https://doi.org/10.5061/dryad.kh18932f7

In our RCT, the unit of analysis are the households. Treatment consists of seminars delivered to household members, in order to induce behavior change in maternal and child health practices, attitudes, and knowledge. In the RCT, there are approximately 24,000, living in 10,000 households, across 176 villages. We employed a 2x8 experimental design: first, assigning villages to 16 treatment arms, and then selecting households within each village to receive treatment. Treatment arms each contain 11 villages, and are defined by a combination of two aspects of treatment: (i) whether households were selected using random-nomination targeting or friendship-nomination targeting, and (ii) the fraction of the households to be selected for treatment in each village (namely, 0%, 5%, 10%, 20%, 30%, 50%, 75%, and 100%). Responses were collected for many outcomes, using a survey instrument, delivered to individuals.

All the Figures in the paper are the results of analyses on household-level data, and the results are then aggregated over villages within each treatment arm. To preserve anonymity and confidentiality of the study participants, here, we publish data aggregated over villages within each treatment arm. At this level of granularity, there are 11 villages in each of 16 treatment arms. This data release and code allow to replicate, for instance, panels in Figure 2 of our study, which display dose-response curves for selected individual outcomes, and for subgroups of outcomes.

Description of the data and file structure

There are two main data files in excel format.

The first file includes data for subgroups of outcomes: this is file dataABC.xlsx. There are three data sheets in that excel file, corresponding to subgroups of interest; namely;

• the “all outcomes” data sheet, which provides data on all the outcomes
• the “sig_direct” data sheet, which provides data on the subgroup of outcomes for which the direct effect is significant, and
• the “” data sheet, which provides data on the subgroup of outcomes for which the spillover effect is significant.

In each sheet, there are four rows and eight columns of data, corresponding to the 2x8 design described above. More specifically, each data point represents the average behavioral adoption over a set of outcomes measured on households located in villages in a specific treatment arm, and we also provide the standard deviation of the behavioral adoption. So two rows for the data from villages where the households where targeted using friendship-nomination and random-nomination, plus two rows for the corresponding standard deviations, for a total of four rows, and eight column corresponding to the fraction of households targeted in each village  (0%, 5%, 10%, 20%, 30%, 50%, 75%, and 100%). The column are labeled, but the rows are not. The row labels are as follows: friendship-nomination data in row 1, friendship-nomination standard deviations in row 2, random-nomination data in row 3, and random-nomination standard deviations in row 4.

The R code in the file plotABC.R will take this file as input and generate panels ABC in Figure 2, in the paper.

The second file includes data for individual outcomes: this is file dataDEF.xlsx. There are five data sheets in that excel file, corresponding to data for the different treatment arms and additional information for plotting; namely,

• the “outcomes” data sheet has three rows and four columns. Each row corresponds to an individual outcome, and the columns, in order, correspond to the outcome index in the master list of outcomes, the variable name for the outcome, the name of the outcome used in the plots, and the number of households for which said outcome was collected
• the “fnt.dose.vdata” data sheet, provides average behavioral adoption for three individual outcomes on the three rows. Each data point represents the average behavioral adoption over households located in the set of villages where friendship nomination was used to target households. The columns denote the fraction of households that were targeted in these villages.
• the “fnt.dose.sd” data sheet, has the same structure of “fnt.dose.vdata” data sheet, and provides standard deviations for each of the corresponding averages in the “fnt.dose.vdata” data sheet.
• the “rnt.dose.vdata” data sheet, which provides average behavioral adoption for three individual outcomes on the three rows. Each data point represents the average behavioral adoption over households located in the set of villages where random nomination was used to target households. The columns denote the fraction of households that were targeted in these villages.
• the “rnt.dose.sd” data sheet, has the same structure of “rnt.dose.vdata” data sheet, and provides standard deviations for each of the corresponding averages in the “rnt.dose.vdata” data sheet.

The R code in the file plotDEF.R will take this file as input and generate panels DEF in Figure 2, in the paper.

Sharing/Access information

We are committed to preserving the anonymity and confidentiality of the study participants and their behaviors. We are also committed to the reproducibility of scientific results, and to offering other scholars the ability to advance science using the data we collected as part of this study.

Other village-level data reported in this paper will be made available on request to the corresponding author from established investigators in accordance with then-existing data-sharing protocols at the Yale Institute for Network Science and Yale University.

Code/Software

The provided files plotABC.R and plotDEF.R will help replicate the results in Figure 2, in the related paper.