Duranton / Henderson / Strange Handbook of Regional and Urban Economics

1. Auflage 2015
ISBN: 978-0-444-59539-3
Verlag: Elsevier Science & Techn.
Format: EPUB
Kopierschutz: Adobe DRM (»Systemvoraussetzungen)

E-Book, Englisch, Band Volume 5A-5B, 2064 Seiten

Reihe: Handbook of Regional & Urban Economics

Developments in methodologies, agglomeration, and a range of applied issues have characterized recent advances in regional and urban studies. Volume 5 concentrates on these developments while treating traditional subjects such as housing, the costs and benefits of cities, and policy issues beyond regional inequalities. Contributors make a habit of combining theory and empirics in each chapter, guiding research amid a trend in applied economics towards structural and quasi-experimental approaches. Clearly distinguished from the New Economic Geography covered by Volume 4, these articles feature an international approach that positions recent advances within the discipline of economics and society at large. - Emphasizes advances in applied econometrics and the blurring of 'within' and 'between' cities - Promotes the integration of theory and empirics in most chapters - Presents new research on housing, especially in macro and international finance contexts

Autoren/Hrsg.

Duranton, Gilles

Henderson, Vernon

Strange, William

Leseproben

Chapter 1 Causal Inference in Urban and Regional Economics
Nathaniel Baum-Snow*; Fernando Ferreira†    * Department of Economics, Brown University, Providence, RI, USA
† The Wharton School, University of Pennsylvania, Philadelphia, PA, USA Abstract
Recovery of causal relationships in data is an essential part of scholarly inquiry in the social sciences. This chapter discusses strategies that have been successfully used in urban and regional economics for recovering such causal relationships. Essential to any successful empirical inquiry is careful consideration of the sources of variation in the data that identify parameters of interest. Interpretation of such parameters should take into account the potential for their heterogeneity as a function of both observables and unobservables. Keywords Casual inference Urban economics Regional economics Research design Empirical methods Treatment effects JEL Classification Code R1 1.1 Introduction
The field of urban and regional economics has become much more empirically oriented over recent decades. In 1990, 49% of publications in the Journal of Urban Economics were empirical, growing to 71% in 2010. Moreover, the set of empirical strategies that are most commonly employed has changed. While most empirical papers in 1990 only used cross-sectional regressions, articles in 2010 were more likely to use instrumental variables (IV), panel data, and nonlinear models. Furthermore, special attention is now paid to the employment of research designs that can plausibly handle standard omitted variable bias problems. While only a handful of papers attempted to deal with these problems in 1990, more than half of the empirical publications in 2010 used at least one research design that is more sophisticated than simple ordinary least squares (OLS), such as difference in differences (DD), matching, and IV, to recover causal parameters. However, the credibility of estimates generated with these more sophisticated techniques still varies. While, in general, the credibility of empirical work in urban economics has improved markedly since 1990, many studies continue to mechanically apply empirical techniques while omitting important discussions of the sources of identifying variation in the data and of which treatment effects, if any, are being recovered. Table 1.1 details the percentages of publications in the Journal of Urban Economics that were empirical and the distribution of empirical methods used for the years 1980, 1990, 2000, and 2010. Table 1.1 Prevalence of empirical methods in the Journal of Urban Economics, 1980–2010 1980 57% 87% 10% 3% 0% 0% 0% 0% 1990 49% 79% 17% 13% 4% 0% 0% 0% 2000 62% 64% 32% 36% 14% 4% 0% 0% 2010 71% 77% 46% 26% 62% 8% 3% 5% Notes: Authors calculations from all published articles in the Journal of Urban Economics in the indicated years. This chapter discusses the ways that researchers have successfully implemented empirical strategies that deliver the most credible treatment effect estimates from data sets that describe urban and regional phenomena. Our treatment emphasizes the importance of randomization, which has been more broadly recognized in other fields, most notably development economics. Randomized trials are an important tool to recover treatment effects, especially those of interest for policy evaluation (Duflo et al., 2008). However, it is typically more challenging and expensive to implement field experiments in settings of interest to urban and regional economists, as it is in other fields such as labor economics. General equilibrium effects, which contaminate control groups with influences of treatment, are more likely to arise in urban settings. Moreover, the nature of such general equilibrium effects is more likely to be the object of inquiry by urban and regional researchers. Labor economists have typically adopted higher standards for evaluating the credibility of estimated causal effects in research that uses nonexperimental data. Here we explore identification strategies that have been successfully used to recover credible estimates of treatment effects, typically in the absence of experimental variation. These include DD, various fixed effects methods, propensity score matching, IV, and regression discontinuity (RD) identification strategies. We also discuss treatment effect heterogeneity and how differences in results across identification strategies may simply reflect different causal relationships in the data. We emphasize that especially without experimental variation (and even often with experimental variation), no one identification strategy is ever perfect. Moreover, when considering causal effects of treatments, it is useful to think in the context of a world in which a distribution of treatment effects exists. Selection into treatment (on both observable and unobservable characteristics) and treatment effect heterogeneity makes empirical work complicated. One recurring theme of this chapter is the following principle, which applies to all empirical strategies: it is crucial to consider the sources of variation in the treatment variables that are used to recover parameters of interest. Distinguishing this “identifying variation” allows the researcher to consider two central questions. First, could there be unobserved variables that both influence the outcome and are correlated with this identifying variation in the treatment variable? If such omitted variables exist, coefficients on the treatments are estimated as biased and inconsistent. We typically label such situations as those with an “endogeneity problem.” Second, how representative of the population is the subset of the data for which such identifying variation exists? If clean identification exists only in a small unrepresentative subset of the population, coefficients on treatment variables apply only narrowly and are unlikely to generalize to other populations. Throughout the chapter, we discuss the key properties of various identification strategies mostly assuming a simple linear data-generating process which allows for heterogeneous treatment effects. Each section cites articles from the literature for readers interested in the details of more complex applications. This structure allows us to easily explain the relationships between different empirical strategies while leaving space to cover applications in urban and regional economics. In each section, we illustrate best practices when implementing the research design by discussing several recent examples from the literature. Given the importance of the use of economic models to aid in the specification of empirical models and interpret treatment effect estimates, we view the material on structural empirical modeling in Chapter 2 as complementary to the material discussed in this chapter. Chapter 2 also considers the recovery of causal relationships in urban and regional data, but through making use of model formulations that are more involved than those considered in this chapter. The advantage of the structural approach is that it allows for the recovery of parameters that could never be identified with observational or experimental data alone. Estimates of a model's “deep” parameters facilitate evaluation of more sophisticated counterfactual simulations of potential policy changes than is possible with the less specific treatment effect parameters considered in this chapter. However, structural models are by their very natures full of assumptions that are most often stronger than the assumptions needed to make use of randomization to recover treatment effects. Additionally, because models can always be misspecified, such theory-derived treatment effects may be less credible than those whose data-based identification we discuss here. When possible, we present a unified treatment of causal relationships that can be interpreted in the context of an economic model or as stand-alone parameters. While the field of urban economics has made considerable progress recently in improving its empirical methods, we hope that this chapter promotes further advances in the credibility of our empirical results by encouraging researchers to more carefully consider which particular treatment effects are being identified and estimated. In defense of our field, it is fortunately no longer acceptable to report regression results without any justification for the econometric identification strategy employed. Nonetheless, we hope we can go beyond this admittedly low bar. This includes dissuading ourselves from simply trying several instruments and hoping for the best without careful thought about the conditions under which each instrument tried is valid or the different causal effects (or combinations thereof) that each instrument may be capturing. This chapter...