Full opinion text
POGUE, Judge. This matter is before the court following partial remand. In its prior opinion, the court reviewed the Department of Commerce’s (“Commerce’s”) affirmative less •than fair value determination for subject merchandise and the antidumping duty order and dumping margins subsequently imposed. Dorbest Ltd. v. United States, — CIT—, 462 F.Supp.2d 1262 (2006) (“Dorbest ”); see also, Wooden Bedroom Furniture From the People’s Republic of China, 69 Fed.Reg. 67, 313, 67, 317 (Dep’t Commerce Nov. 17, 2004)(final determination of sales at less than fair value)(“Fmai Results ”) amended by Wooden Bedroom Furniture From the People’s Republic of China, 70 Fed.Reg. 329, 330 (Dep’t Commerce Jan. 4, 2005)(notice of amended final determination of sales at less than- fair market value and antidumping duty order). During the investigation leading to the Final Results, Commerce used various methods to value the factors of production of the subject merchandise in order to approximate the normal value of the merchandise, and to make its determination regarding dumping. See Dorbest, 462 F.Supp.2d at 1265, h. 1. A number of these valuations were remanded for redetermi-nation pursuant to the court’s order. Id. at 1321-22. Commerce’s remand determination as to the following issues are now before the court: 1. Labor wage rate 2. Valuation of specific factors of production a. Hooks and connectors b. Resin c. Mirrors d. Cardboard e. Metal spare parts, non-scope metal canopies and other metal parts 3. Selection of surrogate companies to calculate financial ratios 4. Calculation of financial ratios 5. Calculation of Separate/Section A rate For the reasons discussed below, the court sustains in part and remands in part Commerce’s redetermination pursuant to court remand. STANDARD OF REVIEW The court reviews remand determinations for compliance with the court’s remand order. NMB Sing. Ltd. v. United States, 28 C.I.T. 1252, 341 F.Supp.2d 1327 (2004)(affirming International Trade Commission’s determinations on remand where the determinations were in accordance with law, supported by substantial evidence, and otherwise satisfied the remand order); see also Olympia Indus., Inc. v. United States, 23 CIT 80, 82, 36 F.Supp.2d 414, 416 (1999)(affirming after “review[ing] Commerce’s compliance with these instructions in its Remand Results” and finding the determination to be supported by substantial evidence and in accordance with law). In addition, any factual findings on remand must be supported by substantial evidence and the agency’s legal determinations must be in accordance with law. 19 U.S.C. § 1516a(b)(l)(B); see, e.g., AG der Dillinger Huttenwerke v. United States, 28 CIT 94, 95, 310 F.Supp.2d 1347, 1349 (2004)(holding remand determination to legal and factual standards set out in 19 U.S.C. § 1516a(b)(l)(B)). DISCUSSION 1. Labor wage rate In Dorbest, the court analyzed the Department of Commerce’s (“Commerce’s”) use of its linear regression model to calculate an approximation of the People’s Republic of China’s (“PRC’s”) wage rate. This method uses the reported Gross National Products (“GNIs”) and wage rates of the market economy countries meeting Commerce’s criteria to create a linear function that is then used to calculate approximations of wage rates based on a country’s per capita GNI. Commerce then specifically determines the wage rate that corresponds to the PRC’s reported GNI, and uses that wage rate as an input in further calculations. As-applied invalidity: In its initial analysis, the court first found that Commerce’s use of a data set that excluded countries that met its standards for inclusion did not constitute use of the best available information and was arbitrary and therefore not supported by substantial evidence. Dorbest, 462 F.Supp.2d at 1295 (2006). Thus, the court instructed that Commerce either “(a) justify why its data set constitutes the best available information; or (b) incorporate those countries meeting its criteria into the data set____” Dorbest, 462 F.Supp.2d at 1321. On remand, Commerce “reconsidered the data set used in this case and ... determined to include all data that meet [Commerce’s] suitability requirements and that were available at the time the 2004 wage rate was calculated.” Final Results of Redetermination Pursuant to Court Remand 4 (“Remand Results ”). In addition, Commerce engaged in notice and comment rulemaking to arrive at a revised methodology, detailed at Antidumping Methodologies: Market Economy Inputs, Expected Non-Market Economy Wages, Duty Drawback; and Request for Comments, 71 Fed.Reg. 61, 716 (Dep’t Commerce Oct. 19, 2006) (“Antidumping Methodologies ”); see also Expected Non-Market Economy Wages: Request for Comment on Calculation Methodology, 70 Fed.Reg. 37,761 (Dep’t Commerce June 30, 2005)(“Expected NME Wages ”). Commerce applied the results of its rule-making procedure to address the court’s data selection concerns here. Commerce has complied with the court’s order, and this aspect of the redetermination is not challenged by any of the parties, see Comments Pis. Dorbest on Commerce’s 2d Remand Redetermination 6 (“Dorbest Comments on Remand Re- • suits ”); the court therefore affirms this aspect of the redetermination. Distortions in Wage Rate Calculation: In Dorbest, the court also found Commerce’s use of its wage rate methodology unreasonable, absent sufficient justification for its calculations, particularly as Commerce had not engaged in notice-and-comment rulemaking when developing its methodology. The court chose not to examine each aspect of Commerce’s chosen methodology without the benefit of the agency’s reasoning. Nonetheless, the court noted that, while in theory the regression analysis used by Commerce could be considered a reasonable use of the best available information, Commerce’s regression model appeared to produce distortions. For example, when used to predict wage rates of market economies such as India (whose reported wage rate is known), the model predicts a wage rate three times higher than India’s reported wage rate. Dorbest, 462 F.Supp.2d at 1296-97. (Commerce found India’s economy to be closest to the PRC’s. Id.) Further, in performing its regression analysis, Commerce chose not to use regression through the origin (“RTO”) methodology, which employs the assumption that countries with a GNI of zero have wage rates of zero. The court noted the criticism that because Commerce used a regression model that had a positive y-intercept, rather than the RTO method, its regression model appears to overstate wage rates of low-income countries, so that the calculated wage rate is higher than the wage rates of any countries Commerce might otherwise look to as surrogates, making the methodology neither the most accurate nor fair. The court instructed Commerce to “reconsider its use of its methodology or an alternative method for determining the labor rate for the PRC in this case.” Dorbest, 462 F.Supp.2d at 1321. Commerce has now engaged in notice and comment rule-making regarding data selection and methodology for determining labor wage rates for non-market economies. See, e.g., Expected NME Wages, 70 Fed.Reg. 87,761; see also, Antidumping Methodologies, 71 Fed.Reg. 61,716. During this process, Commerce considered the parties’ suggestions for altering its methodology. Ultimately, Commerce determined to employ the same methodology (Ordinary Least Squares linear regression) as it had in its Final Results using a data set of all countries meeting its selection criteria. As such, Commerce’s regression model assumes that wage rates and GNI have an approximately linear relationship. See Expected NME Wages, 70 Fed.Reg. at 61,720 (stating that there is a “consistent relationship over time between wage rates and GNI”). This methodology was applied during the remand proceedings now at issue. In its remand determination, Commerce concluded that “there is no inherent distortion in the regression model that would lead to systematic overestimation of wages.” Remand Results 7. As to arguments regarding India’s calculated versus actual wage rate, Commerce asserts that India’s reported actual wage rate is anomalously low, and, as a result, any estimate based on data from other countries will produce an over-estimate of India’s wage rate. No party challenges this aspect of Commerce’s remand determination. In making its determination, Commerce considered, but ultimately rejected, methods for determining the wage rate, “including choosing a single wage rate from an economically comparable market economy, averaging the wage rates of economically comparable market economies, and running the regression only on economically comparable countries.” Remand Results 12. Commerce concluded that “none of these alternatives reduces the potential for distortion or increases either fairness or predictability.” Id. Ultimately, Commerce chose to continue with its prior methodology, considering it preferable to rely on “the broadest data set possible to arrive at a wage rate that is directly tied to each NME’s GNI.” Id. Commerce rejected use of data from a single, surrogate country because there was so much variation in wage rates that Commerce found the use of a single surrogate would “undermine the accuracy, fairness and predictability of [Commerce’s] calculations.” Remand Results 13. According to Commerce, the regression technique enhances predictability because the variation in wage rates at low GNIs means that choosing just one would lead to great variability in data used for different NME countries. Commerce also noted that because not all countries reported suitable wage rates every year, using a larger group and “averaging” the data by means of regression analysis was a more desirable option. Remand Results 16. The court finds that Commerce weighed this method and, on this record, reasonably chose to reject it. Commerce also considered other averaging methods. Commerce found that a method of averaging the wage rates of the eight reporting countries with the closest GNIs to a given NME was flawed for countries with small GNIs, because there were no reported wage rates from countries with GNIs lower than U.S. $420. Thus, all countries with lower GNIs would have wage rates determined based on averages for countries that all had higher GNIs (very likely leading to an overestimate). Commerce therefore found that its regression analysis was superior because it allows for wage rates to be calculated for NME countries with smaller GNIs than Commerce has data for. The court finds this conclusion to be reasonable. Commerce also considered (and rejected) limiting its regression analysis to data from a range of economies at comparable levels of development to each NME. Remand Results 18. Commerce found that “a basket of ‘economically comparable’ countries could be extremely small.” Id. Commerce found this problematic because the resulting regression “would be highly dependent on each and every data point.” Id. This, in turn, was problematic because for countries with GNIs less than U.S. $1000, “observed wage rates did not increase in lockstep with increases in GNI.” Id. In other words, the relationship between wage rate and GNI for countries with low GNIs is not necessarily linear. Rather, there is a “global relationship between wage rates and GNI.” Id. Commerce therefore concluded that this global relationship was better captured by “a larger basket [of data].” Id. Here, Commerce reasonably executed the task of choosing between two, imperfect data sets. Next, Commerce considered and rejected Dorbest’s argument that use of RTO methodology would result in less distortion for countries with low GNIs. Commerce gives multiple explanations for its choice not to use RTO, including the fact that the relationship between GNI and wage rates might not be linear near the origin. Remand Results 19. Commerce states that it is not trying to find the correct econometric model for wage rate as a function of GNI, but rather, is using all the data available and assuming linearity; thus, Commerce is simply “using a statistical tool, regression analysis, to generate a variable average of wages.” Remand Results 28 (emphasis omitted). Commerce explained that it does not “theorize on the precise nature of the relationship between wage rates and GNI near the origin. Rather, [Commerce] notes that the universe of relevant data does not indicate that the intercept relating to this universe of data is zero.” Remand Results 29. Commerce asserts that setting the y-axis intercept to zero would be inappropriate, because it would “impose a theoretical constraint on the calculation of expected NME wage rates.” Remand Results 10 (citing Dorbest, 462 F.Supp.2d at 1296-97). Lastly, Commerce explains that “because no country has a zero wage rate, the intercept will never, in reality, be zero.” Remand Results 10. Dorbest argues that here, there is “a strong reason to believe that when the independent variable is zero, the dependent variable is also zero.” Dorbest Comments on Remand Results 9. Dorbest points to discrepancies between reported wage rates for market economies and the wage rates predicted by Commerce’s regression, arguing that these discrepancies show that Commerce’s methodology still leads to distorted rates. Commerce examined Dorbest’s arguments during its remand determination, as well as Petitioner’s argument that “predicted wages based on regression without employing RTO are, on average, $0.02 higher than actual wages. In contrast, the regression results with RTO underestimate wages for the same lowest income countries by over $0.27 .... ” Remand Results 27. Accordingly, on remand, Commerce examined its methodology, and ultimately chose to use only “real world data” rather than include a hypothetical data point, such as the origin, and thus decided against using RTO. Commerce stated that “[Commerce] believes that it is neither reasonable nor fair to place a constraint on the calculation that is ‘theoretically consistent for a hypothetical country’ over a method that provides the best fit with the actual data, with the result to be applied to the actual NME countries and respondents involved.” Remand Results 28. Thus, Commerce chose not to hypothesize as to whether the relationship maintained its general linearity near the origin. The court finds that, based on the record here, Commerce’s determination to use only real-world data and not to perform RTO was reasonable. As a legal matter, on this record, Commerce may reasonably choose to rely solely on real-world data. As a question of fact, Commerce’s weighing of the record evidence was not unreasonable. In Dorbest, the court also noted Respondents’ allegation that a Generalized Least Squares (“GLS”) model would be more appropriate than an Ordinary Least Squares (“OLS”) model, as GLS could correct for the heteroscedasticity apparent in the data set used by Commerce. Dorbest, 462 F.Supp.2d at 1298, n. 33. The court did not examine the reasonableness of Commerce’s choice at that time, but noted that the suggestion (as well as the underlying issue of heteroscedasticity) was “an example of a step Commerce could consider on remand.” Id. On remand, Commerce considered the effect of heteroscedasticity in its data set, and Dorbest’s arguments that because of heteroscedasticity, a GLS regression technique would yield more accurate results than the OLS technique used by Commerce. Remand Results 31-32. Commerce chose to use OLS rather than GLS because of the difficulty of detecting and correcting for heteroscedastity. Commerce also determined that to take heter-oscedasticity into account, it would have to determine from year to year whether its data set was heteroscedastic; if it were, Commerce would have to use a Generalized Least Squares model of regression rather than an Ordinary Least Squares model; if not, it would use the Ordinary Least Squares model currently in use. Commerce found that such a system would not be predictable from year to year. Dorbest argues that “[u]se of GLS would correct for heteroscedasticity when present, and leave the result unaltered where no heteroscedasticity is present.” Dorbest Comments on Remand Results 18. Thus, Dorbest suggests that Commerce would not need to determine, from year to year, which methodology to use, because for homoscedastic data sets, the GLS equation becomes the same as the OLS equation. See, e.g., DamodaR N Gujarati, Basio EoonometriCS ch. 11 (4th ed. 2003)(“Bcmc Econometrics ”). Dorbest further argues that Commerce’s statutory obligation is “to calculate dumping margins as accurately as possible,” and that GLS regression methodology represents a more accurate calculation. Defendant-interve-nors AFMC argue that there are drawbacks to using GLS, particularly that “for GLS to work accurately, the nature of the heteroscedasticity must be known in advance.” AFMC’s Rebuttal Comments on Final Results of Redetermination Pursuant to Court Remand 17 (“AFMC Rebuttal Comments on Remand Results ”). In other words, “[i]n order to implement GLS correctly ... one needs to know the form of the heteroscedasticity — on which variables the variance depends and how it depends on those variables.” Id. Thus, AFMC points out that without more information on the relationship between wage rate and GNI, GLS might not be practical to implement. On remand, Commerce stated that it considers itself simply to be engaging in an exercise in averaging, and that “[t]he only relevant aspect of the regression is the line itself, which represents the variable average of the entire universe of data.” Re mand Results 82. Dorbest argues that Commerce’s protestations that it is not engaging in econometrics do not make it so. According to Dorbest, “the underlying assumptions of econometrics are extremely relevant to Commerce’s purpose, and Commerce cannot just define those assumptions away.” Dorbest Comments on Remand Results 17. It is true that Commerce cannot use econometric methodology in its calculations and then refuse to verify that its methodology produces accurate results by claiming that it is merely “utiliz[ing] regression analysis as a tool to arrive at a variable average.” To the court, Commerce’s tool is econometrics. However, oh remand, Commerce further explained that “even if [Commerce’s] regression analysis were to be characterized as an econometric model, the potential for heter-oscedasticity would not affect the unbi-asedness of the estimator.” Remand Results 32 (citing Basic Econometrics 381). As explained in footnote 11, supra, in the presence of heteroscedasticity, OLS is still a linear, unbiased estimator. As a legal matter, therefore, Commerce may, on this record, use OLS. It may not, however, as a factual question, be the “best” estimator. Commerce, however, also stated that “het-eroscedasticity is not easily detected or corrected for, as evidenced by the many tests and potential correction methods presented in the econometric and statistical texts in [Dorbest’s exhibits].” Remand Results 32. On this remand record, the court cannot disagree. While Dorbest cites its preferred statistical analysis, claiming that it provides a more accurate result, the record here discloses that, because of the “many facts and potential methods presented” in the econometric and statistical texts in [Dorbest’s Exhibits], Commerce was required to choose between a number of imperfect alternatives. Thus, considered in this context, a reasonable mind can conclude that Commerce chose the best information available. Commerce considered Dorbest’s arguments as to the merits of using GLS rather than OLS regression methodology, and reasonably rejected GLS methodology, and the data it produced, as it did all the alternatives proposed to its chosen approach. Accordingly, the court will affirm its choice. 2. Valuation of Specific Factors of Production A. Hooks and Connectors In its Final Results, Commerce valued Dorbest’s hooks and connectors using the Indian Harmonized Tariff System (“HTS[I]”) subheading 8302.4200, HTS[I] as the appropriate category for comparison. Dorbest challenged this determination, arguing that Commerce should have used expired HTS[I] category 8302.1009. The court noted that “Commerce [had] not articulated in what way the inputs ‘closely resemble’ the HTS[I] subheading description,” and remanded the choice of a category for the valuation of hooks and connecters in order for Commerce to “explain how the chosen subheading is rationally connected to the factor input” or to “find another suitable subheading or data set.” Dorbest 462 F.Supp.2d at 1309. In its remand determination, Commerce considered Dorbest’s description of its connectors and hooks, compared the subheading chosen in the Final Results to the inputs, and concluded that 8302.4200 was indeed the appropriate category for valuation. Remand Results 36. Commerce also “considered whether other HTS subheadings would be more specific to the inputs in question.” Id. at 37. Commerce found that Dorbest’s suggested subheading was not appropriate because, in addition to being expired, it was limited to “hinges and parts thereof,” whereas, after considering the definition of hinges, Commerce found that “the record evidence supports a finding that hooks and connectors are not hinges.” Id. at 37-38. Dorbest does not contest this conclusion, and the court finds that Commerce’s determination on remand is reasonable. B. Resin During Commerce’s initial determination, Dorbest suggested that Commerce value Dorbest’s resin applique input using expired subheading 3926.4009, HTS[I]. In its Final Results, Commerce chose not to use an expired subheading, but rather a heading it later argued before this court “most closely resembles Dorbest’s proposed but expired category, and comports with Commerce’s preference that the factor value information be contemporaneous.” Dorbest, 462 F.Supp.2d at 1310 (quoting Defs Resp. Court’s March 10, 2006 Qs 23). Specifically, Commerce valued Dorbest’s resin applique using HTS subheading 3926.3090, HTS[I]. The administrative record showed that Commerce based its decision on the expired nature of Dorbest’s proposed subheading. Memorandum from James H. Jochum to Jeffrey A. May, Issues and Decision Memorandum for the Less-Than-Fair-Value Investigation of Wooden Bedroom Furniture from the People’s Republic of China 170-71 (Cmt.l9)(Dep’t Commerce Nov. 8, 2004), P.R. Doc 1933, available at http://ia.ita.doc.gov/frn/summary/prc/04-25507-l.pdf (“Issues & Decision Mem.”). The court remanded, stating that Commerce had not “explained in what way the selected subheading resembles the suggested subheading, or, more importantly, the factor input.” Dorbest, 462 F.Supp.2d at 1311. The court ordered that Commerce provide an explanation or analysis for its choice on remand. The court also noted that subheading 3926.4009, HTS[I] “appears to apply to ornamental articles while subheading 3926.3090, HTS[I], appears to apply to furniture fittings,” Dorbest, 462 F.Supp.2d at 1310, and therefore further ordered that if Commerce were to find that the factor input was ornamental in nature, it would “need[ ] to then determine what the appropriate subheading would be to reflect that it is ornamental.” Dorbest, 462 F.Supp.2d at 1311. During the remand proceeding, Commerce determined that the resin applique input was ornamental in nature. This determination was based on information in the record, including, inter alia, Dorbest’s description of the input as “PVC and polymer used for decorating.” Remand Results 39 (citing Dorbest HTS submission at attch. l)(emphasis added). Commerce further determined that the proper classification for the input was neither Dorbest’s originally suggested subheading, due to it having already expired and Commerce’s stated emphasis, on contemporaneity, nor was it Commerce’s initially chosen classification, as that did not “appearf ] to include articles that are ornamental in nature.” Remand Results 40. Rather, Commerce found that subheading 3926.4099, HTS[I] most closely resembled Dorbest’s description of its input. In making its decision, Commerce asserted that it had found “a subheading that more directly relates to the input in question,” and that its new, chosen subheading was “specific to the input in question (ie., covers plastic ornamental articles) and is contemporaneous with the POI.” Remand Results 40. Commerce further noted that “there is a close overlap in the description of the HTS subheading selected, 3926.4099, and the HTS subheading proposed by Dorbest, 3926.4009, in that both appear to include plastic ornamental articles.” Id. Dorbest challenges Commerce’s determination on remand. Although Dorbest agrees with the determination that the input was ornamental, Dorbest Comments on Remand Results 38, Dorbest argues that Commerce continues to reject 3926.4009, HTS[I] solely on the basis that it predates the period of review, and that this is not in accordance with the court’s instructions. Id. (quoting Dorbest, 462 F.Supp.2d at 1310 (“the use of contemporaneity as the sole justification for its decision does not comport with Commerce’s statements that contemporaneity is but one of several criteria when selecting surrogate value information.”)). Dorbest further argues that Commerce improperly reopened the record on remand when it chose a subheading — 3926.4099—that was not proffered by any party in the proceeding below. Dor-best Comments on Remand Results 40. Dorbest argues that import statistics related to the subheading and considered by Commerce on remand were not part of the record during the administrative proceeding and should therefore not have been examined during the remand redetermination. Dorbest further argues that it was prejudiced by the short amount of time it had to respond to Commerce’s choice of a new subheading, stating that three weeks was simply not sufficient, particularly when compared to the length of the initial investigation. Despite the short amount of time, Dorbest raised a number of concerns with subheading 3926.4099, HTS[I] before Commerce. Dorbest’s argument characterizes the court’s decision as “overrul[ing] Commerce’s reliance on contemporaneity as the sole basis for rejecting the value proposed by Dorbest.” Dorbest Comments on Remand Results 46. This statement claims too much. The court’s remand order states that “[wjhile the court agrees that contemporaneity is an important factor to consider when evaluating surrogate value information, the use of contemporaneity as the sole justification for its decision does not comport with Commerce’s statements that contemporaneity is but one of several criteria when selecting surrogate value information.” Dorbest, 30 CIT at-, 462 F.Supp.2d at 1310 (citations omitted). The court further criticized Commerce’s determination for not explaining or analyzing how its chosen subheading reflected Dor-best’s description of the factor input. However, the court did not, as implied by Dorbest, set any specific standard for Commerce to meet in order to reject Dor-best’s suggested subheading. Although the court rejected Commerce’s choice of a subheading when the sole justification for that choice was Commerce’s preference for contemporaneous data, on remand, Commerce made its choice based on Dorbest’s description of its input, the overlap of its chosen subheading with Dor-best’s proposed subheading, and the subheading’s status as un-expired. Contemporaneity cannot therefore be said to be “the sole justification for [Commerce’s] decision” on remand. Dorbest, 462 F.Supp.2d at 1310. Rather, Commerce appears to have properly exercised the “wide discretion” accorded to it by Congress “in the valuation of factors of production” pursuant to 19 U.S.C. § 1677b. Nation Ford, Chem. Co. v. United States, 166 F.Bd 1373, 1377 (Fed.Cir.1999). Nothing in this court’s order limited Commerce to a choice between the two subheadings examined during the initial investigation. Further, as Commerce noted in its remand determination, “[Dorbest] has not argued that the [subheading chosen by Commerce] is not specific to the input in question.” Remand Results 46. Thus, Dorbest did not dispute the accuracy of the subheading, although it did advance several arguments as to why data regarding the subheading was distorted or inaccurate. Specifically with regard to the subheading data, on remand, Dorbest argued that data from InfoDrive India indicated that the subheading Commerce chose included mis-classified product. Dorbest Comments on Remand Results 44. Dorbest submitted data from InfoDrive India during the Remand proceedings “to illuminate the contents of HTS 3926.4099.” Id. at 44 (quoting Dorbest’s Comments on Draft Remand Redetermination (Apr. 5, 2007)(2d Remand Pub. Doc. 21 at 26-29 & Exh. 12)). Specifically, Dorbest claims that the data show inclusion in 3926.4099, HTS[I] of numerous items that should be classified in other categories, including plastic pellet samples (boxes), plastic beads, resin clocks, figures (made of polyresin) and photo frames. See Dorbest Comments on Remand Results 44-45. Dorbest states that “[i]f Commerce’s new [chosen subheading] underwent the same scrutiny ensured by the procedural safeguards of the full proceeding below ... additional information could have been developed to illuminate the problems in this [subheading].” Id. at 45. Regarding the problems Dorbest “illuminated” during the remand proceeding, Commerce found that the Infodrive data Dorbest put forth “cannot be relied upon because they fail to account for a significant percentage of imports reported under the HTS subheading and because the data are not reported in a uniform, measurable quantity.” Remand Results 47. Commerce noted that the Infodrive data excluded data from several countries which “represent 54 percent of total imports, by quantity, excluding imports from the PRC, Thailand, and South Korea.” Id. at 48 (citing Dorbest’s Comments on Draft Remand Redetermination (Apr. 5, 2007)(2d Remand Pub. Doc. 21 at Exh. 12)). Commerce also found that imports for all but two countries listed were incomplete. Remand Results 48. Commerce also argued that Infodrive data was not reliable because of the multiple units of measurement under which the Infodrive data is classified. Id. These problems with the data, according to Commerce, differentiated the situation of resin from that of mirrors, see infra at 27-29. Commerce therefore reasonably determined that the Infodrive data was not reliable enough to call into question the MSFTI data. Nor is Dorbest’s argument that three weeks was an insufficient amount of time for it to properly comment on the draft redetermination persuasive. As Defendant notes, remand determinations typically involve “a limited number of issues ... as compared with a full antidumping investigation.” Def.’s Resp. to Comments upon Commerce’s Redetermination Pursuant to Remand 58 (“Def.’s Resp. to Comments on Remand Results ”). Although Dorbest states that it did not have sufficient time, it does not allege specific prejudice from having such a short time to comment. Nor, as the AFMC notes, did Dorbest request an extension of time during the remand proceedings. AFMC’s Rebuttal Comments on Remand Results 45. Dor-best admits that the objections it raised to Commerce’s choice of subheading 3926.4099 were “not ‘foolproof to Commerce’s counterattack.” Dorbest Comments on Remand Results 43, and does not suggest that, since the remand proceeding, it has come up with other objections which it was unable to exhaust before the agency due to a lack of time. However, Commerce addressed this argument below, and found that there was no reason to exclude those values. Lastly, Dorbest argues that if the court affirms Commerce’s choice of subheading, Commerce should still be required to exclude aberrational data from the United States and Singapore. Id. at 47. Dorbest bases its assertion of distortion on the fact that goods from each of these countries were only imported during one month of the POI, and because the values were higher than imports from other countries. Id. (citing Hebei Metals & Minerals Imp. & Exp. Corp. v. United States, 28 CIT 1185 (2004)). Commerce addressed this issue during the remand proceedings, and made a factual finding that the values were not distortive, because “the average unit value from the United States and Singapore are 3.1 and 4.2 times the average unit value from all other countries, respectively,” and because the volume of imports from the two countries was relatively large, notwithstanding the fact that imports were made during only one month. Remand Results 49 (contrasting the situation with that in Hebei, where values were 8.5 times the average value, and the aberrational imports were in small quantities). On this record, Commerce’s determination cannot be said to have been unreasonable, as Commerce weighed the evidence and chose between imperfect alternatives. For the reasons explained above, the court finds that Commerce’s valuation of the resin applique input is reasonable. C. Mirrors In its Final Results, Commerce used MSFTI data to value mirror inputs, claiming it to be the best available information. Dorbest challenged the choice of data, arguing that the MSFTI data for the relevant subheading, 7009.9100, was distorted by its inclusion of specialty mirrors from Taiwan. The court found that Commerce’s determination that the MSFTI data was the best available information was not supported by substantial evidence. The court remanded the issue to Commerce with instructions to evaluate the inaccuracies of the MSFTI data set, which it had not previously done. Dorbest, 462 F.Supp.2d at 1280 (noting that “[ijnstead of addressing the Respondents’ concerns with the MSFTI data, Commerce chose to attack the quality of the data proffered by Respondents, claiming that the unreliability of the data negated its ability to serve even as a means of evaluating the MSFTI data.”). The court also instructed Commerce to evaluate the Glass Yug data, stating that “[a]t the very least, none of Commerce’s arguments with respect to Glass Yug address why this data should not be viewed as probative towards a view that Commerce’s chosen valuation is too high and/or inaccurate.” Dorbest, 462 F.Supp.2d at 1282. On remand, Commerce determined that specialty mirrors from Taiwan were included in the MSFTI data and likely produced distortion in that data. Remand Results 51-52. Commerce further determined that excluding Taiwanese data from the MSFTI data would not remedy the dataset. Id. at 52-53. This determination was based in part on the fact that a large portion of the MSFTI data consisted of the Taiwanese data. Commerce also noted that automotive parts made up at least a portion of the data from other countries, and that these data were miselassified. Id. at 53. As a result, Commerce felt it could not “presume with any degree of certainty that the remaining imports within this category are not similarly misclassified.” Id. Commerce then determined that this Glass Yug data represented the best available information. Commerce found that the two companies whose prices were reported therein were “large Indian producers that are significant players in the Indian mirror marketplace”, and that their price information was “reasonably representative of the cost of mirrors in India.” Remand Results 56. Commerce further found that the data were contemporaneous, id., specific to the product described by Dorbest, id., likely to have excluded taxes, id. at 57, and corroborated by other sources on the record. Id. In its determination, Commerce rejected Petitioners’ arguments that the Glass Yug data are “not contemporaneous, not country-wide, not representative and that the prices are dis-tortive of the market for mirrors in India.” Id. at 64. No party objects to this determination, and the court finds that it is reasonable. D. Cardboard In its Final Results, Commerce valued Dorbest’s packing cardboard factor of production under subheading 4808.1000, HTS[I]. Issues & Decision Mem., at 170 (Cmt.19) P.R. Doc. 1933. Dorbest challenged the results, based on an argument that subheading 4808.9000, HTS[I] was more appropriate to its cardboard, which Dorbest claimed was not perforated. This led Commerce to issue an Amended Final Determination, using subheading 4808.9000 to value the cardboard input. Amended Final Determination and accompanying Issues & Decision Mem. (Cmt. 5). Before the court, AFMC argued that the original classification of 4808.1000, HTS[I] was more appropriate, alleging that Dor-best’s packing cardboard was corrugated, and that therefore subheading 4808.1000 was specific to the input, and further arguing that the phrase “perforated or not” made the question of perforation irrelevant to valuation. Petitioner’s Br. 22-23. The court explained that “[l]ogic would dictate that the use of the word ‘other’ in the basket subheading indicates that this subheading should only be used if all other subheadings within that heading are exhausted and have been deemed inappropriate.” Dorbest, 462 F.Supp.2d at 1313. The court further explained that under well-established classification principles, “it is only appropriate to use a basket provision when no other subheading applies.” Id. (citing Witex, U.S.A., Inc. v. United States, 28 CIT 1907, 1916-17 & n. 16, 353 F.Supp.2d 1310, 1319 & n. 16 (2004)). In its discussion, the court noted that nonetheless, there was a difference between Customs classifications and the issue here, stating that “Commerce’s goal here is different from Customs’. Whereas in Customs cases determining the proper specific classification is paramount ... in Commerce ■ cases, Commerce is using the HTS[I] merely to approximate the cost of a factor of production.... ” Dorbest, 462 F.Supp.2d at 1308 (citations omitted). After concluding that, in subheading 4808.1000, “whether” likely meant “whether or not,” and that perforation was therefore probably not dispositive to a classification determination, the court found that Commerce’s decision was not supported by substantial evidence. Dorbest, 462 F.Supp.2d at 1313. Dorbest’s cardboard input valuation was therefore remanded to the agency, with the order to “either explain, with reference to the description of the input, why 4808.9000 is the appropriate classification, or to change the classification accordingly.” Dorbest, 462 F.Supp.2d at 1313. Upon remand, Commerce again found that subheading 4808.9000 was appropriate for valuing the import. Remand Results 90. Commerce agreed that “whether or not perforated” did not exclude either perforated or non-perforated cardboard from subheading 4808.1000. Id. However, Commerce still found that the perforation language distinguished the two categories, stating that the language allows for the inclusion of perforated cardboard in subheading 4808.1000, while perforated cardboard would rarely be included in subheading 4808.9000. This second observation, that perforated cardboard would not be included in basket provision 4808.9000, was based on Commerce’s determination that perforated cardboard would more likely be classified under 4808.1000, as that subject heading specifically mentioned perforated cardboard. Id. at 91. Thus, Commerce chose to use the basket classification “in an attempt to eliminate any distortions that may potentially be caused by valuing Dorbest’s non-perforated packing cardboard using an HTS subheading that likely includes perforated paperboard.” Id. In response to comments from the parties during the remand proceedings, including Dorbest’s argument that other entries under 4808.1000 would be distortive of the input cost, Commerce further stated that “when valuing factors of production for Dorbest, it is not classifying Dorbest’s corrugated packing cardboard within either of these subcategories. Rather ... it is evaluating which HTS subcategory better reflects Dorbest’s packing cardboard input.” Remand Results 94 (emphasis added). Although Dorbest cited Infodrive data to support its argument that 4808.9000 was the most appropriate valuation for its input, Remand Results 93, Commerce appears not to have relied on this data. Id. at 94-95. Although Commerce defends its decision by arguing that it chose the subheading that “better reflects” the input, it appears to the court that Commerce’s justifications for the decision are only related to analysis of the classification. This issue is therefore unlike the mirror inputs, discussed above, in which Commerce made a factual finding that certain inclusions in the appropriate subheading caused distortion. See supra at 1334-35. Rather, here Commerce did not make a finding, as would have been proper, as to whether perforated cardboard items classified under subheading 4808.1000 were distortive to its valuation. Commerce has now requested a voluntary remand “in order to permit Commerce to consider whether any cardboard would likely be present in the 9000 basket subheading given that all cardboard would likely be placed in the more specific 1000 subheading.” Def.’s Resp. to Comments on Remand Results 71 (emphasis added). Commerce suggests that remand is “particularly appropriate because, for the first time, in AFMC’s June 29, 2007 comments brief, petitioners argued that the proper title of the 9000 basket subheading is ‘other’ rather than ‘other paper and paperboard corrugated.’ ” Id. Commerce therefore requests a remand “to determine the actual title of the subheading and determine the correct HTS subheading to employ for the valuation of cardboard.” Id. at 72. Dorbest opposes remand, arguing that numerous documents on the record show that subheading 4808.9000 is unquestionably “other paper and paperboard corrugated,” and that AFMC has just now begun to mis-characterize it as “other.” Rebuttal Comments of Pis. Dorbest on Remand Redetermination 6-11 (“Dorbest Rebuttal Comments ”). Dorbest further argues that Commerce applied subheading 4808.9000 to all other respondents in this case, and that it is unfair to Dorbest for Commerce to use a different rate. The court understands, as put forth by Defendants, that “[a]n agency may request a remand to reconsider its previous position without confessing error.” Def.’s Resp. to Comments on Remand Results 71 (quoting SKF USA Inc. v. United States, 254 F.3d 1022,1028 (Fed.Cir.2001)). However, it is difficult to fathom how a failure to ascertain the meaning of the subheadings at issue on three separate occasions could be characterized as anything but error. Nonetheless, the court will grant Commerce’s request for another remand, particularly as it appears that Commerce has yet to follow this court’s instructions to “explain ... why 4808.9000 is the appropriate classification, or to change the classification,” Dorbest, 462 F.Supp.2d at 1313, due to the logical inconsistencies in Commerce’s proffered “explanation.” See, e.g., supra, n. 24. Upon remand, Commerce shall determine under which subheading Dorbest’s input would properly be classified, and further determine whether the data put forth by Dorbest regarding distortion to data in subheading 4808.1000 necessitates alteration of the data used or the selection of a different subheading. E. Metal Spare Parts, Non-Scope Metal Canopies and Certain Raw Materials Expenses Commerce requested a voluntary remand on the issue of its treatment of spare parts, elimination of metal parts and canopies and valuation of incoming raw material expenses after “realizing that certain elements of Dorbest’s margin calculation were made in error.” Remand Results 95. The court granted Commerce’s request. Upon remand, Commerce made certain changes to these factors, altering Dor-best’s resulting margin calculation. Dor-best concurred with Commerce’s resolution of these issues, and the Petitioners did not comment on them, either below or before the court. Therefore, the court will accept without review Commerce’s remand decision. S. Selection of surrogate companies to calculate financial ratios A firm’s “general expenses and profits” not traceable to a specific product are also considered factors of production for purposes of calculating normal value. These expenses and profits include (1) factory overhead (“overhead”), (2) selling, general and administrative expenses (“SG & A”), and (3) profit. 19 U.S.C. § 1677b(c)(l); see also Dorbest, 462 F.Supp.2d at 1300. In order to attribute the appropriate amount of expenses and profits to the merchandise in question, Commerce uses a ratio that describes the relationship between expenses and profits and a firm’s overall manufacturing cost. These calculations are performed with data from financial statements of one or more surrogate companies. In its Final Results, Commerce used the financial statements of nine companies to calculate financial ratios. Dorbest, 462 F.Supp.2d at 1302. The parties challenged various of Commerce’s determinations as to which companies to include and which to exclude. The court remanded to Commerce its determinations concerning the inclusion of data from Jayaraja and Evergreen upon AFMC’s challenge, and also remanded Commerce’s determination to include data from Swaran, Nizamuddin, Fusion Design, and DnD upon Dorbest’s challenge to those data. Jayaraja and Evergreen In remanding Commerce’s determination to include data from Jayaraja, the court ordered Commerce to address AFMC’s objections on remand, noting that “[djespite the fact that Commerce outlined [Results] does not directly or indirectly refute these arguments.... ” Dor-best, 462 F.Supp.2d at 1304 (citations omitted). Specifically, the court ordered Commerce to address AFMC’s contention that Commerce, in using Jayaraja’s financial statement, had departed from its “well-established practice to reject” financial statements that contain “no notes” and “no auditor’s statement.” See Dor-best, 462 F.Supp.2d at 1303. The court noted that Commerce had also not addressed AFMC’s alternative argument that Jayaraja’s statement was “a significant outlier and a cause of distortion” because “[Jayaraja] report[s] zero depreciation in the profit and loss statement.” See id. at 1303-04. The court went on to state that “[a]lthough Commerce’s argument [that Jayaraja’s financial statement was approved by an auditor and was sufficiently detailed such that no auditor’s statement or notes were necessary] may be compelling, the agency must adopt this position on the agency record if the court is to affirm it here.” Id. In remanding Commerce’s determination to include financial statements from Evergreen, the court found that Commerce had not justified its decision “to include statements which it admits are of questionable reliability and thereby unlikely to constitute the best available information.” Id. at 1304. Specifically, Commerce had not addressed AFMC’s concerns that Evergreen’s status as a “significant producer of leather garments as well [as] a manufacturer of furniture,” id. (quoting AFMC’s Br. 24), would lead to distortions in Commerce’s calculation of Evergreen’s financial ratio. In the Final Results, Commerce had excluded “identifiable manufacturing expenses related to the production of leather goods from the MLE denominator,” because of its finding that “Evergreen outsources almost the entire production of its leather goods.” Dorbest, 462 F.Supp.2d at 1304 (quoting Memorandum from Jon Freed, Case Analyst, to File Re: Final Determination Financial Ratio Memorandum: Wooden Bedroom Furniture from the People’s Republic of China, P.R. Doc. 1931 at 2). However, Commerce had “conceded that Evergreen’s leather production was more problematic in the calculation of the other ratios,” because its “SG & A and profit relate to both leather and furniture goods.” Id. (citations to record omitted). Nonetheless, “Commerce included leather related expenses in the denominator of the SG & A and profit ratios.” Id. The court found that Commerce had not explained “why the inclusion of Evergreen’s financial statement, in spite of the complication identified by Commerce, adds to the accuracy of its calculation of the surrogate ratios.” Id. The court went on to explain that “[pjarticularly problematic is the fact that other financial statements, without such problems, exist,” Id. On remand, Commerce made a determination to exclude financial statements from Evergreen and Jayaraja. Remand Results 72. In determining not to use Jaya-raja data, Commerce found that “due to the lack of the auditor’s report, schedules, the auditor’s opinions and notes to the financial statement,” there was not sufficient data to “allocate Jayaraja’s expenses among direct expenses, overhead, and SG & A with any level of certainty.” Id. at 73. Commerce explained that “the Jayaraja statements do not report an amount for depreciation,” and that therefore Commerce had “no means to conclusively determine if Jayaraja had any depreciation expenses during the period.” Id. at 73-74. Commerce further noted that this decision “is consistent with [Commerce’s] practice to normally disregard surrogate financial statements if they are incomplete and lack certain key reports (e.g., schedules, notes).” Id. at 74 (citing Silicomanganese from Kazakhstan, 67 Fed.Reg. 15,535 (Dep’t Commerce Apr. 2, 2002)(determina-tion of sales at less than fair value)). In determining not to use Evergreen data, Commerce took into account the issue of disaggregation of profits and SG & A expenses, and determined that “[u]pon further consideration of this issue, we find that sufficient evidence does not exist for a finding that the profit margin for producing and selling leather garments is the same as that for producing and selling furniture.” Remand Results 72. Dorbest argues that although Commerce stated concern regarding distortion resulting from the use of Evergreen’s data, it did not identify the nature of the distortions. Dorbest Comments on Remand Results 29. Dorbest points to Fuyao Glass Industries for the proposition that “a company’s SG & A supports its entire operations, including both self-production and outsourced merchandise.” Dorbest Comments on Remand Results 31 (citing Fuyao Glass Indus. Group Co. v. United States, 2005 Ct. Int’l Trade LEXIS 29, 56-57, 2005 WL 280437 (CIT Jan. 25, 2005)). However, the issue in that case was that expenses related to the purchase of traded goods were reflected in the numerator, but not the denominator, of the ratio. Fuyao Glass Indus. Group Co., 2005 Ct. Int’l Trade LEXIS at 56-57, 2005 WL 280437. The court noted that there were numerous ways to correct this problem, and once Commerce had included those expenses in the denominator, the court found Commerce’s methodology to be supported by substantial evidence. Dorbest also cites Tapered Roller Bearings, arguing that if there is distortion, Commerce should be able to correct for it, rather than eliminating the data. Dorbest Comments on Remand Results 31 (citing Tapered Roller Bearings and Parts Thereof, Finished and Unfinished, from the People’s Republic of China, 71 Fed.Reg. 75,936 (Dep’t Commerce Dec. 19, 2006) (final results of 2004-2005 administrative review and partial rescission of review)). Dorbest points out that in the Final Results in this case, Commerce also included the leather manufacturing data in the denominator of the SG & A and profit ratios. However, neither in Fuyao nor in Tapered Roller Bearings was the issue raised, as it has been here, that the products being included would have a different profit margin, based on their differences from the subject merchandise. In Fuyao, the court found that Commerce’s calculation had resulted in an “apples to apples” comparison. Similarly, to achieve such a results here, Commerce reasonably chose to exclude Evergreen so as not to make a furniture-to-leather-goods comparison. Commerce could not find evidence to show that the profit margins of leather goods and wooden furniture would be similar, and as such determined that the inclusion of the data— even if included in both the numerator and the denominator — could likely be distor-tive. Where other data without these drawbacks existed, Commerce’s determination to exclude data from Evergreen is supported by substantial evidence. Regarding Jayaraja, Dorbest notes that Commerce’s assertions on remand “are the polar opposite of those made in [sic] previously to the Court in justifying use of Jayaraja [data].” Dorbest Comments on Remand Results 34. Dorbest further argued that IFP similarly did not have an auditor’s report, speculating that Commerce had simply chosen to dispose of the smaller of the two numbers. Id. at 36. However, the court finds that Commerce followed its practice by choosing not to use incomplete information that lacked an auditor’s report, and its determination was supported by substantial evidence where Commerce pointed to specific data that it was missing, such as depreciation data, and explained the significance of that data. Because the inclusion of IFP was not challenged here, the court will not now address this issue. The court also remanded Commerce’s determination to use data from financial statements from Swaran, Nizamuddin, Fusion Design and DnD in its calculation of financial ratios. The court noted that Respondents’ concerns that the smaller size of these companies would lead to inclusion of unrepresentative data appeared to be relevant, given that these four companies “had an average SG & A ratio of 32.22% compared with the 14.37% average SG & A rate of the other allegedly similar five companies included in the data set.” Dor-best, 462 F.Supp.2d at 1306. Though Commerce justified its conclusion by pointing to the SG & A ratios of Jayaraja and Akriti to show that small production and sales volumes did not always correlate with high SG & A expenses, the court found that this observation did “not demonstrate the opposite conclusion: that the size of Nizamuddin, Fusion Design, Swar-an, and DnD are irrelevant.” Id. The court further explained that “where, as here, there are (perhaps) other surrogate companies which better approximate the manufacturing experience of the importers’ businesses, the comparability test may require limiting the data set to those surrogate companies which reasonably approximate the importers’ manufacturing experience.... ” Id. at 1307. On remand, the court ordered Commerce to explain its inclusion of these four companies’ statements, noting that “Commerce is free on remand to find that these financial statements are as reflective of the Respondents’ manufacturing experiences as the other financial statements upon which it relies,” but adding that “Commerce must uniformly apply whatever criterion it ultimately adopts.” Id. On remand, Commerce again chose to use financial statements from Swaran, Ni-zamuddin, Fusion Design and DnD. Remand Results at 68-69. Commerce found that “none of the seven Indian companies used in [Commerce’s] calculation of the surrogate financial ratios approximates the size of Dorbest,” and that the revenues of all but one of the companies were under $1 million, with revenues of all the companies significantly less than that of Dorbest. Remand Results 69-70. As a result of all the surrogate companies being smaller than Dorbest, Commerce did not find “that their relative size in relation to Dorbest is a basis for the inclusion or rejection of a financial statement.” Id. at 70. Commerce further found “no evidence that relative size is a primary driver in the differences in financial ratios of the Indian surrogate companies on the record.” Id. Commerce based this conclusion on the fact that the largest of the surrogate companies, Indian Furniture Products (“IFP”) “represents the midpoint of the calculated SG & A ratios,” and that “if relative size drove the SG & A ratio,” the largest company “would be expected to have the lowest SG & A ratio.... ” Id. (citations omitted). Commerce also found that Swaran’s high SG & A ratio skewed the average ratio for the four companies, and that exclusion of Swaran’s data would result in comparable ratios to those of the other companies. Id. at 71. Commerce went on to defend the inclusion of Swaran’s data based on “[Commerce’s] preference to use multiple financial statements when they are not distortive or otherwise unreliable, in order to eliminate potential distortions that may arise from using those of a single producer.” Id. (citing Issues & Decision Mem. (Cmt.3)). Dorbest contends that Commerce has “cherry-picked” its data, and that a “clear, dividing line exists between the small versus larger companies, with the smaller companies having a markedly higher SG & A ratio.” Dorbest Comments on Remand Results 19. Dorbest further suggests that if none of the surrogate companies is comparable to Dorbest, the court should revisit its decision to affirm Commerce’s determination not to use Indonesian data. Lastly, Dorbest suggests that at the least, Commerce should remove Swaran data from its calculation, as it has admitted that this data is distortive. In support of its argument that Commerce’s determination was unreasonable, Dorbest presents its analysis showing that there is a relationship between company size (as measured by cost of manufacture ) and SG & A ratio, which is more clearly visible when the data from Evergreen and Jayaraja are present, as they are in the second of the two tables, below, taken from Dorbest’s briefing. See Dorbest Comments on Remand Results, 23-25. Table 3: MLE and SG & A% for 7 Companies Used in Remand Company SGA% MLE Fusion Design 34.39% 1,491,058 Nizamuddin 31.51% 1,970,177 Swaran 47.30% 2,350,013 DND_15.66%_4,031,745 Akriti 13.53% 6,426,611 Raghbir_10,44% 28,387,927 IFP 24.38% 345,994,333 Table 4: MLE and SG & A% for 9 Companies Analyzed in Remand Company SGA% MLE Fusion Design 34.39% 1,491,058.00 Nizamuddin 31.51% 1,970,177.00 Swaran 47.30% 2,350,012.62 DND 15.66% 4,031,744.89 Akriti_13.53% 6,426,611.39 Jayaraja 16.61% 8,080,148.00 Raghbir 10.44% 28,387,927.27 IFP_24.38% 345,994,333.00 Evergreen 6.90% 424,690,459.12 It appears to the court that, apart from data from Swaran and IFP (both of which have large values for SG & A compared to the other data points), there is indeed a general, inverse relationship between the size of a company and its SG & A expenses. Furthermore, it appears that Swaran and IFP do not fit this trend. Thus, Commerce’s findings, which include comparisons of the two smallest companies’ SG & A ratios with IFP’s, and which further involved excluding data from Swar-an in order to show comparability, while still using the Swaran data in its determinations because of its “preference to use multiple financial statements,” do not adequately address the court’s analysis and conclusions in Dorbest. Indeed, the government states that “Dorbest’s graph illustrates ... that there is no relationship linear or otherwise, between its measure of a company’s size (i.e., MLE) and its SG & A ratio, because two of the seven data points — IFP and Swaran — bear no relationship to the line.” Def.’s Resp. to Comments on Remand Results 49. However, as Swaran’s data has been questioned for its possible distortive effect, it is unclear how its lack of a relationship to the line can be used in defense of its inclusion. It appears to the court, therefore, that Commerce’s determination — that there is no relation between company size and SG & A-is unsupported. Nonetheless, Commerce’s determination to include SG & A ratios which it has determined are “comparable” to those of companies of other sizes may be within the agency’s discretion, if based on proper findings regarding the effect of including much smaller companies in its data set. Commerce’s remand determination, however, does not contain such findings. Unlike the go