Opinion for the Court filed by Circuit Judge WALD.
WALD, Circuit Judge:
Petitioners seek review of decisions issued by the Federal Energy Regulatory Commission (“FERC” or “Commission”) on remand from this court’s decision in Process Gas Consumers Group v. FERC, 866 F.2d 470 (D.C.Cir.1989) (“PGC I”). At issue is FERC’s approval of a natural gas rate surcharge to support “end-use” research by the Gas Research Institute [206]*206(“GRI”). Upon a thorough review of the extensive record, we find aspects of the Commission’s order inconsistent with FERC regulations and the standards set forth by this court in PGC I. Accordingly, we grant the petition for review, affirm in part and vacate in part the order on remand.
I. Background
GRI is a collaborative research organization of the natural gas industry which performs extensive research, development and demonstration (“RD & D”) activities. These projects are currently “financ[ed] more or less exclusively via direct FERCapproved surcharges on jurisdictional throughput, i.e., gas supplies delivered through regulated pipelines to consumers.” PGC I, 866 F.2d at 471. In 1988, the Process Gas Consumers Group challenged FERC’s approval of a surcharge of 1.51 cents per Mcf1 to support GRI’s 1988 budget.2 In particular, the Group challenged FERC’s approval of surcharges to underwrite “end-use” RD & D aimed at developing more efficient gas appliances and applications and new consumer uses for gas.
In PGC I, this court granted the Group’s petition and remanded to the agency for further proceedings. In doing so, we first distinguished GRI’s research in past years which had as its “broad goal the conservation of dwindling gas supplies.” Public Utilities Commission of Colorado v. FERC, 660 F.2d 821, 825 (D.C.Cir.1981), cert. denied, 456 U.S. 944, 102 S.Ct. 2009, 72 L.Ed.2d 466 (1982). In contrast, we observed, end-use research was “aimed [in part] at generating new demand for natural gas” and thus was likely to increase the demand for and the price of gas. PGC I, 866 F.2d at 474 (emphasis in original). Accordingly, such research was less likely to have “a reasonable chance of benefitting the rate-payer in a reasonable period of time,” as required by FERC regulations. 18 C.F.R. § 154.38(d)(5)(iii)(ri). We found that FERC’s decision that the end-use research projects proposed by GRI met the required standard was inadequately supported.
Our directions to FERC on remand were specific; as they are central to this proceeding, we quote them at length:
[T]he Commission need not undertake scientific “peer review” or otherwise attempt to determine with precision whether the efficiency gains from an end-use application will outweigh the costs to ratepayers of the research. It is enough for the Commission rationally to conclude that the research contemplated is by its nature likely to benefit ratepayers if successful.
When determining whether end-item research ... “has a reasonable chance of benefitting the ratepayer in a reasonable period of time,” FERC, we think, is required to find that the research, if successful, will on balance work to the benefit of existing classes of ratepayers— those customers paying for the research in the first place. On the benefit side of the equation, FERC may include all economic gains that might inure to existing classes of ratepayers through the employment of gas-saving devices, thus excluding those efficiency gains that flow to consumers who switch to gas by reason of ratepayer-financed research. On the cost side, FERC not only must account for the initial expense of the research but also must make a rational judgment about the probable economic effect of the research, if successful. FERC must at least consider whether a [207]*207project is intended to create a new market for gas among present consumers of an alternative fuel, and whether that new demand is likely to work to the disadvantage of existing ratepayers by pushing up gas prices.
PGC I, 866 F.2d at 474, 476-77 (emphasis in original).
On remand, FERC consolidated the remanded proceedings on GRI’s 1988 budget with its review of GRFs 1989 budget. FERC requested and received new submissions from GRI justifying its end-use research proposals in light of PGC I. FERC also invited and accepted responses from other participants, but denied a request by certain participants for discovery of background documents supporting GRFs new submissions.
In its order on remand, Gas Research Institute, 48 F.E.R.C. 1161,048 (1989), FERC adopted GRFs analysis and approved both GRFs 1988 and 1989 budgets and the attendant surcharges. The Commission found, inter alia, that:
• end-use RD & D would increase gas demand by 1.1 Tcf, but that increases in supply resulting from GRFs supply-related research (not at issue in this remand) would offset this increase;
• such increased demand would, taken alone, increase prices by 15 cents per MMBtu;
• increased demand would also reduce per-unit transmission and distribution costs by 8.3 cents per MMBtu (through the more complete use of existing capacity), leaving a net cost increase of 6.7 cents per MMBtu;
• based in part on these findings and on other evidence in the record, the proposed end-use research satisfied the criterion set forth in PGC I, i.e., that the expected benefits of the research exceed the expected costs.
The petitioners, the Process Gas Consumers Group, the American Iron and Steel Institute, and the Georgia Industrial Group (collectively “the Industrial Groups” or “the Groups”), represent industrial consumers of natural gas; they now seek review of FERC’s decision on remand.
II. Analysis
A. Procedural Challenges
The Industrial Groups first raise several challenges to the procedures FERC used on remand and contend that FERC unfairly structured its procedures in favor of GRI. We find these claims unsubstantiated and affirm the Commission’s procedures.
The Industrial Groups first take issue with FERC’s denial of the Groups’ requests for discovery from GRI of certain information that the Groups sought in preparing their response to GRFs submissions on remand. In assessing this claim, we begin from the premise that FERC “has broad discretion ... to decide what procedures to use in fulfilling its statutory responsibilities.” Kansas Power & Light Co. v. FERC, 851 F.2d 1479, 1484 (D.C.Cir. 1988). In this case, consistent with its established practice for GRI-budget proceedings, FERC did not hold a hearing, nor does the record suggest that the participants sought a hearing. This fact is significant because the Commission’s discovery regulations “appl[y] to discovery in proceedings set for hearing ... and to such other proceedings as the Commission may order.” 18 C.F.R. § 385.401(a) (emphasis supplied). Therefore, the use of discovery outside of the hearing context lies within the discretion of the Commission. In this case, FERC, noting that the matter was on remand and that the remand standard called for a “common-sense assessment,” chose not to grant the Groups’ discovery requests. On these facts, we cannot find that FERC’s denial constitutes an abuse of discretion.
The Industrial Groups also challenge FERC’s failure to undertake an independent staff evaluation of GRFs analysis and assumptions. FERC found such an analysis was unnecessary because “the Commission staff had already prepared two reports in 1987 and 1988 addressing end-use R & D.” Gas Research Institute, 49 F.E.R.C. 1161,004 (1989). Moreover, because of its experience reviewing other proposals that used similar methodologies, the [208]*208Commission was familiar with the project assessment methodology used by GRI. Again, on these facts, we cannot conclude that FERC’s failure to undertake on remand an independent staff analysis constituted an abuse of discretion.
Finally, the Industrial Groups contend that FERC incorrectly shifted the burden of proof to the Groups. This claim is unfounded. Although FERC ultimately adopted most of GRI’s conclusions, its decision reflects that FERC weighed and rejected each of the participants’ major challenges to GRI’s findings. 48 F.E.R.C. II 61,048 at 61,225-36. Accordingly, we reject the Industrial Groups’ procedural contentions and turn to their substantive challenges to FERC’s order on remand.3
B. Substantive Challenges
As noted above, in PGC I this court directed FERC to consider on remand whether GRI’s proposed research “is by its nature likely to benefit [existing class of] ratepayers.” PGC I, 866 F.2d at 474 (emphasis in original). To do so, FERC must determine whether the research “will on balance work to the benefit of existing classes of ratepayers,” id. at 476, and make a rough assessment of the various costs and benefits to existing ratepayers of the proposed research. We also noted that in certain cases — such as in weighing “a project that is aimed at creating captive gas demand” — the “Commission would face a heavy burden in justifying ratepayer financing of research.” Id. at 477.
The petitioners contend that, on remand, FERC failed to comply with this court’s directions in several respects; FERC, in turn, defends its actions against this accusation. We first examine the nature of the proposed research, and then consider the petitioners’ contentions.
1. GRI’s Research Taxonomy
In its order on remand, FERC adopted a taxonomy of research (first set forth by GRI) that distinguishes among three classes of research projects.4
Class-1 technologies involve improvements in equipment currently used by ratepayers, such as residential and commercial gas appliances and advanced industrial heat transfer systems.
Class-2 technologies involve new technologies which existing customers may add to their existing equipment. [One] example[ ] of this class of technology [is] residential cooling as an added option for residential customers already using gas for space heating, water heating, clothes drying and/or cooking____
Class-3 technologies would create new uses for natural gas. The users of these technologies would primarily be customer classes which have not traditionally relied on natural gas, such as vehicular transportation customers.
48 F.E.R.C. at 61,222-23. The Commission approved all three of these classes of research. Below, we first address two arguments raised by the parties that concern all three classes. Thereafter, we consider each class in turn to determine “that the [209]*209Commission’s decisionmaking is reasoned, principled, and based upon the record.” Columbia Gas Transmission Corp. v. FERC, 628 F.2d 578, 593 (D.C.Cir.1979).
2. General Arguments
a. Effects on Demand and Prices
Reiterating an argument first raised in PGC I, the Industrial Groups contend that because end-use research will increase demand for gas (and therefore drive up gas prices) such research is categorically inconsistent with FERC’s authority under the Natural Gas Act (“NGA”), 15 U.S.C. § 717 et seq., and FERC regulations. In support of their argument, the Groups cite FERC’s findings that the contested RD & D will increase gas demand by 1.1 Tcf. 48 F.E.R.C. at 61,224. This argument, however, is misguided. Both the terms of the Commission’s regulations and our decision in PGC I (interpreting the NGA and FERC regulations) expressly indicate that the relevant analytical touchstone for evaluating end-use research is not gas demand or gas prices, but rather benefits to existing classes of ratepayers. See 18 C.F.R. § 154.38(d)(5)(iii)(£i); PGC I, 866 F.2d at 475-76.
The claims of the petitioners notwithstanding, research that increases demand may, in fact, benefit existing classes of ratepayers. For example, research that enhances the efficiency of existing appliances benefits current ratepayers even if such research also increases demand (e.g., by encouraging the purchase of the improved appliances), so long as the efficiency gains offset the costs (to existing ratepayers) of increased demand. As this court concluded in PGC I, the relevant mode of evaluation is not a categorical assessment that bars all research that increases demand, but rather a more discriminating “net-benefits” analysis. Accordingly, we reject the petitioners’ suggestion that the Commission’s order should be vacated in toto because end-use research increases gas demand and prices.
b. Aggregation of Supply Research and End-Use Research
FERC also offers a broad argument in support of its order. Noting that, in addition to its end-use research (the focus of this case), GRI also conducts research designed to enhance the supply of natural gas, FERC contends that the two research programs should be aggregated. More precisely, the Commission maintains that any increase in demand resulting from GRI’s end-use research is more than offset by GRI’s supply research. In the terms of PGC I, FERC suggests that even if end-use research taken alone does not “on balance work to the benefit of existing classes of ratepayers,” any negative demand effect of end-use research will be more than offset by GRI’s other good works.
Such aggregation is both inconsistent with this court’s directions on remand and analytically unsound. First, the record suggests that both GRI and FERC treat GRI’s supply research and its end-use research as two separate programs. Building on this practice, the petitioners in PGC I and the PGC I court addressed only end-use research. Accordingly, our remand to the Commission was limited to such research.
Moreover, as a logical matter, the Commission’s argument compares apples on the one hand with apples and oranges on the other. A cost-benefit analysis should compare the costs and benefits of the same proposal; FERC suggests that the costs of GRI’s end-use research should be weighed against the benefits of both GRI’s end-use and its supply research. Thus, even if it were appropriate for FERC to aggregate the benefits of GRI’s end-use and supply research programs, FERC would also be required to aggregate the costs of those programs. For these reasons, we reject FERC’s aggregation of GRI’s supply and end-use research programs as unreasonable and erroneous.
3. Specific Class-Related Arguments
a. Class-1 Research
In approving Class-1 research, FERC embraced GRI’s assessment that the benefits from this class of RD & D would accrue exclusively to current classes of [210]*210ratepayers and concluded that therefore the research would by its nature benefit those ratepayers. FERC also accepted GRI’s contention that the research would generate very little new demand (0.09 Tcf) and would yield very substantial consumer savings ($2,437 billion). 48 F.E.R.C. at 61,-236. On these grounds, the Commission concluded that Class-1 research met the net-benefits test set forth in PGC I.
The Commission’s conclusions are well supported in the record and reflect its consideration of the relevant issues, as set forth in PGC I. The Industrial Groups’ primary challenge to this class of research is that such research will increase both the demand for gas and gas prices and therefore is not “by its nature” likely to benefit current ratepayers. As demonstrated above, however, this argument fails. Accordingly, on the record before us, we find that the Commission’s approval of Class-1 research was supported by substantial evidence, consistent with PGC I, and not arbitrary and capricious.5
b. Class-2 Research
Class-2 research differs from Class-1 research in that Class-2 research “would, to a greater extent than class-1 technologies, also benefit users of the new technology who would otherwise not have used natural gas at all." 48 F.E.R.C. at 61,238 (emphasis supplied). In assessing Class-2 research, FERC first determined that this difference does not “render the R & D projects ‘by [their] nature’ unbeneficial [sic] to existing classes of ratepayers.” 48 F.E.R.C. at 61,226. In this regard, we agree with the Commission.
By definition, Class-2 technologies are targeted at current gas consumers, creating “add-on” uses of gas. Accordingly, the benefits of Class-2 research, such as gas-powered residential cooling systems, will largely accrue to current ratepayers, for “[i]t is unlikely that a consumer without existing gas service would choose gas cooling.” Response of Gas Research Institute (March 23, 1989) (“GRI Response”) Appendix A, Class-2 Projects. Therefore, although Class-2 technologies do create some savings for “fuel-switchers,” the record demonstrates that such research would “predominantly benefit existing classes of gas customers.” GRI Response at 4 (emphasis in original).' On this basis, FERC properly rejected the Groups’ contention that Class-2 research was by its nature inconsistent with the interests of current ratepayers.
The Commission then determined that Class-2 research would on balance work to the benefit of existing classes of ratepayers; in doing so, it relied on GRI’s two measures of the benefits of the proposed end-use research. First, GRI calculated the net consumer benefits to be generated by each project by (i) adding direct consumer savings and indirect consumer savings due to reduced transmission and distribution (“T & D”) costs,6 and (ii) subtracting from that sum increased costs arising from increased gas demand. GRI Response, Appendix A at 6. Second, GRI calculated the benefit-cost ratio for each project, “using the following equation:
[211]*211Consumer Savings (NPV [Net Present Value]) divided by the sum of the following factors: [R & D Costs plus Potential Demand-Induced Acquisition Cost (NPV) minus Transportation and Distribution Cost Savings (NPV)].”
GRI Response, Appendix A at 5. Relying on these measures, the Commission found that Class-2 research (with one exception discussed infra) met the net-benefits test, citing GRI’s estimates of project-by-project benefit-cost ratios of between 2.6-to-l to 11.1-to-l. 48 F.E.R.C. at 61,238-40.
Although the Commission’s determination is, in general, reasoned and reasonable, we note one apparent flaw in GRI’s measures.7 Both the “net consumer benefits” and the “benefit-cost ratio” measures appear to include both benefits to current classes of ratepayers and savings to new ratepayers (fuel-switchers). GRI Response, Appendix A at 5-6. Such inclusion is inconsistent with our direction in PGC I that the net-benefits analysis “exclud[e] those efficiency gains that flow to consumers who switch to gas by reason of ratepayer-financed research.” PGC I, 866 F.2d at 476. However, because the benefits that accrue to “fuel-switchers” comprise only a minority of the net consumer savings, this apparent error does not undermine the Commission’s ultimate conclusion.8
Although the Commission’s analysis of Class-2 research is generally sound, its treatment of one contested project — the “industrial refrigeration project” — merits special attention.9 This project is noteworthy because GRI estimated that the industrial refrigeration project would produce negative net consumers savings — in other words, the project would cost the ratepayers more than it saved them. 48 F.E.R.C. at 61,239-40. FERC, however, found that these net costs were offset by several “unquantifiable” benefits, including reduced transmission and distribution costs and environmental benefits such as reduced pollution. These countervailing benefits, however, are specious. First, transmission and distribution savings have already been included in GRI’s calculation of net consumer savings. Thus, FERC’s invocation of these savings to offset net consumer costs amounts to double counting. More significantly, the Commission’s consideration of “common goods” such as cleaner air, is inappropriate. In PGC I, this court was quite clear in interpreting the NGA and FERC regulations: the relevant considerations in determining the net benefits of proposed research are benefits to ratepayers qua ratepayers. Although the incrementally cleaner air resulting from the proposed research might, in some marginal way, benefit ratepayers as citizens, it was not contended that ratepayers as ratepayers — i.e., as consumers of natural gas— would secure special benefits from cleaner air and that that benefit somehow compensates for the fact that industrial refrigeration research will generate more economic costs than benefits for them.
Without such makeweight justifications the proposed industrial refrigeration research cannot be said to “on balance work to the benefit of existing classes of rate[212]*212payers” for the simple reason that the costs of that project to existing ratepayers exceed the benefits of the project for those groups. Accordingly, we find the Commission’s approval of the industrial refrigeration project arbitrary and capricious and inconsistent with our directions in PGC I.
c. Class-3 Research
Class-3 research consists of two projects: an effort to develop gas technologies that can be combined with existing uses of other fuels to reduce environmental pollution (the “emission-controls project” 10) and research into the development and refinement of natural-gas-powered vehicles. The purpose of Class-3 research is to develop a captive market — to nurture “new consumer and industrial uses of gas that are highly dependent upon the gaseous form of the product (‘form value’), i.e., uses that do not easily accommodate a shift to alternative fuels.” PGC I, 866 F.2d at 474; see also 48 F.E.R.C. at 61,240-41 n. 159. The benefits created by Class-3 research accrue almost exclusively to fuel-switchers, not to existing classes of ratepayers. Indeed, with regard to the emission-controls project, GRI estimates that only eight percent of the net consumer savings will accrue to existing classes of ratepayers; with regard to the natural-gas vehicle project, GRI does not anticipate that any meaningful savings will accrue to current ratepayers. 48 F.E.R.C. at 61,241-46. In fact, “GRI concedes that ... class-3 technology ... is not, by its nature, likely to benefit existing classes of ratepayers, because class-3 technology, by its very definition provides less than 50 percent of its consumer savings to existing ratepayer classes.” 48 F.E.R.C. at 61,240 (emphasis in original).
As noted in PGC I, the Commission bears “a heavy burden in justifying ratepayer financing of research of this sort.” PGC I, 866 F.2d at 477. Accordingly, the Commission “g[a]ve its closest scrutiny to” Class-3 research. 48 F.E.R.C. at 61,240. Finding that other factors outweighed the maldistribution of the benefits of Class-3 research, FERC approved both Class-3 projects. Upon review, however, we find FERC’s analysis in both cases to be flawed; below we consider each project in turn.
(i) The Emission-Controls Project
In approving the emission-controls project, the Commission relied heavily on GRI’s estimate that “existing classes of ratepayers will enjoy a net benefit from this project.” 11 48 F.E.R.C. at 61,242. At first glance, GRI’s conclusion in this regard appears counterintuitive — especially when one considers that only eight percent of the benefits of this project would accrue to existing classes of ratepayers. See id. However, GRI anticipates that the research will generate very large benefits ($5.66 billion per year) and therefore concludes that the benefits to current ratepayers (about $453 million, or 8 percent of $5.66 billion) will exceed the costs of the research (including the cost of increased demand, net of T & D savings). GRI Response, Appendix A, Class-3 Projects.
GRI’s estimates (relied upon by FERC) of the costs of the proposed research are, however, problematic. First, GRI forecasts that this research will create very substantial new demand (almost three times the demand increase generated by all Class-1 projects combined). However, GRI does not calculate the cost of this increased demand, because it contends that use of the new research (and therefore the resulting new demand) is “highly price-sensitive.” Accordingly, GRI discounts the anticipated new demand costs by more than eighty percent. GRI then concludes that the anticipated benefits of this project outweigh the anticipated new-demand costs (so adjusted).
[213]*213Such an analysis is flawed. GRI improperly compares undiscounted benefits with discounted costs. If the use of the new application is “highly price-sensitive,” then both the anticipated costs and the anticipated benefits must be appropriately discounted.12 Instead, GRI treats the benefits as a certainty, but the costs as a mere possibility. Such an analysis is erroneous and the Commission’s reliance on this analysis is unreasonable.13
(ii) The Natural-Gas Vehicle Project
FERC found that, unlike the emission-controls project, the natural-gas vehicle project did not generate any quantifiable benefits for current classes of ratepayers. Nonetheless, the Commission approved that research because it found that the project will “provid[e] unquantifiable benefits such as reliable energy supplies, a healthier gas industry, enhanced competition with other energy sources ..., and (most important as to this project) a cleaner environment.” 48 F.E.R.C. at 61,248. As we have already indicated, however, such common goods do not accrue to existing ratepayers qua ratepayers and thus are not properly included in a net-benefits determination.14 Accordingly, FERC’s approval of the natural-gas vehicle project was unreasonable and inconsistent with this court’s mandate in PGC 7.15
Conclusion
In sum, we find that in its evaluation of GRI’s Class-1 and Class-2 research (with the exception of the industrial refrigeration project) FERC properly and reasonably applied the standards set forth in PGC I and that its factual findings were supported by substantial evidence in the record as a whole. We also find that FERC’s evaluation of the Class-2 industrial refrigeration project and both Class-3 projects was inconsistent with the standards set forth in [214]*214our original opinion. We therefore grant the petition for review, vacate those portions of the Commission’s order approving Class-2 industrial refrigeration research and all Class-3 research, and remand to the Commission for further proceedings consistent with this opinion.
It is so ordered.