From Neighborhoods to Nations: The Economics of Social Interactions

By Yannis Ioannides

Just as we learn from, influence, and are influenced by others, our social interactions drive economic growth in cities, regions, and nations--determining where households live, how children learn, and what cities and firms produce. From Neighborhoods to Nations synthesizes the recent economics of social interactions for anyone seeking to understand the contributions of this important area. Integrating theory and empirics, Yannis Ioannides explores theoretical and empirical tools that economists use to investigate social interactions, and he shows how a familiarity with these tools is essential for interpreting findings. The book makes work in the economics of social interactions accessible to other social scientists, including sociologists, political scientists, and urban planning and policy researchers.


Focusing on individual and household location decisions in the presence of interactions, Ioannides shows how research on cities and neighborhoods can explain communities' composition and spatial form, as well as changes in productivity, industrial specialization, urban expansion, and national growth. The author examines how researchers address the challenge of separating personal, social, and cultural forces from economic ones. Ioannides provides a toolkit for the next generation of inquiry, and he argues that quantifying the impact of social interactions in specific contexts is essential for grasping their scope and use in informing policy.


Revealing how empirical work on social interactions enriches our understanding of cities as engines of innovation and economic growth, From Neighborhoods to Nations carries ramifications throughout the social sciences and beyond.

• Language: English
• Publisher: Princeton University Press
• Release date: Oct 14, 2012
• ISBN: 9781400845385

Book preview

Index

PREFACE

Individuals share information; we self-select into social groups; most of us live and work in close proximity in cities and in firms, both important features of modern economic life. Economists, influenced by other social scientists and recognizing that disciplinary boundaries are sometimes arbitrary, have developed new theoretical models and empirical tools for understanding the social interactions that underlie interpersonal and community life.

This book offers a synthesis of research on the economics of social interactions, a body of knowledge made up of strands from several areas of economics. My goal is to provide a set of tools that can be used to structure empirical investigations and to interpret empirical findings in ways that make recent research in economics accessible as a tool to scholars in other social science disciplines. In other words, the book is designed to enrich our set of metaphors for understanding and modeling the fabric of communities, their neighborhoods, and their consequences for studying larger regional and national economies. Identifying and measuring the importance of social interactions is a challenging task because of the inherent difficulty in separating personal, social, and cultural forces from purely economic ones. Social interactions have important impacts on phenomena ranging from the diffusion of norms to how students learn from one another, and from causes of urban decay to explanations for economic growth.

The concept of social interactions has already shown its value in exploring many facets of interdependence between actors in the modern economy. In economics, social interactions are defined as direct agent-to-agent interactions that are not mediated by price. My overarching theme in this book is proximity in all of its dimensions and its impact on interactions among individuals and firms in society and in the economy. chapter 1 introduces highlights of the significance of social interactions. chapter 2 sets out the basics of the analytical language that I then use throughout the book to describe social interactions. The subsequent chapters use that analytical language. chapter 3 examines location decisions of individuals and emphasizes the study of neighborhood effects in housing markets and their interaction with the role of prices in rationing admission to communities and neighborhoods in market economies. chapter 4 looks at the impact of interactions on firms’ location decisions, focusing on the effects of proximity to other firms, the size of the total urban economy, the availability of a suitable labor force, and risk pooling. chapter 5 builds on the foundations laid down in earlier chapters when economic agents interact in physical space. It examines how the interactions of individuals and firms in their vicinity and in broader communities help us understand the spatial structure of cities as self-organization by agents. chapter 6 documents spatial patterns in productivity, wages, and incomes and addresses the origin of the idea that spatial concentration causes higher productivity. The chapter starts with aggregative spatial measures, such as economic activity at the level of states, regions, and counties, and moves to the smaller scale of cities and their neighborhoods. In chapters 7–9 the city is ultimately the unit of analysis. Those chapters address urban structure, industrial specialization and diversification, and urban growth in the context of national economic growth. Each chapter provides its own microfoundations and moves progressively from static settings to dynamic economies in steady states, such as the model of labor market turnover in chapter 7 and the empirics of urban evolution in chapter 8. chapter 9 explores models of long-run growth with factor accumulation and endogenous technological change.

Finally, chapter 10 speculates about the prospect of a deeper understanding of social interactions in urban settings, introducing broader sets of tools for describing the entire social fabric. I cogitate about ways the interplay of actors in the physical, economic, and social space allows interactions to make the global local. It ends by comparing individuals and their social interactions to an archipelago. Components of the urban economy and social structure interact in numerous ways, sometimes reaching far and other times concentrating locally as they react to economic and social forces. The models can allow an economy to self-organize in the face of vicissitudes within an ever-changing environment, as adverse shocks alternate with payoffs from increasing returns.

My goal is to emphasize that our knowledge of social interactions rests on data, on the empirical findings that derive from them, and on the applied economics that made those findings possible. It also reflects my view that the only way to do justice to the empirical findings is to present their theoretical underpinnings. Each chapter interweaves original material with syntheses of the existing literature, going back and forth between theory and empirics.

The book comes at a time when a torrent of new research has become available. Among several particularly elegant new books, those by Glaeser (2008), Jackson (2008), and Zenou (2009a) stand out. My goal is to provide a synthesis for economist and noneconomist readers that organizes the interacting areas of this very active research topic. Of course, I hope that others will build on my synthesis.

I am truly grateful to a great number of friends, some of whom also happen to be colleagues and research collaborators (from whom I have learned enormously, and especially from Vernon Henderson and Christopher Pissarides), who have shown great selflessness and immeasurable patience in reacting to my work over many years. Many offered suggestions and corrections during presentations of parts of the research that led to this book. Some generously offered thoughtful suggestions on earlier related work and on drafts of parts of the book. They include Tom Bender, Marcus Berliant, Larry Blume, John Boulton, Yann Bramoullé, Drusilla Brown, the late Toni Calvó-Armengol, David Cuberes, Linda Harris Dobkins, Gilles Duranton, Steven N. Durlauf, Dennis Epple, Yannis Evrigenis, Xavier Gabaix, Dominique Goux, Bryan Graham, Hans Haller, Bob Helsley, Vernon Henderson, Wen-Tai Hsu, Panle Jia Barwick, Matt Kahn, Tomoo Kikuchi, Alan P. Kirman, Anne Laferrère, the late Linda Datcher Loury, Stelios Michalopoulos, Tomoya Mori, Henry G. Overman, Theodore Palivos, Christopher A. Pissarides, Diego Puga, Danny Quah, Esteban Rossi-Hansberg, Kjell Salvanes, Kurt Schmidheiny (and his and Giacomo Ponzetto’s students at Pompeu Fabra), Tracey N. Seslen, Spyros Skouras, Adriaan Soetevent, Michael Sobel, Enrico Spolaore, Takatoshi Tabuchi, Chih Ming Tan, Heiwai Tang, Giorgio Topa, David Warsh, Bruce Weinberg, Jeff Zabel, Marios Zachariades, Giulio Zanella, Yves Zenou and Junfu Zhang. I benefited from a wonderful research environment provided by my colleagues at Tufts and by the MacArthur Research Network on Social Interactions and Economic Disparities, directed by Kenneth J. Arrow and Steven N. Durlauf during 1998–2005. The interactions in the network helped me decisively in clarifying my ideas. I acknowledge with gratitude resources from the MacArthur Foundation, the Max and Herta Neubauer Chair in Economics at Tufts, and the National Science Foundation under grants SBR-9618639 and ACI-9873339. I benefited greatly from the regular compilations of working papers produced as New Economics Papers: Urban and Real Estate Economics, part of Research Papers in Economics (RePEC), edited by Stephen Ross, and Economics of Networks eJournal, part of the Social Science Reasearch Network (SSRN), edited by Nicholas Economides. I thank Thad Beal whose Chance Construction 2 is so brilliantly evocative of how human networks overlay urban geography.

I wish to especially acknowledge my intellectual gratitude to Alan Kirman for encouraging me early on, and to Steven Durlauf, whose own research in related areas and whose comments and friendship over more than 15 years have had an extraordinary influence on much of my work reflected in this book. My friends Costas Azariadis, Dimitri P. Bertsekas, and Christopher A. Pissarides taught me the importance of setting high standards for oneself. I am grateful to the anonymous reviewers at Princeton University Press whose comments improved the manuscript enormously. Peter Dougherty, Tim Sullivan, and Seth Ditchik at the Press have been enthusiastic, very encouraging, and extraordinarily patient, and so has Janie Chan. Very special thanks go to Carol Dean for superb copyediting, and to Natalie Baan for meticulous care of the manuscript. Finally, Anna Hardman and Kimon Ioannides in different ways have been wonderfully helpful to me throughout this undertaking: Anna, with her tireless advice and editorial help, and Kimon, whose steadfast advice that writing a book is a different and worthy kind of challenge, kept me going.

September 25, 2011

From Neighborhoods to Nations

CHAPTER 1

Introduction

PHILOSOPHY MASTER: [E]verything that is not prose is verse, and everything that is not verse is prose.

MONSIEUR JOURDAIN: And when one speaks, what is that then?

PM: Prose.

MJ: What! When I say, Nicole, bring me my slippers, and give me my nightcap, that’s prose?

PM: Yes, Sir.

MJ: By my faith! For more than forty years I have been speaking prose without knowing anything about it, and I am much obliged to you for having taught me that.

—Moliére, Le Bourgeois Gentilhomme, 1670, act two. scene 4

We engage in social interactions without knowing anything about it throughout our lives; these interactions teach us new skills and influence our choices. Examples are easy to find: recycling and composting practices; sending a child to a charter school; ideas for software innovations that come from a chance encounter in a Silicon Valley, California, or Austin, Texas, bar; learning from class mates—about schoolwork or about getting pregnant or how to avoid it; gaining weight; attending a church, synagogue, or mosque; joining a gym or a country club; supporting a sports team; getting involved in a civic association or spending time working for a nonprofit; keeping up with college friends in person or on Facebook; enforcing, or failing to enforce, building code and zoning regulations; dying one’s hair to hide the gray. These are just a few examples.

Economic models of cities increasingly focus on the microfoundations of the multitude of interactions underlying innovation and creativity as well as on the pollution and congestion associated with cities as places where social interactions are most dense. Empirical work using data made more accessible by modern technologies of interpersonal communication has followed suit and is expanding the set of metaphors we can use to understand cities and urbanization. While social interactions are most dense in cities, this is not the only place where they are found.

Scholars in recent years have begun to explore the ways these social interactions influence our behavior and their broader implications for policy, asking questions like: How does access to mobile telephones in Africa influence farmers’ productivity and the farming techniques they use; does that in turn influence the size and growth of settlements? What is the reach and influence of places where urban buzz occurs? Is obesity—or depression or acne—contagious? How do racial and ethnic prejudices start and evolve, and can we deter them? Can interactions between neighbors help revitalize a decaying urban neighborhood, and why do they cause urban decline in one neighborhood and not in another? Did Edinburgh’s streets and urban form contribute to the interactions that led to the Scottish Enlightenment in the eighteenth century?

Some of our actions change prices. When families move to a community with good schools, property values rise, and that in turn is relevant for people who do not have school-age children. In other words, prices record the value of social interactions and can signal their quality. Economists’ questions about interactions started from but have moved well beyond direct influences on prices and markets.

In all the examples above social interactions are present, making individuals’ actions interdependent and in turn affecting their lives. Sometimes spatial proximity implies interaction, as in keeping up with the Joneses. Other times the links are professional, social, or familial, and agents interact at long distances from each other. Widespread adoption of information and communication technologies means that personal and social interaction tempt some to claim the death of distance. Travel (still costly albeit cheaper than in the past) is also growing, allowing the physical proximity we sometimes need to clinch deals or collect ideas, to share unique events, or just to spend time together. International migrants now use email and the Web, and make telephone calls using Skype—but that communication is a complement and not a substitute for visits home and from family members. Academics work on joint papers on the Web, but conferences become even more important as an opportunity for face-to-face contact that consolidates the trust needed for long-distance collaboration.

The United Nations has already defined more than half the world’s population as being urban, with rapid further growth forecast in urban populations. Face-to-face interpersonal interactions remain indispensable, and research on social interactions has strengthened the argument that the close proximity of economic, social, and cultural forces (and the density of social interactions) in cities is one, perhaps the most, important reason for cities’ continued growth and economic relevance.

1.1 FROM URBAN EXTERNALITIES TO URBAN INTERACTIONS

Economists typically emphasize the role of markets. Thus, urban economists focus on housing and labor markets and on the economic activities of households, firms, and public institutions that define modern economies. A common concern of economists is what to do if markets are not functioning well. A common cause of dysfunctionality in urban markets is widespread externalities—direct agent-to-agent interactions that are not mediated through the markets. Externalities are pervasive and naturally generated in urban settings with their high density of population and economic activity. Market outcomes in such cases are typically socially inefficient. It is possible to rearrange things and make some individuals better off without hurting others. An earlier urban economics and policy literature used the pervasiveness of allegedly negative externalities to justify the massive interventions in cities in the 1960s and 1970s that came to be known as urban renewal in the Western world and slum clearance in developing countries.¹

Some of these projects rejuvenated urban downtown areas; many others were disastrous. The character of the urban neighborhoods and urban life and lives destroyed has since been mourned as a lost positive force in those cities’ economic and social spheres. Economists and other social scientists now see many kinds of urban externalities instead as instances of social interactions. This broader term refers to preferences or tastes that individuals have for the types of other individuals near whom they live and for those individuals’ actions. Interactions may be undesirable, but they cannot be ignored. Urban amenities are not only attractive scenery, parks, and natural settings but also the characteristics, habits, and activities of individuals’ neighbors. The examples at the beginning of the chapter all involve such direct agent-to-agent interactions. Urban places acquire a life of their own as magnets for formal and informal activities. Some of these activities are so persistent that they confer some specialization on their particular locales, contributing to the vibrancy and variety of life in large cities. Some come to be seen by outsiders as characterizing the larger city. Such places attract professionals, tourists, and locals in varying proportions. Well-known locales in this sense include Soho and the City in London; the Left Bank and the Marais in Paris; Wall Street, Greenwich Village, the Garment District, and Brooklyn Heights in New York; Harvard Square in Cambridge, Massachusetts; Hollywood in Los Angeles; Ginza in Tokyo; the Grand Bazaar and Istiklal Caddesi in Istanbul; and Darb Al-Ahmar (the historic city) and Tahrir Square in Cairo.

Why do some urban activities produce great things? Peter Dougherty (2002, 19) urges economists to talk about cities not in the same way that psychologists talk about sex, that is, without taking the fun out of it. How can the tools of economics help explain the role of cities in bringing the vast variety of human creative resources together in an ongoing spontaneous and combustible mix? (Dougherty 2002, 18). Can rigorous theory support Florida’s (2002) claim that imaginatively selected measurable variables (such as the percentage of gays or of people with bohemian lifestyles) can explain a big part of a city’s attractiveness. Can economists marry thought to feeling so as to help in reaffirming the exciting connections that unite the historic wisdom of Adam Smith with city life? (Dougherty 2002, 19).

An answer needs to combine economic variables, such as prices, with noneconomic ones. Education and health are critical in individuals’ social personas and yet are components of human capital, an economic concept par excellence. The distinction between economic and noneconomic variables has become increasingly blurred, but in the analysis explored here the strength of economics is the rigor and discipline afforded by its theoretical and empirical tools.

The contemporary theory of social interactions is an important example of how these tools provide a powerful framework. Becker (1974) was one of the earliest economists to talk explicitly of social interactions; subsequently they were used extensively in empirical work. Loury (1982) pioneered using variables to measure the impact of community and family background on educational achievement. Yet, it was the Manski (1993, 2000) model that provided the canon for empirical modeling of social interactions. Manski’s approach provides a typology of social influences within individuals’ social milieus and raises key identification issues.

The Manski model distinguishes influences that emanate from: one, the decisions of members of one’s reference group (endogenous social effects), such as keeping up with the Joneses; two, the effects on an individual of characteristics of members of one’s reference group(s) (exogenous or contextual effects), as when individuals value living close to others with similar ethnic backgrounds, or with other characteristics they view as conducive to practices they themselves value; and three, individuals acting similarly because they have similar observable or unobservable characteristics, or face similar institutional environments (correlated effects). This book adds the role that prices play in conveying social effects to the categories proposed in Manski’s paper. It is precisely because individuals take the price of a good as given and beyond their control, making their decisions accordingly, that equilibrium prices ultimately reflect the characteristics of all market participants.

The fact that the actions of individuals in social contact with one another are interdependent is an important notion, and the concept of social interactions can be a powerful tool, as the following examples demonstrate. In seeking to explain one individual’s actions, we can no longer use just the actions (or choices) of neighbors as explanatory variables in a regression. Such magnitudes are not independent of the error. Instead, more elaborate econometric approaches are called for. Nonetheless, even when individuals choose their neighbors and thus their neighborhood effects, results by Brock and Durlauf (2001b) establish that it is possible to actually identify different social effects separately. To do that we need to correct appropriately for the selection bias associated with individuals’ having chosen their neighbors. Sometimes interactions are group-based, in which case individuals value aggregates describing entire communities and aggregates of the actions of the members of those communities. At other times, interactions are one-to-one. In the second case social network models can provide a critical focus on the microstructure of interactions. Heterogeneity in interactions across individual pairs is an important focus of the econometric analysis.

1.1.1 Location Decisions of Individuals

In deciding whether or not to locate in a particular city or neighborhood, each individual weighs numerous factors from their own perspective. These factors can be classified neatly as market variables, endogenous social effects, and contextual variables. When individuals decide where to locate, pursuing equilibrium strategies, their own individual characteristics contribute to defining the equilibrium values of prices and the distribution of characteristics by location. In the process individuals sort themselves into neighborhoods. Some of the sorting is sorting on observables. As Rosen (2002) underscores, it is important to assess such sorting in order to, inter alia, understand the social valuation of neighborhood amenities when individuals are heterogeneous. For example, if some people value neighborhood safety more than others, then those who value it less will sort to less safe neighborhoods. Estimates of the average value to society of neighborhood safety based on those who sort to more safe neighborhoods will be biased upward, while estimates based on those who locate in less safe neighborhoods will be biased downward. Most realistic settings with social interactions involve sorting on unobservables as well as observables. Social interactions models help us understand individuals’ location decisions, as well as membership decisions more generally.

The inherent difficulty in determining what drives the growth of cities is an example of the problem of correcting for sorting on unobservables. We want to know whether the factors that drive location decisions are due to the direct attraction of being near many others (agglomerative forces) or to underlying (unobserved) factors that those who make the location decision have in common. Economic geography provides examples where we can distinguish between the attraction of natural features of the landscape, first nature, and spatial features of the economic system, which include but are not limited to the effects of the landscape, second nature. I discuss the relative importance of first nature versus second nature and how it motivates empirical research at length in several chapters.

1.1.2 Location Decisions of Firms

Decisions made by firms, like those made by individuals, are influenced by factors resembling social interactions; this book exploits this similarity methodologically and links decisions of firms, in particular, with the theoretical underpinnings of new economic geography (NEG) (Fujita, Krugman, and Venables 1999). The case of firms introduces a new angle—spatially dispersed social interactions. The idea that firms interact in the context of the urban economy is an old concept, but to fully understand the benefits firms derive from being near other firms we need to articulate the origin of those benefits. In particular, economists since Marshall (1920) have asked whether proximity to other firms in the same industry generates an effect that is different from proximity to firms in other industries or from other factors such as proximity to a larger city or to a particularly suitable labor force. Moreover, numerous firms may be attracted by the same advantageous local factor, such as attributes of the local labor force. Similarly, workers may be attracted to a location by a factor in common with firms such as good weather and/or other physical amenities in addition to the job opportunities at that location. In such cases it may appear that a single common factor operates as a force of attraction for both individuals and firms.

Yet to understand what is really happening we must distinguish among the multiple types of attractions that are in fact involved. Distinguishing the attraction of other firms, for example, from the attraction of labor force characteristics or of first nature attributes of a place, such as the weather, can be critical for public policy choices that set out to encourage local economic development. If firms are attracted by the presence of a skilled labor force and those workers in turn are attracted to Silicon Valley by the weather, then investments that attempt to reproduce other aspects of that region in a midwestern city are likely to fail.

1.2 ECONOMIES OF CITIES AND NEW ECONOMIC GEOGRAPHY

Since individuals and firms benefit by locating in close spatial proximity to one another, it is fruitful to apply the analysis of social interactions in examining the economies of cities. The social interactions approach to the study of economies made up of cities is contributing much improved microfoundations that allow us to understand and predict how individual economic agents benefit from the size of the city where they live and work. Urban concentrations generate costs as well as benefits. The most obvious costs are those due to pollution and congestion. Two natural questions follow: How large should cities be? Will cities in free market economies attain their optimum sizes? The system-of-cities literature has dealt elegantly with these questions (Henderson 1974, 1977a, 1988a).

Questions about city size have attracted attention for a long time, at least since Plato and Aristotle (Papageorgiou and Pines 2000, 520). In The Laws, Plato (ca. 350 BC) sets the optimal city size precisely at 7! = 5,040 (male) citizens.² This number does not include optimal support personnel (women, children, slaves, and alien residents) whom we would include in the population and who would make the size of Plato’s city considerably larger. According to Aristotle’s (ca. 340 BC) Politics, optimal city size should be constrained from below by self-sufficiency: a city only comes into being when the community is large enough to be self-sufficing. If then self-sufficiency is to be desired, the lesser degree of unity is more desirable than the greater. And it should be constrained from above by efficiency. Too small a city cannot satisfy all the needs of its citizens; if it is too large, it becomes unwieldy. Thus, You cannot make a city of ten men, and if there are a hundred thousand it is a city no longer. But the proper number is presumably not a single number, but anything that falls between certain fixed point (Aristotle, Nicomachean Ethics, Book IX, 10, ca. 330 BC). Chapters 7 and 9 offer more modern perspectives on this issue. Using the tools of new economic geography and casting them in a system-of-cities model, Au and Henderson (2006) take a modern stand and show that Chinese cities are too small.

The system-of-cities approach I cited above adopts a market-based approach to optimal city size. Different industries located in a city all benefit from external economies. People need to commute to their places of work. That creates congestion costs (time wasted in traffic, noise, and air pollution). Each individual contributes more to total congestion than he or she experiences, thus generating a social cost of congestion. When cities specialize in producing a single product or a group of related products, congestion costs are lower: the software industry is not saddled with the social costs generated by the metal-processing industry, as it would be if both industries were to locate in the same city. It follows that cities should specialize once their survival is ensured. It is hard nowadays to think of cities without industries or marketable services. It is thus interesting to contrast with Plato’s proscription (accompanied by severe penalties) against the citizens’ being retail traders or merchants!

In most modern economies governments cannot directly regulate what different cities produce or who lives where. A variety of city types emerge including both industrially diversified and specialized cities. Local and national governments defer to political realities generating favorable treatment for particular cities and their hinterlands, especially via subsidized transportation and other infrastructure. It is thus important to be able to assess how such policies impact the urbanization process and the nature of outcomes in large economies.

Just as local increasing returns to scale are a driving force of the urban economy, similar forces underpin endogenous growth theory, that is, growth driven by endogenous technological change (Lucas 1988; Romer 1990). This research has built on increasing returns-to-scale technologies from plausible assumptions without ending up with an extreme and counterfactual market structure, such as an economywide monopoly. Spatial economics has dealt with a similar challenge so as to navigate carefully between a high concentration of activity in some locations and a low concentration in the rest of space. In hailing the value of proximity, Lucas (1988) credits Jane Jacobs (1969), whose writings had been treated as anathema by the earlier generation of economists.³ The use of increasing returns in these literatures is conceptually related to Adam Smith’s (1776) famous analysis of the division of labor and its being limited by the extent of the market. Urban economics also owes a lot to Alfred Marshall’s (1920) trilogy, now part of the canon. Local increasing returns could arise because of knowledge spillovers, linkages between input suppliers and final producers, and thick local labor market interactions.

1.2.1 New Economic Geography

Paul Krugman’s research and its early popularization in his Geography and Trade (Krugman 1991b), eloquently outlined in his Nobel lecture (Krugman 2008), contributed to the momentum of new economic geography. The approach seeks to integrate urban and regional economics, both in a national as well as an international context, and takes the form of economists’ directing their traditional tools to questions with space as a key dimension.

The emergence of regional disparities within an economy, especially when different regions share the same institutional framework (Kaldor 1970), is emphasized as a key puzzle, as are the origins of international inequalities. Recent interest by economists in European economic integration and in globalization has renewed interest in the study of regional, as opposed to national, phenomena. In the context of European integration, more generally, it is often argued that the abolition of economic borders will shift the playing field of economic interactions to regional entities. New economic geography addresses concerns such as, for example, whether improvements in transportation links intended to break the isolation of lagging regions may have the opposite effect, strengthening the forces of agglomeration in leading regions and thus further exacerbating regional inequalities.

1.3 URBAN STRUCTURE AND GROWTH

Urban agglomeration is a social invention determined by the interplay between the value of concentration relative to the cost of congestion. If the former dominates, spatially uniform steady states cannot sustain themselves. Agglomerations were originally limited by the need for genetically related individuals to live close to one another and to avoid encounters and unnecessary conflicts with strangers, a situation that reduced the attractiveness of large agglomerations (Seabright 2004). Social interactions within cities give rise to innovative ideas. The advantages of interactions themselves, as well as their fueling of technological progress and especially the advent of improvements in public health (Cairns 1997), however, came to outweigh the disadvantages of close proximity. Increasing interactions accommodate an ever finer division of labor that in turn mitigates hostility among unrelated individuals [cf. (Seabright 2004)].

An economy’s urban system is not a static entity. Populations grow, in part, for reasons that are endogenous to the economies that host them. A growing population will be accommodated in growing cities as well as in newly created urban settlements of all kinds. Technological change and infrastructural development can make existing cities function better and accommodate increased populations and diverse industries. Casual observation suggests that there is considerable arbitrariness in the location of cities. Why should Santa Fe, New Mexico, be where it is? For visitors and residents today, its charm is directly due to its location in the mountains of New Mexico. But is that why the city developed there? Natural features of the geographic landscape, such as access to waterways and natural harbors, are important. Proximity to natural or historically given hubs and being in a place where transshipment occurs (boat to rail; air to truck) allow a city to function as a cusp in total transport costs. Once established, a new city itself serves as a cusp for further development of the urban system. Even if the original cause is no longer present, a city rarely disappears.

Even within a mature urban system, existing cities may renew their prominence by reinventing themselves. Cities can also become obsolete, often because they are perceived as unattractive places to live, and when their industries relocate to more attractive sites nationally and internationally. Urban structure adapts through the birth, growth, and death of cities. Urban reinvention may not always prevent urban decay. Urban growth under certain conditions provides a margin that eliminates local increasing returns to yield constant returns to scale at the level of the national economy. This outcome helps reconcile the exploitation of increasing returns in an economy with non-explosive national economic growth (Rossi-Hansberg and Wright 2007). In this context, it is interesting to ask whether urban growth imposes restrictions on national economic growth.

1.4 URBAN INTERACTIONS, POLITICS, AND URBAN DESIGN

The interplay between the spatial and social configurations of cities is important in much of the book. The serendipity of interactions among urban dwellers is a big part of urban living. That public opinion formation is influenced by the topology of social interactions within existing social milieus is long-standing. For example, consider the observation by Doxiadis (1970, 398): Pericles in ancient Athens could get a reasonable sample of public opinion by meeting 100 to 150 people while walking from his home to the Assembly. Ober (2008) interprets the famous political reforms in classical Athens instigated by Cleisthenes by means of modern social network theory. He studies how the administrative rearrangements of the Cleisthenes reform, whereby urban, suburban, and rural communities were grouped together, allowed for artful mixing of opinions as representatives from distant communities sampled public opinion on their way to the agora in the central city. Nowadays, it is the media and social networking that help form public opinion, in addition to locally hosted interactions facilitated by civic associations and local governmental institutions, especially in Anglo-Saxon countries.

1.5 MOVING FORWARD

Many though not all of the questions rhetorically posed at the beginning of this chapter are dealt with formally in the book. Social interactions are the overarching theme that allows me to structure the book and helps embed it within the economics literature. While urban economics lends basic components to social interactions as an organizing principle, it is not the only branch of economics in which the social interactions approach is leading to significant advances. Labor economics, the economics of health, and the economics of education have benefited enormously from this perspective. So too have spatial economics and the economics of international trade. For example, individuals and firms benefit from being in a larger city because its economy can accommodate a greater variety of goods and services. They in turn allow for more attractive lifestyles, greater ability to innovate, and improved ways to mitigate risk. The role of city size serves as an important analytical link between the microbased chapters of the book and the more aggregative ones. Understanding international trade through the lens of an economy’s urban structure is a promising area of research, and so is understanding the forces of urban business cycles, a new area of research, where several chapters of the book propose promising new inroads. Yet above all, the book aims at integrating empirical findings, mainly by economists, and thus helps establish social interactions as a central tool of modern economics.

CHAPTER 2

Social Interactions

Theory and Empirics

2.1 INTRODUCTION

This chapter addresses the role of the social context in individual decisions. Many important markets continue to coexist with nonmarket arrangements. Social interactions, that is, nonmarket interactions, are ubiquitous, and social institutions do matter to an extent not fully appreciated by economics (Arrow 2009). Understanding the social consequences of economic decisions requires that we acknowledge their social context. With economics increasingly venturing into the traditional realms of other social sciences, recognizing the importance of social interactions can be particularly helpful in understanding a diverse set of phenomena, from obesity and cigarette smoking to economic inequality.

In the canonical case of individual decision making when goods and services are procured from markets, individuals are assumed to choose quantities of goods and services to maximize utility subject to a budget constraint. The basic model has been extended to allow for externalities, that is, direct effects from an individual to another that do not involve market transactions. In the presence of externalities, market prices may not reflect the full social value of the respective goods and services. For example, my neighbor’s playing loud music bothers me, and there is no direct market-mediated way for my unhappiness to be transmitted to him and hence to affect his behavior. This might prompt me to leave the area and perhaps to move near people whom I think are less likely to engage in behaviors that I find unpleasant or perhaps who are like me. When I rent a particular apartment in a multiunit complex or buy a home in a suburban subdivision, I can expect that my daily life will be affected by the behavior of my neighbors as they, too, go about their daily lives. Such effects are bundled with my choice of residence. My own actions will in turn affect the welfare and perhaps actions of my neighbors who are sensitive to them.

The part of the marginal value of a good that is due to its being appreciated by those consuming it is equal to the marginal cost to them of acquiring it. In competitive markets, it is also equal to the cost of producing an additional unit. Yet, an additional unit of the good may have adverse effects on some individuals and beneficial effects on others. Individuals’ preferences differ. Externalities can also be beneficial. My neighbor’s male winterberry plants help my female winterberry plants produce lovely berries profusely, and such neighborly habits improve the productivity of the apiary further down the street.

Even though the case of music playing bothering me does connote physical proximity, this need not be so for all externalities. There are examples of consumption practices by people far away raising objections on deeply felt ethical or religious grounds. Some people object to the hunting and consumption by others of meats of certain species even though these acts occur far away. This is the case of Japanese consumption of whale meat raising objections in some quarters in the United States. This example may be an instance of someone else’s consumption affecting my enjoyment, as a matter of principle or because I like to have the option of going on whale-watching trips.

It is arguably less well understood that externalities from some aspects of consumption (broadly construed) are critical for defining social structure and cohesiveness. These range from patriotic activities such as raising flags, displaying national symbols, and celebrating national holidays, to participating in music, sports, and other performances and cultural events. Such activities suggest sharing of values and personal tastes. It would be natural to suppose that people tend to cluster near others with similar values and tastes.

For example, the availability of a variety of different ethnic foods in supermarkets and restaurants is attractive for some but off-putting for others. Therefore, one would think that to the extent possible individuals who are free to choose where to locate will seek to be near others with like tastes and values and far from others with different ones. This may be due to several reasons: either pure preference for the values of others or anticipation that being near others with similar preferences will make it more likely that desirable goods will be readily available. These effects may coexist with externalities. For example, I might want to live near others who take good care of their yards and gardens or decorate their balconies and windows with beautiful flowering plants and keep up with maintaining their houses. I value living near others who are highly educated or artistically inclined because I enjoy engaging in intellectual or artistic casual conversations with my neighbors. Firms seeking to locate near other firms is a similar phenomenon.

I have implied so far that interpersonal effects are passive. They can also be deliberate. Individuals derive satisfaction from displaying their consumption activities conspicuously, perhaps regardless of whether or not others are positively influenced. This is an important phenomenon sometimes referred to as Veblen effects in consumption (Leibenstein 1950).

This book is about social interactions. As we shall see, distinguishing between different types of effects is important for drawing reliable conclusions from observing individual behavior and for designing policy. It is important to have a theory to guide us in interpreting the evidence from a variety of settings where individuals may seek deliberately to mix or to segregate. It is also important to be able to design different types of policy interventions.

The canonical formulation that I develop in this chapter can accommodate, in particular, phenomena that have been emphasized recently by such a diverse set of scholars as Christakis and Fowler (2009) and Wilson (2009). Specifically, Wilson (2009, 5) distinguishes two types of structural forces, social acts and social processes, and two types of cultural forces, national views and beliefs on race, and cultural traits, that is, shared outlooks, modes of behavior, traditions, belief systems, world views, values, skills, preferences, styles of self-presentation, etiquette, and linguistic patterns. These are seen, Wilson (2009, 15) adds, [as they] emerge from patterns of group interaction in settings created by discrimination and segregation and that reflect collective experiences within those settings. Prevailing outcomes associated with the phenomena that Wilson emphasizes as having race as a key salient factor can be modeled as group equilibrium outcomes for analytical convenience. However, they can reflect the full range of concerns described by Wilson. Social acts that Wilson defines as the behavior of individuals within society, including stereotyping, stigmatization, discrimination, and others, may be modeled as contextual effects or endogenous social effects, as when individuals conform to the behavior of others.

As another example, consider one of the phenomena discussed by Christakis and Fowler where it is vitally important to distinguish the spread of behavior from the spread of norms. Reaction to particular behaviors by others in individuals’ social milieus and adherence to norms are both instances of endogenous social interactions. In the case of obesity, as Christakis and Fowler (2009, 105–112) argue (and I discuss in further detail in section 2.7.2.3 below), it may be possible to distinguish between the spread of behavior and the spread of norms as the main force driving its social incidence provided that additional information on physical versus social proximity (and its direction) is utilized.

I proceed next by introducing a sequence of models that highlight applications in different empirical social interactions settings. I start with a simple static model, which I use to demonstrate the basic concepts of the social interactions approach, and then apply it to the case of coexistence, in a market context, of individual actions that are private with actions that have social consequences, and to endogenous networking. Social networks are jointly determined with individual actions. A special case of this model where the endogenous social structure is probabilistic allows me to link social interactions theory with social networks theory (including, in particular, random graph theory). I follow up with a dynamic model where the social structure accommodates a variety of social interaction motives. It is solved as a dynamic system of evolving individual actions. The solution links social interactions theory with spatial econometrics. I conclude with an appendix that surveys available data sets that lend themselves particularly well to social interactions studies.

2.2 A SIMPLE LINEAR MODEL

The empirical economics literature on social interactions addresses the significance of the social context in economic decisions. Decisions of individuals who share a social milieu are likely to be interdependent. Recognizing and identifying the origin and nature of such interdependence in a variety of conventional and unconventional settings and measuring empirically the role of social interactions pose complex econometric questions.

The actions of different individuals in a group are interdependent if they reflect the actions, or expectations of the actions, of all others in the group. This is known as an endogenous social effect (or interaction). This is the case when individuals care not only about the kinds of cars they themselves drive or the education they acquire but also about the kinds of cars or the education obtained by their friends. Therefore, their own decisions and those of others in the same social milieu are simultaneously determined. Individuals may also care about personal characteristics of others, that is, whether they are young or old, black or white, rich or poor, trendy or conventional, and so on, and about other attributes of the social milieu that may not be properly characterized as deliberate decisions of others. Such effects are known as exogenous social or contextual effects. I address below the particular difficulties that these different effects pose for estimation. In addition, individuals in the same or similar social settings tend to act similarly because they share common observable and/or unobservable factors or face similar institutional environments. Such interaction patterns are known as correlated effects. This terminology is due to Manski (1993), who emphasizes the difficulty of identifying econometrically endogenous effects separately from contextual effects in linear-in-means models, and social effects, endogenous or exogenous (contextual), from correlated effects.

Theorizing in this area lies at the interface of economics, sociology, and psychology and is often imprecise. Terms like social interactions, neighborhood effects, social capital, network effects, and peer effects are often used as synonyms although they may have different connotations. Empirical distinctions among endogenous, contextual, and correlated effects are critical for policy analysis because of the social multiplier, as I explain in more detail further below.

Joint dependence among individuals’ decisions and characteristics within a spatial or social milieu is complicated further by the fact that in many circumstances individuals in effect choose their own context. That is, in choosing their friends and/or their neighborhoods, individuals also choose their neighborhood effects. Such choices involve information that is in part unobservable to the analyst and therefore require making inferences among the possible factors that contribute to decisions (Brock and Durlauf 2001b; Moffitt 2001).

Let individual i’s action yi be a linear function¹ of a vector of observable individual characteristics, xi, of a vector of contextual effects, zν(i), which describe i’s neighborhood (or social milieu) ν(iamong the members of i’s neighborhood ν(i), the endogenous social effect, conditional on information known to i, ψi. That is,

where parameters α and θ are row vectors, α0 and β i is independent and identically distributed across observations.

I note that the endogenous social effect is defined with respect to the expectation of the average action within group ν(i). Abstracting at the moment from the issue that individual i may have deliberately chosen her group (or neighborhood), ν(i), and stating that conditional on individual characteristics, contextual effects, and the event that i is a member of neighborhood ν(ii is zero, allows me to focus on the estimation of such models. I assume social equilibrium within the group and that individuals hold rational expectations over ε[yi|Ψi]. That is, individuals’ expectations are confirmed; they are equal to what the model predicts. So, taking the expectations of both sides of (2.1) and setting the expectation of yi allows me to solve for this expectation, an endogenous variable. Substituting back into (2.1) yields a reduced form, an expression for individual i’s outcome in terms of all observables (xi, xν(i), Zν(i)):

Suppose that yi is i’s educational attainment. One’s socioeconomic characteristics, xi, typically do affect educational attainment. The socioeconomic characteristics of adult neighbors, including measures of economic success, are often used as contextual effects and are included in zν(i). They could stand for role model effects. In contrast, the effect of educational attainment by one’s peers in schools and neighborhoods, an endogenous social effect, is an example of a peer group effect. Note that these effects are associated with distinct populations and can be fully articulated in a dynamic model. See chapter 6, section 6.5.4.1, below.

Comparison of model (2.1) and its reduced form (2.2) shows clearly that endogenous social effects generate feedbacks that magnify the effects of neighborhood characteristics. That is, from (2.1), the effect of zν(i) on yi and thus magnified, if 0 < β < 1, relative to θ. Consider the effect on the academic performance of a particular medical student caused by the presence of women in the classroom, measured as a percentage. This problem is addressed by Arcidiacono and Nicholson (2005).² According to (2.1), the partial effect is given by θexactly as shown in equation (2.2).

Following the pioneering work of Datcher Loury (1982), a great variety of individual outcomes have been studied in the context of different notions of neighborhoods. This chapter seeks to show how to interpret findings of significant coefficients for contextual effects. The model in equation (2.1) is the bare minimum of interactions needed in order to express essential complexities of social interdependence. In practice, empirical researchers deal with models considerably more complex than (2.1). For example, it is possible that the marginal effect of a neighbor’s actions may depend on neighborhood characteristics. This can be expressed by an additional term zn(iin (2.1). See sections 2.3 and 2.6 below. Linearity obscures the richness that comes with nonlinear social interactions models like multiplicity of equilibria; see section 2.4 below.

2.2.1 Econometric Identification and Manski’s Reflection Problem

Including as contextual effects only neighborhood averages of individual effects, zν(i) ≡ xν(i), is a common practice but may cause failure of identification of endogenous separately from exogenous interactions. That is, we may not be able to estimate separately coefficients β and θ by means of a linear model like (2.1). Manski (1993) terms this the reflection problem: it arises because the direct effect of the social context variables zν(i. By imposing in equation (2.1) that zν(i) ≡ xν(i), that is, contextual effects coincide with neighborhood averages of individual characteristics, (2.2) becomes

.

The coefficient of xν(i. A statistically significant estimate of this coefficient in a reduced-form regression of individual outcomes on individual characteristics and neighborhood averages of individual characteristics (xi, xν(i)) allows a researcher to infer that at least one type of social interaction is present: β is nonzero and there is an endogenous effect, or θ is nonzero and there is a contextual effect, or both. Therefore, partial identification is possible for some type of social effect. This instance of failure of identification is a direct consequence of the linearity of the endogenous social effect in the behavioral model and of the unobservability of the expectationin (2.1).

If the underlying economic model suggests that some neighborhood averages of individual covariates should be excluded from zν(i), then the econometric model is identified. More precisely, for the identification of (2.1), the vector xν(i) must be linearly independent of (1, xi, zν(i)). It is thus necessary that there be at least one element of xν(i) whose group-level average is not a causal effect and therefore not included in zν(i).

When individuals belong to different groups, there could well be group-level heterogeneity that might not necessarily arise from group-level social interactions. Graham (2008a) proposes a method that separately identifies the social interactions component of any excess variance from that due to group-level heterogeneity and/or sorting. To see this in simple terms, suppose groups come in singletons or in pairs. Let the outcome for singletons be yi i. The outcome for individual i in a pair {i, −i} is yi = βy−i ii , identifies β.

Graham (2008a) reports an application, based on data from Project STAR, where kindergarten students and teachers were randomly assigned to large and small classrooms. The performance of talented students is typically offset by that of below-average students, resulting in little variation in mean student ability in large classrooms. In small classrooms, however, groups composed of mostly above- or below-average students are more frequently observed, generating greater variation in mean ability. As a result, the variance of peer quality is greater across the set of small than across the set of large classrooms, while the random assignment of teachers ensures that the distribution of their characteristics is similar across the two types of classrooms. Graham decomposes the unconditional between-group variance of outcomes into the sum of three terms. The first term equals the variance of any group-level heterogeneity. In Graham’s application, that could be due to teacher quality. The second term equals the between-group variance of any individual-level heterogeneity. In Graham’s application, that is the variance of average student ability across classrooms. It is the third term that reflects the strength of any social interactions. When social interactions are present, between-group variation in outcomes should mirror between-group variation in peer quality. The third term therefore depends on the variance of peer quality across groups. When group sizes differ, as they do in the Project STAR-based data that Graham uses, it is possible to identify the endogenous social effect. Graham (2008b) reports evidence of social interactions; that is, differences in peer group quality were an important source of individual-level variation in academic achievement for Project STAR kindergarten students. Lee (2007) examines in detail the econometric properties of models where group sizes differ exogenously.

2.2.1.1 The Social Multiplier

The fact that social interactions, exogenous and endogenous, help amplify differences in average neighborhood behavior across neighborhoods can itself serve as a basis for identification. Glaeser, Sacerdote, and Scheinkman (2003) use patterns in the data to estimate a social multiplier.⁴ For an incremental change in a particular fundamental determinant of an outcome, the social multiplier is defined as the equilibrium effect in the social group to the direct effect on each individual. In addition to the direct effect on an individual, this includes the sum total of the indirect effects through the feedback from the effects on others in the social group.

To see this clearly, consider the group-level counterpart of equation (2.1) with θ = 0, that is,

yν(i) = α’0 + Xν(iν(i),

ν(i) is suitably defined. For simplicity let x be a scalar. Put crudely, an estimate of the multiplier could be seen as the ratio of the group-level coefficient to the individual-level coefficient, the coefficient of xi . The group-level regression may be seen as being obtained by summing up the reduced forms according to equation (2.2) over all members of each group. As a result, the coefficient of xν(i. The multiplier is

A more precise estimate of the multiplier requires that one account for sorting. Blume, Brock, Durlauf, and loannides (2011, 885) show that the ratio of the coefficient associated with xν(i) in agroup-level regression of neighborhood outcomes on neighborhood attributes (yν(i) on xν(i) to the individual-level coefficient associated with xi when regressing yi, on xi is equal to 1/1 − β + σsβ, where σs = Cov(xi, xν(i))/Var(xi) corrects for the portion of the variation in individual attributes due to the group-level variation. With random sorting, σs = 0, we are back at (2.3). Therefore, one can obtain an estimate of β from the ratio of group-level to individual-level regression coefficients and an estimate of σs. If sorting is perfect, on the other hand, that is, groups are perfectly segregated, σs = 1 and the multiplier is equal to 1 and thus smaller.

It follows that an estimated social multiplier greater than 1 implies magnification of the direct effect and thus endogenous social interactions, 0 < β < 1. This estimate is positive if the underlying social equilibrium is stable, a condition that Glaeser, Sacerdote, and Scheinkman (2003) term moderate social influence. As Burke (2008) emphasizes, while much of the literature on the social multiplier so far rests on linear models in static settings, the concept may be extended to dynamic settings (Binder and Pesaran 2001), to nonlinear settings (Brock and Durlauf 2001a), to settings with complete versus incomplete information (Bisin, Horst, and Özgür 2004), and to economies with more complex interaction topologies (Ioannides 2006). So far, my emphasis is on measuring the strength of social interactions but not necessarily their topology or the dependence of the feedback on incomplete access to information by different agents. Bisin, Horst, and Özgür (2004) show that incomplete information has the effect of dampening the aggregate effects of the agents’ preferences for conformity, thus reducing the social multiplier relative to complete information.

The above discussion shows that in measuring the social multiplier one must deal, in practice, with dependence across decisions of individuals belonging to the same group. This occurs with nonrandom sorting in terms of observables and of unobservables. If educated people prefer to have other educated people as neighbors, the effect of one person’s education (in an individual-level regression) will overstate the true impact of education because it includes spillovers. So, with sorting on observables and positive social interactions, the individual-level coefficient will overstate the true individual-level relationship and the estimated social multiplier will tend to underestimate the true level of social interactions. On the other hand, correlation between aggregate observables and aggregate unobservables will cause the measured social multiplier to overstate the true level of social interactions.

The social multiplier approach is particularly useful in delivering a range of estimates for the endogenous social effect especially when individual data are hard to obtain, as in the case of crime data. Glaeser, Sacerdote, and Scheinkman (1996) motivate their study of crime and social interactions by the extraordinary variation in the incidence of crime across U.S. metropolitan areas over and above apparent differences in fundamentals. If social interactions in criminal behavior are present, variations in observed outcomes are larger than what would be expected from variations in underlying fundamentals, precisely because of the social multiplier. Their results show that the estimated interactions coefficient is highest for petty crimes and declines for more serious crimes to become negligible for the most serious ones. Across cities, the implied extent of interactions is roughly constant.

Glaeser, Sacerdote, and Scheinkman (2003) report results using a multiplier-based model with three different alternative outcomes. One is fraternity/sorority participation by students at Dartmouth College. This setting exploits the advantage that students are randomly assigned to residences at Dartmouth College; in other words, there is no sorting. So, aggregating at the room, floor, and dormitory levels allows these researchers to apply the multiplier technique in the presence of random group assignments. The coefficient of having drunk beer in high school as an explanatory variable in regressions with fraternity/sorority participation as a dependent variable rises with the level of aggregation due to reduced sorting, exactly as the model predicts. This allows them to predict the endogenous social interaction effect of beer drinking associated with large multipliers. A second outcome they study is crime, for which individual data are not reliable. These researchers regress actual crime rates against predicted crime rates, which are formed by multiplying percentages of U.S. individuals in each of eight age categories by the estimated crime rate of persons in that category. They perform such regressions at the level of U.S. county and U.S. state cross-sectionally, and for the entire United States over time. Their results imply large social multipliers that increase with the level of aggregation, specifically from 1.72 at the county level to 2.8 at the state level to 8.16 at the national level. The basic theory would predict that these estimates are consistent with large endogenous social interaction coefficients. Working with data on wages and human capital variables, these authors again find further evidence of large social multipliers. The authors are aware of the fact that their results should be accepted cautiously because they do not control for sorting on unobservables, which may increase with the level of aggregation.

2.2.2 Identification of Social Interactions with Self-Selection to Groups and Sorting

The presence of nonrandom sorting in terms of unobservables is a major challenge for the econometric identification of social interactions models. The deliberate choice of a neighborhood, ν(i), by individual i i|xi, zν(i); Ψi; i ∈ν(i)] ≠ 0 and