# Credit Scoring The credit score is the oldest mass-deployed algorithmic assessment of American individuals, and it established every convention that later scoring systems inherited: a single scalar summarizing a person, computed from behavioral data, by a private firm, using a proprietary model, applied to consequential decisions, with sharply limited rights of inspection or appeal. Understanding credit scoring is understanding the template. [[Algorithmic Bias]] debates about hiring, policing, and content ranking are re-running arguments that credit scoring settled — or failed to settle — decades earlier. ## What it was built to replace The pre-scoring alternative was the local loan officer, exercising judgment about applicants they often knew personally. That system was **arbitrary and overtly discriminatory**: outcomes turned on the officer's impressions, relationships, and prejudices, with no record and no recourse. Statistical scoring genuinely improved on this in measurable ways. It was consistent, auditable, and it removed a large volume of individual bigotry from the decision. Women and minority applicants had been denied credit by loan officers exercising exactly the discretion scoring took away. **This history matters, because the argument for algorithmic decision-making is not naive — it is a response to a documented failure of the discretionary system, and dismissing it requires engaging with what discretion actually produced.** ## What it carried forward anyway Scoring inherited the disparities of the system it replaced, through the data. Credit scores are computed from credit history. Credit history reflects prior access to credit. Prior access was shaped by explicit racial exclusion — the [[Redlining and Housing Discrimination]] regime, restrictive covenants, discriminatory FHA underwriting, and the loan-officer discretion scoring displaced. **The model is not biased in any technical sense; it is accurately summarizing a history of exclusion, and then applying that history forward as a forecast.** The mechanics of the "thin file" show how the loop closes: a person with little credit history cannot be scored reliably, is treated as high-risk, is offered credit only on adverse terms or not at all, and therefore continues to have little credit history. This is the [[Statistical Discrimination]] self-fulfilling loop in its cleanest form. ## Scope creep — the more important story Credit scores were designed to predict repayment of a specific loan. They are now used for: - **Insurance premiums.** Credit-based insurance scores affect auto and homeowners rates in most states, on the empirical finding that credit history correlates with claims. It does. It is also not what the score was built for, and it means a job loss raises your car insurance. - **Employment screening.** Used in hiring for many positions; restricted in some states. - **Rental applications**, utility deposits, and mobile-phone contracts. Each expansion was justified by predictive correlation. Cumulatively they produce a regime in which **a single number, computed by a private firm from a data set the subject cannot fully see, gates housing, employment, insurance, and credit simultaneously.** A single adverse event propagates across all of them at once. There was no decision to build that system; it accreted through a series of individually reasonable correlations. ## Rights and their limits The Fair Credit Reporting Act (1970) and the Equal Credit Opportunity Act (1974) established the basic protections: access to your report, dispute rights, adverse-action notices, and prohibition of protected-characteristic inputs. The gaps are structural rather than incidental: - **Reason codes are not explanations.** "Too many recent inquiries" describes a feature weight, not a route to a different outcome. - **Dispute resolution is largely automated** on both sides, with documented error rates in consumer reports. - **Proprietary models are not publicly auditable**, so disparate impact must be established from outcomes rather than from the model. - **Input prohibition does not prevent proxy reconstruction** — the same limitation described in [[Actuarial Science and Risk Classification]]. ## Why this is the branch's case study Credit scoring is where the modern arrangement was first established: **a mathematical model, defensible on accuracy grounds, replacing human judgment that was worse, while encoding historical injustice into a form that is harder to contest precisely because it is quantitative.** Every subsequent scoring controversy — [[Predictive Policing]], [[Criminal Risk Assessment Instruments]], algorithmic hiring — repeats this structure. The credit-scoring case is the one with fifty years of evidence about how it plays out. ## Related [[Statistical Discrimination]] · [[Actuarial Science and Risk Classification]] · [[Predictive Policing]] · [[Criminal Risk Assessment Instruments]] · [[Fairness Impossibility Results]] · [[Algorithmic Bias]] · [[Redlining and Housing Discrimination]] · [[Surveillance Capitalism]] · [[Mathematics and Power Hub]]