Development and analysis of alternative dispatching methods in truckload trucking

Transcription

1 Transportation Research Part E 35 (1999) 191±205 Development and analysis of alternative dispatching methods in truckload trucking G.D. Taylor a, *, T.S. Meinert a, R.C. Killian a, G.L. Whicker b a Department of Industrial Engineering, University of Arkansas, 4207 Bell Engineering Center, Fayetteville, AR 72701, USA b J.B. Hunt Transport, P.O. Box 130, 615 J.B. Hunt Corporate Drive, Lowell, AR 72745, USA Received 3 February 1998; received in revised form 29 December 1998; accepted 14 January 1999 Abstract In this paper, the authors discuss industry±university collaborative research aimed at improving the performance of truckload trucking operations for a major North American carrier through the development of alternative dispatching methods. The research detailed in this paper develops regularly scheduled delivery capacity in the form of delivery lanes, hubs and zones which regularize driver tours while providing performance bene ts for the carrier. The impact of the alternative methods developed are studied through extensive experimentation using discrete event simulation and actual demand data provided by the industrial partner. Ó 1999 Elsevier Science Ltd. All rights reserved. 1. Introduction and research motivation This paper focuses on one important component of business logistics systems: the truckload (TL) trucking industry in North America. According to Schwartz (1992), customer service expectations and industry paradigms are rapidly evolving in the TL arena. The most signi cant of these changes in the TL trucking industry in the United States has occurred as result of industry deregulation following the 1980 Motor Carrier Act. Previous to this time, Schwartz argues that government regulation caused ine ciencies in routes, service authority, commodity authority, carrier selection by shippers, and rates. Because of the relatively small investment required to enter the truckload trucking market, the post deregulation industry is very competitive. Market share is gained or lost based on cost and service. Pragmatic research must therefore address ways to reduce cost and increase performance and value to customers. * Corresponding author. Tel.: ; fax: ; gdt@engr.uark.edu /99/$ ± see front matter Ó 1999 Elsevier Science Ltd. All rights reserved. PII: S (99)

2 192 G.D. Taylor et al. / Transportation Research Part E 35 (1999) 191±205 In this paper, the authors undertake research on some daunting problems in the truckload trucking industry in North America. We focus on improvements to the driving job and to service quality to customers. We make extensive use of case information supplied by J.B. Hunt Transport (JBH), the largest publicly held TL trucking company in the United States, in seeking improvements to driver job quality and customer service needs (Hunt, 1996). Schwartz (1992) cites driver recruiting and retention as a key truckload trucking business strategy in the 1990's. This is an opinion shared by most researchers and by o cials in the truckload trucking industry. As pointed out in Mele (1989a, b), turnover rates can range from 85% to 110% per year in the TL industry. There are two reasons for this turnover. First, random over the road (OTR) drivers often have very long tour lengths (See Powell et al., 1988) for more information on OTR dispatching). Second, the quality of life on the road is low. Generally speaking, the driver is either driving, eating, or sleeping at all times. In contrast, Mele points out that Yellow Freight System, a less-than-truckload (LTL) trucking company, has a turnover rate of 4.5% for city drivers and slightly above 10% for linehaul drivers. The reason for this increased loyalty seems to be the fact that the LTL driving job is much more regular, with driving routes to support the same end-of-line or breakbulk terminals daily. The resulting short tour lengths make the job much more attractive. The authors of this paper have examined several alternatives to random OTR dispatching in an e ort to regularize the TL driving job and to improve driver retention. The rst such e ort was the consideration of a hub & spoke (H&S) network similar to those employed in LTL settings. Although di erently-motivated and designed, this work was inspired by hubbing in LTL (See Braklow et al., 1992) and airline industries (See Kanafani and Ghobrial, 1985 or Morrison and Winston, 1986). See Taha et al. (1996), or Taha and Taylor (1994) for information about this problem, and for information about the HUBNET simulation tool which was developed. The ndings of research with H&S networks indicate that while tremendous savings are possible in terms of driver tour length, the improvement comes at the expense of miles per driver per day, circuity and rst dispatch empty miles. These ndings have led the academic and industrial investigators to conclude that limited implementation seems to be the best alternative for H&S usage in the TL environment and have motivated this research into new network design and dispatching methodologies. (See Taylor et al. (1995) for more information regarding experimentation with H&S networks using the HUBNET system). A crucial observation made by the authors during research into H&S networks has driven subsequent research. Almost any solution methodology that restricts drivers to regular lanes or service areas achieves the goal of tour length reduction. Unfortunately, almost any solution methodology that abandons point-to-point deliveries adds to load circuity. Furthermore, depending upon the solution methodology and associated parameters, rst dispatch empty miles may increase or decrease, and the miles driven per driver per day may increase or decrease. This is to say that there exists a basic performance tradeo between driver job quality metrics and carrier performance and customer service metrics. This research addresses the development of dispatching methods which provide good performance relative to these multiple performance metrics. From a driver viewpoint, quality is a function of several things, but can be well addressed by the miles per driver per day metric, which is directly proportional to pay. From a company viewpoint, one may be more interested in driver retention, equipment utilization and the on-time

3 G.D. Taylor et al. / Transportation Research Part E 35 (1999) 191± delivery performance. Fortunately, improvements in driver job quality relative to miles per driver per day and in tour regularity simultaneously provide improved driver job quality and reduce service provider costs associated with driver turnover. From a customer perspective, quality is largely a function of on-time pick-ups and deliveries. Some of these multi-criteria goals may appear to compete with one another, but strategies developed in this research seek to concurrently improve quality relative to the driver, the customer, and the trucking company. Certainly, any strategy that helps with driver retention or customer satisfaction is a good candidate for consideration as a long-term strategy for the company. The following section discusses an experimental design which addresses several concerns. This experimentation seeks to provide design guidelines for alternatives to point-to-point delivery such as lanes or zones in an e ort to concurrently provide driver tour length reduction while achieving acceptable performance relative to carrier and customer metrics. Performance evaluations are made relative to standard industry metrics which address carrier, driver and customer goals. Included in this experimental plan is sensitivity analysis around key parameters and examination of the impact of alternative geographical areas to ensure breadth of applicability for results. 2. Research experimental plan In this section, the authors describe the experimental research plan and brie y discuss the tools used to complete the research tasks. Experimental dispatching methods are developed which seek to provide good performance with respect to several metrics while addressing driver tour length reduction. Because of the costs associated with driver turnover, any operational techniques that improve the regularity or quality of the driving job contribute to strategic cost reduction goals. Also, depending upon the methodology selected for tour reduction, the potential exists to concurrently improve on-time delivery performance to customers, thus directly serving customer service goals. The primary research tool for use in this research is discrete event system simulation using the SIMNET II language. All simulation scenarios described herein are data driven using actual JBH historical load information. Statistical output processing has been completed using default SIMNET II output and the SAS statistical package. For initial study, we have selected and experimentally isolated a region in the Southeast (SE) United States for consideration. In the primary experimental design, seven alternative formulations have been considered in an e ort to nd ways to improve dispatching operations and to convert OTR to regional driving jobs with the goal of improving overall system performance: 1. A `baseline' model featuring point-to-point OTR tours for comparison purposes. 2. A `zone' model that makes use of six SE zone perimeter hubs as drop points. The idea is that SE regional drivers pick up and deliver loads within the region while OTR drivers stop at the zone boundary. This would help to ensure that SE regional drivers could have greatly reduced tour lengths. The zone hubs are located near the region border in locations that provide access to major highways and existing freight corridors. 3. A `key lane' model that moves a percentage of baseline loads along a well-de ned delivery lane into and out of the SE region. The lane is selected to be conducive to a one-day drive and because of already existing freight density that provides a reasonable volume and balance. One

4 194 G.D. Taylor et al. / Transportation Research Part E 35 (1999) 191±205 such lane considered is Atlanta, Georgia to Memphis, Tennessee. Loads that can traverse down the lane without encountering more than 20% circuity are selected for lane travel. 4. A second `key lane' model that makes use of an Atlanta, GA to Richmond, VA lane. The reasoning behind the selection of this lane is similar to that for the Atlanta±Memphis lane. Existing freight density and a useful one-day driving time are characteristics which led to this selection. Both of the key lanes addressed utilize Atlanta, GA. This choice re ects the fact that Atlanta is both the largest city and largest freight center in the SE region. The Memphis and Richmond endpoints for these lanes represent major freight centers located on major highways with Richmond being a key point en-route to the Northeast (NE) and the Atlantic seaboard. Similarly, Memphis is located on a major highway from Atlanta and provides good access to much of the West and Midwest. 5. A third `key lane' model that concurrently considers both key lanes. 6. A `key hub' model that uses a single Atlanta, GA hub instead of the six zone hubs as a transshipment point. 7. A `hybrid' model that combines the key hub and zone models for an integrated solution. The experimentation clearly shows the e cacy of the various design approaches, physical resource con gurations, and operational rules used in the experimental design. Fig. 1 shows the location of the SE region, the major hub in Atlanta, GA, the six zone perimeter hubs and the lane locations used in the primary experimental design. Sensitivity analysis is accomplished using additional experimentation and a secondary experimental design. The sensitivity analysis includes examination of the allowable circuity for lane participation for both single lane and 2-lane models in the SE region. Additional sensitivity analysis is performed to determine the optimal number of perimeter hub locations to use in the zone model in the SE region. The experimental design also Fig. 1. The southeast region.

5 G.D. Taylor et al. / Transportation Research Part E 35 (1999) 191± speci es the study of an additional geographical region in the NE United States to ensure that solutions are not speci c to a particular set of geographic and operational constraints. 3. Research ndings We begin with the primary experimental design in the SE region. A global comparison of all metrics of interest appears in Table 1. Because all performance information presented in this section is considered to be very proprietary by JBH, all metrics are published in comparison to the baseline, point-to-point OTR scenario. One of the complexities involved in the analysis of large scale logistics systems is the identi cation of meaningful performance measures. This is particularly true in the case of this research because we are seeking to simultaneously address performance relative to service provider, customer and driver goals. Some performance measures are important to more than one of these entities but not all are compatible. For instance, the use of Table 1 Southeast region results summary Baseline ATL-MEM ATL-RVA 2 lane ATL HUB Zone Hybrid Key performance measures % CIRC DISP MILES DR LATE% Secondary performance measures LATE HR CIRC % AM LN ± ± ± ± % AR LN 0.00 ± ± ± ± % HUB 0.00 ± ± ± MILES AM ± 0.96 ± 0.98 ± ± ± MILES AR ± ± ± ± ± MILES SE ± ± ± ± MILES HB ± ± ± ± MAX DR AVG DR MAX AM ± 0.21 ± 0.21 ± ± ± AVG AM ± 0.17 ± 0.18 ± ± ± MAX AR ± ± ± ± ± AVG AR ± ± ± ± ± MAX HUB ± ± ± ± AVG HUB ± ± ± ± MAX SE ± ± ± ± AVG SE ± ± ± ± IMBAL AM IMBAL ± 0.23 ± 0.23 ± ± ± AR IMBAL ± ± ± ± ±

6 196 G.D. Taylor et al. / Transportation Research Part E 35 (1999) 191±205 regular driving lanes improves driver job quality but may result in circuitous miles with an associated cost unacceptable to the service provider. On the other hand, driver job quality in terms of tour length and regularity is a critical factor in driver turnover which is a signi cant cost for service providers as well. The following is a description of each of the scenarios and performance measures analyzed during this work. The measures are presented in four sections: key service provider metrics, key driver job quality metrics, key customer service metrics and secondary performance metrics. Scenarios: 1. BASELIN ± the baseline point-to-point OTR scenario. 2. ATL±MEM ± the Atlanta, GA/ Memphis, TN lane scenario. 3. ATL±RVA ± the Atlanta, GA/ Richmond, VA lane scenario LANE ± the scenario featuring both SE lanes. 5. ATL HUB ± the Atlanta, GA key hub scenario. 6. ZONE ± the six SE zone perimeter hub scenario. 7. HYBRID ± the zone/key hub hybrid scenario. Key service provider performance measures: 1. % CIRC ± the average percent circuity (actual miles compared to point-to-point miles) for those loads that do not travel point-to-point. This measure is important to the service provider because circuitous miles clearly have a direct impact on the cost of a load. Driver wages, fuel costs and maintenance for vehicles are directly related to miles driven DISP ± rst dispatch empty miles compared to a 1.00 baseline. This measure is the average number of empty miles encountered by a vehicle between dispatch and load pickup. Clearly, the service provider has an interest in minimizing the number of empty miles encountered by its drivers and vehicles. Key driver performance measures: 1. MILES DR ± the average number of miles per driver per day compared to a 1.00 baseline. This metric is crucial to drivers because it directly e ects their job quality. It is also important to the service provider because it a ects driver turnover costs. Key customer performance measures: 1. LATE % ± the percent of all loads which arrive at the customer late compared to 1.00 baseline values. This metric clearly indicates the customer service level of a particular scenario. In addition to the principle performance measures discussed above, there are several secondary performance measures used in the evaluation of each experimental scenario. These metrics are of interest to the industrial sponsor for this research and also provide the researchers with information related to overall solution quality and performance tradeo s resulting from a particular dispatching method. Secondary performance measures: 1. LATE HR ± the average number of late (or early) hours per load. In conjunction with the percent late metric, this measure can be used to evaluate the magnitude of customer service performance. 2. CIRC ± the average circuity (excess miles per trip) for all loads in comparison to point-topoint distances (assume circuity equals zero in the baseline model). 3. % AM LN, % AR LN, and % HUB ± the percent of total loads that travel along the Atlanta- Memphis lane, the Atlanta-Richmond lane, or that travel through one of the zone hubs,

7 G.D. Taylor et al. / Transportation Research Part E 35 (1999) 191± respectively. This measure of participation in lane or hub dispatching methods is helpful in evaluating the impact of the % CIRC measure above. It is also a good measure of general solution quality in that it provides information about the percentage of all network tra c that can be moved using a dedicated lane or zone. 4. MILES AM, MILES AR, MILES SE, and MILES HB ± the average number of miles per driver per day for Atlanta-Memphis lane drivers, Atlanta-Richmond lane drivers, SE regional drivers, and external OTR drivers, respectively, compared to the 1.00 MILES DR baseline. These measures provide a basis for comparison between the tour lengths for all types of drivers. 5. MAX DR and AVG DR ± the maximum and average number of active drivers used in the simulation study compared to 1.00 baseline values. 6. MAX AM and AVG AM ± The maximum and average number of Atlanta±Memphis lane drivers compared to 1.00 baseline values. 7. MAX AR and AVG AR ± The maximum and average number of Atlanta±Richmond lane drivers compared to 1.00 baseline values. 8. MAX HUB and AVG HUB ± the maximum and average number of OTR drivers external to the SE region making deliveries to/from the zone hubs compared to 1.00 baseline values. 9. MAX SE and AVG SE ± the maximum and average number of SE regional drivers making deliveries to/from the zone hubs compared to 1.00 baseline values. 10. IMBAL ± the number of loads into the SE region minus the number of loads out of the SE region compared to a 1.00 baseline value. 11. AM IMBAL and AR IMBAL ± the lane imbalance into/out of the SE region on the Atlanta± Memphis and Atlanta±Richmond lanes, respectively. 4. Results The explicit collection of driver tour length is not undertaken in this research because detailed eld data for various point-to-point and regional alternatives are available through JBH and because the computational overhead associated with collection is considerable. Field data indicates that `typical' OTR tour lengths are on the order of 14±21 days while tour lengths for regional jobs of the type presented herein are on the order of 2±4 days. Thus, our goal for this research is to determine which of the regional dispatching alternatives performs best in terms of the four key metrics and several secondary metrics presented above. The results presented in Table 1 are based on the mean value obtained from ve replications of the simulation models developed for each scenario. Each replication simulates one steady-state week of TL trucking operations. Additional information regarding the development of the baseline and key lane models can be found in Killian (1998). Key hub and zone model development for the SE region is discussed in Ganglu (1998). While Table 1 provides comprehensive information regarding mean the values for all performance measures, the primary discussion of metrics will focus on four of the key metrics: miles per driver per day, the percent circuity, rst dispatch empty (deadhead) miles, and average percent of late (or early) loads. In Table 1, it is clear that the Atlanta key hub scenario produces the highest level for miles per driver per day. Even so, many of these miles are circuitous miles caused by forcing the loads

8 198 G.D. Taylor et al. / Transportation Research Part E 35 (1999) 191±205 through the hub. Therefore, although the key hub scenario appears strong relative to this metric, the informed observer can recognize some ine ciencies. The only other scenario comparing favorably with the baseline scenario relative to this metric is the zone model. The zone model, however, simultaneously provides a much smaller level of circuitous miles than in the key hub model. Means tests on the results indicate, however, that the zone model is not statistically different from the baseline at a 95% con dence level. With respect to the percent circuity metric the key hub model is a very poor performer. This is one of the observations which prevented the authors from giving serious consideration to H&S networks for truckload trucking applications. The baseline is assumed to produce no circuity because all loads are delivered using point-to-point OTR drivers. The percent circuity is therefore caused by those loads that travel along lanes or through hubs. The hybrid model compares most favorably to the baseline model for the percent circuity metric. The key lane models perform adequately, but have a relatively small lane participation (approximately 4±22%). The zone model produces only 10.70% circuity for zone loads with more than 90% zone hub participation. Relative to rst dispatch empty miles, the zone model performs very well. It o ers statistically signi cant improvement over the baseline model at the 95% con dence level. The key hub and hybrid models also o er statistically signi cant improvement compared to the baseline, but do not o er the same level of performance relative to other measures of e ectiveness that continue to make the zone model an attractive alternative. The reason for the strong performance of the hub, zone, and hybrid models is that the positioning of the hubs is well coordinated with freight pick-up and delivery locations. Drops at zone hubs leave trucks well positioned for a next load pick-up. When considering the percentage of loads late, only the key hub model performs worse than the baseline. The hybrid model performs best relative to this important customer service indicator, but the results miss being signi cantly di erent from the baseline scenario at the 95% con dence level by a small margin. Similarly, the other scenarios that o er improvement do not do so with statistical signi cance in comparison with the baseline scenario. The reason the baseline model is not necessarily good relative to on-time performance stems from the nature of US Department of Transportation (DOT) rules regulating OTR drivers. The DOT requires that drivers adhere to strict rules regarding sleep and driving time. As a result, for a given load to move a distance greater than that covered in one driving cycle by a single driver, the load must sit idle while the driver sleeps. Each of the dispatching methods studied during this research convert OTR jobs to regional driving jobs which have been designed around distances which can be achieved in a single day. These regional jobs allow a single load to be deposited at a hub or lane endpoint by one driver and retrieved by another without having to encounter an entire sleep cycle. As a result, those loads participating in a zone or lane model can have substantially reduced transit times and associated improvements in on-time performance. A summary of the results from the SE region is now presented. All of the proposed alternatives to the baseline model would o er tremendous opportunity to convert OTR to regional or local jobs. If this were the only concern, each would be a viable candidate for further development. When considering all metrics, however, the selection of the best alternative becomes less clear. We now discuss the relative merits of each alternative considered. The hub model produces so much excess circuity that almost all other metrics are also adversely a ected. Therefore, the single hub model should not be considered further.

9 G.D. Taylor et al. / Transportation Research Part E 35 (1999) 191± The three di erent lane models examined led to improvement in some areas at the expense of others. The improvement o ered by these models (in addition to controlled driver tour lengths for lane drivers) is primarily in terms of customer service. Lateness and percent lateness is generally reduced for those models that have reasonable lane participation. The problem is that this service improvement is achieved while increasing circuity and rst dispatch empty miles, and while reducing the critical operational metric of miles per driver per day. The miles per driver per day metric is especially signi cant because a reduction here means that not only does the strategy reduce equipment utilization, but it reduces driver pay. Thus the problem contributes to the driver retention problem. Subsequent sensitivity analysis relative to the lane models is completed to determine whether or not lane participation rules a ect performance. Speci cally, the original lane models permit lane participation on a lane for those loads that do not encounter more than 20% circuity in traversing the lane. The authors have performed a secondary experimental analysis to determine the sensitivity of the various performance metrics to the lane participation rule. In this secondary analysis, lane participation has been permitted for 15% and for 25% circuity. Table 2 presents the results from the lane participation sensitivity analysis. The experiments have been completed for all three lane models. The Atlanta±Memphis scenarios are labelled `AM', the Atlanta±Richmond scenarios are labelled `AR', and the 2 lane models are labelled `BOTH' in Table 2. The 20 columns in Table 2 are considered `baseline' scenarios for the sensitivity analysis so all performance comparisons are relative to the 1.00 values in the 20 baseline. Although Table 2 Sensitivity analysis for allowable circuity in the southeast region AM 15 AM 20 AM 25 AR 15 AR 20 AR 25 BOTH 15 BOTH 20 BOTH 25 Key performance measures % CIRC DISP MILES DR LATE % Secondary performance measures LATE HR CIRC % AM LN ± ± ± % AR LN ± ± ± MILES AM ± ± ± MILES AR ± ± ± MAX DR AVG DR MAX AM ± ± ± AVG AM ± ± ± MAX AR ± ± ± AVG AR ± ± ± IMBAL AM IMBAL ± ± ± AR IMBAL ± ± ± ±

10 200 G.D. Taylor et al. / Transportation Research Part E 35 (1999) 191±205 somewhat arbitrarily selected initially, the 20 baseline value performs well during sensitivity testing. Reducing this value to 15 generally leads to improvements in terms of miles per driver per day and in terms of the number of drivers required. Likewise, the circuity measures are improved. This, of course, is an obvious result. The rst dispatch empty miles performance metric appears indi erent to the reduction in allowable circuity. The only negative performance observed for the reduction is in the customer service metrics of late hours and the percent of late loads. This negative performance is most pronounced for the AM scenarios, but is present to a lesser extent in the other scenarios. The results of increasing allowable circuity to 25% are predictable, with the only real improvements coming in the customer service metrics. Therefore, it would appear that the 20% allowable circuity rule results in a compromise between customer service and company or driver performance metrics for the lane models. The zone and hybrid models appear to provide fairly robust improvements in some areas and do not perform as well in others. Both the zone and hybrid models perform worse than the baseline in terms of average lateness but perform better than the baseline scenario in terms of the percent late. Therefore, these alternatives lead to a smaller percentage of later deliveries. Both models lead to a statistically signi cant improvement in rst dispatch empty miles, have strong participation in the hub programs (more than 90%), and have `acceptable' levels of circuity given that the conversion from OTR to regional driving jobs would necessarily be accomplished with some circuity increases. The zone model appears to perform slightly better than the hybrid model in general. The zone model operates with fewer drivers, fewer rst dispatch empty miles, and less late hours. The hybrid model performs better in terms of circuity and the percent of late jobs, but most of these di erences are not statistically signi cant. The most critical di erence is in terms of miles per driver per day, where the zone model performs statistically better at a 95% con dence level in comparison with the hybrid. Based on these results, it is concluded that the zone model appears to be the best dispatching technique in the SE region from among the alternatives examined when simultaneously considering key driver, carrier and customer performance measures. Sensitivity analysis has also been completed for the number of hubs used in the SE region to determine the e ects of using a large or small number of hub locations. Figs. 2±5 present mean performance results and 95% con dence intervals for percent circuity, miles per driver per day, rst dispatch empty miles, and average percent of late (or early) loads criteria for SE zone models. This experimentation makes use of 2±10 hub locations. The circuity metric presented in Fig. 2 is characterized by tight con dence intervals and mean values that decrease as the number of hubs increase. It is clear in Fig. 2 that solutions with 2 or 3 hubs have substantially greater circuity than those with 4 or more hubs. The miles per driver per day criterion presented in Fig. 3 seems to perform especially well with 2 hubs but produces no results which di er signi cantly between 3 and 10 hubs. Recall that 2 hubs produce more circuitous miles. Thus indicating that 4 hubs may be a better lower bound. The rst dispatch empty miles results in Fig. 4 indicate that 3±5 hubs result in the best values for this metric. Scenarios with 7 or more hubs perform signi cantly worse than those with 3 or 4 hubs. Finally, the percent of late loads, illustrated in Fig. 5, seems to be lower for scenarios with 6 or more hubs but none of the scenarios signi cantly di er from the others. These sensitivity analysis results seem to indicate that scenarios with 4±6 hubs in the SE region seem to o er the best compromise solutions relative to all four of the key metrics. These scenarios gain most of the steep improvements in circuity possible from hub increases yet do not

11 G.D. Taylor et al. / Transportation Research Part E 35 (1999) 191± Fig. 2. Comparison of SE circuity for various hub con gurations. Fig. 3. Comparison of SE miles per driver per day for various hub con gurations. pull drivers away from dense freight regions for strong rst dispatch empty miles performance and good miles per driver per day performance. To determine if these results are robust in general terms, an additional analysis has been undertaken using a second geographical region in the NE United States. Fig. 6 presents pertinent information about the NE area in terms of geographical boundaries, hub locations, and lane locations. Table 3 presents the results of testing for an identical set of scenarios and performance measures to those used in the SE study. Some description of the NE region scenario names is required for understanding Table 3. The EBR±EMP lane is from East Brunswick, NJ to Emporia, VA. The EBR±AKR lane is from East Brunswick, NJ to Akron, OH. The EBR hub model is analogous to the Atlanta, GA key hub model in the SE region. These key lanes and key hub were

12 202 G.D. Taylor et al. / Transportation Research Part E 35 (1999) 191±205 Fig. 4. Comparison of SE rst dispatch empty miles for various hub con gurations. Fig. 5. Comparison of SE average percent late for various hub con gurations. selected for reasons similar to those discussed for the SE region. East Brunswick, NJ is an existing freight depot and is central in a dense freight region. Akron, OH is a one day drive via an existing freight corridor on major highways with access to the Midwest while Emporia, VA provides similar access to the SE. Some metric titles have also changed. Obviously, NE has replaced SE in these titles. Also, EE represents East Brunswick±Emporia and EA represents East Brunswick±Akron. Table 3 tends to support the thesis that the results are transferable across regions. The maximum and average driver statistics are very similar. The circuity performance is also similar, but in the NE region both lanes are heavily travelled and circuity is therefore generally higher for the individual lane models. First dispatch empty miles performance has similar trends to those observed in the SE region, but the zone and hybrid models perform even better in the NE region.

13 G.D. Taylor et al. / Transportation Research Part E 35 (1999) 191± Fig. 6. The northeast region. Table 3 Northeast region results summary BASELINE EBR-EMP EBR-AKR 2 LANE EBR HUB ZONE HYBRID Key performance measures % CIRC DISP MILES DR LATE % Secondary performance measures LATE HR CIRC % EE LN ± ± ± ± % EA LN 0.00 ± ± ± ± % HUB 0.00 ± ± ± MILES EE ± 0.96 ± 0.76 ± ± ± MILES EA ± ± ± ± ± MILES NE ± ± ± ± MILES HB ± ± ± ± MAX DR AVG DR MAX EE ± 0.19 ± 0.13 ± ± ± AVG EE ± 0.16 ± 0.11 ± ± ± MAX EA ± ± ± ± ± AVG EA ± ± ± ± ± MAX HUB ± ± ± ± AVG HUB ± ± ± ± MAX NE ± ± ± ± AVG NE ± ± ± ± IMBAL EE IMBAL ± 0.20 ± 0.14 ± ± ± EA IMBAL ± ± ± ± ±

14 204 G.D. Taylor et al. / Transportation Research Part E 35 (1999) 191±205 The customer service metrics involving lateness are also similar, but again the zone model performs even better in the NE than in the SE. In fact, the negative performance in terms of lateness in the SE is reversed for the zone model in the NE. In the NE region, both lateness and percent lateness are improved. Both of these metrics o er statistically signi cant improvement relative to the baseline at the 95% con dence level. Circuity is reduced in comparison with the SE, primarily because the NE region is more easily isolated from the remainder of the United States with less area to cover with perimeter hubs over a smaller collection of major roads. Consequently, almost 98% of loads traverse through a hub in the zone and hybrid models. One could argue that circuity with `zone' methods could be further reduced by using a larger number of drop points, even to the extreme of using so called `drop & swap' software to match one load entering the zone with one load exiting the zone. In practice, these techniques have resulted in logistical nightmares and are subject to the scheduling inconsistencies inherent to stochastic systems. The use of a smaller number of drop yards eliminates some of these stochastic ine ciencies and permits smooth ow of goods across the zone boundary. The miles per driver per day are slightly less for the zone and hybrid models than in the NE baseline scenario, but not alarmingly so and certainly not at a statistically signi cant level. The examination of the second geographical area not only adds credibility to the e ort in terms of making claims of generality, but in fact demonstrates that some regions may be even more well suited to the dispatching strategies than the original scenario. 5. Research relevance and implications This research has investigated TL network design and operation methods which simultaneously address the improvement of driver job quality, customer service and service provider operations. Several alternative dispatching methods have been studied using discrete event simulation and their performance has been quanti ed relative to several metrics. This research has provided promising compromise solutions. The research discussed herein is a signi cant contribution to the literature in determining viable dispatch and delivery alternatives to the OTR system employed by most TL trucking companies. The strategies presented herein have also turned into practical reality within the JBH delivery network. At this writing, a `key lane' is in use in the Eastern United States, linking northern and southern marketing areas with a single east coast lane. Also at this writing, a `zone' system is fully operational in the NE United States. As demonstrated in this paper, conversion of OTR to regularized jobs can have many positive e ects including e ects to the bottom line. Currently, JBH driver turnover rates are on the order of 53% per year for OTR drivers but only 22% for drivers that have been converted to regular routes or zones. With an OTR driver base approaching 6000 drivers, and conservatively estimating hiring costs at $2000 per driver, the savings associated with conversion to regular routes is very signi cant. References Braklow, J.W., Graham, W.W., Hassler, S.M., Peck, K.E., Powell, W.B., Interactive optimization improves service and performance for yellow freight system. Interfaces 22 (1), 147±172.

15 G.D. Taylor et al. / Transportation Research Part E 35 (1999) 191± Ganglu, G.E., Alternative delivery methodologies for truckload trucking operations. Forthcoming Master's thesis, Department of Industrial Engineering, University of Arkansas, Fayetteville, AR. Hunt Transport Inc, J.B., Internet Home page: jbhhome.htm. Kanafani, A., Ghobrial, A.A., Airline hubbing±some implications for airport economics. Transportation Research 19A, 15±27. Killian, R.C., Development and analysis of e cient delivery lanes. Forthcoming Master's Thesis, Department of Industrial Engineering, University of Arkansas, Fayetteville, Arkansas. Mele, J., 1989a. Carriers cope with driver shortage. Fleet Owner 84 (1), 104±111. Mele, J., 1989b. Solving driver turnover. Fleet Owner 84 (9), 45±52. Morrison, S.A., Winston, C., Economic e ects of airline deregulation, Brookings Institute: Washington DC. Powell, W.B., She, Y., Nickerson, K.S., Butterbaugh, Atherton, S., Maximizing pro ts for north american van lines truckload division: A new framework for pricing and operations. Interfaces 18 (1), 21±41. Schwartz, M., J.B. Hunt: The Long Haul to Success. University of Arkansas Press, Fayetteville, AR. Taha, T.T., Taylor, G.D., An integrated modeling framework for the evaluation of hub and spoke networks in truckload trucking. The Logistics and Transportation Review 30 (2), 141±166. Taha, T.T., Taylor, G.D., Taha, H.A., A simulation-based software system for evaluating hub-and-spoke transportation networks. Simulation Practice and Theory 3, 327±346. Taylor, G.D., Harit, S., English, J.R., Whicker, G.L., Hub and spoke networks in truckload trucking: Con guration, testing, and operational concerns. The Logistics and Transportation Review 31 (3), 209±237.