How We Evaluate Fair Lending Outcomes
Fair lending risk seldom reveals itself in summary metrics alone. Institutions need to know not just whether disparities exist, but where they arise, how significant they are, and whether legitimate underwriting factors explain them. CrossCheck’s fair lending statistical analysis addresses those questions through structured testing of lending outcomes across products, channels, and decision points. The result is a practical, evidence-based assessment of potential fair lending risk that compliance, legal, credit, and executive stakeholders can use with confidence.
Our approach combines statistical disparity testing with regression-based analysis to assess approval and denial outcomes, compare prohibited basis groups with control groups, and identify the factors most strongly associated with observed disparities. Where appropriate, we also use layered analysis to examine how disparities shift as applications move through different stages of the decision process, helping institutions distinguish among baseline underwriting risk, policy-driven effects, and manual review impacts. This analysis supports both ongoing fair lending monitoring and more targeted reviews related to new products, underwriting changes, or examination readiness.
- Statistical disparity testing
We test lending outcomes across applicant groups to identify whether observed differences in approvals, denials, pricing, exceptions, or other outcomes are statistically significant and may warrant further review. Statistical disparity testing is a core part of CrossCheck’s methodology for evaluating fair lending outcomes. - Regression analysis
We use regression-based methods to determine whether disparities persist after controlling for relevant credit and underwriting variables. This allows institutions to move beyond raw comparisons and better assess whether observed differences may indicate residual fair lending risk. - Decision funnel analysis
Where lending decisions involve multiple stages, we analyze outcomes across the full decision funnel to identify where disparities emerge, increase, or decline. This layered approach is used to distinguish the effects of baseline underwriting criteria, model outputs, policy thresholds, and manual review. - Benchmark and challenger analysis
We establish a baseline using legacy criteria, benchmark approaches, or challenger frameworks to determine whether disparities reflect existing underwriting patterns or changes introduced by new processes. - Attribution of disparity
Our analysis is designed to do more than identify disparities. It helps pinpoint the factors driving them, enabling more targeted governance, more effective remediation planning, and better-informed internal decision-making. - Targeted file review support
When statistical results point to specific patterns or marginal outcomes, we identify areas for deeper comparative file review.
Meet Your Fair Lending Experts
Rebecca Escario
Ryan Addabbo
Hamza Abourifq
Jonathon Neil
Representative Engagements
Statistical Analysis FAQs
What is fair lending statistical analysis?
Fair lending statistical analysis is the use of quantitative methods to evaluate whether lending outcomes differ across groups in ways that may indicate potential discrimination. It goes beyond simple summary comparisons by measuring where disparities appear, how large they are, whether they are statistically meaningful, and whether they persist after accounting for legitimate credit and underwriting factors. The purpose is not just to identify that a disparity exists, but to provide an evidence-based view of what may be driving it.
What types of lending outcomes can be tested?
Statistical analysis can be used to test outcomes such as approvals, denials, pricing, score patterns, auto approvals, exceptions granted, manual review outcomes, and final credit decisions. The appropriate scope depends on the product, underwriting structure, and available data.
Why is statistical testing important for fair lending compliance?
Summary metrics alone may not show whether disparities are meaningful or where they arise in the decision process. Statistical testing helps institutions move from broad observations to a more structured understanding of whether differences in outcomes may present fair lending risk.
How is regression analysis used in fair lending reviews?
Regression analysis helps evaluate whether disparities persist after relevant credit and underwriting factors are considered.
Can statistical analysis show where disparities emerge in the lending process?
Yes. A layered or funnel-based approach can help identify whether disparities appear in baseline underwriting criteria, model scoring, configuration thresholds, manual review, or final decisioning. This supports a more targeted understanding of potential root causes.
What is decision funnel analysis?
Decision funnel analysis evaluates how applicants move through each stage of a lending process and where disparities may increase, decrease, or reappear along the way. This approach is used to distinguish the effect of legacy underwriting criteria, AI scores, configuration rules, and manual review.
What is the difference between disparity testing and outcome scenario analysis?
Disparity testing measures whether differences in outcomes between groups are statistically significant. Outcome scenario analysis goes a step further by evaluating how applicant outcomes shift between stages of the process, helping identify where similarly situated applicants may be disadvantaged and where targeted file review may be appropriate.
What happens if statistical analysis identifies disparities?
If disparities are identified, the next step is usually deeper analysis to understand possible drivers and determine whether additional review is warranted. That analysis may include regression testing, evaluation of policy thresholds or decision rules, and targeted comparative file review of marginal or transitioned cases.
Is fair lending statistical analysis only relevant for AI or automated underwriting?
No. Statistical analysis can be applied to traditional underwriting, policy-based decisioning, manual reviews, automated models, or hybrid processes. The same core testing concepts can help institutions evaluate fair lending risk across many different lending environments.
How often should fair lending statistical analysis be performed?
The appropriate cadence for analysis depends on the institution’s product mix, lending volume, operational complexity, and pace of change. In many cases, fair lending statistical analysis is most effective when conducted on a recurring basis, allowing institutions to monitor trends over time, assess the impact of new underwriting practices, and identify emerging fair lending risk more proactively.
Featured Resources
