by Yuri de Melo Costa
Abstract
I study how Brazil’s Sistema Nacional de Adoção e Acolhimento (SNA) (National Adoption System)
matches Children and Adolescents (CAs) to Prospective Adoptive Parents (PAPs). Despite
a ratio of roughly seven certified PAPs per available CA, thousands of children remain in institutional
care. Using administrative microdata from the SNA linking 38,535 unique CAs and 77,132
PAPs through 613,694 choice-set observations (2008–2024), I introduce a multi-label acceptance
vector Ai ∈ {0,1}10 and estimate a conditional logit with Kaplan–Meier-corrected child-type waiting
times as the shadow price of adoption. All acceptance-alignment coefficients are positive and
significant, yet 21.7% of completed adoptions fall outside PAPs’ stated preferences. Willingness to-
wait ratios quantify the system’s implicit valuation of each alignment dimension, pooling across
CNA and SNA periods, and range from 102.9 months (race composite) to 15.9 months (treatable
health conditions). The sample captures the transition from manual caseworker matching (CNA,
2008–2018) to algorithmic matching (SNA, 2019–2024). A Chow test confirms a structural break:
CNA-era caseworkers overrode stated preferences more frequently as PAPs waited longer, while
the SNA algorithm mechanically respects stated constraints regardless of queue duration. Gender
preferences are inviolable under both regimes; age and sibling preferences are soft constraints
whose treatment depends on whether the intermediary is human or algorithmic. These findings
suggest that algorithmic rigidity may leave willing PAPs waiting indefinitely, and that a queue duration
trigger expanding age and sibling consideration could recover the caseworker’s beneficial
flexibility while preserving algorithmic transparency.
Atila Abdulkadiroglu, Faculty Advisor
