U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
2023664000
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: FHWARD95176
Date: November 1996 
Development of Human Factors Guidelines for Advanced Traveler Information Systems and Commercial Vehicle Operations: Task Analysis of ATIS/CVO Functions
APPENDIX B. FUNCTION AND SCENARIO SELECTION
FUNCTION SELECTIONSCENARIO SELECTION
FUNCTION SELECTIONThe following paragraphs summarize the procedure and results of each analysis used for function selection, initially for the private vehicle operations and then for the commercial vehicle operations. Functional Characteristics Included in the Private Vehicle Task Analysis Functional Characteristics Included in the Commercial Vehicle Task Analysis
Functional Characteristics Included in the Private Vehicle Task Analysis A casual examination of the functional characteristics suggest that some may be more critical than others for overall system operation. In addition, it is obvious that certain combinations of ATIS functional characteristics may interact more strongly than others. A detailed examination of the information flows ensures that the scenarios used for the task analysis include important functional characteristics and meaningful groups of functional characteristics. This detailed examination shows which functions are most central to ATIS/CVO and which functional characteristics form highly coupled groups, based on the information flows that link them. This analysis depends on identifying the information flows to and from each functional characteristic. The functional characteristics and information flows can be considered as a network, with functional characteristics representing the nodes of the network and information flows representing the links between nodes (see figure 3). This network can also be considered as a matrix of input/output relations. Such a matrix of input/output relations (see table 24) was created to summarize the information flows between functions. To better understand what is meant by input/output relations, it is necessary to define a frame of reference for these inputs/outputs. In this particular analysis, the frame of reference used is the ATIS itself. In other words, an "input" is defined as information and data coming to the function, while an "output" is defined as information and data coming from, or produced by, the function. Thus, the ATIS function acts as a transfer function, transforming information and data from inputs to outputs. In this matrix, an entry was made each time a given function served as either an input or output to another function. In other words, each time there is a "1" in the matrix, it indicates a link between the two functions. The first row and the first column of numbers represent each one of the functions, and the matrix reads as follows: Function 5.1 provides information to functions 5.2, 5.3, 5.6, 6.1, 6.3, 7.1, 7.2, and 8.2. Similarly, the first column reads as follows: Functions 5.7, 6.2, and 8.2 provide information to function 5.1. In other words, functional characteristics listed horizontally receive input from the functional characteristics listed vertically. To make it easier to understand, figure 25 provides a graphical representation of this first row and column, illustrating the functions that are considered inputs and the ones that are considered outputs. Table 24. Information flows between functions for private vehicle operations.
Using this matrix of input/output relations, it is possible to calculate several measures of the importance of a functional characteristic and, as a consequence, determine its level of centrality in the system. Some of the measures calculated include a simple frequency count of the number of interactions between functions, while other measures of centrality are derived from network analysis techniques. The following paragraphs summarize the findings for each one of these analyses. Frequency Count A simple measure of the importance of a functional characteristic is the number of times it either receives from or provides information to other functions. The matrix depicted in table 24 offers the information necessary to calculate this frequency count. The bottom shaded row of table 24 indicates the total number of times a function receives information from other functions. This row shows that destination coordination (6.3) and route navigation (5.6) receive input from more functional characteristics than any other (eight times). Similarly, the right–most column shows the total number of inputs sent to other functional characteristics; for example, road condition information (8.2) provides information to more functional characteristics than any other (nine times). By adding the row and column totals, it is possible to obtain a crude estimate of functional characteristic centrality. Those functional characteristics requiring the most inputs from other functions and providing the most outputs to other functional characteristics represent important nodes in the network of information flows that make up the ATIS. This analysis identifies pre–drive route and destination selection (5.3) and road condition information (8.2) as the two most central functions to an ATIS. The last row of table 24 shows the rank of each functional characteristic. Table 25 summarizes these results, showing the relative centrality of each functional characteristic. Table 25. Rank ordering of the private functional characteristics.
Network Analyses In addition to the simple frequency count, other methods exist to determine the relative centrality of functional characteristics. These measures provide an added level of sophistication that may generate a more accurate estimate of centrality. The frequency count estimates centrality by independently considering the input and output of each functional characteristic. In many cases, this may not be an appropriate approach because a functional characteristic might be more central than another one. For instance, a functional characteristic that has several inputs coming into it has a heavier weight and should be considered more central than a functional characteristic that has only one input. For example, figure 26 shows that the function characteristic "A" has three inputs coming to it, while function characteristic "B" has only one input. In this instance, the output coming from "A" has a heavier weight than the one coming from "B" and should be given more importance when considering its implication in regard to function characteristic "C." To overcome the limits of the frequency count, a network analysis was conducted using the same matrix that generated the frequency counts.
Measure of centrality. A variety of network analysis measures is available to estimate centrality of ATIS functional characteristics. Many of these measures make no distinction between inputs and outputs of a network. Because the network represented in table 24 does make this distinction (i.e., it is asymmetrical), a measure was chosen that accommodates the asymmetry of the matrix. As a consequence, this measure reflects the importance of functions based on information flows in and out of each function. Thus, each function has two measures of centrality, one as an input and one as an output. The input and output measures of centrality can be used to anticipate how people might need to interact with the system. Functions with high levels of centrality, based on their output, represent functions that provide information to other functions. With these functions, the task analysis should focus on how the transfer of information from function to function may be facilitated by minimizing the amount of recoding and memory required of the driver. Conversely, functions that are highly central, based on the number of functions providing input, represent functions that tend to combine information from a variety of sources and then relay it to the driver. Passing information between functions will involve the driver in an entirely different set of tasks, compared to the tasks involved in acting on information provided by the system. Thus, a task analysis of these functions should focus on how the system conveys information to the driver, rather than on how information is passed between functions. In general, functions that are highly connected with other functions may merit special attention in the task analysis, because many aspects of the system may depend on them. Freeman's measure of centrality (1979) estimates the centrality of the functional characteristics based on their output and input information flows. This estimate of centrality reflects the number of adjacent vertices to a given vertex, divided by the maximum possible vertices, and expressed as a percentage (Borgatti, Everett, & Freeman, 1992). For ease of understanding, a numerical example is provided. As an example, five ATIS/CVO functions are considered. As table 24 shows, these functions are linked to other functions, but for simplicity, these links will be ignored for this example. The links between these five functions can be shown graphically, as in figure 27.
The links can also be shown as a matrix, as in table 26.
Using the data in this table, the input and output centrality of each function can be calculated. To determine the input centrality, the number of inputs to a function are added (adjacent vertices). This number is divided by the total possible inputs (total vertices). For function 5.1, the input centrality is the number of inputs divided by the total possible inputs (0/4 = 0 percent). This results because function 5.1 has no inputs and a possible total of four inputs. For function 5.2, the input centrality is 1/4 = 25 percent, because table 27 shows one input to function 5.2 out of a possible four inputs. Output centrality is calculated in the same manner. For example, function 5.1 has outputs to functions 5.2 and 5.3. Since the output centrality is the total number of outputs divided by the total possible inputs, the output centrality for function 5.1 is 2/4 = 50 percent. The input and output centrality for each function in this example is summarized below.
Table 28 summarizes this measure of centrality. Input centrality is proportional to the number of functions that a given function receives, while output centrality is proportional to the number of functions to which a given function provides information. In other words, a function that has a large number of outputs has a greater output centrality than a function that has fewer outputs. For example, table 28 shows that destination coordination (6.3) and route navigation (5.6) are very central in terms of the inputs they receive from other functional characteristics. Table 28 also shows road condition information (8.2) to be central based on the input it provides to other functional characteristics. These two facts are illustrated in figure 28. To ensure that the scenarios accurately represent the ATIS, several scenarios should include functional characteristics that are highly central; otherwise, scenarios may fail to capture a representative set of tasks associated with the functional characteristics critical to system success. Table 28. Centrality measures for private driver functional characteristics.
In conclusion, based on the network analyses for the private group, the centrality measures indicate that seven measures were found to be highly central. They are:
Cliques In addition to estimates of the centrality of each functional characteristic, an estimate of how information flows link functional characteristics into groups is an important element in identifying representative scenarios. If scenarios contain arbitrary sets of functions, they are unlikely to reveal representative information transfers between functional characteristics. Two network analyses were performed to explain how information flows link groups of functional characteristics. One analysis identified cliques and one identified clusters or "factions." Each uses a different criteria to group functional characteristics, but provides converging measures of groups of functional characteristics linked by information flows. A clique is a formal description of the density of links between nodes of a network. The density of links is equal to the total number of links divided by the number of nodes present. For example, if the number of links between two groups of functions is equal, the group that has the smallest number of functions (nodes) would have the greatest density of links. A clique also represents a maximally complete subgraph, which is defined as a coherent grouping of nodes connected by links or information flows. In other words, a clique is defined as a set of nodes that are directly linked to each other. Each member of a clique must have at least as many connections to other functional characteristics as there are members in the clique. Specifically, in a clique of three nodes, each node must have links to each of the other nodes in the clique (three) plus any other links to other nodes in the network that are not part of the clique. Thus, a clique shows a group of functions that share information directly between each other. For the private application of ATIS, the network analysis (see table 29) identifies 20 cliques of three functions or more (a clique of two is trivial, simply a pair of linked nodes). These results suggest that the selection of scenarios should include those that consist of groups of functional characteristics from the 20 combinations. Clusters A cluster or faction (Borgatti, Everett, & Freeman, 1992) analysis represents another network analysis method that can identify groups of functional characteristics linked by their information flows. The cluster analysis extracts "factions" and does not impose the same strict, formal definition on the members of a group that a clique does. This analysis optimizes a cost function that is based on the extent that a group of functional characteristics consists of linked clique–like structures (Borgatti, Everett, & Freeman, 1992). Thus, the cluster analysis identifies groups of functional characteristics in a manner vaguely analogous to factor analyses in traditional statistics. Table 30 summarizes the results of the analysis when five clusters were specified. Two of the five clusters were not considered meaningful as they contained functions that interacted only with themselves. The five clusters for the private operations consist of the functions listed in table 31. Table 29. Cliques for private functions.
Table 30. Clusters for private functions.
Table 31. Private vehicle cluster analysis.
These clusters of functions can help guide the task analysis by identifying scenarios that consist of functions within a single cluster and scenarios that require functions from several clusters. A task analysis of a scenario that consists of functions from a single cluster can reveal the requirements of supporting a driver with what should be an integrated set of functions (from the perspective of the information flows that generated the clusters). One interesting result of the cluster analysis is that the resulting clusters do not correspond only to the ATIS subsystems (IRANS, IMSIS, ISIS, IVSAWS). Nevertheless, these distinctions do indeed capture an essential feature: IRANS––Cluster 1, IVSAWS––Cluster 3, and IMSIS––Cluster 4.
FHWARD95176
