Node Publications

BIAS homepage

BIAS II finished its work in 2011.


Modelling biases in survey non-response

Non-response in surveys can arise for many different reasons, and biased inference can result if factors associated with non-response are also associated with the question under study. Researchers at BIAS II developed different approaches to modelling non-response bias. In one approach, they used graphical models to develop a type of influence diagram representing both the process under study and the potential causes of non-response. The conditional independence assumptions encoded by such diagrams allowed them to separate the process of non-response from the mechanism of inferential interest. BIAS II researchers could then develop biased-corrected estimates (e.g. using post-stratification strategies) of the parameters of interest by conditioning on appropriate variables separating the parts of the graph. their second approach borrows ideas from the propensity score literature to model non-response bias using latent variables. Instead of directly introducing variables responsible for non-response in a graphical model, BIAS II researchers modeled the effect of the bias globally through a latent "bias parameter" which is used to adjust the parameter estimates of interest. As part of this work, they investigated, within the Bayesian framework, how to incorporate external information, e.g. via a validation data set, in the specification of the prior distribution of this bias parameter.

Duration: July 2008 - March 2011

People working on the project: Dr Sara Geneletti, Prof Nicky Best, Prof Sylvia Richardson Collaborators: Dr Lawrence McCandless (Simon Fraser Uni), Prof Paul Gustafson (UBC)


Spatio-temporal Modelling of Small Area Data to estimate social changes in space and time

Researchers at BIAS II developed Bayesian space-time models to characterise stability and estimate change over time in small area indicators. Their methods were designed to distinguish random temporal fluctuations in stable areas from areas in which real change has occurred. Their approach was to include different types of space-time interaction terms in the basic small-area model with spatial and temporal main effects, with a particular emphasis on choosing an appropriate hierarchical structure for these terms that facilitates classification of areas/time periods as 'predictable' (i.e. smoothly changing) or 'abruptly changing'. Together with their collaborators at the Office for National Statistics (ONS), BIAS II researchers used these space-time models to identify time trends and areas with low levels of income and employment, poor health and housing conditions, using small area data from e.g. General Household Survey, the Integrated Household Survey and the Family Resources Survey. BIAS II researchers also collaborated with Prof Bob Haining at Cambridge University to apply their methods to analyse the space-time pattern of criminal offences in Cambridgeshire. They investigated how stable patterns are and whether there was evidence of repeat victimisation and "spatial" repeat victimisation (where an offence such as burglary does not occur in the same household but within some radius). In addition, BIAS II researchers attempted to detect sudden increases in the crime activities by modelling space-time interactions.

Duration: July 2008 - June 2011

People working on the project: Prof Nicky Best, Prof Sylvia Richardson Collaborators: Mr Philip Clarke (ONS), Prof Bob Haining (Cambridge)


Generalised Evidence Synthesis for Longitudinal Data

Researchers at BIAS II collaborated with Prof Scott Hofer at Oregon State University USA, who directs a large international collaborative research network on longitudinal studies of ageing - IALSA - to develop and apply a Bayesian hierarchical modelling approach for the synthesis (meta-analysis) of multiple cross-national longitudinal datasets. Their basic synthesis model assumed a common set of variables measured in all studies and has a hierarchical structure with random effects capturing differences between, e.g. countries, studies with different population bases etc. Their model could then be elaborated to include constructs and covariates that are measured differently across the studies, by relating these to underlying latent variables, and to include several outcomes simultaneously. Substantive questions that BIAS II researchers were addressing using these synthesis models included: (a) to document general patterns of population average (between person age differences) and individual variation (within person changes in age) in change in cognitive capabilities; (b) evaluate the effects of age, education, and sex on intra-individual trajectories of cognitive outcomes across studies; (c) identify/describe the level and rate of change in Mini Mental State Exam at the individual and aggregate population level prior to diagnosis of dementia.

Duration: July 2008 - June 2010

People working on the project: Dr Jassy Molitor, Prof Nicky Best, Prof Sylvia Richardson Collaborators: Prof Scott Hofer (Oregon State University)


Combining individual and aggregate data to analyse electoral behaviour

During phase 1 (BIAS I, 2005-08) BIAS II researchers developed hierarchical related regression (HRR) methods for combining random samples of individual level data with aggregate data on the same variables in order to reduce ecological bias and increase power compared to analyses based on a single data source. BIAS II researchers collaborated with Dr Steve Fisher at Oxford University to apply these HRR methods to analyse electoral behaviour data. One question they were addressing concerns ethnicity and vote choice. The vote choice of ethnic minorities in Britain is hard to estimate with opinion polls, or even with the British Election Study (BES) surveys, because the sample sizes are too small to yield sufficient numbers of ethnic minorities. Researchers at BIAS II used HRR models to combine the individual level BES data with aggregate census data, which includes ethnicity and religion, and election results for parliamentary constituencies, to improve estimates of the strength of association between ethnicity and vote for the 2001 and 2005 general elections and thereby assess the extent to which there was a realignment of British ethnic minorities away from Labour.

Duration: July 2008 - June 2010

People working on the project: Dr Jane Key, Prof Nicky Best, Prof Sylvia Richardson Collaborators: Dr Steve Fisher (Oxford University)