Information, Participation and Deliberation: Experimental Evidence from Bangladesh

The project is conducted within an arsenic mitigation program implemented in rural Bangladesh. Key aspects of the program are decided by unanimous consensus during community meetings open to the whole community.
We evaluate the impact of an SMS reminder initiative on the number and characteristics of attendees at the community meeting, and how these effects vary with the content of the SMS.
We randomize communities across three SMS-treatments:
(i) No SMS;
(ii) SMS with individual-level information: reminder on the date, hour and location of the community meeting, plus a reminder on the arsenic and bacteria contamination level of the household’s water source;
(iii) SMS with community-level information: reminder on the date, hour and location of the community meeting, plus information on the share of arsenic- and bacteria-contaminated water sources in the community.

2015-10-15

2018-01-31

Number of participants at the community meeting.

Characteristics of participants at the community meeting in terms of: gender, arsenic and bacteria contamination of the water source used by the household, household network size, household leadership status, household poverty index, household distance to the meeting location.

Characteristics of the deliberation: number of people speaking during the meeting; length of the meeting; number of proposals; number of community meetings before reaching an agreement; share of households in the community providing cash/labour contributions.

Project outcomes: share of households within 5 minutes walking from the new water source among those using an arsenic-contaminated water source at baseline; inequality in the distribution of benefits from the project; characteristics of the households benefiting the most from the project; self-reported use of the new water source; quality of drinking water as measured by arsenic and bacteria tests at the water source and in the household.

We plan to extend the empirical analysis to evaluate qualitative data on the dynamics of the negotiation process during the community meeting(s) and the details of the project implementation. We will use the audio recording from the community meeting(s) as well as detailed reports from our project staff on the different phases of the implementation. We will define these outcomes variables once the relevant data is available. We plan to analyze these qualitative data in order to elucidate mechanisms rather than as part of the main analysis.

We conduct our main study in 171 communities eligible for receiving the arsenic mitigation program. The mean community size is 153 households. We selected communities for inclusion based on their baseline level of arsenic contamination: (i) communities where more than 25% of sources of drinking water are arsenic-contaminated; (ii) or communities where arsenic-contaminated water sources are geographically concentrated in few clusters. We randomly select 129 communities for receiving the program. Treated communities are further divided across three requirements for co-funding the installation costs: no contributions; cash contributions; labor contributions.

Before the implementation of the arsenic mitigation program we collected baseline data in all communities. We conducted a census of all sources of water used for drinking or cooking. We also randomly selected 40 households per community for an in-depth household survey.
As part of the baseline water source census, we offered a water quality testing program. 99.9% of water source owners or caretakers agreed to participate in the water source testing program. We also requested water source caretakers or owners to share their phone numbers with project staff. We used their phone numbers at baseline to send automatically generated SMS messages with the results of the water quality tests.

We used the phone number of caretakers of water sources tested at baseline in order to implement our SMS treatments. Water source owners are not a random selection of the population of households, but in our context, on average 71% of households own a water source. As a result, the sample of water source owners generally represents a large majority of the households in most communities. We also requested water source owners to share the message with other households who use their water sources, meaning that the reminders should have reached a large majority of households in the treatment unit.

In communities selected for the SMS-treatments, we send two reminders. We sent one message at 7 pm the evening before the community meetings, and one message at 7 am on the morning of the day when the meeting was held. The English translation of the text of the SMS reads:
1) Individual information treatment: “Hello {name}. You are welcome to attend the community meeting organized by NGO Forum for Public Health today/tomorrow in your community, {location}, at {time}. Your participation is important! Remember, your tubewell tested bacteria contaminated/not-contaminated and arsenic positive/negative! Please share this message with households who share your tubewell.”
2) Collective information treatment: “Hello {name}. You are welcome to attend the community meeting organized by NGO Forum for Public Health today/tomorrow in your community, {location}, at {time}. Your participation is important! Remember, of water sources in your community {share}% are bacteria contaminated and {share}% are arsenic contaminated! Please share this message with households who share your tubewell.”

Our design allows us to perform two sets of tests in order to evaluate: (i) the impact of sending SMS-reminders; (ii) whether the effects vary with the content of the information included in the SMS.

We randomly assign communities to receive the arsenic mitigation program and the contribution requirements by public lottery, stratified by Union Parishad.

We randomize the SMS-treatment status in office, using STATA pseudo-random number generator. The randomization is stratified by Union Parishad, and by assignment to treatment under a given contribution regime.

The randomization is performed at Treatment Unit level.

Treatment Units are defined within the arsenic mitigation program in order to identify target communities. Based on previous experience with similar interventions, implementing the project in large communities presents practical challenges that might compromise the success of the program. We deal with this constraint by defining Treatment Units as follows:
- Treatment Units correspond to villages, defined by administrative boundaries, if the village size is less or equal than 250 households;
- We divide larger villages in smaller Treatment Units along geographic boundaries. Villages with 250-500 households are divided in two Treatment Units, with 500-750 households in three Treatment Units, etc.

We refer to communities or treatment units interchangeably in this document.

No

129 Treatment Units.

Outcomes using household survey data: approx. 5,160 (approx. 40 households per Treatment Unit). Treatment Unit outcomes: 129.

Non-SMS: 51 Treatment Units.
Individual information treatment: 39 Treatment Units.
Collective information treatment: 39 Treatment Units.

The reported numbers refer to the original design of the intervention. We expect the actual sample used in the empirical analysis to shrink because of two constraints.
First, in a few cases we faced unexpected technical problems, and we were not able to send the SMSs according to our plans.
Second, in a few villages we were not able to provide project tubewells due to geological constraints that impeded installation. We learned about these limitations only after the community meeting(s) was held and after the community reached an agreement and raised the required contributions.

We will test that the reasons for unsuccessful delivery of the SMSs, as well as the reasons for installation failures, are orthogonal to the SMS-treatment status.
We will exclude the Treatment Units where we were not able to send the SMS-reminders from the final sample, and the Treatment Units where we failed to install the project tubewell from the analysis on outcomes observed only conditional on installation.

We report the most conservative power calculations relative to community-level specifications. When estimating the effect of the SMS-reminder (pooling together the two SMS treatments), the minimum detectable effect relative to standard deviation at the standard 80% power is 0.5. It increases to 0.6 when we disaggregate the two SMS treatments in order to examine the differential effects of sending SMS reminders with individual- or community-level information. We plan to estimate household-level specifications for outcome variables observed at the household level, improving the power of the study. We stratify the SMS-treatments by contribution requirements in order to ensure balance of the study, but we do not expect to have enough power to exploit the cross-randomization structure to estimate heterogeneous effects.