Targeting High-Potential Youth Entrepreneurs: A Head-to-Head Comparison of Alternative Approaches in Ethiopia
How to cite
Abay, Kibrom; Nigus, Halefom; Abate, Gashaw; Gebrekidan, Bisrat; Chamberlin, Jordan; Mockshell, Jonathan. 2026. "Targeting High-Potential Youth Entrepreneurs: A Head-to-Head Comparison of Alternative Approaches in Ethiopia." AEA RCT Registry. AEARCTR-0018883. https://doi.org/10.1257/rct.18883-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
While many governments and development partners continue to promote entrepreneurship to facilitate job creation for the youth, identifying high-ability entrepreneurs remains challenging for many reasons. Youth entrepreneurship programs in low-income countries, where program implementers badly need to allocate limit capital and credit to where they are needed most or impactful, are particularly affected by this challenge. The substantial heterogeneity in returns to capital among small enterprises imply that who receives grants can matter as much as whether grants are provided at all. Two emerging and promising methods being used to identify high-ability entrepreneurs include: community-based targeting (Hussam et al., 2024) and data-driven (machine learning) methods (McKenzie and Sansone, 2019; Bryan et al., 2024). This project compares these two emerging methods in an experimental setting involving actual selection of high-ability youth entrepreneurs. While these two targeting methods have each been validated in prior research, they have never been placed in head-to-head competition. We implement both methods simultaneously in an experiment involving distribution of modest seed capital for youth in Ethiopia. Our findings will speak directly to the question of which approach is more effective, scalable, and for identifying “high-ability” entrepreneurs. Beyond comparing returns associated with the seed capital, we also compare whether either of these targeting methods affect trust and social cohesion among peers. To address these research questions, we exogenously vary the targeting methods across established youth groups in Ethiopia.
Intervention
This intervention involves transfer of modest seed capital to rural youth in Ethiopia. This experiment and intervention builds on a sample of Youth Agricultural Innovation Groups (YAIGs) in Oromia region of Ethiopia. These YAIGs were operating and being supported by the SAfA (Foundations Alliance for Africa) project implemented by the Hanns R. Neumann Stiftung (HRNS). The SAfA project aimed to enhance the livelihood prospects of young people in Ethiopia through an integrated and multi-sectoral approach. The purpose of the intervention is to test two methods for identifying high-ability youth entrepreneurs. We test this by providing a one-time cash grant to support entrepreneurship and business creation. For this purpose, we work with youth aged 18–35 who are organized across 183 YAIGs. These YAIGs have been functioning for several years and are well-suited to compare both targeting methods.
YAIG leaders and members will participate in the targeting of the seed capital to high-ability youth. YAIG leaders and members will be informed that the NGO conducting the survey aims to identify a small group of youth to support them with seed capital to start or expand a business or support livelihood activities. The transfer of the seed capital follows a two-stage selection process. First, beneficiary YAIG will be randomly selected. Next, within each selected YAIG, the NGO we are working with will transfer to the seed capital to the four highest-ranked youth members as determined by the committee members. If a YAIG is selected, the NGO will award 20,000 Birr seed capital to each of the four youth members with the highest potential within each YAIG. For this purpose, we implement the following targeting interventions:
Community (peer) ranking: All YAIG members will participate in peer ranking sessions. Each participant privately ranks all eligible youth by perceived entrepreneurial ability. This follows Hussam et al. (2022). The core motivation and rationale builds on the fact that community members hold accurate and accurate information about who in their neighborhood is a high-ability entrepreneur — an important information that program implementers cannot easily access. However, community-based targeting is prone to elite capture and favoritism. The participants include both YAIG leaders as well as YAIG members. Specifically, the YAIG members asked to prioritize and rank the YAIG members based on the following criteria: (i) Youth who may have potential entrepreneurial qualities to succeed in business compared to other peers in the same group; (ii) Youth who has or may have business ideas or are in the process of starting a business; (iii) Youth who know or have experience running a business, and (iv) Youth who have the potential to turn a seed capital into in a profitable business with high potential returns. We aim to implement alternative aggregation methods to generate weighted ranking (including different weights to ranking from leaders and self-ranking) and those top-four ranked youth within each YAIG receive the seed capital to support, establish, expand or strengthen their business.
Data-driven (machine learning) targeting: With the objective of implementing data-driven targeting we also administer comprehensive baseline survey to capture detailed psychometric indicators (including measures of big five personality, self-efficacy, grit and locus of control) as well as other measures of entrepreneurial tendencies and talents. For this purpose, we aim to administer these psychometric (big-five personality, self-effectiveness, locus of control) modules and entrepreneurial skill measures among all members of the 183 YAIGs in our baseline survey. The entrepreneurial modules include: Measure of Entrepreneurial Tendencies and Abilities (Meta) as well as Abridged Entrepreneurial Attitude Orientation (EAO) Scale. These attributes will be utilized for the predictive-targeting approach of McKenzie and Sansone (2019) and Bryan et al. (2024), which allows us to construct a data-driven score that predicts a youth's entrepreneurial performance using a supervised machine learning model trained using the baseline data.
Research Objectives and Research Questions
This research project aims to address four related research questions. First, we aim to provide comparative evidence on the relative performance of community versus data-driven targeting methods for identifying high-ability entrepreneurs. This builds on previous evidence and successful application of these methods in India (Hussam et al., 2022), Nigeria (McKenzie and Sansone, 2019) and Egypt (Bryan et al., 2024). Second, we investigate whether these methods are prone to elite capture and favoritism by comparing the observable characteristics of youth identified through these two methods. Third, we also aim to investigate whether these targeting methods as well as associated seed capital transfer affect business creation, youth labor market outcomes and associated welfare. Finally, we compare the perceived legitimacy associated with these targeting methods as well as their impact trust and social cohesion among peers. For example, data-driven approaches such as PMT based targeting methods are sometimes considered as lacking transparency (Hanna and Olken, 2018), an important element that can affect trust and social cohesion among community members. More specifically, we aim to address the following research questions:
1) Which targeting methods is more effective at identifying high-ability micro-entrepreneurs?
2) Do these targeting methods vary in terms of the composition of youth identified and which of them are more prone to elite capture and favoritism?
3) What are the impacts of the seed capital and associated targeting methods on youth labor market outcomes and associated welfare?
4) What happens to those not selected by each targeting method and which targeting method is perceived as relatively more legitimate by participants and do these targeting methods affect trust and social cohesion?
Experimental design
The target group for this experiment are YAIG and their members, rural youth members organized for conducting joint farm and off-farm activities. These YAIGs have been operating for the last five years and they were being supported by the SAfA (Foundations Alliance for Africa) project implemented by the Hanns R. Neumann Stiftung (HRNS). These groups have been working together closely, some of which have joint ventures and businesses, which allow them to know each other. This offers a unique opportunity to test the potential of community-based targeting to identify high-ability youth entrepreneurs.
Our intervention includes targeting processing to identify high-ability entrepreneurs, which ultimately leads to transfer of modest seed capital. The main purpose of the intervention is to test two methods for identifying high-ability youth entrepreneurs while also quantifying the impact of the seed capital on a number of labor market and entrepreneurial outcomes. Our sample includes youth aged 18–35 who are organized across 183 YAIGs, with an average size of 15 members. The targeting process and ultimate transfer of the seed capital follows a three-stage selection process. First, we administer a comprehensive baseline survey that involves community-based (peer) ranking while also capturing comprehensive list of psychometric indicators and entrepreneurial ability measures. The community-based peer rankings are used to identify and rank youth based on perceived entrepreneurial ability and following the criteria given above. Similarly, we aim to use the psychometric and measures of entrepreneurial ability to identify and rank youth according to their predicted entrepreneurial ability and using data-driven (machine learning) methods. Second, YAIGs will be randomly assigned into three equal groups, which vary in terms of the targeting process for selecting the ultimate beneficiaries of the seed capital. Below we describe the three arms and the targeting process followed (see Figure 1). In the final stage, four youth from each of the YAIG selected in the treatment group will receive 20,000 Birr seed capital to support business creation or career progression.
Arm 1 (Community Targeting). YAIGs assigned to this group, a third of the YAIGs, will follow community-based targeting. Each participant privately ranks all eligible youth by perceived entrepreneurial ability. This follows Hussam et al. (2022). The core motivation and rationale builds on the fact that community members hold accurate information about who in their neighborhood is a high-ability entrepreneur — an important information that program implementers cannot easily access. We aim to implement alternative aggregation methods to generate weighted ranking and those top-four ranked youth within each YAIG receive the seed capital to support, establish, expand or strengthen their business. We particularly aim to test alternative weighing of rankings by members, leaders and self-ranks. To facilitate this, we also elicit relationships (and acquaintances) between peers, those ranking and being ranked.
Data-driven (machine learning) method: In another third of the YAIGs we follow a data-driven targeting method building on machine learning methods proposed by McKenzie and Sansone (2019) and Bryan et al. (2024), using psychometric indicators (including measures of big five personality, self-efficacy, grit and locus of control) as well as other measures of entrepreneurial tendencies and talents. This approach follows the predictive-targeting approach of McKenzie and Sansone (2019) and Bryan et al. (2024), through which construct a data-driven score that predicts a youth's entrepreneurial performance using a supervised machine learning model trained entirely on baseline data. The procedure is as follows. We define a baseline performance label as a standardized composite of current entrepreneurial and labor-market outcomes (business ownership, revenue, profit, and income), constructed for every youth. We then train a cross-validated, regularized supervised model that predicts this performance label from the full set of psychometric and enterprising ability and tendency indicators, along with basic demographic and business-experience covariates. We use k-fold cross-validation so that each youth's score is an out-of-fold prediction, eliminating overfitting and in-sample leakage. The estimated model yields a single predicted entrepreneurial-performance score for each youth; within each data-driven YAIG. Similar to the community-based ranking, this method will be utilized to rank youth based on their predicted entrepreneurial potential. Like the first arm, those top-four youth with the highest score within each YAIG receive the seed capital to support, establish, expand or strengthen their business.
Pure Control: A third of the YAIGs will remain as control group and no one in these YAIGs will receive seed capital. However, YAIG leaders and members will administer the community and peer-ranking exercise in the baseline survey. Similarly, we will administer the comprehensive modules used for capturing psychometric and entrepreneurial skills data, which will be used to construct comparable control groups for each of the targeting methods described above. We note that in the baseline survey, all three groups of YAIGs conduct similar surveys and ranking exercises. YAIG members in control clusters are enumerated and surveyed at baseline and endline but receive no grant.
Design details
Cited by
No citation data available for this registration yet.