Which Factors Contribute to the Dismissal of Eviction Filings? a Computational Analysis of Court Cases

Background/Purpose: Eviction remains one of the most pressing issues today, affecting millions of households. On average, 3.6 million eviction filings are made annually across the US. A major challenge in addressing this issue is the inaccessibility of eviction data, as a vast amount of detailed information resides in unstructured court records, primarily in PDF format. Although filing data provides an initial insight into the eviction process, the subsequent events are largely understudied. This oversight is critical because not all eviction ends with the same result. If the case is dismissed, it indicates that the case was resolved early, potentially giving tenants more flexibility. Additionally, the movement to seal eviction records, such as Oregon's Senate Bill 873, allows for dismissed cases to have restricted public access. Thus, this study aims to identify the individual, community, and macro-level factors associated with the outcomes of dismissal or judgment in eviction filings.

Methods: This study employs advanced document layout analysis using Natural Language Processing and Computer Vision to extract information from PDF files. This information is then linked to broader property, community, and county datasets. Using classification models, including logistic regression and random forest, the study identifies key factors associated with the outcomes of eviction filings. The analysis encompasses 56,070 unique cases are from 772,629 PDF files from 2004 to 2022 in a western state county. A descriptive analysis is also presented to illustrate dismissal and judgment outcomes distribution.

Results: Except in 2004 and 2005, the dismissal rate of eviction filings ranged from 20-30%, indicating that many cases were resolved before reaching judgment. The rates were notably higher in 2021. This increase may be attributed to more cases being resolved outside of court, as court processes experienced delays due to the moratorium. Furthermore, several significant variables were identified, including racial identity, property sale records within the same year as the eviction filing, property taxable value, landlord type (organization vs. individual), percentage of rental properties in the community, and the overall fair rent price. For instance, a one-unit increase in the percentage of rental properties was associated with a 13% increase in the estimated odds of dismissal (AOR=1.13, p<0.01). This suggests that more rental properties in the same tract may facilitate easier relocation for renters facing eviction, without significant disruption. The model also revealed that Asian and Black individuals were less likely to have their cases dismissed compared to White individuals, particularly when the proportion of White people in the community increased.

Conclusions/Implications: Each eviction case represents an individual or household, yet research often aggregates these cases at the census scale, which may obscure individual experiences. This study innovatively extracted individual-level data from PDFs, enabling the examination of how personal and property data contribute to the likelihood of case dismissal. This approach can be expanded to other research topics related to court files, such as medical debt, thereby creating possibilities for collaborative research. The findings can also assist in developing interventions that can be provided post-eviction filing to promote stable housing within the community.