close× Call Us
close×
Title Preventing Violent Extremism - Evidence from a Youth Job RCT Study in Somalia
Post date 10/29/2018
C1 Background and Explanation of Rationale What are the determinants of civilian support for international forces in the fight against violent extremism? Of the many non lethal instruments available, “providing employment to youths” has been the buzz word to dissuade idle youths from supporting violent extremist groups.1 Since research, regarding the widely presumed causal relationship between jobs and support to extremism, so far is still inconclusive, the primary aim of this study is to address this with a randomized control trial (RCT) designed job employment program to present a convincing experimental evidence. In November 2017, the UN funded a stabilization project, in support of the African Union Peacekeeping troops (AMISOM) across 38 districts in South Central Somalia by hiring Community Liaison Officers (CLOs) to serve as the contact and resource persons for the UN and AMISOM and were also tasked to help raise awareness and understanding of the root causes of violent extremism in the rural communities. I seize this opportunity to design a pairwise RCT where we randomly selected 48 individuals, in a public lottery for our treatment group for a job with the UN till June 2021, while 55 individuals in our control group, did not receive a job.
C2 What are the hypotheses to be tested?

→Primary Outcomes
1. Ho/Ha: No impact (positive impact) of getting a job on attitudes towards violent extremism. → Variables to be used: from phone survey- part four (Q. 11.1, 11.3, 11.4, 11.6, 11.7, 11.9),31 and part six (Q. 13.4, 13.5); from baseline survey- ( Q. 38 and 39). 1.1. H0/Ha: No impact (positive impact) of getting a job on behavior towards violent extremism. → Variables to be used: from phone survey- part ten (Q. 17, 20, 21, 23 and 24)

2. Ho/Ha: No impact (positive impact) of getting a job on attitudes towards Western backed efforts. → Variables to be used: from phone survey- part two (Q. 9.1-9.6). Three questions from part two will also be used to test if subjects support Al-Shabaab (Q. 9.1,9.3 and 9.4).

3. Ho/Ha: No impact (positive impact) of getting a job on attitudes towards democracy, Western liberal values and governance. → Variables to be used: from phone survey- part eight (democracy and liberal values, Q. 15.1-15.7) and part five (governance, Q. 12.1-12.5); from baseline survey- (democracy and liberal values, Q. 43.1, 43.2, 43.6, 48.3, 49.3, 49.4) and (governance, Q. 37.3, 42.1, 42.3, 42.4, 42.5)

3.1. Ho/Ha: No impact (positive impact) of getting a job on behavior towards Western liberal values in particular human rights, corruption and women rights. → Variables to be used: from phone survey- part ten (Q.18, 19 and 22) 5.1.2

→Secondary Outcomes
1. Ho/Ha: No impact (positive impact) of getting a job on attitudes towards social cohesion. → Variables to be used: from phone survey- part three (Q. 10.1-10.4).

2. Ho/Ha: No impact (positive impact) of getting a job on attitudes towards gender diversity. → Variables to be used: from phone survey- part nine (Q. 16.1-16.5); from baseline survey- (gender parity, Q. 37.4-37.6, 42.4 and 42.5)

3. Ho/Ha: No impact (positive impact) of getting a job on attitudes towards Islamic religiosity.32 → Variables to be used: from phone survey- part four (Q.11.2, 11.5, 11.8, 11.10. 11.11); part six (Q. 13.1-13.3) and part seven (Q. 14.1, 14.4); from baseline survey- (Q. 40.1, 41.4,

C3 How will these hypotheses be tested? *

(The original response to this field was updated and re-submitted on December 4, 2018. The updated response was submitted before researcher access to outcome data.)

→Analysis One: Difference between Treatment and Control

For our primary analyses, based on Imai, King and Stuart (2008), our measure of impact is an intent-to-treat (ITT) estimate, using the difference between treatment and control (evaluated over the experimental period) to test the above eight hypotheses. Using variables from our second follow up phone survey titled “September 2018 Phone Survey” (conducted 10 months post treatment in Sept 2018), we will run the following OLS regression:
Outcome idt = γ + β ∑ Xid + δT + µr + Єit

where ∑ Xid are control variables for each individual and his/her household characteristics, which will include, age, gender, marital status, clan, indices of dwelling quality and asset ownership. T is the treatment indicator, and µr is the randomization pair dummies since randomization was stratified by pairs (BruhnandMcKenzie,2009), and Єit is the error term. Impact of the project, the intention to treat (ITT) effect of winning the job lottery, is captured by the coefficient, δ.

→Analysis Two: Multiple Observations

After the third survey in September 2018 (first phone survey), we plan on conducting the same survey at least twice before the project closure in June 2021, which gives us four panel waves. Therefore, in addition to the first analysis stipulated above in section 6.1, I will carry out a second regression analysis where the variables from our phone survey, questions from part five to part nine (see annex-September 2018 phones survey), exactly match with those from the baseline, I will use two approaches and present results side by side.

→Analysis 2.1: ANCOVA Specification

Outcome idt+1 = γ + β ∑ Xidt + δTid + OutcomeBaseline idt + µr + Єit

For the subsequent second and third phone follow up survey data, the ANCOVA specification at time t=2,3,4 will be:

Outcome idt= γ + β ∑ Xid+1 + OutcomeBaseline idt=1 + δ2Tid*F t+1 + δ3Tid*F t+2
+δ4Tid*F t+3 + ∑ s=2,..4 σs1(s = t) + µr + + Єit

where Outcomeidt+1 is the relevant impact indicator in the first follow up phone survey in eq(2) and in round t in eq(3), OutcomeBaselineidt=1 is the outcome variable’s baseline value, ∑ Xid+1are control variables for each individual and his/her household characteristics, which will include, age, gender, marital status, clan, indices of dwelling quality and asset ownership. Tid is the treatment indicator,F t+1,2,3 in eq(3) are dummies taking the value one in follow up phone surveys in rounds 1-3, σs are survey round dummies, µr is the randomization pair dummies since randomization was stratified by pairs (Bruhn and McKenzie, 2009), and it is the error term. Impact of the project, the intention to treat (ITT) effect of winning the job lottery, is captured by the coefficient, δ by testing if δ, (ten months post winning the job lottery)= δ2 (20 months) = δ3 (30 months) to determine whether the impacts differ over time, and whether δ=δ2=δ3=0 to determine whether there is no impact over time.

→Analysis 2.2: Difference in Difference between Treatment and Control

In my second approach, as specified above and after the completion of the first phone survey (September 2018 phone survey), I will use difference in difference specification to evaluate effects of having a stable job for ten months by estimating the following equation for individual $i$ using OLS:

Outcome idt = γ + β ∑ X id + δT + θTF + πF + µr + Єit

where ∑ X idt are control variables for each individual and his/her household characteristics, which will include, age, gender, economic situation of the village, clan. T is the treatment indicator, F denotes the follow up observation as opposed to the baseline period, µr is the randomization pair dummies since randomization was stratified by pairs , and Єit is the error term.

After the end of the project in June 2021 when multiple follow up survey data is available, I will use the following:

Outcome idt = γ + β ∑ X id + δ1Ti + δ2TiF2t + δ3TiF3t + θ2F2t + θ3F3t + µr + Єit

where ∑ X idt are control variables for each individual and his/her household characteristics, which will include, age, gender, education, economic situation of the village and clan. T is the treatment indicator taking the value of 1 if subject i was won the job lottery, F2 and F3 are dummy variables for the second and third follow-up phone surveys, respectively. µr is the randomization pair dummies since randomization was stratified by pairs, and Єit is the error term.

C4 Country Somalia
C5 Scale (# of Units) 103 (55 control; 48 treatment)
C6 Was a power analysis conducted prior to data collection? No
C7 Has this research received Insitutional Review Board (IRB) or ethics committee approval? Yes
C8 IRB Number IRB FY 2018-1966
C9 Date of IRB Approval July 11, 2018
C10 Will the intervention be implemented by the researcher or a third party? The United Nations in Somalia and Ukroboronservice (https://en.uos.ua/)
C11 Did any of the research team receive remuneration from the implementing agency for taking part in this research? No
C12 If relevant, is there an advance agreement with the implementation group that all results can be published? Yes
C13 JEL Classification(s) C93, D71, D74, J62