U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
|This report is an archived publication and may contain dated technical, contact, and link information|
Publication Number: FHWA-RD-96-143
Date: April 1997
Development of Human Factors Guidelines for Advanced Traveler Information Systems and Commercial Vehicle Operations: Definition and Prioritization of Research Studies
CHAPTER 4. EXPERIMENT 3
ATIS will provide a broad range of support to drivers. Most concepts of ATIS for private vehicles include navigation aids, safety systems, traveler's aid services, and communications. CVO also includes administration, tracking, and management functions. Regardless of the exact functional composition of ATIS, there are overriding concerns about whether such a system will be accepted and purchased by drivers of private vehicles and whether commercial drivers will accept or reject systems installed by fleet owners. As documented in our earlier literature review (Kantowitz, Becker, & Barlow, 1993), user acceptance of new technology is a complex, multi–faceted problem but one which may be tractable to experimental analysis. In the study reported here, our goal has been to examine a variety of methodologies and analytic techniques that may prove useful in assessing user acceptance of ATIS/CVO functions.
The current study addressed CVO function acceptance issues, independent of implementation. Both local and long–haul commercial drivers served as participants in the study. In this study, only paper and pencil questionnaires were used, coupled with verbal explanations and examples of function application. The current study also used a direct magnitude estimation task, a psychophysical forced–choice analysis, and a relatively new link–weighted network analysis (Schvaneveldt, 1990).
Throughout the study, the participants were asked to assess the ATIS functions for their job–related value. The variations between this study and the prior studies increase the range of methodological alternatives considered under this task.
The link–weighted network analysis is an attempt to apply the emerging technology of knowledge engineering to the task of understanding user acceptance issues. The network analysis can yield detailed structures for the concepts under investigation leading to greater specificity in data interpretation, alternative hypotheses, and perhaps even conclusions. The Pathfinder algorithm (Schvaneveldt, 1990) for network analysis has been chosen for use in this study for two reasons. First, it is well founded in mathematical graph theory providing a form of representation that is shared with many system engineering disciplines. Second, Pathfinder was developed for the purpose of more explicitly representing the structures of human memory and the contents of mental models. As such, Pathfinder is also well founded in psychological measurement. Earlier studies using this analysis have identified differences in the networks produced by Air Force instructor pilots, by pilot trainees, and by current fighter pilots (Schvaneveldt et al., 1985), as well as network differences between users of a documentation preparation system and the model used to define the system (Kellog & Breen, 1991). In using Pathfinder, we will attempt to identify how local and long–haul drivers evaluate ATIS functions as job performance aids and whether there are differences between the types of drivers.
Sixty–five commercial truck drivers were recruited from the Sacramento area by the California Trucking Association (CTA). Participants signed up for one of four 4–h group meetings conducted at the CTA building in Sacramento on the weekend of November 20–21, 1993. The four groups ranged in size from 15 to 17. Participants were paid $50.00 for their participation. Fifteen participants were eliminated from the study because of missing data or a failure to follow instructions. Table 56 summarizes the demographic data for the 50 participants whose data were included in the analyses.
Table 56. Demographic data for local and long–haul drivers.
Materials and Procedure
A 25–page booklet of concept definitions, ATIS/CVO explanations, task instructions, and answer sheets was prepared specifically for use in this study (see appendix F). The booklet contained five sections. The first section included an introduction to ATIS/CVO as a concept, and definitions of 16 ATIS functions. That section served as the primary training on ATIS systems, functions, and usefulness. The second section of the booklet contained the instructions and answer sheets for a direct magnitude estimation task. The third component contained additional training materials that highlighted the possible interactions among the 16 functions. The fourth section included the instructions and answer sheets for a forced–choice paired comparison task. The final section was a short demographic questionnaire.
At the beginning of each session an introductory explanation of the study was given, and the participants were asked to read and sign a consent form. The booklets were distributed, and the participants were instructed to read the first five pages. The first page included a general characterization of ATIS, attempted to set the context for the drivers' evaluations of the ATIS components, and reiterated some of the points on confidentiality from the consent form. The next four pages contained brief definitions of 16 ATIS functions. When all participants finished reading these materials, a focused discussion addressed any questions that they raised, and attempted to elicit their reactions and comments. During the discussions, the experimenter tried to maintain a focus on the definitions of the 16 individual functions, provided additional information as required to answer questions, and attempted to engage participants in the discussion. In three of the groups, at least half of the participants actively discussed the functions. Participants contributed specific experiences in which an ATIS function might have proved useful, offered comments on the uselessness of some functions, and considered how implementation details might increase or decrease the usefulness of a given function. In the fourth group, about one–third to one–half of the participants were contributors to the discussion with two participants trying to focus the discussion on their concerns. The discussion typically lasted about an hour. Some of the participants' comments are incorporated into the results and discussion sections.
The list of functions published in this experiment was culled from the ATIS capabilities described in a previous report (Lee et al., 1997) to create a smaller set of functions more amenable to this study. Specifically, similar capabilities were merged into single functions (e.g., in–vehicle signage was separated based on four types of information in the previous report, but merged into a single function here). Capabilities that could not stand alone were combined to form a single function (e.g., trip planning and dynamic route selection presuppose a navigation capability; so, these were combined into a single route navigation function). These changes produced 15 of the 16 functions. The final function, Vehicle Location Update, was added to the set for two reasons. First, it appeared as an enhanced implementation component for several of the other functions. Second, it is already available and implemented in commercial vehicle operations and, therefore, may serve as an anchor point for some of the estimation tasks.
The following 16 functions were used in this experiment:
Following the discussion, participants performed a direct magnitude estimation task in which they rated each function against Vehicle Location Update as the standard. The basis of the rating was "the value to you in performing your job as a commercial vehicle operator." Participants were encouraged to refer back to the function definitions during the task. The complete written instructions for this task are given in appendix F (p. 304). When this first task was completed, the group went on to a second magnitude estimation task which used Vehicle/Cargo Condition Monitoring as the standard for comparison. Again, the instructions, given in appendix F (p. 306), emphasized rating the functions on their value for job performance. The participants retained their ratings from the first task while doing the second task, but they were folded into the rest of the materials. Participants were instructed not to refer back to their first ratings when making their second rating. After completing the second rating task, participants were given a 15 to 20 min break.
To begin the second half of the session, participants were instructed to read pages 10 to 13 in the booklet which described how the various functions interact with each other and how they could be combined to provide better total capabilities. These materials described four "option packages" that were designed to provide different services for drivers. The option packages were oriented toward Driver Safety, Driver Services, Management Services, and Navigation support. Because the Vehicle Location Update function interacts with functions in each of the option packages, this function was included in all of the packages. The group discussed the new information, commenting on topics like work overload caused by too many system components, the diverting of attention to process potentially unimportant information, the ease with which law enforcement officials could generate speeding tickets, and the fact that your dispatcher would be able to figure out how much time you spent in Winnemucca. At the end of the discussion, participants were instructed to rate the job–related value of the four ATIS option packages against a standard package that included all 16 functions.
The next task in the survey was a forced–choice, paired–comparison task which required a participant to respond to each of the 120 unique function pairings. Each of 10 answer sheets contained 12 pairs. The order of functions within a pair was randomly determined. Ten different orders of the answer sheets were used. The goal of this task was to generate scaled differences between the members of all possible pairs of ATIS functions for use in a link–weighted network analysis. Therefore, participants were asked to do more than simply make a forced choice of the function most valuable on the job. The instructions asked the participant to assign the number 100 to the member of a pair of functions that they deemed to be most valuable in doing their job. Then, they were instructed to assign a number between 1 and 99 to the other member of the pair. The second number was to reflect how much less valuable the other member of the pair was. For example, for the function pair Vehicle Location Update <–> Services Directory, one participant assigned 100 to Vehicle Location Update and 50 to Services Directory indicating that an in–vehicle "yellow pages" was about half as useful on the job as the vehicle locator function. After participants read the instructions, the experimenter reviewed them and answered any questions. Participants were told to complete this task and then fill in the demographic questionnaire. This completed their participation. Participants were paid at the end of the session.
As stated above, 15 participants were eliminated from the study. Three participants omitted one or more responses in one or more of the tasks. Six participants produced ties in the forced–choice task assigning the same number to both members of at least one pair. Six participants assigned absolute ratings in the forced–choice task (e.g., 40 to one member and 50 to the other member of a pair) instead of following the relative rating instructions. Aberrant responses in the main tasks of this study were assumed to reflect a misunderstanding of the instructions for generating and assigning numerical values to functions in the various tasks. This was assumed to reflect other misunderstandings, as well, and hence no attempt was made to repair the data. Nine of the eliminated participants were local drivers, and six were long–haul drivers. One additional participant missed the first magnitude estimation answer sheet entirely. Instead of dropping this participant from the analyses, we will present a formal analysis of only the second direct magnitude estimation task. These errors were detected only at the end of the sessions, at the earliest. We chose not to have participants make corrections.
Magnitude Estimation Data
In the direct magnitude estimation task, each participant rated the ATIS functions twice, against different standards, once against Vehicle Location Update, and once against Vehicle/Cargo Condition Monitoring. These two functions were chosen as the standards because they are already in use in early forms. Several existing systems provide vehicle tracking services, and a variety of devices currently monitor such things as brake air pressure and trailer refrigeration. A preliminary inspection of the data from this task showed no differences between the first and the second ratings. The findings from the second rating task are reported.
The data from each individual participant was standardized using a z–score transform to minimize the skewing effects of a rating scale that was bounded only on one end. These data are shown in table 57. A Sheffé test was used for each participant group to identify any differences among the functions. For the long–haul drivers, the small number of participants (N = 12) produced a large minimum significant difference; hence, only two groups of functions were detectable with considerable overlap between them. In general, though, safety functions appeared at the top of the list, navigation and communications functions filled out the top half of the list, control and administrative functions were next, and driver services functions were at the bottom of the list. For the sample of local drivers, three function groups were detected. Again, safety–related functions were at the top of the list, followed by communications, control, navigation, administrative, and finally driver services functions. Across the two groups of drivers, there are some differences in the rank–ordering of the functions. We will focus on these differences later in our analyses.
Option Package Magnitude Estimation Data
The main purpose of this section of the materials was to introduce the ways in which the 16 individual functions interact to enhance the overall ATIS capabilities. Four combinations of functions, or option packages, were described and then discussed. Participants then performed a magnitude estimation of the job–related value of each package compared to a complete package containing all functions. The means of the raw ratings are shown in table 58.
Table 58. Mean ratings of option packages.
The Complete Package was assigned a standard value of 100 by the experimenter. One participant in the long–haul driver group changed that value, and the change was entered into the data. In these data, both local and long–haul drivers rated the Driver Safety and the Navigation Packages as more valuable than the Complete Package. Both groups rated the Management Package below the Complete Package but disagreed on the relative value of the Driver Services Package. The definitions of the option packages can be found on pages 308–310 of appendix F.
Paired Comparison Preference Data
From the paired–comparison task, we extracted the proportion of times each function was preferred over other functions for each participant. For comparability with the magnitude estimation data, the raw scores were converted to z–scores with the means reported in table 59. Again, a Sheffé test was used to identify significant differences among the means. The ordering of the functions here is somewhat different from that obtained for the magnitude estimation task. For long–haul drivers, the most valued functions now include safety and communications, with lower–values assigned to navigation, management and control, and driver services. For the local drivers, the safety functions still comprise the most valued group, with lower ratings for navigation, communications, management and control, and driver services.
In an overall ANOVA, we examined the effects of FUNCTION, DRIVER GROUPS (local vs. long–haul), and TASK (magnitude estimation vs. paired comparisons). Because of the normalizing, there was no effect of DRIVER GROUP (F < 1). Unfortunately, there was a very small but apparently consistent effect of TASK, F(1,48) = 4.12, p = <0.05, and a significant interaction of TASK with DRIVER GROUP, F(1,48) = 4.40, p = <0.05. The Magnitude Estimation task resulted in very small negative means, –0.0000625 and –0.00125 for local and long–haul drivers, respectively. The Paired–comparison task produced means of zero, as expected. We suspect a small but consistent rounding or truncation error lays behind these significant results, and we are crosschecking the data. There was a significant main effect of FUNCTION, F(15,720) = 15.83, p < 0.0001, supporting the differences discussed above. Finally, there was a marginally significant interaction of TASK with FUNCTION, F(15,720) = 1.56, p = 0.08. The source of the interaction probably lies in the shifts for functions rated near the midpoint of the scales. For instance, in the magnitude estimation task, Vehicle Location Update received a mean z score of –0.405, and in the paired–comparison task , the mean z score was –0.139. In contrast, Fleet Resource Management had a mean z score of –0.197 in the magnitude estimation task and a mean z value of –0.427 in the paired–comparison task. There was no significant three–way interaction (F[15,720] = 1.08), p = > 0.05.
This configuration of statistical results includes the one bothersome finding of significance where no effect was expected. The main effect of TASK does appear to be the result of a truncation error in the manipulation of the magnitude estimation data. Since that truncation occurs in a commercial data analysis package, we do not have the luxury of repairing the problem, nor do we have the luxury of writing our own analysis routine. The marginal interaction of TASK with FUNCTION may be an artifact of the same problem, or it may reflect a more fundamental difference between the rating tasks. In magnitude estimation, the entire suite of functions was considered at the same time; whereas, the paired–comparison task pitted one function against another. The major finding, that there are clear–cut preferences for certain functions, indicates that some proposed ATIS capabilities may be an "easy" sell to commercial drivers while others may have a long road to driver acceptance. These data begin to separate one type from the other.
In the paired–comparison task, participants assigned a value of 100 to the member of the pair which had the greatest on–the–job value. They then generated a second number to indicate the relative on–the–job value of the other member of the pair. The absolute value of the difference between the two numbers is a measure of the distance between the members of a pair in the rating space. The task required participants to rate all 120 possible pairs of ATIS functions resulting in a half–matrix of distance estimates between all possible pairs of the 16 ATIS functions. From the distance data, the Pathfinder algorithm (Schvaneveldt, 1990) produces a mathematical graph showing the edges or links that exist among the ATIS functions.
Depending on the parameter values, the resulting graph can contain all pair–wise links (parameter q = 1) or exactly n–1 links (q = n–1), where n is the number of nodes in the graph (the number of ATIS functions). The q parameter specifies the number of edges or links over which triangle inequalities are resolved. Mathematically, the graphs for each value of the q parameter are equally valid. For our purposes there is a level of graph richness and complexity that is informative somewhere between the two extremes of q = 1 and q = n–1. The second Pathfinder parameter is the Minkowski r–metric which determines how distance is computed between indirectly linked nodes; that is, there are intervening nodes along the path. For a value of r = 1, Pathfinder uses the sum of all edge distances between directly linked nodes along the path between separated nodes. For r = 2, multiple–edge path distances are calculated as Euclidian distances, and for r = infinity, the distance of a multiple–edge path is the maximum of the individual edge values. In generating a given graph, Pathfinder calculates distances based on the r parameter and eliminates possible paths by satisfying triangle inequalities over q edges.
Selecting parameter values for use in this analysis is partly rational, partly based on what others have done, and partly empirical. For the r parameter, the underlying physical space models for the smaller values of r seem too concrete. Our procedures were intended to capture distances in an abstract space for which we do not have objective measures of distance, nor do we have a full understanding of the dimensionality of the space. When r = infinity, distance becomes more abstract, and an appropriate analogy could be that of electrical resistance. A multiple–edge path will be included in the final network if its maximum individual edge distance is less than the maxima of other paths. In other words, a path will be included if it is the path of least resistance among those considered. Therefore, we will assign the r parameter a value of infinity. The q parameter will be used to create a family of graphs that varies from the sparsest graph to the most complex graph. From this family of graphs, we will select the value of q that produces an interpretable level of richness. All values of q greater than or equal to 4 produced the sparsest graph, one that corresponded to the rank ordering derived in the forced–choice analysis. For q = 3, only one or two additional links emerged, and for q = 1, the graphs contained so many links that they appeared visually disorganized. Therefore, the best choice for the q parameter for these data seems to be 2. The Pathfinder analyses below will use parameter values of q = 2 and r = infinity.
Our goal in using the Pathfinder analysis was to provide a numerical solution to organizing the ATIS/CVO functions. What remains is interpreting the resulting networks of functions in the context of ATIS and the judgments of the participants. Figure 79 shows the Pathfinder graph for the 38 local drivers in our sample. The absolute value of the distance between members of a pair was calculated for each participant for each pair and then averaged across participants into a single half–matrix. The network for local drivers includes 24 total pair–wise links using the parameter values q = 2 and r = infinity. The 14 links shown in the figure as thick lines are the strongest links with the shortest distances between functions. These links identify how the concept nodes are connected within cohesive groupings. The 10 thin lines represent greater distances and hence weaker, more tenuous connections. The weaker links indicate connections across the primary groupings. The distinction between shorter and greater distances is the result of the first step in trying to understand the structure of the network. The weaker links were progressively removed from the network until some cohesive grouping emerged. In figure 79, there appear to be three groups, two of which are relatively cohesive and one which is fairly loose. One group includes Road Condition Information, Emergency Aid Request, and Immediate Hazard Warning. These are three safety functions that nearly all drivers rated as valuable in doing their job. Note that two of the concept nodes are not strongly connected to anything.
The second group includes eight functions that are navigation and communications related. The exception is the Vehicle/Cargo Condition Monitoring function. By examining its connections to other functions, we may be able to understand how it is incorporated into the group. The function itself is a combination of sensor systems that signal out–of–tolerance conditions, faults, and failures to the driver. In a local driving context, these error conditions may not be much of a safety hazard, but they do require attention. After isolating the problem, correcting it starts with knowing your exact location, communicating your problem to those in a position to make repair decisions, and then getting to a repair point or getting a repair service to you. Bear in mind that these speculations go far beyond the data in attempting to specify the attributes of the connections in the domain of job–related value. There could be other reasons why these functions are near neighbors.
The final, loosely connected group includes those functions that were rated lowest in job–related value. Included here are functions that seem outside of the day–to–day activities of a local driver. Within a single jurisdiction, the value of the Regulatory Administration function could be almost Non–existent. Fleet Management may be too high a level to have immediate value to the individual driver. Services Directory and Broadcast Services may offer nothing new to a local driver who operates in the same area every day, and Cargo Transfer Scheduling may not enter into a local driver's domain. As one participant put it, "What does any of this matter to me? I just haul rocks."
In addition to assessing the groupings that emerge, it is informative to consider how the groups interconnect. There are three connections between the group of safety functions and the navigation/communications Group. Road Condition Information is connected to Voice and Message Communication. This link may result from the drivers' current practices in using citizen band (CB) radios to tell each other about the disruptions, congestion, and other delays that often go unreported Elsewhere. Emergency Aid Request is connected to Vehicle Location Update. This may reflect information from the training in which these two functions were linked in the description of one of the options packages so that an aid request would automatically include the vehicle's current location. Finally, Immediate Hazard Warning is connected to Route Selection and Guidance. This warning function provides information about hazards within several hundred meters around the vehicle. When a warning is received, the driver may have to unexpectedly deviate from his route and plan his way around the hazard.
The Pathfinder analysis of the data for the 12 long–haul drivers is shown in figure 80. The 15 strongest links are shown as thick lines, and the 4 weakest links are shown as thin lines. The strong links identify how the concept nodes are connected within cohesive groupings. The weaker links indicate connections across the primary groupings. Using the same technique of progressively eliminating the weakest links, three function groups emerge. Again, there is what could be labeled a safety grouping, but for the long–haul drivers, it consists of five functions. Emergency Aid Request, Immediate Hazard Warning, and Road Condition Information are connected with Voice and Message Communication and Vehicle/Cargo Condition Monitoring. This set of five functions appears to provide drivers with complete information about their immediate environment coupled with functions that support both normal and emergency communication.
The second group of functions could be labeled navigation and control with the central functions of Fleet Management and Dispatch Control. The final group includes Services Directory and Broadcast Services which, from some of the comments, drivers seemed to consider as sources of disruptive workload rather than as effective job aids.
Comparing the networks for local and long–haul drivers is not so simple as counting the pair–wise links that occur in both networks. Of the total 34 unique links, eight occur in both networks yielding a similarity measure of 0.235. By chance, this level of similarity would be expected to occur less than one time in one hundred. Thus, superficially, the two networks share some common ground. A closer inspection of the eight shared links, however, shows that four of them are strong links in both networks, and the other four are strong in one network but weak in the other. One of the strong links, Voice and Message Communication with Vehicle/Cargo Condition Monitoring, changes groups across networks. Clearly, there are potentially meaningful differences in the larger structures that are not captured in the simple similarity measure. Techniques for comparing networks for "neighborhood" similarities and structural differences are being investigated (Goldsmith & Davenport, 1990), but those preliminary techniques will not be applied here.
The differences between the local and long–haul drivers' networks lie both in different pair–wise links and in different overall structure. An approach to explicating the differences would be to isolate potential ATIS/CVO usage patterns for local and for long–haul drivers and then try to relate different usage to different structures. For example, the weak link between Emergency Aid Request and Vehicle Location Update in the local driver network does not exist in the long–haul driver network. This may be the result of long–haul drivers' resistance to vehicle tracking systems as they are currently implemented for dispatcher control. Traditionally, long–haul drivers have had the independence to manage the several days drive time between loading and unloading. As one driver put it, "you don't really want your dispatcher to know how long you spent in Winnemucca." The negative reaction to being monitored by vehicle tracking systems may have outweighed any benefit accruing to an Emergency Aid Request that includes the exact vehicle location. For local drivers, the total time for loading, delivering, unloading, and the return trip is usually measured in hours rather than days, and the driver is often in continuous radio contact with the dispatcher. Driver monitoring and control is already integrated into the local drivers' job because of the smaller distances and shorter times involved. Here, the perceived threat of Vehicle Location Update may be much smaller, and some of its value may surface. The exercise of identifying usage patterns is context dependent, and therefore, we will not apply this approach exhaustively. Later, though, we will consider how the differences in network structure could affect a technology introduction and training plan.
There is one difference between the local and long–haul networks that bears mentioning. The local–driver network is somewhat richer in including several additional links. The added complexity may be the result of the greater number of participants who contributed a greater diversity to the data. Alternatively, the job of the local driver may simply be more diverse than that of the long–haul driver. As time and resources allow, we suggest adding more participants to the sample of long–haul drivers.
For commercial drivers, there are differences in the perceived job–related value of the 16 ATIS/CVO functions investigated here. Some functions seem to be consistently highly valued, some are consistently judged to be of little or no help on the job, and the ratings for other functions appear to be changed by the presentation of new information. The network analysis identified some of the ways that the functions interconnect in the rating space of job–related value. With these data, we may be able to make some preliminary suggestions about how to configure ATIS/CVO functions for introduction into commercial vehicles and about how to structure training programs.
In the best of all possible worlds, ATIS/CVO technology would be introduced into commercial vehicles, embraced as a good thing, and used successfully and effectively by all drivers. From past product introductions, we know that this will not happen. Commercial drivers will resist some functions that may be inappropriately introduced, like some of the current vehicle tracking systems. For other functions, like the voice communications currently supported by CB radios, there may be instant acceptance of a new mechanism that improves upon a highly valued function. If we follow the general principles that the highest rated functions stand the best chance of being accepted and that the acceptance of less valued functions can be improved by tying them to more highly valued functions, then the data presented here may provide a useful starting point for identifying some of the usable links between functions.
Safety functions received the highest ratings out of all the measures taken here for both driver groups. Immediate Hazard Warning, Emergency Aid Request, and Road Condition Information are valued functions for both local and long–haul drivers. Therefore, these functions should probably be included in any initial ATIS/CVO release for commercial drivers. From some of the driver comments, the implementation of these functions should be reliable enough so that drivers, for example, do not need to verify road condition information through independent sources. Invalid or stale information will seriously undermine the eventual value of this function. Also, information should be available for any alternative routes that the driver may wish to consider. These drivers pointed out that all drivers approaching traffic congestion would get the same re–routing suggestions which would result in traffic congestion along the alternate route.
Beyond the safety functions, the networks of local and long–haul drivers diverge. Long–haul drivers placed a high value on Voice/Message Communications and on Vehicle/Cargo Condition Monitoring. These functions are in turn linked to Route Navigation, Route Selection And Guidance, and Route Scheduling. In many ways, this set of eight functions meets the needs of a long–haul driver who is making his own decisions and managing his own driving schedule over a time span of several days. The safety functions, vehicle monitoring, and navigation functions support the driver in knowing the status of his vehicle and in handling both normal routing and route deviations. The communications function allows contact with others as needed. A training program for long–haul drivers could emphasize these functions and the independence and personal control that they afford. The remaining ATIS functions provide external control over the driver. Company control, customer coordination, and regulatory control could be cast as those necessary evils of commercial driving most of which are buffered through the Dispatch Control function. The Broadcast Services and Services Directory functions seem to have little place in commercial driving. On this point, both the local and long–haul drivers' networks agree. Even after discussions about filtering out unwanted information, the drivers' comments addressed concerns about the workload induced by these two functions and about the likelihood of distraction from their primary task of driving.
As stated above, the network extracted for local drivers is more complex than that for long–haul drivers. Each safety function has one relatively weak link to a function in the navigation and control group. A training program designed around this type of network must accommodate the variability, or focus on a single link. We could focus on the link that connects the functions closest to the center of their respective groups. In this instance, that link is the one between Emergency Aid Request and Vehicle Location Update. After presenting the safety functions, the enhanced value of adding vehicle location to an aid request could bridge the safety functions and the navigation and control functions. Additional functions can then be presented by selecting a path through the strong links in the sub–network. As Voice/Message Communications and Route Selection and Guidance are encountered, there may be opportunities to emphasize their links to the safety functions and perhaps better tie the navigation and control functions to the more highly valued safety functions. As with the long–haul drivers, there seem to be some functions that have little part in the job of local commercial driving. In addition to the services functions, the local driver network leaves Fleet Management, Cargo Transfer Scheduling, and Regulatory Administration fairly unconnected. As suggested earlier, perhaps because local drivers operate within a single regulatory jurisdiction and because they operate over short distances and short time periods, the higher–level control functions may not directly affect them. We should consider whether to eliminate these functions from a local driver ATIS/CVO system, or at least to move them into the background.
These preliminary recommendations are based on a first look at driver preferences for ATIS functions as they are currently defined. Even so, we suggest the differences between network structures for local and long–haul drivers are meaningful and they imply differences in the structure of technology introduction and training programs. Before proceeding we will need to affirm the differences and consider other important influences. Probably the most important unexamined influence is the preference and requirements of the trucking companies that would purchase ATIS/CVO systems. Although we have no data from the company perspective, we might speculate that a network of ATIS/CVO functions for trucking company managers would differ from the networks generated by drivers. The company view could well emphasize the management and control functions that would help in minimizing costs. If the company takes a narrow view of costs, the functions might be limited to Dispatch Control, Fleet Management, Vehicle Location Update, Regulatory Administration, and, perhaps, Route Scheduling incorporated into dispatching operations. These are the functions that the long–haul drivers rated relatively low in the domain of job–related value. For some companies, the safety function could receive relatively high ratings because, in the long run, unsafe operation is costly to the company. Depending on the details of a network of the company perspective, the safety functions could provide a focus for introducing ATIS/CVO technology to decision makers in trucking companies as well as to drivers. The plan here could start by noting the often hidden costs of driver resistance, suggest that the safety functions stand the best chance of technology acceptance, and then link into the other functions that are perhaps more highly valued by the company.
Much of the preceding discussion is tentative, at best. We have assumed the veracity of the Pathfinder networks and we have speculated about networks for which we have no data. Nevertheless, we believe that if there is merit in the methodology used here, the preceding discussion illustrates how the results can be used.