POLICY OPTIONS FOR ALCOHOL PRICE REGULATION: RESPONSE TO THE COMMENTARIES
We would like to thank the contributors for their excellent observations on our paper [1]. As Wagenaar [2] suspects in his piece, responses to our work have indeed been two-fold, with feedback from media, industry, policymakers and academics focusing either on methodological issues or on what the impact on policy practice should be. Our four commentators are no different in this respect. We would like to focus our reply on some key points related to data, methods and further work.
Room and Livingston [3] rightly point out that data availability and quality are maybe the greatest challenge, and that any model is only as good as the evidence that is used to build it. Like most population-level alcohol research, we were heavily reliant on data from surveys and cohort studies with less than perfect response rates. Nevertheless, the UK is quite fortunate in that there is individual-level purchasing data with information on price per volume, by beverage and outlet type, data that is not routinely available in many jurisdictions. Consumption surveys also show a (relatively) good match with alternative population-level data sources [4]. Whilst it is plausible that heavy drinking may be underestimated in population surveys, there is no robust evidence that allows us to determine the distribution of unrecorded alcohol consumption by age, gender and consumption groups. We decided not to inflate figures, since the risk functions used in the model are also based on self-reported consumption rather than true underlying consumption, and harm estimates are scaled to match routine available data on actual population-level observed harm. Nevertheless, increasing concerns about the representativeness of survey samples together with the field's reliance on evidence that only surveys currently provide presents us with a significant challenge. Work to find solutions on how we will deal with this issue in future, when survey response rates drop below what could be considered a ‘minimum acceptable’ representativeness threshold, is urgently needed.
We agree that individual-level longitudinal data would be ideal for our econometric modelling [3]. So far, modellers have had access either to individual-level cross-sectional or longitudinal population-level data, both with well-recognised drawbacks. Thus, we encourage public health and economics researchers to continue efforts to instigate collection of better datasets. The idea of looking at what happens to elasticities derived from cross-sectional data when prices are not stable [3] is something that we would wish to explore as additional years of purchasing data become available.
We agree that timing effects are important especially when we have modelled separate age/sex subgroups [5]. The model projects the reduction in harms over the ten years following policy implementation. For chronic conditions, we assume a time lag of ten years to full effect with a linear trend over the intervening years. We also consider general survival over the ten-year period for each individual and use age/sex adjusted quality of life estimates for survivors to account for the fact that people living to older ages tend to suffer chronic condition like dementia. However, Van Baal and Feenstra [5] are correct that the cost side of this is currently excluded, partly because there is a substantial debate as to the extent to which costs of added life years should be considered. Whilst we account for several time-related dimensions in the model, we concur fully that a more dynamic and in particular time trend oriented modelling of both the past and the future consumption and harms would be an interesting, important and substantial further development to our work.
Room and Livingston [3] also suggested that an alternative pathway, a price-to-harm rather than a price-to-consumption-to-harm model, may shed light on the way in which price changes may affect harm without necessarily affecting consumption. The problem here was that the evidence base linking price changes directly to harm changes is largely confined to estimates of how average price changes affect average population-level harms, which is not sufficient when considering policies that differentially affect parts of the alcohol market and where policy makers require information at the level of population subgroups.
The reported policy effects relate to the direct impact of a price rise, for example in beverages currently retailing at less than the minimum price threshold. As noted [2], we do not account for any indirect effects resulting from supply-side responses to the policy. It is not clear whether these would lead to greater or lesser impacts on consumption and harm.
Between them, the commentators make many helpful suggestions about opportunities for expanding and improving this kind of modelling work. These include the broad areas of international adaptation of the model as well as moving towards comparative modelling capabilities across diverse jurisdictions [6], modelling specific levels of government (local, regional as well as national) [2], modelling industry responses [2], an investigation of how price responsiveness changes in situations of price instability [3], considerations of the combined effects of policy options [2], considering timing issues related to long-term policy effects [5] and ultimately moving towards cost benefit analysis [5]. We feel that this spells out an ambitious and exciting programme of future research.
Declarations of interest
None.
