US Defends Use of 'd' Test in Lead Case at Federal Circuit

The U.S. addressed all three of the U.S. Court of Appeals for the Federal Circuit's concerns regarding the Commerce Department's use of the Cohen's d test to find "masked" dumping in antidumping duty proceedings, the government told the appellate court. Submitting its reply brief on Nov. 6 in the lead case on Commerce's use of the test, the U.S. claimed that data sets with non-normal distribution "do not inevitably lead to inaccurate results," the ratio test used as part of its differential pricing analysis accounts for data sets with small amounts of numbers and the test runs effectively involving data sets with small variance (Stupp Corp. v. United States, Fed. Cir. # 23-1663).

Also addressing the Federal Circuit's concern regarding thresholds set by the test's founder, Dr. Jacob Cohen, the government said that the use of the 0.8 threshold is a "measure of practical" and not "statistical significance." Cohen developed this threshold "independently of his assumptions regarding normal distribution, equal variance, and sufficient numbers of data points as part of a statistical or 'power' analysis," the brief explained.

The U.S. railed against exporter SeAH Steel Corp.'s "remaining arguments" in the suit, telling the appellate court that just because the exporter "can concoct a contrived hypothetical which purports to demonstrate flaws in the Cohen's d test does not make" the differential pricing analysis "unreasonable."

The Federal Circuit initially raised its concerns about the statistical test in July 2021, taking issue with the fact that Commerce used the test with data sets that did not adhere to basic statistical assumptions, including normal distribution, equal variances and large sample sizes (see 2107150032). The agency saw its position sustained, though, once the case returned to the trade court. The matter is now back before the Federal Circuit, where the government is claiming that its further explanation of the test assuages the appellate court's concerns.

One such concern dealt with the non-normal distribution of the data. The U.S. said this distribution doesn't inevitably lead to inaccurate results, noting that there is no evidence, nor does SeAH claim there to be any evidence, that SeAH's sales led to inaccurate results "with the sufficient frequency as to change the outcome of Commerce's analysis." Even if there were issues, the ratio and meaningful difference tests compensate for any potential issues, the brief added.

In its previous opinion, the Federal Circuit raised a hypothetical situation whereby data sets with small variances skew the test's results, allowing all the data to "pass" the test. In response to this concern, the government said the test continues to work as intended in this circumstance, "consistent with the direction that 'Commerce will proceed on a case-by-case basis, because small differences may be significant for one industry or one type of product, but not for another.'" In this situation, the U.S. also touted the soothing effects of the ratio and meaningful difference tests.

Central to the government's defense in the present matter, and all other cases involving the Cohen's d test, is the idea that the statistical assumptions aren't required since the agency used the entire population of data and not just a sample size. In making this claim, the government said SeAH's arguments to the contrary are "purely negative" and fail to refute the point that nothing in the academic literature specifically says that it's "improper to use the Cohen's d test in the manner utilized by Commerce."