Impact of Transportation Network Companies on Ground Access to Airports: A Case Study in Austin,Texas

Abstract

This study proposes using intelligent transportation systems (ITS) and open-data sources to evaluate the impact of transportation network companies (TNCs) on ground access to airports. The unexpected interruption of the TNCs services in Austin, Texas, U.S., in 2016, is used as a natural experiment to provide a before-and-after analysis of the changes in the traffic conditions of the access area to the Austin-Bergstrom International Airport (ABIA). An analysis of variance (ANOVA) is implemented to determine whether the difference in speeds across periods is statistically significant, and the value of time for TNC-induced delay is estimated, using values of passengers’ willingness to pay for airport access travel time savings. Furthermore, a speed linear model is developed to assess the impact of TNC demand on ground access areas using trip information from an Austin-based TNC service. The main results suggest that airport ground access speeds were higher during the period that the TNCs were out of the city. The re-introduction of the services resulted in a speed reduction of 9% for the airport morning and 18% for the afternoon peak hours, translating to a total passenger cost of approximately $150+ (morning) and $400+ (afternoon) per hour. Furthermore, it was found that the number of TNC pick-up trips is a predictor of airport access speed and that the flight schedule can potentially be used to develop predictive speed models.

Transportation network companies (TNCs) redefined the way that passengers choose to arrive and depart from airports. The introduction of TNCs has led to many significant changes at airports, including curbside design and configuration, service provider relationships (e.g., contracts with TNCs versus taxis and limousine services), operating rules and regulations, and revenue models (i.e., declining parking garage income) ( 1 – 3 ). Because of the relevance of these emerging transportation technologies on airport operations and maintenance, it is imperative that airport agencies understand the effect of TNCs on ground access to make informed decisions. However, research studies on the impact of TNCs on ground access are limited, and the main challenge remains the availability of publicly available data to provide empirical evaluations ( 2 ).

This research presents an analysis of the effect of TNCs on ground access to airports using a case study in Austin, Texas, U.S., where the major TNCs left for a year. Although it is widely recognized that TNCs have a direct impact on airport curbside congestion, there are limited empirical analyses, and the leading research studies are based on surveys ( 1 , 2 , 4 ). It is proposed to use intelligent transportation systems (ITS) to assess the congestion conditions on airport access and evaluate the impact of TNCs. The unexpected interruption of TNC services during one year in the city of Austin is used as a natural experiment to assess traffic conditions with and without the presence of TNCs.

Background

In 2012, the San Francisco International Airport (SFO) became aware that a new business was operating on airport roadways without permits ( 1 ). Users were arriving and departing the airport using an app-based transportation service. It was not until 2014 that the first permits were issued, allowing TNCs to operate on airports through pilot programs ( 2 ). Nashville International Airport and SFO were among the first primary airports in the U.S. to develop permit agreements with these companies ( 1 , 2 ). By the end of 2016, TNCs were permitted at 60 airports, and, in 2019, Uber had more than 170 agreements with U.S. and Canadian airports ( 4 ).

Parking and ground transportation activities represent essential revenue sources for U.S. airports, representing 42% of the non-aeronautical revenue for all U.S. commercial service airports in 2018 (refer to Figure 1a). For the Austin-Bergstrom International Airport (ABIA), this percentage represented 51% in 2018, but it has decreased steadily over the previous decades from 67% in 2000 (Figures 1b and c ). Therefore, airport operators need to fully understand all activities involving ground access areas, including TNC services.

Figure 1.

Non-aeronautical revenue (6): (a) all U.S. commercial service airports (2018) total: $10,682 million; (b) Austin-Bergstrom International Airport (ABIA) (2018) total: $96 millions; and (c) parking and ground transportation percentage of non-aeronautical revenue by year.

TNCs have had numerous impacts on airport operations, including the need for additional staff to oversee TNC operations, a decrease in taxicabs, shared-ride vans, limousines, and rental car operations, parking usage, and an increase in curbside or roadway congestion ( 2 ). Airport ground access congestion is a result of the increasing volume of TNC traffic and some drivers unfamiliar with airport regulations. It is also a consequence of staging and holding actions, including drivers driving slowly past curbside areas waiting for arriving passengers or improperly parking on roadway shoulders near the terminals ( 2 ). Furthermore, despite allowing shared rides, TNC services have caused switches from shared modes to private trips, which also contributes to curbside congestion ( 6 ). Airports have begun to employ measures such as reassigned pick-up and drop-off locations and short-term parking garages to reduce the traffic effects ( 4 ). Despite the significant consequences of congestion on ground access areas, airport operators base their congestion metrics on surveys and visual inspections (C. E. Thomas, personal communication). ITS allows for a direct measure of traffic conditions, such as speed and vehicle volumes, and can potentially help operators manage current traffic conditions and inform short- or long-term decisions providing data-driven approaches.

Objective and Contributions

This research proposes the use of ITS data and open-data sources to evaluate the impact of TNCs on airport ground access. The methodology consists of the analysis of a case study in Austin, Texas, where the leading TNC companies, Uber and Lyft, left the City of Austin and the ABIA airport area for one year. The unexpected interruption of the service will be used as a natural experiment to analyze the effect of TNCs on airport access. To achieve this goal, a three-pronged analytical framework for understanding speed differences is proposed:

First, diverse data sources are used to analyze the demand patterns during the period of study. TNC and transit demand data are included, and the airport enplaned and deplaned passenger volumes and the mode distribution are evaluated.

Second, the traffic conditions are evaluated before and after the TNC service interruption, using historical speed data. Transit automatic vehicle location (AVL) data is used, using transit vehicles as traffic probe sensors, to determine traffic speeds and travel times on the airport ground access ( 7 ). Also, ITS traffic information (INRIX data) is used to monitor the airport highway access during the same period.

Finally, statistical methods are implemented to determine whether the difference in speed across periods is statistically significant, and estimate the time value of TNC-induced delay using values of passenger willingness to pay for airport access travel time savings. Furthermore, a speed model is developed to assess the impact of TNC demand on ground access speed, using trip information from an Austin-based TNC service.

This study contributes to the literature by providing an empirical analysis of the effect of TNCs on airports, using data that can potentially be obtained by other airports. Previous research on the topic is mostly based on surveys or simulated scenarios. Although informative, surveys are subject to bias and provide limited information, while historical data can be used to develop predictive models to support data-driven analysis for decision-making processes. The main contributions of this research include: (i) a methodological framework to obtain speed measures based on transit vehicle trajectory; (ii) empirical evidence of the influence of TNCs on airport congestion; and (iii) a proposed method to use ITS and other diverse data sources to monitor airport access area traffic conditions.

Outline

The subsequent sections of this paper are organized into five parts. First, the literature review is presented, where an overview of the previous work in the field is provided. Next, the data description section presents an overview of the primary data sources used in the study. The methodology describes the processes made to obtain the speed data. In the results and discussion, the main findings are presented. The last section summarizes the main findings and proposes future lines of research.

Literature Review

In recent years, there has been substantial work on TNCs in transportation research studies ( 8 – 12 ). However, there is a lack of research focusing on the TNC impact on airport operations.

The primary research in the topic includes the work by Mandle and Box, and Leiner and Adler ( 2 , 4 ). Mandle and Box conducted a survey during fall 2016 of the airport staff members responsible for ground transportation operations at the 100 largest U.S. airports, with responses received from 72 of these airports ( 2 ). Their study provides valuable insights into the state of regulations and impacts of TNCs on airport revenue and operations. They highlighted the lack of before/after data availability and the lack of relevant literature at the time of the study. Among their main findings, 46% of the airports reported an increase in roadway congestion as a result of TNC operations. Furthermore, operators stated that there was a significant decrease in parking and rental car revenue. Although 98% of airports had a TNC permitted charged trip, activation, or annual fees, the study suggests that these revenues are less than the foregone revenues at most airports More recently, Leiner and Adler provide a reference guide designed to help airport operators develop and implement practical approaches to managing TNCs within the context of commercial ground transportation policies and programs ( 4 ). It includes a survey of airports, interviews with landside managers, and a review of the revenue and business impacts of TNCs. The study suggests that managing curb congestion and enforcing permit conditions is one of the most persistent issues that airports face.

Previous research studies analyzing airport curbside congestion include the work by Hermawan and Regan ( 6 ). They evaluated the effect of TNCs on the use of shared modes (such as buses, light rail, shared vans, and shuttles) using data from the 2015 passenger surveys in SFO, Los Angeles (LAX), and Oakland (OAK). Results suggest that TNCs replace shared rides more than they complement them, and the net effect is that TNCs add to congestion as they increase the number of low-occupancy vehicles at the airport. Furthermore, Hermawan and Regan provide an analysis of passenger sensitivity to changes in TNC travel time and cost by developing estimates of ground access mode-choice decisions in LAX ( 13 ). They found that, if TNC fares were raised to match the price of taxi services, their demand would fall approximately 20% from the initial TNC shares. Furthermore, business passengers appear to be slightly more sensitive to travel time, while leisure passengers are more susceptible to travel costs.

Recent studies analyzed the mode changes after the introduction of TNCs. Dong and Ryerson investigated how taxi trips, AirTrain ridership, and parking transactions have changed since the launch of UberX and Lyft at two major New York airports ( 14 ). Their results showed that the number of taxi trips decreased at both airports. At the same time, AirTrain ridership continued its growing trend, and the number of parking transactions at John F. Kennedy International Airport (JFK) continued the mild decline that began before the launch of these TNC services. Similarly, Wadud studied the effect of the entry of TNC services on airport parking patronage at New York airports using monthly parking data, finding a statistically significant reduction in the numbers of cars parked at the airports after the entry of TNCs ( 15 ).

In terms of studies in the Austin, Texas, area, Gurumurthy and Kockelman§ used an agent-based simulation approach (using MATSim) with TNC data available for the City of Austin, to approximate the impact of a futuristic fleet of shared autonomous vehicles (SAVs) on ABIA revenue and operations ( 16 ). The main result suggested that the airport may lose 30%–60% of revenues from airport access fees levied on TNCs in a future world of SAVs, where dynamic ride-sharing reduced the number of SAVs serving the airport.

In summary, research studies analyzing the impact of TNCs on airports focused on mode replacement and parking analysis. There is limited evidence of the direct measure of traffic congestion on the airport ground area, and researchers have based their results on surveys and observational studies. This study aims to provide a data-driven method for estimating traffic congestion in this area and proposes the use of publicly available ITS data to provide diverse analysis.

Data Description

The aim is to to evaluate traffic conditions at the airport before, during, and after the leading TNCs interrupted their service, and to analyze travel demand information for different transportation modes during this time. First, a period of analysis of 5 years was selected, from January 2015 to December 2019. Information was collected from different sources during this period. Figure 2 presents a timeline, including the data source availability (on the right side) and significant events that affected the airport area (on the left side).

Figure 2.

Timeline of data sources, events, and periods of analysis.

The TNC service offered by Uber and Lyft was interrupted between May 9, 2016, and May 28, 2017 (shaded area in Figure 2), after residents voted to maintain strict regulations, including fingerprinting of drivers. Because of this interruption, other companies, such as RideAustin and Fasten, started their service. The demand for these emerging TNCs began to increase exponentially after October 2016 ( 2 ). In September 2016, ABIA began the construction of nine additional gates. However, this construction did not affect the traffic accessing the ground area, since it was isolated in the airport’s southeast area. Based on these events and on the data availability, three periods of analysis were selected, including 5 months per year (May to September), to compare across periods without carrying high seasonal variations. The periods are described as follows:

Period 1 (“before”)—between May 1 and September 30, 2015—is in the time before Uber and Lyft left the city.

Period 2 (“during”)—between May 9 and September 30, 2016—is in the time during which Uber and Lyft were absent from the city.

Period 3 (“after”)—between May 29 and September 30, 2017—is in the time after Uber and Lyft returned to the city.

It is important to mention that, during November 2018, the ABIA managers permanently moved the ride-sourcing pick-up location to a nearby garage. Picking up passengers from the ground access area is no longer allowed. This change reduced significantly the curbside congestion, based on observations made by ABIA managers (C. E. Thomas, personal communication).

ABIA Passengers and Flight Schedule

Information about deplaned and enplaned passengers was obtained from the ABIA Activity Reports, publicly available through the State of Texas Open Data Portal. This information is aggregated by month and includes data from January 2015 to December 2019. Furthermore, the daily flight schedule for arrivals and departures of commercial, general aviation, military, and cargo operations flights is approximated, using the ABIA aviation activity forecast ( 17 ).

TNC Information

The TNC information used in this analysis corresponds to a dataset that RideAustin, an Austin-based company, made available in early 2017, publicly available at Data [dot] World. It consists of 1,494,125 rides between June 2, 2016, and April 13, 2017. The dataset provides a description of the trip, rider and driver (anonymized), payment, cost, location, and other trip information. Furthermore, the total ABIA TNC demand is approximated, using information from the fiscal year trip reports (October to September) for the years 2016 to 2019, provided by ABIA (C. E. Thomas, personal communication). The annual demand was divided into monthly ridership. An estimated percentage of 60% drop-off and 40% pick-up trips were used based on ratios obtained from the RideAustin dataset.

Transit Information

Two data sources were used to obtain information about public transit: the automatic passenger counter (APC), and AVL. AVL and APC data are widely used in transit analyses because they provide a standardized data source, which allows the analysis of data across different transit agencies ( 18 ).

Automatic Passenger Counter (APC)

The Capital Metro APC ridership dataset provides information about transit vehicle location and its corresponding boarding and alighting at the stops level. Sensors are mounted in the doorways to count passengers boarding and exiting, and information on the bus stop location and time is acquired. The dataset is provided by Capital Metro twice per year through the State of Texas Open Data Portal. It presents 47 columns, including details about the route identification number, dwell time, vehicle identification number and maximum load, timestamp, latitude, and longitude. The APC dataset allows the estimation of transit demand information—for example, the number of passengers boarding and alighting. For this study, data from January 2015 to December 2019 is analyzed.

Automatic Vehicle Location (AVL)

The Capital Metro AVL system automatically determines and transmits the geographical location of a transit vehicle. It provides information about transit vehicle location. The data, also available at the State of Texas Open Data Portal, consist of location measures every 2 min for each of the Capital Metro routes. The information provided includes the trip, vehicle, and route identification numbers, among other variables that allow for fusion with different transit datasets to provide a broader analysis. The AVL information can provide insights into the transit travel time, average speed, and delays. However, the main limitation of the data is that, because of the 2 min interval, the location data can vary greatly, and there is no additional information within this time window. This study develops a method to estimate the transit vehicle speed based on this data. Another limitation is that, from 2016 to 2018, the transit agency provided instantaneous vehicle positions but did not store the data. The historical data available were automatically stored in a GitHub repository only from March 2015 to January 2018. Currently, the transit agency stores and shares the information from the year 2019 until the present day, on request. Also, the interval was reduced from 2 min to 5 s, allowing for more granular analyses.

Traffic Information

The traffic information on the roadway network surrounding the airport is obtained using INRIX data. INRIX a private-sector provider of travel time and speed information. INRIX collects anonymized data on congestion, traffic incidents, parking, and weather-related road conditions. They gather real-time, predictive, and historical data from different GPS tracking data sources.

Methodology

ITS information is used to estimate airport ground access traffic conditions. This section provides details of the methodology used to achieve this estimation.

Airport Traffic Conditions

Two different approaches are used to approximate the airport traffic conditions. The first consists of the estimation of speed in the airport ground access area using a transit bus as a probe vehicle. The second is the estimation of the speeds in the highway access areas, using information provided by ITS traffic sensors.

Using Transit Bus as a Probe Vehicle

Probe vehicles are typically used to collect traffic data in real time. This study aims to approximate the airport ground access traffic condition using transit buses as traffic probe vehicles. It is proposed to use the position of route 100, the transit bus that connects the City of Austin with ABIA, to approximate the airport ground access speed. The bus shares its path with vehicles entering the airport, providing insights into the ground access area’s traffic conditions.

The analysis using the available AVL data has limitations because of the low sampling rate of 2 min intervals. Estimating speed values require at least two consecutive location points and, because of the low sampling rate, a large distance separates these location points. Therefore, a large segment has to be considered to obtain two to three transit location points at the airport accessing area. The analysis uses a segment of 2.3 km (1.4 mi) along Presidential Boulevard, from the intersection with Hotel Drive to the drop-off point. (See Figure 3a for more details). The selected path does not include any traffic signal that would affect the bus speed. The last 600 m (0.3 mi) includes only the pick-up zone located in the lower level, which is the bus’s path. The upper level consists of the drop-off zone. However, the bus shares a significant distance (1.8 km or 1.1 mi) with vehicles accessing both drop-off and pick-up areas. Therefore, the speed is considered a reasonable estimation of vehicles’ overall speed entering the airport. The last 300 m (0.2 mi) of the bus route is located in the arrival area. This area splits into an inner road for personal vehicles and TNC pick-ups, and an outer road shared between the bus service and the taxi/van services.

Figure 3.

Location of the study area: (a) Austin-Bergstrom International Airport (ABIA) ground access total segment; and (b) highway access INRIX segments.

The bus speed estimated using AVL data captures the congestion caused by the increase in traffic volume in the shared road segment of the access area (87% of the path selected) and the spillover effect of this area’s pick-up activities. However, it is likely that the speed changes in the shared segment are not as significant as the speed changes in the immediate pick-up service area, where the TNC services had a greater influence, based on observations made by ABIA managers (C. E. Thomas, personal communication) and as reported by the ABIA master plans ( 17 ). Therefore, this metric may underestimate the congestion on the inner road location, directly related to TNC pick-up activity. Still, because of the lack of any other data source capable of measuring speed in this location, the speed of the bus is a good approximation of the overall conditions of the airport accessing area. New developments in the transit data collection system allow a higher sampling rate on the current vehicles. Currently, Austin Capital Metro collects and stores the position of the vehicles every 5 s. Therefore, future studies in similar conditions could be performed at a granular level, and speed estimates could be limited to shorter sections.

INRIX Traffic Information

INRIX data, aggregated by 15 min intervals, is used to obtain the highway vehicles’ speed information for two segments of approximately 2 km (1.2 mi), each connecting the Texas highway TX 71 with ABIA (refer to Figure 3b for more details). Since INRIX directly provides speed data, no additional process for speed estimation is applied.

Transit Vehicle Speed Estimation

Transit vehicle speed estimation is based on geographic vehicle position data obtained from AVL. This section describes the process followed to obtain this information. The process is divided into a set of steps, described as follows:

Step 1: Organize and Clean the Data

In this step, the daily AVL raw CSV files for the period of analysis, corresponding to 1,041 records (1,041 days between March 2015 and January 2018), are merged. Then, the information for the airport route filtering is selected by the route identification number 100 and by trip direction (in this case, only trips arriving at the airport are of interest). Furthermore, the last segment of the journey is selected, corresponding to the last 2.3 km, depicted in Figure 3a.

Step 2: Define Valid and Invalid Points

Each trip contains a set of points describing the location of the bus within the last segment of 2.3 km (1.4 mi) before arriving at the airport. There were several location points recorded while the bus was stopped because the AVL sensor continues providing the location every 2 min, even after the arrival at the transit stop. It was necessary to define the location point corresponding to the arrival to avoid including invalid data points in the speed estimation. This point was found by defining a buffer of 50 m around the transit stop and selecting the first point of the trip located within the area. The location points after this were marked as invalid, since they do not describe the trajectory of the bus (refer to Figure 4a for examples of invalid points).

Figure 4.

Examples of speed estimation: (a) location points; and (b) speed estimation based on distance and time (weighted average).

Step 3: Speed Estimation

The speed for every sub-segment of the trip, formed by two consecutive location points, was estimated. First, the distance traveled between them is estimated. The Google Maps API is used to find the distance traveled, to account for the semicircular geometry of the path. In total, 75k sub-segments were identified and processed through the distance estimation process. Finally, the time it took the bus to cross the sub-segment, which is defined as the difference between the timestamps of two consecutive points, is calculated. The sub-segment speed is estimated as the ratio of distance and time. The average speed of the trip is defined as the weighted average (based on the longitude) of the sub-segment speed. Figure 4a shows examples of the speed estimation for two trips.

Results and Discussion

This section presents the principal results and a discussion of the main findings. First, an analysis of the historical airport demand by transportation mode is presented. Second, the results of the estimated speed for the defined periods of study are analyzed.

Airport Demand by Transportation Mode

Using the information available, the historical airport demand by mode is approximated. The results are shown in Figure 5, along with the historical airline passengers enplaned and deplaned. The airport data shows a growing number of passengers, with an increment of approximately 50% in 5 years. However, this increase does not affect the analysis drastically because the change is not significant for the selected periods. There is a 5% increase in passengers between Period 1 and Period 2, and a 12% increase between Period 2 and Period 3. The difference between Periods 2 and 3 may affect the traffic conditions, with Period 3 potentially showing higher congestion than Period 2. An extended discussion is added in the speed analysis section.

Figure 5.

Austin-Bergstrom International Airport (ABIA) airport demand, based on data available.

TNC usage at the airport presents an unbalance of the ratio between pick-up and drop-off trips. Using information from trips made by RideAustin, it was found that 60% of the trips correspond to drop-off rides and 40% to pick-up rides. The “ABIA-rep” monthly demand is estimated based on the fiscal-year demand information provided by ABIA and using the same ratio of pick-up and drop-off trips obtained from RideAustin data.

As shown in Figure 5, the TNC demand has increased exponentially over the years, with three times as many trips in 2019 compared with 2017. There is no information on TNCs before April 2016. Therefore, the TNC usage during Period 1 of the analysis is not known. During Period 2, the leading companies were out of the city, and small, newly formed companies started their service with low demand, corresponding to less than 5k trips. Approximately 0.3% of the airport users arrived by TNC in this period. Therefore, it is considered as the treatment period, where there was not an extensive presence of TNCs in the airport. For Period 3, there were a total of 170k TNC trips, with an average of 3.3% of passengers arriving by this mode. By 2019, the percentage of passengers arriving by TNC increased to 10.3%, an increment of 312%.

Transit usage has also increased over time, with a 115% change between 2017 and 2019. Austin’s transit agency implemented significant improvements to the transit network in June 2018. The airport route increased its coverage area and frequency, translating into higher transit usage, with 53% more boardings between 2019 and 2018. There is no information about transit usage during Period 1 of the analysis, while Period 3 presented 25% more transit demand than Period 2. The number of passengers arriving by transit increases at the same rate as the number of airport users. Therefore, the mode share percentage remains the same over the years. In 2016, only 0.7% of the airport users arrived by bus, while, by 2019, and after the improvements made to the airport bus route, 0.9% of users arrived by transit. These percentages are similar to the national average, where approximately 1% of business travelers and 2% of leisure travelers use public transit to access the airport ( 19 ).

Speed Analysis

This section summarizes the results obtained from the estimation of speed. The analysis is focused on comparing the two highest periods of the day, corresponding to the morning peak, between 5:00 and 7:00 a.m., and the afternoon peak, from 3:00 to 5:00 p.m. ( 17 ). Furthermore, only weekdays are included, and weekends and holidays are removed, to keep uniform the comparison across periods. The results are summarized in box plots for which Figures 6a and b show results for the airport ground access speed, estimated using transit information. Figures 6c and d show the speeds on TX 71 which connects to the internal access road, obtained from INRIX data.

Figure 6.

Average speed for airport ground access (automatic vehicle location [AVL]) and highway access (INRIX): (a) average AVL speed per period; (b) average AVL speed per period by month; (c) average INRIX speed per period and segment; and (d) average INRIX speed per period by month.

Speeds on the internal airport ground access road range from 20 to 55 km/h, approximately. During the morning peak, Period 3 shows the lowest speeds, while Periods 1 and 2 are similar to each other. For the afternoon peak, Period 2 is notably higher compared with the others. This is an important finding because it means that, when the major TNCs were out of the city, the airport ground access speed was higher than when they were serving the area. Figure 6b shows the disaggregation of the data into months, May to September. The results show that Period 1 has a higher variability (a wider spread across the y axis), mainly during May and September during the afternoon peak hours. Also, Period 2 shows low variability among all the months for the afternoon peak, but the morning peak presents considerably more variability. To provide a more comprehensive evaluation, in the next section an analysis of variance (ANOVA) is developed to test if the differences across the three means are statistically significant.

The speed estimated for the airport highway access road (TX 71 eastbound) shows higher variability, especially for Periods 1 and 2, during the afternoon peak. The speed ranges between 60 and 105 km/h, and, overall, the values obtained for Period 3 seem to be higher than for Periods 1 and 2. The monthly disaggregation, shown in Figure 6d, indicates that all months in these two periods present significant variation. These results can be related to the construction of the TX 183 segments near the airport, which could have affected the traffic conditions mainly during the afternoon in the eastbound direction. Since there are external factors that cannot be isolated for the TX 71 speed measures near the airport entrance, it is not possible to draw conclusions from this analysis.

Analysis of Variance (ANOVA)

A one-way ANOVA is implemented to test the statistical significance of the differences across the means of ground access speeds for the three selected periods. The analysis is divided into morning and afternoon peaks and compares the variation across the means of Period 1, Period 2, and Period 3. The ANOVA model can be written as:

Y_{ij} = μ_{i} + ε_{ij}

(1)

where:

$Y_{ij}$ is the speed value of the $j^{th}$ sample for the $i^{th}$ period;

$μ_{i}$ are parameters;

$ε_{ij}$ are independent $N (0, σ^{2})$ ;

$i = 1, 2, 3$ are the periods; and

$j = 1, \dots, n_{i}$ are the samples per period, approximately 200.

The ANOVA tests the hypothesis of equal means across periods. The null and alternative hypotheses can be written as shown in Equations 2 and 3. Furthermore, the Tukey honest significant difference (HSD) test is used to provide a pairwise comparison of the means. The ANOVA model and the Tukey test are estimated using R; the results are shown in Table 1.

H_{0} : μ_{1} = μ_{2} = μ_{3}

(2)

H_{1} : Means are not all equal .

(3)

Table 1.

Analysis of Austin-Bergstrom International Airport (ABIA) Ground Access Speed Variance Across Periods

Sources of variation	Sum of squares	df	Mean square	F	p-value
Morning peak
Between periods	1,319	2	659.5	21.04	0.00*
Error	17,364	554	31.3
Total	18,683	556
Afternoon peak
Between periods	6,576	2	3,288.0	76.98	0.00*
Error	19,991	468	43.0
Total	26,567	470
Tukey honest significant difference (HSD) test results
	Speed (km/h)				p-value
Pairwise comparison	Mean difference	CI lower	CI upper		p-value
Morning peak
Period 2 versus period 1	1.23	−0.26	2.73		0.13
Period 3 versus period 1	−2.36	−3.69	−1.02		0.00*
Period 3 versus period 2	−3.59	−4.95	−2.23		0.00*
Afternoon peak
Period 2 versus period 1	6.08	4.04	8.12		0.00*
Period 3 versus period 1	−2.78	−4.60	−0.97		0.00*
Period 3 versus period 2	−8.86	−10.55	−7.18		0.00*

Significance at 95% confidence level.

The ANOVA results indicate that the p-value for the F-test is significant at a 95% confidence level. Therefore, the ANOVA null hypothesis can be rejected, and it can be said that not all means are equal for both the morning and afternoon periods. Furthermore, an analysis of the sum of squares shows that 7% of the variation in speed can be explained by the differences across periods for the morning peak, while in the afternoon peak, 24% of the change in speed can be explained by the periods.

The ANOVA is used to test if the means across periods are equal, but it does not provide a pairwise analysis. Therefore, the Tukey HSD test is used, as shown in Table 1. The results showed a mean speed increment of 6 km/h for the afternoon period when the TNC services left the city (Period 2). This difference is significant at a 95% confidence level, while the same change for the morning period is not significant. Also, when the services came back to the city (Period 3), the mean speed decreased nearly 9 km/h (from 47.8 to 39.0 km/h ), corresponding to 18% for the afternoon peak, and 4 km/h (from 44.4 to 40.8 km/h), or 9%, for the morning peak. The difference between the morning and afternoon peak results can be related to the arrivals/departures imbalance in the airport in the morning and afternoon peaks. The schedule shows that, during morning peaks, most of the flights are departures, while the afternoon peak is high on both arrivals and departures ( 17 ). The speed may be more related to the arrivals because the bus trajectory is highly influenced by the location of the pick-up area, occupied by the passengers arriving from a flight.

The reduction of speed after the return of the leading TNCs can be related to the mode change. Based on the demand information, TNC users are likely not using the public transit system as a replacement mode. Therefore, the demand corresponds to personal vehicle users, who tend to use the parking garages. Unlike TNCs, vehicles located in the parking garages do not directly access the airport ground area for pick-up or drop-off. When the major TNCs returned to the city, it is possible that personal vehicle users moved to TNCs, leading to an increment in the number of vehicles accessing the airport ground area for drop-off and pick-up trips.

Furthermore, it is important to highlight that, during Period 3, ABIA had an increment of 12% of passenger traffic. Higher passenger demand can also be a contributing factor to the speed reduction of 9 km/h. However, Period 2 showed an increment of 5% as well, and the results depicted an increase of 6 km/h in speed. Therefore, speed changes do not seem to be significantly affected by the small changes in demand, and can still be a good indicator of congestion resulting from TNC activity. The increase in demand likely corresponds mostly to personal vehicles. These users tend to access the airport from the designated garages, avoiding a significant section of the ground access area.

Time Valuation

Assuming that TNCs caused the reduction in speeds, the previous results can be used to quantify the cost that all the airport users would incur because of the curbside congestion induced by the presence of TNCs. The value of time is a dollar amount assigned to value the benefit of a change in expected travel time resulting from transportation projects, policies, programs, or events ( 19 ). A national air passenger survey by Landau et al. estimated passengers’ value of time for ground access and egress times ( 19 ). Since ABIA airport does not have this data, the approximations are appropriate for the development of the analysis. These values were converted from 2013 to 2017 dollar value using a 5.22% of cumulative price change based on the U.S. Bureau of Labor Statistics consumer price index ( 20 ). The converted value is used to determine the approximate cost of the delay induced by the re-introduction of TNC companies in Austin in 2017. Landau et al. provided a guidebook designed to enable a more accurate application of benefit-cost studies for proposed airport improvement projects by expanding and enhancing passenger value-of-time calculations ( 19 ). However, time valuation is still an ongoing research issue, and the results presented in this study should be taken with caution in light of the variability of time valuation definitions.

The delay associated with the leading TNCs’ return is approximated, using the average speed values from Period 2 and Period 3 over the 3 km access segment that connects the highway exits to the airport main entrance. For the morning peak, the average speed changed from 44.4 to 40.8 km/h (a reduction of 3.6 km/h) and this corresponds to an additional 0.27 min per vehicle, while for the afternoon peak, the average speed changed from 47.8 to 39.0 km/h (a reduction of 8.9 km/h) corresponding to an additional 0.66 min per vehicle, approximately.

During Period 3 (May to September 2017), there were, on average, 607,635 monthly passengers enplaned (refer to Figure 5). Approximately 10% of the total flight departures are made during the highest demand hour of the morning and afternoon peaks ( 17 ). Therefore, about 60,764 passengers per month, or 2,025 passengers per day, used the airport during the highest demand hour of morning and afternoon peaks. The departing passengers are divided into business (40.4%) and leisure (59.6%) used by Landau et al. and also provide the total or airport composite values ( 19 ). Using these values, the time value for the TNC-induced delay at the airport ground access during the return of the TNC services is estimated. The results are summarized in Table 2.

Table 2.

Time Valuation for the Transportation Network Company (TNC)-Induced Delay at the Airport Ground Access

Category	Number of passengers departing per hour	TNC-induced delay (hours)	Value of time per hour (2017 dollars)	Daily TNC cost per hour (2017 dollars)
Morning peak
Business	818	3.68	$19.57	$72.06
Leisure	1,207	5.43	$17.83	$96.86
Airport composite	2,025	9.11	$18.52	$168.80
Afternoon peak
Business	818	9.00	$19.57	$176.15
Leisure	1,207	13.28	$17.83	$236.76
Airport composite	2,025	22.28	$18.52	$412.62

In 2017, the daily passenger cost of the delay induced by TNCs was approximately $170 per hour for the morning peak and $410 per hour for the afternoon peak. The total annual cost is approximately $210,000, considering only the two highest hours of the day. Business passengers reported a higher value of time. Therefore, the impact on them is higher than for leisure passengers.

Modeling Airport Ground Access Speed

This section analyzes further the speed information to understand which factors influence changes in the speed at the airport ground access level.

Modeling speed allows us to analyze how useful the TNC demand is for predicting speed and determining whether the relationship between speed and TNC demand is due. For the modeling, data from the TNC company RideAustin when the number of trips was significant, between October 2016 and April 2017, is used. Also, the ground access speed values estimated for the same period are used. The speed estimated for this study is obtained using transit vehicle location information. During this time, buses and TNCs used the same pick-up area. Therefore, the model is based on pick-up trips only.

The simplified model is shown in Equation 4, where the independent variable corespondents to hourly speed estimates, and the dependent variables are the number of hourly TNC pick-up trips and a time variable in units of months. Although the number of pick-up trips by RideAustin does not include the total TNC demand, it was nearly a third of the market, making it possible to provide a basic analysis ( 8 ). It is important to mention that this model ignores multiple other variables that influence speed, such as traffic volume and roadway geometry. Rather than developing a robust speed model, the aim is to analyze the nature of the relationship between TNC trips and speed. The linear model is estimated using R, with a sample size of 3,173 hourly speed measures. The results are summarized in Table 3.

y_{ground access speed} = β_{0} + β_{1} x_{pick - up trips} + β X_{months}

(4)

Table 3.

Airport Ground Access Speed Model for RideAustin Trips (October 2016–April 2017)

Variable	Estimate	p-Value
Intercept	41.79	0.00*
Transportation Network Company (TNC) pick-up trips (hourly)	−0.04	0.00*
November 2016	−3.53	0.00*
December 2016	−2.19	0.00*
January 2017	−1.54	0.00*
February 2017	−0.18	0.71
March 2017	−1.48	0.00*
April 2017	−0.49	0.43
Goodness of fit
$R^{2}$	0.03
F-statistic	11.63
p-value	0.00
Sample size (n)	3,173

Significance at 95% confidence level.

The model indicates that, as the number of TNC trips increases, the ground access speed decreases by a factor of 1.2 km/h for every 30 RideAustin trips per hour. Furthermore, this relationship is statistically significant at a 95% confidence level. Based on the estimates of TNC pick-up trips during the 2019 fiscal year, an average of 180 trips per hour were made at ABIA. Using the simplified model to provide an approximate projection, this demand would translate into a speed decrement of approximately 7 km/h. Although the correlation coefficient of 3% is low, other multiple variables are not considered, including two-thirds of TNC demand from other providers estimated in the same period. Yet, the model F-statistics and p-value are significant, confirming the validity of this model approach. Furthermore, the sample size of 3,173 data points also adds validity to the model estimates.

Hourly changes in TNC demand are now compared with the airport flight schedule. Figure 7a shows the hourly estimated flight schedule of ABIA and Figure 7b presents the average hourly TNC demand estimated with RideAustin data ( 17 ). Using this information, it was found that the number of TNC drop-off trips is correlated to the number of flight departures with a 70% correlation coefficient. In comparison, the TNC pick-up trips are related to flight arrivals by a 55% correlation coefficient. Although simplistic, this analysis suggests that the airport flight schedule can be used to estimate TNC demand and to approximate changes in curbside speed.

Figure 7.

Hourly flight schedule and transportation network company (TNC) demand: (a) Austin-Bergstrom International Airport (ABIA) approximate hourly flight schedule (17); and (b) average hourly ABIA TNC demand (RideAustin).

Conclusions, Limitations and Future Work

This study uses the unexpected interruption of TNC service in Austin, Texas, in 2016 as a natural experiment to analyze the effect of TNCs on airport access. The use of public transit buses as probe vehicles is proposed to approximate the ground access speed and a methodological framework is implemented to process the raw bus location points and obtain approximate speed values in the area of interest. This study highlights how various data sources can be merged to answer a question of interest unrelated to the original purposes of the collected data. It also proposes the use of ITS sensors to better assess traffic conditions at the airport accessing area.

The main findings suggest that, when the major TNC companies returned to the city, the internal access road speed was reduced. The passenger cost of these speed reductions was approximately $170 per hour to morning peak passengers and $410 per hour to the afternoon peak ones. Furthermore, the speed model indicates that, as the number of TNC pick-up trips increases, the airport ground access speed decreases. Although expected, the results can help the development of improved models that can predict changes in traffic conditions using information widely available for airport managers, such as the flight schedule. The highway traffic conditions did not provide valuable insights because of multiple external factors affecting speed in the surrounded areas. However, this method is still a valuable resource that airport officials can use. The introduction of ITS sensors on the airport roadway can provide direct measures of the traffic conditions and help airport managers guiding decision-making processes with data-driven analysis.

The results and methods proposed in this study can serve multiple purposes. First, from the airport agencies’ point of view, the proposed method to measure traffic conditions can help airport operators to understand the volume and nature of activities occurring at curbside and ground access areas to adequately address operational challenge presented by TNCs. An analysis of the congestion impact can inform the need for near-term and long-term adjustments in decision-making about resource allocations (such as curbside capacity expansions), permits agreements, regulations, and the establishment of TNC fees (e.g., pick-up and drop-off fees, congestion fees, or annual fees). Second, from the transit agency point of view, this study can help evaluate the benefits of improvements to the transit system as an investment for reducing congestion consequences. Finally, our results also have relevance in transportation research by providing data analytics methods and empirical findings to fill gaps in the literature on airport ground access.

This study used AVL data collected by transit vehicles to approximate the speed at the airport accessing area. Although innovative, the results should be taken with caution. The AVL data was not collected with the purpose of analyzing the speed changes in the airport area, and the low sampling rate (2 min intervals) limited the resolution of the estimates. The selected 2.3 km segment includes a large section where the congestion may not have been as critical as the immediate pick-up area. Furthermore, the arrival lane of the bus route was separated from the TNC area in the last part of the 2.3 km segment of analysis. Therefore, it is possible that the speed obtained does not reflect the exact conditions of the airport inner arrivals roadway. However, the AVL data provides a good approximation of the traffic conditions in the area. New developments in data collection and storage allow higher sampling rates. Future studies in similar situations can use the methodology developed in this paper to approximate speed conditions in before-and-after scenarios using data with higher resolution.

The limitations of this study also include the lack of disaggregated information at a temporal level. Future analysis can be improved by using more detailed information, such as hourly passengers enplaned and deplaned, parking occupancy by hour of the day, daily flight schedules, and transit vehicle location with higher time resolution. This information can help the development of improved speed models that can be used by airport officials to predict changes in traffic conditions.

Footnotes

Acknowledgements

The authors thank Larry Goldstein and the project advisers and mentors for their beneficial comments, guidance, and feedback. The authors also thank Carlton Thomas for his valuable collaboration in providing insights about the ABIA ground access management.

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: N. Zuniga-Garcia, R. Machemehl; data collection: N. Zuniga-Garcia; analysis and interpretation of results N. Zuniga-Garcia, R. Machemehl; draft manuscript preparation: N. Zuniga-Garcia. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by TRB ACRP through the Graduate Research Award Program on Public-Sector Aviation Issues.

ORCID iDs

Natalia Zuniga-Garcia

Randy B. Machemehl

References

Davol

A New Model for Airport Ground Transportation: Transportation Network Companies at San Francisco International Airport. Journal of Airport Management, Vol. 11, No. 2, 2017, pp. 147–153.

Mandle

Box

ACRP Synthesis 84: Transportation Network Companies: Challenges and Opportunities for Airport Operators. Number Project A11-03, Topic S03-11. Airport Cooperative Research Program (ACRP), Washington, D.C., 2017.

Nix

E. J.

Mundy

R. A.

Airport Drop-Off and Pick Up Charges in the United Kingdom: Will They Come to the United States?

Journal of Airport Management, Vol. 11, No. 3, 2017, pp. 309–326.

Leiner

Adler

Transportation Network Companies (TNCs): Impacts to Airport Revenues and Operations. Number 215. Airport Cooperative Research Program (ACRP), Washington, D.C., 2019.

Federal Aviation Administration (FAA). Certification Activity Tracking System (CATS) Report 127, 2020. https://cats.airports.faa.gov/Reports/reports.cfm.

Hermawan

Regan

A. C.

Impacts on Vehicle Occupancy and Airport Curb Congestion of Transportation Network Companies at Airports. Transportation Research Record: Journal of the Transportation Research Board, 2018. 2672: 52–58.

Cathey

F. W.

Dailey

D. J.

Transit Vehicles as Traffic Probe Sensors. Transportation Research Record: Journal of the Transportation Research Board, 2002. 1804: 23–30.

Lavieri

P. S.

Dias

F. F.

Ruiz-Juri

Kuhr

Bhat

C. R.

A Model of Ridesourcing Demand Generation and Distribution. Transportation Research Record: Journal of the Transportation Research Board, 2018. 2672: 31–40.

Tec

Zuniga-Garcia

Machemehl

R. B.

Scott

J. G.

How Likely are Ride-share Drivers to Earn a Living Wage? Large-scale Spatio-temporal Density Smoothing with the Graph-fused Elastic Net: Application to Ride-Sourcing Driver Productivity Analysis, 2019. arXiv Preprint arXiv:1911.08106 https://arxiv.org/abs/1911.08106

10.

Yahia

C. N.

de Veciana

Boyles

S. D.

Rahal

J. A.

Stecklein

Book-Ahead & Supply Management for Ridesourcing Platforms. Transportation Research Part C: Emerging Technologies, Vol. 130, 2021, p.103266.

11.

Zuniga-Garcia

Spatial Pricing Empirical Evaluation of Ride-Sourcing Trips Using the Graph-Fused Lasso for Total Variation Denoising. Master’s thesis. University of Texas, Austin, 2018.

12.

Zuniga-Garcia

Tec

Scott

J. G.

Ruiz-Juri

Machemehl

R. B.

Evaluation of Ride-Sourcing Search Frictions and Driver Productivity: A Spatial Denoising Approach. Transportation Research Part C: Emerging Technologies, Vol. 110, 2020, pp. 346–367.

13.

Hermawan

Regan

A. C.

On-Demand, App-Based Ride Services: A Study of Emerging Ground Transportation Modes Serving Los Angeles International Airport (LAX). Journal of the Transportation Research Forum, Vol. 56, 2017, pp. 111–128.

14.

Dong

Ryerson

M. S.

Taxi Drops Off as Transit Grows Amid Ride-Hailings Impact on Airport Access in New York. Transportation Research Record: Journal of the Transportation Research Board, Vol. 2675, No. 2, 2020, pp.74–86.

15.

Wadud

An Examination of the Effects of Ride-Hailing Services on Airport Parking Demand. Journal of Air Transport Management, Vol. 84, 2020, p. 101783.

16.

Gurumurthy

K. M.

Kockelman

K. M.

Impacts of Shared Automated Vehicles on Airport Access and Operations, with Opportunities for Revenue Recovery: Case Study of Austin, Texas. Under Review, 2020.

17.

Austin-Bergstrom International Airport (ABIA). Austin-Bergstrom International Airport (ABIA) Master Plan. Technical Report, 2019. http://www.abiamasterplan.com/theplan/.

18.

Furth

P. G.

Hemily

Muller

T. H. J.

Strathman

J. G.

Transit Cooperative Research Program (TCRP) Report 113: Using Archived AVL-APC Data to Improve Transit Performance and Management, 2006.

19.

Landau

Weisbrod

Gosling

Williges

Pumphrey

Fowler

Passenger Value of Time, Benefit-Cost Analysis and Airport Capital Investment Decisions, Volume 1: Guidebook for Valuing User Time Savings in Airport Capital Investment Decision Analysis. Airport Cooperative Research Program (ACRP), Washington, D.C., 2015.

20.

US Bureau of Labor Statistics (BLS). Consumer Price Index (CPI), 2020. https://www.bls.gov/cpi/.