Community-based Skills Training, Employment and Social Transformation: Evidence from a Field Experiment in Northern Nigeria

This study will evaluate the impact of a unique vocational training program delivered through Community Skills Development Centers (COSDECs) to over 900 young men and women. The program is unique in targeting highly marginalized individuals, including many Almajiri. The intervention utilized a classroom-based approach in which fully equipped workshops and production facilities were used to train students in select trades.

2017-01-01

2018-03-31

1. Income generating activities (in past 30 days)
a. Worked in wage employment (t3>0)
b. Worked self-employed (u3>0)
c. Hours worked in wage employment (t4)
d. Hours worked in self-employment (u4)
e. Total income from wage employment (t12)
f. Total profits from self-employment (u9a)

2. Participation in political and religious violence. For our main analysis of this outcome, we will use responses from a self-completed audio module. We will also do a confirmatory analysis using responses to the same questions elicited in a list experiment (see Section 4.5 for details).
a. Used force or violence for a political cause (r13)
b. Participated in a group that sought to administer justice to someone for behaving badly/immorally (r12)
c. Used force or violence for reasons related to your religion (r9)
d. Participated in any riot, protest or demonstration that turned violent (r10)

3. Attitudes toward religious adherence and religious enforcement. We will test the primary hypothesis for this outcome only for the subsample of Muslim respondents, since the religious laws measured the vignette questions only apply to Muslims. We will also report effects for Christians as part of the heterogeneity analysis.
a. Believes it’s important that government ensures people follow religious laws (e6)
b. Vignettes on enforcing violations of religious rules. (For calculation of the index, the outcome will be defined as choosing either “report to authorities” or “beat up” as the answer. In additional analysis, we will also report effects of treatment on the probability of choosing each individual answer)
i. Willing to punish alcohol consumption (e91)
ii. Willing to punish miniskirt wearing (e101)
iii. Willing to punish blasphemy (e111)
4. Female empowerment
a. Desired/actual age of marriage / fertility (bc2, b2c, b2d, y32)
b. Experienced domestic violence (o21-o215)
c. In a marriage, who do you think should have the greater say in each of the following decisions (outcome defined as probability that decision is taken by wife or both partners equally) (y1-y10)
d. Who decides how income will be used in household (y11)
e. Would like daughters to work outside the home (y12)
f. Various questions on women’s role in society (y14-y21)
g. OK for a husband to beat his wife under various circumstances (y22-y28)
h. OK for a wife to have her own opinion, even if it is different from her husband’s (y29)
i. Would like to be working in 5 years’ time (ha1)
j. Confidence in ability to solve Raven’s matrices (B11)

5. Caregiver female empowerment. In the primary hypothesis we will examine the sample of caregivers who have female wards. Treatment will be defined based on whether the caregiver has any female ward enrolled in Mafita. We will also conduct heterogeneity analysis with boys, to determine if the response of caregivers is significantly different for caregivers who have male wards enrolled in (as compared to those who have female wards enrolled in Mafita).
a. Ok for unmarried and married women of ages 14, 18 and 22 to be earning an income inside home and outside home (b2a-e)
b. Age at which women should get married (b3_1); Age at which women should have their first child (b3_3)
c. Various questions on what women should be allowed to do in society and what boys and girls should have to do and have prioritized toward them (b5_1 – b5_8)
d. Various questions on equality in household decision making (b6_1-b6_10)

6. Anti-social behavior. For our main analysis of this outcome, we will use responses from a self-completed audio module. We will also do a confirmatory analysis using responses to the same questions elicited in a list experiment (see Section 4.5 for details).
a. Used tramadol in past 3 months (r2)
b. Used kayan maye in past 3 months (r1)
c. Stolen objects or money from someone, when that person was not present (r5)
d. Stolen objects or money from someone, when that person was present (r6)
e. Gotten into a fight where I tried to physically hurt someone (r3)
f. Did some work for a criminal group (d’an ta’adda) in the past 6 months (r14)
7. Generosity toward other religions: donation experiment. See section 4.4 for details.

For the directly measured outcomes (all outcomes except 7) we will combine all survey responses related to an outcome into a mean effects index, following Kling et al. (2007). To do this, we will first express responses in terms of standard deviations from the control group mean. We will then sum all standardized responses related to an outcome into an index switching signs if necessary to ensure that the positive direction always indicates a “better” outcome. We will also report a robustness test using the method of Anderson (2008), which weights the index items by their inverse covariance matrix. We will also present estimates of individual indicators within each family to better gauge how various indicators contribute to overall effects within families.

8. Attitudes towards political violence
a. Use of violence is justified in protesting an injustice (I3)
b. It is good to use violence to resolve problems (I5)
c. It is ok to use force or violence for political cause (I6)

9. Religious social network (for the purpose of this outcome, a respondent’s social network is defined as the five people he/she spends the most time with)
a. Number of people from other religion in social network (e22)
b. Trusts people from other religions (e45)
c. Trusts leaders from other religions (e57)

10. Religiosity
a. Time spent on religious activities (c22c)
b. Importance of religion in respondent’s life (e35)
c. Other things are more important than religion (r16)
d. Often has a strong sense of God’s presence (h211)

11. Subjective well-being
a. Cantril’s Ladder (h11)
b. MHI-5 (mhi1-mhi5)

12. Self-esteem
a. 7 item index (h212, h31-h39)

13. Skills
a. Literacy assessment (Ia1-Ia5)
b. Numeracy assessment (na1-na7)

14. Assets and consumption
a. 12-item module (Q1-Q7, Q11-Q12)
b. Money spent on various categories (d22a-d22c, d22f-d22k)
15. Job search behavior
a. In the past 6 months, did you actively look for a job? (v1)
b. How many months out of the past 6 months did you actively look for a job (v2)
c. In the past 30 days, did you actively look for a job? (v3)
d. How many days out of the past 30 days did you actively look for a job? (v4)
e. In the past 7 days, did you actively look for a job? (v5)
f. How many days of the past 7 days did you spend actively looking for a job? (v6)
g. In the past 6 months, did you try to start your own business (v7)?
h. In the past 30 days, did you try to start your own business (v8)?

16. Range of social networks (for the purpose of this outcome, a respondent’s social network is defined as the five people he/she spends the most time with)
a. Number of people in social network who are currently employed (e24)
b. Number of people in social network who live in same neighborhood as respondent (e26)

For the directly measured outcomes (all outcomes except 7) we will combine all survey responses related to an outcome into a mean effects index, following Kling et al. (2007). To do this, we will first express responses in terms of standard deviations from the control group mean. We will then sum all standardized responses related to an outcome into an index switching signs if necessary to ensure that the positive direction always indicates a “better” outcome. We will also report a robustness test using the method of Anderson (2008), which weights the index items by their inverse covariance matrix. We will also present estimates of individual indicators within each family to better gauge how various indicators contribute to overall effects within families.

The study is a randomized control trial that targets a pool of 1,824 individuals, 900 of whom were randomly assigned into the treatment group while the remaining 924 were assigned to the control group.

See pre-analysis plan attached.

Randomization was done electronically in office by a computer using Stata.

Applicants were randomly assigned into a treatment group that received training starting in April 2017 and a control group that was excluded from training. The pool of applicants comprised 1,824 individuals, 900 of whom were randomly assigned into the treatment group while the remaining 924 were assigned to the control group.
When applying to the program, individuals were asked to specify a trade they wanted to be trained in by the COSDEC program. The Mafita program asked to make sure that no more than 30 individuals per center would be assigned to treatment in the same trade. To ensure this, we stratified the randomization by the intersection of center (the six COSDEC centers in Daura, Funtua, Katsina, Kagoro, Mando and Sabon Gari) and chosen trade (Brick-laying, Carpentry & Joinery, Electrical installation, Fashion Design, Hospitality, Office Management, Welding & fabrication).

Unit of randomization is at the individual level.

No

1824 individuals

1824 individuals

Treatment: 900 Individuals
Control: 924 Individuals

0.132 standard deviations.

Pre-analysis plan