Richard B. Davies

New methods for the analysis of sex ratio data independent of the effects of family limitation

A stochastic model has been described which allows for arbitrary Poisson, Lexian and family limitation effects. This model not only provides a framework for hypothesis tests concerning the determinants of sex ratio data but provides parameter estimates of considerable interest. Application of the model to Renkonens data suggests the presence of Lexian and family limitation effects but the absence of Poisson effects.

Longitudinal versus Cross-Sectional Methods for Behavioural Research: A First-Round Knockout

It is explained that statistical problems over omitted variables and endogenous factors cast doubt upon the wisdom of attempting inference about many forms of human behaviour from cross-sectional data. Simulations are used to confirm that cross-sectional analyses can produce seriously misleading results. The problems are avoided by using longitudinal methods.

A Generalised Beta-Logistic Model for Longitudinal Data with an Application to Residential Mobility

The beta-logistic model described by Heckman and Willis in 1977 is generalised to include feedback effects and time-varying variables. The resulting model provides a flexible and analytically tractable framework for inference from longitudinal discrete-time data on recurrent choice. Moreover, consistency with the cross-sectional logit model ensures theoretical plausibility. The statistical and practical problems that have to be confronted in making the model operational are illustrated by an application to data on residential mobility.

Heterogeneity, Nonstationarity, and Duration-of-Stay Effects in Migration

This paper briefly outlines the theoretical problem of the identification of duration-of-stay effects in migration. The empirical studies are critically reviewed with particular emphasis upon their treatment of heterogeneity and nonstationarity. New statistical tests developed by the authors are applied to Wisconsin migration data. These tests suggest that although heterogeneity and nonstationarity are strongly present within the data, duration-of-stay effects appear weak.

SABRE: Software for the Analysis of Binary Recurrent Events

In this paper we describe a GLIM-like software package (SABRE) for modelling binary recurrent events. The package fits a mixture model which allows for residual heterogeneity by the inclusion of a case- specific nuisance parameter in the linear predictor. Gaussian quadrature methods are used to integrate out this nuisance parameter. The use of SABRE is illustrated with data from the Social Change and Economic Life Initiative.

A family of hypothesis tests for a collection of short event series with an application to female employment participation

Data comprising a collection of short event series are increasingly encountered in social science research. Such series may be expected to be heterogeneous and nonstationary precluding conventional inferential methods. Tests are presented for homogeneity, nonstationarity, and zero order, with appropriate controls. The test procedures are based upon the subdivision of each series into a ‘conditioning sequence’ and an ‘experimental observation’. The tests are applied to data on labour force participation by married women.

Identification of some recurrent choice processes

Many social processes may be studied as collections of binary series. The processes which generate such series may be heterogeneous over the population, non‐stationary over time and may exhibit a feedback effect, such as “cumulative inertia.” Methods, using maximum likelihood estimation, are developed which model arbitrary heterogeneity and non‐stationarity to provide an effective test of the presence of “true contagion” or feedback effects. Where absent, a parametric form of the heterogeneity mixing distribution may be empirically identified using moment derived criteria. Application to ten year migration histories for 10,000 residents of Wisconsin, U.S.A., identifies “true contagion” amongst the younger, but not the older, residents for both owners and renters. An Su type distribution is identified as appropriate for modelling the heterogeneity amongst the older renters whereas the older owners are shown to be homogeneous.

The Mover-Stayer Model

Recent developments in nonparametric marginal likelihood have generated a very general, but readily operationalized, method of overcoming the nuisance parameter problem in stochastic models. Theoretical, empirical, and simulation analyses show that the nonparametric approach seriously undermines the modeling advantages traditionally associated with the mover-stayer model. Moreover, the goodness-of-fit success often achieved by the mover-stayer model is shown to have a plausible explanation not requiring a true mover/stayer dichotomy in the population.

Accounting for omitted variables in a discrete time panel data model of residential mobility

The problems of omitted variables in the empirical analysis of dynamic social processes is explained and alternative approaches to its solution briefly reviewed. The beta-logistic model of Heckman and Willis (1977) is introduced and then generalised for application to processes which include event history effects and time-varying covariates. The problem of initial conditions is explained. The generalised beta-logistic model is applied to residential mobility histories of households from the Michigan Panel Study of Income Dynamics.

Modelling the evolution of heterogeneity in residential mobility

Migration probabilities are known to vary over the population (heterogeneity) and over time (nonstationarity). It is shown that if the heterogeneity is represented by an SB mixing distribution and the nonstationarity by a set of arbitrary logistic scaling functions, then not only may all sources of heterogeneity, including tastes, be modelled, but heterogeneity at different points in time may be readily compared. The model is calibrated using data previously published by Clark et al. (1977, 1979) for a sample of 1,176 older renters and is found to fit well. The-changing heterogeneity over the ten-year period of observation is represented graphically.