Handwashing and Habit Formation

We examine three behavioral interventions targeted towards young children in an RCT setup in which control households receive a standard public health information campaign only: (1) a dispenser arm, in which the household is also provided with a foaming dispenser and soap; (2) a monitoring arm, in which the household is also informed that their behavior is being tracked by the dispenser and receives biweekly reports on performance; and (3) an incentive arm, in which the household is informed it is being tracked and receives discrete incentives for daily handwashing performance. The design is embedded within the classic habit loop of trigger, routine, and reward: the trigger is an agreed-upon mealtime against which performance is evaluated; the routine is handwashing with soap, and the reward is either clean hands only, a satisfied conscience (monitoring), or a monetary incentive. Within the two latter arms, some households are told that they will receive a future boost in their monitoring or incentive services; this allows us to examine the habit formation process against the standard model of rational addiction. External interventions (monitoring and incentives) will be removed after three months, but household dispenser use will continue to be tracked for one year to measure the precise patterns of habit formation. At the six month mark, households will be cross-randomized into an additional reminder arm, in which alarm clocks programmed for the daily meal time are attached to their dispensers, thereby enhancing the saliency of the trigger and examining limited attention. In conjunction with the high frequency dispenser use data, we will collect biweekly child health data on diarrhea and ARI incidence, as well as detailed anthropometric, blood, and stool data at endline.

2015-10-30

2016-10-31

The primary outcomes of interest encompass both beliefs and behavioral changes of households. We capture beliefs through (1) forecasts of own daily handwashing behavior and (2) willingness to pay for the liquid handsoap dispenser. We capture behavioral change through (3) recorded total daily handwashing rates and (4) recorded dinner time-specific daily handwashing rates. Beliefs and behavioral measures can only be collected for those households with dispensers, so we do not have data from the pure control households on these metrics. Finally, we collect child health data in the form of (5) self-reported biweekly incidence of child diarrhea and respiratory illness (all households), (6) bacteria cultures of random hand swabs (a random subset), and (7) child blood and stool reports (a consenting subset).

A. Household Beliefs
1. Forecasts of own daily handwashing behavior are collected biweekly in efforts to elicit how each intervention impacts household beliefs about future behavior. Respondents are asked to forecast how many days in the coming week they anticipate themselves and their children washing their hands with soap prior to dinner time.
2. Willingness to pay for the liquid handsoap device is elicited using a standard BDM mechanism at midline. We present households with a series of sequentially increasing monetary amounts, and ask them to choose between taking the amount in cash or taking (keeping) the dispenser. We then `randomly' choose a number in the range; if the number is lower than the switching point of the household, the household keeps the dispenser; if it is higher, they receive the cash. Given the potential endowment effect, we recognize that WTP estimates in the pure control are not comparable to those in the remaining arms, but among those households who received a dispenser in Phase I, the WTP six months later serves as an informative measure of how effective or useful households believe the product to be over time.


B. Household Behavior
Handsoap dispenser data was collected every two weeks during surveyor visits. Although it was not possible to identify the identity of the user at any given press, we proxied for separate users by collapsing presses that happened two or fewer seconds apart into a single press. In other words, if the device was used in seconds 34, 35, 37,45, and 46, the first three presses were considered a single use by one household member, and the later two presses as a single use by another member. Though not exact, observations from pilots made clear that a user would press several times in quick succession and rarely return for more soap during a single handwashing event, since the water source (usually a bucket right outside the front porch) was not within reach of the dispenser, unlike the familiar setting of sink, soap, and running water common to more developed contexts.1. Daily handwashing rates are calculated as the sum of all `individual' uses over the course of each twenty-four hour period.
2. Mealtime-specific handwashing rates are calculated as the total number of `individual' uses in the interval of 90 minutes before and after the household's reported start of the evening meal time. If a family reported eating dinner every day at 8:00 PM, for example, this outcome would be the sum of all individual presses observed between 7:00 PM and 8:30 PM.
3. Family use at mealtime is derived from the above and is a binary variable which equals one if at least three `individual' uses were observed in the dinner time interval.
4. Evening handwashing rates are calculated as the total number of `individual' uses any time 5pm or later. Use of the dispenser during this time of day is almost surely tied to use before eating, since all other potential uses (laundry, shampoo, or hand washing after defection or cleaning the house) have all been completed by this time. Since we rolled out in the winter season, children often ate earlier than their parents due to the cold, so the reported mealtime is not a consistent measure of child eating habits. This broader measure allows us to capture all handwashing-before-eating events in the evening.


C. Child Health
1. Incidence of child diarrhea and respiratory illness is collected every two weeks by surveyors, and consists of self reports in which mothers are asked how many days each child has experienced diarrhea in the past two weeks, and likewise for respiratory illness (cough, cold). When possible, the surveyor examines the child in person and asks him/her directly about symptoms.
2. Anthropometric outcomes were collected at baseline, at the four month mark, and at the eight month mark. These included child weight, height, and mid-arm circumference.
3. Bacteria cultures of hand swabs were collected during random, unannounced visits to households two months after the distribution of the dispensers. Subject to budget constraints, a subset of households will randomly be chosen for the audits.
4. Child blood and stool analysis is conducted for a small subset of children whose parents consent to the process.

A baseline survey collecting information on sociodemographic characteristics, health behavior, health knowledge, and aspirations and discount factor measures, is first conducted across all mothers (or primary caretakers of the children in the household) in the full sample of 2947 households. Surveys on child health (ARI and diarrhea episodes) are conducted every two weeks. A midline survey measuring WTP of the dispensers is conducted six months after roll-out, and an endline survey with blood and stool analysis is collected at the twelve month mark. Funds permitting, household behavior among dispenser households are tracked every four months over three years.

The randomization process proceeds in five steps:

1. 105 villages are randomized into Monitoring Villages (MV) and Incentive Villages (IV).
2. Households in MV are then randomized into two groups: (MV1) control and (MV2) dispenser. Households in IV are likewise randomized into two groups: (IV1) control and (IV2) dispenser + monitoring + incentive. This occurs after baseline, before roll-out of dispensers.
3. After roll-out, households in MV are randomized into those who remain dispenser only (MV2) and those who will receive the monitoring service (MV2a); likewise, households in IV are randomized into those who keep the standard incentive (IV2) and those who will experience the thirty-day incentive bump (IV2a). Both the monitoring service (in the case of MV2a) and the incentive bump (in the case of IV2a) are scheduled to begin approximately 1.5 months after the day the household received the dispenser.
4. On the date of the monitoring or incentive change (1.5 months after dispenser distribution), half of MV2 households (MV2b) and half of IV2 households (IV2b) are randomly selected to also receive the monitoring service or incentive boost, respectively.
5. Six months after roll-out, the fifth stage of randomization was completed, with households in all arms cross-randomized into a Reminders (R) intervention.

Incentives and monitoring services are withdrawn three months after roll-out, but household behavior continues to be tracked.

All households receive a basic information campaign regarding the importance of washing their hands with soap, especially prior to eating. They also receive a calendar with the SHDS logo as a token for participation. They are notified that they will be visited biweekly for several months (time left unspecified) to collect information on child health and (for those who received dispensers) check soap supplies, which will be replenished as needed free of charge.

The randomization is conducted in five stages. First, the 105 villages are randomized into Monitoring Villages (MV) and Incentive Villages (IV). Households in MV are then randomized into two groups: (MV1) control and (MV2) dispenser. Households in IV are likewise randomized into two groups: (IV1) control and (IV2) dispenser + monitoring + incentive. These first two stages of randomization are determined after baseline, prior to the roll out of the dispensers. (Households were first randomized at the village level in order to limit the scope for inter-household tension: surveyors expressed concern that control households would be angered if they had some neighboring households who received a dispenser and others who received a dispenser and incentives. It would be easier to justify the interventions through the limited resources lottery framework if all dispenser-receiving households within a village received a consistent package of goods (i.e. the dispenser either always came paired with incentives or never did).) During roll-out, MV2 households are notified about an upcoming lottery in which selected households will receive, along with the dispenser, a monitoring service. Willingness to pay for this monitoring service is elicited, with households informed that a higher willingness to pay will result in a higher chance of winning the lottery for the service (to ensure incentive compatibility). IV2 households are notified about an upcoming lottery in which selected households will receive a larger incentive for thirty days.

After the completion of roll-out, the third stage of randomization is run, determining which households in MV will remain dispenser only (MV2) and which will receive the monitoring service (MV2a), and which households in IV will keep the standard incentive (IV2) and which will experience the thirty-day incentive bump (IV2a). Both the monitoring service (in the case of MV2a) and the incentive bump (in the case of IV2a) are scheduled to begin approximately 1.5 months after the day the household received the dispenser; since roll-out is staggered, the specific date varies by household and is clearly circled on the SHDS calendar by the surveyor. Importantly, households are notified of their future service or incentive immediately after the completion of roll-out. The staggered roll-out implies that some households are told about their future reward two weeks after receiving the dispenser while others are told two days after receiving the dispenser. We embed this variation in timing of the announcement in an effort to disentangle the effects of learning and experimentation from those of habit formation.

The fourth randomized allocation occurs on the date of the monitoring or incentive change (1.5 months after dispenser distribution). At this point, half of MV2 households (MV2b) and half of IV2 households (IV2b) are randomly selected to also receive the monitoring service or incentive boost, respectively. These households, being surprised with this additional service or incentive which is then effective immediately, cannot have altered their behavior in anticipation of the change.

Finally, six months after roll-out, the fifth stage of randomization is conducted, with households in all arms cross-randomized into a Reminders (R) intervention. We choose to integrate this final intervention after the withdrawal of all extrinsic incentives for three reasons: (1) we feared that the effectiveness of reminders might soak up power to detect the rational addiction effects; (2) reminders should have a smaller marginal effect in settings where habits are already deeply embedded, implying heterogeneity in response across treatment arms depending on the effectiveness of the earlier treatments; and (3) logistical ease.

Randomization is done in the office by a computer.

Randomization is at the village level for the first randomization and at the household level for all subsequent randomizations.

No

NA

2947 Households

Control: (IV1) 1286 and (MV1) 499
IV2: 191
IV2a: 310
IV2b: 179
MV2: 130
MV2a: 233
MV2b: 119

With randomization at the household level and handwashing outcomes at the household-day level, we conduct our power calculation using households as our cluster unit and household-days as the number of observations per cluster. We calculate power using pilot results on number of evening presses (5pm and later) per day. Mean: 1.6 SD: 4.0 ICC: 0.04 (but overestimate at 0.15) Power: 0.8 Size: 0.95 To detect a standardized effect size of 0.15 with 80% power, we require 1670 household-day observations per treatment arm. For a large standardized effect size of 0.25, we require 600 household-day observations per treatment arm.

Pre-Analysis Plan: Handwashing and Habit Formation