Estimating Multistate Transition Hazards from Last-Move Data

Abstract

Following United Nations recommendations, many countries collect or publish internal migration data in last-move form, despite continuing uncertainty among researchers about how to estimate transition rates and probabilities from such information. “Last-move” data are a form of retrospective event history in which the only available information for each observational unit are the state at the time of a survey (ω), the last previous state (ψ), and the time at which the ψ → ω transition occurred. The statistical literature has addressed special cases, but there is still no general method for estimating transition hazards from last-move data. In this article I propose such a method, analyze its performance in a Monte Carlo simulation study, and apply it to migration data from Brazil's 1980 census.

Key Words:

• Backward recurrence times
• Censored data
• Migration
• Multistate demography
• Open intervals
• Survival analysis