Wage penalty of abstinence and wage premium of drinking–A misclassification bias due to pooling of drinking groups?

Abstract

Several studies have found protective effects of low/moderate (hereafter ‘light’) alcohol consumption compared with ‘abstinence’ on mortality, health and wage. Some of these studies have been criticised because former drinkers have been included among the abstainers, which may overstate the protective effect of light alcohol consumption. It has also been proposed, but not shown, that the commonly pooled group of light drinkers and former heavy drinkers would understate the protective effect of light drinking. We also suggest that former abstainers might cause the same effect when pooled with light drinkers. The aim of this article is to study whether the pooling of consumption groups creates bias in the form of misclassification and confounding. The analysis focuses on: ‘former drinker error’ (pooling of lifelong abstainers and former drinkers); ‘former abstainer error’ (pooling of former abstainers and lifelong light drinkers) and ‘former heavy drinker error’ (pooling of light drinkers with and without a history of heavy drinking). Swedish panel data were used in a multinomial logit model, presenting odds ratios when comparing the subgroups. The results demonstrate that commonly pooled groups are heterogeneous with respect to a number of variables, which may implicate confounding. Given appropriate controls, misclassification bias is likely in the pooled group of light drinkers. The direction of the misclassification bias, however, is to underestimate the beneficial effect of light alcohol consumption on wage and therefore cannot explain the wage penalty of abstinence compared to light drinking.

• Alcohol consumption
• drinking history
• consumption groups
• misclassification bias

Acknowledgements

We would like to thank Martin Nordin, Krister Hjalte, Åsa Ljungvall, Graham Scotland and two anonymous referees for helpful suggestions and comment. Preliminary drafts have been presented at the 7th European Conference on Health Economics and at a Health Econometrics Working Group seminar in the Health Economic Research Unit (HERU), University of Aberdeen and the authors are grateful for all comments received. Parts of the study were conducted while the authors were situated at HERU, University of Aberdeen. Financial support from the Swedish Council for Working Life and Social Research (dnr 2006-1660) is gratefully acknowledged.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the article.

Notes

Notes

1. This article focuses on the alcohol–wage literature. The study question is, however, just as relevant in other areas, e.g. alcohol-health and alcohol-absenteeism literature.

2. Another issue often termed as misclassification bias is the fact that alcohol consumption is often considered being under-reported in surveys. This is especially a problem if consumption is correlated to the bias in reporting. However, this is a separate issue from what is studied in this article. Here we only deal with misclassification bias as defined in the text.

3. The model passes the traditional Hausman test of IIA.

4. A sensitivity analysis was performed where the Danish social tariff were employed in constructing the health variable instead of the UK social tariff. It gave generally larger associations although no effect on significance. A gender-specific model was also estimated, which showed that most of the significant associations found in the main model is connected to women and few connected to men. This indicates a stronger association between women's consumption and life situation compared to men.

5. The magnitude of the misclassification bias in this study could be quite small with respect to lifelong light drinkers due to the relatively small group of former abstainers compared to lifelong light drinkers. As mentioned are the relative proportions of the consumer groups dependent on the time period between waves and a longer period are expected to reduce the number of lifelong light drinkers.

6. A sensitivity analysis showed that there is a risk that ‘very low/occasional drinkers’ are mistakenly defined a former abstainers or former drinkers due to the characteristics of the dataset. The analysis showed that former drinkers and former abstainers are not distinguishable in the model when excluding the top 20% high consumers (current and lagged) in these groups.