Statistical Analysis of UNHCR Peru Graduation Model
1 Description of Intervention
The Graduation Model allows refugees, migrants, and asylum seekers from Venezuela to integrate into host communities by promoting sustainable livelihoods and economic stability. Between 2021 and 2023, the institutions involved in implementing the Model were UNHCR Peru, HIAS, & BPRM.
The intervention was designed to provide aid to Venezuelan families in mobility and vulnerable situations for up to 18 months. Families graduated (i.e. capable of securing the means necessary to live without institutional support) after meeting each program’s criteria for 6 months in a row.
2 Objective of the Study
Identify family profiles that successfully achieve graduation and determine the main factors that could influence graduation outcomes through a statistical analysis.
The study components were:
Descriptive Analysis
Statistical Modelling
Results visualization
2.1 Research questions
To achieve the study objective, different research questions were formulated:
What are the characteristics of the program’s recipients?
Which family features have a greater impact on graduating?
Do these features behave similarly across different groups in the target population?
Is it possible to create family profiles based on shared features between beneficiaries?
Do profiles explain families’ graduation probabilities?
2.2 Limitations of the study
The main limitation to carry out a more robust analysis was data quality. Databases were not homogeneous nor complete, some of them included fewer variables than the rest and, in top of that, inconsistencies in registration led to information losses.
3 Descriptive Analysis
To understand the main features and characteristics of families participating in the Graduation Model, we ran an exploratory analysis of program databases with the following results:
We focused on families that are no longer part of the Graduation Model, leaving out of the statistical analysis families still participating in the program.
4 Statistical Modelling
As previously mentioned, the goal of the analysis was two-fold:
Understanding the relation between the program’s outcome and the characteristics of participating families; and
Identifying traits that allow to cluster families.
However, the study was not designed to infer causality. An experimental design, including control and treatment groups, would have been necessary.
The statistical modelling relied on two classification methods: a logistic regression (logit) and a hierarchical clustering model.
The logit results indicated which family features have the greatest impact on the probability of graduating, while the clustering model was estimated to group households based on family characteristics and program performance, in other words, families’ profiles.
| Input | Data source | Logit regression | Clustering model |
|---|---|---|---|
| Family characteristics | Program | \(\checkmark\) | \(\checkmark\) |
| Intervention performance | Program | \(\checkmark\) | \(\checkmark\) |
| Sociodemographics in host community | Secondary - Official records | \(\checkmark\) |
4.1 Logit results
The following table shows the estimation of the logit regression, remembering that the model only used data from families that finished their participation in the Graduation Model. Internally, the model compared the performance between graduates and non-graduates to estimate the impact of each individual feature in the probability of graduation.
Our theoretical approximation to the logit model can be summarized as this: any feature that makes graduation more likely is associated with a reduction in vulnerability.
| Features increasing the probability of Graduation | |||
| Data source | Feature | Probability change | Result interpretation |
|---|---|---|---|
| Family characteristic | Male focal point | 22% | Probability of graduating increases compared to families with a female focal point. While the program empowers female beneficiaries, gender gaps in terms of employment and salary still prevail, impacting the probability of graduation of female-led families. |
| Family characteristic | Carne de extranjeria (foreigner id card) | 15% | The Carne grants more rights to its holder than other immigration documents, which could bring greater stability to the family and opportunities to integrate to the host community |
| Family characteristic | Number of family members between 31-40 | 8% | An additional family member in that age range increases the probability of graduating. A larger number of working-age individuals in the family may contribute to having an additional source of income, facilitating to meet multiple program criteria |
| Program performance | Participation in personal finance workshop | 44% | Individuals who participated in the workshop acquired skills to improve the household economy (e.g., savings, budget planning, access to the financial system, etc.) |
| Program performance | Support from family/friends networks | 22% | Community support could increase families' chances of meeting their needs, which can include employment, food, housing, or other necessities |
| Program performance | Per capita income (increasing S/ 100 monthly) | 2% | Every additional S/100 increases the probability of graduation. Income enables families to satisfy their basic needs and to meet multiple program criteria for graduation. Moreover, income allows for planning and savings, permitting families to be prepared for external shocks |
| Local context | % of women with tertiary education in host community | 9% | When the percentage of women with tertiary education in the district increases by 1%, the probability of graduation for families in that same district increases. University education could be enhancing employment opportunities or fostering a social context of greater inclusion for women. Some academic references supporting this outcome can be found in the following links |
4.2 Clustering model results
As in the logit regression, the clustering model was estimated using data from families that are no longer participating in the program. This statistical exercise identified how families are grouped according to a set of shared traits.
The clustering algorithm produced 4 family profiles. The following table shows the two of them with the higher probability of graduating.
| Number of families in cluster | Shared characteristics among families | |
|---|---|---|
| Profile 1 | 80 |
|
| Profile 2 | 46 |
|
5 Conclusions
The initial conditions (characteristics) of families have a significant impact on their subsequent performance and the probability of graduation.
During program, creating social networks and learning to manage income contribute to reduce the vulnerability that families face when joining the program.
In spite of logit results, traits that increase the chance to graduate are not isolated and it is rare to find a family that possess all features that favours graduation.
6 Recommendations
6.1 To strengthen future interventions
- Make the financial workshop mandatory, particularly, for families participating in the intervention for over 12 months
- Reinforce communication on the benefits of obtaining a Carne de Extranjeria and providing legal assistance to get it
- Map organizations and community-based associations
- Childcare services for single-mother led families with young children
6.2 On data management
- Standarize the structure of databases
- Promote a data-driven culture and raise awareness among all staff about the importance of data in decision-making
- Collect more data to create comprehensive profiles of beneficiaries at the family and individual levels
- Encourage collaboration and knowledge exchange between stakeholders
- Recognize and reward the use of data