Primary Outcomes (explanation)
Generalized communal trust: we will construct indicator variables for trusting several groups:
• “Other women in my community can be trusted” (Response Options: Strongly agree = 1, Somewhat agree = 2, Neither agree nor disagree = 3, Somewhat disagree = 4, Strongly disagree = 5).
• “Most people can be trusted”
• “I trust my neighbors to look after my house if I am away”
• “Now I’m going to ask you a question about your community - both farmers and herders. To
what extent can people in this area be trusted?”
Whenever a respondent has as the answer “refused to answer” or “don’t know,” we will treat this as missing data. We will create an equally weighted index of these measures, called our Generalized Communal Trust Index (for men, this will exclude the first question in the above list). All raw index components will be standardized by dividing by the standard deviation in the control group. In other words, we will take equally-weighted averages of the standardized index components. We then standardize the average by dividing by the standard deviation in the control group. When there are missing component measures, we will use imputation strategies to minimize the loss of observations.
Next considering bridging trust and social cohesion with out-group members, we will consider an additional set of variables. We will test whether treated women and men report stronger social cohesion with Fulani community members, using the following measures:
• “With regards to some from the Fulani ethnic group, would you feel comfortable if they worked in your field?”
• “With regards to some from the Fulani ethnic group, would you feel comfortable paying them to watch their animals?”
• “With regards to some from the Fulani ethnic group, would you feel comfortable trading goods
with them?”
• “With regards to some from the Fulani ethnic group, would you feel comfortable sharing a meal with them?”
• “With regards to some from the Fulani ethnic group, would you feel comfortable with a close relative marrying a person of Fulani descent?”
• “How much do you trust people from the Fulani ethnic group?”
Response options for each of the questions are as follows: Not at all = 1, A little = 2, Somewhat = 3, A lot = 4, A great deal = 5. We will construct an equally-weighted index of these measures, called our Out-Group Trust Index, as a primary summary outcome, with higher values indicating greater out-group social cohesion. Whenever a respondent has as the answer “refused to answer” or “don’t know,” we will treat this as missing data. Again, all raw index components will be standardized by dividing by the standard deviation in the control group. In other words, we will take equally-weighted averages of the standardized index components. We then standardize the average by dividing by the
standard deviation in the control group. When there are missing component measures, we will use imputation strategies to minimize the loss of observations.
Considering vertical (citizen-state) trust, for both men and women, we will construct indicator variables for trusting the state and authority figures.
• “How much do you trust: National assembly?”
• “How much do you trust: Your local government councilors?”
• “How much do you trust: Your traditional leader?”
• “To what extent do you respect the political institutions of Nigeria?”
• “To what extent do you think that citizens’ basic rights are well protected by the political system
of Nigeria?”
“To what extent do you think that women’s basic rights are well protected by the political system
of Nigeria?”
• “To what extent do you feel your leaders are doing the best job possible for Nigerians?”
Response options for each of these questions are as follows: Not at all = 1, A little = 2, Somewhat = 3, A lot = 4, A great deal = 5. We will construct an equally-weighted index of these measures, called our Political Trust Index, as a primary summary outcome, with higher values indicating greater vertical social cohesion. Whenever a respondent has as the answer “refused to answer” or “don’t know,” we will treat this as missing data. Again, all raw index components will be standardized by dividing by the standard deviation in the control group. In other words, we will take equally-weighted averages of the standardized index components. We then standardize the average by dividing by the standard
deviation in the control group. When there are missing component measures, we will use imputation strategies to minimize the loss of observations. We will also look at each component measure separately in case the trainings impact trust on specific actors and institutions and not others.
Conflict resilience capacity: the items of interest for this measure are as follows:
• “If I am affected by a farmer-herder conflict, I feel confident that I would know what to do”
• “If I am affected by a farmer-herder conflict, I feel women in my community would be allies that
would help me”
• “If I am affected by a farmer-herder conflict, I feel men in my community would be allies that
would help me”
• “If I am affected by a farmer-herder conflict, I feel my baale would help me”
Response options for each of these questions are as follows: Strongly agree = 1, Somewhat agree =
2, Somewhat disagree = 3, Strongly disagree = 4. We will construct an equally-weighted index of these measures, called our Conflict Resilience Capacity Index, with higher values indicating greater conflict resilience capacity. Whenever a respondent has as the answer “refused to answer” or “don’t know,” we will treat this as missing data. Again, all raw index components will be standardized by dividing by
the standard deviation in the control group. In other words, we will take equally-weighted averages of the standardized index components. We then standardize the average by dividing by the standard deviation in the control group. When there are missing component measures, we will use imputation strategies to minimize the loss of observations. We will also look at each component measure separately
in case the trainings impact trust on specific actors and institutions and not others.
Perceived effectiveness index:
The federal government, The state government, Male traditional and religious leaders, Female traditional and religious leaders, The media, Male-led civil society organizations, Female-led civil society organizations
Response options for each of these questions are as follows: Extremely effective = 1, Somewhat effective = 2, Not very effective = 3, Not at all effective = 4. We will reverse-code all outcomes such that a higher overall index score indicates greater perceived effectiveness. We will then construct an equally-weighted index of these measures, called our Perceived Effectiveness Index, with higher values indicating greater perceived effectiveness of a range of institutions. Whenever a respondent has as the answer “refused to answer” or “don’t know,” we will treat this as missing data. Again, all raw index components will be standardized by dividing by the standard deviation in the control group. In
other words, we will take equally-weighted averages of the standardized index components. We then standardize the average by dividing by the standard deviation in the control group. When there are missing component measures, we will use imputation strategies to minimize the loss of observations. We will also use principal component analysis to assess whether these measures load clearly onto two factors corresponding to the horizontal and vertical dimensions of social cohesion, which would argue for constructing two separate indices. We do not pre-assign survey measures to one dimension or the other at this stage, as the mapping of each survey item is not straightforward in the Nigerian context—for instance, it is unclear whether media is best understood as a power-wielding institution more characteristic of vertical social cohesion, or as a check on authority more characteristic of the horizontal dimension.
Freedom of movement: We will construct an equally-weighted index of the following indicator variables (response options are Yes = 1, and No = 0), for both women and men, called our Freedom of Movement and Activity Index. Each index component will be reverse-coded so that higher values reflect greater freedom of movement and activity (coded as 0 for “don’t know” responses, as uncertainty about whether insecurity affected one’s activities most plausibly reflects an absence of meaningful constraint; and coded as missing for refusals):
• In the last two years, were there any areas that you avoided going to or through because of insecurity during the NIGHT?
• In the last two years, were there any areas that you avoided going to or through because of insecurity during the DAY?
• In the last two years, did insecurity ever prevent you from: Working when you wanted to work?
• In the last two years, did insecurity ever prevent you from: Going to the market?
• In the last two years, did insecurity ever prevent you from: Getting water to the household?
• In the last two years, did insecurity ever prevent you from: Going to your field/farm?
In the last two years, did insecurity ever prevent you from: Moving your animals to grazing areas?
• In the last two years, did insecurity ever prevent you from: Moving your animals to water?
• In the last two years, did insecurity ever prevent you from: Earning money or going to work?
• In the last two years, did insecurity ever prevent you from: Going to school?
All reverse-coded index components will be standardized by dividing by the standard deviation in
the control group. In other words, we will take equally-weighted averages of the standardized index
components. We then standardize the average by dividing by the standard deviation in the control
group. When there are missing component measures, we will use imputation strategies to minimize
the loss of observations.