through social media, and specifically, Facebook, are associated with more segregated and extreme news, compared to other news sites. Nevertheless, exposure to opposing news outlets helps to diversify the news consumption of an individual.
2. Exposure to counter-attitudinal news decreases negative attitudes toward the opposing political party, i.e. affective polarization. This is in line with Bail’s finding of exposing people to those with opposite views on the discussion platform.
3. Exposure does not affect people's beliefs and ideological slant.
4. Facebook’s algorithm is less likely to supply individuals with posts from counter-attitudinal outlets, conditional on individuals subscribing to them. Thus, social media algorithms may limit exposure to counter-attitudinal news.

Thus, a comparison of the focuses of these works could help to shed a light on this divergent effect. First, the extreme belief which is emphasized in Bail's works could be the main driver of polarization. Since the share of such people is not high enough, this is not reflected in Levy's findings. The second reason lies in the dichotomy of active interaction and consumption. In my opinion, the first should affect beliefs and affective polarization compared to pure consumption of news since in the interaction with others more tools of persuasion are used. The third potential explanation, which is close to the first one, is the coherence of beliefs. As C. Bail shows in his book and is established in political science literature, people could have contradictory and incomplete beliefs which also could change the way how they are responsive to the exposure of opposite opinions.

This question lies if the area of my research interest, therefore this small post could be treated as a motivation for my thesis. I hope to provide a more rigorous argument on these counter effects later on.

References:

Levy, Ro'ee. 2021. "Social Media, News Consumption, and Polarization: Evidence from a Field Experiment." American Economic Review, 111 (3): 831-70.

C. A. Bail et. al. 2018. Exposure to opposing views on social media can increase political polarization. Proceedings of the National Academy of Sciences, 115(37): 9216-9221

C. A. Bail 2021. Breaking the Social Media Prism: How to Make Our Platforms Less Polarizing. Princeton University Press.

Dynamic polarization

Polarization is a widespread phenomenon in the political sphere, and it affects different groups differently. Particularly, we notice the polarization between elites (parties) and masses (voters). There are great papers that empirically investigate polarization among people (to be reviewed), but there was no one that tries to catch polarization at both levels and provide a theoretical concept to explain polarization dynamics. S.Callander and J. C. Carbajal fill this gap by providing a dynamic voting model with voters who update their beliefs.

They build upon a classic model of electoral competition adding behaviorally justified change to the nature of voter preference. Namely, the voter updates their preferences in a way to justify their choice which is called a causal loop, i.e. they vote for a party that is closer to them and then they become even closer to this party in terms of preferences. Also, unlike other models, they introduce the tolerance of voters which guarantees that voters vote not just for the closest party but the closest party within a particular range. This helps to derive equilibrium with the so-called “missing middle” of the electorate or “disappearing center” Abramowitz (2010) – moderate voters who are left behind because of parties’ polarization.

Since parties face a trade-off between the probability of winning and the policy outcome, the authors find a new type of equilibrium where parties do not compete with each other but with abstention. This strategy is called “no-voter-left-behind”. It means that parties polarize incrementally such that in each step, no voters are lost to abstention. This implies that the speed of polarization is tightly linked to the degree voters update their preferences after voting: the smaller is updating, the more incremental the updating, and the slower and more iterative party polarization. This dynamic reflects a shift in the nature of electoral competition over time: from competition for the swing voter to competition for turnout.

The dynamic nature of this model allows us to conclude that parties would converge faster to their ideal points while voters will update their beliefs until they reach the position of their favorite parties. Note that abstainers do not update their ideals points, so the gap in the middle opens up. This result is supported by empirical literature that finds evidence for elite polarization and very moderate polarization within the masses.


[1] Steven Callander & Juan Carlos Carbajal, 2022. "Cause and Effect in Political Polarization: A Dynamic Analysis," Journal of Political Economy, Vol. 130, N. 4

[2] Abramowitz, A. I. (2010). The Disappearing Center: Engaged Citizens, Polarization, and American Democracy. Yale University Press.

Attitudes change via social media: example of xenophobia

The literature provides evidence that social media could influence opinions and attitudes. Among others, it could drive the recent increase in the expression of hate and xenophobia. Bursztyn et. al. (2019) provides causal evidence in the Russian context. [1] I will briefly summarize the research design, theoretical model, and results interpretation.

To establish the causal relationship between social media usage and xenophobia, they follow the (Enikolpov, at. al. (2020)) approach using the penetration of VK as an instrument for social media usage. [2] Then, the approximate xenophobic views by the city-level vote share of Rodina, an explicitly nationalist and xenophobic party. The authors documented that the impact of social media on hate crime is positive and significantly depends on the strength of pre-existing support of nationalists in the city which is consistent with previous literature about traditional media (Adena et al., 2015). [3] Then, the authors examine the potential mechanism via the online survey.

Channels through which social media could affect the expression of hate are the following: facilitation of coordination that is relevant for illegal and stigmatized activities like hate crimes, change of opinion (exposure to intolerant views that are prone to be exacerbated via echo chambers), and effect on people’s perceptions of the acceptability of expressing hate (and, consequently, the willingness to express hate). The survey was designed to test the last channel, i.e. perceptions of the social acceptability of xenophobia. The authors find a positive effect of social media penetration on elicited ethnic hostility, i.e. the share of respondents that hold xenophobic attitudes, regardless of whether they are willing to openly report them. Bursztyn, et. al. also look at the effect of social media on self-reported ethnic hostility, but the coefficients are mostly insignificant and negative. The difference between elicited and reported ethnic hostility could be used as a proxy for the perceived stigma associated with the expression of such attitudes in a survey. According to the authors' results, there is no evidence that social media reduced that perceived stigma.

In addition to empirical evidence, the authors build a simple theoretical model which captures the idea that social media increases the propensity of individuals to meet like-minded people, thereby resulting in higher polarization of opinions. In this model, individuals have a particular position (for instance, about aborts) which is drawn from some distribution. In each period, they update their views as the result of interactions with others. The change proceeds by the following scheme: the new position incorporates the current one with weight ω, and the positions of other people they talk to with weight 1 − ω. Also, each individual interacts with others whose share r is similar to an individual, and share 1-r is a random set of individuals from the society. Thus, r is considered to be a proxy for penetration of social networks. Also, to avoid convergence, the authors assume that each individual’s position is subject to a random additive shock with the normal distribution of zero mean. This model predicts convergence of the normal distribution of preferences with a mean equal to the mean of the original distribution. The predictions of the model are the following: the more individuals are influenced by people with random opinions (lower r) the faster these preference shocks dissipate leading to the smaller the variance of the limit distribution. Conversely, if people are mostly influenced by themselves (higher ω) or like-minded people (higher r ), as in ‘echo chambers, so we observe the persistence of beliefs.


[1] Bursztyn, L., Egorov , G., Enikolopov, R., Petrova, M. (2019). Social media and xenophobia: evidence from Russia. NBER Working Paper 26567

[2] Enikolopov, Ruben & Makarin, Alexey & Petrova, Maria. (2020). Social Media and Protest Participation: Evidence from Russia. Econometrica,Vol. 88, Issue 4, pp. 1479-1514

[3] Adena, M., Enikolopov, R., Petrova, M., Santarosa, V., Zhuravskaya, E. (2015) Radio and the Rise of The Nazis in Prewar Germany , The Quarterly Journal of Economics, Volume 130, Issue 4, November 2015, Pages 1885–1939, https://doi.org/10.1093/qje/qjv030

Factors of online news demand

Insights into the factors influencing people's choice of news sources could help understand better important questions such as propaganda effectiveness, the dynamic of slants, fakes’ spreading, and many others. Simonov and Rao (2022) investigated the demand drivers in the context of authoritarian Russia which is known for its control of media.

The authors separate two drivers of demand for government-controlled (GC) outlets. One of the drivers is pro-government bias. The second factor includes tastes for other features of the outlets, for instance, website design, video content, referrals by news aggregators, or accumulated brand capital. All these factors authors call “persistent preferences”. Also, they differentiate between two types of content: sensitive and non-sensitive topics. For the non-sensitive topics, there is no difference in terms of ideological slants, while news on sensitive topics is framed following the official position.

Simonov and Rao ground their research in a model of news production and consumption. In this model, consumers choose news outlets relying on their outlet-level persistent preferences and tastes for outlets’ ideological coverage. The importance of outlets’ ideological coverage changes depending on the volume of realized sensitive-news events. For instance, on days with no sensitive-news events, people do not pay attention to news outlets’ ideological positions.

On the production side, news outlets decide on shares of sensitive news commodities in their product. The authors assume that news commodities are costly to produce. Also, it is less costly on days when a lot of topic-related events happen. Given that the government dislikes sensitive-news publications, it exercises censorship by imposing additional costs for the production of sensitive news on independent outlet A.

From this model, two observations could be derived. First, a controlled outlet would choose to produce less sensitive news than an independent outlet. Second, since the government is assumed to care mainly about the first few sensitive stories reported, we should expect that the difference in the amount of sensitive news produced by the independent and controlled outlets is increasing in the share of sensitive news to report. For instance, if there is no sensitive news, it can be very costly for both news outlets to produce sensitive news, so both outlets produce very low sensitive content. However, when there is a lot of sensitive news to report, the cost of sensitive-news production is low, and censorship (as costs) plays a more important role.

Next, Simonov and Rao turn to empirical analysis. They consider the top 48 Russian-language news outlets, classified by the degree and type of government influence. They end up with 3 groups: outlets that are owned by the government or members of the incumbent political party (GC), “potentially influenced” outlets, and independent outlets (owned by either journalists, international media companies, or the opposition to the government). They gathered the content published by these outlets during the reference period. From this data, they found 2 topics of sensitive news: POC (political protests, opposition, and corruption) and news about Ukraine. Also, they conclude that international outlets closely resemble Ukrainian outlets in their ideological slant on sensitive topics, and independent outlets have more “neutral” ideological positions. Some potentially influenced outlets closely resemble independent outlets, while others resemble GC outlets.

Using structural estimation, the authors show that news outlets that report more news about the Ukraine crisis and have a more pro-Ukraine ideological position get the highest increases in their market shares on days with a high share of sensitive news realizations.

Next, Simonov and Rao describe the news consumption patterns. They show that consumers are more likely to arrive at the GC outlets from third-party websites: direct navigation plays a lower role for the GC outlets (50.68%), especially compared to the independent ones (56.28%). Particularly, the GC outlets get more than a quarter of their traffic from Yandex (25.57%), compared to only 15.5% for the independent outlets. GC outlets also tend to cross-refer each other more than other types of news outlets. In addition, a large share of consumers arrives directly at their news article pages and “other pages” such as special projects and videos that support the notion of persistent preferences. Also, they conclude that there is almost no outlet-topic specialization. The authors also estimate the inputs of various third parties on the popularity of GC outlets as well as provide contrafactual analysis for market shares.

To sum up, the authors build a theoretical framework to investigate the importance of two channels for online news demand: ideological preferences and factors related to the design, brand, and others. They conclude that the main source of demand for the GC news outlets comes from persistent preferences, which are largely supported by third-party referrals and the nonsensitive news content on GC outlets’ websites. This could potentially help the government in the propaganda making the customers stick to the controlled outlet.

The role of third-party referrals was discussed in the context of the war in Ukraine which was followed by shutdowns of independent news outlets. Also, one of the critical parts was again the role of Yandex as one of the main news aggregators in Russia.

One of the important comments about this paper is that the authors could measure only visitings of the outlets, not the customers' opinion (or its shift). For instance, some of the people, who follow independent outlets, could read the referred GC news not because they agree with the official position but because they want to critically assess the arguments of the opposite side. Therefore, some of the visits do not construct the demand for GC outlets but, for instance, show the demand for comparison and gaining additional information for the critique.

Simonov, A. & Rao, J. (2022) Demand for Online News under Government Control: Evidence from Russia. Journal of Political Economy, 130(2), 259-309

Ideological bias and trust

People with different ideological preferences disagree not only about particular issues but also about the trustworthiness of each other's information sources. Thus, the root of the polarization might be grounded in both people’s biases and their trust in different sources of information. This mechanism is addressed in the theoretical model proposed by M. Gentzkow, M. Wong, and A. Zhang. Their contribution to this work is explaining the biases in the learning process without an assumption about any behavioral deviation, i.e. with Bayesian individuals. The model is built upon Berk (1966), who provides a general statement that beliefs need not converge in the long run under misspecified learning, and Acemoglu, Chernozhukov, and Yildiz (2016) conclude that arbitrarily small differences in beliefs about the interpretation of signals can generate large disagreements about an underlying state.

The model describes the learning process of agents who gain knowledge about a sequence of unobservable states in several periods. There is an ideological valence on the state of the world, i.e. consistency of a particular ideological platform. The agent observes signals from information resources that are associated with parameters reflecting the accuracy and bias of signals. In some cases, the agent receives feedback that she believes is unbiased. The authors look at single-homing when there is only one course and multi-homing cases when agents observe all the signals. Introducing the relationship structure between parts of the model, the authors look at the dynamic of several key variables described below.

The results could be divided into two groups. The first group comprises the learning about parameters including correlation between variables, i.e. how the signal of the source is correlated with the state of the world. Another group of the results is devoted to learning about the state of the world. Below I list the main conclusions about different facets of these two directions.

Learning about parameters of signals and feedback
— Confidence: beliefs about the accuracy of her feedback). The authors show that learning from either one or all resources under some conditions overconfidence arises. It means that people could rely more on the feedback than they should.
— Trust: beliefs about the accuracy of the sources. The authors show the trade-off between accuracy and bias and argue that trust-maximizing sources will lie on the border of all possible sets of parameters standing for accuracy and bias. Thus, if the bias of a source is sufficiently high relative to accuracy, the agent will prefer a source with a bias close to one and accuracy close to zero, i.e. a source that reports the ideological valence.
— Ideology and perceived bias: beliefs about the correlation ideological valence and state of the world. Considering the agent with the right-leaning bias, the authors derive that she perceives an unbiased source as oppositely biased. Besides, she perceives a like-minded biased source as less right-biased than it actually is.

Learning about the state of the world
— Accuracy: the position of posterior beliefs about the state of the world compared to the true state. The paper concludes that with the tight priors of an agent the lack of bias leads to perfect accuracy. However, the positive bias distorts the posterior beliefs.
— Polarization: the extent to which agents with opposite biases come to disagree about the value of the state of the world. Supporting the previous empirical evidence, the model predicts that the differences in view on facts between people with the opposite view can arise as a result of information processing biases even when accurate information is widely available and agents’ only motivation is to learn the truth.
— Single- vs. Multi-Homing: The authors test the intuitive result that exposure to an ideologically diverse set of information could reduce polarization. They show that it is possible to have such an effect in multi-homing cases, but there are also negative outcomes with the same or even exacerbated polarization. As for the single-homing setting, posterior beliefs depend on the observed source in each period. If the source has less bias, there is less expected disagreement.
— Learning without feedback: the influence on asymptotic learning of the presence of feedback (ex-ante believed to be the unbiased source). In such a setting, the agent cannot rule out the extreme possibilities that any of the sources are perfectly positively correlated, uncorrelated, or perfectly negatively correlated with the true state. It means, that conditional on a received signal, the expected state of the world is zero.

Thus, this paper provides a rich set of conclusions that explain the underlying reasons for disagreements between representatives of various ideological platforms and elaborate on the forces for the dynamic of polarization related to media.

Matthew Gentzkow, Michael B. Wong, and Allen T. Zhang. (2021) “Ideological Bias and Trust in Information Sources”. Working Paper.

This post opens the series of reviews on theoretical models of learning and dynamic of beliefs. Besides, some literature about network models might be covered.

Theoretical perspective on news consumption and beliefs formation

Here, I decided to share short summaries of several theoretical works about news consumption and belief formation that are relevant to my thesis. I think understanding the learning process is an important key to understanding the roots of peoples' behavior.

The working paper by G. Tabellini and L. D'Amico studies how people comment on political news on Reddit political forums. The authors provide a theoretical model that shows how people allocate their attention to signals about two candidates to form a ranking between them. To test the model predictions, the authors use the comments to post on political forums on Reddit. The paper concludes that partisan users behave very differently from independents: partisan comments on bad news are less frequent on the own candidate, and more frequent on the opponent. Applying text analysis, the authors show that partisan users are less likely to accept bad news about their candidate, and more likely on the opponent that is consistent with motivated reasoning. They point out that emotions play an important role in the propensity to comment on political news. The authors mention that a potential interpretation of their results is that a partisan user tries to protect their self-identity, rationalizing the candidate’s behavior, finding excuses for it, or attenuating its relevance.

In this regard, Fryer et. al. suggest a model explaining the behavior of people when they receive ambiguous news such that different people might interpret them differently. They show that double updating (the interpretation of ambiguous signals, and then the formation of a posterior) leads to confirmation bias and polarization. They also conducted an online experiment in which individuals interpret research summaries about climate change and the death penalty. The authors show a significant relationship between an individual’s prior and their interpretation of the summaries as was predicted by the model.

Another pattern of news consumption leading to polarization is selective sharing presented in (Bowen et. al., 2023). The authors assume that people tend to share only the signals which are in line with their priors. Also, they consider a not fully rational person who does not estimate properly the probability of not receiving a signal. That distorts her updating process and leads to polarization.

(Gentzkow et. al., 2021) considers trust as one of the driving forces of polarization. The authors present a learning process where agents do not know the accuracy of sources ex-ante and rely on noisy feedback. This paper shows that small biases in this feedback can cause large ideological differences both the in the trust in information sources and beliefs about the states of the world.

One more mechanism was presented by (DeMarzo, et. al., 2003). The authors assume that people underestimate or completely ignore the fact that signals they receive are correlated with each other. This fact distorts the way they form their beliefs, and it leads to overconfidence.

References
D’Amico, L., & Tabellini, G. (2022). Online Political Debates (tech. rep.). Center for Economic Studies & Ifo Institute
DeMarzo, P. M., Vayanos, D., & Zwiebel, J. (2003). Persuasion Bias, Social Influence, and Unidimensional Opinions. The Quarterly Journal of Economics, 118 (3)
Fryer Jr, R. G., Harms, P., & Jackson, M. O. (2018). Updating Beliefs when Evidence is Open to Interpretation: Implications for Bias and Polarization. Journal of the European Economic Association, 17 (5)
Gentzkow, M., Wong, M. B., & Zhang, A. T. (2021). Ideological Bias and Trust in Information Sources (Working paper). https://www.matthewgentzkow.com/papers/
T Renee Bowen, Danil Dmitriev, Simone Galperti, Learning from Shared News: When Abundant Information Leads to Belief Polarization, The Quarterly Journal of Economics, 2023; qjac045, https://doi.org/10.1093/qje/qjac045