International Choice Modelling Conference, International Choice Modelling Conference 2015

Font Size: 
Modelling the impact of multi-buy promotions on alcohol purchasing behaviour: a comparison of estimation and implementation of sample selection models and the multiple discrete-continuous extreme value (MCDEV) model
Daly Andrew, Stephane Hess, Hui Lu, Charlene Rohr

Last modified: 18 May 2015


Excessive alcohol consumption is associated with economic and social harm, as well as a major cause of ill-health and mortality, and there is wide recognition that excessive drinking needs to be tackled. The present study evaluates the impact of banning multi-buy promotions, like 2 for the price of 1 promotions, on consumers’ alcohol purchasing, an area of research which is much less developed than those on the alcohol pricing (Anderson et al., 2009; Purshouse et al., 2010). To do this, we develop a series of econometric models, ranging from sample selection models to multiple discrete continuous extreme value (MDCEV) models, to capture the discrete and continuous feature of alcohol purchasing behaviour. We then examine the impact of banning multi-buy promotions on alcohol purchasing. This paper presents the comparison of the model results and contributes to the recent developments that facilitate estimation and implementation of these consumer demand models.

Our study deploys innovative Stated Preference (SP) choice experiments to measure consumers’ potential responses towards price changes and promotions. An SP approach was deemed necessary because of the lack of a database containing information on consumer’s purchasing, information on available multi-buy promotions, and information on consumers, specifically their drinking habits which was a key segmentation variable for the study. The online SP choice experiments presented consumers with a number of hypothetical scenarios, with different alcohol prices and with/without the multi-buy promotions for six types of alcohol products, presented in the form of a supermarket shelf display. The alcohol types considered are three types of wine (described by their price range, i.e. cheap, intermediate and expensive), two types of beer or cider (again, described by the price level, i.e. cheap and less cheap) and spirits; other alcohol types and distinctions were found less relevant to the UK market in preliminary analyses. Respondents were asked to indicate which type(s) of alcohol they would purchase and the volume they would purchase during a four week period. A total of 1,265 respondents across the UK participated in the survey, with approximately 300 respondents targeted for each of four alcohol consumption levels: Moderate A (very moderate), Moderate B (intermediate), Hazardous and Harmful . 

The potential to purchase multiple types of alcohol is considered in the modelling, incorporating both decisions to purchase (or not) –a discrete choice- and the volume to be purchased – a continuous amount.  A key characteristic of the data is the presence of a substantial proportion of observations with zero expenditure. The selection of appropriate econometric techniques is crucial to incorporate all of these characteristics. Tobit (Tobin, 1958) and Heckman (Heckman, 1979) models are developed initially to quantify the impact of price and promotions on purchasing behaviour as well as considering individual’s socio-demographic characteristics.

However both models are restricted to binary choices only, i.e. the decision to purchase a specific alcohol type and the volume to be purchased. The inherent deficiencies of the models prevent them from explicitly predicting the likelihood of substitution among different alcohol types in the presence of multiple promotions on different products. To address this issue, an advanced multiple discrete-continuous extreme value (MDCEV) model(Bhat 2005, 2008)was developed which overcomes the binary restriction of the Tobit and Heckman models. The model results from the different models are compared and are implemented in forecasting models to enable the evaluation of promotions on consumers’ purchasing behaviour.

The comparison reveals that different model assumptions lead to different estimates of the impact of promotions. Despite the differences in the magnitude of coefficients estimated, the socio-demographic and economic variables identified from the MDCEV model are similar to those identified in the Heckman and Tobit model.  However, we observe higher own-price elasticities in the MDCEV models, particularly for the cheaper alcohol products, indicating the importance of competition and switching between products. In addition, we observe a number of interesting findings more generally.  All model results tend to show higher price elasticities and relative impacts for moderate compared to hazardous and harmful drinkers (and this is consistent with the findings of others, e.g. Fogerty (2004)).  Furthermore, we find that promotions have a substantial impact on the alcohol products consumers purchase; however, the overall impact on the total alcohol purchase is rather smaller (again, because of switching between products).

This research contributes to assessment of the impact of multi-buy promotions on alcohol purchases. It illustrates the benefits of incorporating recent modelling developments on practical policy studies incorporating consumer demand. In addition, the implementation of the models enables a series of policy scenario tests, which help to gauge the potential impact of policies on alcohol purchasing, specifically focussed on banning alcohol multi-buy promotions.


Anderson, P, Chisholm, D., and Fuhr, D. (2009), Effectiveness and cost-effectiveness of policies and programmes to reduce the harm caused by alcohol, Lancet 373(9682): 2234–46

Bhat, C.R., (2005). A multiple discrete-continuous extreme value model: formulation and application to discretionary time-use decisions. Transportation Research Part B, 39(8), 679-707.

Bhat, C.R., (2008). The Multiple Discrete-Continuous Extreme Value (MDCEV) Model: Role of Utility Function Parameters, Identification Considerations, and Model Extensions. Transportation Research Part B, 42(3), 274-303.

Fogerty, J. (2004), The own-price elasticity of alcohol: a meta-analysis, Working Paper, University of Western Australia.

Purshouse, R., Meier, P., Brennan, A., Taylor, K. and Rafia, R. (2010), Estimated effect of alcohol pricing policies on health and health economic outcomes in England: an epidemiological model, Lancet 375(9723): 1355–64.

Heckman, J. (1979), Sample selection bias as a specification error, Econometrica 47, 153-161.


Tobin, J. (1958), Estimation of relationships for limited dependent variables, Econometrica 26: 24–36.

Conference registration is required in order to view papers.