Comparison of Large-Truck Travel Estimates from Three Data Sources

Transcription

1 5 TRANSPORTATION RESEARCH RECORD 147 Comparison of Large-Truk Travel Estimates from Three Data Soures DAWN L. MASSIE, KENNETH L. CAMPBELL, AND DANIEL F. BLOWER The number of miles traveled eah year by the U.S. large-truk population is a topi of interest for many reasons, one of whih is safety. Although the number of aidents involving large truks may be easily alulated from aident data, it is often more informative to know their risk of aident involvement per mile of travel. This requires aurate travel data. Compared in this paper are three soures of truk travel data: the Truk Inventory and Use Survey onduted by the Bureau of the Census; the National Truk Trip Information Survey onduted by the University of Mihigan Transportation Researh Institute; and annual estimates published in Highway Statistis by the Federal Highway Administration. Eah data soure yields different estimates of annual travel by large truks, whih is to be expeted onsidering the diffiulty of olleting travel data. The overall onlusion, however, is that the Truk Inventory and Use Survey and the National Truk Trip Information Survey estimates are reasonably lose to eah other, whereas Highway Statistis estimates are signifiantly higher. The impliation of this finding is that the proedures used by the states and the Federal Highway Administration to generate Highway Statistis data lead to artifiially and systematially high estimates of travel by large truks. The Center for National Truk Statistis of the University of Mihigan Transportation Researh Institute (UMTRI) onduted a national survey of medium and heavy truks beginning in January of Termed the National Truk Trip Information Survey (NTTIS), the study produed estimates of the national registered large-truk population and its annual travel. The methodology of NTTIS is desribed in detail in a ompanion paper (J). In this paper, estimates of largetruk travel from NTTIS and two other soures are ompared. One purpose of the omparisons is to assess the degree of orrespondene among the three travel estimates. Another is to illustrate the inherent diffiulty in measuring truk travel and the benefit of onsidering multiple soures of travel data. Despite the diffiulty and assoiated ost of olleting travel data, truk travel information is vitally needed in order to make informed deisions on a host of topis, partiularly those onerning truk safety. NTTIS AND COMPARISONS The omparisons start with two soures of data, NTTIS and the Truk Inventory and Use Survey (). is onduted every 5 years by the Bureau of the Census as part of Center for National Truk Statistis, University of Mihigan, Transportation Researh Institute, 291 Baxter Road, Ann Arbor, Mih ' the Census of Transportation. NTTIS and begin with a ommon base, the R.L. Polk vehile registration files. The sampling frame for NTTIS was formed from the July 1, 1983, Polk files. The two most reent surveys were drawn from the July 1, 1982, and July 1,, versions of the Polk files respetively. NTTIS restrited its sample to truks with a gross vehile weight rating (GVWR) greater than 1, lb. All pik-up truks, regardless of GVWR, were exluded from the sample, as were all passenger vehiles (suh as passenger vans, rereational vehiles, ambulanes, and buses of any type), farm trators., and government-owned truks. Similar to NTTIS, exluded ambulanes, open utility vehiles, motor homes, buses, farm trators, and governmentowned vehiles. Unlike NTTIS, sampled truks of any GVWR, inluding light truks. The implementation phase of NTTIS was arried out in January through May of As part of this phase, survey interviewers ontated truk owners and asked them how far they drove their power unit in a year. Phone interviewers also obtained desriptive information on the truk and the ompany at this time. The implementation phase produed data on 6,35 truks. During the subsequent trip phase of NTTIS, truk owners were ontated by phone four times over the ourse of a year. Eah time, interviewers sought information on all trips made by the truk in a speifi 24-hr period. Detailed physial information on the truk and its argo was olleted, and the routes traveled by the truk were mapped aording to road type, population area (rural/urban), and time of day (day/night). is onduted through survey forms mailed to owners of seleted truks beginning in January of the year after the Polk sample is drawn. Owners haraterize their truks in terms of the typial onfiguration and use over the previous year. This inludes an estimate of the number of miles traveled, as well as information on the number of trailers usually hauled, type of argo usually arried, typial weight of a load, and so on. The 1982 olleted data on a total of 84,334 truks, inluding light truks. The total was 14;66 truks. Before the ompletion of NTTIS, had been the only national data base onerning the use of large truks. Therefore, it is important to onsider whether major differenes exist between NTTIS and, given that NTTIS sampled a smaller proportion of the national truk population than. Whenever possible, NTTIS data elements were designed to be ompatible with in order to failitate omparison between the two. This setion will ompare truk population and travel estimates derived from NTTIS with those from the two surveys.

2 Massie et al. Truk Population Estimates of the registered large-truk population in the ontinental United States by power unit type an be derived from both and. The number of straight truks is estimated at 2,534,973 by 1982 ; 2,185,63 by ; and 3,23,21 by. The estimate is about 14 perent lower than 1982 and 32 perent lower than. The number of truk-trators is 9,884 aording to 1982 ; 919,72 aording to ; and 1,38,13 aording to. estimates about 2 perent more trators than 1982 and about 11 perent fewer trators than. At least three fators affet the degree of orrespondene among the estimates from the three files. One is that the samples were drawn from three different registration years. Generally one would expet a small inrease in the number of registered truks from year to year, assuming favorable eonomi onditions. The other two fators are more omplex and will be disussed in the next few paragraphs. One onerns identifying medium- and heavy-duty truks in the data, and the other involves the time gap in between drawing the sample and onduting the survey. Large Truks and GVWR From the outset, the survey was restrited to mediumand heavy-duty truks, those with a GVWR over 1, lb. In ontrast, samples all truks, inluding light truks. GVWR is enoded in the Vehile Identifiation Number (VIN) for almost all truks manufatured after 198. R. L. Polk has developed deoding algorithms to extrat this information from the VIN, and this ode was inluded in the data supplied for the survey. The Polk-derived GVWR is also inluded in the 1982 file but not in the version. The VINs of some truks, partiularly those from model years before 1981, do not diretly ontain the GVWR. For many of these ases, the Polk-derived GVWR is based on the truk model as derived from the VIN, with the highest GVWR available for that model (as an option, for example) assigned. For many speifi models, the majority of sales are at lower GVWRs. To improve the auray of the 1,-lb GVWR utoff when the sample was drawn, UMTRI speified whether partiular models should be inluded or exluded, in some ases overriding the Polk-derived GVWR. Models and series were identified for inlusion or exlusion based on sales information provided by the manufaturers. If the manufaturers indiated that the majority of sales were at a GVWR of 1, lb or less, then all of that speifi model and series were exluded. The objetive was to prevent the inlusion of an entire series when only a small fration was atually rated over 1, lb. The models most influened by this proedure were small step vans and pik-up truk models sold as a ab and hassis. The latter often have a flatbed or stake body added. To further ensure auray, GVWR was onfirmed with the owner during the implementation phase of. Restriting the sample to truks with GVWRs of more than 1, lb was not an issue for beause light truks are inluded in that survey. The Polk GVWR an be used to identify large truks in the 1982 file, but for the reasons just stated some light truks probably reeive a Polk-derived GVWR over 1, lb. This would inrease population estimates of medium-duty truks, primarily straight truks. The situation is worse for the file beause that version does not inlude a GVWR variable. The file ontains an average gross vehile weight (GVW) variable based on the owner's estimate of the average weight of the vehile when arrying a typial payload. GVW is only loosely related to GVWR, however, and rejeting all ases with average GVW below 1,1 lb would result in the exlusion of many medium-duty truks. The population estimates presented in this paper exlude all vehiles identified as a pik-up, van, minivan, utility vehile, or station wagon on truk hassis. In addition, a vehile was exluded if the empty ombination weight was 6, lb or less and the power unit was oded as having only four tires. This should ensure that only light-duty vehiles are exluded from the analysis. However, it is likely that not all light truks in were exluded, thus inflating population estimates. Medium-duty straight truks are the vehiles most likely to be overstated. To summariz to this point, the diffiulty of aurately identifying large truks in data probably results in inflated estimates of straight truks ompared with. The problem should be less severe for the 1982 file, beause it ontains a Polk-derived GVWR variable that should be only slightly less aurate than the GVWR determinations employed by. The straight truk estimates are undoubtedly more affeted beause that file does not ontain a GVWR variable. The GVWR problem is not thought to seriously affet population estimates of trators in either file. NTTIS Time Gap The third major fator affeting vehile population estimates between and onerns a time delay in between drawing the sample and implementing the survey. The sample was based on the July 1,.1983, R. L. Polk files, but the implementation phase was not onduted until January through May of Vehiles that were junked or srapped in the interim were removed from the sample, and there was no opportunity to replae them with vehiles that were purhased during that time. This means that vehile ounts are low by about a model-year lass and a half-those truks bought during 1984 and the seond half of In the ase of, the sample is drawn from registrations as of July 1 in a partiular year, and survey forms are mailed out over several months of the following year. However, if a vehile has been junked or srapped in the meantime, it is still inluded in the survey. Thus population estimates refer to the date of the registration lists on whih the sample was based, with no loss of ases. Other things being equal, population estimates should ome loser to approximating the entire registered truk population on a given date. Reoniling Population Estimate Differenes It is possible to adjust vehile ount estimates to aount for the year and a half of missed model years. Distribu- 51

3 52 tions of 1982 and vehile ounts by power unit type and model year were examined to see what perent the newest year and a half of model years represent in those two files. Beause samples were based on Polk vehile lists made halfway through a alendar year, truks of the newest model year represent about half of a model-year lass in. The next most reent model year should represent a full modelyear lass. The newest model year and a half of straight truks represent 4.6 perent of straight truks in 1982 and 8.8 perent in. It is impossible to say exatly what perent the missed straight truks in NTTIS represented of the entire straight-truk population when that survey was onduted. The size of model-year lasses varies from year to year, as the two perentages illustrate, beause of eonomi onditions and other fators. However, using the perentages to estimate a range of missed straight truks results in an adjusted NTTIS population estimate of 2,291,17 to 2,395,678 vehiles. This is still 5 to 1 perent below the 1982 estimate and 26 to 29 perent below the estimate. Considering that the three surveys were onduted in different years, that it is problemati to identify large truks in (espeially the ), and that the adjustment is a rough estimate, the agreement among surveys is not bad. The newest model year and a half represents 1.5 perent of trators in 1982 and 13.1 perent in. This results in NTTIS adjusted trator ounts of 1,27,6 to. 1,58,755 vehiles. These estimates are 14 to 18 perent above 1982 trator ounts and between 1 perent below and 2 perent above trator ounts. This is also a fairly good agreement, and the adjustment to NTTIS may in fat be higher than is appropriate beause of the variation in modelyear lass sizes. File Comparisons Leaving aside the question of absolute vehile population estimates, NTTIS and will be ompared based on the distribution of several variables desribing the large-truk population of eah. Both surveys were designed to desribe the U.S. registered truk population. Agreement between the two would indiate that they are haraterizing the same basi population of vehiles. Although the surveys were onduted in different years, many aspets of the large-truk population hange slowly enough that general agreement should be expeted if both surveys are representative of the U.S. truk population. GVWR provides a good basis of omparison between NTTIS and 1982 beause GVWR was inluded in the original sample data provided by R. L. Polk to both surveys. Most of the other information olleted by the two surveys ame from respondents and is therefore subjet to respondent error. A omparison of the distributions of the national truk population by GVWR from 1982 and NTTIS is shown for straight truks in Figure 1 and for trators in Figure 2. In general, the agreement is good, espeially for trators. The main differene is a somewhat higher proportion of GVWR Class 3 to 5 (1,1 to 19,5 lb) straight truks in ompared with NTTIS, possibly a result of mislassifiations in the Polk-derived GVWRs used in the file ::; a Q) TRANSPORTATION RESEARCH RECORD 147 NlTIS ITIJE] Unk GWJRClass FIGURE 1 Straight truks by GVWR in NTTIS and NlTIS a e Q) o ll=:l GWJRClass Unk FIGURE 2 Trators by GVWR in NTTIS and Compared in Figure 3 is the distribution of ab style for trators only in NTTIS and This information is obtained from the survey respondent, and the ategories used were abover, short onventional, medium onventional, and long onventional. (Conventional abs were not subdivided in the, so no distribution is inluded.) The agreement between 1982 and NTTIS is very good. This is partiularly gratifying in view of the lak of a preise definition of what onstitutes a short, medium, or long onventional ab. The last omparison presented here is arrier type for trators only, shown in Figure 4. Again, this information is supplied by the respondent in both surveys. Carrier types are defined aording to whether the ompany operates interstate or intrastate and whether it is private or for hire. Private arriers operate lose to 5 perent of the trators in both of the files, and about 53 perent in NTTIS. In NTTIS, a further breakdown of private arriers is made into interstate and intrastate arriers (not shown in Figure 4). Interstate private arriers operate 32.5 perent of all trators and intrastate 19.9 perent in NTTIS. The remainder of the vehiles in NTTIS and are for hire in one way or another. For-

4 Massie et al. 53 'E :5 a. Cll Cll E :l lower proportion of intrastate for-hire arriers than either file. NTTIS shows relatively fewer daily rental truks as well. The owners in both of these ategories are usually small arriers and diffiult to reah exept at night and on weekends. These response problems may be partly responsible for the smaller proportion of truks operated by intrastate for-hire arriers or in daily rental in NTTIS. Overall, however, the agreement between NTTIS and on arrier type is quite good. Truk Travel Cat)Qver M. Conv. Oth/Unk S. Conv. L. Conv. Cab Style FIGURE 3 Trators by ab style in NTTIS and :5 a a; ' CJJJ NTTIS ' o+--=...--1'""""'..,_""'1;;;.j..._,_ """""""-...-=---1 Private ICC-Exmt Intra-Hire Rental ICC-Auth ICC-Unk For-Hire Unknown Carrier Type FIGURE 4 Trators by arrier type in NTTIS and 1982 and. Self-Reported Average Annual Mileage Comparisons NTTIS estimated average annual travel of truks in three ways: owners' estimates, odometer readings, and mapped mileage from survey alls (1). relies only on estimates from respondents, so NTTIS owner estimates will be used to ompare average annual mileage between the two surveys. Both surveys asked owners essentially the same question about how far their truk is driven in a year. Comparisons are based on average annual mileage per vehile rather than total miles logged by the entire registered large truk population so that the different vehile population estimates produed by NTTIS and will not affet tile evaluation of mileage estimates. As shown in Figure 5, the overall agreement in ownerreported average annual travel between the surveys is quite good. The NTTIS straight truk figure is about 18 perent higher than 1982 and about 13 perent higher than. The estimates for trators are loser, with NTTIS 4 perent higher than 1982 and 2 perent lower than. It is interesting to note that there is a higher degree of orrespondene between the files for trators than for straight truks. This may be related to the inlusion of some light truks in the straight-truk estimates. Light truks would be expeted to travel less in a year, thus lowering the straight truk average. hire vehiles are further subdivided in both NTTIS and into interstate for-hire, in whih ase they are subjet to Interstate Commere Commission regulations, and intrastate for-hire, where they are governed by state publi servie ommission regulations. Interstate for-hire vehiles are also separated into authorized arriers-the ommon and ontrat arriers-and those hauling exempt ommodities. The small group of unknown ICC-regulated arriers in 1982 are those instanes in whih respondents did not speify whether they were authorized or exempt arriers. If these ases were distributed between authorized and exempt arriers, it would bring the 1982 survey into fairly good agreement with NTTIS. A ategory of just "for-hire" arriers is inluded for the file. These are ases in whih the respondent indiated that the ompany was for hire but did not speify whether it was subjet to ICC regulations. The "for-hire" ases would be distributed among the ICC-authorized, ICCexempt, and intrastate for-hire ategories. This redistribution of ases would probably result in NTTIS having a slightly w ' [[] m _ '87 t:.. g, :J o. Straight Power Unit Type Trator FIGURE 5 Owner-reported average annual mileage by power unit type in NTTIS and 1982 and.

5 54 Total Annual Mileage Total mileage estimates by power unit type may also be ompared between NTTIS and. Earlier the degree of underounting of vehiles in NTTIS beause of missed model years was estimated. The orresponding lost travel may be alulated in a similar manner. Straight truks of the newest model year and a half represent 1.3 perent of the total mileage of 1982 straight truks and 15.7 perent of the straight truk mileage in. Making the orresponding adjustment to NTTIS raises NTTIS straight-truk mileage from 26,7 million mi to the range of 29,75 to 31,672 million mi (Figure 6). This plaes NTTIS straight-truk mileage estimates between the 1982 and estimates, as would be expeted beause NTTIS was intermediate in time between the two surveys. The adjusted NTTIS estimates are 11 to 18 perent above the 1982 estimate and 11 to 17 perent below the estimate. The newest model year and a half aount for 16.3 perent bf total trator mileage in 1982 and 19.3 perent in. This adjustment raises the NTTIS trator mileage estimate from 49,921 million mi to the range of 59,632 to 61,879 million mi (Figure 7). This plaes estimated trator mileage in NTTIS 25 to 3 perent above 1982 and 2 to 6 perent above. The adjusted NTTIS mileage is higher than expeted, possibly beause the missed model years in NTTIS represented a lower proportion than was alulated using the files. Disussion of NTTIS and Estimates of national truk population and travel from NTTIS were ompared with 1982 and. The omparisons overed power unit type, GVWR lass, ab style, arrier type, and owner-reported annual mileage. Overall, there is a good orrespondene between the two surveys. Some of the differenes observed may be beause of the different years of registration files from whih the samples were drawn, the 18- month period between the sample year and the survey in NTTIS, and the probable lassifiation of some straight truks 4, , 1982 Data Soure 35,738 FIGURE 6 Total annual mileage for straight truks in NTTIS and 1982 and. TRANSPORTATION RESEARCH RECORD 147 8,ooo , 6, (/) : Q) g> 4, -- _Q! ]j I- 2, Data Soure FIGURE 7 Total annual mileage for trators in NTTIS and 1982 and. with GVWRs below 1, lb as Class 3 or higher in. Aside from these known disrepanies, there is no indiation of systemati differenes between NTTIS and. COMPARISONS WITH FHWA HIGHWAY STATISTICS Eah year the Federal Highway Administration (FHWA) publishes Highway Statistis, a tabulation of national transportation statistis based on data submitted by the states. Highway Statistis ategorizes travel for different lasses of vehiles on different types of roads. This setion ompares national esti.mates of the number of registered large truks and their annual mileage from Highway Statistis with NTTIS and estimates. Data Soures Highway Statistis ategorizes large truks as single units and ombination vehiles. Single units essentially inlude straight truks alone, straight truks hauling utility trailers, and bobtails (trators without a trailer). Combinations inlude trators hauling one.or more trailers, as well as straight truks hauling full trailers. Highway Statistis is published annually, and estimates from one year are revised in the following year's edition. The data ited here ome from the 1986 and 1988 editions of Highway Statistis, Table VM-1 (2), representing the revised estimates for the 1985 and large truk populations respetively. Highway Statistis 1985 will be ompared with NTTIS, and Highway Statistis will be ompared with. Numbers. for single units were not available for 1982 Highway Statistis, so no omparisons will be made with The Highway Statistis data inlude government-owned vehiles and vehiles registered in Alaska and Hawaii. These vehiles should be exluded for purposes of omparison with NTTIS and estimates. Beause the published Highway Statistis data for truks do not indiate the perentage of government vehiles or the distribution of vehiles by state,

6 Massie et al. 55 estimates were made using other soures of information. The Alaska and Hawaii adjustments for vehile ounts were made based on the state distribution in. The Alaska and Hawaii travel estimate adjustments relied on several years of raw and adjusted state-reported mileage figures submitted to FHW A (3, 4). It was more diffiult to estimate the perentage of government-owned vehiles beause they are not inluded in or NTTIS. The vehile ount adjustments for government truks were made based on an UMTRI data base of large truks involved in fatal aidents (5), and the mileage adjustments took into aount figures ited by Mingo (4). The NTTIS vehile ount and mileage estimates used for the omparisons are based on the adjusted figures that aount for the missed model year and a half of truks. The midpoint of the adjusted range was used in eah instane. Estimates were produed following Highway Statistis' single-unit and ombination vehile lassifiation system. NTTIS mileage figures are based on owner-reported estimates. The data were also made onsistent with the Highway Statistis lassifiation system, but this was slightly more diffiult beause produes no estimates for bobtails. Adjustments were made using onfiguration distributions from NTTIS. Vehile Count and Mileage Comparisons Vehile ount estimates of single-unit truks are 2,367 million for NTTIS, 3,79 million for 1985 Highway Statistis (HS), 3,26 million for and 3,668 million for HS. As noted earlier, the straight truk estimate is believed to be too high beause of the inadvertent inlusion of light truks. Given this, it is signifiant to observe that both HS estimates are even higher than the figure. HS estimates more than 14 perent more single-unit truks than for. Vehile ount estimates for ombination vehiles are 1,19 million for NTTIS; 1,393 million for 1985 HS; 1,62 for ; and 1,49 million for HS. For 1985, HS is about 37 perent higher than NTTIS, and for, HS is 33 perent higher than. These estimates suggest good agreement between NTTIS and and a substantial overestimation by Highway Statistis. Total annual mileage estimates are shown in Figures 8 and 9. There is onsiderable variation in the single-unit travel estimates, with NTTIS showing 29.5 billion mi and HS estimating 48.3 billion mi of travel (Figure 8). The HS 1985 estimate is 56 perent higher than NTTIS, and the HS estimate is 37 perent higher than TUIS. The situation is similar for ombination travel (Figure 9). HS 1985 estimates 28 perent more mi than NTTIS, whereas HS estimates 45 perent more travel than. Disussion of Highway Statistis Estimates The vehile ount and travel estimates published in Highway Statistis are based on data provided by the states. The aggregate statistis are alulated by FHW A using proedures that are intended to provide omparability of values among states. In a reent disussion of Highway Statistis large-truk travel estimates, Mingo ( 4) ited several indiations that the 6,o ?-?.... ' 45,84 I 4, (/) ]i r=. 2,.... 1, FHWA Soure of Data FHWA FIGURE 8 Total annual mileage for single-unit vehiles in NTTIS, 1985 and Highway Statistis, and. (j) (/) 1,fh ,99 8, 4,.... I- 2, FHWA Soure of Data 85,522 FHWA FIGURE 9 Total annual mileage for ombination vehiles in NTTIS, 1985 and Highway Statistis, and. estimates are too high. Mileage data submitted by states are based on traffi ounts of 13 vehile lasses on seleted segments of 12 types of roads. Most states use manual and automati vehile ounting proedures, both of whih are problemati. Human error in manual ounting often results in the mislassifiation of vehile types. With automati lassifiation, detetor defiienies an result in losely spaed separate vehiles being ounted as a single ombination vehile or in the unintended ounting of vehiles in adjaent lanes. Beause large truks represent a small proportion of vehiles overall, ounting errors an lead to large perentage errors in vehile lass estimates, espeially if there is a systemati bias in the rriilassifiations. Aside from these problems, states do not all employ the same vehile type lassifiation system. A partiular diffiulty is straight truks with trailers, whih, depending on state and trailer type, may be lassified as either single-unit or ombination vehiles. Another major soure of error is that most states ount truks only on weekdays. Generally no orretion is made for the fat that truk travel is heavier on weekdays than week-

7 56 ends. Compounding the problem is the fat that ounting sites frequently our on routes with a large volume of heavy truks. In addition to these methodologial problems, Mingo desribed other inauraies and inonsistenies in state reporting proedures. State estimates in various travel ategories have a low level of preision, with mileage figures sometimes reported with only a single signifiant digit. In most of the states, vehile-type lassifiations are entirely omitted for at least some of the road-type breakdowns. Mingo observed many instanes of tremendous annual variation in travel estimates within states, inluding one state that reported an annual inrease of more than 5 perent in ombination travel. FHW A attempts to ompensate for some of the problems in the state data by adjusting the estimates. For example, a itation on Table VM-1, 1988 Highway Statistis, indiates that the "stratifiation of the truk figures is based on the 1982 Truk Inventory and Use Survey ()." The problem of making these adjustments is ompounded beause the more reent data did not beome available until nearly January The authors annot evaluate the FHW A adjustment proedures beause they have not had the opportunity to review them. Mingo onludes that FHW A's efforts to orret state-reported data ontribute to an overestimation of large-truk travel. The point here is that the Highway Statistis figures systematially overestimate large-truk travel. This is a matter of onern beause Highway Statistis figures are widely used, both in virtually all FHW A studies requiring truk travel data and in many other studies as well. The following example illustrates the relevane of aurate travel information to traffi safety studies. Sine 198 UMTRI has onduted the Truks Involved in Fatal Aidents (TIFA) survey. The survey ombines information from Fatal Aident Reporting System (FARS) ases, Offie of Motor Carriers aident reports, and telephone interviews to produe a file of detailed desriptions of all large truks in the ontinental United States involved in fatal aidents. In Figure 1 the annual number of fatal involvements of ombination vehiles has been plotted for the 7 years from 1982 through 1988 (5). The frequeny of fatal involvements has remained relatively stable over this period, with a low of 3,376 in 1982 and a high of 3,762 in On the same graph, the original Highway Statistis estimates of the total mileage of ombination vehiles for eah year have been plotted (2). Highway Statistis mileage estimates have risen every year. The 9,149 million mi estimated for 1988 represent a nearly 5 perent inrease over the 6,31 million mi estimated for The ombination of the substantial inreases in estimated travel and the omparatively steady number of fatal involvements results in a sharply delining fatality rate. This is also plotted in Figure 1, against the y-axis on the right edge of the graph. Aording to the Highway Statistis numbers, the fatal involvement rate of ombination vehiles/1 million mi of travel has delined from 5.6 in 1982 to 4.12 in, a drop of 26 perent. Although suh a dramati derease in the fatal involvement rate would be enouraging news, it is possible that muh of this trend is an artifat of systemati error in the Highway Statistis travel estimates. It is reasonable to believe that largetruk travel has inreased from year to year, with the overall expansion of the eonomy. However, estimates only a 1, 9, f' 8, TRANSPORTATION RESEARCH RECORD 147 Total Mileage (FHWA) Q5 7, > 6,.! 5, CD E CD > 4, >.!:: 3, LL 2, 1, Fatal Involvements (TIFA) Fatal Involvement Rates Year FIGURE 1 Fatal involvement rate of ombination vehiles, r=--.4 > t-.3 i VJ.2 23 perent rise in trator travel from 1982 to, whereas Highway Statistis estimates a 43 perent inrease in ombination vehile travel during the same time span. Furthermore, Highway Statistis' 1982 figure was only 27 perent higher than the 1982 estimate, whereas in the Highway Statistis figure was 47 perent above. This suggests that umulative error in the Highway Statistis large-truk travel estimates inreases the amount of overestimation over time. If the Highway Statistis mileage figures are too high, then fatal involvement rates based on those figures will be too low. CONCLUSIONS Auray of large-truk travel estimates is learly an important issue. Evaluating the safety of partiular lasses of vehiles requires information on both the number of aidents they experiene and how many miles they aumulate, so that aident rates per mile of travel may be alulated and ompared with other kinds of vehiles. Travel estimates that are too high will produe aident rates that are too low. Compared in this paper are large-truk travel estimates from three soures. The omparisons are not as straightforward as desired beause of the different times the data were olleted and the different methodologies used by eah soure. However, the overall onlusion is that estimates produed by and NTTIS show muh loser agreement to eah other than either survey does to estimates published in Highway Statistis. Ideally, more nationally representative surveys of large-truk travel will be onduted in the oming years so that their results may be inluded in similar travel omparisons. With more independent studies, the auray of Highway Statistis estimates an be better evaluated. ACKNOWLEDGMENT This analysis of the National Truk Trip Information Survey was supported by a grant from the U.S. Department of Transportation, University Transportation Centers Program..1 * C1l (ii LL

8 Massie et al. REFERENCES 1. D. L. Massie, K. L. Campbell, and D. F. Blower. Large-Truk Travel Estimates from the National Truk Trip Information Survey. In Transportation Researh Reord 147, TRB, National Researh Counil, Washington, D.C., 1993 (in prodution). 2. Highway Statistis ( editions). Offie of Highway Information, Federal Highway Administration, U.S. Department of Transportation, R. D. Mingo. Safety of Multi-Unit Combination Vehiles. Intermodal Poliy Division, Assoiation of Amerian Railroads, R. D. Mingo. Evaluation of FHWA's Vehile Miles of Travel Estimates for Heavy Vehiles. Intermodal Poliy Division, Assoiation of Amerian Railroads, D. F. Blower. Truks Involved in Fatal Aidents, , by Power Unit Type. Report UMTRI University of Mihigan Transportation Researh Institute, Ann Arbor, Publiation of this paper sponsored by Committee on Traffi Reords and Aident Analysis. 57