Disentangling the association between alcohol consumption and employment status: causation, selection or confounding?

Lluis Mangot-Sala*, Nynke Smidt, Aart C. Liefbroer

*Corresponding author for this work

Research output: Contribution to journalArticleAcademicpeer-review

2 Downloads (Pure)


Background Alcohol use constitutes a major health risk and is related to unemployment. However, the direction of this relationship is unclear: unemployment may change drinking patterns (causation), but heavy drinkers may also be more prone to lose their job (selection). We simultaneously examined selection and causation, and assessed the role of residual confounding. Moreover, we paid attention to the subgroup of abstainers and occupationally disabled, often disregarded in the literature. Methods Longitudinal data (three waves collected between 2006 and 2018) of the Lifelines Cohort study from the Netherlands were used (138 875 observations of 55 415 individuals, aged 18-60 at baseline). Alcohol use was categorized as 'abstaining', 'moderate drinking' and 'binge drinking' (>= 5 drinks/occasion for male; >= 4 for women). Employment status included occupational disability, short (= 6 months) unemployment. Random- and fixed-effects multinomial regression models were fitted in order to account for residual confounding. Reciprocal causality was assessed through generalized structural equation modelling with fixed-effects. Results Long unemployment spells increase the risk for both binge drinking (beta = 0.23; 95% CI 0.04-0.42) and abstinence (beta = 0.27; 95% CI 0.11-0.44), and the effects hold after accounting for reciprocal causality and time-constant confounding. Contrarily, the effect of binge drinking on unemployment is weak (beta = 0.14; 95% CI -0.03 to 0.31). Abstinence is strongly associated with occupational disability (beta = 0.40; 95% CI 0.24-0.57). Conclusions We find evidence supporting the causation hypothesis (unemployment altering drinking patterns), whereas evidence for the selection hypothesis is weak and mostly confounded by unobserved variables, such as poor health prior to baseline.

Original languageEnglish
Article numberckac141
Number of pages7
JournalEuropean Journal of Public Health
Publication statusE-pub ahead of print - 10-Oct-2022



Cite this