Question

Difficulty: Very hardLogical Arguments and Causal Relationships

A financial research study across 60 regional credit unions examined the implementation of machine-learning credit risk algorithms and subsequent small-business loan default rates. The study observed that credit unions implementing machine-learning algorithms experienced a 40 percent lower default rate on small-business loans over a three-year period than credit unions relying strictly on traditional manual underwriting. Skeptical financial analysts contend that the algorithm itself did not cause the lower default rate. Instead, they argue that credit unions adopting the algorithm had systematically raised their minimum credit score requirements for all loan applicants immediately prior to software installation, thereby selecting an inherently lower-risk applicant pool.

Based on the information provided, select the statement that most strengthens the financial analysts' alternative explanation, and select the statement that most weakens the financial analysts' alternative explanation (thereby supporting a direct causal relationship between algorithm adoption and reduced defaults).

  • Strengthens Analysts' Alternative ExplanationCredit unions that raised minimum credit score requirements without installing the machine-learning algorithm achieved small-business loan default reductions equal to those of credit unions that installed the algorithm.
  • Weakens Analysts' Alternative ExplanationCredit unions adopting the machine-learning algorithm observed substantial default reductions even among approved borrowers whose credit scores were below the newly raised minimum threshold via special waiver programs.

Answer

The statement strengthening the analysts' argument is that credit unions raising credit score requirements without the algorithm achieved equal default reductions. The statement weakening the analysts' argument is that default reductions occurred even among borrowers with credit scores below the new threshold.
To strengthen the analysts' claim that credit score threshold increases (and not algorithm installation) caused the reduction in defaults, we look for evidence showing that raising thresholds alone produces the same result. The statement noting that unions raising score thresholds without algorithms achieved equal default reductions directly confirms this alternative cause. To weaken the analysts' claim, we look for evidence where default reductions occurred independently of the higher credit score threshold. The statement showing default reductions among waiver-program borrowers below the threshold proves the algorithm worked even when the proposed confounding factor was absent.

Step-by-Step Solution

1
Analyze the causal claim and the counter-argument
Study Claim: Algorithm adoption causes lower default rates. Analysts' Claim: Tightened minimum credit score requirements (confounding variable), not the algorithm, caused lower default rates.
To strengthen or weaken an alternative causal explanation, we must isolate the proposed confounding variable (credit score threshold) from the primary variable (algorithm adoption).
2
Evaluate candidate statements to strengthen the analysts' argument
The statement regarding credit unions that raised credit score requirements without installing the algorithm shows that credit score tightening alone yields an identical default reduction. This proves the confounder is sufficient to explain the outcome without the algorithm.
A control group demonstrating that the outcome occurs to the same degree without the treatment directly strengthens the alternative explanation.
3
Evaluate candidate statements to weaken the analysts' argument
The statement showing default reductions among borrowers below the new credit score threshold demonstrates that the algorithm reduced defaults in a population where the credit score change was absent.
Demonstrating the effect in a subgroup unaffected by the proposed confounding variable directly undermines the claim that the confounder is responsible for the overall effect.

Key Concept

Evaluating Confounding Variables and Alternative Causal Explanations
Estimated Time:2m 30s
Rate this question