Você está na página 1de 10

2e Confrence spcialise en gnie des matriaux de la Socit canadienne de gnie civil 2nd Material Specialty Conference of the Canadian

Society for Civil Engineering Montral, Qubec, Canada 5-8 juin 2002 / June 5-8, 2002

CHARACTERIZING BUS TRANSIT PASSENGER WAITING TIMES


Wei Fan and Randy B. Machemehl Department of Civil Engineering, University of Texas at Austin, USA ABSTRACT: A statistically significant quantity of waiting time data, including related passenger attributes, transit system operational characteristics and bus frequency, were collected by direct observation and video taping in the city of Austin, Texas. The effects of bus line headway, schedule reliability index and many other potential predictors were examined. Mathematical models were developed to predict bus passenger waiting times in both totally random and real-world situations. Traditional, conceptual and new developed models were compared and evaluated. A new predictive linear model W = 2.0 + 0.3*BLH proved to be a reasonable and efficient model for describing passenger wait time using bus line headway as the independent variable. Difference between the traditional random model and the new developed model was examined. A 10 minute vehicle headway was identified as the transition from random passenger arrivals to non-random passenger arrivals. Introduction During the past 30 to 40 years, the transit share of total travel has been declining. Suburban sprawl and dispersion of employment provide great challenges for public transportation systems. Therefore, how to maintain transit ridership is a problem facing many transportation agencies today. Of many methods to expand the transit market, improving the consumer-based transit level-of-service is undoubtedly the key issue. Total transit trip times are often described as including in-vehicle and access time components. As a major part of the access times, passenger waiting times at bus stations are problematic because they increase total trip times and are generally regarded by many transit users as highly undesirable lost time, which is also cited as a factor that hinders the usage of bus transit. In conclusion, knowledge of bus transit passenger waiting times is necessary for demand estimation, operational analysis, transit scheduling, network design and other purposes. Several researchers have studied waiting times experienced by bus transit passengers. They have analyzed different factors that might have effects on passenger wait time and developed mathematical models to predict passenger wait time. However, none of these models could provide a satisfactory picture of passenger waiting time systematically and quantitatively. In order to better understand the characteristics of passenger waiting times, a further analysis of passenger waiting times and a model that can reflect changes in wait time corresponding to changes in factors affecting wait time is needed. The objective of this paper is to examine potential variables that may have effects on passenger wait times and to develop a mathematical model to predict waiting times as a function of bus passenger attributes and transit system characteristics.

2e Confrence spcialise en gnie des matriaux de la Socit canadienne de gnie civil 2nd Material Specialty Conference of the Canadian Society for Civil Engineering Montral, Qubec, Canada 5-8 juin 2002 / June 5-8, 2002

Literature Review A widely used traditional model used to express the average passenger waiting time was the half headway model, which was based on the following assumptions (Holroyd and Scraggs): (1) Passengers arrive at random; (2) Passengers can get on the first vehicles that come; (3) Vehicles arrive regularly. This model was frequently referenced by many later researchers who investigated passenger wait time. However, although most researchers regarded the first two of the above assumptions as acceptable approximations, the assumption of on-time bus arrivals was problematic. If possible bus arrival irregularity was considered, the average times that passengers had to wait before a bus comes was expected to be longer. Welding(1957), Holroyd and Scraggs(1966) and Osuna and Newell(1972) conducted theoretical research in this field and developed a model for the expected waiting time W, which was a function of the average headway and variations in the headway s2:

W = * (1 +

s2

)/2

[1]

where W----- expected passenger waiting times; ----- mean headways between buses; s2 ----- variances of headways between buses; If the headway becomes longer or bus service is more infrequent and there are some passengers who frequently make the same trip, it is expected that these passengers might plan their arrivals at the bus stops in order to reduce their waiting times. That is to say, passenger arrivals would become less random as headways increase. Several empirical studies demonstrated that this supposition was reasonable. OFlaherty and Mangan(1970) developed a model W = 1.79 + 0.14* that fit many observations collected in central Leeds. Seddon and Day14 (1974) found that W = 2.34 + 0.26* was applicable for passenger waiting times in Manchester. Salek and Machemehl (2000) explored a model, which was W = 2.0 + 0.3* based on extensive empirical data. Several researchers investigated related issues of passenger waiting time. Jolliffe and Hutchinson(1975) proposed a behavioral explanation of the association between bus and passenger arrivals at a bus stop, which considered that the passengers were made up of three categories of people: a proportion q who arrive coincidentally with the bus, because they see it coming and run to the stop and thus have zero waiting time; a proportion p(1-q) who arrive at the optimal time; and a proportion (1-p) (1-q) who arrive at random. Turnquist(1978) represented some modifications of the Jolliffe and Hutchinson model. With some designed changes, the effects on passenger wait time of service frequency and reliability for both random and non-randomly arriving passengers were identified. Bowman and Turnquist(1981) developed a model to evaluate the sensitivity of average passenger wait time at bus stops to service frequency and schedule reliability, in which passenger arrival times at bus stops was regarded as the outcome of a decisionmaking choice. Knoppers and Muller(1995) mainly investigated optimized transfer opportunities in public

2e Confrence spcialise en gnie des matriaux de la Socit canadienne de gnie civil 2nd Material Specialty Conference of the Canadian Society for Civil Engineering Montral, Qubec, Canada 5-8 juin 2002 / June 5-8, 2002

transport. Some characteristics of the possibilities and limitations of coordinated transfers in public transit were examined. In addition to these analysis, they also proposed the concept of thresholds between random and non-random passenger arrivals. These reserach efforts presented insights and provided a basis for related research on passenger waiting time. However, none of these models could provide a satisfactory picture of passengers waiting time systematically and quantitatively. A further analysis regarding passenger waiting times and a model that can reflect the changes in wait time corresponding to changes in factors affecting wait time is needed. Based on experimental data, this paper systematically investigates related variables and analyzes their effects. An efficient predictive mathematical model is developed. Data Collection Totally 2491 waiting time observations were collected in the city of Austin, Texas during a six-month period from June through December 1998 and a technique report was written by Salek and Machemehl(1999). To ensure data accuracy and consistency, two different techniques, direct observation and video tape recording, were employed in the data collection process. Passenger general attributes and waiting time data including gender, ethnicity, access mode, the activity passengers preformed during wait time, arrival time and departure time were recorded. Besides these data, transit system operational characteristics such as bus arrival time and bus departure time were also collected. Passenger waiting time components were calculated for each passenger by subtracting the values of Rider Arrival Time (RAT) from Bus Departure Time (BDT). Data Process and Regression Analysis It should be noted that some observations whose total waiting time was more than 1.3 times the bus headway, which were due to unusual cases such as mechanical bus problems or passenger unwillingness to take the first available bus, were excluded in this research. Except for these data, all the collected data set were used to perform statistical analysis. Incorporating all collected variables related to bus passenger attributes and transit system characteristics that may have effects on passenger waiting time, the following model was employed: WAIT = + *BLH + *SRI + *FEMALE + 1*A.M.PEAK + 2*A.M.OFF + 3*P.M.PEAK + 1*ASIAN + 2*AFRICAN + 3* HISPANIC + 4* WHITE + 1* BIKING + 2*DRIVING + 3*RIDING + 4*WALKING + 1*SOUTH + 2*NORTH + 3*HIGHLAND + 4*PAVILION + 5*SANJAC All variables included in the above equation except bus line headway and schedule reliability index are 1, if passengers are female dummy variables. For examples, FEMALE means that: FEMALE = { 0, if passengers are male and so on. Using the SPSS statistical package, regression results were obtained and shown in Table 1.

2e Confrence spcialise en gnie des matriaux de la Socit canadienne de gnie civil 2nd Material Specialty Conference of the Canadian Society for Civil Engineering Montral, Qubec, Canada 5-8 juin 2002 / June 5-8, 2002

Table 1 Regression Results of potential predictors for passenger wait times Unstandardized Coefficients B -.519 .295 .119 .287 .790 1.111 .267 1.383 2.367 1.522 2.124 -.352 -2.150 -3.803E-02 .228 .245 -.804 .292 -.779 -.396 Standardized Coefficients Beta .404 .062 .021 .042 .064 .015 .039 .146 .108 .145 -.005 -.070 -.001 .015 .015 -.042 .017 -.032 -.011 t -.431 20.377 3.091 1.053 1.492 2.362 .737 1.079 2.074 1.351 1.893 -.241 -1.822 -.035 .363 .453 -1.419 .463 -.795 -.444 Sig. .667 .000 .002 .292 .136 .018 .461 .281 .038 .177 .059 .810 .069 .972 .717 .651 .156 .644 .427 .657

Model 1

(Constant) BLH SRI FEMALE A.M.PEAK A.M.OFF P.M.PEAK ASIAN AFRICAN HISPANIC WHITE BIKING DRIVING RIDING WALKING SOUTH NORTH HIGHLAND PAVILION SANJAC

Std. Error 1.204 .014 .038 .272 .529 .470 .362 1.282 1.141 1.127 1.122 1.461 1.180 1.084 .627 .540 .567 .632 .980 .891

The t-values indicate that the effects on WAIT of some variables were not significant and so they were removed. The revised regression model was displayed in Table 2: Table 2 Revised Regression Results of the potential predictors for waiting times Unstandardized Standardized Coefficients Coefficients B Std. Error Beta (Constant) 1.800 .343 BLH .289 .014 .396 SRI .111 .037 .058 A.M.OFF .802 .337 .046 AFRICAN .729 .314 .045 DRIVING -2.338 .601 -.076 t 5.252 20.378 2.979 2.380 2.319 -3.889 Sig. .000 .000 .003 .017 .020 .000

Model 2

The t-values show that all variables included in Table 2 are significant at the 95 percent level, so the revised model is as follows:

2e Confrence spcialise en gnie des matriaux de la Socit canadienne de gnie civil 2nd Material Specialty Conference of the Canadian Society for Civil Engineering Montral, Qubec, Canada 5-8 juin 2002 / June 5-8, 2002

WAIT = 1.8 + 0.289*BLH + 0.111*SRI + 0.802*A.M.OFF + 0.729*AFRICAN 2.338*DRIVING Location and Waiting Time

[2]

Passenger waiting time data in this study were collected in six different representative locations in the City of Austin, which included Capitol Center, South Transfer Center, Northcross Transfer Center, Highland Mall Transfer Center, Pavilion Park & Ride and San Jacinto bus station. The results of ANOVA in Table 1 showed that, at 95 percent confidence level, there was no relationship between waiting time and location. This means that if other conditions are the same, average passenger wait times at different locations are not significantly different. Therefore, location was excluded in the final model. Traffic Period and Waiting Time To find out whether waiting time behavior is different during morning versus afternoon hours or during peak hours versus off-peak hours, daytime was categorized into four periods: A.M. peak hours, A.M. offpeak hours, P.M. peak hours, P.M. off-peak hours. As seen from Table 1, the t-values suggest the exclusion of A.M. peak hours, P.M. peak hours and P.M. off-peak hours as predictors for waiting time, which means that the waiting times in these three periods are not significantly different from each other if other conditions are identical. However, the result of the ANOVA in Table 2 indicates that the average time that passengers waited in A.M. off-peak hours would be 0.802 minutes more than that in any other period. From the original data, it is also found that corresponding to the lower travel demand, the headways in A.M. off-peak were also longer than in any other period. It is expected that this may result in the longer average passenger waiting times during A.M. off-peak hours. Therefore, it was excluded in the final model. Gender and Waiting Time The average waiting time for females (8.0 minutes) was higher than that for males (7.7 minutes), which means that there may be a gender effect. However, the results of ANOVA in Table 1 show that at a 95 percent confidence level, there is no statistically significant relationship between gender and WAIT and therefore gender was not included in the final model. Ethnicity and Waiting Time Passenger waiting times collected in this study regarding ethnicity included five categories of people: Whites, African Americans, Hispanics, Asians and other unidentified ethnicities. The t-values in Table 1 suggest that there is no relationship between WAIT and ethnicity except AFRICAN, which seems to be a significant variable for predicting wait time with the regression coefficient being 0.729. This means that AFRICAN passengers would wait 0.729 minutes more than others if all other conditions are the same. Access Mode and Waiting Time The results of ANOVA in Table 1 suggests that, at a 95 percent confidence level, DRIVING was a significant predictive variable for predicting passenger waiting times. It is noted that the coefficient of

e 22eConfrence spcialise en gnie des matriaux Confrence spcialise en gnie des matriaux 2e Confrence spcialise en gnie descivil de la Socit canadienne de gnie civil de la Socit canadienne de gnie matriaux de la Socit canadienne de gnie civil nd Material Specialty Conference 22ndMaterial Specialty Conference 2nd Material Specialty Conference of the Canadian Society for Civil Engineering of the Canadian Society for Civil Engineering of the Canadian Society for Civil Engineering

Montral, Qubec, Canada Montral, Qubec, Canada 5-8Montral, Qubec,5-8,2002 5-8juin 2002 / /June 5-8, 2002 juin 2002 June Canada 5-8 juin 2002 / June 5-8, 2002

DRIVING is negative, which means that those passengers who drove to the bus transit stops would wait less time to board the bus. The coefficient in this study is -2.338, which means that, under the same conditions, passengers who drove or were offered a ride to the bus stations would wait 2.338 minutes less than those who walk, ride, etc. This is expected for the higher probability that passengers would plan their departure times and drive to the bus stations as late as possible so long as they make sure that they can catch the intended buses. This would happen especially when the headway is long. Schedule Reliability and Waiting Time In this paper, SRI for each observation was defined to measure how each bus service was running according to the published schedules and its value was identified as the number of late or early minutes for the bus that passenger was taking. Passengers who boarded late buses were assigned positive SRI values and passengers boarding early buses were assigned negative SRIs. From the results of the ANOVA in Table 2, Schedule Reliability Index is a significant variable for predicting passenger waiting times. However, compared to the greatest significance of bus line headway, the relative contribution of SRI to WAIT was not so significant. The result in Table 3 shows that excluding SRI doesnt decrease the R-square value, which is 0.179, even a little higher than the previous value 0.172. This suggests that only using BLH as a predictor is practicable. Therefore, for simplicity, although Schedule Reliability Index is a significant variable, it was still excluded in the final model. Table 3 Regression Results of the preferred predictors for waiting times Unstandardized Standardized t Coefficients Coefficients B Std. Error Beta (Constant) 2.039 .279 7.300 BLH .299 .013 .423 23.103 Sig. .000 .000

Model 3

Bus Line Headway and Waiting Time Bus Line Headway was frequently used as an independent variable for predicting passenger wait times in most traditional models. The highest t-value of Bus Line Headway in Table 2 also suggests that the most significant variable affecting passengers waiting time is BLH. As described before, considering that in most practical applications of estimating passenger waiting time, bus service is always assumed to be ontime and the access mode by which passengers arrive at the bus transit stops is assumed to be walking, it seems feasible to exclude SRI, A.M.OFF, AFRICAN and DRIVING in a more practical model. Therefore, a mathematical model only using headway as independent variable for predicting passenger waiting times was developed in Table 3. WAIT=2.0+0.3*BLH (R2 =0.179) [3]

However, it should be noted that an assumption of equal variances of passenger waiting times regarding each bus headway was implicitly made. To demonstrate the feasibility of this model, a theoretical analysis

2e Confrence spcialise en gnie des matriaux 2e Confrence spcialise en gnie des matriaux 2e Confrence spcialise en gnie des matriaux de la Socit canadienne de gnie civil de la Socit canadienne de gnie civil de la Socit canadienne de gnie civil 2nd Material Specialty Conference 2nd Material Specialty Conference of the 2nd Material Specialty Civil Engineering Canadian Society for Conference of the Canadian Society for Civil Engineering of the Canadian Society for Civil Engineering Montral, Qubec, Canada Montral, Qubec, Canada Montral, Qubec, Canada 5-8 juin 2002 / June 5-8, 2002 5-8 juin 2002 / June 5-8, 2002 5-8 juin 2002 / June 5-8, 2002

and the chi-square test were performed and the results indicated that the assumption here is satisfied. Therefore, this linear regression model is feasible. However, the small R-squared value, may cause doubt whether the relationship between passenger waiting times and bus headway is significant or not. Based on the collected data, Duncans multiple range tests, were performed to test whether the bus line headway does have a relationship with passenger wait time. The results showed that, with 95% confidence, the bus line headway does have a significant relationship with passenger wait times, as the scatter plot in Figure 1 clearly suggested. Therefore, it can be used as an independent variable to predict passenger wait time. It is expected that the low R Squared value in model 3 can be explained by the great number of the waiting time data points and the wide range of passenger waiting times, especially when the bus headway is long. This can also be seen from the increasing standard deviation as illustrated in Figure 1, which would result in variability that could not be explained by bus line headway. Based on all these, model 3 is recommended.

50 Passenger Wait Time (minutes) 45 40 35 30 25 20 15 10 5 0 0 20 40 60 80 Bus Line Headway (BLH) (minutes) Scatter Plot Random Arrival Model Predicted Linear Model

Figure 1 Line fit plot of observed data for WAIT and its comparison with the random arrival model Another perspective to this relationship is that presented in Figure 2. At any given cumulative percentage, the figure shows that when the bus line headway becomes longer, the waiting time also tends to become longer.

2e Confrence spcialise en gnie des matriaux 2e Confrence spcialise en gnie des matriaux 2e Confrence spcialise en gnie des matriaux de la Socit canadienne de gnie civil de la Socit canadienne de gnie civil de la Socit canadienne de gnie civil 2nd Material Specialty Conference 2nd Material Specialty Conference of the 2nd Material Specialty Civil Engineering Canadian Society for Conference of the Canadian Society for Civil Engineering of the Canadian Society for Civil Engineering Montral, Qubec, Canada Montral, Qubec, Canada Montral, Qubec, Canada 5-8 juin 2002 / June 5-8, 2002 5-8 juin 2002 / June 5-8, 2002 5-8 juin 2002 / June 5-8, 2002

120%

100%

Cumulative Percentage

8<=BLH<=10 80% 11<=BLH<=15 16<=BLH<=20 60% 21<=BLH<=25 26<=BLH<=30 31<=BLH<=35 40% 36<=BLH<=40 41<=BLH<=60 20%

0% 0 10 20 30 40 50

Total Waiting Time (WAIT) (minutes)

Figure 2 Cumulative Percentage plot of observed data for WAIT Characteristics Analysis and Model Comparisons As described before, a widely used waiting time model suggests that if passengers arrive randomly at the bus station and the bus schedule is perfectly reliable, the mean waiting times will be about 0.5*BLH. Comparisons of the predictive model [3] derived from real data and the random arrival half headway model indicate that if the bus headway becomes longer, i.e., bus service becomes more infrequent, the difference of passenger waiting time predicted by these two models will increase as shown in Figure 1, which suggests that passengers tend to coordinate their arrivals according to the published timetable or available information. Comparing the linear regression model with the random arrival model, the headway at which the two models predict the same wait time can be identified. 2.0 + 0.3 * BLH = 0.5*BLH BLH =10 minutes [4]

This might suggest a 10 minute bus headway marks the transition from random to coordinated passenger arrivals. Actually, previous research on passenger wait time also suggested similar results. Seddon and Day (1974) found a value of 10 minutes in Manchester to be the threshold; OFlaherty and Mangan (1969) suggested that 12 minutes in Leeds to be the transition. According to Bowman and Turnquist (1981), Okrent (1974) also concluded that in general, 12 to 13 minute bus line headways marked transitions, from random to coordinated passenger arrivals. Therefore, it is recommended that an 10 minutes headway is the threshold below which passengers can be considered to be random arrivals. Furthermore, it is noted that all bus line headways in Austin were no less than 8 minutes, which might suggest that all transit users in Austin are non-random passengers. This might happen because passengers in Austin frequently make the same journey so that they are very

2e Confrence spcialise en gnie des matriaux de la Socit canadienne de gnie civil 2nd Material Specialty Conference of the Canadian Society for Civil Engineering Montral, Qubec, Canada 5-8 juin 2002 / June 5-8, 2002

familiar with the route schedule. As a result, they would time their arrivals in order to reduce their waiting times instead of arriving at the bus station randomly. The gain from passenger coordinated arrivals in passenger waiting time can identified by the difference between that predicted by the random arrival model and that by the developed model: G = Erandom - WAIT = 0.5*BLH - 2.0 - 0.3*BLH = 0.2*BLH 2.0 [5]

If the headways become longer, G (the potential gain from passenger coordinated arrivals) will increase, as shown in Figure 3. This is expected because as the headways become longer, more transit users will find it more beneficial for them to learn the schedule and plan their arrival time to reduce their waiting times at the bus station. Therefore, the portion of all the passengers who are non-random arrivals, will increase accordingly which in turn will increase the overall potential gain.
35 30 WAIT (minutes) 25 20 15 10 5 0 10 60 Bus Line Headway (BLH) (minutes) predicted linear model Random Arrival Model Gain from coordinated arrivals

Figure 3 Plots of gain from coordinated arrivals Conclusion Estimating passenger waiting time has many applications in optimum spacing and frequency of bus service, modal split studies and many operations designed to improve the public transit market. The main purpose of this study is to investigate the potential variables that affect passenger waiting time. Mathematical models were developed for both totally random and the real-world situations. A linear regression model, W = 2.0 + 0.3*BLH was developed for predicting average passenger waiting time as a function of bus line headway. Analysis results suggested a 10-minute bus headway as the transition from random arrivals to non-random arrivals. And as headways increase, it was found that the portion of nonrandom arrivals will increase. The results also suggest that most transit users in Austin are non-random arrivers. 9

2e Confrence spcialise en gnie des matriaux de la Socit canadienne de gnie civil 2nd Material Specialty Conference of the Canadian Society for Civil Engineering Montral, Qubec, Canada 5-8 juin 2002 / June 5-8, 2002

Acknowledgement Appreciation must be given for this research was sponsored by GRANT 10727 from SWUTC. Many thanks must be given to Mr. Salek for his extensive work that provided valuable contribution to this paper. References: Barnett, Arnold. (1974) On Controlling Randomness in Transit Operations, Transportation Science Vol.8, No.2, 102-116. Bowman L. A. and Turnquist M. A. (1981) Service frequency, Schedule reliability and Passenger Wait Times at Transit Stops, Transportation Research, Part A: General, 15A, No. 6, 465-471. Ceder, Avishai and Marguier, Philippe H. J. (1985). Passenger Waiting Time at Transit Stops, Traffic Engineering and Control, Vol.26, No.6 327-329. Holroyd E. M. and Scraggs D. A. (1966) Waiting Times for Buses in Central London, Traffic Engineering and Control, Vol.8, No.3, 158-160. Jolliffe J. K. and Hutchinson T. P. (1975) A Behavioral Explanation of the Association Between Bus and Passenger Arrivals at a Bus Stop, Transportation Science, Vol.9, No.4, 248-282. Knoppers, Peter and Muller, Theo. (1995) Optimized Transfer Opportunities in Public Transport, Transportation Science Vol.29, No.1. Marguier, Philippe H. J. and Ceder, Avishai (1984) Passenger Waiting Strategies for Overlapping Bus Routes, Transportation Science Vol.18, NO.3 207-230. OFlaherty C. A. and Mangan D. O. (1970) Bus Passenger Waiting Time in Central Areas, Traffic Engineering and Control, Vol.11, No.9, 419-421. Osuna E. E. and Newell G. F. (1972) Control Strategies for an Idealized Public Transportation System, Transportation Science Vol.6, No.1, 57-72. Ozekici, Suleyman. (1987) Average Waiting Time in Queues with Scheduled Batch Services, Transportation Science Vol.21, No.1 55-61. Polus, Abishai and Eck, Ronald W. (1982) Performance Dependability and Waiting Times of a Personal Rapid Transit System, Journal of Advanced Transportation Vol.16, No.3, 285-301. Polus, Avishai. (1978) Modeling and Measurements of Bus Service Reliability, Transportation Research Vol.12, No.4, 253-256. Salek M. D. and Machemehl R. B. (1999) Characterizing Bus Transit Passenger Wait Times, Southern Region University Transportation Center, 167211-1. Seddon P. A. and Day M. P. (1974) Bus Passenger Waiting Times in Greater Manchester, Traffic Engineering and Control, 15, 422-445. Skinner R. J. (1980) Bus Planning Methods, Traffic Engineering and Control, 554-558. Turnquist M. A. (1978) A Model for Investigating the Effects of service Frequency and Reliability on Bus Passenger Waiting Times, Transportation Research Record, No.663, 70-73. Welding P. I. (1957) The Instability of a Close-Interval Service, Operational Research Quarterly, Vol.8, No.3, 133-148.

10

Você também pode gostar