چارچوب کنترل کیفیت برای قابلیت اطمینان اتوبوس برنامه

This paper develops and demonstrates a quality control framework for bus schedule reliability. Automatic vehicle location (AVL) devices provide necessary data; data envelopment analysis (DEA) yields a valid summary measure from partial reliability indicators; and panel data analysis provides statistical confidence boundaries for each route-direction’s DEA scores. If a route-direction’s most recent DEA score is below its lower boundary, it is identified as in need of immediate attention. The framework is applied to 29 weeks of AVL data from 24 Chicago Transit Authority bus routes (and therefore 48 route-directions), thereby demonstrating that it can provide quick and accurate quality control.

service reliability indicators has been diminished by three problems. The first problem has been their infrequent collection. In order to make the best use of these indicators, it is necessary to frequently collect samples from each bus route, and to quickly make them available for analysis. In the past, for this activity to occur would have resulted in unacceptably high expenses because the data had to be collected and recorded manually (Nakanishi, 1997). The second problem has been the absence of a single, over-all performance indicator that validly aggregates partial measures such as those identified above. One comprehensive service reliability indicator would make it much easier to quickly and validly identify those routes most in need of intervention. With multiple indicators, it may be difficult to determine which routes have the overall worst performance because routes doing well on some measures may be doing poorly on others. This problem is exacerbated when quick decisions should be made. The third problem is determining whether a route’s declining service is caused by systematic new problems, or simply due to random chance. If management is to address problems of routes that are truly in difficulty, it should avoid wasting time on routes whose reported declines are simply random variations. The purpose of this paper is to present a framework for mitigating these three problems, thereby enabling management to more quickly and accurately identify those routes most in need of assistance. The framework involves use of (1) automatic vehicle location (AVL) data to obtain frequent and quickly-available samples, (2) data envelopment analysis (DEA) to aggregate the various service reliability measures into one comprehensive indicator, and (3) panel data analysis (PDA) to develop quality control charts for the performance of each individual route, which will alert management to routes performing worse than normal random variation explains. The paper unfolds as follows. In the rest of this introductory section, background information on AVL, DEA and PDA is presented. Then, application of the framework is illustrated through a case study using archived AVL data provided by the Chicago Transit Authority (CTA). The CTA’s bus route schedule adherence performance measures are defined in Section 2. The assessment framework is presented in Section 3. The case study results are reported in Section 4, including discussion on the DEA scores and their confidence intervals as quality controls for bus schedule adherence performance. Finally, the study contributions, limitations of the study and future research needs are summarized in Section 5. 1.1. Availability of automatic vehicle location data With automatic vehicle location (AVL) devices becoming available on many buses in recent years, the quantity and quality of data have greatly improved and can be made quickly available to transit agencies. According to the US Department of Transportation, two thirds of the 19 largest American transit agencies had their fleet fully equipped with AVL technology by 2004; the Chicago Transit Authority (CTA) is among those 100% AVL equipped agencies (US Department of Transportation, 2007). Therefore, AVL has become widespread and will likely be available at even more transit agencies in the future. 1.2. Data envelopment analysis DEA is widely used in economic analysis for identifying technically efficient operations (Cooper et al., 2004, Färe et al., 1994, Färe and Grosskopf, 2004 and Gattoufi et al., 2004). It is a linear programming method that combines partial efficiency measures into a single comprehensive indicator that provides objective evaluation and consistent comparisons of technical efficiency among decision making units (DMUs), i.e., base analysis units in DEA. Use of DEA to compare the efficiencies of urban transit systems has become increasingly popular in recent years, particularly since 2000. De Borger et al., 2002 and Brons et al., 2005 have given comprehensive reviews of transit DEA studies. Among the articles published since 2000, some analyze the efficiency of public transit in terms of services delivered (Graham, 2008, Karlaftis, 2003, Karlaftis, 2004, Pina and Torres, 2001 and Novaes, 2001); some measure the efficiency in terms of productivity (Odeck and Alkadi, 2001 and Odeck, 2006); others compare technical and social efficiency of transit agencies (Boilé, 2001, Boame, 2004, Nolan et al., 2001 and Nolan et al., 2002). One recent study uses panel data analysis to make statistical inferences about estimated technical efficiencies of Canadian paratransit systems (Barnum et al., forthcoming-a). Most recently, DEA has been applied to compare subunits within a single transit agency. Sheth et al. (2007) evaluate the overall performance of an agency’s bus routes by using DEA and goal programming with artificial data. In one study, Barnum et al. (2007) combine DEA and Stochastic Frontier Analysis to compare the CTA’s park-and-ride lot efficiency, and combine DEA and a reverse two-stage procedure (Barnum and Gleason, forthcoming) to analyze the technical efficiency of bus routes in another (Barnum et al., forthcoming-b). In this study, DEA scores are based solely on outputs (i.e., schedule adherence performance), which is different from all past transit DEA studies. In all of the other transit studies, DEA scores are based on output/input ratios known as technical efficiency indicators (Charnes et al., 1978, Cooper et al., 2004 and Färe et al., 1994). The DEA scores in this study are effectiveness indicators, because they measure goal achievement (Gleason and Barnum, 1982). DEA has been so used in non-transit cases, such as in determining best location in location analysis (Thompson et al., 1986) and in evaluating human performance (Anderson and Sharp, 1997). In these two cases, as in this study, the DEA score is a measure of comparative output performance of each DMU, not a measure of each DMU’s efficiency. Specifically, DEA is used in this study to compare the schedule adherence performance of individual routes. 1.3. Panel data analysis Even with quickly available AVL data from which performance indicators can be aggregated into a single valid measure by DEA, a third problem remains to be solved. Because of random noise in the data (Grosskopf, 1996), a given route’s sample DEA scores would be expected to vary; if a route’s score goes down, or even if the trend in its scores is downward, this does not necessarily mean that the route’s true performance has declined or is declining. In order to determine whether a decrease in a route’s mean performance has occurred, it is necessary to develop statistical tests to determine if the scores have declined from their expected value to a statistically significant degree, or if the observed variations are just due to random chance. Recently, a new method has been developed using panel data analysis on DEA scores to construct confidence intervals (Barnum et al., forthcoming-a). This technique has not been applied to transportation subunits, but, because AVL makes panel data for individual bus routes available, it can be applied in this study.

This study presented a Panel DEA framework for evaluating bus schedule adherence performance. The proposed framework was demonstrated with a case study using 24 CTA bus routes for a 29-week period between January and June 2006. The bus schedule adherence performance indicators, namely running time adherence and headway regularity, were derived from CTA’s archived bus AVL data. Compared to assessing bus schedule adherence based on partial performance indicators historically used by transit agencies, the Panel DEA-based framework demonstrates clear superiority in terms of providing a comprehensive performance measure that identifies problems quickly and accurately. The contributions of the paper to the transit literature are four-fold. First, this paper has demonstrated a new application of AVL data for transit operations. Because AVL data is continuously collected and quickly available, management can use the information to promptly address service reliability problems. Moreover, trend analysis and panel data analysis become practical when AVL data are available. Second, this paper presents a mathematically and economically plausible method to construct a comprehensive measure of service reliability from multiple partial reliability indicators, by using DEA. This DEA indicator is put into even better use than those in all past transit DEA studies because it not only identifies the benchmark DMUs but also prioritizes those in need of attention. Prior transit DEA studies have usually addressed the former but not the latter. In addition, the DEA indicator developed herein is a pure effectiveness measure in that it utilizes only outputs, instead of a ratio of outputs to inputs that have been employed in all past transit DEA studies. It is likely that there are other transit goals that could be best measured by considering only outputs rather than output–input ratios. Third, by coupling PDA with DEA, it has been demonstrated in this paper that the traditional deterministic DEA can be extended to stochastic DEA and thus statistical inferences can be made. This is the second application of Panel DEA to transit and the first to apply the methodology to performance evaluation of a transit agency’s subunits. Unlike all previous transit DEA papers, this paper is built upon the concept of quality control, realized by comparing each DMU’s current performance level to a statistical confidence interval based on its past performance. Lastly, this paper is among the first that have extended the use of DEA from the traditional comparisons among transit organizations to performance assessment of organizational subunits performing parallel activities. Nonetheless, there are limitations in this study that require further research. It is recognized that the four on-time performance indicators used in this study do not reflect every aspect of bus service reliability. For example, no measure of passenger related activity was considered. Nor have measures of traffic conditions or environmental factors been taken into consideration. The running time and headway based indicators were adopted mainly because they have been used by transit agencies (e.g., CTA) and they were readily derived from the archived AVL data. Some of the other measures may be derived from the bus automatic passenger count (APC) data; others may require different data sources such as traffic sensor data. The real-time AVL data is another source of data, which contain travel speed information. With more advanced data collection technologies available in public transportation systems, the framework presented in the paper should readily apply. Future research is also needed to identify causes for poor bus schedule reliability. This is outside the scope of this paper but no-doubt a close-to-the-heart issue to transit managers. Panel DEA tells which routes demand immediate attention; however, it does not identify the causes of the problems. Continuous research effort in the area is desired and will surely elevate the value of the research even further.