When Statistics Matter
Appellate litigation is generally the final throw of the dice, so to speak, in U.S. trade regulatory disputes. Litigants, therefore, should carefully consider both the substance and presentation of their appellate arguments. To illuminate what may seem like an obvious point, this note focuses on two recent decisions of the U.S. Court of Appeals for the Federal Circuit (CAFC).
The CAFC is almost invariably the court of last resort in appeals from decisions by the U.S. trade enforcement agencies – the Department of Commerce and the International Trade Commission. The CAFC tends to exercise a large degree of deference to the decisions of the agencies it reviews. The U.S. Supreme Court in Chevron instructed courts to apply a deferential standard of review toward decisions of administrative agencies – an instruction that the CAFC has taken very seriously, peppering its opinions with statements regarding the “tremendous deference” afforded to Commerce, and the agencies’ “broad discretion”. These statements would appear to bode poorly for parties seeking to overturn such determinations.
There are sources of authority, however, that the CAFC appears willing to evaluate with less deference toward the agencies’ interpretations. One such source is obvious – namely, the statutory texts enacted by Congress to govern trade remedy proceedings – although as I discussed in a recent note, even this source is subject to conflicting modes of interpretation by the Court of Appeals. Intriguingly, another external source of authority for which the CAFC has been willing to employ a searching inquiry is the statistical analytics of numeric data. In two recent decisions, the CAFC has undertaken a probing review of statistical concepts, which resulted in skepticism toward the Commerce Department’s practice.
A closer look at the Court’s treatment of these two categories of sources is worthwhile for parties preparing their appellate strategy.
The Court’s treatment of statutory text has not been linear. The Court has, on occasion, taken a bold position by parsing the statutory text, but then, when blowback erupted and the broad implications of its position became apparent, it has retreated from its interpretation without formally overruling it. One example of this dynamic involved the International Trade Commission’s analysis of causation – i.e., whether the material injury felt by the U.S. industry was “by reason of” the imports subject to an investigation. In Gerald Metals, Inc. v. United States, the CAFC ruled that, at least in an investigation involving a fungible commodity product, the ITC had to consider the possibility that the presence of fairly-traded imports in the U.S. market disrupted the causal link between the subject imports and the injury felt by the U.S. industry. The Court reiterated that ruling in subsequent cases involving commodity products, such as Caribbean Ispat Ltd. v. United States, and Bratsk Aluminum Smelter v. United States, in which it held that “the Commission is required . . . to directly address whether non-subject imports would have replaced the subject imports without any beneficial effect on domestic producers.”
This series of appellate decisions caused an uproar in the trade remedies practice, and the ITC added a step to its injury inquiry, while protesting that it was compelled to do so by Court directive, unmoored from the statutory text. The Court subsequently retreated from the potential breadth of Gerald and Bratsk, in Mittal Steel Point Lisas Ltd. v. United States. There the Court explained that it had not intended to reorient the ITC’s injury inquiry toward the future effectiveness of a proposed antidumping order; rather, it just meant that the ITC must consider, in a market involving a commodity product in which non-subject merchandise was present, whether injury could be attributed to the subject merchandise, as part of its “but for” causation analysis.
Another example of judicial sally-and-retreat is seen in Dongbu Steel v. United States, and JTEKT Corp. v. United States, in the context of a “targeted dumping” / “zeroing” dispute. The Court in those cases wondered how Commerce could interpret a single statutory text, 19 U.S.C. 1677(35), to mean one thing in antidumping investigations and something different in annual administrative reviews. These decisions released a flood of uncertainty among parties to antidumping proceedings as to whether Commerce’s express abandonment of zeroing in investigations (in response to several unfavorable WTO decisions), now compelled the abandonment of that practice in reviews as well. Commerce, however, persevered in applying zeroing in administrative reviews, and explained how a statutory provision could tolerate different meanings depending on the situation. The Court retreated from the implications of Dongbu and JTEKT, and affirmed Commerce’s explanation in Union Steel v. United States.
Recent cases involving statistics, however, demonstrate that the Court of Appeals does not always retreat after its sally, even in important but highly technical applications of the antidumping law. An example involves, again, Commerce’s “targeted dumping” methodology. The Court has affirmed several major aspects of that methodology (despite a series of unfavorable decisions by the WTO Appellate Body), concerning issues such as Commerce’s use of the “average-to-transaction” comparison method in annual administrative reviews even though the statute only authorizes that comparison method in investigations, and Commerce’s reliance on the difference in the magnitude of margins between the average-to-average and average-to-transaction comparison methods to satisfy the statutory test whether the former method can “adequately account for” any targeted dumping that is found to exist.
The Court’s deference to Commerce’s “targeted dumping” methodology appears to have reached its limit, however, when it came to issues regarding the use of statistics. In recent years, Commerce has applied a statistical test, known as “Cohen’s d”, to determine whether a foreign exporter has engaged in “targeted dumping” – i.e., a pattern of export sales that differ significantly among purchasers, geographic regions, or time periods. Put simply, Commerce applies the Cohen’s d test by calculating the difference between the means of the sales prices of the test group and the comparisons group of export sales, and then dividing that difference by the simple average of the two groups’ standard deviations.
Within the past nine months, the Court of Appeals has twice issued opinions rejecting various aspects of Commerce’s methodology. The CAFC found that Commerce’s methodology, although derived from a statistical test to determine whether pricing differences were significant, had deviated from important premises that were necessary to ensure that the statistical test would generate meaningful results. In Stupp Corp. v. United States, the Court reviewed the literature on Cohen’s d, and spotlighted “Commerce’s application of the Cohen’s d test to data that do not satisfy the assumptions on which the test is based.” The Court identified three characteristics of export price data that would render the test results suspect – namely, where the data groups being compared were small, were not normally distributed (i.e., not approximating the shape of a bell curve), and had disparate variances – and noted that Commerce applied the test without evaluating the data to determine the presence of those characteristics.
And just last week, in Mid Continent Steel & Wire, Inc. v. United States, the Court found that the use of a simple average, rather than a weighted average, in the denominator of the Cohen’s d ratio could distort the results, particularly where the groups of sales being compared were of unequal size. The Court relied on statistics-based reasoning to reject Commerce’s defenses of its methodology, and concluded that:
Commerce's job is not to follow a statistical test as explained in published literature for its own sake, but to implement the statutory mandate to determine when prices of certain groups “differ significantly.” In implementing a statutory mandate, an agency is not duty-bound to follow published literature . . . . But here Commerce embraced the Cohen's d statistics measure and relied on the literature for that measure in making its statutory significance assessment . . . .
As a practical matter, Stupp and Mid Continent may have a significant impact on antidumping proceedings, because Commerce applies the targeted dumping (or “differential pricing”) analysis in almost every case in which dumping margins are calculated. The machinery by which Commerce makes its targeted dumping determinations is detailed and complex, so revisions may be complex and far-ranging. Nonetheless, the Court in Mid Continent did not appear to be discouraged from pursuing its logic regardless where it may have led. It linked its statistics-based analysis to the underlying legal standard of judicial review, stating:
. . . Commerce needs a reasonable justification for departing from what the acknowledged literature teaches about Cohen's d. It has departed from those teachings about how to calculate the denominator of Cohen's d, specifically in deciding to use simple averaging when the groups differ in size. And its explanations for doing so fail to meet the reasonableness threshold (a deferential one, in recognition of expertise) for the reasons we have set forth.
Despite the nod to the “deferential” standard of review, this statement reveals a willingness on the part of at least some of the Judges on the Court of Appeals to engage in a rigorous consideration of statistics issues, and are willing to be skeptical of Commerce’s efforts to simplify the application of statistics in discharging its statutory obligations.
The focus of arguments presented to the Court, and the manner of their presentation, do indeed matter.
* * * * *
Through his 38 years of experience in the international trade regulatory field, Neil Ellis has litigated numerous cases involving the U.S. trade laws before the administrative agencies, U.S. courts, and in WTO dispute settlement proceedings. Please contact us at email@example.com with questions that you may have regarding trade litigation strategy.
 Smith-Corona Group v. United States, 713 F.2d 1568 (Fed. Cir. 1983).
 Borusan Mannesmann Boru Sanayi Ve Ticaret A.S. v. Amer. Cast Iron Pipe Co., 5 F.4th 1367 (Fed. Cir. 2021).
 132 F.3d 716 (Fed. Cir. 1997).
 450 F.3d 1336 (Fed. Cir. 2006).
 444 F.3d 1369, 1375 (Fed. Cir. 2006).
 542 F.3d 867 (Fed. Cir. 2008).
 635 F.3d 1363 (Fed. Cir. 2011).
 642 F.3d 1378 (Fed. Cir. 2011).
 713 F.3d 1101 (Fed. Cir. 2013).
 JBF RAK LLC v. United States, 790 F.3d 1358 (Fed. Cir. 2015).
 Apex Frozen Foods Pvt. Ltd. v. United States, 862 F.3d 1322 (Fed. Cir. 2017).
 The calculation is more complex than this, but it will suffice for purposes of this note.
 5 F.4th 1341, 1357 (Fed. Cir. 2021).
 2022 U.S. App. LEXIS 10767 (Fed. Cir. April 21, 2022) (reaffirming and expanding on its decision in a prior iteration of the same appeal, 940 F.3d 662 (Fed. Cir. 2019)).
 Indeed, the Court of Appeals recently remanded another case in light of Stupp. See Nexteel Co. v. United States, 28 F.4th 1226 (Fed. Cir. March 11, 2022).
 2022 U.S. App. LEXIS 10767, at *33 (emphasis added).