Targeting Incentive Contracts in Heterogeneous Populations

2019-05-20

2021-12-15

Daily steps taken during intervention.

Indicators for daily steps above 10,000, 12,000, and 14,000; Distance from assigned step target to “optimal” step target; step target assigned; steps taken during “phase-in period”.

We randomly assign individuals to one of three targeting regimes: choice, targeting on observables, and random assignment. In addition, a small monitoring group is included for benchmarking purposes; this group will not receive any incentive contract, but we will monitor their steps. We will cross-randomize information on which step target we believe is most likely to be most effective at encouraging walking.

Experiment 1: The design will entail three incentive-targeting treatment groups: choice, targeting on observables, random assignment. A fourth treatment group – the monitoring group – will not receive incentives; this treatment is not of primary interest but included so that we can benchmark the size of the other treatment effects against the size of the gap between the groups receiving incentives (pooled) and the monitoring group. An “information” intervention will be cross-randomized among a subset of these treatment groups. Individuals in all treatment groups will be enrolled in a one-week phase-in period followed by a four-week intervention period.
During the phase-in period, all individuals will receive a fitbit, and will be encouraged to report their walking to the investigators through an automated calling system. Some individuals will have a 6-day Phase In, some will have a 13-day Phase in.
During the intervention period, individuals in the monitoring group will receive fitbits, and will be encouraged to continue to report their walking through the calling system. All other individuals will receive a fitbit and an incentive contract that pays them a fixed rate each day they report reaching a fixed step target through the calling system (e.g. 20 rupees per day 10,000 steps are reached). The rate and step target will vary across individuals, and will be assigned differently across each of the treatment groups. The assignment mechanism in each group is explained below.
Choice 1: The main choice group will be given their preferred contract among a menu with three options: a low-target low-payment option, a middle-target middle-payment option, and a high-target high-payment option. The three contracts are:
• Low: 10,000 step target for 16 INR
• Middle: 12,000 step target for 18 INR
• High: 14,000 step target for 20 INR
Choice 2: A small second choice group will be given their preferred contract among a second menu of three-options with the same three targets of 10, 12, and 14 thousand steps but payments of 10, 15, and 20 INR, respectively.
Targeting: The targeting on observables group will receive a contract where the step target is assigned according to average observed walking in a one-week “pre-period”. The target is either 10,000, 12,000, or 14,000 steps. We assign the target among these that is closest to average steps in the first 6 days of Phase in + 5,500. The assignment formula is based on walking patterns in an earlier experiment incentivizing individuals to walk 10,000 daily steps in exchange for 20 rupees (Aggarwal, et al). The payment level will be 20 rupees in all contracts.
Random: The random assignment group will receive a contract with a randomly assigned daily target of either 10,000, 12,000, or 14,000 steps. The payment level will be 20 rupees in all contracts.
All individuals not in the Targeting group make all menu choices, but not all individuals are assigned an intervention based on these choices. To assess how the lack of information on future walking performance in each contract influences contract choice, we implement an “Information” intervention. In this intervention, we inform individuals of the contract in the first menu that we believe will encourage them to walk the most.

Experiment 2:
In order to better understand the role of private information and learning in choice, we will layer a second experiment on top of Experiment 1 beginning on December 9, 2019. This experiment introduces a new treatment group, the "Baseline Choice" group. The Baseline Choice group is identical to Choice 1 except that they make their choice at Baseline, prior to the Phase-in period.

Experiment 3:
Beginning on January 29, 2020, we estimate that we will have enrolled our target sample for Experiment 1. However, we would like to continue to assess the role of private information as in Experiment 2. Finally, we would like to better understand how whether individuals demand weakly "dominated" contracts, which pay the same for higher step targets. At that point, we will re-shuffle our experimental design in the following ways:
1) We will eliminate the information treatment
2) We will introduce a new "Choice" group. The three contracts on the menu will the same step targets as the other menus (10K, 12K, and 14K) but all contracts will pay 20 INR.
3) We will change our treatment group assignment probabilities.
4) We will layer a new experiment on the existing, and cross-randomize the duration of phase-in between 6 and 13 days (implement starting 2/17)

Randomization is stratified based on our forecasted median age (49) and gender. Ordered treatment assignment lists for each stratum cell are randomly generated in the office by a computer using Stata 15, and treatments are assigned to participants on rolling basis.

Individual

No

1

We aim to enroll approximately 4,000 diabetics, pre-diabetics, hypertensives, and prehypertensives in Experiment 1. We aim to enroll approximately 750 diabetics, pre-diabetics, hypertensives, and prehypertensives in the "Baseline Choice" group added in Experiment 2 and continuing into Experiment 3. We aim to enroll approximately 3-4,000 diabetics, pre-diabetics, hypertensives, and prehypertensives in Experiment 3. We will continue enrollment in Experiment 3 as our funding allows.

The following treatment group distribution is approximate, since enrollees will be assigned to treatments according to baseline strata in a pre-determined order, and some may withdraw following randomization but prior to the intervention.

Experiment 1:
Incentive Groups:
T1 – Random, 10,000 step target: 15%
T2 – Random, 12,000 step target: 25%
T3 – Random, 14,000 step target: 15%
T4 – Choice 1: 25%
T5 – Choice 2: 0.5%
T6 – Targeting: 17.5%

Non-incentive Group:
Monitoring Only: 2%

Information Treatment:
60% of the experimental sample excluding the Targeting group will be cross-randomized into the Information treatment prior to deciding which contract they prefer in Choice Menu 1.

Experiment 2:
20% Baseline Choice
80% Allocated to Experiment 1 (no change to those within Experiment 1)

Experiment 3:
Incentive Groups:
T1 – Random, 10,000 step target: 9.3%
T2 – Random, 12,000 step target: 9.3%
T3 – Random, 14,000 step target: 9.3%
T4 – Choice 1: 19%
T5 – Choice 2: 0.5%
T8 – Choice 3: 19%
T6 – Targeting: 10%
T7 – Baseline Choice: 19%
Non-incentive Group:
Monitoring Only: 4.25%

Phase-in period duration cross-randomization:
60% of the experimental sample will be cross-randomized into a 13-day (rather than 6-day) phase-in period.

For comparing walking in the intervention period between our primary groups of interest (T5 - Choice 1 and T2 - Random, 12,000 step target), our MDE is 400 average daily steps. For comparing intervention-period walking between T6 – Targeting and T2 – Random, 12,000, our MDE is 550 average daily steps (here we cannot control for phase-in walking as this will be influenced by the assignment to the T6 – Targeting group). For comparing walking in the one-week phase-in period between T6 – Targeting and all other groups, our MDE is 460 average daily steps. For assessing the impacts of the Information intervention, we are powered to detect changes in the proportion of individuals assigned to a given target of 7 percentage points. Note: While this text was part of our originally-drafted text that we intended to upload to the platform and submit for review, we discovered in March 2023 that it was not uploaded due to a clerical error. At that time, we added this text to the registry, as well as uploaded our original draft and time stamp (to show that we originally intended to include the section) to the “Supporting Documents & Materials” section of the registry.