The Impact of Experiential Learning Interventions on Sustainable Water Management: Evidence from a Randomized Control Trial in India

Last registered on January 23, 2023

Pre-Trial

Trial Information

General Information

Title
The Impact of Experiential Learning Interventions on Sustainable Water Management: Evidence from a Randomized Control Trial in India
RCT ID
AEARCTR-0010765
Initial registration date
January 13, 2023

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
January 23, 2023, 6:02 AM EST

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Locations

Region
Region

Primary Investigator

Affiliation
International Food Policy Research Institute

Other Primary Investigator(s)

PI Affiliation
International Food Policy Research Institute
PI Affiliation
International Food Policy Research Institute
PI Affiliation
International Food Policy Research Institute
PI Affiliation
International Crop Research Institute for the Semi-Arid Tropics
PI Affiliation
Foundation for Ecological Security

Additional Trial Information

Status
On going
Start date
2021-10-15
End date
2023-12-31
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Abstract
This study provides rigorous testing of the effectiveness of a package of experiential learning tools related to groundwater management in India on mental models, norms, and behavior. The experiential learning intervention includes three main tools: 1) collective action games that allow people to experience the effect of their water use decisions on water management and to try out different rules; 2) structured community debriefings to discuss the implications of the games; and 3) crop water budgeting to facilitate participatory water resource planning. We use a collective action groundwater game developed and piloted by the International Food Policy Research Institute (IFPRI) and Foundation for Ecological Security (FES), in which participants choose between crops with different water consumption levels and see the effects of their decisions on water tables over multiple rounds of play. To facilitate full participation of women and men, games take place in separate groups of women and of men. The intention of the games is to create awareness and receptivity to changing water management practices. We then follow game play with a debriefing discussion with the full community. Finally, we complement game play with provision of information about technical and institutional options for more sustainable water management. Specifically, we do so by engaging with the farmers to apply a Crop Water Budgeting tool, which supports community-level discussions for development of collective cropping patterns that save water.

The primary population of interest are farmers using groundwater in Andhra Pradesh, Karnataka, and Rajasthan states in India, where overuse of groundwater has led to falling water tables. All districts also have hard rock (confined) aquifers, such that local water extraction—or refraining from extraction—can affect local water tables. Within each state, the key selection criterion was districts and communities where Foundation for Ecological Security (FES) has an active program in commons management for at least 10 years, so that there would be community resource persons (CRPs) who could be trained to implement the experiential learning tools.

We selected a total of 472 habitations (communities) where the interventions had not been used, stratified by the presence of surface irrigation, and randomly assigned 2/3 of the sample to treatment and 1/3 to control. To select individuals, focus group discussions (FGDs) were held all communities; among FGD members, 5 women and 5 men were randomly selected to play the game, and 3 men and 3 women randomly selected for an individual survey. In control communities, 3 men and 3 women were similarly selected for the survey; no games or debriefing took place. Baseline data collection and first round of interventions took place.

Baseline data collection was conducted over 3 months (November 15th 2021 – 7th March 2022). The intervention was repeated in September-December 2022, and endline data collection is planned to begin on January 17, 2023. Major data sources are baseline and endline key informant interviews and focus group discussions; an individual survey of game players and game data; community debriefing discussion; crop water budgeting data; and habitation records.
External Link(s)

Registration Citation

Citation
Falk, Thomas et al. 2023. "The Impact of Experiential Learning Interventions on Sustainable Water Management: Evidence from a Randomized Control Trial in India." AEA RCT Registry. January 23. https://doi.org/10.1257/rct.10765-1.0
Sponsors & Partners

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information
Experimental Details

Interventions

Intervention(s)
Overextraction of groundwater has led to depletion of this vital resource in many areas in India. This study examines the effect of experiential learning tools related to groundwater management, including: 1) collective action games that allow people to experience the effect of their water use decisions on water management and to try out different rules; 2) structured community debriefings to discuss the implications of the games; and 3) crop water budgeting to facilitate participatory water resource planning. We examine the effects of the experiential learning package on mental models, norms, rules, and behavior.
Intervention Start Date
2021-11-15
Intervention End Date
2022-10-15

Primary Outcomes

Primary Outcomes (end points)
In this study we plan to work with three types of outcomes: 1) individual level outcomes; 2) group level outcomes; and 3) habitation level outcomes. For this, we will use information from five sources: a) Focus group discussion (FGD); b) Key informant interview (KII); c) Individual survey of game players; d) Crop Water Budgeting (CWB) report; and e) Habitation records.

We propose nine hypotheses:
1. Hypothesis 1: Outcomes related to better system understanding (individual and group levels).
2. Hypothesis 2: Outcomes related to stronger norms for sustainable water management (individual and collective level).
3. Hypothesis 3: Outcomes related to women more involved in water-related decision-making (collective level).
4. Hypothesis 4: Outcomes related to water management rules (individual and collective levels).
5. Hypothesis 5: Outcomes related to better rule enforcement of water management rules (collective level).
6. Hypothesis 6: Outcomes related to more likely to invest in water harvesting structures or leverage public funds for the purpose (collective level).
7. Hypothesis 7: Outcome related to conditions of water-related infrastructure (collective level).
8. Hypothesis 8: Outcome related to crop choice with lower water consumption leading to irrigated water savings (collective level).
9. Hypothesis 9: Outcome related to depletion of water table (collective level).
Primary Outcomes (explanation)
Note: II = Individual Interview, KII=Key Informant Interview; FGD=Focus Group Discussion data
Hypothesis 1: Outcome related to better system understanding (individual and group levels):
Our main outcome variable is an index:
• iUnderstanding = sum of iBenefitCoordination, iFamersInteract, and iUseAffectsGW
Where the indicator takes on a 1 if the individual responds that they “Agree” or “Strongly agree”:
• iBenefitCoordination: II7.4 Coordinating and regulating water management in the community is for the benefit of everybody in the community.
• iFamersInteract: II7.5 One farmer’s crop choice affects other farmers in the community?
• iUseAffectsGW: II7.6 Our ground water use now will affect the availability of the resource in the future for the community as a whole.

Hypothesis 2: Outcomes related to stronger norms for sustainable water management (individual and collective):
Our main outcome variable is an index:
iNorms = Sum of iRulesGroundwater, iCooperateToInvest, iNoRulesGroundwaterOwnLand, iAllAccessSurface, iSupportGovtRules, iSupportGvtPolicy1, iSupportGvtPolicy2.
Where for the following 2 indicators =1 if the individual responds that they “Agree” or “Strongly agree”:
• iRulesGroundwater: II7.8 Community members should make rules about use of groundwater, such as about well locations and depth, or when, where, or how much groundwater may be used, for what purposes
• iCooperateToInvest: II7.10 Community members should cooperate to invest in the establishment and maintenance of important community water structures such as dams
The following 2 indicators =1 if the individual responds that they "Disagree" or "Strongly disagree":
• iNoRulesGroundwaterOwnLand: II7.7 Farmers should be allowed to use the groundwater under their land without any rules and restrictions
• iAllAccessSurface: II7.9 Anybody who can access surface water bodies, such as streams or tanks, should be able to use water without any rules or restrictions
The following 3 indicators = 1 if the individual responds that they "Support" or "Strongly support"
• iSupportGovtRules: II6.11 Would you support if the government introduced new rules restricting farmers water use?
• iSupportGvtPolicy1: II6.12 Would you support a government policy which increases electricity charges for agricultural use, esp. pumping of water?
• iSupportGvtPolicy2: II6.13 Would you support a government policy which increases electricity charges for agricultural use, esp. pumping of water if at the same time it increased the minimum support price for water efficient crops?

Hypothesis 3: Outcomes related to women more involved in water-related decision-making (collective level):
• iWWCropChoice = II3.31 Who decides which crop to grow? For women = 1 if answers 1 (I alone), 3 (Jointly with spouse), 4 (jointly with other family members); 5 (Jointly with other community members); =0 if answers 2 (spouse alone); missing if =999 (Refuse to answer) [Dummy variable]
• iMWCropChoice = II3.31 Who decides which crop to grow? For men = 1 if answers 2 (Spouse alone) or 3 (Jointly with spouse), 6 (jointly with other female family members); 7 (Jointly with other female community members); =0 if answers 2 (I alone); missing if =999 (Refuse to answer) [Dummy variable]

Hypothesis 4: Outcomes related to water management rules (individual and collective levels):
Our main individual outcome variable is an index:
• iRules = sum of answers “Agree” or “Strongly agree” to iRulesPrioritizeGwater, iRulesPrioritizeSwater, iRulesFarmerCropCh [Range 0-3].
Where
• iRulesPrioritizeGwater: II7.1 In our community there are rules with regard to prioritizing groundwater use.
• iRulesPrioritizeSwater: II 7.2 In our community there are rules with regard to prioritizing surface water use.
• iRulesFarmerCropCh: II 7.3 In our community are some arrangements or rules related to individual farmer’s crop choices.
Our main community outcome variable is an index:
• cRules1 = sum of answers of “yes” the outcome indicators : cRulesVillageComm, cRulesDigWells, cRulesGrowCrops, cRulesWaterShortage [Range 0-4].
Where
• cRulesVillageComm: KII 9.1 Are there any village committees or other community organizations which deal with the management of groundwater or surface water resources?
• cRulesDigWells: KII 10.1 Are there any rules about who can dig or deepen wells to use groundwater under their land in this community, or what kind of pumps they can use?
• cRulesGrowCrops: KII 10.5 Are there any rules about what crops they can grow under groundwater irrigation in any season?
• cRulesWaterShortage: KII 11.8 In cases of water shortage (e.g. drought years or dry spells), are there discussions in the community on individual farmer’s crop choices?

Hypothesis 5: Outcomes related to better rule enforcement of water management rules (collective level).
3 outcome indicators= 1 if the individual responds that they “Agree” or “Strongly agree”:
• cCropRulesEnforced: KII 10.7 Are these rules [related to what crops they can grow under groundwater irrigation in any season] enforced? [see if this was asked for each rule]
• cOperationRulesEnforced: KII 11.19 Are the rules related to operation or maintenance of the water harvesting structures enforced?
• cIrrigationRulesEnforced: KII 11.4 Are rules related to individual farmer’s crop choices from surface irrigation enforced?

Hypothesis 6: Outcomes related to more likely to invest in water harvesting structures or leverage public funds for the purpose (collective level).
Outcome indicator=1 if the individual responds “Yes”.
• cMaintainKII = KII 11.10 Has water harvesting structures (tank, stop dam, or check dam) in your community received any maintenance within the last 12 months? [Dummy variable]

Hypothesis 7: Outcome related to conditions of water-related infrastructure (collective level)
Our main individual outcome variable is an index:
• cWHPoor = Sum of the community-level variables cReducedCap1, cReducedCap2, cStateEarthWalls [Range 3-15, where higher values indicate good state of infrastructure]
Where
• cReducedCap1: KII 14.2 Capacity reduced due to siltation [Range 1= Very much, 5= Not]
• cReducedCap2: KII 14.3Capacity reduced due to overgrowing vegetation [Range 1= Very much, 5= Not]
• cStateEarthWalls: KII 14.4 State of the earth walls [Range 1= Very poor, 5= Very good]

Hypothesis 8: Outcome related to crop choice with lower water consumption leading to irrigated water savings (collective level)
Our primary outcome indicator is from crop water budgeting:
• cRabiSavings = CWB_X – CWB_AB [in TCM] where
• CWB_X = Total water requirement as per farmers' own plan
• CWB_AB = Total water requirement as per revised plan

Hypothesis 9: Outcome related to depletion of water table (collective level)
Outcome indicators from habitation records:
CRabiWaterTableChangey = CWaterTablePreRabiy - CWaterTablePostRabi y as an estimate of the change in water table during Rabi
Where
CWaterTablePreRabiy = average depth of monitoring wells as recorded in habitation records at start of Rabi seasons in the two years before the intervention started.
CWaterTablePostRabi y = average depth of monitoring wells as recorded in habitation records at start of Rabi seasons in the two years after the intervention was completed.
We can compare CRabiWaterTableChangey between treatment and control communities in the 2022/23 Rabi season, as well as to compare the treatment communities in 2022/23 with the Rabi season water table changes in previous years, controlling for local rainfall in the Rabi and previous Kharif season and dummy variables for years to control for other year-specific factors.
However, the effect of the intervention on the water table may take years to appear. Thus, we suggest that this be monitored on a long-term basis.

Secondary Outcomes

Secondary Outcomes (end points)
Hypothesis 1
We will examine the effect of the intervention on the individual variables iBenefitCoordination, iFamersInteract and iUseAffectsGW
To the extent that there are differences between “agree” and “strongly agree in the individual varuables, we will complement our analysis with an ordered logit of these variables.

All individual indicators mentioned in the primary outcome section will further be aggregated at the group level and condensed to an index by adding all three above, group-level outcome indicators.

Hypothesis 2
We will examine the effect of the intervention on the individual variables iRulesGroundwater, iCooperateToInvest, iNoRulesGroundwaterOwnLand, iAllAccessSurface, iSupportGovtRules, iSupportGvtPolicy1, iSupportGvtPolicy2.
For sensitivity analysis, we will also conduct a principal component using all individual indicators aggregated at the group level and compute an index. These would be mainly variables at the group level and some of them at the community level.

Hypothesis 3
We will generate an indicator using individual variables aggregated at the group level and we will also use responses to questions in the Focus Group Discussions related to women’s involvement in decisions among the community.

Hypothesis 4
We will examine the effect of the intervention on the individual variables used to compute the index.
We will use the habitation records to complement our analysis and will create our second main community-level outcome for this hypothesis.

Hypothesis 6
We will complement the analysis using individual contribution variables. Some of them add new dimensions and others can be used to cross check the information provided by key informants. We will also include community-level indicators created from the FGDs to complement our analysis.
If habitation records indicate the amount of resources mobilized, this will be noted in discussion.

Hypothesis 7
We also define community-level indicators using the Focus Group Discussion information to complement our analysis.

Hypothesis 8
We will use crop choice data from the individual surveys to create additional variables and use community level crop choice data from key informant interviews to test this hypothesis.
Secondary Outcomes (explanation)
Hypothesis 1
All individual indicators will further be aggregated at the group level and condensed to an index by adding all three above, group-level outcome indicators.
gUnderstanding = addition considering: gBenefitCoordination, gFamersInteract, and gUseAffectsGW

Hypothesis 2
For sensitivity analysis, we will also conduct a principal component analysis and use the first component as the outcome variable.
All individual indicators will further be aggregated at the group level:
• gRulesGroundwater = mean of iRulesGroundwater across the group playing the game (i.e., 3 men or 3 women) [Range 0-3]
• gCooperateToInvest = mean of iCooperateToInvest across the group playing the game (i.e., 3 men or 3 women) [Range 0-3]
• gNoRulesGroundwaterOwnLand = mean of iNoRulesGroundwaterOwnLand across the group playing the game (i.e., 3 men or 3 women) [Range 0-3]
• gAllAccessSurface = mean of iAllAccessSurface across the group playing the game (i.e., 3 men or 3 women) [Range 0-3]
• gSupportGovtRules = mean of iSupportGovtRules across the group playing the game (i.e., 3 men or 3 women) [Range 0-3]
• gSupportGvtPolicy1= mean of iSupportGvtPolicy1 across the group playing the game (i.e., 3 men or 3 women) [Range 0-3]
• gSupportGvtPolicy2= mean of iSupportGvtPolicy2 across the group playing the game (i.e., 3 men or 3 women) [Range 0-3]
And we will compute an index
• gNorms = Sum of seven group-level outcome indicators considering: gRulesGroundwater, gCooperateToInvest, gNoRulesGroundwaterOwnLand, gAllAccessSurface, gSupportGovtRules, gSupportGvtPolicy1, gSupportGvtPolicy2.
Then, we will collect information at the community level from the Focus Group Discussion to complement our analysis. In this regard we will include the following three outcome indicator—where the indicator takes on a 1 if the individual responds that they “Agree” or “Strongly agree” to the following proposition:
cNorms = FGD7.9 Do you think people should consult others in the community in making their personal choices about how to use groundwater?

Hypothesis 3
The indicator will further be aggregated at the group level:
• gWWCropChoice = mean of iWWCropChoice across the group playing the game (i.e., 3 men or 3 women) [Range 0-3]
We will also use responses to the following questions in the Focus Group Discussions:
• cWCropChoice = FGD8.11 How strongly are women involved in deciding about crop choices in your community? [Range 1-5: 1=only men decide, 5= only women decide]
• cWMaintenance = FGD8.12 How strongly are women involved in deciding about the maintenance of water infrastructure in your community? [Range 1-5: 1=only men decide, 5= only women decide]
The same question has been asked in the Key Informant Interviews and the results will be compared:
• cKICropChoice = KI11.21 How strongly are women involved in deciding about crop choices in your community? [Range 1-5: 1=only men decide, 5= only women decide]
• cKIMaintenance = KI11.22 How strongly are women involved in deciding about the maintenance of water infrastructure in your community? [Range 1-5: 1=only men decide, 5= only women decide]

Hypothesis 4
All individual indicators will further be aggregated at the group level:
Additionally, we will use the habitation records to complement our analysis and will create our second main community-level outcome for this hypothesis.
• cRules2 = 1 if in village habitation records any rules regarding water management that have been recorded after the intervention [Dummy variable]

Hypothesis 6
We use the following additional individual contribution variables. Some of them add new dimensions and others can be used to cross check the information provided by key informants.
• iContributeMaintenanceMoney: II 8.1 Have you or your family members contributed money to the maintenance of water harvesting structures during the last year
• iContributeMaintenanceLabout: II 8.2 Have you or your family members contributed labour to the maintenance of water harvesting structures during the last year?
We then create a community-level indicator from the last 2 individual-level to complement our analysis.
• cContribute = Community level mean of the individual level total of the following 2 individual-level variables iContributeMaintenanceMoney and iContributeMaintenanceLabout [Range 0-2].
While the foregoing variables are our primary outcome variables for this hypothesis, the following community-level indicators will be created from the FGDs to complement our analysis:
• cProtect = FGD3.2 Efforts of catchment protection or management [Range 1-4].
• cPInvolvement = FGD3.3 Degree of involvement of villagers in catchment protection [Range 1 -2]
• cMaintainFGD = FGD 4.2Efforts of maintenance of water infrastructure [Range 1-4]
• cMInvolvement = FGD4.3 Degree of involvement of villagers in maintenance of water infrastructure [Range 1 -2]
• cInvest = 1 if habitation record mentions community contributions to maintenance of water harvesting structures [Dummy variable]
• cMobilize = 1 if habitation record mentions mobilizing public works funds (MGNREGA) or other external funds for maintenance of water harvesting structures [Dummy variable]

Hypothesis 7
We also define the following community-level indicators using the Focus Group Discussion information to complement our analysis:
• cCatchment = FG3.1 Catchment condition [Range 1-4 where 1 = Very unhealthy/poor and 4 is Healthy and excellent]
• cSurfaceInf = FG4.1 (How many months is there sufficient and good quality water for the whole village?) Choose which is most closely applicable to surface water infrastructure in your village [Range 1-4 where 1 = Very unhealthy/poor and 4 is Healthy and excellent]
• cMitigate = FG5 Capacity of the water infrastructure to help mitigate drought or flood [Range 1-3, where 1= The village can’t withstand even a single year of drought or floods as infrastructure is inadequate; 2= The village can’t withstand even a single year of drought or floods although it takes time to get back to normal; and 3 = The village can withstand consecutive years of drought or floods and can bounce back quickly]

Hypothesis 8
In addition, we use individual crop choice data to create the following variables:
• iKharifCropType = 1 if II3.3 Main kharif crop grown in last Kharif season is classified according to local field staff expertise [Range 1-4 where 1= very water consumptive, 2 = rather water consumptive, 3 = rather water saving, 4= very water saving]
• iRabiCropType = 1 if II3.13 Main Rabi crop grown in last Rabi season is classified according to local field staff expertise [Range 1-4 where 1= very water consumptive, 2 = rather water consumptive, 3 = rather water saving, 4= very water saving]
• iPerennialCropType = 1 if II3.23 Main perennial crop grown in last Rabi season is classified according to local field staff expertise [Range 1-4 where 1= very water consumptive, 2 = rather water consumptive, 3 = rather water saving, 4= very water saving]
We also use community level crop choice data from key informant interviews to test this hypothesis. The following variables will be created.
Taking into account that 1 to x refers to the different crops mentioned under KI8.1 Major crops grown in kharif season in the last season:
• cChangeCropWaterReqKharif = Endline sum of (crop water requirement of crop (1 to x) as estimated by the local field staff times Endline KI8.2 Area under kharif crop (1 to x) in acres) / Endline sum of (crop water requirement of crop (1 to x) as estimated by the local field staff times Endline KI8.2 Area under kharif crop (1 to x) in acres)
• cChangeCropWaterReqRabi = Endline sum of (crop water requirement of crop (1 to x) as estimated by the local field staff times Endline KI8.2 Area under rabi crop (1 to x) in acres) / Endline sum of (crop water requirement of crop (1 to x) as estimated by the local field staff times Endline KI8.2 Area under rabi crop (1 to x) in acres)
• cChangeCropWaterReqSummer= Endline sum of (crop water requirement of crop (1 to x) as estimated by the local field staff times Endline KI8.2 Area under summer crop (1 to x) in acres) / Endline sum of (crop water requirement of crop (1 to x) as estimated by the local field staff times Endline KI8.2 Area under summer crop (1 to x) in acres)
• cChangeCropWaterReqPerren= Endline sum of (crop water requirement of crop (1 to x) as estimated by the local field staff times Endline KI8.2 Area under perennial crop (1 to x) in acres) / Endline sum of (crop water requirement of crop (1 to x) as estimated by the local field staff times Endline KI8.2 Area under perennial crop (1 to x) in acres).

Experimental Design

Experimental Design
This randomized control trial focuses on three states in India: Andhra Pradesh, Karnataka, and Rajasthan. These states were selected due to the severity of their water crises and their high dependence on groundwater. Within each state, the key selection criterion was districts where FES has an active program in commons management for at least 10 years, so that there would be community resource persons (CRPs) who could be trained to implement the experiential learning tools. All districts also have hard rock (confined) aquifers, such that local water extraction—or refraining from extraction—can affect local water tables.

We randomly sampled habitations into treatment and control groups using a stratified (by our six strata, defined by state and presence or lack of presence of surface irrigation infrastructure) randomization procedure designed to achieve two-thirds of the sample being assigned to treatment and one-third to control. In particular, we used the “randtreat” command in Stata, while indicating our use of these stratifying variables and our desire for a 2/3 – 1/3 ratio of treatment to control units.
Experimental Design Details
This impact assessment focuses on three states: Andhra Pradesh, Karnataka, and Rajasthan. These states were selected because they are the driest of our project states, where water crises are most severe, and they have high dependence on groundwater. While this means findings may not generalize to less-dry states, it importantly sheds light on possible solutions in areas that need them most in a climate crisis. Within each state, the key selection criterion was districts where FES has an active program in commons management for at least 10 years, so that there would be community resource persons (CRPs) who could be trained to implement the experiential learning tools. All districts also have hard rock (confined) aquifers, such that local water extraction—or refraining from extraction—can affect local water tables. The process for developing a sampling frame and then selecting the communities and randomly allocating them to treatment and control is discussed below, after presenting the treatments and data sources.

This study focuses on four districts in the three states: Andhra Pradesh (2 districts), Karnataka (1 district), and Rajasthan (1 district). In these districts, project partner FES had targeted 834 habitations (i.e., a named, distinct cluster of houses that constitutes the local community) for future interventions involving experiential learning for water management, but in which they had not yet done so at the time of our sample selection. We had access to geographic identifiers in addition to data on 40 variables related to water systems and poverty for these habitations (these data come from a primary survey done by FES at the start of engagement with habitation communities). Given our interest in habitations with irrigation present, we further eliminated 151 habitations with strictly less than 10 percent of land irrigated, bringing the total down to 683 habitations, which comprise our final sample frame.

Our goal was to select a sample from this sample frame with equal numbers of habitations in each of six strata, defined by the state (among three) in which the habitation was located and whether or not the habitation has infrastructure used for surface irrigation. The sample was stratified by preexisting FES data on availability of surface irrigation because surface irrigation can decrease pressure on groundwater by providing an alternative source of water and increasing recharge to groundwater.
Randomization Method
For the randomization process, we randomly sampled habitations into treatment and control groups using a stratified (by our six strata, defined by state and presence or lack of presence of surface irrigation infrastructure) randomization procedure designed to achieve two-thirds of the sample being assigned to treatment and one-third to control. In particular, we used the “randtreat” command in Stata, while indicating our use of these stratifying variables and our desire for a 2/3 – 1/3 ratio of treatment to control units. Within each state – water infrastructure status (yes/no) combination (i.e., strata), non-sampled habitations were randomly ranked such that they could serve as possible replacements. As needs for replacement habitations emerged, the non-sampled habitation with the lowest rank within the same strata was used as a replacement and given the treatment status of the habitation it was replacing. Where no replacement habitations were available within the same strata, we took the lowest ranked replacement habitation that was from the same state. In this way, we identified 29 replacements.
Randomization Unit
Habitation (community)
Was the treatment clustered?
Yes

Experiment Characteristics

Sample size: planned number of clusters
472 habitations
Sample size: planned number of observations
2,360 individuals
Sample size (or number of clusters) by treatment arms
158 habitations in the control group and 314 habitations in the treatment group
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
We conducted power calculations using two different datasets, one at the habitation level and another one at the individual level. The information collected at the habitation level was part of the assessment of actual management practices of stop dam management of the Mandla District MP/India. This information was collected via two surveys at two different times: 1) baseline information collected between March and June 2017; and 2) endline information, collected during January 2019. We used these data to examine 22 variables providing information regarding status and maintenance of dams. We conducted the analysis to find the effect size (both in mean difference and standard deviation difference) for the study so that the power of a 5%-level two-sided test to detect the specified differences between means is at least 80% (or 90%). More information is shown in Table 6 of the attached pre-analysis plan. Additionally, we used the information at the individual level that was collected in the Mandla District MP/India between March and June 2017. We used these data to examine 3 dummy variables: a) if a player contributed labor to maintain the dam; b) if a player contributed money to maintain the dam; and c) if a player received water from the village dam. We estimated intra-cluster correlations (ICCs) (where a cluster is defined as a habitation), assessed the extent to which demographic characteristics explained the variance in the outcome, and determined the minimum detectable effect (MDE) size given our sample size of 472 habitations and 5 individuals in each (we have 6 individuals, 3 women and 3 men, but were aware of potential refusals to do an interview/other forms of attrition at endline). We then calculated the ICC for each variable. For (a), the ICC was 0.34; taking 80% power, a 0.05 significance level, 133 control clusters, 267 treated clusters, 5 individuals per cluster, and without baseline controls, our MDE size was 0.088 standard deviations, for this binary dependent variable. For (b), the ICC was 0.535; with the same assumptions, the MDE was 0.217. For (c), the ICC was 0.383; with the same assumptions, the MDE was 0.195. More information is shown in Table 7. We consider these MDEs reasonable (even if not particularly small).
IRB

Institutional Review Boards (IRBs)

IRB Name
International Food Policy Research Institute
IRB Approval Date
2021-08-31
IRB Approval Number
EPTD-21-0830
Analysis Plan

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Post-Trial

Post Trial Information

Study Withdrawal

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Intervention

Is the intervention completed?
No
Data Collection Complete
Data Publication

Data Publication

Is public data available?
No

Program Files

Program Files
Reports, Papers & Other Materials

Relevant Paper(s)

Reports & Other Materials