| |
|
|
|
[3001] In this docket the Postal Service renews its proposal to apportion Segment 3 Clerk and Mailhandlers Costs to the Mail Processing, Administration, and Window Service Components. In doing so, it proposes to give priority to MODS information contained in IOCS tallies. As in Docket R97-1, the Commission concludes that in some instances the recorded observations of the IOCS data collectors concerning the nature and location of work activities are more reliable. The reasons for this conclusion are discussed in Section 1.
[3002] The Postal Service also proposes to estimate the volume variability of mail processing labor costs with an econometric model similar to the one presented in Docket No. R97-1. The model presented in this docket yields even lower variabilities for the analyzed MODS pools. As in Docket No. R97-1, the Commission concludes that these results are unreliable. The reasons are summarized in Section 2. The Commission adheres to the established finding that most mail processing labor costs change in proportion to volume.
[3003] The Postal Service proposes to modify its method for distributing volume variable mail processing labor costs to subclasses. It proposes to distribute the costs associated with allied not-handling tallies to subclasses in proportion to the direct tallies from all Function 1 cost pools. The Commission adheres to the established method, which distributes those costs in the same manner that the costs of direct and mixed tallies are distributed within each allied pool. The reasons for doing so are provided in Section 3.
[3004] Cost Segment 3 payroll costs are the clerk and mailhandler labor costs in CAG A-J offices. In BY 1999 these costs exceeded $18.2 billion. The Postal Service associates approximately $13.8 billion of these costs with mail processing. PRC LR-5, CS 3 Worksheet 3.0.1. Approximately $10.4 billion of these costs are incurred in the more than 300 offices that collect Management Operating Data System (MODS) information. Id. Worksheet 3.1.1a. MODS records hours worked by employees while they are clocked into specific mail processing operations, the payroll costs associated with those hours, and the work effort expended in those activities, in terms of the number of piece handlings performed. Employees must be clocked into a specific activity in order to be paid. They are supposed to reclock whenever their work assignment changes. The 21 Bulk Mail Centers (BMCs) have a similar data management system called Productivity Information Reporting System (PIRS).
[3005] Subclass responsibility for Segment 3 costs is determined with data from the In-Office Cost System (IOCS). IOCS data collectors observe randomly selected workers at random instants in time. They record the worker's activity, and, if possible, identify the subclass or special service with which the activity can be associated. Each IOCS sample is called a tally. Samples are drawn by craft within each CAG. Each tally is dollar weighted so that the sum of all the weighted tally dollars equals the total labor costs in the sample frame from which the tallies are drawn. USPS-T-2 at 2-6.
[3006] In MODS offices, the IOCS tally taker also records the code of the MODS activity that the employee is clocked into. In some instances the clocked activity differs from the activity actually observed by the IOCS data collector. When a MODS code is invalid or missing, the Service may use the data collector's response to IOCS Questions 18 and 19, and/or the IOCS operation code to assign tallies to a particular mail processing cost pool. USPS-T-12 at 8.
[3007] Prior to Docket No. R97-1, the Postal Service apportioned Segment 3 costs to its Mail Processing, Window Service, and Administration components using IOCS data. In Docket No. R97-1, the Service proposed relying primarily on MODS data for this purpose. The Commission accepted some aspects of this proposed reapportionment and rejected others. In this docket, the Postal Service continues to use IOCS tallies to apportion Segment 3 costs in BMCs and non-MODS offices into Mail Processing, Window Service, and Administration components. For MODS offices, the Postal Service again proposes to apportion Segment 3 costs to components according to the MODS record of the activity an employee was clocked into even where it conflicts with the activity that the IOCS data collector actually observed being performed. Resolving all conflicts in favor of MODS data would cause $72.2 million of IOCS-defined Window Service and $537.6 million of IOCS-defined Administration costs, to "migrate" to the Mail Processing component. PRC LR-5, CS 3.0 Worksheet 3.01a. Migration of tallies into mail processing costs pools would change the variability factors and the distribution keys from those associated with Window Service and Administration to those associated with Mail Processing. In R97-1, the Commission rejected this proposed migration.
[3008] The Postal Service argues that the MODS operation code recorded in an IOCS tally provides a means for tying the window service and administration activities to the mail processing activities "representing `Function 1' (mail processing plant) or Function 4 (station and branch) support operations." This association with the supported activities allows a more accurate distribution of costs, according to the Service. Tr. 38/17309.
[3009] Witness Degen notes that the Commission rejected this migration in R97-1 on the grounds that it would not be consistent with the Commission's method for determining the variability of Segment 3 costs. Witness Degen argues that such consistency is not necessary, noting that in R97-1 the Commission disaggregated mail processing costs into operation-specific MODS pools, but did not apply the variabilities proposed by the Postal Service to those pools. Postal Service Brief at V-64-V-65.
[3010] Witness Degen asserts that the tallies where the IOCS data collector observed an employee working at the retail window while clocked into a mail processing activity are not likely to reflect the kind of retail sales activity that makes up the majority of window service activity. He argues that temporary hand off of the control of cash drawers and stamp inventories is very unlikely given the time consuming audit procedures that are supposed to be followed each time. Witness Degen states that the employees clocked into a mail processing activity but observed performing window service functions are likely to be retrieving held mail or gathering collection mail from the window. He notes that many of the migrated tallies are not-handling tallies, and argues that they should be tied to the operation into which the employees are clocked. He comments that if employees are temporarily in the window service area when observed and not handling sales items, then it is inappropriate to use a window service distribution key that is based, in part, on sale activities. Tr. 38/17310.
[3011] On behalf of Periodical mailers, witness Stralberg testifies that during his visits to postal facilities as part of the advisory task force on Periodicals costs, managers consistently admit that they use mail processing clerks to fill in for window service or administrative clerks, for example, during lunch breaks, and that they do not always reclock when temporarily performing these non-mail processing activites. Tr. 24/11389. His main concern is that this phenomenon causes window service costs to be inappropriately assigned to Periodicals and other mail classes that generally do not use window service. Ibid.
[3012] The Commission concludes that the MODS operation code that a worker is clocked into is generally less reliable than the IOCS information on the activity that the worker was performing at the time he was actually observed. Witness Stralberg's explanation that workers do not always re-clock when temporarily switching activities is plausible. Witness Stralberg reports that supervisors told him that it is a common management practice to have mail processing clerks temporarily fill in at window service and administration activities without re-clocking. Even witness Degen agrees that mail processing clerks may switch assignments temporarily without reclocking, although he emphasizes that the frequency with which it occurs is unknown. Tr. 38/17310. The Postal Service Inspection Service, however, reports numerous instances of misclocking within the MODS system, as witness Stralberg describes. LR-H-236 AT 18-19.
[3013] While witness Stralberg's testimony is anecdotal, the Inspection Service report tends to corroborate it. Witness Degen's assertion that strict auditing requirements deter temporary assignment of clerks to retail window service activities is speculative. The extent of this deterring effect, like the extent of misclocking, is unknown. Even if this deterring effect were absolute, however, it may not be particularly relevant. If mail processing clerks are retrieving held mail, or performing other work that would otherwise be done by regular window clerks, it is still window service activities they are performing. Likewise the window clerks involved in retail sales are also involved in activities that are part of the window service cost component. It is not relevant that at any moment one window service clerk is performing retails sales and another is performing other window service activities. Both are performing window service cost component activities. Therefore it is appropriate to treat their temporary assistance as window service activity, regardless of the precise form that their assistance takes. The fact that the IOCS tally taker actually observed the employee performing a window service or administrative task, in the Commission's view, is more meaningful than the MODS activity that the employee is clocked into.
[3014] Over 75 percent, or $406.1 million, of the migrated Administration costs come from the single IOCS activity code 6630, which is designated as general administrative services. Given the large proportion of general administrative services for which the Postal Service proposes to override IOCS information, the Service should provide some empirical evidence confirming that the portions it proposes to migrate into mail processing solely support Function 1 or 4 mail processing operations.
[3015] Until then, the direct evidence in the IOCS codes should continue to take precedence over the MODS codes in these cost components. In his calculation of Segment 3 costs, UPS witness Sellick uses IOCS information to reverse the migration of Window Service and Administration costs into the Mail Processing component. The Commission will use this method to reverse the Postal Service proposal.
[3016] In Docket No. R97-1, for the first time, the Postal Service disaggregated mail processing costs in MODS offices into 39 operation-specific pools. Total payroll costs were apportioned to these MODS cost pools using the MODS codes associated with each pool. The IOCS tallies from MODS offices were apportioned to the same pools based on the MODS record in the IOCS tally. IOCS tallies were used to develop distribution keys for each pool. The Commission adopted the disaggregation of mail processing costs in MODS offices into operation-specific pools, and reaffirms this recommendation in this docket. In this docket, the Postal Service retains the R97-1 division of BMCs costs into 6 pools based on IOCS data. For the first time, it proposes to disaggregate mail processing costs in non-MODS offices into 8 pools based on IOCS defined activities.
[3017] No participant opposes the proposed disaggregation of mail processing costs in non-MODS offices. UPS witness Sellick incorporates this disaggregation in his calculation of Segment 3 costs. The Commission believes that it will allow the costs in non-MODS offices to be attributed to subclasses more accurately. It accepts disaggregating the mail processing costs in non-MODS offices for the same reasons that it accepted operation-specific MODS and BMCs pools in Docket No. R97-1.
[3018] Reversing the trend to disaggregate costs into operation-specific pools, the Postal Service proposes to consolidate four current MODS pools into two. It proposes that the cost pools currently labeled "1MISC" and "1SUPPORT" be combined into a single pool labeled "F1 Support." It proposes that the cost pools currently labeled "LD48OTH" and "LD48_ADM" be combined into a single pool labeled "F4 Support." Witness Degen argues that this reorganization is warranted because the F1 Support activities only provide support services to mail processing operations in Function 1 facilities, and the "F4 Support" activities only provide support services to mail processing operations in Function 4 facilities (branches and stations). Consolidating these pools would enable the Postal Service to distribute costs in the new pools using cost drivers from the supported F1 and F4 cost pools, respectively. The Periodical mailers object to this proposed method of distributing the costs in these consolidated pools. Tr. 24/11388.
[3019] The Postal Service has concluded, but not demonstrated, that costs in the consolidated pools are caused only by the Function 1 and 4 activities they are claimed to support. The Commission needs some assurance that this conclusion is valid before it can recommend overriding the direct tally information in these pools and distributing their costs based on tallies found in other pools. Until it receives this assurance, it will treat these cost pools as separate.
[3020] For more than two decades, the Postal Service and the Commission have accepted the operational judgment of postal experts that mail processing labor costs rise essentially in proportion to the volume of mail processed. This implies that the volume variabilities of mail processing labor costs are approximately 100 percent. In this proceeding, the Postal Service uses an econometric model to obtain an estimate that the average volume variability of mail processing labor costs is less than 73 percent. The Commission concludes that the weight of the evidence in this proceeding supports the established finding that the volume variability of most mail processing operations is approximately 100 percent.
[3021] The record on mail processing variability is exceedingly complex. For that reason, this section of the Commission's Opinion briefly summarizes the debate, presents the Commission's conclusions in "bullet" form, and summarizes the Commissions reasoning. Detailed findings and conclusions are presented in Appendix F. The issues fall into three main areas-operational analysis, data issues, and econometric modeling issues. The evidence concerning the nature of the Postal Service's mail processing operations is analyzed in detail in Parts A and C of the Appendix. Data issues and econometric modeling issues are analyzed in detail in Part B of the Appendix.
[3022] Operational evidence. The validity of both the Commission's and the Postal Service's estimates of the variability of mail processing labor costs depends, first, on whether they are based on a realistic view of the way that mail processing operations are actually managed.
[3023] The Postal Service's economic consultants base their econometric model on a novel view of how the Postal Service copes with volume changes over a typical rate cycle. They assert that the Postal Service designs large cushions of excess labor into many of its processing operations in order to meet service commitments. As a result, they argue, much of the increase in volume that occurs over a rate cycle can be absorbed without increasing workhours. This, they say, largely explains the low variabilities obtained from their model, and the large economies of scale that they imply. These economic consultants also assert that the typical rate cycle is too short to allow the Service to change anything but the amount of labor it hires to cope with changes in volume. Accordingly, their model is structured to capture primarily short-run volume effects.
[3024] The view consistently expressed by the Postal Service's managers supports the established finding that mail processing labor costs change in proportion to changes in volume. They assert that when mail processing operations are running, they are typically operating at full capacity, using standardized equipment and methods. They testify that managers carefully and continually match staffing to expected workload, rather than design cushions of excess capacity into particular mail processing operations. They agree that the Postal Service copes with volume changes that are sustained over a typical rate cycle by making adjustments to plant, equipment, and mail flows, as well as their supply of labor. These characteristics of mail processing operations imply that costs will vary in proportion to volume.
[3025] Data issues. In Docket No. R97-1, and again in this docket, the suitability of Management Operating Data System (MODS) data for estimating volume variability has raised considerable controversy. MODS records the mail processing operation a worker is clocked into and associates those hours with costs taken from payroll records. Since the number of unique mail pieces processed in an operation is difficult to identify, MODS records initial and total piece handlings that occur in particular operations. In manual operations, this requires the Postal Service to weigh mail in order to infer the number pieces handled, contributing substantial imprecision to the data. Postal Inspection Service audits conclude that errors in clocking data and in handlings data are common.
[3026] Although obvious errors are common in MODS data, the Postal Service argues that it effectively screens these errors before using them to estimate variability. It argues that its statistical results pass standard diagnostic tests, implying that the data is reasonably error free.
[3027] UPS and the OCA have an opposing view. They argue that because obvious errors in the MODS data are common, but their source is unknown, one must assume that errors that are not obvious, and therefore are not detected, are also common. This, they argue, results in a substantial risk that the Postal Service's econometric estimates exhibit "errors-invariables" bias, which reduces its variability estimates.
[3028] The Commission concludes that a substantial risk of "errors-in-variables" bias remains after witness Bozzo's data screens. As long as this risk remains substantial, the Postal Service's econometric estimates of variability cannot be regarded as reliable.
[3029] Modeling issues. The Postal Service argues that witness Bozzo has corrected the flaws identified by the Commission in the Postal Service's model presented in Docket No. R97-1. UPS and the OCA argue that the most serious flaws remain. They argue that the model still reflects transient effects, such as temporary variations in the intensity of workeffort in manual operations, rather than reflecting cost effects that are sustained over a rate cycle. They also argue that many of the variables that witness Bozzo's model treats as independent of volume are, in fact, affected by volume, rendering its results invalid. The OCA argues, in particular, that witness Bozzo's model erroneously assumes that capital and equipment are not affected by changes in volume over a rate cycle, and that the model is therefore mis-specified.
[3030] The Commission concludes that witness Bozzo's model, like the R97-1 model, reflects short-run, transient effects, and improperly assumes that capital and equipment are not affected by changes in volume that occur over a rate cycle. Evidence is compelling that the Bozzo model is mis-specified, because it implies that capital is reducing, rather than increasing, the productivity of mail processing labor.
[3031] In Docket No. R97-1, the Postal Service presented for the first time an econometric model depicting the relationship of work hours to piece handlings in specific processing operations. This model was fit to data mostly reported by the Service's Management Operating Data System (MODS) by USPS witness Bradley. Based on this model, the Postal Service estimated that the overall volume variability of mail processing labor was only 81 percent. In this docket, the Postal Service has presented similar models fit by witness Bozzo to data from the Postal Service's MODS system. These new models estimate that only 72.8 percent of the costs of mail processing labor are volume-variable. Mail processing labor costs constitute a very large part of the Postal Service's total costs. Replacing the Commission's 100 percent variability with the Service's econometric estimates would reduce the costs attributed to subclasses and increase the costs the Commission regards as institutional. This transfer of about $1.8 billion would be large enough to have a significant impact on the Commission's recommended rates.
[3032] Many parties have contributed testimony that the Commission has considered in its evaluation of the Postal Service's proposal. This proposal is presented or defended by Postal Service economics witnesses Bozzo, Degen, and Greene. The Commission has also considered the testimony of Postal Service operations and management witnesses Kingsley, O'Tormey, and Unger. The proposal is opposed by OCA witness Smith, UPS witness Neels, and MMA witness Bentley. It is supported by witnesses Elliot, Stralberg, and Cohen, representing the Periodical mailers' group.
[3033] Based on the record, the Commission reaches the following findings and conclusions regarding the volume variability of mail processing labor:
· The long-standing conclusion that the cost of mail processing labor varies in proportion to the volume of mail processed flows from the basic analytical insight that each piece or container of mail requires individual handling at each workcenter, and such handlings are replicated as volumes rise. This basic insight applies to automated and mechanized as well as manual processing activities, since throughout the network, processing equipment is typically operated at or near capacity, according to standard operating procedures.
· The Postal Service presents counter theories to explain its estimates of low variabilities in manual sorting operations, allied operations, and some automated operations. It hypothesizes that large proportions of excess labor capacity are designed into these operations, allowing them to absorb increased volumes without increasing workhours. These hypotheses are contradicted by postal managers, who testify that they meticulously match staff to workload, and use overtime to process unexpected surges in volume.
· Startup/shutdown costs are insignificant in manual and most machine-based operations. They might be significant in FSM, SPBS, and parcel sorting. The number of startup/shutdown cycles in these operations, however, is likely to vary partially with volume. For example, rising volumes can cause sort schemes to shift from manual to machine-based processing. They can also increase the number of machine-processed schemes that are run in parallel. Some attempt to quantify the amount of fixed setup/shutdown time should be provided in future proceedings.
· Witness Bozzo's low labor demand variabilities imply labor productivities (percent change in output for a percent change in labor) that are well above 100 percent for most modeled operations. They also imply that there are very large economies of scale in mail processing. Such labor productivities and scale economies are implausibly high. If true, they imply that a major reorganization of the mail processing network is needed to exploit these remarkable labor productivities and scale economies.
· Adopting witness Bozzo's low labor variability estimates would result in greatly reduced estimates of savings from mailer worksharing. This would require dramatically smaller worksharing discounts, and undermine the viablility of the Postal Service's worksharing program. This provides an additional reason that the Commission would need to be assured that witness Bozzo's results are reliable, before it accepts them.
· Witness Bozzo's models are based on MODS data in which obvious errors are common, but whose source cannot be identified. This strongly implies that errors that are not obvious, and therefore are not detected, are also common. This presents a substantial risk that data errors have biased witness Bozzo's results.
· Witness Bozzo's model corrects some of the flaws in the Docket No. R97-1 model, but the most serious flaws remain. The Bozzo model includes key variables (the manual ratio and capital index) that are likely to be codetermined with work hours. This implies that in order to avoid a simultaneous equations bias in the estimate of variability, a method of estimation different from that applied by witness Bozzo must be used.
· Volume-induced changes in the Postal Service's plant, equipment, and mail flows occur over the typical rate cycle. Their contribution to the demand for mail processing labor has been excluded from the Postal Service's econometric model. As a result, the model narrows its focus to a subset of volume effects, yielding the wrong estimate of variability.
· Witness Bozzo's estimated capital productivities are negative, which is nonsensical. Negative productivities imply that capital and workhours could both be reduced without affecting the amount of mail that is processed. A model that so poorly estimates the effect of capital on labor costs is unlikely to have correctly measured the effect of volume on labor costs
· Witness Bozzo's models assume that the number of piece handlings performed in an operation is proportional to the volume of mail processed in that operation. The record contains econometric evidence that the number of handlings performed in various mail processing operations increases faster than the volume of mail processed. There are also operational reasons for expecting that the number of handlings increases faster than volume. The depth of sort attempted in a machine-based operation, and therefore the number of re-handlings performed in that operation, depends on the volume of mail that is destined to a given area. Witness Bozzo's variabilities are not reliable as long as the proportionality of piece handlings to volume is in doubt.
· The large number of econometric models presented on this record and their widely varying results are illustrated in Table 3-1. They range from an average variability of 66 percent for witness Elliott's Model A, to an average variability of 139 percent for witness Neels' shape-level adjusted model. The Commission does not consider any of these econometric model results to be reliable. In the Commission's view they all run a substantial risk of bias from the suspect piece handling data on which they are based.
[3034] Basic economic and econometric theory, coupled with an operational understanding of how the Postal Service processes mail, provides the basis for the Commission's evaluation of witness Bozzo's estimated variabilities. Witness Bozzo describes his equations as short-run derived demand functions. Economic theory prescribes the general properties of the price elasticities, productivities, and returns to scale exhibited by such functions for a firm that is managed efficiently. Econometric theory provides a basis for evaluating the reliability of witness Bozzo's estimates. This is not just a matter of statistical accuracy. If witness Bozzo's estimation methodology is not appropriate for his model or sample, econometric theory indicates that the result will be biased and unreliable estimates of the structural parameters of his models. Witness Bozzo's variability estimates are so far below the proportional relationship that is expected that they are either reflections of a biased model, or the Postal Service operations exhibit an implausible degree of waste.
[3035] Record descriptions of the technology and methods that the Postal Service applies in its mail processing plants supports the finding, long accepted by both the Postal Service and the Commission, that mail processing labor costs vary about in proportion to the volumes of mail processed. This expectation is based upon several fundamental observations about mail processing operations.
[3036] First, the operations are mostly piece-by-piece sorting, cancellation, preparation, and allied processes that run at nearly uniform average output rates per workhour. This seems to be true whether the processes are mechanized or automated. It should be true of manual processing as well, averaged over the period of a typical rate cycle. Second, there is little labor time that can be identified as downtime or slack time, in the sense that the activities are fully staffed but processes are not actually running at full capacity. Mail processing labor is mobile within a plant, so an activity would not normally be staffed when mail is not actually being processed. Third, the activities ought to exhibit nearly constant returns to scale in the long run. Over this time, it should be possible to deal with an increase in volume by increasing proportionately all of the personnel, floor space, machines and other equipment located at a plant or by just replicating processing plants. Fourth, short run variabilities eventually should be higher, not lower, than 100 percent because of the law of diminishing returns. If volumes are increased with some capital inputs held fixed, labor productivities should fall, and labor variabilities rise, as diminishing returns are encountered. Finally, proportionate changes in both labor and capital in response to volume changes are feasible within a 3-to-4-year rate cycle. The capital found at the Service's mail processing plants is fixed only over a run that is shorter than a typical rate cycle.
[3037] The Postal Service's operational witnesses. In the current docket, Postal Service witnesses Kingsley, O'Tormey, and Unger have provided the Commission with fairly complete, generic descriptions of the actual planning, organization and operation of the Service's mail processing plants. These witnesses are Postal Service management personnel who appear not to have focused on the implications of their testimony for attributing mail processing costs. They generally confirm the observations that underlie the long-held expectation that mail processing labor is approximately proportional to volume. For certain processes, they confirm these observations in some detail.
[3038] Postal Service consultant witness Degen has also provided the Commission with descriptions of the mail processing system. Witness Degen's descriptions conflict in several crucial ways with the descriptions provided by the Service's own management personnel. First, witness Degen claims that the configuration and operation of a processing plant is determined almost entirely by the plant's location within the Service's network rather than by the volume of mail that it processes. The testimony of witnesses Kingsley, O'Tormey, and Unger confirms that the Postal Service's processing network is configured as it is as a means of responding to volume in an optimum way. In doing so, it employs standard equipment, standard criteria for staffing, and standard guidelines for organizing mail flows through all of its plants.
[3039] Second, witness Degen regards the capital equipment at processing plants as predetermined and effectively fixed over a typical rate cycle. The testimony of the Postal Service's operations witnesses, however, describes space expansions, equipment installations, replacements, and removals taking place within periods of time that are generally shorter than a rate cycle. Finally, witness Degen argues that the low variabilities that witness Bozzo estimates for specific processing operations are reasonable. He does this by identifying specific kinds of downtime, waiting time, or slack time that he believes exist in various processing operations that cause the operation not to be operated at full capacity a sufficient portion of the time to account for witness Bozzo's low variabilities.
[3040] Postal Service operations witnesses testify that the Postal Service enforces productivity standards throughout its mail processing system with regard to equipment and staffing that essentially rule out under-used equipment or under-employed workers. Plant managers also use their equipment efficiently by not conducting processing runs that are too short to justify the necessary startup and teardown times. One would not expect to see downtime or slack time accounting for over 27 percent of labor time, as witness Degen's explanations of witness Bozzo's variabilities indicate.
[3041] The operational theories of the Postal Service's economic witnesses. The non-volume-variable time hypothesized by witness Degen consists of setup and teardown time for processes that use mechanized or automated equipment, workers assigned to create excess capacity at gateway and backup operations, workers engaged in mail movement and sweeping activities at the end of runs, and manual processing conducted below the optimal sustainable pace ("discretionary effort"). For example, the volume variabilities of 75.1 percent for the Optical Character Reader (OCR) operation, and 64.1 percent for the Small Parcel Bundle Sorter (SPBS) operation imply that the workers engaged in these activities are sorting at less than capacity at least 24.9 percent, and 35.9 percent of the time that the operations are staffed.
[3042] The pattern found in witness Bozzo's variability estimates of specific mail processing operations is also difficult to explain. Witness Bozzo finds that workhours in manual operations, where mail is necessarily handled by workers one piece at a time, are less volume-variable than man hours in automated operations, where it is the machines that do the piece-by-piece work. However, it is the automated operations, if any, which ought to exhibit the lower variabilities. To explain this result, witness Degen hypothesizes implausibly large proportions of excess labor capacity in manual operations.
[3043] Witness Degen's analysis accords with the Postal Service's new mail processing variability estimates but conflicts with the testimony of the Service's own operations witnesses. The testimony of the operations witnesses fails to confirm that the percentages of downtime or slack time in mail processing are large enough to explain witness Bozzo's variabilities. This conflict is clearest with respect to manual operations where witness Degen hypothesizes especially large amounts of slack time in order to explain especially low volume variabilities.
[3044] Witness Degen's analysis depends upon several assumptions that are unfounded. These are, first, that the floor space, equipment and other capital found at mail processing plants are all fixed for the duration of a rate cycle. Apart from building or rebuilding complete plants, the record indicates that space and equipment at mail processing plants are variable within a rate cycle. Second, witness Degen assumes that the elements of downtime or slack time are truly fixed rather than variable with volume. This assumption has been challenged by non-postal witnesses who have pointed out several serious defects in witness Degen's interpretations. The record indicates that most of the downtime or slack time hypothesized by witness Degen will actually vary nearly proportionately with volume over a typical rate cycle. And, third, witness Degen simply assumes that the downtime or slacktime that he identifies is large enough to explain a pattern of variabilities as low, particularly for manual operations, as those proposed by witness Bozzo in this proceeding. There is little empirical evidence in witness Degen's testimony or in the testimony of any other Postal Service witnesses that corroborates his assertions of large blocks of underused processing time.
[3045] Conflicts between the Postal Service's econometric estimates and economic reality. Perhaps the least credible feature of witness Bozzo's estimates are the productivities that they imply for labor and capital inputs at processing plants. Productivity is the marginal rate at which an input contributes to output, with all other inputs held constant. Productivities are the reciprocal of witness Bozzo's variabilities. In the case of witness Bozzo's estimates, these would be the elasticity of piece-handlings with respect to work hours and with respect to his index of capital. All of witness Bozzo's piece handling elasticities with respect to workhours are greater than one, and all-but-one of his piece handling elasticities with respect to capital turn out to be negative. The overall picture that these elasticities present is that the Service's mail processing plants are so badly under-staffed and over-capitalized that they are actually wasteful. For example, the elasticities that derive from witness Bozzo's estimates for Flats Sorting Machines (FSM) are 1.224 (work hours) and -0.061 (capital). According to these elasticities labor and capital can simultaneously be reduced without affecting FSM piece-handlings at the Service's mail processing plants. As it happens the Postal Service is engaged in adding 175 new AFSM-100s to the 812 FSM 881s and 340 FSM 1000s now in service. Witness O'Tormey stated "we have told the field our expectations are you are going to lose 23 employees per machine." Tr. 21/8374. According to witness Bozzo's estimates, processing plants would have to add employees just to process the same number of flats after receiving the additional machines.
[3046] The picture of mail processing operations painted by operations witnesses and the picture painted by witness Bozzo's estimates are different. One can reasonably infer constant returns to scale and volume-variabilities of around 100 percent from the testimony of the operations witnesses. On the other hand, the volume variabilities proposed by witnesses Bozzo are around 72.8 percent. Witness Bozzo's variabilities fall so far below 100 percent that they imply that very large economies of scale in mail processing operations remain unexploited. It is difficult to reconcile unexploited economies of scale with several basic facts. One is that larger mail processing facilities typically require more hours to process a given amount of mail than smaller facilities. If there are large economies of scale, exactly the reverse should be true. The most puzzling fact of all is the obvious lack of concentration we can see in the Postal Service's network of mail processing plants. The Service's processing network is decentralized, consisting of over 375 plants, and is growing at the rate of about one new plant per year. The large economies of scale implied by witness Bozzo's variabilities should mean that the Postal Service has a strong economic incentive to consolidate its mail processing in a much smaller number of efficient large plants. Yet there is not the slightest indication in the testimony of witnesses Kingsley, O'Tormey and Unger that the Service believes it can reduce costs by consolidating plants.
[3047] The Postal Service's econometric model suffers from several serious technical flaws, any one of which could render its results anomalous and unreliable. First and foremost, there is a basic problem with the raw data with which the Postal Service has built its models of mail processing labor variability. There is no effective way to determine whether the accuracy of the MODS measurements for piece handlings and work hours is good, fair, or poor. A high frequency of apparently anomalous observations provides strong circumstantial evidence that errors pervade the piece-handlings data. If they do, and if the screens used by witness Bozzo do not succeed in deleting most of the erroneous data, then econometric estimates based on these data will contain an "errors-in-variables" bias that could be large.
[3048] In Docket No. R97-1, witness Bradley assumed that all of the control variables found in his equations would control only for non-volume effects on work hours. Among witness Bradley's controls was the ratio of manual piece handlings to piece-handlings in all manual, mechanized and automated operations (the Manual Ratio). The record provided grounds for concluding that these variables would be indirectly affected by volume changes over a rate cycle. In this docket, witness Bozzo likewise assumes that the controls only remove non-volume effects from work hours. Witness Bozzo has added a network proxy (possible deliveries) and a plant-level index of capital to the list of controls which are assumed to be unaffected by volume changes over a rate cycle. The same reasons for concluding in Docket No. R97-1 that the Manual Ratio is volume-variable applies in this docket as well. In addition, the testimony of the Postal Service's own operations witnesses provide ample justification for concluding that witness Bozzo's new capital variable also is likely to be affected by changes in volume over the rate cycle. Elsewhere in the Postal Service's Base Year and Test Year calculations, capital is regarded as volume variable.
[3049] Witness Bozzo's index of capital provides particularly compelling grounds for concluding that his model is technically flawed. All but one of the capital elasticities witness Bozzo estimates for mail processing work hours have impossible positive signs; most are also statistically significant. If the Postal Service's investment behavior is rational, it will not have acquired so much capital that capital productivities become negative.
[3050] Industrial capital is difficult to measure and witness Bozzo's index may not be a good index even at the plant level. His index is clearly an imperfect proxy for the capital in use in specific operations. Even if his capital index were a good variable, serious problems would remain. Capital is codetermined with labor rather than predetermined, as assumed by witness Bozzo. In this respect, witness Bozzo's treatment of capital is excessively short run. This mistake adds a well-known simultaneous equations bias to the list of things that might have gone wrong. Any of these flaws, and others, could account for the negative capital productivities implied by his estimates.
[3051] The Commission concludes that the mail processing system that witness Bozzo has modeled the differs substantially from the one observed in the real world. The weight of the evidence in this proceeding continues to support the long-established finding that mail processing labor variabilities are approximately 100 percent.
[3052] In the preceding section on the volume variability of mail processing labor costs, the Commission explained its reasons for concluding that these costs are likely to rise in proportion to the volume of mail sorted to its destination. To directly estimate the portion of variable mail processing labor costs that each subclass causes, it would be necessary to associate the volume of each subclass with the labor hours required to sort that volume to its destination. The Postal Service does not know the volume of each subclass that is processed in particular operations. Instead, its In-Office Cost System (IOCS) "tallies" record subclass information. Tallies are randomly sampled instants in time in which a data collector observes a worker and records the processing operation that he is engaged in, and, if discernible, the subclass of mail that is being processed. Each tally is given a dollar value so that the sum of dollars equals the clerk and mailhandler labor costs for each component of the sample frame.
[3053] To yield an economically meaningful estimate of the marginal costs of mail processing by subclass, IOCS tallies are assumed to be the equivalent of the workload-weighted subclass distribution of pieces in a given operation. It is necessary to further assume that IOCS tallies are proportional to subclass volume. See PRC Op. R97-1, paras. 3153-3155.
[3054] In Docket No. R97-1, the Postal Service and UPS proposed that mail processing activities in MODS offices be separated into over 40 pools that are thought to exhibit a homogeneous response to changes in the volume of mail processed. The sum of the tally dollars in each MODS pool was adjusted to be consistent with the payroll costs that were recorded in that pool. Id., at paras. 3067-3071. Non-MODS offices were treated as a single cost pool, and BMC costs were disaggregated into six pools, according to the information contained in IOCS tallies.
[3055] In Docket No. R97-1, an IOCS tally was assigned to a MODS pool according to a three-digit MODS operation code. Where the MODS operation code indicating the activity a worker was clocked into conflicted with the activity observed by the IOCS tally taker, the Postal Service proposed that the MODS code override the IOCS tally. In most instances, the Commission accepted participants' arguments that the information provided by the IOCS tally was more reliable than the MODS clocking system, and distributed the costs of the tally accordingly.
[3056] An IOCS tally can be associated with a subclass directly or by inference. A worker might be observed with a piece of mail of an identified subclass in his hand, pushing a container holding only one subclass, or handling a tray in which the subclass of the top piece is identified and assumed to be representative of the entire contents. This would be considered "direct" information associating the tally with the subclass. If a worker were observed handling a container of mail whose subclass content could not be determined, this would be a "mixed mail" tally. If a worker was observed not handling any mail (on break, clocking out, etc.), this would be categorized as a "not handling" tally.
[3057] In FY 1998, direct tally costs accounted for 41.8 percent, mixed mail tallies for 12.2 percent, and not handling tallies for 46.1 percent of total mail processing tally costs in MODS pools. Tr. 24/11371. In Docket No. R97-1, the Postal Service proposed that the subclass distribution of mixed mail tallies be assumed to match the subclass distribution of direct tallies within a MODS pool. It assumed that "not handling" tallies matched the subclass distribution of direct and mixed tallies combined, within a pool.
[3058] The Commission distributed tally dollars according to these assumptions of subclass responsibility for operations in which the presence of pieces of a subclass could be assumed to be the cause of the labor costs in that pool. Generally, these consist of sorting operations. For "allied" operations, however, the Commission concluded that the fact that a percentage of directly identified pieces belonged to a particular subclass did not provide a reliable inference that the subclass was responsible for mixed mail and not handling tallies in the same proportions. One reason was that some allied functions are internal to each allied pool, while others primarily serve other pools. The Commission decided that distributing mixed tally dollars in allied pools in proportion to the IOCS direct tallies across all pools would better reflect the diverse sources of workload that drive labor costs in allied pools.
[3059] Another feature of the Postal Service's proposed distribution scheme in Docket No. R97-1 was its stratification of MODS pool tally dollars by the type of "item" and "container" that is associated with the tally. The rationale was that particular mail shapes and subclasses are more likely to be found in particular types of items or containers, so that it is more accurate to distribute unknown item tallies to subclasses in the same proportions as known item tallies of the same type are distributed, and to distribute unknown container tallies in the same proportions as known container tallies of the same type are distributed. PRC Op. R97-1, para. 3165.
[3060] In Docket No. R97-1, the Commission concluded that distributing the cost of mixed mail and not handling tallies in proportion to the direct tallies within a given pool carried with it a greater risk of selection and assumption bias when this procedure is applied to allied pools than when it is applied to distribution pools. Allied pools process all shapes and types of mail, and allied pools are where most mail is handled in bulk and its subclass identity is more likely to be obscured. The Commission recognized that subclasses of mail that are typically presorted might be over-identified in allied operations because they are typically presented in identical containers. The Commission also recognized that presorted mail might be under-identified to the extent that it is bypass mail that appears in mixed items or containers. Id., paras. 3170-3173. It concluded that there is a greater risk that there is such "assumption" bias in allied operations, despite stratification. The Commission concluded that a distribution of unknown allied tallies on direct tallies across all pools would help reduce this risk.
[3061] An added consideration persuading the Commission in Docket No. R97-1 not to distribute allied mixed mail and not handling tallies entirely on direct tallies within each allied pool is the fact that direct tallies are a relatively small percentage of total tallies in those pools. When direct tallies are few, the risk that they do not accurately indicate subclass responsibility for mixed mail and not handling costs is magnified. Id., paras. 3145-3146.
[3062] In this docket, the Postal Service states that, apart from container handling on the platform, it would prefer that mixed mail tallies in an allied operation be distributed to subclasses in the same proportions as the direct tallies in that operation. It says, however, that it would accept a distribution of allied mixed mail tallies over the direct tallies in all Function 1 pools if it were done within item and container strata. The Periodical mailers favor a broad distribution of allied mixed mail tallies, but still express concern that item and container strata in allied pools may increase rather than decrease potential bias. As in Docket No. R97-1, they prefer to use shape inferences from the IOCS rather than rely on item and container strata. Tr. 24/11377-78.
[3063] Postal Service witness Degen purports to have demonstrated that distributing mixed mail tally dollars within item and container type does not present a risk of bias that is large enough to warrant further investigation. He addresses the Periodical mailers' claim that their mail is over-represented in direct tallies in allied operations because it is more likely to appear in identical items or containers, making it easier to identify. He also addresses their claim that Periodicals are more likely to be counted when they appear in mixed items or containers, because they are more likely to have only a few, large pieces per item or container, and therefore are easier for an overworked IOCS tally taker to count. Witness Degen's response, in effect, is that it does not matter how large a bias against Periodicals is imputed by assumption from direct tallies to mixed, non-empty item tallies, since mixed, nonempty item costs have so little effect on the cost of Periodicals. USPS-T-16 at 59-61. He similarly argues that it hardly matters how biased the distribution of non-identified mixed container tally costs to Periodicals might be, since those costs have such a small effect on Periodicals. Id. at 63-64.
[3064] It might be true that the costs represented by mixed, non-empty items and non-identified container tallies have only a small impact on the total mail processing costs of Periodicals regardless of how they are distributed. This, however, is somewhat beside the point. The Periodical mailers' R97-1 arguments were that the direct tallies themselves present a substantial risk of bias against workshared mail, and if this bias is extended to mixed mail and not handling talllies, it magnifies whatever bias is in the direct tallies. PRC Op. R97-1, paras. 3167-3168. Witness Degen has offered nothing new to refute their argument that this risk of bias in the direct tallies is substantial. If it is substantial, then all of the various mixed mail strata that are distributed on the basis of direct tallies involve a substantial risk of bias as well. Witness Degen addressed only a few of those strata, and only the effect of those few on Periodicals. He did not show that bias in other strata would have only a small effect on Periodicals, or on other subclasses. In Docket No. R97-1, the Commission attempted to limit this risk by using direct tallies from all MODS pools to distribute allied mixed mail costs. It will continue to do so in this docket, for the same reason.
[3065] Distributing allied mixed mail costs on direct tallies from all MODS pools, as the Commission did in Docket No. R97-1, was a compromise. This compromise recognizes that the causal link between direct tallies and processing costs in allied pools is ambiguous. It also recognizes that there is a greater risk of selection and assumption bias in the direct tallies in allied pools than in other pools. In order not to extend whatever bias exists in those direct tallies to allied mixed mail distribution, the Commission recommends using direct tallies from all pools to distribute allied mixed mail tally dollars to subclasses.
[3066] The Commission's Docket No. R97-1 distribution was a compromise in another sense. It did not distribute allied mixed mail costs within item and container strata, but used shape information provided by the IOCS instead. Using IOCS shape information rather than item and container shape associations was proposed by Periodicals mailers in Docket No. R97-1, and again in this docket. Tr. 24/11377-78. The Commission uses shape information from the IOCS again in this docket because item and container shape and subclass associations are less reliable in allied operations than in non-allied operations, and carry a risk of bias that is not limited to the non-empty item and unidentified container bias that witness Degen asserts are de minimus.
[3067] Other areas of potential bias in item and container stratification remain a concern despite witness Degen's arguments to the contrary. For example, witness Degen cites a 1995 platform study by Christensen Associates that sampled the subclass distribution of items in containers at eight processing plants to see how closely they matched the subclass distribution of direct items not in containers. The results are presented in Table 8 at page 66 of USPS-T-16. Witness Degen asserts that they show no evidence of bias. The sample results, however, are substantially different from the IOCS results for a number of subclasses. The subclass share of Priority and Express mail combined was 2.6 percent of the total in the IOCS compared with 11.4 percent of the total in the sample. The subclass share of Standard A was 32.7 percent in the IOCS compared with 25.3 percent in the sample. The share of First-Class Mail was 50.6 percent in the IOCS, but 45.7 percent in the sample. For what it is worth, the Christensen study suggests that distributing the cost of items in containers on the tallies for direct items outside of containers rather than relying on IOCS information substantially benefits Priority and Express Mail and significantly penalizes First-Class and Standard A mail. The Christensen study does little to reduce concerns that such stratification might be biased.
[3068] Witness Degen concedes that the use of full containers within a cost pool may not accurately indicate how empty containers are used in that pool. To address this concern, he demonstrates that it makes little difference whether the costs of empty containers are distributed using tallies for full containers from within pools or across pools, except to Special Standard Mail. Id. at 68. He also purports to show that distributing the cost of empty items on non-empty item tallies yields a distribution of Periodical costs among items that appears "reasonable" to him. Id. at 62. This falls well short of a demonstration that the inferred subclass distribution of empty items is accurate.
[3069] Taken together, witness Degen's arguments confirm the risk of bias as much as they dispel it. They suggest that if each item and container combination is viewed in isolation, it doesn't matter how its costs are distributed. They do not provide sufficient grounds for changing the Commission's approach to distributing allied mixed mail costs.
[3070] The participants agree that they would prefer to see allied not handling costs distributed to subclasses on the basis of direct tallies in all Function 1 MODS pools. They do not agree on the reasons. The Postal Service proposes that allied not handling costs be distributed on direct tallies across pools to compensate for its proposal to treat allied mail processing labor costs as 100 percent volume variable. Witness Degen explains that he believes the variability of allied labor is less than the variability of labor in distribution operations, but that the Postal Service did not develop econometric models of allied operations that it is willing to sponsor. It characterizes this as a "compromise" that yields results similar to those that it would have proposed had it sponsored models of allied cost pool variability. Id., at 69.
[3071] The Periodicals mailers support a broad distribution of allied not handling costs on the theory that not handling costs reflect excess staffing of allied operations, the need for which is caused by preferential mail. They argue that distributing these costs on the basis of direct tallies across all Function 1 pools helps to distribute these costs back to the preferential classes that cause them. Tr. 24/11373-76.
[3072] In Docket No. R97-1, the Commission distributed the costs of allied not handling costs using a combination of direct tallies within each allied pool and the broad distribution of allied mixed mail tallies that is based on direct tallies across all pools. Not handling costs consist largely of break time and the time it takes to clock in and out of operations during a shift. It was the Commission's view in Docket No. R97-1, and it remains the Commission's view, that such costs are more causally related to the number of employees in a given allied operation, and the time that those employees spend in that operation, than they are to mail processing operations in general. Accordingly, the Commission continues to distribute allied not handling costs on the combination of within-pool direct tallies and across-pool direct tallies used to distribute the cost of mixed mail. This partially spreads allied not handling costs beyond the allied pool where they were incurred, but it does not completely sever the relationship between allied not handling costs and the pool where they were incurred.
[3073] In Docket R97-1, and again in this docket, the Commission recommends apportioning Segment 3 mail processing labor costs to mail processing, window service, and administrative components, using the information provided by IOCS tallies to determine what operation a worker was part of when tallied. Where the IOCS information was in conflict with the clocking information provided by the MODS system, the Commission concluded that the IOCS data collector's direct observations were more likely to be accurate than the self-clocking information provided by the MODS system.
[3074] Since the Commission leaves costs in the mail processing, window service, or administration component where IOCS information indicates they belong, it is appropriate to use direct tallies from the MODS pool where the IOCS indicates they belong as the basis for distributing the mixed mail and not handling costs found in each pool.
[3075] Postal Service witness Degen argues that it is conceptually superior to distribute costs in the MODS Function 1 and Function 4 "support" pools according to the subclass distribution of the volume variable costs in the pools that they support. He argues that the direct tallies indicating that administrative clerks were observed processing mail or performing window service functions are "incidental" to their administrative duties. Therefore, he argues, they are not accurate indicators of the cause of the costs incurred in those support operations. He proposes that such direct tallies be ignored when distributing the costs of these support pools. USPS-T-16 at 55-57. The Periodical mailers prefer that costs be distributed on the basis of these direct tallies. Periodical Mailers' Initial Brief at 28.
[3076] The Commission rejects this proposal, but its rejection is not definitive. The reasons offered by witness Degen for concluding that direct tallies in these support pools do not indicate causation were thin and conclusory. The Commission needs a more informative discussion of the activities actually performed in these support pools in order to independently evaluate witness Degen's assertion that direct tallies found there do not indicate causation.
[3077] Question 19 data provides an indication of the location of a worker at the time he was tallied by the IOCS data collector. The Periodical mailers argue that this information sometimes can be used to infer what operation, and what shape of mail, a mixed mail or not handling tally should be associated with. Tr. 24/11384-86. Witness Stralberg argues that Question 19 data can be extended beyond its traditional use to support inferences about employees tallied while on break or handling empty containers. Id., at 11378-79. He argues that this can be done separate from, or in conjunction with, the Postal Service's use of item and container strata to draw similar inferences. Witness Stralberg shows that the Question 19 data often conflict with the item and container inferences drawn by the Postal Service. He concludes that this provides grounds for doubting the accuracy of inferences based on item and container. Periodical Mailers' Initial Brief at 27.
[3078] Witness Degen argues that Question 19 data provide no information on the causal relationship between accrued costs and mail subclasses. He argues that this is especially true when the worker is clocked into Function 4, non-MODS, allied, or support cost pools. He asserts that activities in those pools are not location specific because workers in those activities are required to move among activities transporting full or empty containers. He argues, for example, that when a platform worker who is creating a parcel sorting corral is tallied retrieving an empty hamper from a BCS operation, Question 19 will report the BCS location. He argues that it would be erroneous to infer from Question 19 location information that letter mail should be charged for the costs of setting up a parcel sorting operation. Tr. 38/17324-25.
[3079] The Commission agrees that care must be taken to avoid drawing weak and unsupported shape and subclass inferences from the location information in Question 19. Accordingly, it does not endorse a mechanical application of Question 19 data to distribute the costs of location-ambiguous activities to subclasses.
[3080] The Periodical mailers also argue that IOCS Question 18 provides information indicating that employees clocked into mail processing operations were actually working at postal windows. Witness Stralberg proposes to trust what the IOCS data collector saw, and distribute the costs according to a window service, rather than a mail processing distribution key. This proposal would be relevant if the Commission were migrating window service tallies into mail processing cost pools on the basis of clocking information from the MODS system. The Commission leaves window service tallies where the IOCS indicates they belong, and applies a window service distribution key to these costs. Witness Stralberg's proposed adjustment is therefore moot.
 Postal Rate Commission http://www.prc.gov Voice: (202) 789-6800 Fax: (202) 789-6886 prc-admin@prc.gov |
| |
|
|
|