Incentivizing Behavioral Change: The Role of Time Preferences

Abstract

How should the design of incentives vary with the time preferences of agents? We formulate predictions for two incentive contract variations that should increase ecacy for impatient agents relative to patient ones: increasing the frequency of incentive payments, and making the contract "dynamically non-separable" by only rewarding compliance in a given period if the agent complies in a minimum number of other periods. We test the e cacy of these variations, and their interactions with time preferences, using a randomized evaluation of an incentives program for exercise among 3,200 diabetics in India. On average, providing incentives increases daily walking by 1,300 steps or roughly 13 minutes of brisk walking, and decreases the health risk factors for diabetes. Increasing the frequency of payment does not increase e↵ectiveness, suggesting limited impatience over payments. However, making the payment function dynamically non-separable increases cost-e↵ectiveness. Consistent with our theoretical predictions, agent impatience over walking appears to play a role in non-separability's e cacy: both heterogeneity analysis based on measured impatience and a calibrated model suggest that the non-separable contract works better for the impatient.

‘Pay-Later’ vs. ‘Pay-as-You-Go’: Experimental Evidence on Present-Biased Overconsumption and the Importance of Timing

“…3: "Any Noticeable Short-Term Discounting is Evidence of Present Bias" (p. 274). The same identification strategy of present bias has also been used by Kaur, Kremer, and Mullainathan (2015), Aggarwal, Dizon-Ross, and Zucker (2020) and Kim (2017). I decided against identifying present bias from time-inconsistent choices due to practical concerns.…”

“…To tackle such suboptimal consumption, this paper builds on insights from incentive-design studies, and tests whether pay-as-you-go schemes decrease consumption. The theoretical argument of that literature is that hyperbolic discounting agents increase effort with decreasing time lag to payday (Cutler and Everett 2010, Kaur, Kremer, and Mullainathan 2015, Aggarwal, Dizon-Ross, and Zucker 2020. Under the assumption of βδ-discounting (Laibson 1997), only immediate payments will increase effort provision (Aggarwal, Dizon-Ross, and Zucker 2020).…”

“…The theoretical argument of that literature is that hyperbolic discounting agents increase effort with decreasing time lag to payday (Cutler and Everett 2010, Kaur, Kremer, and Mullainathan 2015, Aggarwal, Dizon-Ross, and Zucker 2020. Under the assumption of βδ-discounting (Laibson 1997), only immediate payments will increase effort provision (Aggarwal, Dizon-Ross, and Zucker 2020). To extend these findings to a bill payment set-up, I run a lab experiment in which participants can buy 'energy' to ease the difficulty of a real effort task.…”

“…Finally, prior literature on how to design contracts in presence of self-control problems, focuses on set-ups with immediate effort costs and future earnings. Examples are medication intake (Cutler and Everett 2010), physical activity (Aggarwal, Dizon-Ross, and Zucker 2020) and work effort (Kaur, Kremer, and Mullainathan 2015). However, findings are rather mixed.…”

Banerjee, Duflo, Kremer, and the Rise of Modern Development Economics*

“…This can happen in a multitude of ways. Some authors develop an explicit theory and find a reduced-form way to test it, such as the following studies for example, Hanna et al (2014) studied limited attention with multiple dimensions to observe, which they examine by studying seaweed farmers; Casaburi and Willis (2018) examined an explanation for low insurance demand, in which they posit that regular insurance conflates moving money across time and states, and then design an experimental test of this idea; Aggarwal et al (2020) looked at the effect of behavioral biases on incentive contract design, which they study in the context of a program to incentivize exercise for diabetics. Other papers have used theory to explicitly design an intervention (e.g., Khan et al, 2019), to structurally estimate parameters using experimental moments (e.g., Bai, 2016;Kreindler, 2020), or to use theory as a more general guide to help interpret the results and provide guidance for further tests (Banerjee et al, 2019).…”