Chapter3C-D.html

[3237] Introduction. Cost Segment 10 covers the salaries, benefits, and related costs of rural carriers. Rural carriers primarily provide delivery, collection, and retail services to customers on rural routes.

[3238] Rural routes are divided into three categories based on the way the carrier is paid. The majority of rural routes are evaluated routes for which a carrier's salary is based on time standards. These time standards are developed annually from the National Mail Count (NMC), based on route factors such as route length, number of boxes served and quantity of mail by shape. In FY 1998, 91 percent of volume variable and 83 percent of total accrued Segment 10 costs were generated by evaluated routes. Mileage routes are low-density routes for which compensation is based on route length. Auxiliary routes require less then 35 hours per week and compensation is based on the number of hours worked.

[3239] Since R90-1 the Postal Service has calculated Rural Carrier costs using data from both the NMC and the Rural Carrier Cost System (RCCS). The NMC provides rural carrier costs by activity but does not break down volumes by subclass. The RCCS is used to distribute these costs to the subclasses of mail. Before using the data to distribute costs to subclasses the Postal Service applies a mail shape adjustment to the RCCS data. The mail shape adjustment is necessitated by differences in how the NMC and RCCS define flats. The NMC defines rural letters as 5" in height or less. However, by RCCS standards a letter can be up to 61/8" in height. This causes the RCCS system to code some NMC flats as letters. The mail shape adjustment is applied to the RCCS volumes to ensure that the RCCS flats percentage is consistent with the NMC flats percentage.

[3240] In Docket No. R97-1 the Commission accepted the mail shape adjustment proposed by the Postal Service. In that case the Postal Service determined the number of letters to reclassify as flats by comparing the RCCS flats volumes from the same four-week period during which the NMC is conducted to NMC flats volumes. The Commission also accepted the Postal Service's use of systemwide rather than rural-specific DPS and Sector Segment percentages to create a separate DPS and Sector Segment distribution key. However, the Commission noted that the information systems used to develop rural carrier costs could be modified to gather more information and therefore reduce the number of sources from which data must be combined. PRC Op. R97-1, para. 3358.

[3241] Since R97-1 the Postal Service has modified the RCCS to record DPS and Sector segment volumes separately from other volumes so that distributed costs would be more uniformly related to the evaluated costs. However, in this docket the Postal Service again uses RCCS data from the four-week period during which the NMC is performed to develop the mail shape adjustment.

[3242] Intervenor Opposition. Periodicals Mailers witness Glick proposes the use of annual RCCS data to determine the RCCS flats percentage. Witness Glick argues that annual RCCS data are more reliable than the RCCS data for the four-week period during the NMC. The "RCCS was `designed to produce precise annual estimates, with a sample size of over 6,000 tests', not precise estimates for any four week time period." Tr. 24/11225. The use of only four weeks' data reduces the number of RCCS tests from 6,000 to 333. Tr. 46C/20841. Moreover, use of annual RCCS data results in a lower coefficient of variation for the RCCS flats percentage. Witness Glick concludes that "because the NMC is performed during September-a month that USPS views as annually representative there is no drawback to using annual RCCS data to perform the adjustment." Tr. 24/11225.

[3243] In rebuttal testimony the Postal Service accepts witness Glick's recommendation Postal Service witness Kay confirms that "[w]itness Glick made a compelling argument to use a full year's Rural Carrier Costing System (RCCS) volume in the Mail Shape Adjustment." Tr. 39/17777 (citations omitted). Witness Kay also revises the parcel and sector segment evaluation factors to 0.500 and 0.0610 respectively. Id. at 17780. This correction accounts for errors in the CRA spreadsheets filed in USPS-LR-I-80 and USPS-LR-I-130. These revisions cause an increase of $8.8 million in Standard Mail (B) Base Year 1998 cost. USPS-LR I-450 and LR-I-80.

[3244] The Commission accepts the recommendation of witness Glick to use annual RCCS data to develop the mail shape adjustment. The Commission also finds witness Hay's revised rural carrier cost as presented in LR-I-50 acceptable and uses them in the cost calculation.

[3245] Introduction. The costs for highway transportation services purchased under contract are included in Cost Segment XIV. The Highway Contract Support System (HCSS) contains information on "virtually" all of these contracts. USPST19 at 27. The contracts in HCSS fall under 27 account numbers, each of which relates to a category of transportation, such as Intra-BMC, Intra-P&DC Regular, and Intra-CSD Regular. The basic unit of observation used in the study of highway transportation is the "contract cost segment." Each segment "is a discrete part of a highway contract that has its own transportation specifications and its own payment type." A total of 16,791 observations are contained in HCSS. USPST18 at 22, 23, and 25. This number is large enough to permit detailed empirical analyses.

[3246] In this case, Postal Service witness Bradley, (USPST18) and (USPSLRI84-86), FGFSA witness Ball, (FGFSAT1), and Periodicals Mailers¹ witness Nelson, (MPAT3) and (MPALR5) provide testimony concerning how to analyze highway costs. The testimony of witness Bradley is similar to testimony he presented in Docket No. R971, which was adopted by the Commission. Bradley also submitted rebuttal testimony on this topic, (Tr. 43/18380 and USPSLRI452), as did witness Neels, Tr. 46E/21895-917. With the exception of UPS, rather extensive discussion of this testimony is included in the briefs and the reply briefs of the parties involved.

[3247] The Commission has reviewed the testimony, the workpapers, and the arguments on brief. The Commission finds the analysis of witness Bradley to be the best available on the record. No basis has been found for making improvements to his analysis, which is analytically similar to that adopted by the Commission in Docket No. R971. See the extensive discussion in PRC Op. R971, para. 3360 et seq. The alternative analyses presented on the record are not sufficiently developed to be useful. In fact, the final alternative supported on brief by the Periodicals Mailers relates to a modified Nelson model, developed for curiosity's sake by witness Bradley, and Bradley warns against its use. Therefore, the model is not adequately supported by any witness and has not been subject to review on the record.

[3248] Although the Commission does not adopt an alternative analysis, some of the arguments raised by the interveners are interesting and could be developed further. The Commission does not take the position that the analytical framework or the empirical analysis adopted in this Opinion represent perfection. Empirical analyses are difficult at best and it is often the case that more than one approach seems likely to yield defensible results.

[3249] Review of the analyses proposed and the controversy on the record will be facilitated if certain aspects of the methods used are outlined up front. The goal of the analysis has been to develop attributable costs for the subclasses of mail that use the transportation. Attribution has been premised primarily on concepts of volume variability, as developed and discussed extensively by the Commission since its inception in 1970. It has not been found feasible to address directly and in one step the cost effects of specific volume changes. For example, an equation has not been developed with transportation cost as the dependent variable and the volumes of each subclass of interest as separate dependent variables.² This does not, however, reduce the usefulness or the relevance of a thought process guided by the essentials of the volume variability question. The approach selected thus far for highway transportation, and developed extensively in hearings before the Commission, has been to break the analysis process into three steps.

[3250] The first step focuses on the elasticity of the cubic-foot-miles (CFM)³ of capacity purchased relative to a change in the overall volume of mail using the transportation segment being analyzed. The second step focuses on the elasticity of the cost of purchased transportation relative to a change in the cubic-foot-miles of capacity purchased. This step is viewed as relating to the behavior of pricing in the transportation markets in the sense that the cost at which transportation can be procured is related to the cubic-foot-miles of capacity involved; in the parlance used in recent years in such analyses, cubic-foot-miles of capacity is called a "cost driver" of transportation costs. The product of these two elasticities and the transportation cost is the volume variable cost. The third step utilizes a distribution key to apportion the volume variable cost to the particular subclasses of mail. The Postal Service's TRACS system provides this distribution key.⁴

[3251] The elasticity of the first step has not received extensive empirical analysis and has thus far been taken to be 1.0. Some observers tend to view 1.0 as an upper limit for an elasticity of this kind, although it is not, since a 10 percent increase in volume could lead to a greater-than-10 percent increase in the CFM of capacity purchased. The use of 1.0 for this elasticity is based on descriptions of postal operations and contracting practices. These descriptions suggest that a CFM-of-capacity figure is developed as a requirement based on such things as length of contract, peak loads, volume fluctuations, and the nature of round trip contractor runs. The argument is that if there is a 10 percent increase in overall postal volume, there will need to be an increase in the CFM-of-capacity purchased of about 10 percent. The Commission does not view this as a final answer; certain questions about it have been raised in this case.

[3252] The elasticity of the second step has been based on complex empirical analyses, using advanced econometric techniques. The Commission has used this approach since Docket No. R871 and it has received extensive review on the record. Both witnesses Bradley and Nelson provide econometric evidence relating to this elasticity. These will be reviewed further below.

[3253] The distribution key used in the third step is developed from data collected in the Postal Service's Transportation Cost System (TRACS), which was first introduced into rate proceedings in Docket No. R901. In this system, postal data collectors sample mail as it is either loaded on air transportation or unloaded from highway and rail transportation. The distribution keys are relative proportions from this sampling effort. Witness Xie, (USPST1), provides testimony concerning this system. TRACS is discussed in Section D. 2 of this Chapter.

[3254] The history of highway transportation analysis is contained in previous Commission Opinions and is summarized in the testimony of witness Bradley. USPST18 at 1-13. In early dockets, transportation costs were considered to be 100 percent volume variable. This was based on arguments that if the volume increased, a proportionate increase in transportation would be needed. This analysis did not allow for the realization of scale economies as volume increased.

[3255] An econometric analysis utilizing limited data was presented in Docket No. R841 and one utilizing a more extensive database was presented in Docket No. R871. In the latter case, the Commission based its recommendation on the econometric analysis. Although the details of the analysis became more complex and (with the HCSS in Docket No. R971) the database became more extensive, the roots of the analysis used in R97-1 may be traced to R871. The Commission summarized these roots in R971, briefly as follows: 1) the HCSS data are suitable for statistical purposes; 2) the models are consistent with perceptions of operational practice; 3) the translog model applied cross-sectionally to the HCSS data reflects the entire range of cost-affecting changes and gives the kind of "longer run" volume variability considered appropriate by the Commission; 4) the translog model is preferred due to its flexibility and to the freedom it provides the data to influence the functional form; 5) CFM of capacity is an acceptable cost driver and the elasticities should be evaluated at the mean of the observed values; and 6) it is best to include route length in the equations, but it doesn't vary much. PRC Op. R971, para. 3366.

[3256] The analysis of witness Bradley, USPST18, follows closely the analysis adopted by the Commission in Docket No. R971. The only structural difference relates to an adjustment to accommodate a new organizational structure of the Postal Service. Beginning at the national level, the old structure broke down to Region, Division, MSC, and SCF. The new structure breaks down to Area, Cluster, Processing and Distribution Center (P&DC), and Customer Service District (CSD). In R971, the principal categories of transportation accounts were Inter-BMC, Intra-BMC, Inter-SCF, Intra-SCF, Intra-City, and Box-Route. Under the new structure, the first two and the last two are unchanged, and the middle two (Inter-SCF and Intra-SCF) are disaggregated into Intra-P&DC, Intra-CSD, Inter-P&DC, Inter-Cluster, and Inter-Area.

[3257] Bradley indicates that one response to this change would be to map the new categories into the old categories, and to keep the same basic equations. However, what he does instead is to work with the new, more disaggregated categories and to test whether the new equations are statistically different from each other. His conclusion is that they are. Within the framework of the new account categories, Bradley estimates 17 translog equations. He describes his steps as being preestimation, estimation, testing the structure, and checking unusual observations. USPS-T-18 at 19.

[3258] On behalf of the Florida Gift Fruit Shippers Association, witness Ball provides testimony concerning the appropriate attribution of purchased highway transportation costs. He refers to the need to focus on the causal relationship between costs and volume. In reference to Bradley's analysis, he says: "[he] did not take into account mail volumes being transported or any changes in the mail volume. Vehicle capacity cannot properly be used as a proxy for mail volumes. Actual or estimated mail volumes, and changes in those mail volumes, are essential elements in the determination of variability or attribution of costs." Tr. 30/14302-303. He then discusses actual trends and says that "the utilization of vehicles for transporting mail has been low for several years, and continues to diminish." Then, without explaining how one would go about doing so, he proposes that the attribution of Inter-BMC and Intra-BMC transportation be based on some figures relating to the average rate of utilization of capacity, which were provided by Postal Service witness Xie. Ibid. at Table A.

[3259] The argument of witness Ball relates primarily to the elasticity of CFM capacity purchased with respect to changes in the volume of mail. This elasticity was identified above as the first step in the analysis. Aside from a statement that a "proxy" relationship may not exist, Ball provides no basis for assuming that the elasticity is on one side or the other of the assumed value, i.e., on one side or the other of 1.0. The Commission believes that further analysis would be desirable. It is possible, for example, that increases in volume allow increases in the rate of utilization of capacity, and thus do not cause the need for a proportionate increase in the CFM capacity purchased. The assumption that this elasticity can be approximated as 1.0, however, has been discussed extensively in the past and it is reasonable.

[3260] Witness Ball discusses a downward trend in the rate of utilization of capacity but does not say whether this trend is caused by a trend in volume. Also, the figures he relies on cover only two years. FGFSA argues that "[t]he low and declining utilization of the transportation vehicles demonstrates that the purchased transportation costs, which are increasing, are not variable with volume." FGFSA Brief at 5. The nature of the causal argument here could be that volume changes (a fact not established or mentioned) are causing changes in utilization rates. Since costs and utilization rates are affected by a number of factors, a more complete analysis would be needed to establish reliable relationships.

[3261] On behalf of the Periodicals Mailers, witness Nelson provides testimony covering, among other things, purchased highway transportation costs. Tr. 28/13408-13 and MPALR5. His work may be separated into three categories. First, he makes certain observations relating to the first step in the analysis, the determination of elasticity of CFM capacity relative to changes in volume. Second, he raises several questions concerning witness Bradley's analysis. Third, he presents an alternative analysis to that of Bradley. These may be dealt with separately.

[3262] Witness Nelson provides a discussion of the ways in which transportation systems adjust to changes in volume. One of the options is to reconfigure the system. Suppose, for example, that a truck leaves office A and makes stops at office B and C. It might off-load part of its volume at B and then drive on to C to off-load the remainder. Given a volume increase, it might be profitable to contract for two trucks, having one go the short distance from A to B, and back, and having the other go from A to C, and back. In a complex system, changes of this kind might allow, for example, a 10 percent increase in volume to be handled with only an 8 percent increase in the CFM capacity purchased. That is, the elasticity of CFM capacity purchased with respect to changes in volume may be less than 1.0. Tr. 28/13410, 13413.

[3263] Nelson is correct that these kinds of changes can be made. They relate to the first step in the analysis described above and it is possible that this elasticity is different from 1.0. No empirical analysis is available to allow an improved estimate of this elasticity and Nelson has not provided one. The Commission would welcome such an analysis, but notes the possibility that the correct answer could be 1.0. Indeed, the assumption that it is 1.0 is based on considerable testimony concerning how the transportation system functions.

[3264] On rebuttal, Bradley discusses Nelson's speculations on the elasticity in the first step and agrees that there may be some "merit" in his arguments. Tr. 43/18388. Then Bradley goes on to note that the possibility of this elasticity being different from 1.0 does not imply that there are any problems or weaknesses in the econometric models used to estimate the elasticities in the second step. Bradley is correct on this point.

[3265] Witness Nelson discusses four potential weaknesses in Bradley's econometric models used to estimate the elasticity for the second step in the analysis. The first concerns Nelson's belief that the models do not allow for the effect of increasing vehicle size. The second concerns Bradley's use of mean-centered data. The third concerns possible circularity in the handling of power-only contracts. The fourth concerns the way Bradley handles outliers. Bradley does exclude a few observations as outliers, but the number is under 1.5 percent. USPST18 at 29. Also, Bradley discussed his exclusion of these observations and analyzed the resulting effects. The Commission considers Bradley's testimony on this issue to be credible and the effect to be relatively small. Further, Bradley's treatment of outliers is similar to that of the Commission in Docket No. R971. Therefore, the outlier issue will not be discussed further.

[3266] Vehicle size. Witness Nelson describes vehicle-contracting practices, as he sees them, and explains that increased volume can often be handled by increasing the size of the truck used on a specific run. He says many runs have small trucks and that the percent increase in cost for increasing the size of the truck can be small relative to the percent increase in CFM capacity. Nelson says Bradley's econometric models are specified in a way that makes them "inconsistent" with a system that adjusts in this way. Tr. 28/13408-11.

[3267] On rebuttal, witness Bradley argues that the data set used to estimate his model is effectively a "census of all Postal Service purchased highway transportation contracts" and thus reflects actual experience and "all historical changes in both vehicle size and trip frequency (as well as routing)." Tr. 43/18387. The results found by Bradley show variabilities well below 100 percent on many route types. Such results imply economies of scale. There must be an explanation for why such economies are available. The option to increase vehicle size is one possible explanation. Nelson has apparently not considered that his truck-size argument may be a part of the explanation for the variabilities that Bradley found, instead of an argument for why they should be lower. The Commission agrees with witness Bradley that the data set covers the kind of adjustments described by Nelson. Whether or not a different model specification might do a better job of quantifying the truck-size effect is a different issue and will be discussed further below.

[3268] Mean-centered data. There has been an extraordinary amount of discussion on this record of the practice of mean-centering data, which Bradley did. Nelson says he attempted to refine Bradley's models and "encountered immediate difficulties with witness Bradley's evaluation method, in which the model is estimated on mean-centered data and the variability is taken from the coefficient on the relevant first-order term." Tr. 28/13412. In response to an interrogatory positing that the difficulties were in Nelson's model and not Bradley's, Nelson said:

As shown in my Workpaper WP1 and explained in further detail in my Workpaper WP4, the inter-BMC model yielded statistically insignificant (and negative) results for the CFM variable, but good statistical significance for the squared and cross-product terms that contain the CFM variable. I concluded from this that witness Bradley's approach of evaluating the elasticity only from the (mean-centered) first-order term may produce implausible and unusable results in the context of the modified specification being estimated, and that the results from the translog specification may be quite sensitive to the evaluation method chosen.

[3269] On brief, the Periodicals Mailers contend that "the Postal Service's own witness, Dr. Greene, and UPS witness Neels have demonstrated the flaws in Bradley's `mean-centering' method of calculating translog elasticities from only the first-order coefficients." Periodicals Mailers Brief at 32-33. Their Reply Brief reiterates that "Bradley's theory for relying on `mean-centering' to simplify the computation of variabilities has been refuted by Greene and Neels, and Neels asserts that Bradley's program does not correctly implement the computation that Bradley was trying to perform." Id. at 22.

[3270] The Commission has reviewed this issue carefully, partly because mean-centering has been used by the Commission in the past. Postal Service witness Greene specifically states, at Tr. 46E/22078, that mean-centering would give different coefficients but that the elasticities would be the same. UPS witness Neels states that mean-centering is a "computational convenience" and "shouldn't change the result." Id. at 21925. Short of calculating the elasticity with an erroneous formula, there does not appear to be support for the Periodicals Mailers' position.

[3271] The Commission does not accept that there are difficulties with the use of mean-centered data. Using mean-centered data is a common practice. It involves dividing each point in the data by a constant, which happens to be the mean of the data set. If mean-centering changed the elasticity, then one would get different volume-variable costs by measuring the costs in cents instead of dollars, an obviously absurd implication.

[3272] Circularity in power-only contracts. Power-only contracts exist primarily in BMC and plant load accounts. Nine of the 13 areas report using them in BMC contracts. USPST18 at 26-27. Of 552 contracts in the BMC accounts, 185 are power only. Tr. 6/2405. Witness Nelson indicates that Bradley's "treatment of power-only contracts appears to be circular at best." Tr. 28/13411. He says this because Bradley (as did both Bradley and the Commission in R971) used an average number of cubic feet for the trailers in each region for the power-only contracts, and then used a dummy variable to account for differences between the areas.⁵ The averages range from 2,596 cubic feet to 3,228 cubic feet. Also, witness Bradley reports being informed that the cost of the trailer is less than 5 percent of the total cost of a tractor-trailer contract. USPST18 at 24-26

[3273] On rebuttal, witness Bradley explains that there are only a limited number of trailer sizes under power-only contract in each area and that the variable being used is cubic-foot-miles, not just cubic feet. Tr. 43/18389-91. Nelson has not provided a clear explanation of the nature of the problem and the effect of Bradley's handling of power-only contracts. The Commission agrees with Bradley that they should be included in the regressions and that the use of an estimate based on a Price-Waterhouse study is appropriate.

[3274] Because he believes that the specification of Bradley's model is inconsistent with the way route adjustments are actually made and is therefore unable to provide elasticities that quantify the actual behavior of the costs of highway transportation operations, Nelson proceeds to develop models of his own. Before actual estimation, however, he stratified the data into three groups and normalized the data for the number of runs.

[3275] Nelson reasons that contracts supplying the largest vans cannot respond to volume increases by using an even larger van. He says that the only way to expand these contracts is to add runs, and that 10 percent more runs generally cost 10 percent more. Therefore, he separates these contracts from the others and assumes that they have a variability of 100 percent. Tr. 28/13411-12. On rebuttal, witnesses Bradley and Neels note this stratification but do not criticize it. Tr. 43/18393-94 and Tr. 46E/21906. The Commission does not find this stratification objectionable. Such a stratification or a similar one could be part of a useful analysis. On the question of whether the assumed variability of 100 percent is reasonable, the Commission takes no position. Variabilities can be both above and below 100 percent.

[3276] Another stratification made by witness Nelson is to separate power-only contracts and to piggyback them on the variability of the remaining contracts. Tr. 28/13412. Questions relating to power-only contracts were addressed above. The Commission does not find their inclusion in the data set used for the models to be objectionable. Based on further analysis, this issue could be addressed in a future case.

[3277] Before estimating any models, Nelson normalized the data for the number of runs. This means, for example, that the cost and the CFM capacity for a contract providing 9 runs were each divided by 9. Tr. 28/13412. This procedure, in and of itself, is not objectionable. It could be part of useful analysis, depending on the modeling exercise. However, this process did give rise to questions on the record about whether the resulting observation should be weighted by the number of runs in the contract, as Nelson did.⁶ Ibid. On rebuttal, witness Neels finds the weighting scheme to be "inappropriate." Tr. 27/12789-90. The Commission does not take a position on weighting schemes. It is certainly true, as Neels indicates, that "[a] contract is still only one contract, regardless of how many runs it covers." Ibid. at 12790 and 12787. On the other hand, if 5 separate runs, each fitting a certain cost-CFM relationship with a random disturbance term, were collected into one contract and normalized, the size of the error term would be expected to be smaller. Arguably, then, one could be more certain that that point lies near the estimated relation. This issue should receive greater attention if models are presented that focus on normalized data.

[3278] Witness Nelson's actual modeling occurred in three steps. The first model adopted a translog specification and took the form cost/run = f(CFM capacity/run, route-length). Nowhere does he provide any discussion of why a model of this form should be able to represent how actual operations are adjusted, given a volume change. He reported difficulties with this model and rejected it. His second model adopted a log-log specification and took the form cost/run-mile = f(cube, 1/route-length). For this model, the data were stratified by route length, another potentially interesting stratification. Again, no discussion was provided justifying this particular specification. Tr. 28/13412. See also Tr. 43/18393-96. Again, Nelson rejected the results from this model.

[3279] Nelson's third model, and the one on which he relied for his proposed variabilities, also adopted a log-log specification and took the form cost/run = f(CFM capacity/run, run length). He says this model "generally exhibited a high degree of explanatory power, and high statistical significance for the variables needed to estimate the relevant elasticity." Tr. 28/13413.

[3280] As explained in considerable detail by witness Neels, Tr. 46E/21906-11, this model is a restricted form of the translog model, the latter being a model that has been used by the Commission for transportation analysis since Docket No. R871. When a restricted form is used, it is incumbent on the analyst to explain why the constraints and limitations in that form make it more likely to be able to infer the elasticity information desired. Aside from statements that his model corrects inconsistencies in Bradley's model, Nelson has not provided this explanation. The Commission notes that both of the "independent" variables in the equation (CFM capacity/run and run length) are influenced by changes in the run length. Therefore, the equation will encounter collinearity problems.⁷ Such problems cannot always be avoided, but one would prefer to face them only when the model can be strongly justified on other grounds.

[3281] Extensive rebuttal testimony focusing on this third model is provided by witnesses Bradley, Tr. 43/18398-418, and Neels, Tr. 46E/21905-29. According to witness Bradley, the model: 1) is not justified on a relationship recognized in economic theory; 2) is not justified on a description of how actual operations adjust to volume changes; 3) does not have a sound mathematical basis; 4) is not estimated with state-of-the-art econometric techniques; 5) was estimated erroneously; and 6) does not yield results that are robust and consistent. Neels supports many of these contentions.

[3282] Bradley then goes on to take an unusual step, which he says he takes at least partly for the sake of curiosity. In effect, he says: if I were going to build a log-log model with cost/run as the dependent variable, and avoid all of Nelson's errors, how would I do it and what would the results be? He believes the most defensible starting point would be a cost function specifying:

Tr. 43/18401. He estimated this equation omitting power-only contracts, using Nelson's filters, and using a corrected version of Nelson's segregation by truck capacity. Note that estimation provides a value for a, a value for b-1, and a value for b. The variability results are provided in Bradley's Table 6. Tr. 43/18414. Additional details were provided by Bradley in response to a question asked during cross examination. See Id. at 43/18462-64.

[3283] The Periodicals Mailers now take the position that Nelson's original analysis should be discarded and that witness Bradley's results in Table 6 should be used. Periodicals Mailers Brief at 33-34 and Reply Brief at 23-24. They take this position despite the fact that Bradley specifically warns against using these results. In fact, Bradley explains that since he gets a value for b-1 that differs from the value for b, there is a strong statistical suggestion that the specification of the model should be rejected. Tr. 43/18462-64.

[3284] The Commission is left with a recommendation on brief that a result be used, Bradley's Table 6, that is not supported on the record by any witness, that has not been reviewed or tested on the record, and that is denounced by the witness who prepared it. The Commission does not find it necessary to provide a specific assessment of every point made on rebuttal by witnesses Bradley and Neels. The Commission uses the analysis provided by Bradley in his original testimony.

[3285] In this docket, as in previous cases, the Postal Service uses the Transportation Cost System (TRACS) to distribute transportation costs to mail categories. USPS-T-1 at 2. TRACS consist of five subsystems: Commercial Air, Passenger Rail (Amtrak), Network Air, Highway, and Freight Rail. Ibid. In this docket discussion of the distribution of transportation costs focused on purchased highway transportation. Regarding purchased highway transportation, TRACS sampling is a three-stage sample with a primary sampling unit, stop-days, defined as "all mail unloaded from a truck at one facility on a specific trip, on a specific day." USPS-LR-I-52 at 3. In the first stage, a random sample of stop-days, stratified based on facility type and whether the segment is on an out-bound or an in-bound trip, is selected. The second stage is a stratified sample of off-loaded containers, pallets, and loose items. The third stage is only applicable to wheeled containers and is a stratified sample of items from each selected container.

[3286] The sampling process gathers data on truck utilization, mail class proportions, and miles traveled. The truck utilization data are collected before any mail is off-loaded and consists of the percentage of empty floor space, space occupied by mail being unloaded, and space occupied by mail remaining on the truck. The height, length, and width of pallets are recorded along with weight and volume information related to sampled items. These data, coupled with the origination facility and truck utilization information, are used to estimate cubic-foot-miles by mail class.

[3287] The estimation of cubic-foot-miles by mail class involves allocating empty space to the classes of mail. The Postal Service accomplishes this by `expanding' the sampled mail classes to include empty space. The empty space is distributed to off-loaded mail in proportion to the estimated truck space occupied by the sampled mail.

[3288] The cost of the sampled trip is multiplied by the cubic-foot-mile proportions to estimate the cost of the trip by mail category. These costs are then expanded to represent all trips in the quarter. The distribution key is the expanded cost of a mail category divided by the total expanded cost.

[3289] Intervenor Proposals. Two parties, United Parcel Service (UPS) and Florida Gift Fruit Shippers Association (FGFSA) propose modifications to the manner in which TRACS distributes purchased highway transportation costs. FGFSA witness Ball criticizes TRACS in several respects. First, he contends that the allocation of Intra-BMC samples fails to reflect relative mail volumes and is biased. Tr. 30/14299; see also FGFSA Brief at 7. Second, FGFSA opposes the manner in which TRACS allocates empty container and vehicle space. Tr. 30/14299-300; see also FGFSA Brief at 7. Witness Ball argues that "expansion" penalizes mail on in-bound trips which has lower vehicle and container utilization. Tr. 30/14299-300. Third, FGFSA argues that Inter-BMC samples erroneously reflect a distribution key for DBMC parcels, which, FGFSA asserts, do not utilize inter-BMC transportation. Similarly, FGFSA contends that Intra-BMC samples erroneously reflect DBMC parcels on the in-bound trip back to the BMC, which, according to FGFSA, defies reason. Tr. 30/14300-01; see also FGFSA Brief at 9. Finally, FGFSA advocates the use of cubic feet in lieu of cubic-foot-miles to distribute DBMC purchased highway transportation costs for Standard (A) and Parcel Post. The predicate for this proposed redistribution is that cubic-foot-miles differs dramatically from the cubic feet reported for these mail categories. Tr. 30/14301-302; see also FGFSA Brief at 9-10.

[3290] UPS witness Neels criticizes the Postal Service's current method of allocating empty space, which involves the expansion of mail actually found on the sampled truck to fill the unused space. Tr. 32/16006. He argues that the "procedure places greater weight in the cost distribution process on the mail mix on trucks with lower capacity utilization." Id. at 16008. Witness Neels' proposal gives greater weight to the classes and subclasses of mail on the more fully loaded trucks. His approach involves determining a distribution key for the mail mix found on more fully loaded trucks, allocating empty space on the basis of this key, and then determining an overall key for the distribution of transportation costs. This approach requires identifying the more fully loaded trucks, which he accomplishes by arraying the sampled segments and finding the segments with the highest capacity utilization. To determine the fraction of segments to include, he uses the inverse of the average number of segments per trip. Tr. 32/16014. See UPS Brief at 33-34.

[3291] In addition, witness Neels argues that TRACS appears to underrepresent time-sensitive mail because it fails to include emergency or exceptional contracts, and TRACS inspectors may bypass time-sensitive mail in an effort to avoid disrupting the movement of this mail. Tr. 32/16019-22. Witness Neels outlines several problems concerning the representativeness of the TRACS distribution keys, e.g., the misallocation of samples to strata, and that the sample design requires updating to reflect the Postal Service's current operating environment. Id. at 16025-33. Finally, witness Neels suggests possible improvements in the TRACS sampling, e.g., supplementing the TRACS data with data from the Transportation Information Management Evaluation System (TIMES), and changing the sampling procedures so that all segments on a trip are sampled. Id. at 16034-38.

[3292] Postal Service Rebuttal. On brief, the Postal Service responds to each of FGFSA's contentions. The Postal Service dismisses FGFSA's claims that the Intra-BMC samples are biased, explaining that the Horvitz-Thompson type estimator used by TRACS reflects the selection probabilities and therefore produces unbiased estimates of cubic-foot-miles. Postal Service Brief at III3. The Postal Service also rebuts FGFSA's claims that Intra-BMC and Inter-BMC samples erroneously reflect DBMC parcels, asserting that FGFSA misperceives the transportation network. According to the Postal Service, intermediate stops are made under both Inter-BMC contracts and the Intra-BMC network. In addition, DBMC parcels that are mis-sorted, mis-sent, mis-entered, or returns may be found on Inter-BMC and Intra-BMC transportation. The Postal Service asserts that the presence of DBMC parcels in the sample is indicative of the robustness of the sampling. Postal Service Reply Brief at III-9-III-10.

[3293] Witness Eggleston rebuts FGFSA's proposal to use cubic feet as a distribution key. She contends, inter alia, that witness Ball's comparison of cubic feet and cubic-foot-miles for Parcel Post and Standard Mail (A) is invalid because it improperly compares transportation modes and rate categories, and further because it incorrectly presumes that cubic feet and cubic-foot-miles are directly related. Tr. 41/18163-64; see also Postal Service Brief at III6-III-7, and Postal Service Reply Brief at III4-III-7.

[3294] Witness Bradley criticizes witness Neels' proposal to revamp the allocation of empty space, contending it is flawed. However, he proposes a compromise method for allocating empty space, one that uses data on both the tested legs and the more fully loaded trucks. Witness Bradley assets that this method is more accurate than either Neels' or the current method, and partially addresses the Commission's desire, espoused in Docket No. R971, to separate the TRACS calculation of cubic-foot-miles from the expansion process. This approach yields transportation cost distributions that are approximately midway between the Postal Service's and UPS's methods.

[3295] On brief, the Postal Service dismisses witness Neels' contentions that TRACS appears to under-sample time-sensitive mail as purely speculative. Postal Service Reply Brief at III8-III-9.

[3296] Commission Analysis. In Docket No. R971, the Commission expressed several concerns about TRACS, including its documentation, potential sampling bias, and highway distribution keys. The Postal Service has endeavored to address each. First, it has improved the documentation for TRACS, adding sample design data, estimation formulas, processes, and procedures. The Postal Service also produced final analysis files (Z-file) that facilitates review of the system by both the Commission and participants. See Postal Service Brief at III-2-III-3; and Postal Service Reply Brief at III-3-III-4. Second, the Postal Service modified its sampling of in-bound and out-bound Inter-BMC and Intra-BMC trips from 70 percent and 30 percent, respectively, to 51 percent and 49 percent, respectively. See Postal Service Brief at III-3, and Postal Service Reply Brief at 2-3. Third, in response to the Commission's concern about possible bias on in-bound and out-bound trips due to mail mix, the Postal Service revised its procedure for computing the highway distribution key to conform to the cost driver for highway transportation, namely, cubic-foot-miles. The new procedure uses separate weights for in-bound and out-bound trips. See Id. at III4-III-5. In addition, the Postal Service revised its variance formula for estimating the coefficients of variation for the highway distribution keys. The revised formula results in lower CVs for the major mail categories. Id. at III-5-III-6. The Commission commends the Postal Service for addressing these concerns.

[3297] FGFSA's various criticisms of TRACS are unavailing. It offers only general assertions that TRACS is biased without the necessary statistical evidence demonstrating the point. Moreover, FGFSA's principal claim, that TRACS is biased because it fails to reflect relative volumes, is misplaced. As the Postal Service notes, witness Xie provides detailed information about TRACS that refutes the claim that the sample allocation must be proportional to mail volumes. See, e.g., Tr. 17/6751, 6796, and 6845-48; see also Postal Service Brief at III-3 and Postal Service Reply Brief at III1-III-3. The Commission finds this testimony to be persuasive.

[3298] Similarly, the Commission is also persuaded by the Postal Service's response to FGFSA's contention that Intra-BMC and Inter-BMC samples erroneously include DBMC parcels. See Postal Service Reply Brief at III-9-III-10. Finally, for reasons outlined by witness Eggleston, the Commission rejects FGFSA's proposal to use cubic feet as a distribution key. See Tr. 41/18163-64 and Postal Service Reply Brief at III4-III-7.

[3299] Witness Neels' analysis of TRACS is useful, focusing attention not only on the allocation of empty space, but also on sundry related issues. On brief, UPS limits its discussion to two of those issues, namely, Neels' allocation proposal and the sampling of time-sensitive mail. UPS Brief at 33-35. For its part, the Postal Service endorses Bradley's alternative proposal. Postal Service Reply Brief at IV81-IV-83.

[3300] Witness Bradley's compromise proposal is well taken. It produces a distribution key that better reflects actual usage of transportation capacity. As noted, the Postal Service endorses it; UPS, while preferring Neels' proposal, finds it acceptable. Accordingly, the Commission adopts Bradley's compromise method of allocating empty space.⁸

[3301] UPS makes a passing reference to emergency contracts and exceptional service highway movements, arguing first, that they are likely to contain a higher proportion of time-sensitive mail, and second, that if sampled they likely would result in higher Priority Mail attributable cost levels. UPS Brief at 34. While the first inference may appear reasonable, there is, as the Postal Service argues, no evidence corroborating it. Nor is there any evidentiary support for the second inference. Witness Neels indicates that these movements comprise approximately 16 percent of total transportation costs. Tr. 32/16020. On brief, the Postal Service suggests that reliably sampling these movements poses certain problems. Postal Service Reply Brief at III-8-9. Nonetheless, the Postal Service should attempt to develop a means of better reflecting the costs of these movements in its sampling results.

[3302] Finally, witness Neels' various suggestions regarding TRACS go unaddressed by both UPS and the Postal Service. See Id. at 16025-38. While the record is not developed regarding these suggestions, they would appear to merit more than cursory rejection, and the Postal Service is encouraged to review them substantively.

[3303] Since Docket No.R90-1, a portion of the intra-Alaskan air transportation costs has been classified as institutional and removed from the attributable cost base of Parcel Post. This adjustment recognizes the unique nature of mail delivery to the parts of Alaska where road access is limited. This adjustment was reviewed and sustained in United Parcel Service v. U.S. Postal Service, 184 F.3d 827 (D.C. Cir. 1999). In this proceeding witness Bradley proposes a change in the calculation of the Alaska air adjustment. Rather than calculating average cost per cubic foot and cubic-foot-mile by an unweighted average as was done previously, he suggests using a weighted average. USPS-T-18 at 59-61. The proposal is unopposed and the Commission accepts it.

[3304] The Postal Service operates three dedicated air networks devoted to the transportation of mail. The Eagle network is a hub and spoke operation, located in Indianapolis, Indiana, which links approximately forty cities. See USPS-T-1 at 13, USPS-T-19 at 1, and Tr. 32/15596. The Western network is also a hub and spoke operation, located near Oakland, California, connecting approximately a dozen cities in the western United States. Ibid. The Christmas network, which operates for the two weeks prior to Christmas, is a daytime operation designed to transport expedited mail volumes. Ibid.

[3305] The Postal Service divides the costs of each of these networks between network premium costs and volume variable, non-premium costs. For each network, the network premium cost is the difference between the actual cost of each network and the hypothetical cost of providing the same service via passenger (commercial) air. As proposed by the Postal Service, the premium costs associated with the Eagle and Western networks are treated as incremental to Express Mail. See, e.g., USPS-T-22 at 34. As initially reported by the Postal Service, the network premium costs were $102 million for the Eagle network and $22.7 million for the Western network. USPS-LR-I-57 at 1. The premium cost of the Christmas network is treated as incremental to Priority Mail. USPS-T-19 at 2.

[3306] In this proceeding, the Postal Service proposes two changes in the manner in which it calculates premium costs. First, for each network, the Postal Service proposes to calculate the linehaul portion of the premium costs based on origin-destination great circle miles in lieu of aircraft route miles. This will make the calculation of premium costs consistent with the manner in which passenger air linehaul costs are incurred. Ibid. The second change concerns the use of certain Eagle planes during the daytime, "Daynet turns" in the Postal Service's lexicon. These daytime operations, which began in PQ 2 of FY 1998, "were designed to substitute for passenger air transportation, to better meet the service commitments of so-called two and three day mail (non-local First-Class and Priority Mail)." Id. at 4. The estimated cost of the Daynet turns was eliminated from the Eagle network costs and assigned, within Cost Segment 14, to the Passenger Air cost pool. Ibid. See also USPS-LR-I-60 and USPS LR-I-1 at 14-1 et seq. for a general discussion of Cost Segment 14.

[3307] Witness Neels, on behalf of UPS, proposes that the network premium costs associated with the Eagle and Western networks should be attributed to Priority Mail in addition to Express Mail. See Tr. 32/15996-16004. In support of this position, witness Neels compares the base year volumes of Express Mail and Priority Mail carried on these networks, and concludes that the Postal Service would be unlikely to incur the premium costs of operating these networks, as currently configured, solely to transport Express Mail. According to witness Neels, Express Mail represents 24 percent of the Eagle network volumes and 9 percent of the Western network volumes compared to 47 and 54 percent, respectively, for Priority Mail. Id. at 15998. Witness Neels contends that the networks are sized to handle both Express Mail and Priority Mail volumes. He gives no credence to the Postal Service explanation that its decision to upgrade capacity on the Western network using Boeing 727s was a product of the bid specifications. He argues that the configuration of that network was driven by the need to improve Priority Mail service. Id. at 15999-16001. In addition, he argues that smaller aircraft could serve both networks. Based on these arguments, witness Neels proposes to increase the BY 1998 allocation of domestic air costs to Priority Mail by approximately $65 million, while reducing the allocation to Express Mail by approximately $93 million. Id. at 16004. In percentage terms, the reallocation results in a 60 percent decrease in domestic air costs allocated to Express Mail and a 13 percent increase to Priority Mail. On brief, UPS reiterates witness Neels' testimony, while characterizing the Postal Service's arguments as red herrings. See UPS Brief at 29-33, and UPS Reply Brief at 35-38.

[3308] The Postal Service, through the rebuttal testimony of witness Pickett, contends that witness Neels' arguments are predicated on misunderstandings of postal operations. See Tr. 43/18531-38. For example, witness Pickett indicates that both aircraft speed and load characteristics, i.e., the ability to accept containerized loads, are critical considerations in the efficient operations of both networks. Witness Pickett states that the Boeing 727, while not the only aircraft that would have satisfied the Postal Service's requirements, proved, on balance, best suited for the Postal Service's needs for reasons, among others, of compatibility between the networks, cost, and flexibility. Regarding the volumes transported on each network, witness Pickett notes that the relative share of Express Mail increased in FY 1999, but, more importantly, that the average annual volumes mask the need to serve swings substantially exceeding the average. Witness Pickett also points to the recent implementation of the Priority Mail Processing Center (PMPC) network as an indication that the Eagle network is not caused by Priority Mail. This follows, according to witness Pickett, because the advent of the PMPCs has not caused a reconfiguration of Eagle flights in the affected areas. Id. at 18536-37. Witness Pickett concludes that "overnight dedicated air networks are absolutely needed to support a guaranteed overnight product." Id. at 18538; see also Tr. 17/6718-19.

[3309] Addressing this issue in Docket No. R97-1, the Commission concluded that a causal link exists between the guaranteed overnight service and the network fixed costs that required the fixed costs of the Eagle and Western networks be attributed to Express Mail. PRC Op. R97-1, para. 3399. In addition, the fixed cost of the Christmas network was attributed to Priority Mail. Based on testimony in that docket, the Commission concluded that fixed network costs should be attributed solely to Express Mail since if it were eliminated, the Eagle and Western networks would not be retained, and Priority and First-Class Mail would be transported via commercial air without degrading service quality.

[3310] As UPS notes on brief, the Commission's finding in Docket No. R97-1 represented a departure from earlier decisions. UPS Brief at 29, 31. The Commission's conclusion, however, was based on testimony that the Eagle network would cease operations if Express Mail were eliminated.

[3311] The Commission is not persuaded by witness Neels' analysis. Several factors influence this conclusion. First, the current allocation is based on causality, specifically on the evidence that the networks would no longer operate if Express Mail were eliminated. In this proceeding, the Postal Service reiterates this point, stating that the overnight networks are configured for Express Mail and without that product the overnight network would be superfluous. Tr. 43/18538; see also Postal Service Reply Brief IV-84-IV-87. Witness Neels fails to demonstrate that this is not the case. For example, focusing largely on the Western network, witness Neels argues that larger aircraft, Boeing 727s, are used to accommodate transportation of both Express and Priority Mail. In rebuttal, however, witness Pickett outlines several inadequacies of possible alternatives ranging from the general, e.g., from cruising speed and containerization, to the more specific, i.e., the cities common to both networks. See also Postal Service Reply Brief at IV-85. Moreover, witness Pickett adequately explained why witness Neels' reliance on a 1995 memorandum for the proposition that the Western network was reconfigured to provide, inter alia, improved service for Priority Mail was misplaced. Tr. 43/18537-38; see also Postal Service Brief at V-155 and Postal Service Reply Brief at IV-85-IV-86.

[3312] Second, UPS argues that the networks could be operated with smaller planes. UPS Brief at 31-32 and UPS Reply Brief at 35. On redirect, witness Neels did suggest that smaller jet aircraft could be used. Tr. 32/16133-34. Notably, however, his prepared testimony referenced only turbo props, which were not shown to be a suitable substitute for jets on both networks. See id. at 16116-17. While smaller aircraft were used on the Western network prior to August 1999, they are subject to limitations, e.g., range, lift, and avionics, which restrict their ability to serve the network. Moreover, there has been no showing that they could be used on the Eagle network. Finally, even if smaller jets were considered, it does not necessarily follow that network operating costs would decline. Id. at 16106-07; see also Tr. 45/19597 and APMU Brief at 30.

[3313] Third, witness Neels' analysis suffers, comparatively, from a less than full understanding of the Postal Service's network operations. See, e.g., Tr. 32/16101-07, 16109-10; see also Tr. 43/18535 and Tr. 32/16113. This is not meant so much as criticism but as fact, which makes the Commission reluctant to reallocate substantial amounts based on speculative, even if well-presented, theories. Fourth, under witness Neels' premise, commencement of the PMPC operations should, logically, have caused downsizing or reconfiguration of Eagle flights in the affected areas. Witness Plunkett indicates that neither occurred. Tr. 43/18537; see also Postal Service Reply Brief at IV87. Finally, while average relative volumes are not inconsequential, the more critical inquiry, given the time constraints attendant operation of a guaranteed overnight service, is the ability to accommodate variations in demand. In that regard, the evidence is insufficient to refute the Postal Service's position, as developed by its witnesses, that, but for the need to support a guaranteed overnight product, the overnight network would be unnecessary. Tr. 43/18538; see also Postal Service Reply Brief at IV-84.

Nelson's testimony is sponsored by MPA, ANM, ABM, CRPA, Dow Jones, McGraw-Hill, NNA, and Time Warner, referred to in this section as the Periodicals Mailers.

If such an equation were available, the unit attributable cost of, say, subclass A would simply be the partial derivative of the cost with respect to the volume of A.

Note that CFM could refer to cubic-foot-miles of capacity purchased but could also refer to cubic-foot-miles of capacity actually used. Often, CFM is unmodified but is meant to refer to the former measure. As clarified on discovery, witness Nelson refers to the former as "gross" CFM and to the latter as "net" CFM. Tr. 28/13410 and 13460.

⁴

For further discussion of this method of analysis, see: PRC Op. R97-1, para. 3376, USPST18 at 44, and Bradley, Colvin and Smith, "Measuring Product Costs for Ratemaking: The United States Postal Service," in Regulation and the Nature of Postal and Delivery Services, Kluwer, 1993, 133-157.

⁵

Nelson's concerns are taken to apply to BMC contracts. Bradley indicates that the trailers used in the plant load contracts are all the same size due to TOFC (Trailer on Flat Car) specifications. USPS-T-18 at 27.

⁶

Another question, not raised on the record, concerns whether the range of variation of normalized data would be smaller. Econometric models are often viewed as approximating reality over a limited range of variation of the variables involved.

⁸

FGFSA witness Ball contends that the current method of allocating vehicle and container empty space penalizes mail on inbound trips. Tr. 30/14299; FGFSA Brief at 8. Neither contention is adequately supported. FGFSA took no position on witness Bradley's compromise. FGFSA Brief at 8. It provided unsubstantive evidence to buttress its claim regarding the allocation of container empty space. See Postal Service Reply Brief at III-11-III-12.