This was a two-year research project (June 2003 - May 2005) under the ESRC Research Methods programme. The project involved the development of multilevel simultaneous equations models for the analysis of correlated event histories. The research was motivated by a study of the interrelationships between partnership (marriage or cohabitation) durations and decisions about childbearing, using event history data from the 1958 and 1970 British birth cohort studies. Methodology developed under the project was implemented in MLwiN.
This project was rated ‘outstanding’ by the ESRC.
The aim of the project was to develop a general framework for the analysis of correlated event histories. The research was motivated by a substantive research question, the link between partnership transitions and childbearing among British women. Of particular interest was the effect of the presence of children on the probability that a marital or non-marital union is dissolved and the probability that a cohabiting partnership is converted to marriage. Previous research had found that women with children are less likely to separate than childless women. However, most of this research was based on the assumption that prior fertility outcomes are exogenous to the risk of separation. This assumption will be invalid if the processes of childbearing and partnership dissolution are correlated, and the correlation is not explained by covariates included in the model. Failure to account for this source of correlation between processes may lead to biased estimates of the effects of fertility outcomes on the hazard of separation.
A US study (Lillard 1993) used a multiprocess (simultaneous equations) model to allow for potential correlation between the risk of marital dissolution and the probability of having a child within marriage. We extended their approach to include non-marital partnerships and the childbearing that takes place within these unions. While marriage remains the favoured setting for childbearing in North America and much of Western Europe, the increasing popularity of unmarried cohabitations has contributed to a rise in extra-marital births. It is therefore important to consider the relationship between childbearing and dissolution of both marital and non-marital partnerships. Also of interest was the link between childbearing and the movement from cohabitation to marriage.
We proposed a multilevel multistate competing risks event history model (Steele et al. 2004) to study predictors of the dissolution of marital and non-marital unions and movements from cohabitation to marriage. A multilevel model allows for correlation between the durations of multiple partnerships, and the intervals between multiple births, contributed by the same woman. A multistate model was used to distinguish between two partnership states: marriage and cohabitation. Competing risks arise when considering transitions from the cohabitation state, since a period of cohabitation may end in separation or marriage to the same partner. Using a simultaneous equations model these partnership transitions are modelled jointly with fertility.
The multiprocess model consists of a system of equations, one for each partnership transition and for childbearing within marital and non-marital unions. Each equation contains a woman-specific random effect or residual which represents unobserved characteristics of the woman. These random effects are permitted to correlate across equations to allow for correlation between the unobserved woman-level characteristics that affect each process.
The discrete-time event history model used can be framed as a multilevel multinomial model which can be estimated using existing software, e.g. MLwiN.
The project used data from the 1958 and 1970 British birth cohort studies. Partnership and pregnancy histories are available for ages 16-42 for the 1958 cohort, and for ages 16-29 for the 1970 cohort. One aim of the project was to compare the experiences of the two cohorts in early adulthood.
Lillard, L.A. (1993). Simultaneous equations for hazards: Marriage duration and fertility timing. Econometrics, 56: 189-217.
Steele, F., Goldstein, H. and Browne, W. (2004). A general multilevel multistate competing risks model for event history data, with an application to a study of contraceptive use dynamics. Statistical Modelling, 4 (2): 145-159.
A workshop on multilevel event history analysis was held in February 2005 at the Institute of Education and repeated in March 2006 at the University of Bristol.
These workshops introduced discrete-time methods for the analysis of event history data. Advanced topics such as modelling transitions between multiple states and correlated event histories were discussed. The emphasis was on applying these methods in practice and interpreting the results.
The practical exercises used data derived from the National Child Development Study and the 1970 British Birth Cohort Study. These datasets are available for download via the UK Data Archive.
Multilevel Event History Analysis Training Datasets - Go to data-archive web site to download