Effect of Passenger Encumbrance and Mobility Aid Use on Dwell Time Variability in Low-floor Transit Vehicles

Abstract

Public transit serves users with a broad range of physical capabilities and design needs. However information about the operational effects of diverse users interacting with the transit system is scarce. This paper examined the occurrence and effects of boarding and alighting passengers with mobility aids (wheelchairs, scooters, walkers and canes), or with large items (carts, strollers, bicycles, or carrying an infant) on bus stop dwell time in a fixed-route bus service. On-board video data from low-floor public transit buses serving Ann Arbor, Michigan were used from 199 bus stops with at least one passenger boarding or alighting with a mobility aid or encumbered with a large item, and an additional 1642 bus stops without any mobility aids or encumbrances. A sequence of linear regression models examined the relationship between dwell time and the addition of variables representing passengers with mobility aids and encumbrances, and use of the on-vehicle access ramp, beyond explanatory variables typically used in dwell time analysis. Accounting for passengers boarding/alighting with mobility aids and encumbrances (p < 0.001) and use of the access ramp (p < 0.001) increased the variance explanation of a dwell time model based on boarding passengers by fare payment, alighting passengers by door use, and passenger load from 46% to 56%. Results indicate distinct patterns in the durations for boarding and alighting by passengers with vs. without mobility aids and encumbrances, and when a ramp is used by wheeled mobility users vs. ambulatory passengers with walking aids. The findings suggest that accounting for the presence of passengers with mobility aids or encumbrances and ramp use in dwell time analyses could help transit operators make their service operationally more efficient and inclusive for all passengers and encourage more use of fixed-route transit among individuals with disabilities.

1. INTRODUCTION

Travel time is an important measure of transit performance. Dwell time, the time a bus spends at a scheduled stop without moving to load and unload passengers takes a considerable portion of bus travel time along a fixed route and can vary considerably depending on many factors. Shorter dwell times are universally desirable (Levinson, 1983). Clearly, the numbers of passengers boarding and alighting affects the dwell time, as do policies and practices pertaining to fare payment, use of doors, lifts, or ramps and the design and capacity of the vehicles which affect the ease with which passengers can move into and out of the vehicle. Fluctuations in demand for the transit service over the day, reflected in passenger load and door crowding, affect the available circulation space inside vehicle which in turn affects the ease of passengers’ movements and thus, the dwell time (Fernandez et al., 2015). Passengers with disabilities, users of mobility devices, or those with large objects may require more time to board/alight than fully ambulatory individuals, and their presence among boarding or alighting passengers has an effect on the time the bus spends at a stop (Kittelson et al., 2013). However, empirical data on the effects of passengers with encumbrances and disabilities on dwell time variability are lacking. An understanding of the variability in dwell time from the presence of passengers with disabilities and encumbrances is needed for guiding effective and efficient operations and policies.

1.1. Background

Since the passage of the Americans with Disabilities Act (ADA) (Title II, Part B) in 1990, public transit systems have become increasingly more accessible to persons with disabilities. The introduction of electromechanical lift systems, folding ramps, wheeled mobility securement systems and low-floor vehicle designs have enabled most people with disabilities, including those who use wheelchairs and scooters to ride fixed-route transit (King, 1998; Lenker et al., 2017). However, multiple studies report the accessibility barriers experienced by older adults and passengers with mobility impairments associated with boarding, interior circulation, and alighting on low-floor transit buses (National Council on Disability, 2005; Nelson/Nygaard Consulting Associates, 2008; Goldman and Murray, 2011). Users of wheeled mobility devices (e.g., manual and powered wheelchairs, electric scooters) in particular are affected disproportionately more than ambulatory passengers due to the presence of steep access ramps during boarding/alighting and limited space for on-board maneuvering in low-floor transit buses (Frost et al., 2010; Frost, Bertocci and Smalley, 2015; Nelson/Nygaard Consulting Associates, 2008; Syed et al., 2013). One study conducted in a mid-sized US city found that 43% of wheeled mobility device related incidents on public transit occur when the bus is stopped presumably when boarding and disembarking (Frost and Bertocci, 2010). Understandably, a majority of the prior research has focused on aspects of transit accessibility and safety.

Consideration of accessible and universal design for wheeled mobility devices in transit also benefits people with other mobility assistive devices such as walkers and canes as well as people with larger items such as strollers and carts to use fixed-transit bus services (Zajac, 2016). To encourage mode shifts to transit, and to promote bicycling for health, transit systems are increasingly allowing the transport of bicycles on their buses (Pucher, 2009). Statistics for the number or proportion of fixed-route passenger trips by persons with disabilities or with large items are difficult to find. The portion of fixed-route ridership by passengers with disabilities in the US has been estimated to range from about 2.5 % to 7%, based on a 2011 survey of seven transit systems serving different sized communities in various geographic parts of US (Thatcher et al., 2013). Similar estimates are unavailable for the number of fixed-route passenger trips with items such as strollers, carts, bicycles. However, the existence of policies about these items at various transit systems indicates that passengers have brought such items on board or tried to. A different 2011 survey of a sample of 42 transit agencies with regard to policies and practices on bringing large items on board reports that 81% of the agencies surveyed had policies regarding strollers, and 54%, 35% and 35% had policies on luggage, bicycles, and other large items, respectively (Goldman and Murray, 2011).

1.2. Consideration of Encumbered Passengers in Current Dwell Time Models

Accurate estimates of dwell time are important for designing efficient and reliable transit operations and policies (Kittelson et al., 2013; HCM, 2010; Jayaprakash et al., 2015). Models for estimating dwell time are typically developed from empirical data. Early models of dwell time included only the number of passengers boarding and alighting (e.g., Levinson, 1983; Feder. 1973; Guenther and Sinha, 1983). Because the speed of the boarding process is affected by the way passengers pay their fares, fare structures were included in some models (e.g., Kraft and Bergen, 1974, Guenther and Hamat, 1988). The demand for bus service on some routes is high during peak periods, so later models included variables that reflect the time of day (e.g., Milkovits, 2008) or passenger load and door crowding (e.g., Lin and Wilson, 1993; Puong, 2000; Tirachini, 2011; Fletcher and El-Geneidy, 2013). Even after the passage of the ADA the presence of passengers with mobility aids was not considered in dwell time analyses because their ridership on fixed-route transit was very low.

Until relatively recently the general practice was to simply consider data on passengers with disabilities on fixed-route transit as outliers or atypical events (Jayaprakash et al., 2015). A few research studies accounted for passengers with disabilities explicitly by including lift operations as a binary predictor of dwell time (Rajbhandari et al., 2003; Dueker et al., 2004). However, these models did not account for passengers with other mobility impairments that may not necessitate use of the lift. In addition, these models would need to be updated as electromechanical lifts are becoming less common on fixed-route transit vehicles in the US in favor of low-floor vehicles with folding access ramps due to lower equipment maintenance costs and shorter deployment times (Lenker et al., 2016). It has also been recognized that passengers with large items need more time to board and alight a transit vehicle, and a few models accounted for the effects of passengers with large items (e.g., luggage, bicycles) and those with disabilities by treating them as one encumbered category (e.g., Grise and El-Geneidy, 2017).

Current practice acknowledges the presence of passengers with disabilities. The Transit Capacity and Quality of Service Manual (Kittelson et al., 2013) notes that the “presence of infrequent events such as wheelchair or bicycle loadings have the potential to significantly increase a given bus’s dwell time at a stop” (pg. 6–69). The Manual uses dwell time variability, represented by the coefficient of variability at a stop for developing an operating margin that ensures the bus stop will meet a desired level of reliability by allowing for more than usual dwell times. For stops with regular wheeled mobility users the Manual recommends using either the directly measured dwell times or the average dwell time if direct measurements are not available. If bicycle usage is frequent, then the Manual recommends using the bicycle loading/unloading time if it is greater than the passenger service time. If the presence of wheelchair users or bicyclists is rare, then the Manual states that their effect is accounted for in the dwell time variability. The effect of users of other mobility aids or passengers with items other than bicycles is not specifically mentioned. Presumably, they would be accounted for in the dwell time variability component of the dwell time calculations.

A preliminary examination of dwell times from a fixed-route bus service in Ann Arbor, Michigan found that classifying bus stops by whether the boarding/alighting passengers included an individual with a mobility aid (manual or power wheelchair, scooter, walker, cane) or a large object (bicycle, stroller, cart or was carrying an infant) and whether or not a ramp was deployed yielded distinct differences in the distribution of measured dwell times (Kostyniuk and D’Souza, 2018). On occasion, the access ramp was deployed to serve ambulatory passengers with mobility aids or large items leading to longer measured dwell times than those for passengers with such encumbrances not requiring a ramp (Kostyniuk and D’Souza, 2018), and was never addressed in prior studies. This suggests that combining the approaches of considering mobility aids and large items as encumbrances and also including the use of the ramp in dwell time considerations may provide a method for considering the effects of a broad diversity of transit passengers, both with and without disabilities, into travel time estimates, and in turn into more inclusive measures of transit operations. For simplicity the current study uses the term encumbered to denote passengers with luggage, bicycles, strollers, and other large items.

1.3. Objective

The objective of this study was to quantify the effects of passengers boarding and alighting with visible mobility aids and encumbrances together with ramp deployment on dwell time at on-route bus stops in a fixed-route transit service. The study compares the addition of variables describing the presence of passengers with mobility aids and encumbrances, and ramp use to those variables conventionally used in dwell time models by comparing the respective contribution to variance explained in a series of linear regression models of dwell time. Implications of the findings are discussed.

2. DATA AND METHODS

Data for this study are from a larger research effort concerned with the use of public transportation by people with mobility impairments in Ann Arbor, Michigan. The Ann Arbor Area Transportation Authority, a.k.a. TheRide, provides fixed-route and paratransit services in the cities of Ann Arbor and Ypsilanti and nearby townships. At the time of this study, TheRide’s fixed-route operational fleet consisted of 80 low-floor vehicles, having a kneeling feature to reduce the height difference from the curb, equipped with electromechanical ramps that deploy for no-step access, and can accommodate persons who use mobility aids such as wheelchairs, scooters, and walkers. If requested, drivers assist passengers in wheeled mobility devices during boarding, alighting, and device securement in one of the two designated wheeled mobility securement areas. Passengers eligible for TheRide’s paratransit service or a special senior category ride the fixed-route bus for free, and reduced fares are available to individuals with other mobility impairments, other older adults, and people with low income. Fares can be paid with prepaid fare cards, and coins. If needed, passengers receive change in the form of fare cards of $0.25 value.

The buses are configured to accommodate passengers with items such as strollers, and bicycles. TheRide created a designated area on-board for strollers and walkers by removing one of the curb-side seats on all of its buses. Buses are equipped with two exterior bike racks on the bus anterior available without charge or special permit. No driver assistance is provided for securing and unloading bicycles.

2.1. Data Collection

On-board surveillance videos from a random sample of week-day bus trips (a bus trip being a one-way trip between terminals) recorded between November 2014 and December 2016 on six service routes were provided to the research team by TheRide staff for the larger study. The routes were selected because they serve key medical facilities, rehabilitation clinics, and disability service organizations in the community. TheRide’s operational policies and services on these routes remained unchanged during the study period. The study was approved by the University’s institutional review board.

Six video camera views were used in the analysis, including, two cameras each directed towards the front and rear doorway, one camera directed towards the fare payment device and driver station, one camera directed out the front windshield (capturing aspects of the anterior bicycle rack), and three cameras of the interior passenger compartment. A systematic task analysis of boarding and alighting in the video records was carried out by the research team using the methods described by Jayaprakash and D’Souza (2015). Task analysis is a human factors research methodology for documenting how a task is completed by deconstructing the task into smaller sub-tasks or steps, with each step being a segment of the operation defined by sub-goals and start and end events that occur during task completion (Shepherd and Stammers, 2005). Our analysis process involved manually observing the videos and extracting or coding key characteristics of passengers, and conditions at each bus stop. The analysis data for this study were developed from the systematic task review files. The data record for each bus stop on each bus trip included: route, date, time of day, bus-trip, stop identification and location, start of dwell time, end of dwell time, duration in seconds that the bus was early or late arriving at stop, passenger load (i.e., the number of passengers on board), the number of passengers boarding and number of people alighting by their mobility status (i.e., ambulatory with no encumbrances; ambulatory with mobility aids such as cane or walker; passengers seated in a wheeled mobility device such as a wheelchair or scooter; and ambulatory but encumbered with a large item such as bicycle, cart, stroller or carrying an infant), use of the access ramp, method of fare payment for each boarding passenger, doors used when alighting, and any unusual incidents.

For the present study, we selected records of bus trips that contained at least one stop at which a person with a mobility aid, large item, or infant in arms was among those passengers boarding or alighting. This resulted in 95 bus trips on six routes with a total of 2042 stops, of which 201 were stops at terminals and 1841 were at regular on-route stops. Dwell times at terminal stops are determined by pre-defined schedules rather than passengers boarding and alighting, and hence terminal stops were excluded from our analysis. In 199 (11%) of the on-route stops, at least one person with a mobility aid or large item was observed among passengers boarding and alighting.

The final dataset consisted of 1841 bus stops of which 1,642 bus stops had no passengers with visible mobility aids or encumbrances, and 199 that had at least one person with a mobility aid or large item, or infant in arms. Of these stops, eleven had both, at least one person with a mobility aid and at least one person with an item or carrying an infant among those boarding and alighting.

2.2. Statistical Data Analysis

The number of bus stops in each category stratified by the presence of passengers with mobility aids or encumbrances among those boarding and alighting and ramp use was tabulated. Descriptive statistics (mean, SD, range, coefficient of variation) of the dwell times for each category of bus stops were computed. Distributions of measured dwell time tend to be highly skewed toward short dwell times, and this was also the case in our data. Natural log transformations have long been used in studies of dwell time to achieve a more normal data distribution (e.g., Gonzalez et al., 2012; Rashidi and Ranjitkar, 2013; Guenter and Hamat, 1983). Log-linear models also help to correct for violations of the homoscedasticity assumption (i.e., equal variance of residuals) encountered in linear regression models of untransformed dwell time, leading to more consistent and efficient estimates of regression coefficients (Glick and Figliozzi, 2019).

Five regression models were progressively constructed to examine the relationship between natural log dwell time at bus stops and a set of explanatory variables that accounted for the number of passengers boarding and alighting passengers, their methods types of fare payment, the passenger loads, time of day, adherence to schedule, passenger mobility aids and encumbrances, and use of ramp. The progression of models started by using only the total number of passengers boarding and alighting as explanatory variables (Model 1), adding variables for peak period, passenger load, and whether bus was late vs. on schedule (Model 2), then replacing the number of passengers boarding and alighting with variables for front vs. rear door use (Model 3), fare payment method for passengers boarding (Model 4), and lastly adding variables for the number of passengers with mobility aids and encumbrances, and ramp deployment (Model 5). TheRide deploys the access ramp for passengers seated in wheeled mobility devices (e.g., manual and powered wheelchairs, and scooters) as well as for ambulatory passengers such as with a walker or a large object who request for the ramp. Wheeled mobility device users need additional time to maneuver to the wheeled mobility securement area, and then secure the wheeled mobility device often with driver assistance. Thus, ramp deployment was coded as a categorical variable with 3 levels, no deployment, deployed for passenger using a wheeled mobility device, and deployed for an ambulatory passenger (such as with a walker or other device or object).

Variables with significant parameter estimates (p < 0.05) were retained. To avoid multi-collinearities in the models, correlation analysis was first conducted on the explanatory variables, and significant highly correlated variables were excluded from the analysis. The variance explanation of dwell time for each model by the coefficient of determination (R²) was noted and the distributions of residuals were examined to ensure homoscedasticity. SAS 9.4 was used for all analyses with Proc GLM used for the linear regressions and Proc CORR for correlation analyses.

3. RESULTS

In total, 2215 passengers boarded and 2327 passengers alighted at the 1841 stops. Among the boarding passengers, 2102 were unencumbered, 14 had wheelchairs (3 manual and 11 power), 1 had a scooter, 13 had walkers, 29 had canes, 25 had carts, 19 had bicycles, 8 had strollers, and 4 had infants in arms. Among the 2329 passengers alighting, 2218 were unencumbered, 18 had wheelchairs (3 manual and 15 power), 1 had a scooter, 13 had walkers, 26 had canes, 22 had carts, 19 had bicycles, 7 had strollers, and 5 had an infants in arms. The average (± standard deviation; SD) number of passengers boarding was 1.20 (±1.79) and the average (SD) alighting was 1.26 (±1.89). The ramp was deployed 55 times.

Table 1 shows the number of bus stops in each category based on the presence of passengers with encumbrances and mobility aids among those boarding and alighting, and those needing the access ramp and the descriptive statistics of the dwell times for each category of bus stop. Overall, stops with passengers with no mobility aids or encumbrances had the lowest mean dwell time (14 s). Stops with ramp use had the highest mean, 88 s). However, there was a clear difference in the mean dwell times when the ramp was deployed for passengers using wheeled mobility devices (118 s) vs. for other ambulatory passengers (39 s).

TABLE 1.

Descriptive statistics of dwell time at bus stops by passenger encumbrance and mobility aid category

Type of Stop by Encumbrance and Mobility Aid Category	Number of stops (% all stops)	Range, s	Mean (±SD), s	Coefficient of variation
All stops	1841 (100)	2.4 – 317.7	17.3 (±25.2)	1.45
Unencumbered only	1642 (89.2)	2.4 – 174.3	13.8 (±17.1)	1.24
Total Encumbered & Mobility Aid	199 (10.8)	5.1 – 317.7	46.2 (±49.5)	1.07
Mobility Aid*	113 (6.1)	5.1 – 317.7	56.7 (±58.2)	1.03
Encumbered with large item	86 (4.7)	6.2 – 259.4	32.4 (30.3)	0.94
Mobility Aid & Encumbered - no ramp	144 (7.9)	5.1 – 259.4	30.1 (±31.4)	1.04
Ramp deployed	55 (3.0)	10.2 – 317.6	88.1 (±62.4)	0.71
Ramp deployed – wheeled mobility	34 (1.9)	29.0 – 317.6	118.2(±58.9)	0.50
Ramp deployed - other	21 (1.1)	10.2 – 131.6	39.4 (±27.2)	0.69

^*The 11 stops with both, at least one person with mobility aid and at least one person with an encumbrance are accounted for in this category

We examined the cases with the longest dwell times in the data file by reviewing the original videos. The maximum dwell time at a stop with unencumbered passengers only was 174 s (approaching 3 minutes). In this case 12 passengers boarded and 4 of them paid with coins, which appeared to slow the boarding process. The maximum dwell time for stop with encumbered or mobility aid passenger boarding was 318 s (over 5 minutes). At this stop 13 people boarded, including a person using a powered wheelchair and a person with a walker. Three people alighted the bus, two of whom used the front door to exit the bus, which consequently slowed down the boarding process.

3.2. Dwell Time Models

Table 2 shows the definitions of the entire set of variables that were considered in the dwell time models. The results of the dwell time regression analyses are summarized in Table 3.

TABLE 2.

Variable definitions and descriptions used in the regression models

Continuous Variable	Definition	N	Mean	SD
Ln Dwell	Ln Dwell time at on-route bus stop (s)	1841	2.42	0.82
Total boarding	Number of passengers boarding	1841	1.20	1.79
Total alighting	Number of passengers alighting	1841	1.26	1.89
Boarding-Fare card	Number of passengers paying with fare card	1841	0.71	1.40
Boarding-Coin	Number passengers paying with exact change	1841	0.31	0.68
Boarding-Coin/Card	Number passengers paying with coin and receiving change in fare cards	1841	0.01	0.09
Boarding- No Fare	Number of passengers not paying	1841	0.17	0.59
Alighting - Front door	Number of passengers using front door to alight	1841	0.42	0.81
Alighting - Rear door	Number of passengers using rear door to alight	1841	0.84	1.49
Total Boarding/Alighting Encumbered or Mobility Aid	Number of passengers with a mobility aid or encumbrance boarding and alighting	1841	0.12	0.39
Passenger load	Number of passengers on-board at arrival on bus stop	1841	15.4	7.06
Late	Actual departure minus scheduled departure (s), negative if early	1841	206.67	229.90
Categorical variables	Definition	N	Distribution
Peak	If time of bus arrival at stop is between: 6:00 – 8:59 or 15:00 – 17:50 hours, then peak = 1, else 0	1841	1: n = 591
			0:n = 1250
Latecat	If actual departure – scheduled departure was	1841	−1: n = 43
	< −120 s, then Latecat = −1		0:n = 718
	−120 s < Latecat < +120 s, then Latecat = 0		1:n = 1080
	>120 s, then Latecat = +1
Ramp Use	If ramp deployed for passenger using wheeled mobility device, Ramp Use = 1,	1841	1: n = 34
			2: n = 21
	If ramp deployed for an ambulatory passenger, Ramp Use = 2		0:n = 1786
	Else Ramp Use = 0
Route	Route number for the six routes used in this study	1841	2:n = 156
			4:n = 441
			6:n = 783
			10:n = 68
			11: n = 119
			20:n = 274

TABLE 3:

Regression model results for the natural logarithm of dwell time

Model	Model 1: Passengers In & Out		Model 2: + Passenger Load		Model 3: + Front & Rear Door		Model 4: + Fare Payment Method		Model 5: + Mobility aid or Encumbrance & Ramp Use
Parameter	Estimate (SE)	t-value	Estimate (SE)	t-value	Estimate (SE)	t-value	Estimate (SE)	t-value	Estimate (SE)	t-value
Intercept	1.933 (0.021)	92.74**	2.069 (0.035)	58.28**	2.050 (0.035)	58.03**	2.015 (0.034)	59.14**	1.996 (0.030)	65.10**
Total boarding	0.268 (0.008)	32.46**	0.278 (0.008)	32.66**	0.280 (0.008)	17.20**	-	-	-	-
Total alighting	0.129 (0.008)	16.09**	0.135 (0.088)	16.61**	-	-	-	-	-	-
Passenger Load			−0.011 (0.002)	−4.59**	−0.010 (0.003)	−4.38**	−0.008 (0.002)	−4.08**	−0.007 (0.002)	−3.78**
Alighting - front door			-	-	0.207 (0.019)	10.99**	0.207 (0.018)	11.38**	0.155 (0.017)	9.34**
Alighting - rear door			-	-	0.105 (0.010)	10.02**	0.104 (0.010)	10.39**	0.092 (0.009)	10.94**
Boarding - Fare card			-	-	-	-	0.198 (0.010)	19.22**	0.187 (0.009)	20.01**
Boarding - Coin			-	-	-	-	0.453 (0.021)	21.35**	0.426 (0.019)	22.12**
Boarding - Coin/Card			-	-	-	-	0.526 (0.156)	3.37	0.551 (0.141)	3.92*
Boarding - No fare			-	-	-	-	0.398 (0.024)	16.55**	0.279 (0.022)	12.40**
Total Boarding/Alighting Encumbered + Mobility Aid			-	-	-	-	-	-	0.317 (0.037)	8.34**
Ramp Use - wheeled mobility device			-	-	-	-	-	-	1.497 (0.101	14.80**
Ramp Use - Other									0.362 (0.124	2.92**
R2	0.398		0.404		0.410		0.457		0.562

^*p < 0.05;

^**p < 0.001

Model 1, the simplest model with total passengers boarding and the total passengers alighting as the only explanatory variables explained 39.9% of the variance. A categorical variable for the bus route was added to this model, but was not significant and was excluded from subsequent models. Variables for peak period (peak), adherence to schedule (i.e., late and latecat tested in separate models), and passenger load (passload) were added next. Because statistically significant correlations were found between the variables peak and late (rho = 0.08601, p < 0.001) and between late and passload (rho = 0.23656, p < 0.001), each of these variables was added separately. In each case about 40% of the variance was explained but the parameter estimates for late, latecat, or peak were not significant at p = 0.05. However, the parameter estimate for passload was significant at p < 0.001 but negative (Model 2). The reason for this may well be that passenger loads on buses arriving at most of the bus stops in the data were well below the capacity of the buses. The number of passengers on board ranged from 2 to 46, with a median of 15 passengers, 75^th percentile of 20 passengers and only one occurrence of the maximum of 46 passengers. Although, the negative sign was problematic, the variable passload was significant and was retained in subsequent models.

Model 3 replaced the number of passengers alighting by the number alighting through the front and rear doors, increasing the variance explained to 41.0%. The modest improvement in variance explained is most likely due to the fact that exiting through the front was not a frequent behavior. In a previous analysis we found that only 15% of passengers used the front door to exit when passengers were waiting to board (Kostyniuk and D’Souza, 2018).

Model 4 included fare payment methods where the total number of passengers boarding was replaced by the numbers paying by each fare payment method, namely, fare card, coin, combination of card and coin transaction, and those that did not pay. The variance explained increased to 45.7%.

Model 5 included the number of passengers boarding or alighting with a mobility aid or encumbrance, and ramp use differentiated by whether the ramp was used by a passenger using a wheeled mobility device or was ambulatory (e.g., using a walker). The variance explained increased to 56.2%. An examination of the residuals showed that some of the outliers were a result of the model under-predicting dwell times for a set of cases with no passengers with mobility aids or encumbrances, and small numbers of boarding/alighting passengers. Because unusual incidents were noted and coded in the analysis data file, we were able to discern what added to the dwell times in these cases. In one case, a passenger boarding tripped and fell, in another situation a passenger boarded the bus, talked with the driver and then alighted; in several cases there was a change of drivers or the door remained open as the bus started to drive off. Adding a variable for these unusual events could potentially increase the variance explained and improve the residual distribution. However, that would be fitting the data to this particular set of observations and would be difficult to generalize.

4. DISCUSSION

This study quantified the effects of boarding and alighting passengers with wheeled mobility devices (manual and powered wheelchairs, and scooters), ambulation aids (e.g., walkers and canes), or with large items such as carts, strollers, bicycles, or carrying an infant on bus stop dwell time in a fixed-route transit service. On-board surveillance video data from bus stops with and without passengers with mobility aids and encumbrances using TheRide, the public transit system in Ann Arbor, Michigan were utilized. A sequence of linear regression models examined the relationship between dwell time and the addition of variables representing total passengers with mobility aids and encumbrances, and ramp use to a set of explanatory variables typically used in dwell time analysis.

Broadly, the analyses indicated significant increases in dwell time from use of the access ramp and presence of passengers boarding/alighting with mobility aids and encumbrances. In the US, passengers using wheeled mobility devices require substantially more time for boarding and alighting on low-floor buses in-part due to design constraints in equipment such as the access ramp (e.g., driver initiates the deploying and retracting of the ramp), and the use of active wheelchair securement on-board the vehicle. These tasks require some level of communication and coordination with the bus driver, and occasionally cooperation from other passengers (e.g., to vacate the folding seats located in the wheelchair securement area). Interestingly, the study also found other passenger groups that required additional boarding/alighting time, and with 38% (n = 21 of 55) of all ramp deployments occurring for ambulatory passengers. While prior survey based studies have reported the occurrence of such events on public transit (e.g., Goldman and Murray, 2011), to our knowledge this is the first empirical study based on naturalistic transit use that quantifies differences in dwell time associated with specific passenger characteristics, particularly mobility aids and encumbrances.

From a systems perspective, the results demonstrate that transit service performance characterized by dwell time variability is inextricably linked to boarding and alighting practices of diverse passenger groups. These findings have important implications for dwell time estimation. The inclusion of the new variables associated with mobility aid use and other encumbrances increased the variance explained by the dwell time model considerably from 45.7% to 56.2%. Mean dwell times at stops with ramp use vs. stops with encumbered passengers and no ramp use were substantially different. For example, our final dwell time model (Model 5) estimates an average dwell time at a bus stop with a single passenger boarding in a wheelchair and no passengers alighting of 59.7 s (= exp (1.996 + 0.279 + 0.317 +1.497)), compared to 19.2 s = (exp (1.996 + 0.279 + 0.317 + 0.362)) for an ambulatory passenger boarding with a walker using the access ramp, and 12.2 s (= exp (1.996 + 0.187 + 0.317)) for boarding an ambulatory passenger with an encumbrance paying by fare card, compared to 8.9 s (= exp (1.996 + 0.187)) for boarding an ambulatory unencumbered passenger paying by fare card.

However, several of the variables such as peak period and on-time arrival which have been shown to affect dwell time in other studies (Dueker et al., 2004; Milkovits, 2008) were not significant in our analysis. This is not an inconsistency, but an artifact of the temporal travel patterns of passengers with mobility aids and other encumbrances. A review of the time of boarding by passengers with mobility aids and large items in our data showed that these passengers tended not to travel in the morning peak period, and wheelchair users are most likely to travel during the morning and afternoon off-peak period between 0900 and 1500 hours (Kostyniuk and D’Souza, 2018). A criteria for selection of bus trips for inclusion in our dataset, was that there was at least one bus stop with a passenger with an observable mobility aid or encumbrance. As a result 68% (n = 1250) of the bus stops in our dataset were in the off-peak period. This reinforces that people with encumbrances may choose to not travel in peak periods but tend to make their trips at less busy times. The fact that users of wheeled mobility devices travelled mostly in the morning and afternoon off-peak periods can be that they prefer to travel in the less busy and crowded conditions, or that their activities are scheduled at those times, or that travelling during peak periods is difficult and they may feel that they are disruptive to other passengers due to the long dwell time. The lack of travel during peak periods could also explain the negative direction of the passenger load parameter, i.e., a decrease in dwell time with increasing passenger load. As noted earlier, passenger loads were light in this dataset, so it is possible that drivers and passengers were less rushed in the boarding and alighting process during the off-peak periods. This implies that an approach stratified on peak and non-peak periods could better explain the effects of the various contributing factors.

The findings from this study also reflect severe shortcomings in the usability of current transit vehicles for wheeled mobility device users, resulting in among the longest dwell times by user group. The location of the entry doorways and fare payment system, the seating arrangement, and the sizes and positions of securement areas are all interrelated, and affect the size and shapes of maneuvering space for wheeled mobility devices available during boarding/alighting, and consequently the difficulty experienced and time needed for boarding/alighting (Lenker et al., 2017; D’Souza et al., 2017, 2019). In the US, the majority of access ramps on low-floor buses are mounted at the front of the bus where the space is highly constrained by the driver’s seating area, fare payment system, and the wheel-wells. Longer ramps that result in gentler slopes may reduce the floor space available for fare payment and maneuvering from the ramp to the passenger seating area. The securement of wheeled mobility devices in large transit buses continues to be a time-consuming process, with significant accessibility and safety concern for many wheeled mobility users and transit vehicle operators (Buning et al., 2007; Frost et al., 2013). Until many of these systemic and physical design issues get addressed, accommodations through inclusive policies (e.g., driver assistance) and adequate dwell time would need to be at the forefront.

4.1. Methodological Limitations

Although the variance explained increased with the addition of variables for encumbered passengers and ramp use, about 44 % of the variation in dwell time still remains unaccounted for. The review of outliers found that our final model underestimated the dwell time for set of cases with no passengers with mobility aids or encumbrances but in which some unusual event occurred. These included a passenger who tripped and fell while boarding, a passenger boarding and then alighting (possibly due to boarding the incorrect bus), and change of drivers, and other events that occurred randomly. These events are difficult to predict but have an effect on the time a bus spends at a stop. Another reason contributing to the unaccounted variance may be that this analysis used video review to identify passengers as mobility-impaired or encumbered and needing more time if they had visible mobility aids and large objects or an infant in arms, compared to other ambulatory passengers without observable encumbrances or disabilities. Video review could not identify individuals with other disabilities who do not use a mobility aid. For example some passengers, such as older adults, might also require additional time to board and alight the bus due to limitations in strength, joint range of motion, and balance. Such age-related functional decrements are difficult to categorize reliably from on-board videos. Specific inclusion of these passenger characteristics in dwell time models may further increase the explained variance.

The analysis used data from one transit agency. Some of the behaviors, particularly with respect to fare payments and use of door for alighting might be different in other locations with different fare structures, low-floor vehicle sizes and interior configuration, and policies for accommodating individuals with mobility aids and other encumbrances. For example, TheRide requires that passengers using wheeled mobility devices have their device secured with a four-point active securement system, which also involves driver assistance (Frost et al., 2013). Some US transit agencies give the wheeled mobility user the option for wheeled mobility securement, with many users opting out to avoid potential delay or inconveniencing the bus driver. Thus, the results are not readily generalizable without further similar analyses at other transit agencies. However, the need for and challenges of accommodating passengers with disability, older adults, and passengers with large items on public transit vehicles resonates with many transit providers in the US (Thatcher et al., 2013), as well as in other countries such as the UK (Gabriel et al., 2004), Sweden (Wretstrand et al. 2009), Australia (Broome et al., 2009). The reliance of manual methods to extract information from on-board video is also time-intensive and thus imposed limits on the sample size used in this study. Nevertheless, this study demonstrates the benefits of manual methods using video analysis for studying passenger activity, particularly user groups and travel behaviors that are extremely difficult to capture from direct in-person observations (e.g., people with mobility impairments since few presently use fixed-route transit; Thatcher et al., 2013) and by automated methods that rely on Automatic Vehicle Location (AVL), Automatic Passenger Counters (APC) and electronic fare payment methods alone (e.g., Grise and El-Geneidy, 2017; Milkovits, 2008). Unlike observational studies, studies using AVL and APC data are amenable to large sample sizes, but do not provide information on attributes of the passengers (e.g., older adults, passengers with a wheelchair or stroller) that can help explain variability in dwell times across an identical number of passengers. Additional studies that shed light on the experiences of and challenges encountered by people with disabilities and other large items using fixed-route transit are necessary to inform and update new policies to accommodate better the diversity of potential transit users.

5. CONCLUSIONS

Travel time is one of the important measures of transit performance and a key factor in quality of service perceived by trip makers. Dwell time, the time spent at bus stops loading and unloading passengers, can be a considerable part of the travel time in fixed-transit bus operations. Furthermore, the variability of dwell time depends on many factors, and knowledge of these factors and when and how they affect dwell time can be of great use for transit system operations.

Among the factors affecting dwell time on fixed-route buses that is often overlooked is the presence of people with disabilities requiring mobility aids or the presence of people with items such as bicycles, strollers, and carts, or carrying an infant. Although the frequency of finding these passengers is presently quite low, they need more time to board and alight than passengers without any encumbrances. The potential for delaying or inconveniencing others could also deter some people, particularly people with disabilities, from using fixed-route public transit (Goldman and Murray, 2011) leading to low frequency counts. Our results suggest distinct similarities in the durations for boarding and alighting by passengers with mobility aids when a ramp is not used and those with passengers with large items, and distinct differences when the ramp is used. Accounting for the presence of encumbered passengers and ramp use greatly affects the variance explained by dwell time models, and in turn into more inclusive measures of transit operations. Incorporating knowledge of the ridership composition on different routes or specific stops along a route that potentially serves a high proportion of older adults and persons with disabilities (e.g., hospitals, rehabilitation services, public libraries, Independent Living Centers) or with items such as carts (e.g., grocery stores) could yield more accurate and reliable schedules. This knowledge can help transit operators make their service more efficient and inclusive for all passengers and encourage more use of fixed-route transit among individuals with disabilities who are otherwise capable of using public transit.

HIGHLIGHTS.

• Effect of passengers with mobility aids and encumbrances on dwell time was examined

• Video data on 1841 on-route bus stops from a fixed-route service were analyzed

• Dwell time significantly increased with ramp use and number of passengers with mobility aids and encumbrances

• Adding variables on ramp use and passengers with mobility aids or encumbrances increased explained variance in regression models of log dwell time from 46% to 56%