Sampling Auto Insurance Claims to Deny a Class Action Suit

An insurance carrier faced a class action affirming improper denials of claims for out-of-state accidents. To determine if the Plaintiffs represented the class, the insurer's statistician proposed sampling recent out-of-state claims. Then, senior case managers would review the claims and policies, looking for criteria relevant to each accident, to see if the claims were “typical” of those from other class members.

A paramount concern was ensuring that enough cases were sampled, because many would be rejected as irrelevant. Moreover, the results would only be persuasive if the review procedures were demonstrated to be objective, reliable and valid. Faced with the challenges and expense of such time-consuming data analysis, the insurance carrier retained statistical consultants Analysis & Inference to help design and execute the sampling.

To ensure reliable and valid claim reviews, a protocol was created. Dr. William Fairley, senior statistician at A+I, randomly selected 300 cases for review from the 11,303 out-of-state claims by in-state residents. Exactly 222 proved relevant – not duplicating another case, and not excluded under the policy.

Then, via sampling, guided by four criteria to rank similarities, Dr. Fairley estimated what percentage of the 222 cases were similar to the Plaintiff’s accidents and policies – within a reasonable margin of error. How many cases met all four criteria? Exactly zero.

Counsel for the insurance carrier used the Analysis & Inference report to prepare for the hearing. The motion for class action certification was denied.

While basing conclusions on a sample size of less than 2% may seem extreme, in fact, the margin of error provided by a random sample, and supported by probability calculations, depends hardly at all upon the size of the universe The actual margin of error was less than 4%.