The effect of short-term subsidies on demand for potable water in rural Bihar, India: Pilot study

This pilot study tests the impact of subsidies for week-long subscriptions of daily potable water delivery in 20-liter jugs to participating families.

2019-03-24

2019-04-06

Our primary outcome of interest is uptake of subscription renewals between high price vs. low price treatment groups at end-line.

For an intent to treat analysis, we will compare the roughly 50% of households receiving a low price (10 - 20 INR) vs. those receiving a high price (30 - 40 INR). As a sensitivity test we can adjust the threshold for high vs. low prices at baseline by comparing a low price of 10 INR vs. a high price of 20 INR and above, as well as a low price of 30 INR and below vs. a high price of 40 INR. To further test sensitivity of our threshold, we may also include 5 INR increments, testing at each possible point.

In this pilot we will use a random price auction to offer random levels of subsidy (at 10 rupee increments between 42.9%-85.7% off the market price) for a 1-week potable water delivery subscription. At end-line, we will return to households and offer the same slightly discounted price for the water delivery subscriptions. The purpose will be to test whether higher subsidies increase or decrease demand for the product when it is offered again later.

At end-line interviews, families who received week-long subscriptions starting at baseline will all be offered one additional week of deliveries at the same price. Our outcome of interest will be the uptake of an additional week of deliveries among households who received high subsidies at baseline versus those who received low subsidies at baseline. The thresholds for high vs. low will be those drawing 10-20 INR in the low price arm, and those drawing 30-40 INR in the high price arm. Prices for subscriptions at baseline will vary randomly between (but not including) zero rupees and 70 rupees (the market rate). The potential random prices will be increments of 10 rupees (10, 20, 30, 40, possibly 50). We do not include 60 INR as we anticipate that uptake will be extremely low at these levels. The main purpose of the study will not be to test our core hypotheses around anchoring vs. learning effects. Instead we will use the pilot to determine a demand curve and identify predictors of positive demand among households who participate. For the full trial we will both need to keep from including households with no willingness to pay for the product (those that bid zero, for example, during the random price auction) and to establish the precise thresholds at which we can expect families to purchase water delivery subscriptions at baseline. We will want a "low" price versus a "high" price as the two treatment arms for the full trial so that we can ensure a sufficient sample size to compare the two arms.

We must determine the optimal price at which to offer subscriptions between a "high" vs. a "low" price group in the full RCT such that a "high" price would achieve roughly 20 - 50% uptake during the subsidy period, and a "low" price for which we can achieve a much higher uptake. The reason for this is so that in the follow-up period of the full study, when introducing the same "high price" to both groups, we will have sufficient space both above and below the control group uptake (which we expect to remain constant) to be able to measure either a relative increase (learning effects) or decrease (anchoring effects) in uptake among low price the treatment group.

The drawing of a price for the week-long subscriptions during the random price auction constitutes our method of randomization.

Individual households.

No

No clustering.

A minimum of 200 households. Our IRB approval allows for a maximum of 300 households, but budget constraints will likely keep us to 200.

Presuming that the in-person randomization process establishes balanced prices, we expect the following groupings to emerge:
10 INR: 50 households
20 INR: 50 households
30 INR: 50 households
40 INR: 50 households
For an intent to treat analysis, we will compare those 100 households receiving a low price (10 - 20 INR) vs. those 100 receiving a high price (30 - 40 INR). Increasing the sample to 300 households would yield a sample of 150 households per arm.

With alpha 0.05, beta 0.8, for a two-tailed test, assuming a 20% uptake among the "high price" group at end-line, and with 50%, of the sample assigned to either arm, we will be able to detect a 15.9 percentage points MDE in uptake between arms. This also assumes 0% attrition in either sample arm. Increasing the sample to 300 households with all other parameters remaining the same would yield an MDE of 13.0 percentage points difference between arms.