Federal Circuit Says Commerce Can't Use 'd' Test Without Adhering to Statistical Assumptions

Prost, writing for the court, said the present case picks up where the appellate court "left off" in its 2021 Stupp Corp. v. U.S. decision, which initially called Commerce's use of the test into question. Prost said that Commerce's claim that the agency didn't have to adhere to the assumptions since it was using the entire population of data as opposed to samples "strains credulity."

In remanding the antidumping duty investigation on utility-scale wind towers from Canada, the Federal Circuit also sent back Commerce's rejection of a supplemental cost-reconciliation item submitted by respondent Marmen that was meant to correct certain purchase information that hadn't been properly converted from U.S. dollars to Canadian dollars. However, the court upheld Commerce's decision to "weight-average (or smooth)" Marmen's reported steel plate costs.

The Cohen's d test is the first step in Commerce's differential pricing analysis and is used to measure whether a respondent's U.S. prices to a given "purchaser, region, or time period differ significantly from the prices for all other purchasers, regions, and time periods." If the Cohen's d value is equal to or greater than 0.8 for any test group, the sales "pass" the test, meaning the sales are "significantly different from the sales prices in the comparison group." From there, Commerce uses the results of the test as an input in a ratio test to find the proportion of the respondent's U.S. sales that pass the test to see if a pattern exists.

If a pattern does exist, Commerce calculates the company's dumping margin by comparing its average home market sales price with an individual transaction in the U.S.

The Stupp court initially questioned Commerce's use of the d test, remanding so the agency could explain if it still works when basic statistical assumptions such as normal distribution and roughly equal variances are not met (see 2107150032). On remand, the agency offered two justifications for its use of the test: it didn't need to adhere to the assumptions when it used the entire population of data and that the assumptions didn't matter, since the agency doesn't use the test to determine dumping margins but "only to show whether prices differ significantly" as one part of its analysis in determining whether to use an "alternative comparison method" to address targeted dumping.

Addressing the second justification first, Prost said the court declines to "conclude that a flawed output in step one of a calculation can be transformed into meaningful data by further manipulating it.”

Regarding the first argument, the CAFC said it matters not "whether the data comes from a sample or a population" in producing an "accurate Cohen's d coefficient." Citing the work of the test's founder, Dr. Jacob Cohen, Prost said the d coefficient is a measure of “effect size” between the means of two groups and for it to have any meaning, it's necessary to maintain the statistical assumptions. In addition, it's possible to "define measures of non-overlap" associated with d, though the relationship between the non-overlap and d figures "is based on the assumption that the data sets being compared have normal distributions.”

The Federal Circuit said Commerce, on remand, "may re-perform a differential pricing analysis," though it may not rely on the Cohen's d test for data sets "like those here." This finding doesn't bar the agency from "fashioning and justifying a statistical analysis that uses some of the ideas underlying Cohen’s analysis of group differences as long as the resulting analysis is itself justified as sound for gauging differences in the data sets at issue," the decision said.

The appellate court also remanded Commerce's rejection of a minor correction to Marmen's cost-reconciliation worksheet to account for an exchange rate. In 2018, Marmen didn't account for the exchange rate of U.S. dollars to Canadian dollars for its U.S. dollar-denominated purchases, though it did so in 2019. Relying on its 2019 records, Marmen "inadvertently omitted the conversion from USD-to-CAD in the cost-reconciliation worksheet for Item L," which covers affiliated purchases of wind sections from Marmen Energie, for the first half of the investigation period (July-December 2018).

To correct the error, Marmen added a new line to adjust the exchange rate to Canadian dollars, though Commerce rejected it as untimely. The Court of International Trade overturned this rejection, though it ultimately sustained the agency's refusal to accept the correction on the grounds that the exchange rate conversion would be "duplicative" of other adjustments and that Marmen's specific exchange rate is "unreliable."

The Federal Circuit rejected both of these claims on appeal. In claiming the rejection would be duplicative, Commerce relied on representations from Marmen's auditors that the cost reconciliation "was complete." However, the court said these statements "were primarily made prior to the discovery of the omitted exchange-rate conversion." As for the correction's reliability, Prost said Commerce took an "unnecessarily complicated view of the record." The records Marmen relied on were meant to show that the 2018 sales were recorded in U.S. dollars and should have been converted. "That is what these records show," the court said.

Lastly, Prost sustained Commerce's decision to smooth Marmen's reported steel plate costs. The agency "determined that the reported steel costs did not reasonably reflect the cost of production," since Marmen's suppliers didn't charge different prices for plates of different "grade, thickness, width, or length." This means there should be "little difference in plate costs for different dimensions and grade based on record evidence on a per-unit weight basis," yet the agency did find differences unrelated to physical characteristics, such as the timing of production.

The court said Commerce didn't arbitrarily disregard its standard test for when smoothing may be used, since it’s standard practice for addressing cost reporting for finished products and not inputs, as the agency assessed here. Prost also held that Commerce's factual findings weren't unsupported by substantial evidence.

(Marmen v. United States, Fed. Cir. # 23-1877, dated 04/22/25, Judges: Sharon Prost, Richard Taranto, Raymond Chen; Attorneys: Jay Campbell of White & Case for plaintiffs-appellants Marmen, Marmen Energie and Marmen Energy Co.; Robert Kiepura for defendant-appellee U.S. government; Maureen Thorson of Wiley Rein for defendant-appellee Wind Tower Trade Coalition)