The decisions-by-the-numbers guys certainly had their day in court. The free advertising wasn’t bad either.
A recent U.S. Court of Appeals case is timely not only because it involves corporate liability for ocean pollution when everybody in this country is morbidly tracking the BP spill in the Gulf but because it is a case in which the judge highlights and corrects some common misconceptions about Monte Carlo simulation.
In a consolidated case involving hazardous waste dumping in the Houston Ship Channel, the codefendants, Tenneco and Occidental, acknowledged liability for the pollution cleanup, but they appealed a lower court’s decision partly on the basis of the court’s method of allocating costs. The court had called an environmental engineer as expert witness and statistical analyst. The engineer used Monte Carlo software and court-established inputs for his model. The defendants challenged the court’s inputs in the risk analysis model, and the Circuit Court decision rebutted their objections in clear terms.
Writing for the Fifth Circuit Court of Appeals, Judge Patrick Higginbotham said, "Monte Carlo measures the probability of various outcomes, within the bounds of input variables; to calculate Occidental’s waste volume,. . . Instead of simply averaging the input values, Monte Carlo analysis uses randomly-generated data points to increase accuracy, and then looks to the results that those data points generate. The methodology is particularly useful when reaching an exact numerical result is impossible or infeasible and the data provide a known range—a minimum and a maximum, for example—but leave the exact answer uncertain."
Responding to the charge that this method of statistical analysis is unreliable and untestable, Higginbotham responded,". . .the cited cases at most stand for the proposition that Monte Carlo analysis is unreliable when injected with faulty inputs, but nothing more. . . . Monte Carlo simulation is not inherently untestable. . . . If anything, Monte Carlo provides greater certainty than the basic alternatives: using one of the three data or using the arithmetic average of all three."
Countering the challenge that the model results were "equivocal" the judge continued, " The Monte Carlo analysis—though it produced a statistical range of likely outcomes and not one determinative answer—supports choosing one result over another, and certainly assisted the district court in its decisionmaking."