Commentary on Rossow<i>et al</i>. (2014) and Norström &amp; Svensson (2014): We want to believe-or do we have to?

The work of Bruun and colleagues 1 was a breakthrough for the reduction of total consumption of alcohol as a major prevention strategy. With the total consumption approach, however, theories were needed to explain that restricting the availability of alcohol is not just an approach that punishes everybody for the sins of a few, the heavy drinkers. Two main theories were supportive in this respect. The preventive paradox 2 states that the reduction of exposure would be more effective when conducted in the entire population rather than in high-risk groups only, because low- and medium-risk drinkers create even more harm for the society than the high-risk individuals, who comprise a smaller group. Kreitman's preventive paradox has been questioned for ignoring heavy drinking occasions 3. More recent research in Europe has found that heavy drinking (60+ g per day for men; 40+ g per day for women) accounted for almost 90% of disability adjusted life-years lost and almost 80% of alcohol-attributable deaths 4. The second main theory is the ‘collectivity of drinking cultures’ 5, which suggests that a change in the mean consumption will affect drinkers at all levels and thus also the heavy drinkers. The two papers by Rossow and colleagues [6] and Norström & Svensson 6, 7 revitalize the theory by overcoming two main critiques 8; namely, that most studies (i) compared the differences between societies and not changes within a society over time and (ii) investigated the link between mean consumption and heavy drinking mainly for periods of increasing but not decreasing consumption (e.g. 9). Skog's theory is relevant to the link between mean consumption and heavy drinking as part of the originally named Ledermann-string (e.g. 10), namely the causal path from (restriction of) availability to (reduction of) the prevalence of harm through (reduction of) mean consumption and (reduction of) the proportion of heavy drinkers. The causal path from mean consumption to heavy drinking is generated through social interactions between drinkers who influence each other in their social networks directly or indirectly via the society as a larger network 5: ‘… each actor is influenced, directly or indirectly, by practically every other member of his culture’ (p. 88). ‘Hence, we are led to expect that each individual drinker in a culture will adjust his drinking according to the mean consumption level in his culture’ (ibid., p. 89). This interaction leads to a synchronization so that the members of a society move more or less in concert up and down the consumption scale, and consequently, if there is a change in the mean consumption, the upper end (the heavier drinkers) will change in the same direction. Although Skog perceived his theory as less deterministic and causal, it is important to consider the common understanding of the theory. Both present papers 6, 7 agree with the two general principles of causation and synchronization. Rossow et al. 6 attempted to strengthen the causal aspect by providing an analysis to eliminate the tautological part that heavy drinkers determine the mean consumption, and they stated that universal policies were less justified if heavy drinkers drive the connection between heavy drinkers and mean consumption. Norström & Svensson 7 tested the polarization hypothesis (different direction of change among the heavy drinking group compared with other drinking groups), which is stated as being in contrast to Skog's synchronization. One common test of Skog's theory is to look at collective changes within larger subgroups of a society. In a society with changing consumption one would expect no large groups or subpopulations to deviate from the general trend (9, p. 48). There is, at best, inconclusive evidence (examples given by 6, 8, 11) that population subgroups change collectively. Men and women are large subgroups, but there is no agreement as to whether both genders should be tested separately. Norström & Svensson 7 tested for gender differences but analysed both genders together, because there were no differences at an age of 15–16 years. Rossow and colleagues 6 showed analyses sometimes pooled for men and women, sometimes separately for both genders. Skog's original analysis was conducted on combined men and women populations without giving separate estimates for them. At least, it seems unclear whether the theory should apply to men and women in the same way. Given the large differences between males and females in drinking behaviour in many parts of the world, and particularly in countries with low gender equality, one may want to know whether the theory applies to a population as a whole or to mainly men, who are more likely to fall at the high end of the consumption scale. Under many circumstances, the result will be a displacement of the whole distribution. However, the theory also predicts when we may expect exceptions from the overall pattern. If there are barriers for the diffusion process, for instance between different social-economic strata, exceptional patterns may result. Secondly, informal social control (i.e. the extent to which people influences each other) may vary across cultures or between substrata. For instance, there may be gender differences in this respect, and it has been demonstrated that there are systematic differences in the dispersion of the distribution among males and females (p. 327). Therefore, Skog's theory would be correct even if differences in the collective displacement occurred, particularly between men and women. However, subgroup analysis would barely be a good test of the theory in this case. Similarly, it is also unclear whether or not abstainers should be included in the analysis of collectivity. While Rossow and colleagues 6 understood that Skog analysed only drinkers because he assumed that social interaction, which influences individual's drinking, occurs among drinkers only, Norström & Svensson 7 argued that Skog published the data based on surveys that included drinkers only, and thus the exclusion of abstainers was just not a deliberate choice. They analysed total population data (using a sensitivity analysis excluding abstainers). Edwards and colleagues (13, p. 90) stated that abstainers were not included by convention, but also, ideally, that it should be determined whether changes in mean consumption emerged because of changes in abstention rates or change in the consumption level of drinker. The exclusion of abstainers may be irrelevant in societies in which the majority consumes alcohol. However, far more than half of the world is abstaining 14, and thus the relevance of a theory that excludes abstainers is questionable. If all individuals within a society influence each other, why should then drinkers not influence abstainers and vice versa? Per capita consumption is related more strongly to abstention rates than to the mean consumption among drinkers, and very clearly the distribution of alcohol use changes dramatically when abstainers are included or excluded 14. One should not forget that Skog's theory is not only a philosophical work; instead, it has important implications for prevention. Thus, it should make a difference whether the reduction in the mean consumption is achieved by increasing abstention or decreasing consumption among drinkers. In addition, Skog did not even say that heavy drinkers must necessarily change in the direction of the mean change. ‘In fact, some of these individuals will no longer be among the top 10% of the distribution—they have been replaced by individuals who belonged initially to a lower consumption stratum. It is the location of the 10th percentile that should be expected to change by a similar percentage, not these particular individuals’ (12, p. 329). Although whether heavy drinkers change in the same direction as the mean may not matter for the theory, it matters for the preventive implications of the theory. It can be shown relatively easily that a change of the top 10% of drinkers to almost no consumption, and therefore becoming the 10% of drinkers with the lowest consumption (e.g. due to particularly effective measures among the heavy drinking population), can result in exactly the same consumption distribution as a change in drinking of a relatively equal amount among all consumers, i.e. the collective change. The difference to the collective change would be that 90% of consumers have not changed their consumption; they simply moved on to higher percentiles of the distribution. The distributional statistics, however, would be exactly the same as with the collective change. As a result, the famous figures used to demonstrate a collective change (figure 3 in Skog 5; figure 3 in Rossow et al. 6 and figure 4 in Norström & Svensson 7) would all look identical, regardless of whether a replacement of the top 10th percentile has taken place or whether a collective change occurred. The example may be unrealistic, but more realistic examples have been given by Gmel & Rehm 15 in their response to Skog's critique 12 of Gmel & Rehm 8. The graphical test is not a test of collective displacement. Skog's theory has been enormously helpful in justifying general population approaches to prevention. The theory may be correct, but many open questions, such as who to include in the study and how to test the hypotheses empirically, remain. The main question is whether we actually still need that theory. The Ledermann-string goes from availability to harm through mean consumption and heavy drinking. Do we actually need the link between mean consumption and heavy drinking? Would it not be sufficient to show that reduction of availability reduces harm and thus justifies general population approaches? The link between availability and harm has emerged repeatedly. As Skog has stated repeatedly 12, whether a certain change in per capita consumption causes a parallel change in the prevalence of heavy drinking or vice versa is hardly a meaningful problem. None.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.