A statistics expert witness advises on a case involving an insurance company in South Dakota that brought suit against a mortgage capital company over a residential mortgage-backed security transaction. It alleges the mortgage company induced it to enter into the insurance agreement and issue the policy by knowingly making material misrepresentations and providing false warranties. These false misrepresentations and warranties concerned, primarily, the attributes of the loans and the manner in which the defendant’s loan conduit operations originated, acquired, and reviewed them.
The plaintiff retained a statistician who proposed a sampling methodology that plaintiff intends to use to prove liability and damages for its claims. The expert selected a sample of 400 loans from the total population of 15,615 loans in the transaction using stratified random sampling.
Question(s) For Expert Witness
- 1. Is the sampling methodology of plaintiff’s expert reliable?
- 2. Will the sampling be helpful to the trier of fact?
Expert Witness Response
Plaintiff’s expert failed to present enough information to allow assessment of his sampling and extrapolation plan. He is silent on such critical issues as:
· How he intends to stratify the population
· How he will select the sample from each stratum
· How he will handle loans that cannot be re-underwritten in the event loan documents are unavailable
· How he will deal with questions that are not binary, “yes/no” questions
· How he will extrapolate from the sample to the population of loans in the transaction
Given the paucity of details that he provides, I have no basis to conclude whether his sampling and extrapolation plan could achieve what he sets out to achieve. For that reason, I believe that his sampling plan is inconsistent with professional standards in statistics.
Further, plaintiff expert’s sampling proposal meets his accuracy standard of ±5 percentage points at a 95 percent confidence level only under restricted conditions. In suggesting his sample of 400 loans will meet the ±5 percentage point accuracy standard laid out in his affidavit, the expert also assumes that no question in these actions will pertain to a specific originator. However, certain allegations brought by plaintiff suggest otherwise.
The expert’s description of how he evaluated the representativeness of his sample is incomplete at this stage, including because he does not explain why he performed representativeness tests on some variables but not others. Furthermore, the representativeness tests that he says he performed may well be irrelevant.
The expert states that he stratified the population of loans before selecting the sample, and he suggests that stratification will be at least as accurate as simple random sampling and might reduce the sampling error. Yet he provides no details about his stratification plan, including the variables on which he stratified and his method for selecting the sample within each stratum. As presented, the stratification could increase sampling error compared to simple random sampling.
He also does not explain adequately how he would extrapolate findings from his sample to a broader population of loans.
Given my conclusions, I have serious doubts about whether the sampling proposal would generate useful information for a trier of fact. I further believe that there is no way to tell whether the expert’s sampling and extrapolation proposal will lead to reliable results without observing its application to the securitization at issue in this action.
The expert is a university statistics professor who specializes in applied statistical analysis research. He has taught probability and statistics at a leading institution for more than 35 years.