Shaping aspirations among the poor through digital self-diagnosis and mentoring: Evidence from a randomized trial in Honduras

The study evaluates a digitally enabled “Poverty Stoplight” (PSL) intervention integrated into Unbound’s existing sponsorship program in Honduras. The PSL is a structured goal-setting tool that guides participating households through a self-assessment of their living conditions across multiple dimensions of poverty and supports them in identifying and prioritizing concrete, step-by-step goals.

Using a mobile or tablet-based platform, households assess their situation across 50 indicators covering six domains, such as education, health, housing, income, and financial security. Based on this self-diagnosis, participants identify areas of deprivation and set achievable goals to improve their conditions over time.

The study compares four versions of program delivery. A control group receives standard Unbound programming, which includes light-touch goal discussions and unstructured group meetings. Three treatment groups receive the PSL tool in addition to standard programming: one group uses the PSL without formal mentoring, one group receives individual mentoring to support goal-setting and follow-up, and one group participates in group-based mentoring sessions designed to encourage peer interaction, shared learning, and collective problem-solving. Mentoring sessions occur quarterly, with more frequent remote follow-ups.

The intervention lasts approximately 13 months and is delivered at the community level through Unbound’s existing network of social promoters.

2025-09-01

2026-10-01

The primary outcomes of interest capture changes in aspirations, individual agency, collective agency, and multidimensional poverty conditions.

Individual (Critical) Agency
Measures of participants’ perceived ability to set goals, make decisions, and take actions to improve their living conditions. This includes perceived control over life outcomes, confidence in achieving goals, and self-efficacy related to economic and social decisions.

Collective Agency
Measures of participants’ perceived collective efficacy and capacity for joint action, including trust in group processes, coordination with others, and beliefs about the ability of groups to solve shared problems and achieve common goals.

Aspirations
Aspirations across key life domains, including education, occupation and income generation, housing quality, and savings or financial security. Outcomes capture both the level and clarity of aspirations.

Multidimensional Poverty Indicators
Disaggregated indicators of deprivation across multiple domains (e.g., housing, education, health, income, and financial security), analyzed at the indicator or domain level rather than as a single composite index.

Several primary outcomes are constructed from multiple survey items to capture latent concepts related to aspirations and agency.

Individual (Critical) Agency will be constructed as an index combining survey measures of perceived control over life outcomes, self-efficacy, goal-setting capacity, and confidence in carrying out actions to improve household conditions. Items will be standardized and aggregated into a summary index, with higher values indicating greater individual agency.

Collective Agency will be constructed as an index capturing perceptions of collective efficacy and group-based problem-solving. This includes items measuring trust in others, willingness to engage in joint action, perceived ability of groups to achieve shared goals, and confidence in collective decision-making. Items will be standardized and aggregated into a summary index.

Aspirations will be measured using domain-specific questions capturing desired future outcomes in education, occupation and income generation, housing quality, and savings or financial security. For each domain, aspirations will be operationalized using indicators such as desired levels, time horizons, and perceived feasibility. Domain-specific aspiration measures may be analyzed separately and/or combined into a standardized aspirations index.

Multidimensional Poverty Indicators will be measured using self-reported indicators corresponding to the Poverty Stoplight domains, including housing, education, health, income and employment, and financial security. Rather than relying on a single composite poverty score, outcomes will be constructed at the indicator or domain level to capture changes in specific dimensions of deprivation.

For all constructed indices, individual components will be standardized using the control group distribution, and aggregation methods (e.g., simple averages or weighted indices) will be pre-specified in the analysis plan.

Secondary outcomes capture intermediate behaviors, perceptions, and program engagement that may mediate the effects of the intervention on aspirations, agency, and poverty-related outcomes.

Goal-Setting and Planning Behaviors
Measures of whether households set specific goals, the number and type of goals identified, and progress toward completing planned steps.

Follow-Through and Action
Self-reported actions taken toward stated goals, such as investments in education, housing improvements, income-generating activities, or savings-related actions.

Engagement with Mentoring and Group Activities
Participation in mentoring sessions, attendance at group meetings, and frequency of follow-up interactions with mentors or facilitators.

Social Interaction and Peer Support
Measures of peer learning, information sharing, mutual support, and perceived benefits from group participation, particularly relevant for the group-mentoring arm.

Psychosocial Well-Being
Measures of hope, optimism, future orientation, and perceived stress or discouragement, which may respond to changes in aspirations and agency.

Household Economic Behaviors
Intermediate economic behaviors such as saving practices, budgeting, or participation in income-generating activities, without requiring short-run income changes.

Secondary outcomes are constructed from survey responses and administrative records to capture intermediate mechanisms and behavioral responses to the intervention.

Goal-Setting and Planning Behaviors will be measured using indicators of whether participants articulated specific goals, the number of goals identified, and the extent to which goals are described in concrete and time-bound terms. These measures may be combined into summary indices reflecting goal clarity and planning intensity.

Follow-Through and Action will be measured using self-reported and, where available, administratively recorded actions taken toward achieving stated goals, such as educational investments, housing improvements, initiation of income-generating activities, or savings-related actions. Indicators may be analyzed individually or aggregated into an action index.

Engagement with Mentoring and Group Activities will be measured using administrative records and self-reports capturing participation in mentoring sessions, attendance at group meetings, and frequency of follow-up interactions with mentors or facilitators.

Social Interaction and Peer Support will be constructed from survey items capturing peer learning, information sharing, mutual support, and perceived usefulness of interactions with other participants, particularly in group-based mentoring settings.

Psychosocial Well-Being will be measured using survey items capturing constructs such as hope, optimism, future orientation, and perceived stress or discouragement. Items may be standardized and aggregated into summary indices.

Household Economic Behaviors will be measured using indicators of savings behavior, budgeting practices, and engagement in income-generating activities. These outcomes are intended to capture behavioral changes that may precede measurable income or consumption effects.

For constructed outcomes, individual items will be standardized and aggregated using pre-specified methods detailed in the analysis plan.

The study uses a clustered randomized controlled trial to evaluate a digitally enabled Poverty Stoplight (PSL) intervention delivered through Unbound’s existing sponsorship program in Honduras. Households are assigned to one of four study arms: an active control group receiving standard Unbound programming, and three treatment groups receiving the PSL tool with varying levels of mentoring support (no mentoring, individual mentoring, or group mentoring).

Randomization is conducted at the social-promoter cluster level to minimize spillovers between participants. Each cluster consists of approximately 7–14 households, and clusters are assigned to treatment arms in approximately equal proportions. The intervention is implemented over a period of approximately 13 months, and outcomes are measured using baseline and endline household surveys.

The primary outcomes of interest include aspirations, individual agency, collective agency, and multidimensional poverty indicators.

Not available

Randomization is implemented in Stata using block and stratified assignment at the cluster level, with a reproducible random seed. Following Morgan and Rubin (2012), a re-randomization procedure is applied whereby candidate random assignments are repeatedly drawn and evaluated against pre-specified balance criteria on baseline covariates. Only assignments that satisfy these balance thresholds—defined in terms of standardized differences across treatment arms at the cluster level—are retained. From the set of acceptable assignments, one allocation is selected at random. Assignment is approximately balanced across the four experimental arms.

Randomization is conducted at the social-promoter cluster level. Each cluster consists of approximately 7–14 households served by a social promoter. All households within a given cluster are assigned to the same treatment arm. There is no individual-level randomization.

Yes

343 social-promoter–level clusters (operational clusters of approximately 7–14 households each).

2,623 households (caregivers), distributed across 343 social-promoter–level clusters.

Control (standard Unbound programming): 659 households in 86 social-promoter clusters

PSL without mentoring: 662 households in 86 social-promoter clusters

PSL with individual mentoring: 655 households in 85 social-promoter clusters

PSL with group mentoring: 647 households in 86 social-promoter clusters

Total: 2,623 households across 343 social-promoter–level clusters.

Treatment assignment in the study follows a re-randomization procedure as proposed by Morgan and Rubin (2012). Under this approach, random allocations of clusters to treatment arms are repeatedly drawn and evaluated against pre-specified balance criteria on baseline covariates at the cluster level. Only allocations that satisfy these balance thresholds are retained, and one acceptable allocation is then selected at random for implementation. This procedure is designed to improve baseline covariate balance relative to simple random assignment. Power calculations are conducted using simulation-based methods in Stata and do not explicitly model the rerandomization acceptance rule. Instead, they assume random assignment at the cluster level with equal allocation shares across arms. This choice is standard in practice and yields conservative power estimates, since rerandomization weakly improves precision by reducing imbalance in baseline covariates. The simulations implement an ANCOVA specification with baseline and endline outcomes and standard errors clustered at the randomization unit (social-promoter cluster). The data-generating process includes both cluster-level and individual-level components, with serial correlation over time. Specifically, cluster-level persistence is set to rho=0.3 and individual-level persistence to rho=0.5. The design assumes perfect compliance, no attrition, and no additional covariates beyond the lagged outcome. The main power scenario reflects the planned study design: 343 clusters, an average cluster size of 8 households, and four experimental arms with approximately 25% allocation per arm. Power is evaluated for treatment–control contrasts using a two-sided significance level of alpha=0.017 to account for multiple primary comparisons, and a target power of 80%. Each scenario is simulated using 1,000 Monte Carlo replications. Intracluster correlation coefficients (ICCs) are varied across values informed by baseline data. Under this design, the minimum detectable effect size (MDE) for main continuous outcomes (e.g., standardized indices of aspirations and agency), expressed in standard deviation units, is approximately: ICC = 0.05: MDE ≈ 0.18 SD (18% of one standard deviation) ICC = 0.075: MDE ≈ 0.19 SD (19% of one standard deviation) ICC = 0.10: MDE ≈ 0.20 SD (20% of one standard deviation) These MDEs are reported in standard deviation units of the outcome variable. Because the simulations do not explicitly incorporate the rerandomization rule, the reported MDEs should be interpreted as conservative estimates. To the extent that the Morgan and Rubin (2012) rerandomization procedure improves baseline balance and reduces residual variance, the true statistical power of the study is expected to be at least as large as, and potentially larger than, that implied by these estimates.