Building State Capacity and National Unity with Market Design: The Problem of Volunteer Assignment in Kenya’s G-United Program

Greatness United (G-United) is a national volunteering program implemented by the Government of Kenya and the Ministry of Education, in partnership with Evidence Action. The program recruits and trains recent university graduates to assist in low-performing public primary schools throughout the country to increase learning and build national unity. Applicants may be assigned to any county beside their home county, so the program is partly designed with the intention of promoting inter-ethnic contact. The program is meant to provide university graduates, who may not have good labor market opportunities, with a chance to develop new skills and build networks while, at the same time, improving government-provided educational services for young children.

In 2018, G-United assigned 1,227 volunteers to over 600 schools across 20 of Kenya’s 47 counties. In 2019, G-United will assign approximately 1,600 volunteers to 800 schools across 22 counties in Kenya. The program has increased capacity each year; at scale, it is intended to recruit, train, and deploy 10,000 volunteers annually, reaching 200,000 learners. The cost-effectiveness of this program is contingent upon improving volunteer retention at multiple stages of recruitment, and upon placement.

Like many civil service programs, G-United faces high rates of attrition, and qualitative findings suggest this dissatisfaction among volunteers may stem from the placements; prior to 2018, all volunteers were matched to counties via random assignment. This low retention increases recruitment and per-student costs, and limits the potential positive returns of the program. Attrition from G-United occurs at various stages: (1) applicants being assigned a school but subsequently not appearing for training and dropping out of the program as a result; and (2) applicants being assigned a school, initially working in that school, but then leaving the school during the school holidays and subsequently not returning. Others considering leaving the program may exert low effort, mirroring a common issue observed among public servants, especially in developing countries (e.g. Chaudhury et al., 2006).

2018-03-01

2020-03-01

Our evaluation will measure the following primary outcomes:
a. Volunteer retention in the program
b. Literacy and numeracy outcomes of primary students
c. Volunteer national unity attitudes and behaviors, specifically in terms of inter-ethnic prejudice, levels of identification (ethnicity vs. national) and national pride.
d. Host school and community national unity attitudes and behaviors
e. Job performance of volunteers in terms of volunteer productivity, and satisfaction with the G-United program on the part of school officials

Our hypothesis is that mismatch between volunteers and their assigned schools contributes to the high levels of attrition seen in the program. Accordingly, in the 2019 cohort, volunteer applicants have been randomized into two placement mechanisms:

(i) Deferred acceptance allocation (DA): In the first arm, a subset of applicants have been randomly selected to be assigned to counties through a deferred acceptance algorithm that takes volunteer preferences into account. This group completed a baseline survey to elicit their preferences among counties, excluding their home counties (county of ancestry and county of residence). These responses were then used to assign them to a placement county. Allocation of volunteers to schools within each county will be done randomly. Within each county, a randomly selected subset of eligible schools will receive two volunteers placed through the deferred acceptance algorithm. Both volunteers in each school are assigned in the same way.

(ii) Random allocation (RA): In the second arm, a subset of applicants have been assigned randomly to county placements, as had been done in previous years. This group completed the same baseline survey of preferences. These volunteers were then randomly assigned to a county, excluding their home counties (county of ancestry and county of residence). Allocation of volunteers to schools within each county will be random. Within each county, a randomly selected subset of schools will receive two volunteers placed through the random allocation mechanism. Both volunteers in each school are assigned in the same way.

The only differences between the two groups is in the determination of their assigned post. This study will be the first to compare deferred acceptance to random assignment (RA) using experimental variation.

In addition to comparing retention, job performance, and measures of national unity under the random assignment procedure to those under deferred acceptance, we will also compare a subset of outcomes of those assigned through either mechanism to a pure control group. A randomly selected subset of applicants will not be assigned at all, but will instead be placed in a control group, allowing examination of the effect of program participation on applicants’ interethnic prejudice, levels of identification (ethnic vs. national) and national pride.

In addition, we will assess the impact of the G-United program and of volunteer effort on student learning outcomes, by collecting data on the literacy outcomes of students taught by volunteers allocated by DA, those students taught by volunteers allocated by RA, and those students in pure control schools.

We will also focus on host school and community national unity measures by collecting data on inter-ethnic prejudice, levels of identification (ethnicity vs. national), and national pride.

All randomization was and will be done in office by a computer.

(i) Individual volunteers; and (ii) Public schools to be assigned these volunteers

Yes

Approximately 1,600 schools

i. 1,600 public schools ii. 2,400 volunteers iii. 9,600 students (at an average of 6 students sampled per school across 1,600 schools) iv. 800 public school head teachers, whose schools were assigned a volunteer

i. 800 treatment schools and 800 control schools
ii. 800 volunteers assigned to the deferred acceptance mechanism, 800 volunteers assigned randomly, and 800 volunteers as pure control
iii. 4,800 treatment students and 4,800 control students (at an average of 6 students sampled per school across 800 treatment and 800 control schools)