Do incentives crowd out motivation? A feasibility study of a community vector-control campaign in Peru

Abstract

Incentives are a useful tool in encouraging healthy behavior as part of public health initiatives. However, there remains concern about motivation crowd out—a decline in levels of motivation to undertake a behavior to below baseline levels after incentives have been removed—and few public health studies have assessed for motivation crowd out. Here, we assess the feasibility of identifying motivation crowd out following a lottery to promote participation in a Chagas disease vector control campaign. We look for evidence of crowd out in subsequent participation in the same behavior, a related behavior, and an unrelated behavior. We identified potential motivation crowd-out for the same behavior, but not for related behavior or unrelated behaviors after lottery incentives are removed. Despite some limitations, we conclude that motivation crowd out is feasible to assess in large-scale trials of incentives.

Introduction

Offering financial incentives to motivate health-related behaviors has become popular over the past decade as an effective strategy to improve health and health care. ^1–3 A growing body of evidence supports the use of incentives in many forms — including lotteries, discounts, contingent rewards, bonuses, and voucher programs — to change patient, provider, and consumer behavior. ^4,5 This has been demonstrated in both high-income ^4,6–11and in low- and middle-income countries (LMICs) ^1,2,12–14 for a wide range of health-related behaviors. As incentive-based programs have been trialed and scaled, and evidence has accrued pointing to the potential for incentives to change behavior, stakeholders have raised questions about what happens once incentives are removed. Specifically, policy makers and funders have had concerns about the potential for incentives to “crowd out” intrinsic motivation. ^15,16 Crowding out occurs when extrinsic rewards such as incentives displace intrinsic motivation to the point where adherence to or performance of the target behavior declines to below-baseline levels after incentives are removed. ^17–19

Previous studies of motivation crowd out have been inconclusive ²⁰. Two meta-analyses of laboratory studies of crowd out reach different conclusions about whether incentives decrease intrinsic motivation ¹⁹ or do not. ²¹ The finding of no motivation crowd out is consistent with neoclassical economic theory, which suggests that incentives change behavior by simply shifting the costs and benefits attached to engaging or avoiding the particular behavior^3,22. In this model, motivation levels (and observed behaviors) should return to baseline levels after incentives are removed. The presence of motivation crowd-out can be explained by behavioral economics, a subfield of economics that pays careful attention to the cognitive, psychological, and decision-making factors underlying behaviors, including time-inconsistent preferences and the importance of “choice architecture” of the decision environment. ^23–25 Insights from behavioral insights suggest that incentives design makes a difference (lotteries vs. fixed payments, for example) ^14,26,27 and that incentives may be particularly effective in shifting behaviors short term en route to building longer-term habits and achieving persistent behavior change.

In the context of mixed evidence, concerns about motivation crowd-out may dampen enthusiasm for deploying incentives in clinical settings and public health programs. The possibility of motivation crowd-out adds uncertainty to risk-benefit calculations for incentive-based interventions that may also be deemed politically problematic, unethical, or inefficient Unfortunately, few pragmatic trials of incentive-based interventions to encourage healthy behaviors evaluate motivation crowd-out. This study aims to address that gap by demonstrating the feasibility of including motivation crowd-out analyses in evaluations of incentive programs. Following a pragmatic trial of behaviorally informed interventions designed to increase community members’ willingness to participate in an indoor residual spray campaign to prevent Chagas disease in Arequipa, Peru, we evaluated whether community members who had received lottery incentives in the original campaign were less likely to participate in the same behavior (the second round of the same phase of the campaign), a related behavior (participation in a later phase of the campaign), and an unrelated behavior (participation in mass dog vaccination campaigns or MDVC for canine rabies).

Methods

Parent study and motivation crowd-out hypotheses

Our assessment of the feasibility of motivation crowd-out was done in the context of a parent study of household participation in a Chagas disease vector control campaign in Arequipa, Peru. The campaign, the intervention designs, and the main trial results are described in detail elsewhere. ^28–31 Briefly, the existing campaign served as the usual care control condition in our trial. In addition, three behaviorally-informed interventions were designed and trialed alongside the usual campaign. Ethical approval for the parent study for the parent study was obtained from the Universidad Peruana Cayetano Heredia, the University of Pennsylvania, and Tulane University.

One of the three interventions tested in the parent study was a lottery incentive, in which households could win a prize contingent on participation in the campaign. During the design phase for this incentive, we wondered, in partnership with our Ministry of Health colleagues responsible for the ongoing campaign, if being offered incentives (in the form of a lottery) to participate in the insecticide spraying phase of the Chagas disease campaign would lead to lower willingness to participate in subsequent public health activities. We were also interested in testing whether offering incentives for one type of behavior might affect a continuum of other “related” subsequent behaviors, based on the hypothesis that incentives might crowd out motivation to undertake a very similar behavior but would not affect broader participation in any public health campaign.

Our feasibility study evaluated four hypotheses: First, we hypothesized that it would be feasible to conduct an analysis of motivation crowd-out in the context of a pragmatic trial. We evaluated this hypothesis through conducting the motivation crowd-out analysis described below. Second, we hypothesized that participation rates among community members randomly assigned to receive incentives to participate in the first phase of the Chagas disease campaign (by allowing their homes to be sprayed with insecticide) would not decline for subsequent public health activities when no incentives were offered. We evaluated this hypothesis by analyzing participation rates for the same behavior (a second round of spraying in the same phase of the Chagas campaign), a related behavior (participation in post-spray vector surveillance in a later phase of the campaign), and an unrelated behavior (participation in MDVC for canine rabies).

Study populations and sites

This study examines Chagas disease and canine rabies prevention campaign outcomes from two districts in Arequipa: Cerro Colorado (CC), and Alto Selva Alegre (ASA). In both districts, eligible households were approached by campaign staff and encouraged to participate in the campaigns by either permitting insecticide spraying in the home and subsequent search for presence of insects (Chagas disease campaign) or by vaccinating owned dogs (canine rabies campaign). See ²⁹ and ³² for more details.

Randomization and study sample

In Cerro Colorado, 1,610 households were randomly assigned to either a lottery intervention or the usual campaign during the first spray cycle of the attack phase of the campaign (figure 1). In ASA, 4,974 households were randomly assigned to the lottery intervention (1,215), the usual campaign (1,280), or one of the other two behavioral interventions (2,447). Houses where an adult was not present to consent to the insecticide application after multiple attempts were observed as “closed” during each intervention.

Figure 1.

Map of Cerro Colorado clusters assigned to treatment and control intervention strategies during the first indoor residual spray (IRS) campaign.

Incentives intervention

The lottery intervention was a contingent group regret lottery, as follows: Households randomized to the lottery intervention in the Chagas disease campaign were assigned to lottery groups of 5–9 contiguous houses along a block face. Each group was randomly assigned one number (from 0–36) and informed that the lottery drawing for the Chagas campaign would be linked to a specific drawing of the national lottery about 2–6 weeks in the future. For example, a lottery group of 7 households might be told on February 20th that they had been assigned the number 17, and that their lottery drawing would be linked to the national lottery drawing on March 28th. If the group’s number was later drawn in the national lottery, households whose entire neighborhood participated in the campaign (agreeing to have their homes sprayed with insecticide) received a voucher redeemable at a local hardware store worth PEN 100 (approximately USD 30). If only the individual household participated, they received a voucher for PEN 50 (approximately USD 15). Receiving the lottery prize was therefore contingent on having participated in the campaign. This design also incorporates regret aversion, because households could have their lottery number drawn even if they had not participated in the campaign. The anticipation of this regret has been shown in prior lottery designs to be highly motivating. ^33,34

The lottery incentive was hypothesized to increase participation by increasing the benefit to the household for participating and reducing the salience of perceived costs (time investments, hassle factors, inconvenience of moving furniture and leaving the house for several hours). The group participation incentive, the regret aversion provision, and the use of lotteries in general instead of a fixed payment were all design features informed by behavioral economics.

Outcome measurement

We evaluated the potential for motivation crowd out following an offer of lottery incentives for three behaviors: the same behavior for which lottery incentives were offered, a related behavior, and an unrelated behavior. The lottery incentives were offered for participation in the first (CC) or second (ASA) cycle of the spray phase of the Chagas disease vector control campaign in Arequipa. To measure the same behavior, we observed participation in the second cycle of the insecticide spraying in CC as recorded in the campaign records collected by our team in partnership with the Ministry of Health campaign. During this second cycle of insecticide spraying, every household was approached by one of two mobilizer teams, a group of public health professionals from the Peruvian Ministry of Health that explained the general procedure of insecticide spraying. To measure a related behavior, we observed participation in the surveillance phase of the Chagas disease campaign in ASA, as recorded in campaign records. For the unrelated behavior, we observed participation in the MDVC against rabies in ASA as recorded in that campaign’s records.

Statistical analysis

We used the following models, performing all two-sided statistical tests in R ³⁵ with a significance level of 0.05:

Same behavior

Using a generalized linear model via the GEE function of R’s geepack, we estimated the log odds of participating in the second cycle of the spray campaign as a function of receiving a lottery offer (vs. control) in the first spray cycle, controlling for other known predictors of participation including the infestation or participation status of the household in the first spray cycle (insect-infested, not insect-infested, or refused to participate), and mobilizer team assigned to the household.

Related behavior

Using a generalized linear model via the GEE function of R’s geepack and entomological surveillance data from ASA approximately two years after the second spray cycle, we estimated the log odds of participation in the surveillance phase of the campaign as a function of receiving a lottery offer (vs. control) in the second cycle of spray phase of the campaign. We also controlled for household participation/refusal during the second cycle of the spray campaign.

Unrelated behavior

We adapted the proportional odds logistic model we developed in order to evaluate participation in the MDVC ³² to assess the impact of being offered a lottery incentive for the second cycle of the spray phase of the Chagas campaign on participation in the MDVC. The adapted model used the R package MASS ³⁶ to estimate the proportional log odds of MDVC participation as a function of being offered a lottery (vs. control). We also included controls for known predictors of MDVC participation, including Euclidean distance to the closest vaccination point, having a dog leash at home, number of children under five years old at home, time living in the area, and rabies outbreak status of prior residence. MDVC participation was operationalized as an ordinal outcome: no participation (no dog vaccinated in the house), partial participation (some, but not all dogs in the house, vaccinated), and full participation in the MDVC (all dogs in the house vaccinated).

The models for same and related behavior were hierarchical to account for spatial autocorrelation within locality. This was not possible for the unrelated behavior because there was at most one vaccination point in each locality and the variable distance from the house to the vaccination point was unidentifiable under a hierarchical model.

Results

We observed participation results for 1,560 CC houses in the same behavior analysis, 959 ASA houses in the related behavior analysis and 1,078 ASA houses in the unrelated behavior analysis. Participation rates for these three behaviors by exposure to lottery interventions are shown in table 1.

Table 1:

Household actions in Cerro Colorado and Alto Selva Alegre in response to the same behavior, related behavior, and unrelated behavior.

District	Cerro Colorado		Alto Selva Alegre
Behavior	Same behavior		Related behavior		Unrelated behavior
N	Control (N=759)	Lottery (N = 801)	Control (N=493)	Lottery (N=466)	Control (N=565)	Lottery (N=513)
% Participated	87.2	78.8	44.8	45.5	40.0	38.8
% Not Participated	12.8	21.3	55.2	54.5	60.0	61.2
% Closed	5.7	8.4	43.2	45.7	N/A	N/A
% Refusals	7.1	12.9	12.0	8.8	N/A	N/A

Same behavior: participation in spray phase of Chagas campaign

CC households that received the lottery intervention during the first cycle of the spray phase of the Chagas disease vector control campaign had 40.9% lower odds of later participating in the same behavior (the second cycle of household spraying) once incentives were removed (odds ratio (OR) = 0.59, CI = 0.44–0.79, p <.01) (table 2), after controlling for first-cycle participation (table 2). The odds of participating in the second cycle of spray was 57.1% higher in houses visited by one of the mobilizer teams compared to the other (OR = 1.57, CI = 1.17–2.10, p = 0.03), after controlling for other covariates. Infested houses that participated in the first cycle of spray had 3.15 times higher odds of participating in the second spraying (OR = 3.15, CI = 1.28– 7.77, p = 0.02), after controlling for other variables. Uninfested houses that participated in the first cycle of spray had 3.38 times higher odds of participating in the second spraying (OR = 3.38, CI = 2.16–5.28, p < 0.001), after controlling for other variables. Houses that did not participate in the first cycle of the spray had 64% lower odds of participating in the second spray (OR = 0.36, CI = 0.21–0.61, p < 0.001), after adjusting for other variables (table 2).

Table 2:

The effect of a lottery intervention to increase participation in an indoor residual spraying (IRS) campaign on participation in a subsequent IRS campaign. Estimated exponentiated coefficients with a logistic regression model.

	OR1	95% confidence interval	p-value
Lottery group = 1	0.59	(0.44, 0.79)	0.00
Mobilizer team A (Ref. group = team B)	1.57	(1.17, 2.10)	0.03
Status in first spray cycle (Ref.=Household did not answer door)	ref
Infested, participated	3.15	(1.28, 7.77)	0.02
Uninfested, participated	3.38	(2.16, 5.28)	0.00
Refused to participate	0.36	(0.21, 0.61)	0.00

Related behavior: Chagas surveillance

We did not find a significant difference in the odds of participating in a related public health behavior—the Chagas disease surveillance campaign—by lottery incentive exposure status in the second cycle of the spray phase of the campaign among ASA households (OR = 0.99, CI = 0.76–1.28, p = 0.95) (table 3). We found that the odds of participating in the surveillance campaign if the house participated in the second spray campaign was 70% higher compared to houses that did not participate in the second spray campaign (OR = 1.70, CI = 1.05–2.74, p = 0.01). We did not find a significant difference in the odds of participation in the surveillance campaign whether the house participated or was closed during the previous second spray campaign (OR = 1.26, CI = 0.66–2.41, p = 0.56) (table 3).

Table 3:

The effect of a lottery intervention to increase participation in an indoor residual spraying (IRS) campaign on participation in a door-to-door surveillance campaign for Chagas Disease vectors. Estimated exponentiated coefficients of a logistic regression model.

	OR1	95% confidence interval	p-value
Lottery Group	0.988	(0.763, 1.279)	0.95
Closed in previous second spray campaign	ref
Sprayed in previous second spray campaign	1.698	(1.052, 2.740)	0.01
Refused to participate in previous second spray campaign	1.256	(0.656, 2.406)	0.56

Unrelated behavior: mass dog rabies vaccination campaign

Lottery exposure in the second cycle of the spray phase of the Chagas disease IRS campaign was also not associated with participating in an unrelated public health behavior-- vaccinating owned dogs during the MDVC (OR = 0.92, CI = 0.67–1.26, p = 0.61) (table 4). As expected, some adjusting variables were associated with vaccinating owned dogs during the MDVC: having a dog leash at home (OR = 1.37; 95% CI = 1.01, 1.84; p = 0.04) and time living in the study area (OR = 0.99; 95% CI = 0.97, 1.00; p <.01). Having children under 5 years of age and rabies status of the previous area of residence were not associated with vaccinating dogs even though they were correlated in previous analyses.³²

Table 4:

The effect of a lottery intervention to increase participation in an indoor residual spraying (IRS) campaign on participation in a dog rabies campaign. Estimated with a proportional odds logistic regression model.

	OR1	95% confidence interval	p-value
Lottery Group	0.920	(0.671, 1.262)	0.61
Distance to campaign	0.998	(0.996, 1.000)	0.09
Having a leash at home	1.365	(1.013, 1.840)	0.04
Previous area of residence
Endemic for rabies	ref

Discussion

In this study, we assessed the feasibility of evaluating motivation crowd-out associated with an incentive-based intervention in a pragmatic context. Contrary to our hypothesis of no or minimal motivation crowd-out, we found some evidence of motivation crowd out for the same behavior as the originally incentivized behavior, participation in a Chagas disease vector control campaign in southern Peru. We found that participation in the second cycle of the household insecticide spray phase of the Chagas disease campaign was significantly lower among households that had been exposed to the lottery incentives in the first cycle compared to households that had not been incentivized in the first cycle.

While not a strict test of motivation crowd out (which would require a baseline measure of participation prior to incentives being offered, not collected in the parent study), the difference in second-cycle participation rates by first-cycle lottery exposure is stark and is consistent with decreased motivation to participate once incentives are removed. This result was particularly surprising given that the lottery incentive did not appear to boost participation (relative to other interventions including the usual campaign promotion) in the parent trial. ³¹ Importantly, we found no evidence of motivation crowd-out for a related behavior (participation in the surveillance phase) or an unrelated behavior (dog vaccination in a rabies campaign) among those exposed to lottery incentives during the spray phase of the Chagas campaign.

Our mixed results are consistent with conflicting conclusions in two different meta-analyses of motivation crowd-out from the psychology literature, ^21,37 and may reinforce concerns in the public health and global health communities about the potential for incentives to crowd out intrinsic motivations to contribute to public health efforts, including altruism and civic engagement. While much of the concern about incentives has been raised in the context of wealthy country healthcare systems, motivation crowd-out may be particularly relevant and detrimental in LMICs, where intrinsic motivation may play a relatively larger role in decision-making in the face of fewer healthcare resources. ³⁸ Our results are therefore a useful contribution to the sparse literature on motivation crowd out in LMICs.

Our finding that a lottery intervention crowded out motivation to participate in the same behavior when incentives were removed, but not in related behaviors or unrelated behaviors, is instructive in terms of the types of behaviors for which we might expect motivation crowd-out. Several previous studies of incentives for health-related behaviors reported that incentives either had no effect on behaviors or actually increased motivation once incentives were removed. ^9,39For example, students who were given financial incentives to attend the gym exercised more often and continued to attend the gym more frequently than controls after the intervention had concluded. ⁴⁰ Children who were given stickers when they ate vegetables reported liking vegetables more even when the study was over. When healthy foods at a university were subsidized during a study (a form of incentive), the rate of choosing healthy alternatives increased during the study and further increased after the subsidy was removed. ⁴¹ A smoking cessation program reported higher abstinence rates 15 months after incentives were removed compared to those that hadn’t received incentives. ⁴² These studies suggest that incentives that encourage habit formation around repeated behaviors may enable persistence beyond the removal of incentives. The behaviors incentivized in this study were one-off or infrequent behaviors that may be more susceptible to motivation crowd out, particularly when an identical behavior is being prompted without the previously offered incentives.

We can look to the behavioral economics and psychology literature for additional insights about our results. It is possible that the lottery incentive signaled to study participants that they were being compensated because the insecticide was dangerous or had substantial side effects, essentially framing the lottery as a risk payment. Previous studies have demonstrated the salience of this frame. For example, Cryder et al. had participants read a flyer about a fictitious study that would reimburse either $25 or $1000 for undergoing transcranial magnetic stimulation, a new technology that carries some risks for healthy subjects. Participants offered $1000 for participation perceived the procedure as more risky than those offered $25 as compensation. ⁴³ If our participants viewed the lottery as remuneration for risk, a diminished trust in the safety of insecticide could have led to lower participation during the second cycle of spraying without changing participation in the related and unrelated campaign behaviors.

Self-determination theory has also been involved to explain varied results of motivation crowd out analyses. Jacobsen (2014) suggests that the extent of motivation crowd out may depend on whether the incentive is perceived as “controlling” (diminishing the ability of individuals to make decisions about their own actions) or “supportive” (encouraging of people to change their own actions). If a person perceives the incentive as controlling, they may substitute extrinsic for intrinsic motivation, thus leading to crowd out. An incentive that is perceived as supportive is more likely to be integrated into one’s sense of self and less likely to lead to motivation crowd in. ^44,45 Our results suggest that the lottery incentives for the spray phase of the Chagas disease vector control campaign may have been viewed as controlling rather than supportive.

Another explanation for observed motivation crowd-out only in the same behavior is higher baseline levels of intrinsic motivation surrounding the insecticide spraying for Chagas disease prevention compared to ongoing surveillance activities or dog vaccination. Previous work has demonstrated that higher levels of intrinsic motivation are more susceptible to crowd out. ⁴⁵ While intrinsic motivation has traditionally been understood as motivation that originated from within rather than from an outside source, social psychologists have recently updated that definition to encompass motivations that are congruent with a set of goals, values, and preferences. ⁴⁶ The intensity and duration of the spray phase of the Chagas campaign (campaign teams typically spend up to a year in a single district and contact households multiple times to motivate participation) may have shaped community members’ goals and values over time, leading to higher levels of intrinsic motivation--and therefore more crowd out after incentives were offered and then withdrawn. By contrast, the surveillance phase of the Chagas campaign and the MDVC are less intensive, and in the case of the MDVC, much shorter. Cerasoli also reports in the meta-analysis that directly salient incentives, or those that directly reward a specific change in behavior, like bonuses for productivity, are more likely to crowd out intrinsic motivation than indirectly salient incentives, like changes in base salary based on performance. ⁴⁶ The fact that our lottery incentive directly rewarded a specific behavioral change could be another explanation for the finding that it crowded out motivation in the same behavior.

Our assessment of the feasibility of studying motivation crowd-out in this context uncovers several limitations to our analysis. Some key limitations are related to the challenges of running large, pragmatic field trials of behavioral interventions. Several circumstances beyond our control led to changes in our trial protocol, including the fact that the main trial was conducted in ASA during the second cycle of spray, precluding a better-powered analysis of the same behavior. When we selected MDVC as the target unrelated behavior, we did not anticipate an active canine rabies outbreak in Arequipa, which certainly changed the importance and salience of dog vaccination compared to an annual campaign conducted in the context of no active outbreak. We recommend that future assessments of motivation crowd-out embedded in intervention trials identify potential threats to motivation crowd-out analyses and have contingency plans in place.

Perhaps the biggest limitation to studying motivation crowd-out in this context is that our behaviors are not repeated, meaning that we do not observe the same households before, during, and after incentive-based interventions. We therefore proxied crowd-out by comparing participation rates in groups that did and did not receive incentives, a strong but not watertight counterfactual. It is also possible that studying the same behavior in one site, while studying the unrelated and related behaviors in a different site, introduced confounding effects. Assessing motivation crowd-out in pragmatic trials is more feasible when the outcome is measured longitudinally and in the same population.

Conclusions

In this feasibility pilot assessment of motivation crowd out for incentives offered for participation in a public health campaign, we found possible crowd-out for the same behavior as the originally-incentivized behavior, but no crowd-out for subsequent related and unrelated behaviors. Specifically, our results suggest that a lottery incentive to promote participation in the first cycle of a door-to-door indoor residual spray campaign to combat Chagas disease may have decreased participation in the second cycle. Our results are very surprising, especially given the very limited effect of the lottery incentive on participation when it was originally conducted. While our study design cannot definitively rule out other possible explanations for our findings related to motivation crowd-out, we think this is an important contribution to a controversial and inconclusive literature on responses to incentive-based interventions. Our key conclusions are that it is critical to both ask and answer questions about motivation crowd-out in pragmatic trials of incentive-based interventions, and that it is feasible to do so—in this case with the useful addition of related and unrelated behaviors given that the main outcome was measured only once. We recognize continued concerns from public health officials about using incentives to promote participation and engagement in public health campaigns, and hope that thoughtfully-designed incentives will continue to be trialed and, where effective, deployed at scale.