Exporter Argues Against US Defense of 'd' Test in Case Returned to CAFC

The Commerce Department's use of thresholds proposed by Dr. Jacob Cohen -- the creator of the Cohen's d test -- for evaluating the d statistic in the agency's analysis to detect "masked" dumping isn't line with "normal statistical practice," exporter SeAH Steel Corp. argued in a Jan. 8 reply brief at the U.S. Court of Appeals for the Federal Circuit (Stupp Corp. v. United States, Fed. Cir. # 23-1663).

The appellate court previously remanded Commerce's use of the test since the agency did not adhere to certain statistical assumptions, including the normal distribution of data when using the test. The government explained on remand that it didn't need to satisfy these assumptions since it was using the whole population of data instead of a sample. The Court of International Trade sustained that explanation, prompting the case's return to the Federal Circuit.

If the U.S. is to claim that Commerce is not bound by normal statistical practice since it isn't using the d test in the way statisticians use it, the government "must demonstrate that its use of Cohen's d in a manner inconsistent with statistical practice is nevertheless reasonable," the brief said. There's nothing in the academic literature which allows the use of the Cohen's d test when the basic statistical assumptions aren't satisfied, SeAH argued.

The government pointed to a textbook written by Cohen which says that "effect size is a phenomenon of the population and it exists independent of any statistical analysis based on data sampled from that population." Citing Cohen, the U.S. said it was fine using the test for the d statistic since it used the full population.

SeAH dubbed that argument "fundamentally dishonest." For starters, Cohen's reference to effect size was only included in his discussion of "power analysis" and not the d statistic, the exporter claimed. "Professor Cohen’s general statements about effect sizes in a population in Chapter 1 of his text simply have no bearing on the proper use of the d statistic, as described" later in his book, the brief said.

Additionally, Cohen's comment about an inherent effect size "does not mean that it is appropriate to rely on a d statistic, and the specific thresholds that Professor Cohen proposed for use with that statistic." This is true especially since Cohen said the test was meant to be used with a t-test, which even Commerce admits, is to be used with normally distributed data, equal variances and sufficient data points, the brief said.

SeAH added that Commerce's use of the test "can classify imperceptible differences as significant, just as this Court predicted." In its previous decision, the Federal Circuit laid out a hypothetical that showed that small differences with no practical significance could lead to a "large" d value when the data variances are small. The U.S. claimed the "meaningful difference" test resolves this issue since the small price differences would lead to an insignificant difference in the dumping margins under the different comparison methodologies.

This is true "if there is only one product under consideration," SeAH said, adding that if there are multiple products at issue, an incorrect d test result for one product "will affect the ultimate outcome of Commerce's 'Ratio Test,' which depends on the percentage of sales with 'large' price differences."