Oral argument — Stupp Corporation v. United States

Our initial brief in this case had a lot of discussion of statistics, but I want to assure you that we're not asking you to decide this case based on statistical principles. I wish I could, because I think they're pretty clear here, but I recognize that you're a member of law, not a council of math teachers. Our claims here are really about administrative procedure, and in particular, what an agency has to show when it purports to apply fixed numerical thresholds that have not been adopted through the statute or through notice and comment rulemaking. The statutory provision at issue here concerns the method used to calculate dumping margins. Commerce is normally required to use an average-to-average, what we call A-to-A methodology, but it's permitted to switch to what's called an average-to-transaction, or A-to-T methodology, if certain findings are made. Now, where the statute establishes conditions that must be met before commerce can take specific actions, it is commerce that has the burden of proof to establish that the statutory conditions are satisfied, and in that regard, I point to your 1994 decision in a case called Cresswell Trading, which addressed a different statutory provision, but established that principle. Now, what commerce does to determine if the statutory conditions have been met, whether there's a pattern of prices that differ among purchasers, regions, or time periods, it's something that's called the differential pricing analysis, or the DPA. Can I ask you, this is Judge Toronto. How much, let's assume, or assume for purposes of this question, that commerce acted within its discretion in rejecting certain portions of your case brief. How would that assumption affect your, let's call them the merits criticism arguments you are making? It, in my view, of course, in my view, it doesn't, but it doesn't, because our argument is, first of all, the burden was on commerce to reduce substantial evidence in support of its position, which it hasn't done. Second, that commerce has an obligation, since it hasn't adopted this DPA through notice and comment rulemaking, it's required as a matter of law to decide each case as if the policy didn't exist. They have to come to each case fresh. And that means any factual determinations they make in the case have to be based on substantial evidence, as factual determinations always have to be based. When you apply a numerical threshold, for example, if you say that a dumping margin of less than 0.5% is always de minimis, what you're really saying is a factual claim that a dumping margin at that level is always insignificant. And in Washington Red Raspberry, which we rely on, this court said, look, if you're going to use those numerical thresholds without establishing a rule, then you have to show in each case that the thresholds are appropriate. And it's not just that you have to show that the dumping margins, which were the issues in Red Raspberry, were less than 0.5%. You have to establish that 0.5% is the appropriate standard for de minimis in this particular case. And so in that case, you affirm, but only after commerce showed that compared to the margin of underselling and prices in the market, the dumping margin, which was, I think, in that case, 0.42%. But I want to see if I understand this right. So you don't think that if commerce, for example, in Xantham-Gum or some of the other early-ish cases, had actually set out a sound basis for the use of every aspect of differential pricing analysis that is relevant here in circumstances that are present here, that commerce would have to, in this proceeding, do more than, say, the only challenges to our use of the analysis here have already been addressed? See the analysis that we adopted in the earlier proceeding, or would that incorporation by reference be enough? The incorporation by reference would not be enough because the standard is, in this case, commerce has to find that a pattern exists. And in this case, that finding has to be based on the evidence in this case. And so the fact that in a different case, a Cohen's deal... I was trying to build that into my question, that the facts would be the same, but in which case an incorporation by reference would just be a shorthand for repeating everything, subject, of course, to your having introduced new factual evidence criticizing, say, a methodological point previously adopted. Well, see, I disagree with you because... I know I'm not supposed to disagree with judges, but... No, no, no, go on. The 0.8 standard for Cohen's deal, for example, the facts are never the same. Our case is different than Xantham Gum. And what Washington Red Raspberry says is you have to explain in this case why that's The fact that in Xantham Gum, to go to the Raspberry, if commerce said in Xantham Gum, 0.5 is de minimis, that doesn't mean 0.5 is de minimis in this case because, as you said in Red Raspberries, you have to look at the market and what's going on. And so the notion that commerce could say, look, 0.8 is the appropriate cutoff, because they haven't gone through notice and comment rulemaking, they have to come to our case as if the policy doesn't exist. That is what you said in, I'm forgetting the case, Unwired Planet, which is based on a DC Circuit case called Pacific Gas, and we've quoted it extensively in footnote 38, making the explanation by the DC Circuit is terrific at laying out why an agency can't circumvent the requirement of substantial evidence by saying, well, it's our policy. And the other problem is the Xantham Gum evidence is not on the record in my case. So I can't respond to it. And I don't know what, there are literally thousands of Commerce Department cases. I've had cases in the past where I've addressed all of this and Commerce has left the information on the record. I've put in evidence that this method doesn't work, that it finds patterns in purely random data. Commerce hasn't rejected that, said you can't put that on the record, but they've said it doesn't matter. Which cases are actually the same? The legal standard is when Commerce doesn't go through notice and comment rulemaking, it comes to each case as if a new, and it has to satisfy the substantial evidence test. Mr. Whetton, this is Judge Bryson. Let's assume we don't agree with you on that point and instead view what Commerce has done as to adopt an interpretive rule within the meaning of the APA, and that that rule can be upheld upon a finding of reasonableness. Now, assume that's the case with respect to the DPA and each element of the DPA, that still leaves us, of course, with the problem of what is the proper application of that approach in this case? Unfortunately, I'm afraid that gets us into the statistics. So I would like you to explain, at least for my benefit and perhaps for the others, why it is that the Commerce argument, which was made in this case and in many of the previous cases, that your argument that the Cohen's D approach is not valid unless there is a normal distribution, roughly equivalent variance, and a sufficient size. Commerce is arguing that those don't apply because it is looking at the entire universe of sales and not simply looking at a sample. Can you respond to that? Sure. The problem is Commerce could, let me back up, Commerce could use the DPA exactly as it is if it justifies it. But the problem here is Commerce justified the 0.8 by saying, this is what big boy statisticians do. So we're just doing the same thing statisticians do. And when we said, no, you're not, which was not rejected as new factual information, the arguments were left on the record. The citations were taken off. Commerce's response was, well, but we're not using it the way that Professor Cohen and all these other people have described it. That leads right into your decision in Midcontinent where you said, if Commerce is going to say our methodology is justified because it's an application of widely adopted statistical principles, then practices, then they've got to use those widely adopted statistical practices. They can't have it both ways where they say 0.8 is appropriate because that's what statisticians do. But the limitations that statisticians require don't matter because we're not doing what statisticians do. OK, but I really want you to focus, if you would, on the question of whether the Cohen's analysis is applicable without regard to the limitations that you argue for in your brief in a situation in which Commerce has the entire universe of sales before it. OK, well, let me just say no. The answer is no. It's not applicable where it has the entire universe. Commerce's statements about normal distributions and Cohen's d is what statisticians refer to as a parametric test, and it can only be used when the data follows the bell-shaped curve. It's a mathematical—oh, sorry. The cutoffs that Professor Cohen adopted are rules of thumb based on looking at that precise type of distribution. He never said 0.8 makes sense in other situations, and as we laid out in our initial brief, it doesn't make sense to apply it when the data is not normal. So, Mr. Winton, can I just focus on what I think I'm hearing Judge Bryson's question is, and I would have expected you to give—to open your answer to his most recent question by saying the opposite. I would have expected you to say the Cohen's d preconditions apply whether you are treating a sample or whether you are treating the entire universe, and the problem is that those conditions do not—have not been shown to be met here. This is why you have been appointed to the court, and I'm just an advocate. That was what I was trying to say. Okay, but sampling has nothing to do with the question that Cohen's d addresses, which is within some set that you have, whether it's a subset or the universe, there are significant differences within the group you're looking at. Yes. I think my time is up. I have reserved time for rebuttal, but I can keep talking or just— Well, it would take another couple of minutes to add what you think is worth adding to with being conscious that you're on borrowed time. Thank you. I feel on borrowed time. The issue here—the other part of the test is the ratio test, the 33 and 66 percent numbers, and those numbers are purely made up, and there's no justification for them at all, except Commerce says, we think these are the right numbers. Again, that is by definition arbitrary. We've had cases where I said, why do you say 33 and 66? Why isn't it 20 and 80 or 40 and 90 or something like that? Commerce said, well, you didn't explain why those are better numbers. But the problem is, we can't explain why our numbers are better than Commerce's because we don't know why Commerce chose these numbers. All we know is that Commerce said, when these numbers are satisfied, we think the methodology should change, but not why those cutoffs require the change. It's just, at these levels, we change. We've pointed out in other cases that this doesn't work. Commerce rejected that. We showed that if you ran down purely random numbers, you would find a shockingly high pass rate. Commerce doesn't seem familiar with the concept of false positives. Can I just double check something with you? We've addressed bits and pieces of this overall differential pricing analysis in a few of our cases, two aspects, I guess, in mid-continent steel, the 0.8 saying, OK, in general, that's problem of weighted averaging or not. We addressed some aspect of it in, what is it, Apex. Have we addressed the, I guess what they call, is this the ratio test, the 33-66 breakdown in any of our decisions? I don't believe so. How about Apex? It's 862 at 1349 to 50. I think that addresses that issue. I don't have it in front of me, so while the opposing counsel is speaking, I will quickly. All right. The last point I would make is the meaningful difference test. I think we've laid out in our brief our position on that. It's clear that in, or it's probable in some circumstances, the average-to-average methodology can do what Commerce calls masking dumping. But it's also clear that the average-to-transaction methodology can, in certain circumstances, create dumping. And so when you have a difference between the results, it's not enough to say there's a difference. You need to say, why is there a difference? Is it because average-to-average was masking, or is it because average-to-transaction is creating? That's a required part of the explanation, I would submit, and one that doesn't exist in this case. Thank you very much for the additional time. And now I have to go. Okay. Thank you, Mr. Winton. And we'll hear from Mr. Kipura now. Thank you, and good morning, Your Honors, and may it please the Court. Appellants has raised three arguments on appeal, all of which have previously been presented and rejected by this Court. An appellant presents nothing new which would distinguish this case from any of the Court's previous decisions. Where have we addressed the questions of the preconditions of a sufficient size, comparable variance, and normality of the distribution for the Cohen's d-test? Your Honor, the Court's mid-continent decision addressed the Cohen's d-test and the use of the 0.8 threshold for determining that. I don't think that's the same point. The 0.8 is quite a different point, it seems to me, from the question of whether Cohen's d requires the preconditions of normality, roughly equal variance, and sufficient size. Can you tell us why those requirements are not required for purposes of applying Cohen's d in a case such as this one? Well, yes, Your Honor. It goes to the point that was being discussed earlier where the entire universe of sales is available to commerce when it's making this determination. Why does that make a difference? This is the question Judge Toronto asked of your opposing counsel, and I need to understand why that makes a difference. It may require us to get into the math a little bit, but please explain it to me, because I don't think that in any of the various decision issues and decision memos that commerce has written on this, they actually explain why that is true. They have said that it's true, but I haven't seen an explanation. So this is your chance. Yes, Your Honor. And the purpose of it is that the Cohen's d test is to examine whether the price differences that are found to exist between different purchasers, regions, or time periods are significant. And then commerce turns to the ratio test to determine whether or not those differences are sufficient to find a pattern. So in other words, the purpose of the test is to identify the differences within the universe of sales to determine if differences in prices are significant. And so that's why it's significant or why it matters that the entire universe of sales is available, because we're not comparing those sales against some uniform or randomized group of sales. We're comparing them to all of the sales within the United States and trying to determine whether or not sales to particular regions or purchasers or time periods are different from the average throughout the region. And then the second step comes in with the ratio test to determine whether or not those differences are sufficient to find a pattern. Well, I don't understand why, I still don't understand why something like the normal distribution isn't important, even for the purpose that you're using it, that you're using Cohen-Sea. For example, if you had a number of sales that were all at a specific amount with only the tiniest variations among those sales, tiny variations, the standard deviation for that group would be extremely small. I think we can agree on that, right? Yes, sir. Right. Okay. And therefore, that would tend to drive the Cohen's d number up. So it does seem that the normality issue is important to Cohen's d, just for that example alone. Why not? Well, again, Your Honor, because what is being compared, what commerce is trying to determine is, again, this concept of mask dumping. And so the comparison is attempting to sort out whether or not the differences are sufficient that the different prices that are going on in the different regions are a significant change from the overall average. I understand that's the purpose. I'm just not seeing how Cohen's d achieves that purpose when the conditions that Cohen himself admits are applicable are not met in a case like this. Well, Your Honor, again, the court has previously upheld the use of the differential pricing analysis. To my knowledge, and I've read all the cases here, the court has not grappled with the issue that we are discussing right now. The question of whether Cohen's d is applicable to a situation in which the preconditions are not satisfied. I don't see any case law other than cases from the CIT that address that issue directly. Well, Your Honor, the question before the court is not so much whether or not there's a perfect reliance or how Professor Cohen's analysis fits in. It's whether or not commerce's application of the methodology comports with the statute and whether or not it's a reasonable method of determining whether or not there are significant differences that demonstrate a pattern. So can you tell me why it's reasonable? Yes, Your Honor, because it is commerce's methodology to to determine whether or not the the differences that it's seeing in the prices are significant. So what the court has previously, so you're saying that the you're saying that the method is reasonable because commerce uses it to achieve a particular end. But that skips over the critical question of why is it reasonable? Your Honor, as the court has previously stated, the statute is silent as to how commerce should make this determination. And so commerce has explained that this is the methodology that it has employed in order to make that comparison. Now, if the court's question is about, you know, whether or not, you know, this particular mathematical model is is reasonable compared to some other mathematical model, again, we would say that this particular one achieves the purpose of making the determination as to whether or not the differences are significant. This is Judge Chen. Are you saying it's reasonable to do this differential pricing analysis as articulated in Xantham-Gum because commerce has to pick some kind of framework and this is the framework it's chosen and it has support out there among statisticians and and that's all it needs. The commerce needs to articulate. It doesn't need to articulate why the framework it has adopted is, I don't know, working for its intended purpose in every different circumstance. I guess that's what I'm wondering. I mean, we have to figure out how to understand our review objective here on reasonableness of commerce's selection of using this methodology. Is it just the adoption of the methodology in general or is it the adoption of this methodology in the context of this case where, assuming for purposes of this question, the premises for why you would use Cohen's deed are not really present here on the facts of this case? Well, Your Honor, I would begin by referring the court back to its Thai pineapple decision in where commerce receives that deference when selecting the methodology. And in this case, commerce has explained why it's employing the differential pricing analysis and how it's doing that and the results that it achieves. What appellant has been arguing is that there's some other way that the Cohen's deed test could be performed or might reach a different result. But what the question before the court is, is whether or not commerce's explanation is sufficient to explain what it did and whether or not that procedure is reasonable under the statute. So you agree that commerce's explanation has to be sufficient to justify what it is doing here. And I think you just said that. And it seems to me, as I understand the discussion, there are kind of two possible things going on here. One is commerce said, we rely on it because some well-regarded statisticians rely on what we're doing. And then the question is, well, do the conditions here satisfy what the well-regarded statisticians have laid out? And then the second possibility is that we're actually departing from what the statisticians do because we think there's a reason to depart from it. And I think the only aspect that we've talked about this morning, at least, is that the full universe analysis case, not a sampling case. And I'm not sure there's been given a, at least to my mind, a trackable explanation for why that's an actual difference here. Well, Your Honor, I think just to address the point as to whether this issue of the preconditions for it, there's nothing in the statute that would require commerce to specify that in order to use a particular mathematical model, that it has to meet the certain preconditions that were laid out by a particular statistician. What commerce has explained here is that it is using this model, this Cohen-Z test, but specific facts of this case in order to make that determination as to whether or not there are significant differences. And as to the question of a departure from what may have been cited in the academic text about using a normal distribution, again, we're not comparing this to a random sample. We're comparing it to the sales in the United States as a whole. So commerce's application of the mathematical model is acceptable because it's using it to apply the statute in order to make its determination as to whether or not there are significant differences that are sufficient to establish a pattern. And on that point, I just wanted to respond to one issue that was raised about whether or not this is a determination that requires notice and comment. As the trial court explained, in fact, twice, in both its initial determination and its decision... not to be that this had to be done by notice and comment, but rather, given that it was not done, that the justification has to appear in the particular as-applied case. He hasn't made a separate argument, I think, that this had to be done by notice and comment just about what the consequences are given commerce's choice not to do it. Well, yes, Your Honor, but the issue there is that it's been described as a policy. And as the trial court made clear, it's the methodology being employed by commerce is not a question of policy. It's a question of statutory interpretation. And if the court... I would just like to briefly conclude, if that's all right. Okay. And again, what the trial court found was that this was not a question of policy, that the differential pricing analysis is a methodology that is used by commerce in order to find facts. And that's the distinguishing characteristic between the analysis of the differential pricing analysis versus what the court was employing in Washington Red Raspberry. Just one quick follow-up question. You said you understood this DPA as a matter of statutory construction. I thought that your view was it was really more of a gap-filling measure, rather than construing a particular term in the statute, that it was the case that the legislature left it unresolved as to what the right approach to, or for that matter, the right numbers would be for determining whether there is dumping or whether the AA or AT methods should be used, but instead left it to commerce to come up with a device for doing that as a matter of gap-filling. Is that correct? Is that your position? Yes, that's correct, Your Honor. In section 1677F1, and we go down to paragraph D, so the relevant portion is whether there's a pattern of export prices for comparable merchandise that differs significantly among the purchasers, areas, and regions. Those terms that differ significantly and the pattern, this court and the Court of International Trade have found that Congress was silent on those terms, and so the differential pricing analysis and then subsequently the meaningful difference test is the method by which commerce determines whether or not that provision has been met. Okay. Okay. Thank you, Mr. Kipora. Mr. Winton, you have three minutes reserved for rebuttal. Thank you. I'm really feeling inadequate as a lawyer because I've been looking at the apex decisions that were, there are two apex decisions, and neither of them seems to address the argument we're making about the ratio test. They talk about how commerce calculates margins under the mixed methodology once it's satisfied that the ratio test is 33%, but it's never addressed, are 33% and 66% reasonable numbers? And to be reasonable, there's got to be a reason for them, and there isn't one. All commerce has said is, we think these are the right numbers. That's all they've ever said. That's not something you can review. It's not something we can argue, and I would submit that's something totally arbitrary. I'll just finish up on this issue of whether this methodology is an interpretation of the statute. Well, first of all, you're right. Commerce has said it's a gap-filling interpretation, but if it's an interpretation, I ought to be able to use some canon of construction to interpret it, and commerce has done that. They looked at what does the word pattern mean, and they looked at the dictionary definition of pattern, and they interpreted that, and I think they were right. I had suggested that interpretation, so I'm very happy with it, but they don't have any sort of tying this to the statute. Why is it 0.8? Why is it 33% and 66%? There's nothing in the statute that leads you to those. Those are factual conclusions. The methodology that we're going to calculate something that we'll call Cohen's d, that we'll calculate a pass rate, yes, those are methodologies, and if they're reasonable, they're reasonable. But in evaluating the output of those methodologies, commerce is making a factual determination that 0.8 is large, that something happens at 33%, and something else happens at 66%, although commerce has never told us what it is. And so those are factual findings, and they have to be supported by evidence. You know, you asked earlier about what if commerce and Xantham gum had put on evidence. Well, you know, we should address it, but if you look at the evidence in Xantham gum, it's the effect size stupid article, because that's the name of the article, not co, yeah, co, I think that's a better way of putting it, which says it only works, you know, Cohen's d can only be used if the variances are approximately equal and if there's a normal distribution. So, you know, there's no evidence here. Mr. Winton, this is Judge Chen. I was wondering if commerce shouldn't use Cohen's d here for the reasons that you had given, then what should commerce have done? What would have been the better alternative? Well, there are two... Just for the Cohen's d part. Right. Let me give you two possibilities. One is they could have used Cohen's d if they justified it somehow, but not by saying this is what statisticians do. If they said, you know, we've done calculations and empirically we can show that a Cohen's d of 0.8 means the following, you know, they're allowed to do that. What they're not allowed to do is say we're doing what statisticians do. And if you were saying, well, how would statisticians do this? There are entire books written on nonparametric methods of measuring effect size. And commerce could have used this. I'm not a statistician and I'm, you know, until somebody puts it in front of me, I'm too lazy to try to learn all this. I understand that. But you have to understand commerce has a railroad to run, right? They have a lot of different cases and they have to set up frameworks to review different cases and fact patterns. And you're saying, well, under these particular conditions, in my particular case, this framework that you have announced and adopted and say you want to apply everywhere, it just won't work under my factual scenario. So, but then commerce, I would think would expect you to come forward and say, don't use that. Use this other framework. And here's the other framework that really is far more rational than your unreasonable application of Cohen's d. Well, you know, commerce asked for comments after Xanthan gone. And the comment I made, I submitted 50 something pages of comments, which, and the conclusion was you should use a nonparametric test. I didn't say which one. Commerce, you know, could evaluate them. But there are tests that can be used when the data doesn't follow normal distribution. And, you know, commerce simply ignored that. That's, you know, I understand they have a railroad to run. But on the other hand, we have administrative procedures to follow. And commerce didn't follow them here. Thank you. The, yeah, the portion of APEX I was referring to is the very last section of the second APEX case where they had the question before them as to whether the AT and the AT test was applicable. And they said it was because the number was in the 33 to 66 range. It was 65.31, which the court blessed. So it seems to me, unless there's a reason that that's not, doesn't constitute a blessing of the three part test. It looks to me like that's what they're saying is that's fine. I would say it's an issue that wasn't litigated in that case. The appellant didn't say to you this 33 and 66 percent have no basis. In fact, they accepted that a mixed methodology was appropriate. And they argued about how you calculate the results. They said that, you know, if you didn't address, if the issue wasn't raised, it's not a binding decision. All right. Let me, I don't know, were you done? Because I had a couple of administrative issues I wanted to raise that just that I don't want to use up your time. No, I think I've said enough. Okay. There were a couple of things that were missing from the joint appendix that seemed to me to be at least potentially important. One was the Kelly article. You cite that in your brief as one of the few articles that's pertinent to this whole question of Cohen's D. And you have a cite to 10626 of the joint appendix. 10626 does not exist in the joint appendix. So if you would send us a copy of that article, I would appreciate it. The other thing is that you cite at page 9789 of the joint appendix, your redacted brief that you submitted after the department struck your initial case brief. But you give us only one page of that in order to understand the argument you were making to it would be helpful to have the entire brief that you submitted after the redactions. So if you could submit that to us as well, I would appreciate it. Thank you. We'd be delighted to. All right. Okay. With that, I'll thank both counsel and say that the case is submitted.