ERIC Number: ED173212
Record Type: RIE
Publication Date: 1978-Dec
Reference Count: 0
Final Report for Dynamic Models for Causal Analysis of Panel Data. Alternative Estimation Procedures for Event-History Analysis: A Monte Carlo Study. Part III, Chapter 5.
Carroll, Glenn R.; And Others
This document is part of a series of chapters described in SO 011 759. The chapter examines the merits of four estimators in the causal analysis of event-histories (data giving the number, timing, and sequence of changes in a categorical dependent variable). The four procedures are ordinary least squares, Kaplan-Meier least squares, maximum likelihood (ML) and partial likelihood (PL). Eight sections comprise the report. Section I introduces the problem. Section II presents a short description of the event-history model and its appropriate use. The model is appropriate when data containing exact timing of events is available across units. For example, a political sociologist may want to test the hypothesis that the rate of collective violence in nation-states increases with the power of a state when the level of economic development is held constant. A report of Monte Carlo experiments designed to compare the four procedures follows. Topics include small sample properties, normal, lognormal, and uniform exogenous variables, random gamma disturbance, and time-dependent disturbances. Results indicate that both ML and PL procedures yield similar estimates, especially at high levels of censoring. ML estimation is slightly superior for time-independent data with or without random disturbances and PL estimation performs slightly better when the rate is time-dependent. (Author/KC)
Publication Type: Reports - Research
Education Level: N/A
Sponsor: National Inst. of Education (DHEW), Washington, DC.
Authoring Institution: Stanford Univ., CA.
Note: For related documents, see SO 011 759-772