Physical activity after commitment lotteries: examining long-term results in a cluster randomized trial

The World Health Organization has identified physical inactivity as the fourth leading risk factor for global mortality, accounting for 3.2 million deaths annually (Forouzanfar et al., 2016). While many people know that regular physical activity (PA) is beneficial for their health and can contribute to weight management (Hildebrandt et al., 2007), most Americans (79%) and Europeans (66%) do not meet recommended levels of PA (CDC, 2014; Lee et al., 2012; Eurobarometer, 2014). Besides, overweight and obese individuals generally exercise less than normal-weight individuals (CBS & RIVM, 2014) and the prevalence of obesity has more than doubled since 1980 (WHO, 2015): approximately 70% of Americans and 62% of Europeans are currently overweight or obese. Consequently, research in the field of health promotion has yielded numerous effective ways for people to change their health behaviors (Hoeymans et al., 2014). Still, most short-term interventions only have short-term effects (Van den Berg & Schoemaker, 2010). Although many people intend to improve their health by exercising on a regular basis for longer periods, the majority fails to follow through (Kooreman & Prast, 2010).

The progressing field of behavioral economics has identified systematic and predictable decision patterns that can explain why people make decisions that deviate from their own long-term health goals. Failures of self-control have been associated with present bias: the human tendency to disproportionally overweigh costs and benefits that are immediate (e.g., exercising vs. relaxing) over those that are delayed (e.g., good health in the future) (Ainslie, 1975; Laibson, 1997). Consequently, people often intend to exercise on a regular basis, but eventually attend their gym less frequently than they had planned to (DellaVigna & Malmendier, 2006; Schumacher et al., 2017). In the present work, we aimed to test whether some of the same decision-biases that contribute to unhealthy behaviors can be used to durably assist individuals who have trouble sticking to their PA goals.

There is increasing behavioral economic evidence for the notion that people who foresee their self-control troubles can benefit from -and are willing to use- interventions that are known as commitment devices (Bryan et al., 2010). Commitment devices are defined as voluntarily imposed arrangements that restrict future behavior to avoid temptation (Rogers et al., 2014). For example, people cut up their credit cards or embrace withdrawal penalties on their savings account to avoid overspending and undersaving in the future (Beshears et al., 2015; Bryan et al., 2010) or request nearby deadlines for their work to preempt procrastination (Ariely & Wertenbroch, 2002). Likewise, Dutch employees are happy with mandatory pension savings because they fear that otherwise they would not save enough for retirement (Van Rooij et al., 2007).

Psychologically grounded commitment devices have also been tested as a tool to support health behavior change (Rogers et al., 2014). Based on the principle that humans dislike losses more than they like gains of equal size (Kahneman & Tversky, 1979), individuals have agreed to forfeit their monetary deposits at voluntarily imposed deadlines if they do not quit smoking or fail to stick to their diet (Giné et al., 2010; Halpern et al., 2012; Volpp et al., 2008). Recently, lottery-based commitment devices have also been demonstrated to support weight loss, medication adherence and walking (Kimmel et al., 2012; Patel et al., 2016a, b; Volpp et al., 2008). The lotteries aimed to tap into the human tendency to avoid regret (Christy et al., 2016; Ferrer et al., 2012; Zeelenberg & Pieters, 2007) by only awarding prizes to lottery winners who attained their health goals and informing unsuccessful lottery winners on their forgone prizes.

As we presented in previous work (Van der Swaluw et al., 2018), in the current sample, 13 weekly lotteries (short-term lottery arm) supported regular gym attendance. After the 13 weekly lotteries, participants in our trial attended their gym considerably less. Adding an additional long-term lottery deadline after 26 weeks (long-term lottery arm) partly averted the decline in gym attendance, indicating that a long-term lottery deadline can help sustain regular gym attendance up to 13 additional weeks. The adjusted probability of goal attainment (week-gym attendance ≥ 2) in weeks 1–13 was 57% in the short-term lottery arm, 66% in the long-term lottery arm and 25% in the control arm. In both lottery arms, 8 out of the 13 weekly winners were eligible for their prize. Between weeks 14–26, the probability of week-goal attainment was highest in the long-term lottery arm (51%) versus 23% in the other arms. The majority of participants (55%) in the long-term lottery arm was eligible to claim the prize in week 26 if they would have won it.

For the present paper, we analyzed gym attendance after all lotteries ended.

Although there is evidence that commitment devices (e.g., voluntary imposed deadlines with consequences) in the health domain are effective in the short-run, the long-term effects are often unsatisfactory (John et al., 2011; Patel et al., 2016a, b; Royer et al., 2012). Consequently, one of the key challenges in the application of commitment devices is either safeguarding maintenance of behavior change after an intervention or continuing its application (Halpern et al., 2012; Rogers et al., 2014). We focused on the former and with this purpose we examined individual goal attainment (week-gym attendance ≥ 2) after the interventions.

After all deadlines had passed, gym attendance was monitored up to week 52 in all arms to examine maintenance of behavior change. Specifically, we questioned whether goal attainment differed between the long-term lottery arm, the short-term lottery arm and the control arm between weeks 27 and 52. We hypothesized that goal attainment in the lottery arms would decline between weeks 27 and 52, but still be significantly higher than goal attainment in the control arm and would be highest in the long-term lottery arm (Van der Swaluw et al., 2016).

We additionally studied weight patterns over the course of the trial. Next to –among others– genetic, sociocultural, economic and environmental factors, modifiable behaviors (diet and PA) markedly contribute to the development of overweight and obesity (Bray et al., 2016). As such, regular PA can contribute to weight loss and weight management (Fogelholm, 2010; Fogelholm & Kukkonen-Harjula, 2000). Furthermore, overweight has been associated with self-control problems (Ikeda et al., 2010; Schlam et al., 2013), while commitment lotteries aim to combat this. Hence, we also explored the effect of the interventions on the development of weight over the 52 weeks of this trial.

The 52-week follow-up analyses of our cluster randomized trial show moderately sustained levels of goal attainment (gym week-attendance ≥ 2) six months after completing a 26-week lottery intervention. Up to one year after the start of the intervention, participants who entered 13 weekly commitment lotteries (weeks 1–13), followed by an additional lottery 13 weeks later (weeks 14–26) were more likely to attend their gym twice per week than participants in the control arm. In the same follow-up period, goal attainment in the short-term lottery arm (13 weekly commitment lotteries) did not differ from the control arm or long-term lottery arm. Weight declined slightly in the first 13 weeks of the trial in all arms and remained stable from there on.

The present study may contribute to the pursuit of methods for sustainable behavior change. Commitment lotteries have been used relatively sporadically in the field of health promotion. Starting a decade ago, Volpp et al. (2008) have offered lotteries to overweight participants to support their weight loss attempts. While their lotteries were effective for 16 weeks, participants regained weight after the intervention. Likewise, Patel et al. (2016a, b) have effectively used team-based lotteries to stimulate walking for 13 weeks, but differences between intervention and control deteriorated during the 26-week follow-up period. Although there is evidence for their short-term effectiveness, the longevity of commitment lotteries is unsatisfactory. In the present trial, we observe a significantly higher likelihood of goal attainment in the long-term lottery arm than control in the six months after the intervention. The observation that the long-term lottery-arm does not underperform the control arm during follow-up implies that the 26-week intervention has had a net effect on gym attendance over a 1-year period.

Still, in the interpretation of our results, some nuance is warranted. In Fig. 2 it can be observed that levels of goal attainment in the long-term lottery arm dropped after the 26-week lottery deadline. Accordingly, average week-goal attainment in the long-term lottery arm halved during follow-up, relative to weeks 14–26. Therefore, it cannot be concluded that participants in the long-term lottery arm remained equivalently committed to their gym attendance goal as before. Similar to previous studies, behavior change is shown to be challenging to maintain after commitment lotteries end.

Nonetheless, the tested combination of short-term lotteries and a long-term lottery in this trial provides novel and useful insights that can be built upon in future trials to help optimize the long-term effectiveness of commitment lotteries. For example, for the winner to be eligible for the long-term lottery prize, he or she had to attain at least 9 week-goals between weeks 14–26. The majority of participants in the long-term lottery arm did so. Therefore, future studies can experiment by lengthening the long-term lottery deadline (e.g., to 36 weeks (John et al., (2011) to test if this will further stimulate sustained goal attainment. This exploration may result in a deadline that is nearby enough to be salient in the present, while being lengthy enough to promote sustained behavior change after the deadline.

Similar to previous research, behavioral economic and psychological insights on decision-making could be well incorporated in health promotion (Loewenstein et al., 2012). First, to leverage present bias (the overweighing of the present), we imposed nearby deadlines to draw the consequences of procrastination nearer (Ariely & Wertenbroch, 2002). Second, because people tend to overestimate small probabilities, a lottery is an effective and scalable tool to make missing the deadlines potentially costly for participants (Kahneman & Tversky, 1979). Third, we leveraged the human tendency to avoid regret by drawing the lottery prize from all participants and informing non-eligible winners on their forgone prize (Zeelenberg & Pieters, 2004, 2007). Fourth, while we aimed to stimulate long-term behavior change, we used week-goals to fit individuals’ impatience and to facilitate the human tendency of using of temporal landmarks (e.g., Mondays) to relegate misfortune to the past and have a fresh start (Dai et al., 2014; O’Donoghue & Rabin, 1999).

The results of this trial also offer several practical insights for health professionals, policy makers, insurers or employers who aim to support people in achieving their health goals. Next to its effectiveness, the costs of prevention are considered a key aspect in the determination of its value to businesses and society (Van den Berg & Schoemaker, 2010). The use of psychological and behavioral economic insights in the design of the lotteries can enhance the psychological impact of money that is spent and could offer the opportunity to implement commitment lotteries at low costs. To stimulate regular gym attendance up to 52 weeks in the long-term lottery arm, we spent only €2.21 per participant per week (awarded prizes ÷ participants ÷ weeks). In perspective; this is 0.6% of the Dutch minimum week-wage (€361.25). The costs in future applications may be further reduced if participants are also willing to pay for their wish to commit, similar to self-imposed withdrawal penalties on savings accounts (Beshears et al., 2015) or betting one’s money on personal health goals with a deposit contract (Halpern et al., 2012). It would be valuable to explore the characteristics of participants and potential organizers of commitment lotteries that contribute to the optimal balance between attractiveness and effectiveness.

Despite their potential, commitment devices remain underused (Halpern et al., 2012; Rogers et al., 2014)- a fact that might be explained by the range of open issues on their implementation. For example, employees or patients outside the interventions may resent others receiving lottery prizes (Loewenstein et al., 2012) (which may also be circumvented by asking participants to pay for lottery tickets). Besides, commitment lotteries were especially effective while they were active and -analogous to drugs- lost most of their effectiveness once people stopped ‘taking them’. This finding stimulates thinking about the type of commitment device that maintains its impact over repeated (and possibly infinite) application. While goal attainment declined over time in our trial (see Fig. 2), the decline of goal attainment over time did not differ significantly between the lottery arms and the control arm. Therefore, it would be interesting to design and test similar commitment lotteries that, like many commercial lotteries, are endlessly repeated and accessible.

Another issue surrounding the applicability of commitment lotteries is their target population. The majority of our sample was male (69.3%), which was similar to the general population of the six gyms (69.8% male). Nonetheless, future studies would benefit from enrolling an even proportion of females and males. A meta-analysis by Haff et al. (2015) that compared the effectiveness of several commitment lotteries between demographic groups found no gender differences, nor differences in education or broad ranges of income (except a slight reduced effect with incomes > $87.500). In previous analyses, we also observed no differences in intervention-effectiveness between income- or education categories (Van der Swaluw et al., 2018). This may be an indication of commitment lotteries being broadly applicable. A next step in the development of commitment devices could be the assessment of design features that contribute to their efficacy and feasibility across different populations. Hence, although there is increasing evidence for the effectiveness of commitment lotteries, more applied research could enhance their wide-spread practicability.

Commitment lotteries aimed at promoting regular PA did not result in substantial weight loss. After a moderate decline in weeks 1–13, weight remained stable over time in all arms, which is contrary to projections of overweight and obesity progressing over time (Wang et al., 2008). Considering that medical complications and associated health care costs rise progressively as BMI increases (Cawley & Meyerhoefer, 2012), stable weight is not all bad. Nonetheless, no meaningful weight loss was achieved as a result of increased PA, meaning that most participants remained exposed to the increased risks of cardiovascular diseases, diabetes type 2 and cancers that are associated with overweight and obesity (Forouzanfar et al., 2016; WHO, 2013). Our findings are in line with the general conception that regular exercise can contribute to weight loss but often not solely (Fogelholm, 2010; Fogelholm & Kukkonen-Harjula, 2000). In some occasions, increases in PA have also been found to result in weight gain by calorie compensation (McCaig et al., 2016).

To achieve significant and sustained weight loss, regular PA is only one part of the multicomponent (lifestyle training, nutrition and PA) interventions that are acknowledged to be effective (Bray et al., 2016). Still, weight loss interventions that include an exercise component are often more effective than interventions that do not (de Roon et al., 2017; Jeffery et al., 2003). Given that overweight individuals generally exercise less than normal weight individuals (CBS & RIVM, 2014), any multicomponent weight loss intervention that also encompasses PA might still benefit from including commitment lotteries for PA.

This trial is subject to several limitations. First, while it is clear that commitment lotteries did not promote meaningful weight loss, it remains unclear why this was the case in this trial (e.g., due to the type of exercise, calorie compensation, loss of fat that may have been compensated by increased muscle mass). We chose to focus on gym attendance as our main dependent variable, rather than weight loss, because this is under more direct volitional control by the participants. In future instances, it may be valuable to explore the need for additional weight loss guidance when participants enter commitment lotteries for PA. A second limitation is that our trial limited the promotion of regular PA to gym attendance, while activities with lower intensity (e.g., walking and recreational cycling) are also known to promote health outcomes (Fogelholm, 2010). The benefit of our approach is the novel and scalable context in which commitment lotteries were shown to be effective (company gyms), while exercise was supervised by the gym personnel. Third, our trial included no more than six gyms and randomization at the gym level increased the potential influence of intra gym effects in the effectiveness of the interventions. The sensitivity analysis (where we excluded the best performing control-gym from the model) reduced the sustained effect of the long-term lottery to non-significance. We accounted for intra-gym influences in our multi-level analyses and randomization at the gym level had the benefit that treatment contamination was minimized. Still, future studies can potentially avoid these issues by enrolling more gyms.

Regular PA has numerous health benefits. While many people aim to exercise on a regular basis, there are multiple predictable behavioral patterns that hamper the progression of a health-intention to sustained behavior. Effective and scalable support of long-term regular exercise is needed. Commitment lotteries that were designed to leverage psychological knowledge on decision-making can help cope with the challenges of health behavior change at low costs. While the 26-week intervention supported regular PA up to 52 weeks, levels of gym attendance declined after all deadlines had passed. Participants did not remain equivalently committed as before the long-term lottery deadline. Future research in broader populations could reveal to what degree commitment lotteries remain effective over longer time periods or if they are endlessly accessible.