How to apply
List of First Supervisors
Families and children and human capital
- Nicola Barban
- Cara Booker
- Mike Brewer
- how best to measure living standards at the bottom of the distribution, given the large mis-match between reported income and reported expenditure of households who report to have a low income in household surveys
- in analysis of trends in inequality and poverty, including decompositions, and issues of stochastic dominance
- issues to do with the links between income, consumption and wealth, especially in a dynamic setting
- issues to do with the labour supply of women with children
- using microsimulation models of taxes and benefits to understand better the income distribution, non-take-up of benefits, and labour supply behaviour
- various topics in applied labour economics, especially evaluating the impact of labour market or welfare interventions, and including methods for casual inference
- inequalities amongst children, and the dynamics of family formation
- Malcolm Brynin
- Do changing family and gender values and norms affect the emotional disturbance children feel when parents divorce or separate? In other words, is the psychological impact in part socially determined? This analysis would use panel data in more than one country.
- Is it possible to reconcile the different approaches to social mobility of sociologists, who use occupational mobility, and economists, who use wage mobility? The concern is not whether one or other approach is right, or best, but what do the differences between these approaches tell us? The analysis could be either UK-specific or international.
- Is the importance of ethnicity as a basis for inequality declining? While some might argue that this is the case, and perhaps that immigration is a more useful focus for analysis, it is arguable that ethnicity is becoming more rather than less important because it is more complex. Understanding this complexity and its effects on welfare should be an important part of any analysis in this area. The analysis would use a variety of British datasets.
- What is the value of education? We can analyse the relationship between education and class outcomes, wages, job satisfaction, career development, and also overqualification. These all tell us different things. What is their relationship? Which is most salient? What are their theoretical (sociological) implications? The analysis would use the Labour Force Survey, Understanding Society, and UK graduate destinations surveys.
- Emilia Del Bono
- Paul Fisher
- Angus Holford
- Alita Nandi
- Birgitta Rabe
Health and epidemiology
Income, poverty and wealth
Labour market behaviour and labour economics
Microsimulation of taxes and benefits
Migration and ethnicity
Statistics and Econometrics
- Paul Clarke
I am looking for projects involving the development and application of structural mean models in the social sciences.
Longitudinal panel data comprise repeated measurements of variables which potentially allow us to estimate the causal effects of these variables on future realisations. Causal effects can be used to predict the impact of intervening and manipulating these time-varying variables and so crucial for scientific understanding (e.g. if we modify people’s employment statuses, what is the impact of this on their future mental health?). However, associations between these variables estimated from observational data sets confound causal effects with other influences and so adjustments must be made. Standard models for longitudinal data like multilevel (both random and fixed effects versions) and marginal models (or generalised estimating equations) can only estimate causal effects under strong assumptions that rarely hold in the social sciences.
Structural mean models (SMMs) are a powerful family of models for obtaining causal estimates from longitudinal data under less restrictive assumptions (Robins 1994, Communications Statist. A). SMMs allow us to state clearly the crucial assumptions under which confounding (observed and unobserved) will be adjusted for and our causal estimates valid. The application of SMMs to longitudinal data is limited even in biostatistics (Vansteelandt and Joffe 2015, Stats Science) but novel in the social sciences.
I am looking to supervise a PhD student to work on developing and applying these models for social science panel data. Potential applicants should be interested in or prepared to learn about one or more of the following issues: causal modelling; statistical inference using estimating equations; missing data; longitudinal models for nonlinear outcomes; theoretically assessing the behaviour of estimators using Monte Carlo simulation and/or asymptotic theory; and identifying substantive questions/areas in which these models can be usefully applied.