a system dynamic model of drinking events: multi-level ... · research paper a system dynamic model...

■ Research Paper

A System Dynamic Model of DrinkingEvents: Multi-Level Ecological Approach

John D. Clapp1, Danielle R. Madden1*, Hugo Gonzalez Villasanti2,Luis Felipe Giraldo3, Kevin M. Passino2, Mark B. Reed4 andIsabel Fernandez Puentes21USC Suzanne Dworak-Peck School of Social Work, University of Southern California, Los Angeles, CA, USA2Electrical and Computer Engineering, The Ohio State University, Columbus, OH USA3Department of Electrical and Electronics Engineering, Universidad de Los Andes, Bogota, Colombia4 School of Social Work, San Diego State University, San Diego, CA USA

Drinking events are dynamic. The interactions of individuals, groups, and the environmentas they relate to drinking behaviour are overwhelmingly complex. This paper presents anempirically grounded dynamic conceptual model to better understand drinking events.Using a collaborative mixed-methods approach, we developed an aggregated system dy-namic model of drinking events. The process began with identification of system elementsand boundaries. Once the first aspects of the model were completed, we constructed acausal loop diagram, an aggregated causal loop diagram, and stock and flow diagrams.Finally, we developed and ran computer simulations of the dynamical models. The modelpresented here can be used to guide future agent-based, system dynamics, or differentialequation-basedmodels. Suchmodels can help inform future empirical work andmodellingto increase the understanding of drinking events and provide solutions to the problems thathappen proximal to these events. Copyright © 2018 John Wiley & Sons, Ltd.

Keywords conceptual models; drinking events; drinking behaviour; theoretical constructs

WHAT IS A DRINKING EVENT? TOWARD ASYSTEM DYNAMICS MODEL

Over the past half-century, a small but growingbody of research has emerged with the goal ofbetter understanding drinking behaviour as it

naturally occurs. Researchers hoping to betterunderstand how and why drinkers becomeintoxicated and experience related problemshave struggled to untangle the complexities ofan inherently ecological problem. Reflecting themulti-disciplinary nature of alcohol research,event studies vary in conceptual foci, methodol-ogy, and operational definitions. Independently,studies on ‘drinking contexts’, ‘drinking situa-tions’, and ‘drinking environments’ offer related

*Correspondence to: Danielle R. Madden, College of Social Work, TheOhio State University, 1947 College Rd. N., Columbus, OH 43210,USA.E-mail: [email protected]

Received 5 December 2016Accepted 5 July 2017Copyright © 2018 John Wiley & Sons, Ltd.

Systems Research and Behavioral ScienceSyst. Res 35, 265–281 (2018)Published online 15 January 2018 in Wiley Online Library(wileyonlinelibrary.com) DOI: 10.1002/sres.2478

but unique insights into drinking behaviour insitu. As a collective body of work, such studiessuggest the need for and offer the empirical basisof a conceptual modelling approach reflecting thecomplex and dynamic natures of drinkingevents.

Although conceptual models and theory havelong guided social science in general (Lewin,1951; Kuhn, 1974) and alcohol studies specifically(Denzin, 1987; Gusfield, 1996), models fordrinking events rarely build on previous work,transcend theoretical streams, or acknowledgedynamics and complexity (e.g., nonlinearity,random effects, and feedback loops). Althoughthere is a small body of system dynamic alcoholstudies at the community level (Gorman et al.,2001; Gruenewald, 2006; Holder, 2006; Scribneret al., 2009) and some recent notable exceptionsemploying agent-based modelling (Gormanet al., 2006; Fitzpatrick and Martinez, 2012) atthe population and event levels, dynamicalmodelling in alcohol research is still largelyunderdeveloped. At the event level, this maywell be an artefact of the difficultly of measuringdrinking events (Clapp et al., 2007; Kuntscheet al., 2014).

Although recent advances in data collectiontechnologies (Riley et al., 2011; Leffingwell et al.,2013) have the potential to advance our under-standing of event-level drinking behaviour, Rileyet al. (2011, p.54) note that our ability to collectindividualized, context-specific data and to inter-vene in situ has surpassed our current theories.The authors note that ‘health behavior modelsthat have dynamic, regulatory system compo-nents to guide rapid intervention adaptationbased on the individual’s current and past behav-ior and situational context’ are greatly needed.Recent studies have begun to embrace mobilecontinuous monitoring of physiological mea-sures, like heart rate, as a means of monitoringdrug and alcohol relapse triggers prior to havinga solid theoretical understanding of the underly-ing relationship between this indicator andrelapse triggers (Kennedy et al., 2015). A clearerunderstanding of dynamical relationships duringdrinking events will likely complement thefuture uses of such technologies by identifyingkey leverage points for targeted intervention.

In this paper, we offer a dynamic conceptualmodel of drinking events that is grounded inthe extant literature. Before presenting theconceptual framework, we briefly review thehistorical approaches to understanding drinkingevents. We also discuss the conceptualimportance of drinking events for understandingdrinking behaviour as a whole and preventingacute problems.

WHY ARE DRINKING EVENTS IMPORTANT?

Drinking events are direct antecedents to numer-ous acute alcohol-related problems includinginjuries, sexual and other violence, burns, falls,crashes, and crime, among many other problems(NIH, 2000). In aggregate, drinking events repre-sent patterns of consumption that drive diseaseand premature death (Holder, 2006). Acuteproblems have a huge global impact (Rehmet al., 2009); for instance, approximately 25% ofall unintentional and 10% of intentional injuriesin the world can be attributed to drinking events.When alcohol-related disease and death areconsidered, 3–4% of all deaths in the world arealcohol-related (Rehm et al., 2009).

Although heavier drinkers or those withalcohol-use disorders are at higher risk toexperience acute problems, lighter and moderatedrinkers who engage in heavy episodic drinkingaccount for the bulk of acute alcohol-relatedproblems (Stockwell et al., 1996). This so-called‘prevention paradox’ (Kreitman, 1986) suggeststhat universal environmental approaches (e.g.,driving under the influence campaigns, responsi-ble beverage service, taxation, and regulation)have historically been the primary means forpreventing acute alcohol problems. Although asolid evidence base exists for environmentalalcohol interventions (Holder, 2006; Saltz et al.,2010), newer ‘smart’ interventions (e.g., geo-fencing and SMS prompts) have the potential tocomplement universal environmental preventionefforts by targeting group-level or individual-level ‘leverage points’ (Stokols, 2000) in real timewhile considering the current behaviouralenvironment.

RESEARCH PAPER Syst. Res

Copyright © 2018 John Wiley & Sons, Ltd. Syst. Res 35, 265–281 (2018)DOI: 10.1002/sres.2478

266 John D. Clapp et al.

PAST APPROACHES TO UNDERSTANDINGDRINKING EVENTS

The conceptualization, definition, and measure-ment of drinking events have evolved little inrecent decades. Over 30 years ago, the NationalInstitute on Alcohol Abuse and Alcoholismpublished a monograph titled Social DrinkingContexts (Harford and Gaines, 1982). In the intro-duction to that collection of conference papers,Hartford and Gaines noted, ‘While context, orframe of reference, may hold the key to under-standing drinking behavior, no single idiomdescribes context’ (p.1). The authors go on to notethat the multi-disciplinary nature of alcoholstudies related to context reflects a spectrum ofterms and units of analysis. The nomenclatureand taxonomies used to frame drinking eventsstill reflect such diversity.In that same volume, drawing from the basic

social psychology theory of Lewin (1951), Jessor(1982) offered a simple multi-level representation:

DB ¼ f P;Eð ÞIn this formula, drinking behaviour is a func-

tion of the interaction of person-level variablesand environment influences. Jessor explicitly de-fines ‘context’ as ‘environment’. In his discussion,he notes two important considerations. First,‘(the environment) persists in being a concept ofdisturbing complexity’ (p.230; emphasis added).And then, ‘the dynamics of situations give rise tochanges in situations and behavior over time …an obvious source of such change is … alcoholingestion … and its disinhibition effects’ (p. 231;emphasis added).Some three decades later, understanding

drinking events from a system perspectiveremains a vexing problem. Since the publicationof Social Drinking Contexts (Harford and Gaines,1982), there has been great variation in theconceptualization, measurement, and analysis ofdrinking events. To start, it is important to notethat there is no standard definition of ‘drinkingevent’. Consistent with Jessor’s (1982) basicmodel, we conceptualize drinking events asincluding a drinker interacting socially with anetwork of other drinkers (and non-drinkers),embedded in larger social and physical

environments. Conceptually, we view the eventas a system that activates when drinking beginsand achieves entropy when both active drinkinghas ceased and the social purpose of the eventhas concluded. This approach differs from thenow-common practice of segmenting drinkingevents into time-specific (e.g., pre-gaming), social(e.g., drinking games), and/or geospatial (e.g.,bars) elements. Although analytically useful,such segmentation may obscure our understand-ing of the system as a whole and the complexnature of these events (Miller and Page, 2007).For instance, over the course of an individual’sdrinking event, pre-gaming can occur in a smallprivate setting, followed by drinking games in alarger party setting and culminating withdrinking in a public setting like a bar. Eachactivity and setting comes with its own dynamics(Clapp et al., 2008; Clapp et al., 2009; Fitzpatrickand Martinez, 2012), resulting in complexity(i.e., multi-level) and transitory risk (and protec-tion) across an entire event. Individuals interactsocially with peers, while their personal decisionsand desires are potentially influenced by groupdynamics, the larger environment, and theirown level of intoxication. Segmented approachesto studying drinking events miss much of thisbehaviour.

This paper hopes to further the scientific un-derstanding of the dynamics surrounding drink-ing behaviour as it naturally occurs. There areseveral dynamic problems related to drinking.First, although the biological dynamics associ-ated with metabolism and blood alcohol content(BAC) have been modelled, little is known abouthow individual desires relate to drinking effects(i.e., the rate of drinking, peak blood alcohollevels, and blood alcohol concentration curves)or how the consumption of alcohol impacts one’spersonal desires over the course of an event. Fur-ther, our model addresses the dynamical problemrelated to how a drinker’s drinking companionsinfluence a drinker’s desires; in turn, the modelpostulates the dynamics of how a drinkerinfluences their drinking companions. Finally,the model addresses how the drinking groupinfluences, and is influenced by, the drinking en-vironment. As a system, we view these dynamicsas being critical to better understanding the

Syst. Res RESEARCH PAPER


A System Dynamic Model of Drinking Events 267

complex nature of drinking as a social behaviourthat is inherently ecological.

METHODS

Our general approach to developing a dynamic-driven framework for drinking events isconsistent with Pentland’s (2014, pg. 5) approachto social physics: ‘Just as the goal of traditionalphysics is to understand how the flow of energytranslates into changes in motion’, our aim is tounderstand how different social or environmen-tal factors translate into changes in the dynamicsof a drinking event. While there is no standardapproach to developing dynamical models, wefollowed an approach similar to others(Richardson and Pugh, 1981; Sterman, 2002) byengaging in the following steps to develop ourmodel: (i) problem definition; (ii) system concep-tualization; (iii) model formulation; (iv) testingand simulation; and (v) model evaluation. Wenote that steps ii–v follow an iterative process.

The model presented in the succeeding textsrepresents our work to date. Consistent withothers (Richmond, 2004; Sterman, 2002), we areapproaching this modelling as an ongoingprocess that includes both conceptual andempirical stages. Here, we present our concep-tual work in a form we hope will facilitate itsuse by others.

Modelling Context

Our team consists of two full professors (one insocial work, one in engineering) and four doc-toral students (one in social work, three inengineering) working at the same university. Inaddition, we have a social work professorworking at another institution who did notparticipate in all modelling activities but servedas a reviewer—evaluating the logic of the modelsrelative to the existing literature, in addition toother tasks as the model developed. The socialwork members of the team have extensiveexperience in studying drinking events anddrinking behaviour in situ (Clapp andShillington, 2001; Clapp et al., 2003; Clapp et al.,

2007; Clapp et al., 2008; Clapp et al., 2009), andthe engineering members have extensive experi-ence in system dynamic modelling of behaviour(Passino and Seeley, 2006; Passino et al., 2008).

The team met twice monthly (about 90 min persession) to work on modelling. Group memberstook extensive notes and minutes, includinggraphic depictions of concepts, and those noteswere circulated regularly. It was common forgroup members to take on assignments betweenmeetings. An in-depth description of the teamscience aspects of this process is currently inprogress.

Problem DefinitionIn the early phases of the collaboration, the socialwork partners presented alcohol research relatedto drinking events (Clapp et al., 2008; Wells et al.,2008; Thombs et al., 2010; Neighbors et al., 2011;Clapp et al., 2014; Kuntsche et al., 2015). These pre-sentations helped to define the problem by specif-ically identifying the key elements of the drinkingevent system. Examples related to the influence ofvariables at different levels of abstraction werediscussed in depth. For instance, the relationshipbetween a drinker’s motivation and BAC at theindividual level (see: Neal and Carey, 2007;Wetherill and Fromme, 2009; O’Grady et al.,2011) was discussed. Similarly, studies related togroup influence on drinking (see Cullum et al.,2012; Reed et al., 2013; Wells et al., 2015) andenvironmental influences on drinking (see Clappet al., 2008; Clapp et al., 2009) were discussed.Additionally, the engineering team read keypapers related to BAC metabolism (Lundquistand Wolthers, 1958; Wilkinson, 1980; Norberget al., 2003; Jones, 2010).

System ConceptualizationConsistent with Richmond (2004), we developedhypothesized dynamical ‘behavior over timegraphs’ and causal loop diagrams (CLDs) alongwith other visuals to illustrate concepts andpotential dynamics during this stage of the work.In turn, the engineering members presented sim-ulations of computational models—based on thematerials presented by the social work members




—and explained underlying mathematical con-cepts. Early simulations were based on field datacollected by one of the team members in a previ-ous project (Clapp et al., 2009). Mathematicalmodels, proofs, and simulations related to theconceptual elements of the model presented inthis paper have been published in engineeringjournals (Giraldo et al., 2017a, 2017b).

Model FormulationDuring this phase of the work, the group jointlyconstructed stock and flow diagrams based onthe CLDs. We discussed the dynamics relativeto what would be expected logically—for exam-ple, one with strong motives to become ‘drunk’would likely have a higher BAC than a drinkerwith motives to become ‘buzzed’. Once thequalitative assessment of the initial stock andflow diagrams was complete, the engineeringteam developed a computational model toillustrate the dynamics of BAC metabolism andthe group influence aspects of the model (Giraldoet al., 2017a).

Testing and SimulationComputational models were built for environ-mental dynamics as well. Those models andsimulations were all performed in Simulink. Tofacilitate the social work team’s understandingof mathematical models, and to ensure that theengineering team had gotten drinking conceptscorrectly represented, generic versions usingrepresentative parameters of variables were con-structed in STELLA and Vensim, collaborativelyby our team.

Model EvaluationThe final stage of the initial modelling process in-volved qualitative and quantitative assessmentsof the simulation results. Qualitatively, the teamassessed simulation results relative to the extantliterature on drinking events. For instance,researchers (Trim et al., 2011) have shown that adrinker’s desired level of intoxication at thebeginning of an event often fails to match the out-come (i.e., they become more or less intoxicated

than they intended), suggesting that the interactionof the individual’s drinking and endogenous fac-tors (i.e., group influences or environmental influ-ences) are dynamically linked. As such, our modelneeded to allow for both ‘under-shooting’ and‘over-shooting’ of BAC over the course of an event.The models developed in previous stages werereviewed for this characteristic and others de-scribed in the results. Mathematically, the underly-ing differential equations had to have sound anddemonstrated proofs (Giraldo et al., 2017a, 2017b).

RESULTS

Figure 1 represents an overall conceptual modelof drinking events. The figure depicts individualsduring a drinking event (smallest circles), whoare situated in groups (enclosed in the samemedium-sized circle), drinking in specific envi-ronments (the largest circles). Lines between indi-viduals represent communication between groupmembers, across groups, or between people indifferent locations. Our model focuses on theaspects of a drinking event for only one of thesehypothetical drinkers (for example, the individ-ual marked A). The model is conceptually anextension of Jessor ’s simple heuristic path modelof drinking behaviour. That is, we have begun tofill in the various ecological elements at theperson level (i.e., biological variables andpsychological variables) that theoretically drivethe dynamics within the overall drinking eventsystem.

Figure 2 presents the CLD for the drinkingevent system. Conceptually, our model includedfour key stocks: (i) BAC; (ii) desired state (of in-toxication) or desired BAC; (iii) group wetness,and (iv) environmental wetness. From a socialecological framework, BAC and desired stateare micro-level variables, group wetness is amezzo-level variable, and environmental wet-ness is a macro-level variable. Together, thesestocks represent the various levels of abstractionfound in the literature examining drinking events(Jessor, 1982; Clapp et al., 2009; Reed et al., 2013)in a highly aggregated model consistent with a10 000 foot view of the system (Richmond, 2004).




Beginning from the bottom of the CLD (micro-level) in Figure 2, there is a balancing causalfeedback loop between the BAC stock and themetabolic rate (Giraldo et al., 2017a). A delay isintroduced before the BAC stock, representingthe transit of alcohol through the gastrointestinaltract before entering the blood by an absorptionprocess (it could also be delayed further by foodintake). Put in practical terms, decisions bydrinkers whether to have another drink tomaintain or obtain a ‘buzz’ are often based on amisperception of the amount of alcohol they havein their system (Richmond, 2004).

Moving up a level, the feedback loop related todrinking motives, perceptions of intoxication,and drinking behaviour is represented. This loopsegment of the CDL can be thought of as the‘goal seeking’ segment of the model. As notedin the structure of the CLD, there is a tension be-tween the rational desire to obtain a desired stateof intoxication and maintain it (in a balancingfeedback loop) and the ‘alcohol myopia’ or thedisinhibitory effects of alcohol on cognitive con-trol (reinforcing loop), where an individual’sdesire to drink might increase in vivo with thenet effect of over-shooting their original desiredstate of intoxication (Steele and Josephs, 1990;Field et al., 2010). A full operational modelincluding all stocks, flows, and connectors ispresented in the appendix.

Continuing with Figure 2 and moving up alevel of abstraction, the drinker’s desired BAC

stock is also influenced by the group wetnessstock in a reinforcing feedback structure. As de-scribed in Table 1, group wetness is a form ofsocial influence that includes the average BACof a drinker’s companions at the drinking event,the average desired BAC of the drinker’s com-panions, and the relative influence of each mem-ber of the group on the drinker. In our earlierwork (Giraldo et al., 2017a), we modelled howpeer influence varies in strength and interactswith a drinker’s own desired state to alterdrinking trajectories. Through a series of com-puter simulations, we showed that a strong in-fluence within a peer network pulls all butthose with very strong desires toward a BACtrajectory similar to the peer exerting theinfluence. In turn, completing this reinforcingfeedback loop as a member of the group, thedrinker’s BAC and desired BAC also influencethe group wetness stock.

Finally, moving to the top of the CLD inFigure 2 to the group and environmental levels,we posit a reinforcing feedback loop betweengroup wetness and environmental wetness. Ourearlier studies of drinking events (Clapp et al.,2000; Clapp et al., 2009) found that the presenceof ‘many intoxicated people’ (whether observedby researchers or reported by survey respon-dents) consistently contributed to high BAC orself-reported heavy drinking. We also found thatheavier drinkers seek out wetter environments(Trim et al., 2011), suggesting that influence flows

Figure 1 Conceptual model of drinking events. [Colour figure can be viewed at wileyonlinelibrary.com]




http://wileyonlinelibrary.com

in both directions. As group wetness increases,environmental wetness increases. In turn, envi-ronmental wetness, which represents the overallaverage BAC among bar patrons coupled withalcohol availability in the environment (Clappet al., 2009), influences group wetness in a rein-forcing way.Table 1 describes each element in the model,

except for the GAC stock, which simply

represents the aggregate amount of alcohol (forexample, the unit of measure could be standarddrinks) residing in the gastrointestinal tract. Weoffer basic definitions of each element, theoreticalparameters (based on our mathematical models),how the elements might be measured (or havebeen), and some assumptions about how eachelement operates in the system based on theliterature and our previous research.

Figure 2 Causal loop diagram for drinking event system. [Colour figure can be viewed at wileyonlinelibrary.com]





Figure 3 shows a series of our hypothesizedreference behaviour of time graphs for variousBAC outcomes generated via simulation of themodel found in Figure A.1. Although we are in-terested in conceptually understanding the entiredrinking event system, understanding how dif-ferent elements of the model affect the BAC isparticularly important for guiding prevention

efforts. The graphs shown in Figure 3A portraythe effect of metabolism on BAC. The plots inFigure 3B shed light on the reinforcing cognitiveeffects on BAC, while the plots in Figure 3C showthe effect of peer and environment influence onindividual’s intoxication. In all cases, it is as-sumed that the drinking period lasts for 3 h.The model parameters are provided in Table 2.

Table 1 Description of elements in the model

BAC Desired BAC Group wetness Environmental wetness

Ecologicallevel ofabstraction

BiologicalMicro

PsychologicalMicro

SocialMezzo

Social/physicalMacro

Definition Represents thedrinker’s bloodalcoholconcentration

The state ofintoxication a drinkerhopes to obtain atany given pointduring the drinkingevent

Represents the meanBAC of the peergroup over the event.Could also includemean of group’sdesired state

Represents theextent theenvironmentpromotes heavydrinking

Parameters Grams ofethanol/100 MLof bloodValues range from0.0 to 0.30 (boundednot to go below 0.0or above 0.30 toreflect typical values)

A standard drink= 0.02 (+) in BAC

Metabolism = 0.02 (�)in BAC per hour

(Dubowski, 1985)

Measured on Likertscale with valuesranging from ‘drinkbut not get buzzed’to ‘drink to get verydrunk’

Can also measurestrength of desire byLikert scale

(Clapp et al., 2009;Thombs et al., 2010)

Mean values rangingfrom 0.0 to 0.30

Measured by anindex of availabilityincluding averagedollar amount perstandard drink,average time toobtain a drink, andthe number of fixedand temporaryservers. Could alsoinclude the socialaspects of theenvironment such asthe presence of manyintoxicated people(Clapp et al., 2009)

Assumptions Influenced by rateand volume ofdrinking

Influences desiredstate

Food delaysmetabolism.

Delay drinking andability to perceiveBAC level

May be conscious orsubconscious

Can shift over thecourse of event

Initial strength ofdesire relates tolikelihood of change

Influences and isinfluenced by groupwetness

Some groupmembers might havemore influence thanothers.

Influences individualdrinker’s BAC anddesired state

Is influenced byindividual drinker’sBAC and desiredstate

Influences and isinfluenced byenvironmentalwetness andselection

Influences and isinfluenced by groupwetness

Influences and isinfluenced byindividual drinkerBAC and desiredstate

Is selected based onthe weighted mean ofa group’s desiredstate (giving moreweight to influentialgroup members




A

Figure 3 Blood alcohol content (BAC) curves under different conditions. [Colourfigure can be viewed atwileyonlinelibrary.com]





Figure 3 Continued




While the graphs in Figures 3B and 3C onlyillustrate ‘peak’ BAC and do not show the de-cline of BAC back to zero, Figure 3A shows theentire BAC curve for a drinker who desired tobecome ‘very buzzed or drunk’. As presented,the drinker reaches a peak BAC of over 0.1 byhour 3 of the event, before the BAC begins to de-crease slowly. The fluctuations in the decreasingBAC are due to consumption of more alcohol af-ter stopping drinking for a period. The steepgrowth of BAC in the early hours of the eventis related to the rate of drinking. That is, to ob-tain the BAC shown, the rate of drinking wouldbe fairly fast.

The graph shown in Figure 3B illustrates theBAC curve for a drinker with the desire to ‘getbuzzed (desired BAC = 0.06)’, with lowinfluence from the group and environmentwetness, so his desired BAC remains constant.In this graph, the drinker reaches the desiredintoxication level after 80 min of drinking. How-ever, due to the disinhibitory effect that increaseshis desire to drink, the individual does not de-crease the drinking rate to zero but continuesdrinking at a lower rate, increasing his BAC level.

The final graph in the series, Figure 3C, illus-trates a drinker who desires to ‘get buzzed(desired BAC = 0.06)’ but is pulled off thattrajectory later in the event. Conceptually, such‘overshooting’ is a function of group and/orenvironmental dynamics. Empirically, we foundthis to be fairly common during drinking events(Clapp et al., 2009; Giraldo et al., 2017a; Trimet al., 2011). The mathematics underlying thedynamics of overshooting or undershootingdesired intoxication levels is presented in ourmore technical work (Giraldo et al., 2017a).

Finally, Appendix presents potential randomvariables (disturbances), at each level ofabstraction that, theoretically, might alter thedynamics in the model. The list is provided asboth a means to set the exogenous boundariesof the model and to guide potential simulationsin the future.

DISCUSSION

Drinking events remain an important area ofstudy for alcohol researchers. Understandingdrinking events and the complex dynamics thatunderlie them is important both conceptuallyand to help guide prevention efforts that utilize‘smart’ technologies in situ. Our model andprevious work (Giraldo et al., 2017a, 2017b)advance a conceptual approach, which we hopewill aid understanding of drinking behaviourwhile guiding the development of preventionapproaches.

By considering the underlying interactionsamong the biological, psychological, social, andenvironmental interactions related to drinkingbehaviour as it occurs, the model presented hereis one of the few attempts to address the inherentcomplexity of drinking events noted by Jessor(1982) and Harford and Gaines (1982), over threedecades ago. Our conceptual and mathematicalresults thus far begin to illustrate the potentialof dynamics at several levels resulting in individ-uals drinking heavily and more than they ini-tially intended. Initial intentions are important.In our model, we posit a reinforcing feedbackloop among perceptions/cognitive effects, de-sired state, drinking, and BAC. In theory, a

Table 2 Model parameters

Figure Parameter Value

3A Elimination parameter 0.0067Absorption parameter 0.05Effect of food 0BAC rate perception parameter 0.1Decision commitment 0.0225Alcohol in CNS/blood ratio 0.6451Inhibitory effect parameter 0.01Env strength on individual 0Group strength on individual 0Initial desired BAC 0.1Initial GAC and BAC 0

3B Inhibitory effect parameter 0.02Initial desired BAC 0.03

3C Env strength on individual 0.002Group strength on individual 0.005Ind strength on group 0.0005Ind strength on Env 0.0005Group strength on Env 0.0005Env strength on group 0.002Initial desired BAC 0.06

Note: Parameters not listed in Figures 3B and 3C remain thesame as in Figure 3A




drinker with the motivation to have a ‘no buzz’or a ‘slight buzz’ who drank slowly enough tohave fairly accurate perceptions of their intoxica-tion (or stopped drinking after a drink or two)could be represented by a balancing feedbackloop where equilibrium is achieved. However,as noted by other researchers (Trim et al., 2011;Giraldo et al., 2017a), drinkers’ initial motivesfor a level of ‘buzz’ often do not reflect theiractual BAC (i.e., drinkers become more intoxi-cated than intended). Further field workis needed to better understand how thesedynamics might differ based on initial desiresfor intoxication.The modelling efforts and simulation results

presented here (fig. 3) illustrate the importanceof dynamics in drinking behaviour. For instance,figure 3B shows how a drinker’s desire impactsand is impacted by BAC levels. The issue ofhow one’s BAC curve impacts an overall drink-ing event was raised by Jessor over 30 years ago(Jessor, 1982), yet little work to date has focusedon this issue. Figure 3B illustrated an exampleof ‘overshooting’ where the drinker ends updrinking more than intended, resulting in ahigher BAC. Figure 3C illustrates how the GItract results in a delay in BAC, the mechanismthat theoretically accounts for overshooting.Taken together, the simulation leaves in a mannerconsistent with the CLD presented in figure 2.Future work—both empirical and computa-

tional—will be needed to validate and refine theconceptual model. Field studies with high eco-logical validity (Clapp et al., 2007) are needed toexamine social network influences as they relateto desired intoxication and actual drinking out-comes. Similarly, more work is needed to biolog-ically validate BAC curves as they relate toelements of the drinking event system. Likewise,better understanding of how individuals decideto continue or stop drinking to achieve a desiredlevel of intoxication must be better developed.Better measures of environmental wetness mustalso be created and tested. Finally, applying theknowledge generated by the dynamical model-ling and validation process must identify lever-age points to guide the development and testingof preventive interventions. Such work is nevercomplete; nor is it easy.

Our approach required the collective effort of ateam of scientists from disparate disciplinesworking together in a highly collaborative man-ner for a considerable period. Pulling togetherthe relevant aspects of the drinking event, litera-ture with appropriate mathematical formulationsdrawn from physics and engineering requiredboth parties to be simultaneously open and criti-cal. The extant literature on drinking events, witha few notable exceptions (Gorman et al., 2006;Fitzpatrick and Martinez, 2012), rarely considersmultiple levels of abstraction and is almost exclu-sively grounded in static linear models—makingthe jump to developing a dynamical model chal-lenging. Similarly, the application of principlesrelated to physics (like force and attraction) usedin our computational work (Giraldo et al., 2017a)had to be carefully applied to a social behaviour.As this line of work continues and, we hope, ex-pands to other groups, the multi-disciplinarymethod described here must evolve into a trans-disciplinary approach.

In this spirit, others have noted that system dy-namic maps and CLDs are essentially heuristicdevices to explain complex behaviour in an ele-gant and aggregate form (Richmond, 2004) andguide applied intervention work (BeLue et al.,2012). One challenge of both the current workpresented and future work will be to develop acommon system of visual explanation (e.g., con-ventions for drawing CLDs, stock, and flow dia-grams). CLDs and stock and flow diagrams canbe useful when carefully presented. They can,however, be overly complex and confusing. Find-ing the correct balance of system specificationthat is ecologically valid without sacrificing ac-cessibility can be challenging. Similarly, develop-ing a common nomenclature for discussing andstudying drinking events will be important. Be-yond the scientific community, using collabora-tive model building approaches (BeLue et al.,2012; Hovmand, 2013) and visual simulationsusing free software packages like Vensim andMental Modeller might facilitate the understand-ing of these complex systems. Like others work-ing on collaborative models to tackle importantreal-world problems, we hope that such effortswill help move science into applied situationsfaster and in a more ecologically valid way.




ACKNOWLEDGEMENTS

This study was conducted, in part, throughfunding from the College of Social Work at TheOhio State University.

REFERENCES

BeLue R, Carmack C, Myers KR, Weinreb-Welch L,Lengerich EJ. 2012. Systems thinking tools as appliedto community-based participatory research: a casestudy. Health Education & Behavior 39(6): 745–751.

Clapp JD, Shillington AM. 2001. Environmental predic-tors of heavy episodic drinking events. The AmericanJournal of Drug and Alcohol Abuse 27(2): 310–313.

Clapp JD, Shillington AM, Segars L. 2000.Deconstructing contexts of binge drinking amongcollege students. The American Journal of Drug andAlcohol Abuse. 26(1): 139–154.

Clapp JD, Lange J, Min JW, Shillington A, Johnson M,Voas R. 2003. Two studies examining environmentalpredictors of heavy drinking by college students.Prevention Science 4(2): 99–108.

Clapp JD, Holmes MR, Reed MB, Shillington AM,Freisthler B, Lange JE. 2007. Measuring collegestudents’ alcohol consumption in natural drinkingenvironments: field methodologies for bars andparties. Evaluation Review 31(5): 469–489.

Clapp JD, Min JW, Shillington AM, Reed MB, Croff JK.2008. Person and environment predictors of bloodalcohol concentrations: a multi level study of collegeparties. Alcoholism: Clinical and Experimental Research32(1): 100–107.

Clapp JD, Reed MB, Min JW et al. 2009. Blood alcoholconcentrations among bar patrons: a multi-levelstudy of drinking behavior. Drug and AlcoholDependence 102: 41–48.

Clapp JD, Reed MB, Ruderman DE. 2014. The relation-ship between drinking games and intentions to con-tinue drinking, intentions to drive after drinking,and adverse consequences: results of a field study.The American Journal of Drug and Alcohol Abuse40(5): 374–379.

Cullum J, O’Grady M, Armeli S, Tennen H. 2012.Change and stability in active and passive social in-fluence dynamics during natural drinking events: alongitudinal measurement-burst study. Journal ofSocial and Clinical Psychology 31(1): 51–80.

Denzin NK. 1987. The Alcoholic Self. Sage Publications:Thousand Oaks, CA.

Dubowski KM. 1985. Absorption, distribution andelimination of alcohol: highway safety aspects. Jour-nal of Studies on Alcohol, Supplement No. 10: 98–108.

Field M, Wiers RW, Christiansen P, Fillmore MT,Verster JC. 2010. Acute alcohol effects on inhibitory

control and implicit cognition: implications for lossof control over drinking. Alcoholism: Clinical and Ex-perimental Research 34(8): 1346–1352.

Fitzpatrick B, Martinez J. 2012. Agent-based modelingof ecological niche theory and assortative drinking.Journal of Artificial Societies and Social Simulation15(2): 4–17.

Giraldo LF, Passino KM, Clapp JD. 2017a. Modelingand analysis of group dynamics in alcohol-consumption environments. IEEE Transactions on Cy-bernetics 47(1): 165–176.

Giraldo LF, Passino KM, Clapp JD, Ruderman D. 2017b.Dynamics of metabolism and decisionmaking duringalcohol consumption: modeling and analysis. IEEETransactions on Cybernetics 47(11): 3955–3966.

Gorman DM, Speer PW, Gruenewald PJ, Labouvie EW.2001. Spatial dynamics of alcohol availability,neighborhood structure and violent crime. Journalof Studies on Alcohol and Drugs 62: 628–636.

Gorman DM, Mezic J, Mezic I, Gruenewald PJ. 2006.Agent-based modeling of drinking behavior: a pre-liminary model and potential applications to theoryand practice. American Journal of Public Health96(11): 2055–2060.

Gruenewald PJ. 2006. The spatial ecology of alcoholproblems: niche theory and assortative drinking.Addiction 102(6): 870–878.

Gusfield JR. 1996. Contested Meanings: The Constructionof Alcohol Problems. University of Wisconsin Press:Madison, WI.

Harford TC, Gaines LS. 1982. Social drinking contexts:an introduction. In Harford TC, Gaines LS, eds. So-cial Drinking Contexts: Proceedings of a Workshop.Research Monograph No 7. Rockville: NIAAA: 1-7.

Holder HD. 2006. Alcohol and the Community: A SystemsApproach to Prevention. Cambridge University Press:Cambridge, UK.

Hovmand PS. 2013. Community Based System Dynamics.Springer: New York, NY.

Jessor R. 1982. Some problematic aspects of research ondrinking contexts. In Social Drinking Contexts:Proceedings of a Workshop, Harford TC, Gaines LS(eds.), Research Monograph No 7. NIAAA:Rockville; 228–233.

Jones AW. 2010. Evidence-based survey of the elimina-tion rates of ethanol from blood with applications inforensic casework. Forensic Science International200(1): 1–20.

Kennedy AP, Epstein DH, JobesML et al. 2015. Continu-ous in-the-field measurement of heart rate: correlatesof drug use, craving, stress, and mood in polydrugusers. Drug and Alcohol Dependence 151: 159–166.

Kreitman N. 1986. Alcohol consumption and the pre-ventive paradox. British Journal of Addiction to Alcoholand Other Drugs 81: 353–363.

Kuhn A. 1974. The Logic of Social Systems: A UnifiedDeductive System Based Approach to Social Science.Jossey-Bass Publishers: San Francisco, CA.




Kuntsche E, Dietze P, Jenkinson R. 2014. Understand-ing alcohol and other drug use during the event.Drug and Alcohol Review 33(4): 335–337.

Kuntsche E, Otten R, Labhart F. 2015. Identifying riskydrinking patterns over the course of Saturday eve-nings: an event-level study. Psychology of AddictiveBehaviors 39(3): 744–752.

Leffingwell TR, Cooney NJ, Murphy JG et al. 2013.Continuous objective monitoring of alcohol use:twenty first century measurement using transdermalsensors. Alcoholism: Clinical and Experimental Research37(1): 16–22.

Lewin K. 1951. Field Theory in Social Science: SelectedTheoretical Papers. Harpers: Oxford, UK.

Lundquist F, Wolthers H. 1958. The kinetics of alcoholelimination in man. Acta Pharamcologica etToxicologica 14(3): 265–289.

Miller JH, Page SE. 2007. Complex Adaptive Systems: AnIntroduction to Computational Models of Social Life.Princeton University Press: Princeton, NJ.

National Institute of Health. 2000. 10th special reportto the U.S. congress on alcohol and health. U.S.Department of Health and Human Services.

Neal DJ, Carey KB. 2007. Association between alcoholintoxication and alcohol-related problems: anevent-level analysis. Psychology of Addictive Behaviors21(2): 194–204.

Neighbors C, Atkins DC, Lewis MA et al. 2011. Event-specific drinking among college students. Psychologyof Addictive Behaviors 25(4): 702–707.

Norberg A, Jones AW, Hahn RG, Gabrielsson JL. 2003.Role of variability in explaining ethanol pharmaco-kinetics. Clinical Pharmacokinetics 42(1): 1–31.

O’Grady MA, Cullum J, Tennen H, Armeli S. 2011.Daily relationship between event-specific drinkingnorms and alcohol use: a four-year longitudinalstudy. Journal of Studies on Alcohol and Drugs 72:633–641.

Passino K, Seeley TD. 2006. Modeling and analysis ofnest-site selection by honey bee swarms: the speedand accuracy trade-off. Behavioral Ecology and Sociobi-ology 59: 427–442.

Passino KM, Seeley TD, Visscher PK. 2008. Swarm cog-nition in honey bees. Behavioral Ecology and Sociobiol-ogy 62: 401–414.

Pentland A. 2014. Social Physics: How Good Ideas Spread—The Lessons From a New Science. The Penguin Press:New York, NY.

Reed MB, Clapp JD, Martell B, Hidalgo-Sotelo A. 2013.The relationship between group size, intoxicationand continuing to drink after bar attendance. Drugand Alcohol Dependence 133: 198–203.

Rehm J, Mathers C, Popova S, Thavorncharoensap M,Teerawattananon Y, Patra J. 2009. Global burden ofdisease and injury and economic cost attributableto alcohol use and alcohol-use disorders. The Lancet373(9682): 2223–2233.

Richardson GP, Pugh AL. 1981. Introduction to SystemDynamics Modeling with Dynamo. MIT Press:Cambridge, MA.

Richmond B. 2004. An Introduction to Systems Thinkingwith STELLA. isee systems, inc: Lebanon, NH.

RileyWT, Rivera DE,AtienzaAA,NilsenW,Allison SM,Mermelstein R. 2011. Health behavior models in theage of mobile interventions: are our theories up tothe task? Translational Behavioral Medicine 1(1): 53–71.

Saltz RF, Paschall MJ, McGaffigan RP, Nygaard PM.2010. Alcohol risk management in college settings:the safer California universities randomized trial.American Journal of PreventiveMedicine 39(6): 491–499.

Scribner R, Ackleh AS, Fitzpatrick BG et al. 2009. Asystems approach to college drinking: developmentof a deterministic model for testing alcohol controlpolices. Journal of Studies on Alcohol and Drugs 70:805–821.

Steele CM, Josephs RA. 1990. Alcohol myopia. Itsprized and dangerous effects. American Psychologist45(8): 921–933.

Sterman JD. 2002. All models are wrong: reflections onbecoming a systems scientist. System Dynamics Re-view 18(4): 501–531.

Stockwell T, Hawks D, Lang E, Rydon P. 1996.Unravelling the preventive paradox for acute alcoholproblems. Drug and Alcohol Review 15(1): 7–15.

Stokols D. 2000. Creating health-promotive environ-ments: implications for research and theory. In Pro-moting Human Wellness: New Frontiers for Researchand Practice, Jamner MS, Stokols D (eds.). Universityof California Press: Berkeley, CA; 135–162.

Thombs DL, O’Mara RJ, Tsukamoto M et al. 2010.Event-level analyses of energy drink consumptionand alcohol intoxication in bar patrons. Addictive Be-haviors 35: 325–330.

Trim RS, Clapp JD, ReedMB, Shillington A, Thombs D.2011. Drinking plans and drinking outcomes: exam-ining young adults’ weekend drinking behavior.Journal of Drug Education 41(3): 253–270.

Wells S, Mihic L, Tremblay PF, Graham K, Demers A.2008. Where, with whom, and how much alcohol isconsumed on drinking events involving aggression?Event-level associations in a Canadian national sur-vey of university students. Alcoholism: Clinical andExperimental Research 32(3): 522–533.

Wells S, Dumas TM, Bernards S, Kuntsche E, Labhart F,Graham K. 2015. Predrinking, alcohol use, and breathalcohol concentration: a study of young adult bargoers.Psychology of Addictive Behaviors 29(3): 683–689.

Wetherill RR, Fromme K. 2009. Subjective responses toalcohol prime event-specific alcohol consumptionand predict blackouts and hangover. Journal ofStudies on Alcohol and Drugs 70: 593–600.

Wilkinson P. 1980. Pharmacokinetics of ethanol; a re-view. Alcoholism: Clinical and Experimental Research4(1): 6–21.




APPENDIX A.

Figure A.1 Full stock and flow diagram for drinking event system. [Colour figure can be viewed at wileyonlinelibrary.com]





APPENDIX A.1.

TABLE B.1. POTENTIAL RANDOM VARIABLES AND DISTURBANCES BY LEVEL OFABSTRACTION

Level Random variable

Micro/individual •Taking medication•Level of hydration•Level of rest•Drinking history (e.g., binge drinker)•Genetic markers•Tolerance

Mezzo/group •New group member(s) during event•Exiting group members during event•Sexual attraction among group members•Social media or SMS connection among group members during event

Macro/environment •Moving locations during event•Fights or aggression at events•Introduction of music or dancing during event•Influx or outflow of other groups during event

Note: This table provides potential variables and disturbances and does not reflect specifically what is in our model. This is not anexhaustive list.




a system dynamic model of drinking events: multi-level ... · research paper a system dynamic model...

Documents