NEW UPDATE: Completed trials may now upload and register supplementary documents (e.g. null results reports, populated pre-analysis plans, or post-trial results reports) in the Post Trial section under Reports, Papers, & Other Materials.
Changing Harmful Norms through Information and Coordination: Experimental Evidence from Somalia
Initial registration date
September 19, 2019
April 27, 2020 6:20 AM EDT
Other Primary Investigator(s)
University of Queensland
Additional Trial Information
FGC is the practice of cutting or removing part of the external or internal female genitalia for non-medical reasons. FGC is a harmful practice along several dimensions. First, it leads to serious health consequences, both at the time of cutting (e.g., excessive bleeding and increased mortality) and in the long run (e.g., birth-related complications). Second, FGC is associated with lower educational achievement of girls because girls are often pulled out of school after the practice, either due to health complications or because they are seen as having “transitioned” to adulthood. Third, given that FGC is generally performed on young girls without their consent, it is a human rights violation.
There can be several explanations for why a harmful practice, such as FGC, may continue being practised. In this project, we aim to test for two of these explanations. First, FGC is often viewed as the result of a coordination failure. In a setting where individual choices are interdependent, multiple equilibria may exist and one reason for the persistence of bad norms can be a failure to coordinate to the “better” equilibrium. Individuals may know that others do not like FGC but no single individual may want to move unilaterally due to fear of social sanctions. To the extent that individuals worry about social stigma, they may even be hesitant to reveal their private views on FGC to others. Second, individuals may privately dislike FGC but think that others do and therefore they continue doing it. Social psychologists describe as pluralistic ignorance a setting in which a majority of the group privately want to change their behavior, but mistakenly think that the majority of the other group members prefer to keep the existing behavior. In this project, we aim to test to what extent these two potential reasons may explain why FGC persists in Somalia. In particular, our experiment entails two key components: one directed at providing information (hence correcting misperceptions) by enabling the truthful revelation of attitudes towards FGC; and one directed at providing a coordination device that will facilitate collective action among those who privately support the abandonment of FGM. Registration Citation
Gulesci, Selim et al. 2020. "Changing Harmful Norms through Information and Coordination: Experimental Evidence from Somalia." AEA RCT Registry. April 27.
Intervention Start Date
Intervention End Date
Primary Outcomes (end points)
1. Attitudes towards FGC
2. Prevalence of FGC
Primary Outcomes (explanation)
Our outcomes of interest can be grouped into two categories:
1. Attitudes towards FGC: we will elicit respondents’ own attitudes and their perceptions of community members’ attitudes;
2. Prevalence of FGC: this outcome will be measured in 3 ways. (i) Self-reports: we will ask female respondents about the history of FGC in their family and about which of their daughters are cut (or they intend to cut). (ii) Indirect revelation through symptoms: in the survey we will collect information on whether daughters manifest health symptoms most commonly associated with FGC and use these as a proxy. (iii) Indirect revelation using “item lists” techniques.
Secondary Outcomes (end points)
Perceived costs and benefits" of FGC
Secondary Outcomes (explanation)
We randomize the communities in our sample into four groups in order to test the hypotheses explained above. Below we summarize the different treatment groups:
Control: Poll only.
Treatment 1: Poll + Announcement of the result of the poll.
Treatment 2: Poll + Coordination exercise.
Treatment 3: Poll + Announcement of the result of the poll + Coordination exercise.
To identify the effects of the intervention, we will compare the outcomes of interest between treatment and control samples at follow-up survey(s). Due to the randomized design, these differences can be interpreted as causal impacts. In particular, by comparing the outcomes of interest in T1 vs C, we will test if the truthful revelation of community preferences affects individuals’ outcomes; this constitutes a direct test of the pluralistic ignorance mechanism for the continuation of FGC. The comparison of the outcomes of interest in T2 vs C allows us to shed light on whether the coordination mechanism alone can be an effective way for community members to coordinate to a new equilibrium. By comparing outcomes in T3 vs C relative to T1 vs C, we will test whether there are synergies between information provision and the availability of a coordination mechanism. Community members may not be able to shift to a new equilibrium even in the presence of new information. The experimental design allows us to test the joint effect of these mechanisms by comparing outcomes in T3 vs C relative to T1 vs C. The effect that the revelation of communities’ true preferences will have on individuals is likely to differ depending on the ex-ante expectations of each subject regarding the share of community members in favor of FGC. In a Bayesian updating framework, individuals whose prior was lower than the realized share should revise their expectations upwards, and vice versa. We will be able to test for this effect using our baseline survey that elicits individuals’ beliefs regarding the attitudes of others and by testing the heterogeneity of the impact along this dimension.
Experimental Design Details
To test whether information and/or coordination failures explain the continuation of FGC, we organize community meetings in 141 communities located in Somaliland, Puntland and South-central regions of Somalia. Meetings are organized separately for men and women in each community. During the meetings, participants are asked to state (using anonymous “voting” cards) if they favor the discontinuation of infibulation. Before the poll is taken, participants are incentivized (individually) to estimate what fraction of the group members will respond positively. Accurate guessing of the share in support is incentivized by telling participants that those whose estimates are closest to the true rate will receive a payoff (in the form of mobile money).
Communities are then divided into 4 groups. The first consists of 35 communities and is the "Control" group. In this group the only thing that happens is the anonymous poll, and nothing is revealed publicly.
A second group, consisting of 36 communities, receives an "information" treatment. Here a facilitator announces the result of the poll, thus revealing the “true” share of community members who are genuinely in favor of FGC. This revelation should correct individuals’ misperceptions of the attitudes in their communities, thus addressing the pluralistic ignorance mechanism.
A third group of 34 communities receives a "coordination" treatment. Here, after the poll is conducted, facilitators implement another exercise that enables the participants to coordinate their actions in support of the abolition of FGC. The facilitators draw a circle on the ground and ask people who believe FGC should be abandoned to step inside the circle, while those who believe FGC should be continued remain outside it. This should help individuals who favor the discontinuation of FGC but do not want to move unilaterally (for fear of social sanctions) to coordinate to their preferred equilibrium more easily by lowering the cost of communication and coordination. Thus, it addresses the coordination failure mechanism.
Finally, in a fourth group consisting of 36 communities, participants receive both the information and the coordination treatment.
Randomization is done in office by a computer.
Was the treatment clustered?
Sample size: planned number of clusters
We will survey a representative sample of community members who participated in the meetings in 141 communities.
Sample size: planned number of observations
The sample includes 4230 households, equally split by gender (15 men and 15 women in each community).
Sample size (or number of clusters) by treatment arms
Control (35 communities): Poll only.
Treatment 1 (36 communities): Poll + Announcement of the result of the poll.
Treatment 2 (34 communities): Poll + Coordination exercise.
Treatment 3 (36 communities): Poll + Announcement of the result of the poll + Coordination exercise.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
INSTITUTIONAL REVIEW BOARDS (IRBs)
Ethics Committee of Università Commerciale Luigi Bocconi
IRB Approval Date
IRB Approval Number
70131-3 and 70131-7
Post Trial Information
Is the intervention completed?
Is data collection complete?