Partial Likelihood Analysis of Time Series Models, with Application to Rainfall-Runoff Data,
Abstract
A general logistic-autoregressive model for binary time series or longitudinal responses is presented, generalizing the discrete-time Cox (1972) model with time-dependent covariates as well as the recent regression models of Kaufmann (1987) for categorical time-series. Since this model is formulated in terms of the time-series covariates which are not themselves explicitly modelled, the large-sample theory of parameter-estimation must be justified by means of Partial Likelihood in the sense of Cox (1975), using theoretical results like those of Wong (1986). The large-sample theory also justifies goodness of fit tests analogous to the chi-squared tests of Schoenfeld (1980) and to the tests based on sums of (normalized) squared residuals used in logistic regression. These ideas are illustrated by analysis of a rainfall-runoff hydrological dataset previously analyzed by Yakowitz (1987).
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 25, 1988
- Accession Number
- ADA191157
Entities
People
- Benjamin Kedem
- Eric V. Slud
Organizations
- University of Maryland