Challenging Dogma


Friday, April 27, 2012

The Failure of Gamification in Fitness Behavior Modification From The Perspective of Social-Cognitive Theory - James Henry Steinberg

Obesity and metabolic syndrome represent crisis for the American health care system and its public, in terms of both cost burden and comorbidities. In the preceding quarter century, the entirety of the U.S. population weight distribution has shifted upward (1), resulting in more than two thirds of the adult population now being overweight or heavier (2). Diabetes Mellitus, the most prominent manifestation of metabolic syndrome, has increased in its proportion of U.S. adults from 3.7% in 1980 (age-adjusted) to 8.7% in 2010 (age-adjusted)(3).

Behavioral approaches towards weight-loss have thus far failed: over 80% of individuals return to pre-weight loss levels of body fatness after otherwise successful weight loss, from a variety of weight loss approaches(4). There has been an increased move away from an individual focus on obesity, to the examination of the “obesogenic environment” - the genetic and environmental factors that predispose people towards obesity (5,6).

The most recent trend in fitness behavior modification involves “gamification” (7). Gamification involves the integration of game mechanics into marketing, health, web domains, etc. for the purpose of influencing and motivating groups of people. Gamification is built on the essential findings of behavioral biologists regarding operant conditioning. “Game” mechanisms are usually restricted to what are traditionally known as “role playing games” (RPGs). The RPG genre is identified by incremental advances in game difficulty, coupled to incremental advances in the player’s efficacy, ordinarily distributed by the player, and arriving in a predictable and systematic manner – often easily at first, with increasing times between increments later in the game. These incremental increases in efficacy are collectively termed “leveling”. The game here presents a very simple sort of operant conditioning: perform action X (ordinarily “kill the monster”, although “record 5 push-ups” works just as well), and one receives positive reinforcement in the form of increased efficacy in one’s endeavors. Self-efficacy is high here, as action and achievement are linked in an obvious and deterministic manner.

RPGs also involve more complex conditioning mechanisms. Early behavioral biologists found that the schedule of reinforcement is important. If reinforcement ceases after it a behavior has been linked to it, a behavior may be extinguished. Reinforcement can be spaced out (as leveling is in RPGs), though again – this can lead to extinction. It has been found that the ideal schedule involves early deterministic rewards, followed by their increasing spacing out over time, with a transition into probabilistic rewards. If rewards are consistent on average, but highly variable in time, subjects are shown to be quite susceptible to reinforcements that ultimately begin to arrive over long periods of time (8). RPGs often incorporate this via “item drops” – efficacy increases in the symbolic form of accoutrement for one’s protagonist. These occur in a deterministic fashion (guaranteed on the completion of a certain event), coupled with a small probabilistic event throughout the game. The quality of such “drops” varies in a probabilistic fashion, with greater efficacy increases being correspondingly rarer.

Games have also begun to incorporate more social elements. “Achievements” are deterministic events that, upon their completion in the game, usually are not linked to efficacy increases. Rather, they become visible over game-relevant networks, showcasing one’s triumphs to their peers. In a dual role, it also serves to model to peers that the game in question is being played, and with some measure of dedication. This serves as a form of observational learning to encourage others to partake in the game in question.

Gamification has attempted to incorporate each of these mechanisms. The most prominent example to date is the exercise gamification “Fitocracy.” The website involves a facebook-like social network, where users self-label as being interested in particular types of fitness activities (i.e., “powerlifters”). This serves to involve them in conversation with their cohort. The central function of the site, however, is the ability to enter one’s workouts. Each exercise is worth points and these points add up to allow one to increase their “level.” One can also gain Achievements through particular routines (i.e. doing a bench press, a deadlift, and a squat press all in one week). One’s level and achievements are publicly displayed to one’s network. The leveling mechanism is thought to encourage exercise by providing a deterministic positive reinforcement. To quote its creators, “what if fitness could be turned into a game? After all, both [Richard] and Brian understood how addictive it could be trying to get to that next level, beating that next boss, and completing that next quest… They also realized that the addiction that games create was the exact same addiction that drives their fitness efforts every day.” (9) The social components allow groups to act as models for observational learning, as well as incentivizing (one must gain the deadlift-squat-bench achievement not to be an outcast among powerlifters).

Fitocracy serves as a prominent example, having been one of the first attempts to “gamify” public health, but it is hardly an anomaly. Design companies are now framing gamification as the “future of health care” (10). HopeLab hopes to combat child obesity with the “Zamzee”, a device students wear to track their physical activity. More active children get points, level up, show off on a social network, and purchase goods for “winning” (11). Zamzee’s approach is untenable however, in that creating tangible goods as a reward has worked in their limited (12-week, 350-subject) laboratory studies, but is something they are unable to scale for use with the general public. One of the largest has been the iPod-integrated Nike+, which uses a GPS sensor to track runners and later to upload their data, track statistics, join challenges, and connect (and compete) with other runners in the Nike network.

Modern social gaming mechanics are attractive to those attempting to “hook” people into healthy activities. They seem to leverage the power of social influence (observational learning), as well as the various elements of reciprocal conditioning that have made otherwise-mindless games such as “Farmville” and “World of Warcraft” such runaway successes. If something as “boring” as Warcraft could be so addictive as to ruin lives (12), then why couldn’t such mechanisms be used to make fitness and health equally attractive?

Unfortunately, such “gamification” attempts take on the shape of cargo-cult science. Although they utilize the appearance of social cognitive theory, they fail to utilize its full substance so as to actually enroll the full psychological mechanisms which are needed to create behavioral change.

The fundamental principle behind “leveling” is that it fulfills the “self-efficacy” portion of the SCT model. Increased “levels” in a game environment allow a person to better navigate the game world, or compete against others. That is to say, that “leveling” has utilitarian value, but not intrinsic. The actual positive reinforcement is the pleasure derived from increased ability to navigate challenges. Current “gamification” approaches to driving participation have failed to connect gamification mechanisms to intrinsically valuable outcomes, and as such have yet to import “addictive” game-playing behaviors to gaming.

An additional failure, though less severe than the failure to link game elements to real incentives, is the loss of the probabilistic reinforcement element that creates the optimal reinforcement schedule. Games incorporate this almost automatically, as it is often driven by the in-game fiction. Gamified health elements, however, have pursued a strictly deterministic course: perform X pushups, get 2X points, 300 points needed for level 2. This has been so in all of the major health and fitness programs that have used gamification elements: Fitocracy, Zamzee, Nike+, etc. This is bound to reduce the efficacy of these programs. I would argue it could be enough to result in extinction, however, as noted above, these programs have never linked into an actual incentive: without a notable positive stimulus, it can be argued they’re not providing significant positive reinforcement at all.

The standard gaming environment also has a form of reciprocal determinism. Difficulty scales in reaction to player action; if the player chooses not to advance difficulty, but chooses to increment efficacy, the player can do so. The player effectively has control over the challenges he faces, in large part because the game is based on increases in efficacy, and increases in obstacles proceed (usually) independently. This creates a form of reciprocal determinism: the game’s opportunities for advancing efficacy and obstacle lay-out provide challenges, and the player can navigate these challenges in a manner that optimizes difficulty. Gamification platforms have cast this away: insofar as they do not provide challenges, only arbitrary metrics of current performance, there is little “difficulty” to be had. Fitocracy institutes “quests” and “achievements” but these are generally not difficult, and the difficulty does not scale with player efficacy: the player has no ability to shape his environment.

Fitocracy, for instance, provides points, levels, quests, and achievements, all tied into a social network. Theoretically one is driven to advance one’s exercise regime by chasing new quests and achievements, stabilize it as a habit by chasing points & levels, and introduce the element of observational learning and social approval through the built-in social network that shares these various positive reinforcements with a social network (and broadcasts theirs). It is further facilitated by the development of an iOS application, which allows one to keep up with it actively in the gym. However, the only distinction between a level 30 individual (one who is actively committed to fitness and has been for a significant period of time) and a level 4 individual (perhaps one that has been actively working out for two weeks) is the incremented digit beside their name. If the social mechanism is not itself sufficient motivation to attend to one’s workout, the various gamified elements offer no additional incentive: there is no change in outcome expectation by adding the game elements to one’s life. One could substitute it entirely through a social mechanism (i.e., posting one’s exercises to a facebook group where others do the same) and change little or nothing. Recent updates attempt to add a “player vs player” aspect, in that one active exerciser can challenge one another to “combat”, resolved by the workouts they undergo. However, this is unrelated to the existing game mechanics, and so does nothing to add value to the gamification elements. It is tantamount to saying “let’s see who can bench press more”, and building it into a game. Even that is questionable, as the social norms of weightlifting analogize scrutinizing/competing with someone else’s weights (outside of professional competitions) with romantic glances at someone else’s spouse: people may in fact do it, but social norms are stacked against it. Creating a game element for people already in the gym, stacked directly against prevailing social norms, is unlikely to have a large effect on gym-goers. Moreover, the entirely deterministic “leveling” pace of Fitocracy hinders an optimal reinforcement schedule – they have the “levels,” but not the probabilistic “item drops” which maintain interest after levels have been maximized or extremely infrequent.

Much the same could be said of Zamzee, created by the non-profit Zamzee with the explicit aim of helping children and teens manage their weight. It incorporates many of the standard gamification mechanisms, and it would seem to correct the primary deficit: points and levels ultimately lead to a “victory” condition, and that is connected with tangible rewards. However, where Fitocracy scales as high as the number of servers their software can afford, the non-profit Zamzee must now purchase and gift tangible goods. For their 350-subject test, they claim to have provided such rewards, without specifying what the rewards were. This is clearly an unscalable project, however: expanding this to teens en masse would require an enormously expensive outlay of products, or the cheapening of such rewards until they are no rewards at all. Although this nominally attempts to address the lack of intrinsic motivation, it has been designed in a way that is perhaps optimal for a pilot study but hardly for widespread real-world deployment. Moreover, like Fitocracy, Zamzee has lost the element of probabilistic reinforcement that creates an optimal reinforcement schedule.

Nike+ operates in much the same way as Zamzee, though it has no rewards or levels at all. It focuses on the social network and competitive aspects of play. Nike+ differs from Fitocracy and Zamzee in that where the latter are aimed at improving public health through gamification, Nike+ uses gamification not to improve public health, but to increase brand loyalty among the already healthy. Nike+ has been marketed to runners, with the goal of locking them into the iPod/Nike ecology. Nike+ has in fact been one of the few large corporations successful in using social networking to build brand loyalty, by using this approach (13). Although it is not at common purposes with Zamzee and Fitocracy, it is notable because it is one of (if not the) largest gamified health service, influencing all those that come into the field after it (i.e., Zamzee, Fitbit, Bodybugg, etc.) Moreover, it is relevant because of its emphasis on running: although it doesn’t seek to make people healthier, it does need its customers to continue running in order to build the brand loyal social network Nike seeks. As such, its bottom line goal is still compatible with the other programs mentioned: drive people to continue performing a particular fitness activity.

The resolution to these problems is not to scrap the approach of gamification, but rather, to extend it to account for the elements of the SCT model that are missing. The foremost necessity is to amend the lack of true incentives: “levels”, “points,” and so on cannot merely be metrics that take on the appearance of achievement and advancement. The most difficult approach would be to construct actual games: picture Diablo 2 experience-point gain linked to one’s Fitocracy leveling. In this case one would have a real game with all of the mechanisms of modern gaming built-in: social networks for observational learning and social outcome expectations, reciprocal determinism in the form of the game obstacles being faced, and probabilistic “item drop” reinforcement in-game coupled to the deterministic advance of levels with one’s exercise. Most importantly, leveling would now be tied to something positive and enjoyable: advancing in a fun game. As one advanced in levels and achievements, one would successfully replace the “level grinding” that already exists in the game, but all other elements of the game would remain identical. Therefore, all of the elements that actually created an addictive and successful video game would become linked to one’s fitness activities. This would be difficult to create wholesale, as creating long-lasting and successful video games is an extremely rare event, regularly achieved by only a tiny handful of dedicated game companies. On the other hand, licensing old and successful games that now exist in legacy but produce few sales for their parent company (i.e., Diablo 2, still engaging tens of thousands of players daily, a decade after release) could be a successful alternative, wherein the primary costs of development would become software engineering to link the in-game advancement-state to some health metric, and licensing costs. With such software engineering being a relatively simple feat (people create such programs as hobbies)(14), the only true stumbling block would be licensing fees. For games that produce little annual returns (i.e., Diablo 2 sales a decade after release) companies may be induced to ‘donate’ such licensing fees as a tax deduction or public relations maneuver. Player enrollment may even be increased if such health-game projects were released 6 months to 1 year prior to the release of a new sequel in the game series (i.e., the Diablo 2 Health game 18 months prior to the release of Diablo 3). As a form of marketing for both products, this may be even more attractive to the parent company.

After taking into account the need for true incentives, there would also be a need for probabilistic positive reinforcement to create the optimal reinforcement schedule. This could be coupled to in-game probabilistic elements as noted above, but it could also be instituted as an out-of-game affair. For instance, a program such as Fitocracy could partner with unrelated products such as the various simple in-browser games offered by Rovio and Zynga on facebook, or perhaps aim for small “indie” game publishers that are currently creating armies of new games via platforms such as the iTunes App Store and Steam (15). Zynga, for example, offers in-game purchases using real-world money. However, insofar as these in-game items actually have unlimited supply, some small fraction of them could be “donated” to a public health program. Coupled to something like Fitocracy, some class of items may be achieved upon “leveling up” (in a deterministic fashion), with some others having a random chance of occurrence upon completion of achievements and quests (introducing our probabilistic element). The game-external approach avoids the problem of games that become dated, as well as once more reintroducing the social elements, peer observational learning, and the other relevant elements of SCT. Small independent game publishers may even be influenced to involve health metrics in the game proper: for example, the iOS game “Dungeon Raid” has a modular system for creation of novel game character classes. There exists an endless number of “classes” to be expanded into. One could create a single “Health” class that syncs with something like one’s Fitbit walking data, allowing randomized “bonuses” to one’s play according to one’s daily steps walked, with bonuses of greater magnitude being accordingly rarer. These sorts of systems allow the introduction of probabilistic reinforcement mechanisms that current gamification attempts lack and, again, by working in the framework of successful games also include proper positive incentives.

The issue of reciprocal determinism is resolved once these mechanisms are tied to actual game mechanisms. Because the difficulty level now exists inside of a game itself, rather than in the gamified fitness regime, one can now choose the pace of one’s “leveling” to suit the game. If one does not want to press ahead with too much difficulty, they can loiter in easier parts of the game, just as they would traditionally, and level up slowly and surely. Of course they will inevitably wander into novel parts of the game, as players inevitably due, and the concomitant increase in difficulty will create the drive to continue “leveling” (exercising).
All of the various elements of the SCT model are already addressed in modern socially-integrated role-playing games, and as such, properly integrated a gamified mechanic with such games will ensure that the model’s assumptions are being met and the intervention may actually drive behavior change. The ideal situation is to engage with a game that already manipulates operant conditioning to create a Skinner Box effect, and hook the gamified fitness platform into that ecology. For instance, Massively Multiplayer Online Games (MMOGs) are already built around the idea of addicting players with a combination of deterministic positive reinforcement (leveling), probabilistic positive reinforcement (item drops), observational learning of peers (in-game social networks, integrated with out-of-game social networks), reciprocal determinism (in-game difficulty scaling by choice of zoning and leveling rate), self-efficacy gains through repeat successes against incrementally difficult obstacles, collective efficacy (some obstacles must be faced with one’s in-game network), and incentives are provided primarily through the opportunity for item drops (an opportunity to roll the die on the probabilistic reinforcement). These games often lack anything except these elements, arguably becoming not a game at all, and yet are wildly successful. Through the integration of health metrics into such games – perhaps as one of many character statistics tracked, opening up opportunities for a parallel line of optional character enhancements – one can make use of an already extent and extremely effective Skinner Box environment for addicting people into pursuing health metrics. Mechanisms for convincing corporations to include such an element may be to encourage its use as a shield against litigation: game makers are already being sued for the dangers of their addictive games (16); a defense of keeping an eye on players’ well-being and encouraging public health may be a small investment towards deflecting significant costs in litigation.
The current trend towards gamification of health initiatives is based predominantly on the Social-Cognitive Theory of behavior, but fails to take into account the element of reciprocal determinism, does not provide for probabilistic positive reinforcement to optimize reinforcement of target behaviors, and most crucially tends not to incorporate any real incentives – thus creating “positive reinforcement” with no positives. Gamification is lauded by marketing companies as using the adornment of games without the need for actual game-play, and yet in so doing misses out on what it is that actually causes behavioral repetition (17). The simplest method to address these short-comings is to look at extant games that have already successfully made use of these mechanisms, and target game developers and publishers for including health metrics in such games in a manner that does not interfere with core game mechanics. By preserving fun and making use of the full SCT model, health initiatives will be able to draw upon successful methods for driving user behavior, rather than simply the cargo cult equivalent.

References:
1. Ogden CL, Carroll MD, McDowell MA, Flegal K. Obesity among adults in the United States – no statistically significant change since 2003-2004. NCHS Data Brief No.1. 2007. http://www.cdc.gov/nchs/data/databriefs/db01.pdf Accessed April 23, 2012.
2. Centers for Disease Control and Prevention. National Center for Health Statistics. Prevalence of Overweight, Obesity, and Extreme Obesity Among Adults: United States, Trends 1960–1962 Through 2007–2008. http://www.cdc.gov/nchs/data/hestat/obesity_adult_07_08/obesity_adult_07_08.pdf Last Accessed April 23, 2012.
3. Centers for Disease Control and Prevention. National Center for Health Statistics. Crude and Age-Adjusted Percentage of Civilian, Noninstitutionalized Adults with Diagnosed Diabetes, United States, 1980–2010. Atlanta, GA: CDC. http://www.cdc.gov/diabetes/statistics/prev/national/figageadult.htm
4. McGuire MT, Wing RR, Klem ML, Hill JO (1999) Behavioral strategies of individuals who have maintained long-term weight losses. Obes Res 7: 334-341.
5. Obesity ‘not individuals’ fault. BBC News. October 17, 2007. http://news.bbc.co.uk/2/hi/health/7047244.stm Accessed April 23, 2012.
6. Chang BW, Christakis NA. Medical modeling of obesity: A transition from action to experience in a 20th century American medical textbook. Sociol Health Illn. 2002;24:151-177.
7. Ferdinando, M. Development of a multichannel gamification platform. Politesi. http://hdl.handle.net/10589/22961 . Last accessed April 23, 2012.
8. Ferster, C.B., & Skinner, B.F. (1957) Schedules of reinforcement. New York: Appleton-Century-Crofts.
9. Fitocracy. “What is this crazy Fitocracy thing?” NY: Fitocracy. http://www.fitocracy.com/about-us/ Last accessed April 23 2012.
10. Design Mind. “Gamification of Healthcare.” San Fransisco, CA: Frog Design. http://designmind.frogdesign.com/blog/the-gamification-of-healthcare.html. Last accessed April 23, 2012.
11. HopeLab. “Innovative Solutions: Zamzee.” Redwood City, CA: Omidyar Group. http://www.hopelab.org/innovative-solutions/zamzee/. Last accessed April 24, 2012.
12. Peters CS, Malesky LA. “Problematic Usage Among Highly-Engaged Players of Massively Multiplayer Online Role Playing Games.” CyberPsychology & Behavior. August 2008, 11(4):481-484.
13. Schogel M, Herhausen D. Comarketing Capability and its Impact On Marketing Alliance Performance (pp. 350-351). In: Marketing Theory & Applications. Chicago, IL: American Marketing Association, 2010.
14. System Softlab. “Artmoney.” Moscow, Russia: System Softlab. http://www.artmoney.ru/ Last accessed April 25, 2012.
15. David Michael. Indie Game Development Survival Guide (Game Development Series). Charles River Media, August 2003.
16. Kravets, David. Addicted Gamer Sues Game-Maker, Says ‘Unable to Function’. San Fransisco, CA: Wired Magazine. http://www.wired.com/threatlevel/2010/08/lineage11-addiction/ Last accessed April 26, 2012.
17. Bunchball, Inc. Gamification 101: An Introduction to the Use of Game Dynamics to Influence Behavior. October 2010. http://www.bunchball.com/sites/default/files/downloads/gamification101.pdf . Last accessed April 26, 2012.

Labels: , ,

0 Comments:

Post a Comment

Subscribe to Post Comments [Atom]

<< Home