With Paulius Yamin, Luis Artavia Mora and Lorena Levano.

The project asks why Venezuelan migrants in Peru — younger and better educated than the average Peruvian — are regularly excluded from formal financial inclusion (bank account, credit), and which are the behavioral drivers behind the differential treatment. In partnership with the International Finance Coorporation and a large Peruvian commercial bank, we ran an intervention in Lima, Peru, combining in-person capacity building (including an theater of the oppressed exercise) and a large survey experiment with hundreds of bank executives. In our between-subjects 2×2 factorial vignette experiment participants were randomly assigned to vignettes varying only in the national origin (Peruvian/Venezuelan) and gender of an otherwise identical client. Companion modules captured relative trust, beliefs about creditworthiness, and self-perceived discrimination. We found a significant and substantial discrimination of migrants, robust to model specifications, and deeply intersectional. Predicted loan approval drops almost 60pp for Venezuelan women, a penalty that exceeds the additive origin and gender effects. 

With Lina Restrepo-Plaza, Hector Solaz and Julian Pinazo.

In collaboration with the regional government of Valencia (the Generalitat Valenciana's Conselleria de Servicios Sociales, Igualdad y Vivienda), we diagnosed behavioral barriers that prevent equitable access to inclusion policies. The behavioral diagnostic tool combined focus groups with immigrants and members of other vulnerable groups and a pre-registered online survey experiment with 1,700 Valencian residents aged 18+, stratified by gender, province, and income. As part of the survey, participants were assigned in  different vignettes to case in which an applicant with different origins (Colombian, Moroccan, Spanish) and income level (high, low) applied to an inclusion policy she was entitled to. We found s strong and significant reverse discriminatory perception in vignette #1 (as residents responded that it would be easier for immigrants to succeed in accessing the service) and economic discrimination in vignette #2. We also identified six significant behavioral barriers, including knowledge illusion (63% of participants claimed expertise, but only 16% pass a simple test), mistrust of public officials, perceived information complexity, inaccurate beliefs about how applicants are entitled to inclusion policies, and procedural opacity. 

You can read the policy report here.

With Lina Restrepo-Plaza and Paulius Yamin.

In collaboration with the International Labour Organization, we diagnose workplace violence and harassment in Pakistan's healthcare sector and pilots a behavioral diagnostic tool designed for multi-country deployment. The instrument combines two pillars. Eight focus group discussions with healthcare workers across four facilities in Islamabad and Sahiwal capture the linguistic and normative environment through Recurrent Terms Analysis. A behaviorally inspired survey experiment, administered to 406 doctors, nurses, and technicians in six hospitals, embeds three vignettes (relational, physical, and sexual aggression) and randomly assigns victim gender, isolating the causal effect of gender on organizational response, victim-blaming, and accountability attributions.

Results point to a systemic, not individual, failure. Between 60 and 67 per cent of workers lack substantive knowledge of ILO Convention No. 190; nurses, the highest-risk group, report the lowest knowledge and the highest perceived reporting barriers. Female workers report 60 to 160 per cent more organizational harassment, 40 to 45 per cent lower institutional trust, and 15 to 20 per cent higher reporting barriers. In the experiment, identical scenarios trigger more victim-blaming when the victim is female, especially in sexual harassment. 

With Lina Restrepo-Plaza and Juan José Rojas-Constain

Household over-indebtedness in Latin America has grown alongside financial inclusion. The project asks which behavioral and cognitive factors travel with over-indebtedness, and whether the binding constraint on credit-choice quality lies with the borrower or with the decision environment. We deployed a behavioral diagnostic with clients of a partner commercial bank and control group from the general population in Colombia. The diagnostic combines measures of cognitive hijacking, financial literacy, time and risk preferences, and overconfidence with a 2×2 between-subjects vignette experiment that randomly varies disclosure clarity and the protagonist's income variability.

Two findings stand out. First, cognitive hijacking is the dominant correlate of self-reported over-indebtedness (16 pp at the strict threshold, 34 pp at the broader one); financial literacy is muted. Second, in the experiment, clear and behaviorally informed disclosure raises the probability of choosing the lower-cost loan by 56 to 64 pp, and a variable-income protagonist by a further 22 to 25 pp; the two channels are partial substitutes. The dissociation between the field marker and the experimental lever is the central policy finding.

You can read the policy report here.

With Lina Restrepo-Plaza and Paulius Yamin.

In collaboration with USAID, we identified and addressed behavioral bottlenecks holding back civic engagement among residents and diaspora members in Bosnia and Herzegovina's local governance units (LGUs), and embed behavioral-science methods inside two ongoing USAID activities: Local Governance Assistance (LGAA) and Diaspora Invest 2 (DI2). We added validated behavioral items (self-efficacy, risk and time preferences, confirmation bias) in a LGAA's survey run with 3,500 citizens across 10 municipalities and run fieldwork in 6 LGUs plus Sarajevo, (combining KIIs and FGDs). 

Our results suggest that self/outcome efficacy (64.9% see little chance of making a difference), institutional and group trust (LGU trust outranks national, but is still <40%), and the opportunity cost of personal versus collective benefit drive low engagement with LGUs. Risk-tolerance and patience interact non-linearly — only citizens who are both engage materially more (31.2%). Fieldwork added structural and contextual barriers: demographic decline, weak awareness of scorecards/Doing Business guides/budget infographics, and a perception that participation channels serve political interests. We designed two RCTs for DI2 (a social-media A/B test on the next RFA comparing traditional ad copy vs. a behavioral arm using efficacy, social comparison, and ambiguity reduction) and for LGAA (comparing an "information only" vs. "participation festival" invitation test across public hearings, civic scorecards, budget infographics, and public–private dialogues).

With Lina Restrepo-Plaza, Ana María Rojas, and Paulius Yamin.

This intervention targets a persistent gap between Ecuador's constitutional guarantee of universal access to public services and the lived experience of more than 500,000 Venezuelan migrants, who face exclusion rates four times higher than nationals in education and out-of-pocket health expenditures well above WHO ceilings. We address discretionary discrimination by front line civil servants through a scalable, low-cost behavioral intervention, in collaboration with the Government of Ecuador and the eMBeD at the World Bank, with the support of the Government of Canada as donor. 

We designed an online training course from scratch combining behavioral ingredients (perspective-taking, emotional identification, choice architecture, habit formation) with non-behavioral ones (service design, organizational policy, communication systems). Identification rests on a pre-registered RCT (AEARCTR-0014077) with a diagnostic baseline survey and an endline survey studying intention to act and attitudinal change. A correspondence experiment studied actual behavior tracking email-opening and reply behavior in response to fictitious service requests. The intervention raised trust in migrants by 16pp, and increased support for inclusive workplace incentives in 12pp. Treated civil servants increased email-opening rates by 40–46 pp and reply rates by 27–31 pp, with no interaction between treatment and migrant identity — gains accrued to Ecuadorian and Venezuelan applicants alike.