Use of retail data in the assessment of natural experiments: the case of Reducing the Strength,an intervention to reduce alcohol availability

BackgroundRetailers routinely collect data about people's purchasing behaviours and access to consumer products associated with health and wellbeing. Here we discuss how retail data can be used in public health research and consider potential strengths and limitations to such research. To illustrate the discussion we refer to an evaluation of an intervention called Reducing the Strength, whereby off-licence shops and supermarkets voluntarily stopped selling inexpensive superstrength (≥6·5% alcohol by volume) beers and ciders.MethodsMonthly data from a large retail chain (East of England Co-operative Society) were obtained for three UK counties (141 stores). In one county the intervention started 12 months earlier than the others, allowing for a pre–post study design with a delayed implementation comparator. Difference-in-differences analysis of unit alcohol sales controlled for socioenvironmental confounders and shop-level characteristics including shop size, parking facilities, cash machines, opening hours, and other factors.FindingsThe retail data detailed shop-level characteristics and sales data such as prices, quantities, product brands, alcohol content, sales, and factors affecting sales. The wide geographical coverage, shop-level data, including data for potential confounding factors, and frequent timepoints made the retail data well-suited for a quasi-experimental evaluation capitalising on temporal and spatial variations in intervention exposure. Limitations of this study include a lack of longitudinal data for individual customers, and shops that are not covered by the data. Qualitative interviews with shop workers and customers, and triangulation using alternative data sources can help to address limitations. Alternative sources of retail data such as private sector consultants who specialise in collecting shop-level and sales data for a range of companies might also address some limitations; however, there are potential barriers of expense, accessibility, and coverage associated with the use of such consultants.InterpretationIncreasingly, researchers recognise the potential of retail data for evaluating interventions affecting social determinants of health and inequalities, such as local access to alcohol. However, shop-level data have frequently proved difficult for researchers to obtain. By obtaining such data we have been able to assess, using a quasi-experimental design, the effects of removing strong, cheap beers and ciders from shops. We have also been able to explore in more detail how to optimise the strengths and address some limitations of the data in ways that could potentially assist others planning to use this important data source in their research.FundingThe study is funded as part of the School of Public Health Research by NHS National Institute of Health Research. AJ, SA, and JW contributed as employees of Public Health Suffolk, Suffolk County Council.