Leveraging Linked Administrative Data to Examine Long-Term Outcomes of Supportive Housing for Homeless Families Involved with Child Welfare

Background:
Administrative data linkage has long been recognized as a valuable tool for addressing questions related to child maltreatment. However, gaps remain in the literature regarding the technical details of data linkage, its use in intervention evaluation, and the analysis of outcomes across multiple social service sectors. This study addresses these gaps by applying the Guidance for Information about Linking Data Sets (GUILD) framework to transparently report the data linkage processes used to evaluate the long-term impacts of a supportive housing intervention, tested through a randomized controlled trial, for homeless families with children placed in foster care. The intervention aimed to promote family reunification. The objectives of this study were twofold: (1) to demonstrate how administrative data linkage can be used to assess long-term, cross-sector outcomes—including educational achievement, housing stability, child welfare involvement, lead testing, and receipt of public assistance; and (2) to illustrate how adopting GUILD principles enhances transparency, reproducibility, and analytical validity.

Methods:
Using the Child and Household Integrated Longitudinal Data (CHILD) system—a comprehensive integrated data platform—we linked records from the Homeless Management Information System (HMIS), child welfare, public education, and public assistance datasets. We first described how each dataset was generated, maintained, and cleaned, including how unique identifiers were handled. Deterministic linkage was supplemented with probabilistic algorithms. Key components—such as data cleaning protocols, blocking strategies, match weights, and score thresholds—were fully documented to ensure transparency. Modeling approaches incorporated match confidence indicators, and sensitivity analyses were conducted to assess the impact of linkage quality on findings.

Results:
All data linked to the CHILD system were geocoded and standardized. A third-party SAS macro, LinkPro2, was used to perform deterministic and probabilistic matching. Child and parent names and birthdates were the primary matching variables, supported by Soundex phonetic indexes to address spelling inconsistencies. LinkPro2 applied probabilistic weights to estimate the likelihood of a true match between records. Matches above a high threshold were classified as true; those below a low threshold were deemed false; and intermediate scores were labeled as “possible matches” for manual review, along with other unresolved cases. In total, 540 children were included in the sample. All were successfully matched to child welfare records (100%), with 77% matched to public school records, 72% to HMIS, and 84% to lead testing data.

Conclusions and Implications:
By following the GUILD framework, this study provides a transparent and reproducible data linkage process that strengthens the validity of cross-sector outcome evaluations. As public agencies and policymakers increasingly seek rigorous evidence of intervention effectiveness across domains, our approach offers a practical model for using integrated data systems to inform such assessments. The detailed documentation of our linkage procedures can serve as a blueprint for practitioners and researchers aiming to build or utilize cross-system data to support intervention design, evaluation, and policy development. In our next step, we will explore methods like Hashed Linkages to protect personal information while ensuring successful data merging.