Presentation on theme: "Addressing the problem of endogeneity in estimating area effects Simon Burgess University of Bristol Leverhulme Centre for Market and Public Organisation."— Presentation transcript:
Addressing the problem of endogeneity in estimating area effects Simon Burgess University of Bristol Leverhulme Centre for Market and Public Organisation
July 2004www.bris.ac.uk/Depts/CMPO2 Introduction Does where you live matter for your future prospects? Academic and policy interest in neighbourhoods and their potential impact on people. But important statistical problems in estimating these.
July 2004www.bris.ac.uk/Depts/CMPO3 Plan What are the big questions? What is the problem of endogeneity? Addressing endogeneity What is the evidence? Conclusion
July 2004www.bris.ac.uk/Depts/CMPO4 Big questions Distinguish two related questions: Area effects -- Does where you live matter for your future prospects? –When growing up – childhood environment influences adult outcomes –When adult – environment influences later outcomes Peer group effects -- Does who you work with or play with matter?
July 2004www.bris.ac.uk/Depts/CMPO5 Related through suggested behavioural processes, and through the nature of the statistical problem. Behavioural underpinning: –Role models, information transfer, social norms, (education finance), … Statistical problem: –Correlation of error term and key variables
July 2004www.bris.ac.uk/Depts/CMPO6 Definitions –Area effects or neighbourhood effects: the nature of your neighbourhood has a causal influence on some aspect of your life chances. –Contrast with a different line of enquiry – estimating the degree of spatial autocorrelation, that is, the degree of clustering of people with particular characteristics.
July 2004www.bris.ac.uk/Depts/CMPO7 What is the problem? Three main issues: Omitted common variables Simultaneity or reverse causality –Reflection problem Selection Similar formal structure
July 2004www.bris.ac.uk/Depts/CMPO8 Omitted common variables Example: –Individuals discount rate or willingness to invest and take a long view correlated with acquisition of skills and later earnings, and also with decision not to smoke. Simultaneity or reverse causality Example: –School resources influence school outcomes; school outcomes influence school resources.
July 2004www.bris.ac.uk/Depts/CMPO9 Reflection problem Example: –Your behaviour influences mine and mine influences yours (Manski, REStudies, 1993) Selection Example: –Individuals choice of neighbourhood depends on their outcome and their characteristics. All mean that error term in simple regression is correlated with area variable, leading to bias. Extreme case in reflection problem.
July 2004www.bris.ac.uk/Depts/CMPO10 Addressing endogeneity Experiments Quasi-experimental methods Find specific groups with constrained choice Statistical methods: –Instrumental variables –Difference-in-difference –Matching
July 2004www.bris.ac.uk/Depts/CMPO11 How method addresses the core statistical issue Advantages Disadvantages Role of longitudinal data Addressing endogeneity
July 2004www.bris.ac.uk/Depts/CMPO12 What is the evidence? Quasi-experimental and observational studies have different results. Durlauf Neighborhood Effects –http://econ.pstc.brown.edu/faculty/henderson/handbook.htmlhttp://econ.pstc.brown.edu/faculty/henderson/handbook.html Dietz The estimation of neighborhood effects in the social sciences: An interdisciplinary approach –Social Science Research, p Moving to Opportunity results: –http://www.mtoresearch.org/index.htmlhttp://www.mtoresearch.org/index.html
July 2004www.bris.ac.uk/Depts/CMPO13 Conclusion Important set of issues Formidable statistical problems No single best way