Demonstration Modules for Growing Up in Scotland Dataset
Dr Pamela Warner, Centre for Research on Families and Relationships, University of Edinburgh
There is general agreement that policy should be evidence-based, but often the evidence needed is highly subtle/contingent. In most cases the many factors associated with any outcome of interest are themselves inter-related. This can create challenges in analysis, and also in communicating findings to policy-makers and practitioners.
For any longitudinal multi-variable data set there are many ways to approach analysis. The method chosen should reflect the research question being asked, which should in turn suit the data available. The multi-variable nature of the data set, and the three-way interdependence between question, data available and method of analysis, can be very difficult for researchers to navigate, particularly if they are new to multi-variable (MV) methods of analysis. Ideally there should be pragmatism as to the precise research question tackled, to ensure that some analysis can proceed. This is usually preferable to persisting lack of evidence, or requiring investment and time to collect new data.
In longitudinal studies, as each new sweep of data collection is completed, the richness of the data accumulated increases, enhancing the potential to answer questions, in particular those addressing the relationship of ‘outcomes’ to time-varying conditions or events. However, in parallel with the increase in duration of follow-up, there is inevitably an escalation in the complexity of data to be analysed, creating a tension between the scope/depth of the data available to inform policy and practice, and the analytical capacity of researchers to capitalise on it.
This project undertook MV analyses of a longitudinal data set (encompassing pre-existing, time-varying and outcome variables), to create demonstration materials for a set of MV approaches, all focussed on the same set of variables. This provided insight into the distinct combinations of research question and analytical approach that were possible, even on what was essentially the same data. In addition the project collated references/URLs for educational resources for each method, and provided annotated (but slightly simplified) accounts of the process of undertaking the analysis, including software outputs, iterations and interpretations.
The longitudinal data set utilised was the Growing Up in Scotland study (GUS, http://www.growingupinscotland.org.uk), commissioned by the Scottish Executive Education Department (SEED) in 2003, in order to provide information to help develop policies affecting children and their families in Scotland. GUS is led by the Scottish Centre for Social Research (ScotCen) in collaboration with the Centre for Research on Families and Relationships (CRFR) at the University of Edinburgh, and the MRC Social and Public Health Sciences Unit in Glasgow. The study followed around 13,000 children in three age-groups.
The Scottish Government made GUS data freely accessible to anyone wishing to carry out research on children and their families in Scotland. The project aimed to focus on one subset of GUS data – i.e. child outcomes, experience of family poverty across time and family characteristics (pre-existing and time-varying).
Preliminary drafts of materials were developed based on initial analyses undertaken by an assistant analyst. These were then discussed and revised by the project members, and packaged into an online format. A panel of ‘readers’– covering a range of disciplines, and/or analytical experience – were recruited to comment on the materials prior to finalisation.