tailieunhanh - Handbook of Econometrics Vols1-5 _ Chapter 29

Chapter 29 ECONOMETRIC ANALYSIS OF LONGITUDINAL DATA In analyzing discrete choices made over time, two arguments favor the use of continuous time models. (1) In most economic models there is no natural time unit within which agents make their decisions and take their actions. Often it is more natural and analytically convenient | Chapter 29 ECONOMETRIC ANALYSIS OF LONGITUDINAL DATA JAMES J. HECKMAN University of Chicago and NORC BURTON SINGER Yale University and NORC Contents 0. Introduction 1690 1. Single spell models 1691 . Statistical preliminaries 1691 . Examples of duration models produced by economic theory 1695 . Conventional reduced form models 1704 . Identification and estimation strategies 1710 . Sampling plans and initial conditions problems 1727 . New issues that arise in formulating and estimating choice theoretic duration models 1744 2. Multiple spell models 1748 . A unified framework 1748 . General duration models for the analysis of event history data 1753 3. Summary 1759 References 1761 This research was supported by NSF Grant SES-8107963 and NIH Grant NIH-1-RO1-HD16846-01 to the Economics Research Center NORC 6030 S. Ellis Chicago Illinois 60637. We thank Takeshi Amemiya and Aaron Han for helpful comments. Handbook of Econometrics Volume HI Edited by Z. Griliches and . Intriligator Elsevier Science Publishers BV 1986 1690 J. J. Heckman and B. Singer 0. Introduction In analyzing discrete choices made over time two arguments favor the use of continuous time models. 1 In most economic models there is no natural time unit within which agents make their decisions and take their actions. Often it is more natural and analytically convenient to characterize the agent s decision and action processes as operating in continuous time. 2 Even if there were natural decision periods there is no reason to suspect that they correspond to the annual or quarterly data that are typically available to empirical analysts or that the discrete periods are synchronized across individuals. Inference about an underlying stochastic process that is based on interval or point sampled data may be very misleading especially if one falsely assumes that the process being investigated operates in discrete time. Conventional discrete choice models such as logit and probit when .