The association between unemployment trajectories and alcohol consumption patterns. Evidence from a large prospective cohort in The Netherlands

(2021) - Lluís Mangot-Sala, Nynke Smidt, Aart C. Liefbroer

Journal: Advances in Life Course Research
Link:: https://linkinghub.elsevier.com/retrieve/pii/S1040260821000332
DOI:: 10.1016/j.alcr.2021.100434
Links::
Tags:: #paper #NCDS #Health #Unemployment #LabourMarket #Transition
Cite Key:: [@mangot-salaAssociationUnemploymentTrajectories2021]

Abstract

Unemployment is expected to influence alcohol consumption, but studies show mixed results, partly because most studies concentrate on current employment status. However, unemployment could be particularly consequential if it is part of a trajectory of employment pre­ cariousness. Moreover, the association between unemployment and alcohol consumption may not be homoge­ neous across the population, but differ by subgroups (e.g. socioeconomic status). This study longitudinally analyses the association between different employment trajectories and alcohol consumption, and examines if the association is moderated by socioeconomic status (SES), partner status, age and gender. Four waves of data of the Lifelines Cohort study are used.

Notes

“Unemployment is expected to influence alcohol consumption, but studies show mixed results, partly because most studies concentrate on current employment status” (Mangot-Sala et al., 2021, p. 1)

“Evidence suggests that length of unemployment is key in order to grasp its effects in terms of changing drinking patterns” (Mangot-Sala et al., 2021, p. 1)

“Alcohol consumption is one of the five leading risk factors for global disability-adjusted life years (IHME, 2018), and responsible for 6% of deaths worldwide (Collins, 2016).” (Mangot-Sala et al., 2021, p. 1)

“alcohol has been rated as one of the four most harmful drugs on a population level (Nutt, King, & Phillips, 2010; Van Amsterdam, Opperhuizen, Koeter, & van den Brink, 2010), due to a wide spectrum of negative health effects, ranging from neuropsychiatric disorders, cardiovascular and gastrointestinal diseases to acute conditions, e” (Mangot-Sala et al., 2021, p. 1)

“common drawback of the literature is that it often focuses on simple dichotomous comparisons between those who are currently unemployed and those who have a job. However, the life-course perspective suggests that the strength of the association between life events on the one hand and health behaviours and health outcomes on the other hand, may depend on how they are embedded in the life course (Bernardi, Huinink, & Settersten, 2019).” (Mangot-Sala et al., 2021, p. 2)

“Another limitation of the literature is that it often assumes that the consequences of unemployment for alcohol consumption are homogeneous across the population.” (Mangot-Sala et al., 2021, p. 2)

“Social stress theory suggests that socioeconomically disadvantaged individuals are more likely to be exposed to stressors (Mossakowski, 2008). Furthermore, they also tend to be more vulnerable to stress because they have limited coping resources (Mossakowski, 2014) in terms of material, social, and cultural capital (Marmot & Wilkinson et al., 2001). Unemployment is considered to be a major source of stress (Lynch, Smith, Kaplan, & House, 2000; Pampel, Krueger, & Denney, 2010), and at the same time a drain on resources and the capacity to cope with it (Kalleberg, 2018), as it often entails an income drop (Boden et al., 2017; Lantis & Teahan, 2018), but also a loss of sense of control, and a reduction in the size of the social network (Marmot, 2004). Furthermore, the “shock” (Jackson & Warr, 1984) caused by job-loss also leads to a rupture of people’s daily life structures (Virtanen et al., 2016), which could also increase stress levels.” (Mangot-Sala et al., 2021, p. 2) Potentially important for unemployment

“Thus, long-term exposure to unemployment can lead to a situation of “chronic stress”, in which individuals perceive little sense of control over their own life (Mossakowski, 2014). Chronic stress can lead to the release of hormones, such as cortisol, adrenaline and noradrenaline (Marmot, 2004), which in turn account for biological changes affecting health -e.g. metabolic syndrome (Brunner, 2007).” (Mangot-Sala et al., 2021, p. 3)

“the consequences of unemployment for alcohol consumption may depend on the phase in the life course in which unemployment occurs.” (Mangot-Sala et al., 2021, p. 3)

“The negative impact of unemployment may be stronger for some groups of individuals than for others, depending on different factors, such as their available resources to cope -ranging from material resources to social or emotional support- or their reemployment probabilities (Gebel & Vossemer, 2014).” (Mangot-Sala et al., 2021, p. 3)

“Second, men and women may react differently to unemployment, partly because wage labour has traditionally played a different role in defining their “identities”: from that perspective, unemployment may be more stigmatizing for men due to the traditionally assumed “breadwinner” role, whereas women may have alternative roles other than employment (Paul & Moser, 2009). However, women could also be more vulnerable to unemployment due to higher financial insecurity (Norstr ̈ om et al., 2017). Indeed, gender differences in the effects of unemployment have been reported (Backhans et al., 2012; Janlert et al., 2015), although findings are mixed and highly context-dependent (Norstr ̈ om et al., 2014).” (Mangot-Sala et al., 2021, p. 3)

“First, a simple dummy variable (employed/unemployed) was constructed” (Mangot-Sala et al., 2021, p. 4)

“Second, we took the duration of unemployment into account by constructing a trichotomy: “employed”, “short-term unemployed” (<6 months), and “long-term unemployed” (≥6 months) (cut-off based on previous studies (Backhans et al., 2012; Strandh, Winefield, Nilsson, & Hammarstrom, 2014)).” (Mangot-Sala et al., 2021, p. 4) Important

“Third, we constructed an indicator for the “total duration of unemployment”, calculated by summing the number of months a participant reported having been unemployed at each wave, thus obtaining the total duration of unemployment throughout the observation period, which was categorized into “<6 months”, “6 18 months”, “>18 months”” (Mangot-Sala et al., 2021, p. 4)

“we calculated the “number of unemployment spells”, as the number of times a respondent reported having been unemployed across all four waves, categorized as “no unemployment”, “1 unemployment spell”, “2 or more unemployment spells”” (Mangot-Sala et al., 2021, p. 4)

“based on employment status at each wave, a variable focusing on “employment trajectories” was constructed, with five categories: “continuously employed” (employed at all 4 waves), “continuously unemployed” (unemployed at all waves), “upward trajectory” (those unemployed at baseline that became employed later on, thus following the structure: 1110, 1100 and 1000, where 0 is employed and 1 is unemployed), “downward trajectory” (those employed at baseline that became unemployed later on [0001, 0011, 0111], and “intermittent spells” (the rest of individuals with trajectories such as 1010 or 0101, 0100, 1011, etc.).” (Mangot-Sala et al., 2021, p. 4)

“The main aim of MICE imputation is to provide values that will minimize bias in the estimated models (White, Royston, & Wood, 2011).” (Mangot-Sala et al., 2021, p. 5)

“Gender and age at baseline were the only variables with no missing values, and were used as auxiliary variables” (Mangot-Sala et al., 2021, p. 5)

“imputing such a large number of missing values certainly has some downsides, we followed the guidelines by White et al. (2011) for MICE” (Mangot-Sala et al., 2021, p. 5)

“we set the imputation at 50 iterations (resulting in 50 different datasets for each imputed variable), according to the rule of thumb that the number of imputed datasets should be higher than the percentage of missing values.” (Mangot-Sala et al., 2021, p. 5) Is this the case?

“The syntax used on the MICE process is available in the online Appendix.” (Mangot-Sala et al., 2021, p. 5)

“models based on the Complete-Case Analysis (CCA) design (i.e., including only these individuals who responded to all questionnaires) were run as sensitivity analyses” (Mangot-Sala et al., 2021, p. 5)

“Multinomial logistic models were estimated, as the categorical variable for alcohol consumption has three outcome categories” (Mangot-Sala et al., 2021, p. 5)

“These models were estimated in two steps. First, a series of models for each indicator of unemployment separately -controlling for all covariates- were fitted, to ascertain how strongly each unemployment indicator was related to alcohol consumption” (Mangot-Sala et al., 2021, p. 5)

“in a second step, moderation effects were tested by means of two-way interactions with all potential moderators.” (Mangot-Sala et al., 2021, p. 5)

“First, we fitted models that exclude variables that could be viewed as potential mediators in the association, i.e. partner status and health at baseline” (Mangot-Sala et al., 2021, p. 8)

“Second, models based on a complete-case analysis (CCA) design (i.e. excluding all individuals with missing values in one or more questionnaires) were fitted.” (Mangot-Sala et al., 2021, p. 8)

“Moreover, additional categories for occupational activity were included in a third set of sensitivity analyses: full-time students, homemakers, early retired and disabled were included from the second wave onwards.” (Mangot-Sala et al., 2021, p. 8)

“Our findings suggest that recent unemployment is more strongly associated with alcohol consumption than more distal exposures to unemployment (Virtanen et al., 2016). Moreover, individuals recently exposed to long-term unemployment experience a higher likelihood for Heavy Drinking (HD) and for Binge Drinking (BD).” (Mangot-Sala et al., 2021, p. 9)

““moderate/social drinkers” and abstainers” (Mangot-Sala et al., 2021, p. 9)

“Thus, unlike the majority of previous studies that assumed a homogeneous association between unemployment and alcohol consumption across the population, our study shows that it differs by subgroups of individuals, mainly by educational attainment” (Mangot-Sala et al., 2021, p. 9)

“Clear gender differences emerged” (Mangot-Sala et al., 2021, p. 9)

“As for partner status, models with imputed values show no significant differences between the association between unemployment and drinking patterns for single and partnered individuals” (Mangot-Sala et al., 2021, p. 9)

“Finally, we did not find any evidence that any of the associations substantially vary by age group” (Mangot-Sala et al., 2021, p. 9)

“final key contribution of our study is our focus on both HD and BD. Although unemployment increases both outcomes, we observe that those exposed to long unemployment are at higher risk of reacting by drinking intensively (BD), and not only of increasing the number of drinks (HD).” (Mangot-Sala et al., 2021, p. 9)

“First, data on alcohol consumption and unemployment was self-reported and may be affected by desirability bias (Backhans et al., 2012).” (Mangot-Sala et al., 2021, p. 9)

“Second, although our study relied on a large sample, the relatively low prevalence of unemployment often led to small groups when carrying out stratified analyses, thus reducing statistical power” (Mangot-Sala et al., 2021, p. 9)

“Third, in spite of relying on imputed values, there may be some selective loss to follow-up, as individuals with missing values in the outcomes were not included in the models” (Mangot-Sala et al., 2021, p. 9)

“Fourth, although causal inference was not the main focus of this paper, we implicitly tested one direction of the association, and the issue of reverse causality certainly deserves attention as well, as there is evidence suggesting that alcohol abuse may select some individuals into unemployment (Berg et al., 2017; Jorgensen et al., 2019).” (Mangot-Sala et al., 2021, p. 9)