Commerce Failed to Address Flaws in 'd' Test, Exporter Tells Trade Court

The Commerce Department has failed to address the flaws found in the use of the Cohen's d test when using its differential pricing analysis (DPA) to detect "masked" dumping, exporter SeAH Steel Corp. argued in a reply brief at the Court of International Trade. Responding to the U.S.'s argument that SeAH has failed to point to any statistical texts that explicitly address Commerce's claim that it can properly use the test, the exporter said that the burden is on the agency to find supportive texts and not merely rely on the silence of statistical authorities (Stupp Corp. v. United States, CIT #15-00334).

In antidumping matters, Commerce seeks to identify goods that are dumped into the U.S. market through "targeted" or "masked" dumping. Since Commerce typically conducts its dumping investigations by comparing the average home market price of the good in question to its average U.S. price (average-to-average), certain exporters may work around this by dumping the goods in certain areas and selling them at a higher price in another or at another time to get a non-dumped average U.S. price. To combat this, Commerce may compare the weighted average of sales in the home country to individual sales prices (average-to-transaction).

Before running an average-to-transaction test, though, Commerce must first gather data on the export sales and detect the masked dumping using the DPA. The agency breaks down the U.S. sales data into sets based on comparable product groups. Once in the product group, Commerce then breaks that data into various subsets, including the region the U.S. sales took place. Commerce will then pick one subset as the "test group" while aggregating the remaining subset into the "comparison group." This is where the Cohen's d test comes in, as Commerce uses the test to find if the test group is significantly different from the comparison group. If it is, Commerce applies a "ratio test" to see if the ratio of significantly different transactions warrants using the weighted average to individual transaction comparison.

In a July 2021 opinion, the Federal Circuit remanded Commerce's use of this test (see 2107150032). The appellate court called into question Commerce's use of the test since the agency failed to meet three statistical criteria (assumptions of normality, sufficient observation size and roughly equal variances) the court said could be vital to running the test. The Federal Circuit sent the matter back to Commerce to reevaluate its use of the test and to take another look at the statistical literature surrounding this question.

On remand, Commerce said that it can continue to use the Cohen's d test and that the court's concerns over the three statistical criteria are unfounded (see 2204050058). Since the criteria are not part of Dr. Cohen's thresholds that interpret the value of the "effect size," they are not relevant, the remand results said. Further, Commerce looks at the entire population of sales at issue, rather than just a sample, nixing the need to meet the statistical assumptions paired with running the test, the agency said.

SeAH took issue with this characterization. Addressing the claim that the typical restrictions don't apply since Commerce is using the entire population of data -- a claim the agency has also made in other cases over the use of the d test -- the exporter argued that none of the statistical texts addresses the use of the test with an entire population of data.

"Notably, Commerce itself has not identified any texts that support its proposed use of Cohen’s d," the brief said. "Commerce’s argument is purely negative: It contends that none of the texts on the record prohibit its proposed use. And, it faults SeAH for failing to identify any texts that explicitly address Commerce’s argument that Cohen’s d can properly be used to analyze an entire population, even when the assumptions described by Professor Cohen are not satisfied." SeAH continued to say that the d test is a "parametric test," meaning that it is only appropriate when the distribution of the data being analyzed is fully described by certain parameters.

The exporter further argued against Commerce's argument over the effect size since the proposed thresholds can't be used as "universal yardsticks." The U.S. said they can be since they are based on real-world observations, but SeAH dubbed this proposition "illogical and contrary to the evidence."

"Of course, one can propose an infinite number of thresholds based on an infinite number of 'real-world' observations. One might, for example, state that a 'large' effect is measured by the difference between the average weights of elephants and ants. In order for Commerce to justify its preferred 'real-world' threshold, it must demonstrate that there is some reason to believe that the differences Professor Cohen chose to use as the basis for his thresholds provide a reasonable yardstick for determining whether observed price differences are 'large,'" SeAH argued. "... Commerce has failed to do so."