In this study, alcohol purchase patterns and the clustering of alcohol with unhealthy foods and tobacco were described in a large data set of Finnish loyalty card holders. The typical expenditure on alcoholic beverages bought from grocery stores at one time was small. In the large majority of cases, no alcoholic beverages were bought, and less than 1 of 20 customers bought more than two beers or ciders. Beer was the most purchased type of alcoholic beverage, and cider was not bought in larger numbers without buying beer as well. Expenditures on non-alcoholic options were negligible; they were not characteristic to any of the clusters. The expenditure on alcohol were associated with expenditures on tobacco and foods rich in saturated fat, salt and added sugar.
Alcohol consumption and smoking tobacco are known to cluster together [2, 16]. However, their co-occurrence on the level of individual shopping occasions has not been studied before. The shopping basket clusters with alcoholic beverages contained more cigarettes than baskets without alcohol. However, cigarette expenditures did not peak in the baskets with the largest expenditure on alcoholic beverages but were most pronounced in the mid-sized alcohol baskets. It may be that the largest alcohol baskets were bought to be shared with many people at a social event, and cigarettes are not shared alongside alcohol. In addition, it is possible that people in whom unhealthy behaviours may cluster buy alcohol in moderate numbers but at a higher frequency. Socially disadvantaged people may buy their alcohol in smaller quantities at a time. To identify groups with potential problems with alcohol consumption, we will analyse the frequency and temporal distribution of alcohol expenditures in our next study.
The finding that the expenditure on alcoholic beverages was positively associated with the expenditure on food does not suggest that those who buy large numbers of alcoholic beverages would have to trim their food budget to compensate for the alcohol expenditure. However, only mild alcoholic beverages are allowed for sale in supermarkets and grocery stores in Finland, so wines and spirits were not included in our data. Mild alcoholic beverages cover about half of the total alcohol consumption in Finland [17]. Correspondingly, a similar association of less alcohol and smaller food budget was observed in French supermarkets using purchase data over 1 year [11], and the authors speculated that this could be explained by a lower socio-economic status. Over a two-week period, the expenditures for alcohol and food purchases in British households were inversely associated [12]. Our findings could indicate that the larger alcohol baskets were bought for social occasions and were intended to be shared with others. In fact, many of the food groups bought alongside the larger alcohol baskets – for instance, sausages, meats, soft drinks, snacks, cookies, sweets and pastries – are typical party or barbecue foods. The Danish study found that those who bought wine and beer also bought more food items [10]. Purchasing two different types of alcohol could indicate a social drinking occasion with the need for more food.
The food most consistently associated with alcohol expenditure in this study were sausages, soft drinks and snacks, as both the absolute and proportional expenditures on them grew consistently with the expenditure on alcoholic beverages. All three food categories may be seen as unhealthy options. Sausages and snacks tend to contain plenty of energy and fat, whereas soft drinks are major sources of refined sugar [18]. The aim of the Danish supermarket study was to compare food purchases of wine versus beer buyers [10]. The results showed that wine buyers favoured healthier foods, such as fruit and vegetables, poultry, oils and low-fat cheese, than beer buyers, who bought more sugar, chips, pork, lamb, sausages and soft drinks. In the French study in which purchases were aggregated over 1 year, beer buyers made unhealthier food choices than those who bought wine or did not buy alcoholic beverages [11]. Our results add to the earlier research that showed that alcohol consumption tends to cluster with unhealthy dietary choices. In many of the previous studies, a simpler indicator variable, such as fruit and vegetable consumption, or a pre-defined index of healthiness has often been used [2, 13, 19].
The same foods were associated with both beer- and cider-dominated shopping baskets, but expenditure on cider tended to correlate with sweet foods more strongly than expenditure on beer did. Perhaps people who choose cider prefer sweet tastes, or possibly this has something to do with the unequal gender or age distribution of cider versus beer buyers. Diets of people with different alcohol preferences have been compared before, mainly to account for the observed health benefits associated with wine [19, 20], but to our knowledge, comparisons between food choices of beer and cider drinkers have not been published before. In a French study, cider was aggregated with beer [21]. In light of these novel results, cider does not seem to associate with healthier food choices than beer like wine does.
This study, using automatically registered purchase data, has its unique strengths and weaknesses compared with traditional survey methods in alcohol and food research. A major weakness is that we did not know whether the foods and beverages were consumed by the same individual who bought them or if they were for family members or friends. Correspondingly, we did not know what was purchased elsewhere or eaten outside the home [10, 11, 13]. Wine and spirits are available only in specialist shops in Finland; therefore, a significant share of alcoholic beverages is outside the scope of our analyses. In general, however, food purchase data were seen to reflect individual diets reasonably well. The first analyses from this study indicate a very high proportion of expenditure among loyalty card owners in S Group stores [15]. The estimated mean of total annual expenditure in our data was 2322€ per customer, while the average consumption of groceries and non-alcoholic drinks in Finnish households was 2916€ per consumption unit during the same time period (22). Another limitation was that purchase data were on product group level and were imprecise for certain purposes. For example, we could not differentiate between common bulk beer and more exclusive craft beer varieties. The unit of measurement of expenditure was euros, and the actual weight or volume of the purchases was unknown.
A major limitation in this study, shared with survey samples, is the potential selection bias according to sociodemographic factors and alcohol consumption patterns. We know that women were over-represented in the present study population, compared with the gender distribution among residents of the study area, whereas both young and old people were under-represented [15]. The present study design offers no means to assess whether individuals who consume more alcohol were less likely to participate. However, it should be noted that the study was not presented to the loyalty card holders specifically as an alcohol consumption study, but as a study on purchase patterns at large. Therefore, it may be unlikely that alcohol consumption patterns as such would be the main driver of the decision of not to participate. Also, the consent for data use does not require personal contact with e.g. an interviewed, which may lower the “shame bound” for some individuals to release the details.
The major strengths of this study were large sample size, objective measurement of consumption without over- or underreporting [5, 10] and unbiased by recall or reporting error, and a potentially more heterogeneous study population than in traditional surveys [15]. The assessment of the association of food and alcohol on single shopping occasions would not be possible in survey data either. Due to social undesirability, self-reported alcohol consumption is typically underestimated. In this study, the permission for data use was requested from the participants in retrospect; thus, they were unaware of their status as study subjects at the time of data collection. As the study period covered a whole calendar year, the results were not affected by potential seasonal variation [10].