Passenger Value of Time, Benefit-Cost Analysis and Airport Capital Investment Decisions, Volume 2: Final Report (2015)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

3 DATA COLLECTION AND MODELING RESULTS This chapter summarizes the Research Team’s process of data collection through the development and administration of a stated preference (SP) survey of airline travelers who had made recent domestic air trips within the U.S. The primary purpose of the SP survey was to provide suitable data to estimate the value of travelers’ willingness to pay for various individual time components that are part of air travel (as defined in Chapter 2). The components included airport ground access time, time to reach the terminal, time to reach and pass security, time to reach the gate from security, time spent waiting at the gate until boarding, in-flight time, and expected flight delay. It was expected that travelers place different values on the time associated with different aspects of the air trip. For example, time spent waiting in line to clear security may be more or less onerous than time spent getting to the airport, or time getting from security to the departure gate. Understanding the relationships between these different time components allows for better evaluations of potential airport infrastructure investment projects, instead of the current approach to project evaluation that uses the same values of time (VOT) regardless of the nature of the proposed capital investments.16 The SP survey questionnaire gathered information from airline travelers who were able to recall the details of their trip17. The questionnaire collected data on the recent air travel experience of survey respondents and used stated preference experiments to collect data that were used to estimate travelers’ willingness to pay for travel time savings for various time components under a range of different hypothetical conditions. The survey employed a computer-assisted self-interview (CASI) technique developed by Resource Systems Group, Inc. The stated preference survey instrument was customized for 16 The current approach for performing benefit-cost analyses for projects funded by FAA Airport Improvement Program grants provides single values for business travel, personal travel and a hybrid of business and personal travel, when the divisions of the two classes are not known. These values are applicable to all investment types. 17 The Research Team contracted with an online survey sample provider. This provider recruits and maintains a panel of 6.5 million potential survey respondents, cross referenced with demographic attributes that allowed for the selection of respondents that met criteria specified by the Research Team. This approach allowed the Research Team to specify a sample of recent air travelers, and assure adequate samples of business travelers and personal travelers. Because the survey provider begins the process with a panel willing to complete surveys, the traditional metric of response rate in a general population survey is not relevant. 3 Page 34

each respondent by presenting questions and modifying wording based on respondents’ previous answers. These dynamic survey features provide an accurate and efficient means of data collection and allow presentation of realistic alternative scenarios that correspond with the respondents’ reported experiences. The customized, proprietary software was programmed for online administration to targeted audiences. This chapter documents the development and administration of the survey questionnaire and presents the primary survey findings. The methodology and results of a discrete choice model estimation are reviewed to support these findings. This chapter then reports how the Research Team used the data from stated preference experiments to infer the implied values of time of the survey respondents. The resulting values of time are presented, and additional research is summarized to clarify and explore various issues that were identified in the course of analyzing the survey results or estimating the implied values of time. The following sections of this chapter review the: • Final survey questionnaire (Section 3.1) • Administration of the survey (Section 3.2) • Survey results (Section 3.3) • Model estimation to assign values of time to the survey results (Section 3.4) • The resulting values of time implied by the survey results (Section 3.5) • Additional research undertaken to clarify and explore various issues (Section 3.6) Section 3.7 provides a summary and conclusions from the research described in this chapter. The complete set of survey questions and detailed response tabulations appear as Appendices B.1 and B.2 to this report, and documentation of the additional research summarized in Section 3.6 is found in Appendix B.3. 3.1 Survey Questionnaire The survey questionnaire was designed to collect information about a recent air trip made by respondents within the U.S. and to estimate their sensitivities to paying for various trip attributes and air travel-related time components. The survey questions were grouped into four main sections: 1) Trip characteristics questions 2) Stated preference choice experiments (a statistical technique used to determine how people value different attributes that make up a single product or service) 3) Attitude and air travel background questions 4) Demographic questions Page 35

The complete set of survey questions as they appeared on-screen is included in Appendix B.1. Trip Characteristics Questions After being presented with basic instructions about how to navigate the computer-based survey questionnaire, respondents were asked a set of screening questions to ensure that they qualified for the survey. To qualify, respondents had to be 18 years of age or older, not employed by the aviation industry, and must have made a recent trip in the U.S. that used a purchased ticket. The respondents were told that a purchased airline ticket was one for which they or their employer paid for the airline fare and did not include any flights for which they received frequent flyer award tickets, airline vouchers and/or free flights. Respondents who indicated that their most recent air trip was more than six months prior to the survey were thanked for their time and terminated from completing the survey. Qualifying respondents were asked to focus on their most recent trip in one direction that met all of the screening criteria as they continued through the survey. This most recent trip, referred to as the respondent’s reference trip, formed the basis for the rest of the survey. Respondents were asked a series of questions regarding the specific details of their reference trip, including: • Departure date of the trip • Ticket acquisition and payment • Airfare • Trip origin and destination addresses • Origin and destination airports • Trip purpose • Party size • Number of nights away from home • Airport access mode(s) • Airport access time • Airport access cost • Parking lot type used (if any) and parking cost • Time to reach the airport terminal from parking lot or where dropped off • Anticipated total time that would be spent in the airport terminal • Number of bags checked • Baggage fee • Time from entering the terminal to reach security screening • Actual and anticipated security time Page 36

• Time from security screening to reach the gate area • Time spent in gate area before boarding commenced • Activities performed at the gate • Delay (if any) in departure and arrival of the flight • Scheduled local departure and arrival times • Preferred arrival time • Number of flight connections or stop-overs • Connection airport(s) • Airline(s) used • Preferred airlines – most preferred, second-most preferred, and third-most preferred Stated Preference Choice Experiments After completing the trip characteristics section of the questionnaire, respondents moved on to answer a series of stated preference choice experiments to estimate respondents’ travel preferences and behavior under hypothetical alternative trip scenarios. Initial versions of the stated preference experiments included a single design that included both flight itinerary attributes and airport access/in-airport time components combined. However, it was determined that this design would be too complex for respondents to complete. After working through several iterations of the design, the Research Team decided to create two separate sets of choice experiments– one covering attributes related to flight itinerary, and a second covering attributes related to airport ground access and in-airport time components up to the point of boarding the aircraft. The details of each respondent’s reference trip were used to build eight pair-wise choice experiments for each set, each experiment comprising two hypothetical but credible alternatives for making the trip that the respondent had described. Respondents chose which of each alternative scenario they would have preferred, had these been the only options available for their trip. Flight Itinerary Choice Experiments The flight itinerary choice experiments were designed to understand respondents’ preferences for different flight-related factors, including: • Flight time (airport-to-airport) • Airfare • Number of connections and connection time Page 37

• Air carrier • Departure and arrival time • Type of aircraft • On-time performance of the flight(s) • Average amount of delay for delayed flights Factors related to airport access, time spent in the airport terminal, or other non-flight- related factors were intentionally excluded to focus the respondent on the attributes relevant to the choice between different flight itineraries, in particular airfares, flight time, the number of connections, and expected delay. A series of eight trade-off scenarios described by these attributes were each presented to respondents as two hypothetical flight alternatives. The eight attribute values were varied independently across the two alternatives and respondents were asked to select the alternative they most preferred under the conditions that were presented. The attribute values presented in each alternative varied around a set of base values. To ensure that the scenarios were realistic, the trip characteristics of each respondent’s reference trip were used to calculate the base values for each attribute where applicable. The values for the attributes in each scenario were varied from the base value by multiplying or adding one of several factors to give the level required by the experimental design for that particular scenario. Attributes and Levels Airport-to-Airport Time. The flight time attribute represented the total airport-to-airport travel time, including time spent for taxi, takeoff and landing, in-flight time, and time spent at connecting airports, if applicable. The taxi, takeoff, landing, and in-flight times were combined to give the non-stop flight time that varied for this attribute, with connection times included as a separate component that varied independently from flight time. Non- stop flight time varied around a base value that was calculated as the distance between airports at 500 miles per hour plus a fixed value of 30 minutes to account for taxi, takeoff, and landing (Equation 1). The ground speed of 500 miles per hour was selected to represent an average of eastbound and westbound flights, and was used for both turboprop and jet aircraft. Each connection was assumed to add 30 minutes to the non-stop flight time to allow for the additional distance, landing, taxi, and takeoff. Equation 1. Base Flight Time Calculation Base flight time = 60*[airport-to-airport distance]/500mph+30 This base flight time value was varied according to the following levels: 1) [Base flight time * 0.9] + connection time + connections*30 Page 38

2) [Base flight time * 1.0] + connection time + connections*30 3) [Base flight time * 1.1] + connection time + connections*30 4) [Base flight time * 1.2] + connection time + connections*30 The amount of variation was chosen to present differences in travel time between the alternatives that are large enough to get respondents to trade-off against the attribute, but not so large as to be seen as unbelievable or unrealistic. Airfare. The airfare value was calculated from the respondent’s reported airfare using the following levels: 1) Reported airfare * 0.70 2) Reported airfare * 0.85 3) Reported airfare * 1.00 4) Reported airfare * 1.15 5) Reported airfare * 1.30 As with the flight time attribute, these levels were chosen to present enough variation to encourage respondents to trade-off against the attribute, but not as large as to be unrealistic or to dominate the stated preference exercises. Number of Connections. The number of connections presented varied depending on the respondent’s reported trip length. To ensure that the number of connections presented was realistic, respondents were classified into one of two groups using the base flight time that was calculated as part of the airport-to-airport time and presented in Equation 1. For the purposes of this attribute, short flights were defined as having a base flight time of four hours or less, while long flights were defined as having a base flight time of more than four hours. The number of connections for short flights varied between zero connections and one connection, while the number of connections for long flights varied between zero connections and two connections. Connection Time. The connection time attribute represented the total time spent at connecting airports, if applicable. This attribute was not presented for alternatives that presented non-stop flights. Connection time varied between 45 and 90 minutes per connection according to the following levels: 1) 45 minutes 2) 60 minutes 3) 75 minutes 4) 90 minutes Page 39

These levels were chosen to fall within the range of typical flight connection times, but also to present differences large enough to encourage respondents to trade-off against the attribute. Air Carrier. The air carrier, or airline, attribute was customized for each respondent based on their ranking of airlines earlier in the questionnaire. The air carrier attribute was varied according to the following levels: 1) Most preferred airline 2) 2nd-most preferred airline 3) 3rd-most preferred airline 4) Airline reported for reference trip (if not one of the 3 most preferred) Arrival Time. The arrival time attribute varied around each respondent’s preferred arrival time for their reference trip according to the following levels: 1) 2 hours before preferred arrival time 2) 1 hour before preferred arrival time 3) At preferred arrival time 4) 1 hour after preferred arrival time 5) 2 hours after preferred arrival time Because some respondents may report preferred arrival times very early or very late in the day, this attribute was validated to ensure that arrival times were not shown between the hours of midnight and 6:00 AM. While there are certainly commercial flights that operate during this time period, the Research Team felt that they are infrequent enough to not be a viable option for most respondents. If a flight were shown to arrive between midnight and 2:00 AM, the entire flight itinerary was shifted earlier such that the arrival time was shown to be before midnight. Similarly, if a flight were shown to arrive between 2:00 AM and 6:00 AM, the entire flight itinerary was shifted later such that the arrival time was shown to be after 6:00 AM. Departure Time. Departure time was included as an attribute to make the flight itinerary presentation realistic. However, this attribute did not vary independently from the other attributes. Departure time was calculated as the flight time subtracted from the arrival time (accounting for any differences in time zone between the arrival and departure airports). Departure time was also validated to ensure that no flights were shown to depart between the hours of midnight and 6:00 AM. Aircraft Type. Respondents were presented with different aircraft types depending on the length of their reference trip. To ensure that the types of aircraft presented were realistic, respondents were classified into one of three groups using the base flight time that was calculated as part of the airport-to-airport time and presented in Equation 1. The three Page 40

groups were flights less than two and a half hours, flights between two and a half and four hours, and flights greater than four hours. Flights with a base flight time of less than two and a half hours were presented with the following aircraft types: 1) Propeller 2) Regional jet 3) Standard jet Flights with a base flight time between two and a half and four hours were presented with the following aircraft types: 1) Regional jet 2) Standard jet Flights with a base flight time greater than four hours were presented with the following aircraft types: 1) Standard jet 2) Widebody jet On-time Performance. On-time performance was presented as the percent of flights that arrive on-time. On-time was defined as arriving at the gate earlier than 15 minutes after the scheduled arrival time, consistent with the definition of an on-time flight by the Federal Aviation Administration (FAA). On-time performance varied according to the following levels: 1) 90% of flights on-time 2) 80% of flights on-time 3) 70% of flights on-time 4) 60% of flights on-time 5) 50% of flights on-time Average Delay for Delayed Flights. In addition to on-time performance, a second delay- related attribute was presented that represented the average duration of delay for delayed flights. Because flights are not considered delayed until they are at least 15 minutes late, this attribute started at a value of 20 minutes of delay and varied up to 50 minutes according to the following levels: 1) 20 minutes 2) 30 minutes 3) 40 minutes 4) 50 minutes Page 41

Table 6 summarizes the attributes and levels used to generate the alternatives in the flight itinerary choice experiments. Page 42

Table 6. Stated Preference Attribute Levels– Flight Itinerary Choice Experiments Attribute Level Calculation Airport-to-airport timea 1 [Base flight time * 0.9] + connection time + connections*30 2 [Base flight time * 1.0] + connection time + connections*30 3 [Base flight time * 1.1] + connection time + connections*30 4 [Base flight time * 1.2] + connection time + connections*30 Airfare 1 Reported airfare * 0.7 2 Reported airfare * 0.85 3 Reported airfare * 1.0 4 Reported airfare * 1.15 5 Reported airfare * 1.30 Connections Flight <= 4 hours Flight > 4 hours 1 0 connections (non-stop) 0 connections (non-stop) 2 1 connection 1 connection 3 -- 2 connections Connection Time 1 45 minutes 2 60 minutes 3 75 minutes 4 90 minutes Carrier 1 Most preferred airline 2 2nd-most preferred airline 3 3rd-most preferred airline 4 Airline used for reference trip (if not one of the 3 most preferred) Arrival Timeb 1 2 hours before preferred arrival time 2 1 hour before preferred arrival time 3 At preferred arrival time 4 1 hour after preferred arrival time 5 2 hours after preferred arrival time Departure Timeb N/A Arrival time - travel time Aircraft Type Flight < 2.5 hours Flight 2.5 to 4 hours Flight > 4 hours 1 Propeller Regional Jet Standard Jet 2 Regional Jet Standard Jet Widebody Jet 3 Standard Jet -- -- On-time Performance 1 90% of flights on-time 2 80% of flights on-time 3 70% of flights on-time 4 60% of flights on-time 5 50% of flights on-time Average Delay for Delayed Flights 1 20 minutes 2 30 minutes 3 40 minutes 4 50 minutes a Base flight time was calculated as: 60*[airport-to-airport distance]/500mph+30. Page 43

b Arrival time and departure time were validated to ensure that flights were not shown to arrive or depart between the hours of 12:00 AM and 6:00 AM. If a calculated departure or arrival time did fall within this period, the flight itinerary was shifted earlier or later to avoid this period. The specific levels used in each stated preference experiment were determined by using an orthogonal experimental design18, which ensured that information was collected from respondents in a statistically efficient manner. The experimental design was created using Ngene, a software tool designed specifically for this type of application. Given the number of attributes and levels specified by the researcher, Ngene created an experimental design that minimized correlation in the variation of the attributes. The final solution reached by Ngene, given the attributes and levels specified above, was an experimental design with 128 experiments divided into sixteen groups of eight. One of the sixteen groups was randomly chosen for each respondent and the eight experiments were shown to respondents in a randomized order. The final design was evaluated by the Research Team, checked for correlation, and adjusted to eliminate stated preference experiments where one alternative would be clearly dominant over the other, which would result in little useful information obtained from that experiment. Airport Time Components Choice Experiments While the flight itinerary stated preference choice experiments were designed to understand how travelers value various attributes related to air travel, the airport time components choice experiments were designed to understand how travelers value time related to different segments of the on-ground portion of the trip. The on-ground portion of the air trip can be broken down into many different components. However, the Research Team necessarily had to balance the need to collect detailed information with the need to minimize respondent burden and the distinct possibility of presenting too many different time components for respondents to reasonably evaluate. After much discussion, the Research Team grouped the on-ground times into the following components: • Airport Ground Access Time: The travel time from the respondent’s origin location to their parking location, drop-off location, or transit stop/station at the airport. • Terminal Access Time: The time to reach the airport terminal from the parking location, drop-off location, or transit stop/station. • Check-in and Security Time: The time to check-in and check baggage (if applicable), reach security, wait in line, and clear security. Although check-in and security are two separate lines, many respondents do not check-in or check bags at the airport. In addition, the Research Team has no reason to believe that the disutility associated 18 Orthogonal design is an experimental design used to measure the comparative importance of various attributes. Page 44

with waiting to check-in would be different from the disutility for waiting for security screening. • Time to Reach the Gate Area: The time spent traveling to the gate area after clearing security. • Gate Time: The time after reaching the gate area until boarding commences. Respondents were presented with a series of trade-off scenarios, each of which comprised two alternatives describing the on-ground portion of the trip. The alternatives were described by the attributes listed above, as well as by airport access mode and airport access cost. By varying mode, cost, and the different time components independently, the experiments allowed respondents to demonstrate their sensitivities to each of the attributes by making trade-offs between them over the set of eight experiments. The attribute values presented in each question varied around a set of base values. To ensure that the scenarios were realistic, the trip characteristics of each respondent’s reference trip were used to calculate the base values for each attribute where applicable. The values for the attributes in each scenario were varied from the base value by multiplying or adding one of several factors to give the level required by the experimental design for that particular scenario. Attributes and Levels Airport Ground Access Mode. Respondents were presented with the following four modes to get to the airport: 1) Drive and park at the airport 2) Drive and get dropped-off 3) Taxi 4) Transit Access mode and associated costs can be very important to travelers. Indeed, some travelers will not consider certain modes for their trip at all. Because of this possibility, the mode attribute was constrained such that the same mode (which was the mode that the respondent reported for their trip) was presented in each alternative for a minimum of four of the eight stated preference experiments. This helped to prevent the mode and cost attributes from dominating the other travel time attributes and forced respondents to trade-off against these other attributes. Airport Ground Access Cost. Access costs presented in the airport time components choice experiments included parking costs, taxi fares, and transit fares. If the presented mode was drive and park at the airport, parking costs were presented at the following levels: 1) $10/day 2) $15/day Page 45

3) $20/day 4) $25/day If the presented mode was taxi, the fare was calculated as a $3.00 initial charge (flag drop) plus a per-mile fee that was varied to give the following levels: 1) $3.00 + $1.50/mile 2) $3.00 + $2.00/mile 3) $3.00 + $2.50/mile 4) $3.00 + $3.00/mile If the presented mode was transit, the fare was calculated at 13.5 cents per mile plus a base fare that was varied to give the following levels: 1) $2.00 + $0.135/mile 2) $4.00 + $0.135/mile 3) $6.00 + $0.135/mile 4) $8.00 + $0.135/mile No cost was shown for driving and getting dropped-off at the airport. While vehicle operating and maintenance costs are associated with this mode, the Research Team has found in past studies that travelers do not perceive these costs on a per-trip basis (RSG, 2008). Airport Ground Access Time. Airport ground access time varied around the respondent’s reported access time. Reported times were varied using the same levels for similar modes. For example, if the respondent reported an auto trip (drive and park, drive and drop-off, or taxi) and an automobile alternative was presented, or if the respondent reported a transit trip and a transit alternative was presented, access times varied according to the following levels: 1) Reported access time * 0.80 2) Reported access time * 0.90 3) Reported access time * 1.10 4) Reported access time * 1.20 If a respondent reported an auto trip and a transit alternative was presented, the travel times were increased to account for transit access, egress, wait time, and potentially slower travel times in minutes: 1) 15 + reported access time * 1.2 2) 15 + reported access time * 1.3 3) 15 + reported access time * 1.4 Page 46

4) 15 + reported access time * 1.5 The inverse of these levels was used if a respondent reported a transit trip and an automobile alternative was presented: 1) [Reported access time – 15] * 0.833 2) [Reported access time – 15] * 0.769 3) [Reported access time – 15] * 0.714 4) [Reported access time – 15] * 0.667 Terminal Access Time, Check-in and Security Time, Time to Reach the Gate Area, and Gate Time. After arriving at the airport, travel times were split into four categories: 1) Time to reach the airport terminal from the parking lot, drop-off location, or transit stop 2) Time to check-in and check baggage (if applicable), reach security, wait in line, and clear security 3) Time to reach the gate area after clearing security 4) Time waiting at the gate area until boarding commences These times were all varied around the respondent’s reported values according to the following levels: 1) Reported time - 40% 2) Reported time - 20% 3) Reported time + 20% 4) Reported time + 40% Minimum and maximum amounts of variation for the travel time attributes were set to ensure that all time attributes varied within the same general range. This helped prevent a single time attribute from dominating the others if the base time used in the attribute calculation was very large. The minimum variation for any given attribute was 4 minutes, while the maximum was set at 16 minutes. Table 7 summarizes the attributes and levels used to generate the alternatives in the airport time components choice experiments. As with the flight itinerary choice experiments, the experimental design was created using Ngene, a software tool designed specifically for this type of application. The final solution reached by Ngene also resulted in an experimental design with 128 experiments which were divided into sixteen groups of eight. One of the sixteen groups was randomly chosen for each respondent and the eight experiments were shown to respondents in randomized order. Page 47

The final design was evaluated by the Research Team, checked for correlation, and adjusted to eliminate stated preference experiments where one alternative would be clearly dominant over the other, which would result in little useful information obtained from that experiment. Table 7. Stated Preference Attribute Levels – Airport Time Components Choice Experiments Attribute Level Calculation Access Modea 1 Drive and Park 2 Drive and Dropped-off 3 Taxi 4 Transit Access Cost Drive and Park Dropped-Off Taxi Transit 1 $10/day 0 $3 + $1.50/mi $2 + $0.135/mi 2 $15/day 0 $3 + $2.00/mi $4 + $0.135/mi 3 $20/day 0 $3 + $2.50/mi $6 + $0.135/mi 4 $25/day 0 $3 + $3.00/mi $8 + $0.135/mi Access Time If Reported Mode = Auto or Taxi If Reported Mode = Transit Drive/Taxi Transit Drive/Taxi Transit 1 Reported time * 0.8 15 + Reported time * 1.2 [Reported time – 15] * 0.833 Reported time * 0.8 2 Reported time * 0.9 15 + Reported time * 1.3 [Reported time – 15] * 0.769 Reported time * 0.9 3 Reported time * 1.1 15 + Reported time * 1.4 [Reported time – 15] * 0.714 Reported time * 1.1 4 Reported time * 1.2 15 + Reported time * 1.5 [Reported time – 15] * 0.667 Reported time * 1.2 Terminal Timeb 1 Reported time - 40% 2 Reported time - 20% 3 Reported time + 20% 4 Reported time + 40% Security Timeb 1 Reported time - 40% 2 Reported time - 20% 3 Reported time + 20% 4 Reported time + 40% Security to Gate Timeb 1 Reported time - 40% 2 Reported time - 20% 3 Reported time + 20% 4 Reported time + 40% Gate Timeb 1 Reported time - 40% 2 Reported time - 20% 3 Reported time + 20% 4 Reported time + 40% a Respondents were presented with the same access mode in both alternatives in at least 4 of the 8 experiments. b A minimum variation of 4 minutes and a maximum variation of 16 minutes were used for these calculations. Page 48

Attitude and Air Travel Background Questions After completing the two sets of stated preference choice experiments, respondents answered a series of attitude questions that asked the level to which they agree or disagree with five statements related to how they buy air tickets and their general air travel attitudes. Next, they were asked about the number of round trips they have made by air within the U.S. in the past year for business or work and other purposes. They were also asked about their membership status in frequent flyer programs for their preferred airlines. To end this section, respondents reported whether they had missed a flight connection within the past two years. These questions were intended to provide additional context to help understand respondents’ choices in the choice experiments. These were also asked to enable the Research Team to explore factors that could explain differences in willingness to pay for travel time savings among the survey respondents. Demographic Questions In the final section of the survey, demographic information was collected in order to classify respondents, identify differences in responses or values of time among traveler segments, and confirm that the sample contained a diverse cross section of the traveling population. Both individual and household incomes were collected to identify if values of time are related to income and, if so, whether that relationship is better captured by individual wage rates or average household wage rates. All respondents answered demographic questions relating to the following areas: • Gender • Household size • Annual individual income before taxes • Annual household income before taxes • Employment status Before finishing the survey, respondents were given the opportunity to leave open-ended comments about the survey or air travel in general. 3.2 Survey Administration Quotes were solicited from three market research panel firms in October of 2012 to support the survey effort. Based on the understanding of sample request, overall price, and history of providing a dependable sample in similar efforts, the Research Team selected Research Now to produce a generally representative sample of air travelers nationwide in an efficient and timely way. The sampling plan was designed to include a sufficient range of travelers and trip types to support the statistical estimation of the coefficients of a choice model. Page 49

These differences can then be reflected in the structure and coefficients (value of time estimates) of the resulting choice models. Soft Launch Survey Administration The Research Team initiated a “soft launch” on February 20, 2013 and concluded it on February 22, 2013. A total of 105 respondents completed the survey during that time. As a result of the soft launch, the Research Team implemented two changes to the time components conjoint exercise: 1) The same airport access mode was presented for each alternative in a minimum of four of the eight experiments. This helped to prevent the mode and cost attributes from dominating the other travel time attributes and forced respondents to trade- off against these other attributes. 2) Minimum and maximum amounts of variation for the travel time attributes were set to ensure that all time attributes varied within the same general rage. This helped prevent a single time attribute from dominating the others if the base time used in the attribute calculation was very large. After completing the pretest and subsequent adjustments, the revised survey questionnaire was re-launched to collect the remaining data. Full Survey Administration The full survey commenced on April 1, 2013 and concluded on April 20, 2013. Respondents were recruited at random from the general panel population of all 50 states who met the following criteria (as described in Section 3.1, above): • Must be at least 18 years old • Must have made a domestic flight (within the contiguous U.S., Alaska and Hawaii) in the last 6 months • Must not be employed by the airline industry An additional 1,155 responses were collected during this time. Table 8 summarizes the data collection effort. Table 8. Data Collection Summary Purpose of Recent Trip Business Non-business Total Soft Launch 32 73 105 Full Launch 291 864 1,155 Total 323 937 1,260 Page 50

3.3 Survey Results A total of 1,260 respondents completed the online survey, including 105 responses from the soft-launch in February and 1,155 from the full launch in April. The number of records was reduced to 1,171 after completing data checks and outlier analysis during the model estimation work, which is described in more detail in Section 3.4 (Model Estimation) of this report. The descriptive analysis of the data presented here is based on the 1,171 respondents who were included in the model estimation work. For the purposes of statistical modeling and most other analysis in this report, respondents were grouped into two traveler market segments. The segments are described in Table 9 along with the respective number of respondents contained within each segment. Table 9. Traveler Market Segments Market Segment Respondents Trip Purpose Business Travelers 297 (25%) 1. Business 2. Attend conference Leisure Travelers 874 (75%) 1. Vacation 2. Visit friends or relatives 3. Attend school/college 4. Other Many of the tabulations presented in the remainder of this report and in Appendix B.2 are segmented by these categories. A complete set of tabulations of survey questions by market segment is shown in Appendix B.2. Trip Characteristics Questions At the beginning of the trip characteristics section, respondents were asked about where they purchased their air ticket. A large majority of respondents purchased their tickets online, with 70% of overall respondents reporting that they purchased their tickets online using the airline’s website, and 16% reporting that they purchased their tickets online from a web address other than an airline's website. Figure 4 shows reported airline ticket acquisition method segmented by business and leisure travelers. Page 51

Figure 4. Ticket Acquisition Method by Market Segment Overall, the majority of respondents reported that they personally paid for the air ticket (77%). However, only about 30% of business travelers paid for their tickets themselves, while 67% of them were reimbursed by their employers. On the other hand, 92% of leisure travelers paid for their own airfare. Figure 5 shows these differences by trip purpose. Figure 5. Payment Source by Market Segment Next, respondents were asked to enter their airfare. The median airfare for all respondents was $370 and the mean airfare was $438. Figure 6 shows the distribution of the reported airfare for all respondents. 50% 16% 6% 0% 28% 1% 77% 16% 3% 2% 1% 1% Online - using the airline website Online - from a site other than the airline From a travel agent (in person or by phone) By phone - directly from the airline Company travel office or similar organization Other, please specify: Business Leisure 30% 67% 1% 2% 92% 2% 5% 1% I paid personally My company paid or reimbursed me Family or friend Other, please specify: Business Leisure Page 52

Figure 6. Distribution of Reported Airfare Overall, respondents reported 172 distinct origin airports and 148 distinct destination airports throughout the country. The origin and destination airports, stratified by number of respondents, are displayed in Figure 7 and Figure 8, respectively. The top five origin airports reported by the sample respondents were Chicago O'Hare International Airport (ORD), Denver International Airport (DEN), Newark Liberty International Airport (EWR), Boston Logan International Airport (BOS), and Phoenix Sky Harbor International Airport (PHX). The top destination airports reported were Orlando International Airport (MCO), McCarran International Airport (LAS), Fort Lauderdale–Hollywood International Airport (FLL), Phoenix Sky Harbor International Airport (PHX), and Los Angeles International Airport (LAX). Table 10 and Table 11 compare the top five origin and destination airports with data from the Bureau of Transportation Statistics’ Airline Origin and Destination Survey (also known as DB1B) from 2008.19 The tables show that the survey sample matches the DB1B sample reasonably well, although warm weather destinations are slightly overrepresented in the survey sample. This is most likely due to the fact that the survey was fielded during the February to April timeframe, a popular season for visiting warm-weather destinations such as Florida, Arizona, and California. 19 The Research Team had the 2008 dataset processed, cleaned, and in formatted tables spinning on an SQL server from another study. Given that these data were being used as one point of comparison for the survey sample, the marginal benefit of using new data did not outweigh the investment of time in spending several days repeating the data processing tasks. 0% 5% 10% 15% 20% 25% $50 $450 $850 $1,250 $1,650 $2,050 $2,450 Pe rc en ta ge o f R es po nd en ts Reported Airfare Page 53

Figure 7. Location of Origin Airports by Number of Reported Trips Figure 8. Location of Destination Airports by Number of Reported Trips Page 54

Table 10. Comparison of Top 5 Origin Airports with DB1B Data Airport Sample Originating Passengers Sample % DB1B Originating Passengers DB1B % Chicago O’Hare International (ORD) 42 3.6% 898,921 3.6% Denver International (DEN) 41 3.5% 667,916 2.7% Newark Liberty International (EWR) 37 3.2% 713,308 2.9% Boston Logan International (BOS) 32 2.7% 641,235 2.6% Phoenix Sky Harbor (PHX) 29 2.5% 541,358 2.2% Table 11. Comparison of Top 5 Destination Airports with DB1B Data Airport Sample Originating Passengers Sample % DB1B Originating Passengers DB1B % Orlando International (MCO) 72 6.1% 1,172,747 4.9% McCarran International (LAS) 66 5.6% 1,416,725 5.9% Fort Lauderdale–Hollywood International (FLL) 55 4.7% 595,001 2.5% Phoenix Sky Harbor International (PHX) 47 4.0% 646,174 2.7% Los Angeles International (LAX) 39 3.3% 848,285 3.5% About 44% of all respondents traveled alone; however, 68% of business travelers traveled alone, compared to 35% of leisure travelers (Figure 9). Nearly half of the leisure travelers reported a party size of two. The average party size for business travelers was calculated as 1.47, whereas the average party size for leisure travelers came out to be 1.93. Figure 9. Party Size by Market Segment Leisure travelers reported longer visits, with over one-third (34%) reporting a stay of more than seven days compared to 6% of business travelers. As shown in Figure 10, a majority of business travelers (65%) stayed for 2-4 nights. 68% 24% 8% 35% 49% 16% 1 (traveled alone) 2 persons 3 persons or more Business Leisure Page 55

Figure 10. Length of Stay by Market Segment Table 12 shows respondents’ access modes to the airport. About 60% of business travelers and 41% of leisure travelers arrived at the airport by private vehicle and parked at or near the airport for the entire trip. Being driven and dropped-off was cited as the next most commonly used access mode by both groups, with taxi, door-to-door van, and limousine being used less often. A few respondents (24 in total) reported their access modes in their own words by selecting ‘other’ option. Where possible, these responses were re-coded to the correct pre-defined categories based on the respondents’ stated access modes. Table 12. Reported Access Mode by Market Segment (Multiple Responses were Allowed) Access Mode Business Leisure Count Percent Count Percent Private vehicle and parked at/near airport for entire trip 178 60% 358 41% Private vehicle and was dropped off at the airport (did not park) 68 23% 311 35% Taxi 15 5% 40 5% Shuttle bus or door-to-door van 12 4% 45 5% Limo/Town car 7 2% 33 4% Private vehicle parked at/near the airport for a short time and driven away by others 4 1% 32 4% Rental car 9 3% 18 2% Local city or regional bus 1 0% 11 1% Train (commuter rail, Amtrak, etc.) 1 0% 11 1% Rail transit, subway or streetcar 3 1% 20 2% Other 1 0% 3 0% There was very little reported use of public transit or rail by survey respondents for their access trips, particularly in the case of respondents making business trips. While this is consistent with airport access mode use at many airports, it has some important implications for the time component stated preference choice experiments, as discussed further below. The low use of rental cars reported by survey respondents reflects the fact that the airport access trips in question were for the outbound segment of a roundtrip and thus typically starting from the respondent’s home or place of work. Respondents provided their trip origin location either by directly entering the address or locating the origin on a map, either of which allowed the latitude and longitude for the trip origin to be determined. The latitude and longitude coordinates for each respondent’s trip 12% 2% 65% 31% 17% 34% 6% 34% Business Leisure 1 night or less 2-4 nights 5-7 nights More than 7 nights Page 56

origin and origin airport were used to calculate the airport ground access distance using a Google Maps driving directions algorithm. Respondents also reported their ground access time to the airport. Reported access times for business travelers ranged from about five minutes to four hours, with a mean reported access time of 50 minutes for this group. A similar trend was observed for leisure travelers. Table 13 shows the mean and median reported access time and Google-calculated access distance. Table 13. Ground Access Time and Ground Access Distance by Market Segment Region Access Time (minutes) Access Distance (miles) Mean Median Mean Median Business 50 40 32 22 Leisure 50 45 34 23 Figure 11 shows the distribution of reported terminal access time among all respondents, where the value shown on the horizontal axis is the upper bound of each time range. The terminal access time is defined as travel time from the location where the respondent parked, was dropped-off, or alighted from public transportation to the airport terminal entrance. The majority (56%) of respondents reported a terminal access time of 5 minutes or less. The mean and median terminal times reported were 8 minutes and 5 minutes, respectively. Figure 11. Distribution of Reported Terminal Access Time Figure 12 shows the distribution of the reported time from entering the terminal to reaching the security screening area or line at the airport, including any time spent checking in or dropping off checked baggage. The mean and median reported times to reach security were 14 minutes and 10 minutes, respectively. 0% 10% 20% 30% 40% 50% 60% 5 10 15 20 25 30 35 40 45 50 55 60 Terminal Access Time (mins) Page 57

Figure 12. Distribution of Reported Time to Security Screening from Entering the Terminal Figure 13 shows the distribution of the time it took respondents to clear security, including time spent waiting in line, placing hand baggage and other items on conveyer belts, and progressing through any mechanical and/or personal screening systems. As can be seen in the figure, there is more variability in reported security times. The mean and median reported times to clear security were 17 minutes and 15 minutes, respectively. Figure 13. Distribution of Reported Time to Clear Security Figure 14 shows the distribution of the time reported by respondents to reach the gate area after clearing security. 0% 5% 10% 15% 20% 25% 30% 5 10 15 20 25 30 35 40 45 50 55 60 Time to Reach Security (mins) 0% 5% 10% 15% 20% 25% 30% 5 10 15 20 25 30 35 40 45 50 55 60 More than 60Time to Clear Security (mins) Page 58

Figure 14. Distribution of Reported Time to Reach the Gate Area from Security Finally, Figure 15 shows the distribution of the time that respondents reported spending after reaching the gate area until flight boarding commenced, termed the gate time. Figure 15. Distribution of Reported Gate Time The mean and median times are highest for this time component at 49 minutes and 45 minutes, respectively. This is expected since this part of the airport journey is the least onerous and travelers can spend this time patronizing airport services and using mobile technology, such as cell phones or laptop computers. Also, this time represents the difference between the time that travelers allowed to reach the gate in case they 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 5 10 15 20 25 30 35 40 45 50 55 60 Time to Reach the Gate Area (mins) 0% 5% 10% 15% 20% 25% 10 20 30 40 50 60 70 80 90 100 110 120 Gate Time (mins) Page 59

encountered delays and the actual time that it took them to reach the gate, which is generally less. Airlines commonly advise passengers to arrive at the airport at least an hour before their flight, although it often takes much less than that to reach the gate area. The most common activities cited by respondents while waiting in the gate area include purchasing food and drinks, reading a book, magazine, newspaper or business documents, using a mobile device and checking e-mails. A large number of respondents reported on-time flight departure and flight arrival. In accordance with the Federal Aviation Administration (FAA) definition, the flights were considered on-time if they departed earlier than 15 minutes after the originally scheduled time or if they arrived earlier than 15 minutes after the originally scheduled time. Ninety percent of flight departures and 91% of flight arrivals at the destination were reported on- time (Figure 16). Figure 16. Flight Delay at Origin and Destination Respondents were more likely to travel on flights departing earlier in the day. Approximately 64% of respondents traveled on flights departing before noon, with a similar proportion for both trip purpose segments (Figure 17). This high percentage of flight departures before noon can likely be attributed to the fact that the respondents were reporting the outward leg of their trip. Figure 17. Flight Departure Time by Market Segment Figure 18 shows the distribution of preferred arrival time of travelers with respect to actual arrival time. The majority of the sample (60%) would have preferred an arrival time within one hour of their actual arrival time. About one-third of respondents would have preferred 91% 90% 2% 3% 7% 7% Destination Flight Origin Flight No delay Up to 30 minutes More than 30 minutes 35% 37% 27% 28% 29% 25% 9% 11% Business Leisure Early AM (Midnight-8:59 AM) Late AM (9:00 AM-11:59 PM) Early PM (Noon-4:59 PM) Late PM (5:00 PM-11:59 PM) Page 60

to arrive more than an hour earlier than their actual arrival time, whereas only 6% would have preferred to have arrived more than an hour later than the actual arrival time. The distribution is somewhat similar for both business and leisure travelers. Figure 18. Distribution of Preferred Arrival Time with Respect to Actual Arrival Time Overall, about two-third of respondents reported a non-stop flight (Figure 19). The mean connection time for travelers who reported at least one connection was calculated as 85 minutes and the median connection time was calculated as 70 minutes. Figure 19. Number of Connections by Market Segment Because airline choice can be an important factor in flight itinerary selection, respondents were also asked to report their three most preferred airlines. Table 14 presents each airline’s frequency of being ranked as one of the top three choices. The top three most frequently preferred airlines were Southwest Airlines, Delta Airlines, and United Airlines. Responses to this question were used to construct levels for the air carrier variable shown in the flight itinerary stated preference choice experiments later in the survey. 17% 23% 15% 12% 60% 59% 4% 3% 4% 3% Business Leisure More than 2 hours before actual arrival time Between 1 hour and 2 hours before actual arrival time Within 1 hour of actual arrival time Between 1 hour and 2 hours after actual arrival time More than 2 hours after actual arrival time 63% 36% 1% 66% 32% 2% 0% 10% 20% 30% 40% 50% 60% 70% No connections 1 connection 2 connections Business Leisure Page 61

Table 14. Respondents’ Airline Preference Airline Survey Respondents Including Airline in Set of Preferred Airlines Southwest Airlines 602 Delta Airlines 597 United Airlines 464 American Airlines 381 JetBlue Airways 244 US Airways 233 Alaska Airlines 139 Continental Airlines 139 AirTran Airways 135 Frontier Airlines 107 Virgin America 101 Hawaiian Airlines 84 United Express 54 Delta Connection 50 Allegiant Air 35 Air Canada 34 US Airways Express 20 Sun Country Airlines 19 America West Airlines 17 Horizon Air 17 American Eagle 16 Spirit Airlines 13 Express Jet 4 SkyWest Airlines 3 WestJet 3 Atlantic Southeast Airlines 1 Cape Air 1 Questions on Attitudes to Air Travel and Recent Experience Upon completing the stated preference experiments, respondents were asked to answer a series of questions related to their attitudes about flight itinerary selection and their general level of experience with air travel. As noted in Section 3.1 above, these questions were asked to enable the Research Team to conduct additional analyses of factors that could explain differences in willingness to pay for travel time savings among the survey respondents. When presented with a series of questions regarding their attitudes about how they buy air tickets and other air travel behavior, respondents were most likely to agree that they try to fly without checked baggage whenever possible (60%). Conversely, respondents were less Page 62

likely to fly less for environmental reasons, as shown in Figure 20. Respondents also indicated they generally shop for the cheapest flights and do not consider other factors (48%). Nearly, 47% agreed that they regularly search websites for cheap flights and sometimes will fly if they see a bargain. A majority of travelers (54%) disagreed with the statement that the recent changes to airport security have discouraged them from flying (20% agreed with the statement). Figure 20. Air Journey Attitudes Statements When they were asked about the number of round trips they have made by air within the U.S. in the past year for business or work purposes, more than 50% cited they had not made a business/work related trip in the past year. On the other hand, about 81% of respondents had made between one and five leisure or non-work related trips in the past year (Figure 21). Figure 21. Number of Round Trips in the Past Year by Trip Purpose Respondents were also asked about their membership status in frequent flyer programs for their preferred airlines. A majority of the respondents reported that they were enrolled in 60% 48% 47% 20% 4% 12% 18% 22% 27% 25% 29% 34% 31% 54% 71% I try to fly without checked baggage whenever possible I generally shop for the cheapest flights and do not consider other factors I regularly search websites for cheap flights and sometimes will fly if I see a bargain Recent changes to airport security have discouraged me from flying I am trying to fly less for environmental reasons Agree Neutral Disagree 51% 7% 10% 15% 9% 25% 16% 41% 6% 8% 8% 4% Business Travel Frequency Leisure Travel Frequency No trip 1 trip 2 trips 3-5 trips 6-8 trips More than 8 trips Page 63

standard memberships for their preferred airlines. For the respondents’ most preferred airline, approximately 85% of business travelers and 80% of leisure travelers were members of a frequent flier program. Finally, 16% of respondents reported that they had missed a flight connection in the last two years. Demographic Questions Of the 1,171 total respondents, slightly over half were female (55%). The median age of business respondents fell in the category of 45-54 years old and the median age category for leisure travelers was 55-64 years. About 44% of respondents reported being employed full-time, while 32% reported that they were retired. Figure 22. Age by Market Segment The distribution of annual individual income reported by the respondents is shown in Figure 23. The median individual income of business respondents was in the $75,000 to $99,999 income category and the median individual income for leisure travelers fell in the $50,000 to $74,999 category. It should be noted that the sample demographics are not proportional to the U.S. population as a whole. The sample has a slightly higher proportion of female respondents, is older on average, and has higher incomes on average. This is not surprising as the segment of the U.S. population that travels by air can be expected to be significantly different than the general population. In past airline research conducted by the Research Team, we have found that air travel samples are consistently older and have higher incomes than the general population. Unfortunately, there are no publicly available data that give reliable, 3% 9% 14% 26% 25% 20% 3% 4% 9% 9% 15% 27% 32% 4% 18-24 25-34 35-44 45-54 55-64 65-74 75 years or older Business Leisure Page 64

nationwide, current demographic distributions for air passengers making domestic air trips to use as a point of comparison. There are also pronounced longitudinal trends in the air travel passenger mix that will not be apparent from current demographic data. Figure 23. Annual Individual Income by Market Segment 3.4 Model Estimation Statistical analysis and discrete choice model estimation were carried out using the stated preference survey data. Responses from the stated preference scenarios were expanded into a dataset containing 16 observations for each respondent (eight observations from the flight itinerary stated preference survey and eight from the time components stated preference survey). Each observation included the values of the attributes presented in each alternative, the respondent’s chosen alternative, and additional background information about the respondent’s reported trip and demographic characteristics. This dataset formed the basis for the discrete choice model estimation described in this section of the report. The statistical estimation and specification testing were completed using a conventional maximum likelihood procedure that estimated a set of coefficients for a multinomial logit 1% 1% 2% 3% 8% 22% 20% 24% 10% 3% 5% 4% 3% 4% 9% 10% 23% 18% 18% 5% 3% 3% Under $10,000 $10,000 - $19,999 $20,000 - $29,999 $30,000 - $39,999 $40,000 - $49,999 $50,000 - $74,999 $75,000 - $99,999 $100,000 - $149,999 $150,000 - $199,999 $200,000 - $249,999 $250,000 or more Business Leisure Page 65

(MNL) model20 for the two traveler market segments of those making business and those making personal trips. The model coefficients provide information about the respondents’ sensitivities to the attributes that were tested in the trade-off scenarios. The sensitivities were used to calculate the implied values of various time components discussed in Section 3.5. Identification of Outliers The choice data were screened to ensure that all observations included in the model estimation represented realistic trips and reasonable trade-offs in the stated preference exercises. In particular, inputs that were used to build the stated preference experiments were reviewed, including (the categories are not mutually exclusive): • Comparing reported ground access time to ground access distance (24 cases) • Comparing reported flight time to the distance between airports and the number of connections (18 cases) • Comparing reported access costs to access mode and distance (11 cases) • Identifying invalid origins and destinations (same origin and destination airport, international origin or destination airport, etc.) (34 cases) • Evaluating reported values for airport time components (combined terminal access time, time to reach and clear security, and time to reach the gate area greater than 2 hours) (10 cases) • Screening for inappropriate comments provided by respondents at the end of the survey that indicated the respondents were not paying attention or not taking the survey seriously (2 cases) Significant changes were made in the airport time components stated preference choice experiments after the soft launch. Therefore, stated preference data from the soft launch was not used for estimating choice models for the time components choice experiments. Additionally, some choice experiments were removed from the time components choice data used for model estimation where responses from the soft launch showed inconsistent choice behavior or respondents selected options with very high access costs. 20 The multinomial logit model has the general form , where p(i) is the probability that mode i will be chosen and Ui is the “utility” of mode i, a function of service and other variables. See, for example, M. E. Ben-Akiva and S. R. Lerman, Discrete Choice Analysis, MIT Press, 1985 for details on the model structure and statistical estimation procedures. p( i) = U ie Uje AllModes ∑ Page 66

After completing the data cleaning and outlier analysis, a total of 1,171 survey responses were used to conduct the discrete choice model estimation and specification testing for the flight itinerary choice experiments and 1,072 survey responses were used to conduct the discrete choice model estimation and specification testing for the airport time components choice experiments.21 Multinomial Logit Model Estimation and Specification Using the cleaned dataset, several discrete choice model specifications were tested to explain the respondent choices in both the flight itinerary and the airport time components choice experiments. The statistical estimation and specification testing were completed using a conventional maximum likelihood procedure that estimated a set of coefficients for a multinomial logit (MNL) model. The MNL model estimates a choice probability for each alternative presented in the stated preference choice experiments. The alternatives are represented in the model by observed utility equations of the formula: U1 = β 1X1 + β 2X2 + ... + β nXn Where each X represents a variable included in the information presented to the respondents in the choice experiments and each β is a coefficient estimated by the model that represents the sensitivity of the respondents in the sample to the corresponding variable. Several utility equation structures were tested using the variables included in the stated preference scenarios. Other background variables were tested as potential covariates to identify systematic differences in behavior, such as: • Trip purpose (business or leisure) • Party size • Flight distance • Income (individual and household) • Reimbursement for travel expenses As could be expected, many of these interacting variables did explain behavioral differences in the sample. On average, business travelers had higher willingness-to-pay (WTP) values than non-business travelers, those who were being reimbursed for their travel expenses had higher WTP values than those who were not, high income respondents had higher WTP values than low-income respondents, and multiple-person parties had higher WTP values than those traveling alone. 21 Note that in the descriptive analysis presented in the previous section, data from both the soft launch and the full launch were used with a total of 1,171 respondents. Page 67

It is not immediately obvious why someone in a multiple-person party would have a higher WTP than someone traveling alone, other factors being equal. However, it is possible that some respondents making an air trip with others were paying the airfare for the other members of the travel party as well as their own and took this into account in making trade- offs between airfare and travel time. Including all of these interactions in the MNL model specification becomes difficult as the sample gets divided among several dimensions. As a result, only the most significant interactions that were identified, including trip purpose and individual income, are included in the model results presented in this section of the report. Further discussion of several of these issues is included in Section 3.6 below. The final flight itinerary choice model included the following variables: 1) Flight Time: Origin to destination travel time included in-flight time and any connection time(s). 2) Airfare: The airfare was shown either as the one-way or roundtrip fare depending on the respondent’s reference trip. 3) Number of Connections: This variable was entered as a categorical variable with no connections as the reference category and one connection or two connections as other discrete categories. 4) Type of Aircraft: This was also entered as a categorical variable with four categories: propeller (reference), regional, standard, and widebody aircraft. 5) Expected Delay: This measure was calculated as the product of percentage of delayed flights from the on-time performance and the average amount of delay for delayed flights. For example, if the percentage of flights that were shown as on-time in a given alternative scenario was 60% (i.e. 40% of flights were delayed) and the corresponding delay shown was 30 minutes, the expected delay was calculated as 0.4 x 30 = 12 minutes. 6) Arrival Time: The difference between the arrival time shown in the stated preference experiment scenarios and the respondent’s preferred arrival time was assigned to one of five categories. The reference category was when the arrival time in the stated preference experiment scenario fell within one hour before or after the preferred arrival time. The other four categories were defined as: o More than 2 hours before the preferred arrival time o Between 1 hour and 2 hours before the preferred arrival time o Between 1 hour and 2 hours after the preferred arrival time o More than 2 hours after the preferred arrival time 7) Airline: The airlines shown in the stated preference experiment scenarios were selected from the most-preferred airline, the second-most-preferred airline, and the third-most-preferred airline or the airline used in the actual trip (termed the current Page 68

carrier) if this was not either of the three most preferred airlines. This was coded as a three-level categorical variable. Although the Research Team initially estimated the current carrier separately, the coefficient estimate for the current carrier was not statistically different from the coefficient for the third-most-preferred airline. The final airport time components choice model included the following variables: 1) Ground Access Time: This was defined as the travel time from respondent’s trip origin location to where he/she parked, or got dropped off. 2) Ground Access Cost: This was defined as the one-way “out-of-pocket” access cost for the access trip. If the drive and park option was shown, the ground access cost included the parking cost as well. 3) Ground Access Mode: The ground access modes shown in the stated preference experiment scenarios were selected from four options: drive and park, drive and dropped off, transit, and taxi. This was coded as a four-level categorical variable. 4) Terminal Access Time: This was defined as the time to reach the airport terminal from the location where parked, dropped off, or alighted from public transportation. 5) Check-in and Security Time: This time component was defined as the time to check- in and check baggage (if applicable), reach security, wait in line, and pass security. 6) Time to Reach the Gate Area: This was defined as the time to reach the gate area after passing security. 7) Gate Time: The gate time was defined as the time after reaching the gate area until boarding commenced. Model Coefficients The MNL model results and resulting values of time are presented below for business and leisure travelers. Table 15 and Table 16 present the results for the flight itinerary choice models, while Table 17 and Table 18 include the results for the airport time components choice models. In the second set of models, the cost variable was interacted with individual income to identify the relationship between the willingness to pay for travel time savings and traveler income. Separate specifications were tested using household income and individual income in the development of the analysis, and individual income was found to provide the best explanation of the respondent choice behavior. Table 19 and Table 20 present the results with income effects for the flight itinerary choice models, while Table 21 and Table 22 include the results for the airport time components choice models. The coefficient values, robust standard errors, robust t-statistics, and general model statistics are also presented (and defined below). The coefficient values are the values estimated by the choice model that represent the relative importance of each of the variables. It should be noted that these values are unit- Page 69

specific and the units must be accounted for when comparing coefficients (units are specified in Tables 15-22, below). The tables below identify key diagnostics for each model, starting with the standard error. The standard error is a measure of error around the mean coefficient estimate. The t-statistic is the estimated value of the coefficient divided by the standard error, which can be used to evaluate statistical significance. A t-statistic numerically greater than 1.96 (positive or negative) indicates that the estimated value of the coefficient is statistically different from 0 (unless otherwise reported) at the 95% significance level. The model fit statistics that are presented include the number of observations, the number of estimated parameters, the initial log-likelihood, the log-likelihood at convergence, rho- squared, and adjusted rho-squared. The log-likelihood is a model fit measure that indicates how well the model predicts the choices observed in the data. The null log-likelihood is the measure of the model fit with coefficient values of zero (which implies that each option is equally likely to be chosen). The final log-likelihood is the measure of model fit with the final coefficient values at model convergence. A value closer to zero indicates better model fit. The log-likelihood cannot be evaluated independently, as it is a function of the number of observations, the number of alternatives, and the number of parameters in the choice model. The rho-square model fit measure accounts for this to some degree by evaluating the difference between the null log-likelihood and the final log-likelihood at convergence. The adjusted rho-square value takes into account the number of parameters estimated in the model. It should be noted that there are significant differences between an R2 value commonly used to evaluate linear regression models and the adjusted rho-square value used in MNL models. R2 in a linear regression model gives the proportion of the variance in the dependent variable which is explained by the model. However, the adjusted rho-square depends on the ratio of the beginning and ending log-likelihood functions and thus measures the improvement from the null model to the fitted model. Therefore, the adjusted rho-square value in MNL models is expected to be much smaller than the expected R2 value in linear regression models. A value of 0.12 or higher for the adjusted rho-square is considered good for MNL models with panel data (data with multiple responses by each respondent). Page 70

Table 15. MNL Model Results – Flight Itinerary SP Experiments (Business Trips) Parameters Units Description Value Robust Std. Error Robust t- stat ΒFlight_time Min Flight time -0.00522 0.00126 -4.15 ΒFare $ Airfare -0.00614 0.000545 -11.27 ΒConnection_0 (0,1) Number of connections – 0 (reference) --Fixed-- ΒConnection_1 (0,1) Number of connections – 1 -0.59 0.172 -3.43 ΒConnection_2 (0,1) Number of connections – 2 -1.73 0.377 -4.58 ΒAircraft_propeller (0,1) Type of aircraft – propeller (reference) --Fixed-- ΒAircraft_regional (0,1) Type of aircraft – regional jet 0.611 0.127 4.82 ΒAircraft_standard (0,1) Type of aircraft – standard jet 0.763 0.122 6.25 ΒAircraft_widebody (0,1) Type of aircraft – widebody jet 0.722 0.159 4.55 ΒExpected_delay Min Expected value of delay (probability of delay x average amount of delay for delayed flights) -0.0293 0.00473 -6.19 ΒArrival_preferred (0,1) Within +/- 1 hour of preferred arrival time (reference) --Fixed-- ΒArrival_early1 (0,1) More than 2 hours before preferred arrival time 0.25* 0.385 0.65 ΒArrival_early2 (0,1) Between 1 hour and 2 hours before preferred arrival time -0.176* 0.121 -1.45 ΒArrival_late1 (0,1) More than 2 hours after preferred arrival time -0.334* 0.747 -0.45 ΒArrival_late2 (0,1) Between 1 hour and 2 hours after preferred arrival time -0.127* 0.119 -1.07 ΒCarrier_1 Most preferred carrier 0.379 0.0887 4.27 ΒCarrier_2 2 nd most preferred carrier 0.13* 0.0741 1.75 ΒCarrier_3 3rd most preferred/current carrier (reference) --Fixed-- * Not significant at 95% confidence level Fit Statistics Number of parameters: 14 Number of observations: 2,376 Number of individuals: 297 Null log-likelihood: -1646.918 Final log-likelihood: -1299.118 Rho-square: 0.211 Adjusted rho-square: 0.203 Page 71

Table 16. MNL Model Results – Flight Itinerary SP Experiments (Leisure Trips) Parameters Units Description Value Robust Std. Error Robust t-stat ΒFlight_time Min Flight time -0.00572 0.00073 -7.83 ΒFare $ Airfare -0.00983 0.000642 -15.31 ΒConnection_0 (0,1) Number of connections – 0 (reference) --Fixed-- ΒConnection_1 (0,1) Number of connections – 1 -0.772 0.0908 -8.50 ΒConnection_2 (0,1) Number of connections – 2 -2.3 0.25 -9.19 ΒAircraft_propeller (0,1) Type of aircraft – propeller (reference) --Fixed-- ΒAircraft_regional (0,1) Type of aircraft – regional jet 0.663 0.0943 7.04 ΒAircraft_standard (0,1) Type of aircraft – standard jet 0.725 0.0901 8.05 ΒAircraft_widebody (0,1) Type of aircraft – widebody jet 0.674 0.106 6.35 ΒExpected_delay (0,1) Expected value of delay (probability of delay x average amount of delay for delayed flights) -0.0202 0.00288 -7.01 ΒArrival_preferred (0,1) Within +/- 1 hour of preferred arrival time (reference) --Fixed-- ΒArrival_early1 (0,1) More than 2 hours before preferred arrival time -0.286* 0.233 -1.23 ΒArrival_early2 (0,1) Between 1 hour and 2 hours before preferred arrival time -0.138 0.0646 -2.13 ΒArrival_late1 (0,1) More than 2 hours after preferred arrival time -0.98 0.289 -3.39 ΒArrival_late2 (0,1) Between 1 hour and 2 hours after preferred arrival time -0.118* 0.068 -1.74 ΒCarrier_1 Most preferred carrier 0.372 0.0536 6.94 ΒCarrier_2 2 nd most preferred carrier 0.0816* 0.0455 1.79 ΒCarrier_3 3rd most preferred/current carrier (reference) --Fixed-- * Not significant at 95% confidence level Fit Statistics Number of parameters: 14 Number of observations: 6,992 Number of individuals: 874 Null log-likelihood: -4846.485 Final log-likelihood: -3517.118 Rho-square: 0.274 Adjusted rho-square: 0.271 Page 72

Table 17. MNL Model Results – Airport Time Components SP Experiments (Business Trips) Parameters Units Description Value Robust Std. Error Robust t-stat ΒTime_access Min Ground access time -0.0128 0.00422 -3.02 ΒCost_access $ Ground access cost (including parking) -0.0413 0.00473 -8.74 ΒMode_transit (0,1) Ground access mode – transit (reference) --Fixed-- ΒMode_drop-off (0,1) Ground access mode – drive and dropped- off 1.14 0.21 5.44 ΒMode_drive (0,1) Ground access mode – drive and park 1.27 0.25 5.10 ΒMode_taxi (0,1) Ground access mode – taxi 0.301* 0.224 1.34 ΒTime_terminal Min Terminal access time -0.0233 0.00735 -3.17 ΒTime_security Min Check-in and security time -0.0256 0.00431 -5.95 ΒTime_security_gate Min Time to reach gate area -0.0222 0.00688 -3.23 ΒTime_gate Min Gate time -0.0141 0.00358 -3.93 * Not significant at 95% confidence level Fit Statistics Number of parameters: 9 Number of observations: 2,144 Number of individuals: 268 Null log-likelihood: -1486.108 Final log-likelihood: -1183.802 Rho-square: 0.203 Adjusted rho-square: 0.197 Page 73

Table 18. MNL Model Results – Airport Time Components SP Experiments (Leisure Trips) Parameters Units Description Value Robust Std. Error Robust t-stat ΒTime_access Min Ground access time -0.0202 0.00251 -8.05 ΒCost_access $ Ground access cost (including parking) -0.0715 0.00476 -15.01 ΒMode_transit (0,1) Ground access mode – transit (reference) --Fixed-- ΒMode_drop-off (0,1) Ground access mode – drive and dropped- off 1.13 0.145 7.79 ΒMode_drive (0,1) Ground access mode – drive and park 1.23 0.186 6.62 ΒMode_taxi (0,1) Ground access mode – taxi -0.359 0.15 -2.39 ΒTime_terminal Min Terminal access time -0.031 0.00449 -6.90 ΒTime_security Min Check-in and security time -0.0339 0.00284 -11.93 ΒTime_security_gate Min Time to reach gate area -0.0272 0.00411 -6.62 ΒTime_gate Min Gate time -0.021 0.00218 -9.66 Fit Statistics Number of parameters: 9 Number of observations: 6,385 Number of individuals: 804 Null log-likelihood: -4425.745 Final log-likelihood: -3035.381 Rho-square: 0.314 Adjusted rho-square: 0.312 Page 74

Table 19. MNL Model Results with Individual Income – Flight Itinerary SP Experiments (Business Trips) Parameters Units Description Value Robust Std. Error Robust t-stat ΒFlight_time Min Flight time -0.00542 0.00124 -4.36 ΒFare_inc_level1 $ Airfare if Income less than $75,000, else 0 -0.00966 0.00108 -8.91 ΒFare_inc_level2 $ Airfare if Income between $75,000 and $199,999, else 0 -0.00552 0.000628 -8.79 ΒFare_inc_level3 $ Airfare if Income $200,000 or more, else 0 -0.00322 0.000886 -3.64 ΒConnection_0 (0,1) Number of connections – 0 (reference) --Fixed-- ΒConnection_1 (0,1) Number of connections – 1 -0.62 0.17 -3.64 ΒConnection_2 (0,1) Number of connections – 2 -1.76 0.383 -4.60 ΒAircraft_propeller (0,1) Type of aircraft – propeller (reference) --Fixed-- ΒAircraft_regional (0,1) Type of aircraft – regional jet 0.636 0.13 4.90 ΒAircraft_standard (0,1) Type of aircraft – standard jet 0.781 0.124 6.30 ΒAircraft_widebody (0,1) Type of aircraft – widebody jet 0.747 0.162 4.60 ΒExpected_delay Min Expected value of delay (probability of delay x average amount of delay for delayed flights) -0.03 0.00481 -6.24 ΒArrival_preferred (0,1) Within +/- 1 hour of preferred arrival time (reference) --Fixed-- ΒArrival_early1 (0,1) More than 2 hours before preferred arrival time 0.148* 0.424 0.35 ΒArrival_early2 (0,1) Between 1 hour and 2 hours before preferred arrival time -0.193* 0.125 -1.54 ΒArrival_late1 (0,1) More than 2 hours after preferred arrival time -0.245* 0.699 -0.35 ΒArrival_late2 (0,1) Between 1 hour and 2 hours after preferred arrival time -0.13* 0.121 -1.07 ΒCarrier_1 Most preferred carrier 0.408 0.0918 4.45 ΒCarrier_2 2 nd most preferred carrier 0.136* 0.0759 1.80 ΒCarrier_3 3 rd most preferred/current carrier (reference) --Fixed-- * Not significant at 95% confidence level Fit Statistics Number of parameters: 16 Number of observations: 2,376 Number of individuals: 297 Null log-likelihood: -1646.918 Final log-likelihood: -1269.127 Rho-square: 0.229 Adjusted rho-square: 0.220 Page 75

Table 20. MNL Model Results with Individual Income – Flight Itinerary SP Experiments (Leisure Trips) Parameters Units Description Value Robust Std. Error Robust t-stat ΒFlight_time Min Flight time -0.00606 0.00073 -8.31 ΒFare_inc_level1 $ Airfare if Income less than $75,000, else 0 -0.0121 0.00076 -15.86 ΒFare_inc_level2 $ Airfare if Income between $75,000 and $199,999, else 0 -0.00883 0.000868 -10.17 ΒFare_inc_level3 $ Airfare if Income $200,000 or more, else 0 -0.00381 0.00135 -2.82 ΒConnection_0 (0,1) Number of connections – 0 (reference) --Fixed-- ΒConnection_1 (0,1) Number of connections – 1 -0.789 0.0931 -8.48 ΒConnection_2 (0,1) Number of connections – 2 -2.32 0.254 -9.12 ΒAircraft_propeller (0,1) Type of aircraft – propeller (reference) --Fixed-- ΒAircraft_regional (0,1) Type of aircraft – regional jet 0.682 0.095 7.17 ΒAircraft_standard (0,1) Type of aircraft – standard jet 0.743 0.0905 8.21 ΒAircraft_widebody (0,1) Type of aircraft – widebody jet 0.698 0.107 6.54 ΒExpected_delay (0,1) Expected value of delay (probability of delay x average amount of delay for delayed flights) -0.0216 0.0029 -7.44 ΒArrival_preferred (0,1) Within +/- 1 hour of preferred arrival time (reference) --Fixed-- ΒArrival_early1 (0,1) More than 2 hours before preferred arrival time -0.292* 0.236 -1.24 ΒArrival_early2 (0,1) Between 1 hour and 2 hours before preferred arrival time -0.148 0.0657 -2.26 ΒArrival_late1 (0,1) More than 2 hours after preferred arrival time -0.984 0.286 -3.44 ΒArrival_late2 (0,1) Between 1 hour and 2 hours after preferred arrival time -0.13* 0.0688 -1.89 ΒCarrier_1 Most preferred carrier 0.378 0.054 7.00 ΒCarrier_2 2 nd most preferred carrier 0.0879* 0.0462 1.90 ΒCarrier_3 3 rd most preferred/current carrier (reference) --Fixed-- * Not significant at 95% confidence level Fit Statistics Number of parameters: 16 Number of observations: 6,992 Number of individuals: 874 Null log-likelihood: -4846.485 Final log-likelihood: -3448.519 Rho-square: 0.288 Adjusted rho-square: 0.285 Page 76

Table 21. MNL Model Results with Individual Income – Airport Time Components SP Experiments (Business Trips) Parameters Units Description Value Robust Std. Error Robust t-stat ΒTime_access Min Ground access time -0.0136 0.00436 -3.13 ΒCost_access_inc_level1 $ Ground access cost (including parking) if Income less than $75,000, else 0 -0.0586 0.0103 -5.68 ΒCost_access_inc_level2 $ Ground access cost (including parking) if Income between $75,000 and $199,999, else 0 -0.0383 0.00491 -7.80 ΒCost_access_inc_level3 $ Ground access cost (including parking) if Income $200,000 or more, else 0 -0.0212 0.00939 -2.26 ΒMode_transit (0,1) Ground access mode – transit (reference) --Fixed-- ΒMode_drop-off (0,1) Ground access mode – drive and dropped- off 1.13 0.211 5.36 ΒMode_drive (0,1) Ground access mode – drive and park 1.32 0.254 5.19 ΒMode_taxi (0,1) Ground access mode – taxi 0.30* 0.22 1.37 ΒTime_terminal Min Terminal access time -0.0232 0.00735 -3.16 ΒTime_security Min Check-in and security time -0.0271 0.00444 -6.11 ΒTime_security_gate Min Time to reach gate area -0.0221 0.0068 -3.25 ΒTime_gate Min Gate time -0.0139 0.00358 -3.88 * Not significant at 95% confidence level Fit Statistics Number of parameters: 11 Number of observations: 2,144 Number of individuals: 268 Null log-likelihood: -1486.108 Final log-likelihood: -1167.710 Rho-square: 0.214 Adjusted rho-square: 0.207 Page 77

Table 22. MNL Model Results with Individual Income – Airport Time Components SP Experiments (Leisure Trips) Parameters Units Description Value Robust Std. Error Robust t- stat ΒTime_access Min Ground access time -0.0207 0.00257 -8.06 ΒCost_access_inc_level1 $ Ground access cost (including parking) if Income less than $75,000, else 0 -0.0853 0.00718 -11.88 ΒCost_access_inc_level2 $ Ground access cost (including parking) if Income between $75,000 and $199,999, else 0 -0.0747 0.00655 -11.39 ΒCost_access_inc_level3 $ Ground access cost (including parking) if Income $200,000 or more, else 0 -0.0561 0.00618 -9.08 ΒMode_transit (0,1) Ground access mode – transit (reference) --Fixed-- ΒMode_drop-off (0,1) Ground access mode – drive and dropped- off 1.12 0.146 7.67 ΒMode_drive (0,1) Ground access mode – drive and park 1.28 0.184 6.97 ΒMode_taxi (0,1) Ground access mode – taxi -0.338 0.151 -2.24 ΒTime_terminal Min Terminal access time -0.0314 0.00451 -6.95 ΒTime_security Min Check-in and security time -0.0345 0.00286 -12.04 ΒTime_security_gate Min Time to reach gate area -0.0274 0.00411 -6.68 ΒTime_gate Min Gate time -0.0212 0.00219 -9.69 Fit Statistics Number of parameters: 11 Number of observations: 6,385 Number of individuals: 804 Null log-likelihood: -4425.745 Final log-likelihood: -3017.497 Rho-square: 0.318 Adjusted rho-square: 0.316 3.5 Values of Time The willingness to pay for travel time savings, or value of time, is defined as the marginal rate of substitution between time and money–namely, the amount of money that a person would be willing to exchange for a reduction in travel time (or some specific component of travel time), while maintaining the same level of utility, or satisfaction. These rates of substitution are given by the ratios of the time and cost coefficients that are estimated in the MNL models, since these show the change in utility for any given change in the time or cost variables. Thus, the marginal rate of substitution between the various time component and cost variables given by the ratio of their coefficients provides the implied value that travelers would be willing to pay for a given time savings. Table 23 shows the willingness-to-pay (WTP) values for business and leisure travelers without considering income. As can be seen in the table, the values for business travelers Page 78

are consistently higher for each time component as compared to the corresponding values for leisure travelers. Additionally, there are considerable differences in WTP values for various time components. It should be noted that the WTP values for flight time and the expected value of flight delay were estimated from a different set of stated preference choice experiments than the on-ground time components. The flight times and airfares presented in the flight itinerary choice experiments were generally larger in magnitude than the airport access times and costs presented in the on-ground time components choice experiments. Because of potential scale effects (e.g., $5 on a $20 taxi fare could be perceived as much more onerous than $5 on a $600 airfare), this difference in the magnitude of the travel time and cost variables could have an impact on the WTP values. The WTP value for reductions in expected delay (which is calculated by multiplying the probability of being delayed and average amount of delay for delayed flights) appears to be perceived as 5.6 times the WTP value for flight time savings for business travelers and 3.5 times the WTP value for flight time savings for leisure travelers. Table 23. Willingness-to-pay Values (in $/hour) – Business and Leisure Travelers Component WTP - Business WTP - Leisure Airport Time Components Choice Experiments Ground access time $18.60 $16.95 Terminal access time $33.85 $26.01 Check-in and security time $37.19 $28.45 Time to reach the gate area $32.25 $22.83 Gate time $20.48 $17.62 Flight Itinerary Choice Experiments Flight time $51.01 $34.91 Expected value of flight delay $286.32 $123.30 Finally, the values of various time components segmented by income are shown in Table 24 and Table 25 for business and leisure travelers, respectively. The income is the respondents’ annual individual income from the previous year (2012) as reported by respondents in the survey. As expected, the willingness-to-pay values follow an upward trend as the income levels increase. Page 79

Table 24. Willingness-to-pay Values (in $/hour) by Income – Business Travelers Component Individual Income (2012 $ before taxes) Less than $75,000 $75,000 - $199,999 $200,000 or more Airport Time Components Choice Experiments Ground access time $13.92 $21.31 $38.49 Terminal access time $23.75 $36.34 $65.66 Check-in and security time $27.75 $42.45 $76.70 Time to reach the gate area $22.63 $34.62 $62.55 Gate time $14.23 $21.78 $39.34 Flight Itinerary Choice Experiments Flight time $33.66 $58.91 $100.99 Expected value of flight delay $186.34 $326.09 $559.01 Table 25. Willingness-to-pay Values (in $/hour) by Income – Leisure Travelers Component Individual Income (2012 $ before taxes) Less than $75,000 $75,000 - $199,999 $200,000 or more Airport Time Components Choice Experiments Ground access time $14.56 $16.63 $22.14 Terminal access time $22.09 $25.22 $33.58 Check-in and security time $24.27 $27.71 $36.90 Time to reach the gate area $19.27 $22.01 $29.30 Gate time $14.91 $17.03 $22.67 Flight Itinerary Choice Experiments Flight time $30.05 $41.18 $95.43 Expected value of flight delay $107.11 $146.77 $340.16 3.6 Additional Research The previous sections of this chapter identified the following aspects of the survey findings and model results that appeared to deserve further analysis during the remainder of the project: 1) Comparison of the demographics and other air party characteristics of the survey respondents with corresponding data obtained from recent air passenger surveys, in order to determine how representative the survey respondents were to the air passenger population in general. 2) Higher WTP values found for multi-person travel parties compared to one-person parties. Page 80

3) Higher WTP values found for savings of time spent in the airport terminal compared to time spent in ground access. 4) Significant differences in the WTP values found for flight time compared to ground access time and time spent in the airport terminal. 5) Further study of the differences in WTP values for the different on-ground time components. 6) Possible non-linearity in traveler sensitivity to time differences. The remainder of this section summarizes the findings of the additional analyses of the stated preference survey data undertaken by the Research Team to further explore the various issues suggested by the findings of the model estimation. One important aspect of further analysis of potential non-linearity in traveler disutility of different travel time components is traveler sensitivity to flight delay. It would seem reasonable that travelers would have a relatively low WTP for delay savings where the delays are fairly small, since they may already have made some allowance for some amount of delay in their travel plans and the consequences of a small amount of delay may not be particularly serious. However, as the length of the delay increases, the potential consequences become more serious, including the possibility of missed flight connections or ground transportation services at the destination, arriving at the destination at a very inconvenient hour, or being late for, or missing entirely, the event that is the primary (or only) purpose of the trip. On the other hand, beyond a certain point the length of the delay may cease to matter, since there are no more connecting flights that day or the event that motivated the trip is over. Other potential additional analyses, not explicitly identified in the above sections of Chapter 3 but suggested by the findings of the model estimation, include the following: 1) Compare the distribution of the airport-pair markets of the flights taken by the survey respondents to the distribution of itineraries in the DB1B data for the second quarter of 2013 (the comparison of the survey markets discussed in Section 3.3 only considered the distribution of trip origins and destinations, not the airport-pair markets, and used older data). 2) Compare the values of the ground access WTP and modal constants found in the model estimation with corresponding values found in past studies that developed airport ground access models using revealed preference (RP) data. 3) Explore differences in the effect on WTP of basing this on individual income, household income, or household income adjusted for household size and composition. 4) Explore the use of continuous variables for income, rather than simply dividing income into three categories. 5) Determine whether WTP values for travelers on business trips are different if they pay for the trip themselves versus being reimbursed by their employer or client. Page 81

The following sections of this chapter summarize the findings of the additional analysis undertaken to address the foregoing issues: • Comparison of survey respondent reported flights to national data • Comparison of survey respondent air party characteristics to airport survey data • Further analysis of survey results and model estimation • Comparison of ground access coefficients estimated from survey data with ground access mode choice model coefficients in prior studies A more detailed discussion of the analysis undertaken to address each of these topics is included as Appendix B.3 to this report. Comparison of Reported Flights by Survey Respondents to National Data Since the objective of the SP survey was to obtain a representative sample of air travelers who had made a recent domestic air trip, the extent to which the reported air trips in the survey correspond to the distribution of domestic air travel across different markets has important implications for how representative the survey results are for domestic air travel in general. Therefore a more detailed analysis was undertaken to compare the distribution of the reported trip origins, trip destinations, and airport-pair markets flown between the reported trips by the SP survey respondents and the corresponding distributions for all domestic air travel obtained from the U.S. DOT 10% Origin & Destination Survey (also termed Database 1B, or DB1B, data). The results of this comparison can be summarized as follows: 1) The distribution of trips across origin and destination airports between the SP survey responses and the DB1B data corresponds fairly closely. Origin airports with 10 or more air trips in the SP data accounted for 66% of the reported trips in the survey and 68% of the passengers in the DB1B data, while destinations with 10 or more trips in the SP data accounted for 72% of the reported trips in the survey and 69% of the passengers in the DB1B data. The difference between the SP data and the DB1B data becomes even less when considering origin and destination airports with five or more air trips in the SP data. 2) At the airport-pair market level, the distribution of the recent trips reported by SP survey respondents by market size (based on the DB1B data) is generally close to the distribution of reported passengers in the DB1B data, with the larger markets (those with more than 500 DB1B reported passengers) somewhat under-represented (about 8% fewer trips than would be expected if the distribution corresponded to the DB1B data) and smaller markets correspondingly overrepresented. Necessarily some airports or markets in the DB1B data would not appear in the SP survey data due to the survey sample size, but the under- or over-representation of markets in the survey is generally equivalent to one survey response or less. For the 146 largest directional markets, 32% did not appear in the survey trips, but the majority of these would have only been expected to have one reported trip, based on the survey sample size, and the remainder would only have been expected to have two Page 82

reported trips. Only 10% of the markets that were included in the survey trips were over-sampled by more than one reported trip and the largest over-representation was only four survey responses. Therefore, on balance, the distribution of reported trips by market appears to be generally representative of the distribution of air passenger trips across the full range of markets of varying size and geographic location in the U.S. Comparison of Air Party Characteristics by Survey Respondents to Airport Survey Data In order to explore how representative the recent air trips reported by the SP survey respondents were to air travel in general, aside from the distribution of the airport-pair markets in which those trips took place, the air party characteristics for the reported trips were compared to data for domestic air passengers collected in an air passenger survey performed at Los Angeles International Airport (LAX) in 2011. The LAX survey was used because it has a large sample size of 8,984 respondents making directional domestic trips that began at LAX (i.e., excluding passengers making flight connections at LAX). This survey contains one of the largest sample sizes of a recently completed air passenger survey and included respondents making air trips to destinations throughout the U.S. The comparison addressed the following air party characteristics: • Air party size • Trip purpose • Household income • Gender and age However, some caution should be exercised in comparing the profiles of respondents to the LAX and SP surveys because it is not known how closely air travelers using LAX compare to a national profile of air travelers using the full national network of commercial airports that vary widely by size. Moreover, the median household income of residents of the Los Angeles metropolitan area is somewhat higher than the national median household income,22 as reflected in differences between LAX survey respondents who were residents of Southern California and those who were visiting the region. In comparing the findings of the SP survey with those from the LAX survey, it should be borne in mind that the SP survey was a survey of individual air travelers, while the LAX survey (like all intercept surveys) was a survey of air trips. The distribution of respondent 22 According to the U.S. Census Bureau, the 2012 Los Angeles metropolitan area median household income is about 12.5% above the national level ($57,745 in the LA region compared to $51,324 nationally). Page 83

characteristics in an intercept survey will therefore reflect a higher presence in the survey sample of travelers who make more frequent air trips, since they have a higher likelihood of being surveyed. This will affect the observed distribution of those characteristics (such as household income) that influence the propensity for air travel (measured by the number of air trips made per year) or differ across travelers making different numbers of air trips per year. The findings of this comparison are summarized as follows: 1) The most recent SP survey trips were found to have a significantly smaller proportion of single-person parties and a correspondingly higher proportion of two- person parties than observed in the LAX survey. The proportion of air parties of three or more was approximately the same between the most recent SP survey trips and the LAX survey. When adjusted to reflect differences in the number of air trips made per year for different trip purposes, the proportions of single-person and two- person parties in the SP survey results were closer to those in the LAX survey, but still understated the proportion of single-person parties and overstated the number of two-person parties. 2) The most recent SP survey trips had a smaller proportion of trips reported for a business purpose (excluding attending a conference or convention), although a significantly higher proportion of trips to attend a conference.23 The proportion of SP survey trips for a vacation was only slightly higher than in the LAX survey. However, the proportion of trips to visit friends or relatives was significantly higher in the SP survey. The proportion of SP survey trips to attend school or college was significantly lower than in the LAX survey, although this could be a reflection of the age profile of the SP survey respondents discussed below. It might also be partly influenced by the timing of the two surveys, since the first wave of the LAX survey took place in late August when many students would be returning to school, and by the Los Angeles metropolitan area having the second highest concentration of college students in the U.S. (Florida, 2012). When considering the number of air trips in the previous year reported by the SP survey respondents, the proportion of business trips (including attending a conference or convention) was somewhat higher than given by the LAX survey for Southern California visitors and significantly higher than given by the LAX survey for Southern California residents. The proportion of air trips in the previous year reported by SP survey respondents for leisure was correspondingly lower than the proportion of non-business air trips given by the LAX survey. The SP survey asked respondents for the number of business and non-business trips taken in the previous 23 In the model estimation using the data from the SP survey choice experiments the Research Team combined business trips with attending conferences and conventions to form a broader “business” trip purpose category. Page 84

year without any further breakdown by purpose, so it is unclear how the lower proportion of non-business trips reported by the SP survey respondents relates to the various trip purpose categories in the LAX survey. The lower proportion of business trips in the LAX survey may also reflect the timing of the survey, which was conducted in two waves with the first wave in late August, as well as the importance of Southern California as a vacation destination. 3) It was found that overall a higher proportion of the SP survey respondents were reasonably well-off (household incomes between $50,000 and $150,000) compared to the LAX survey respondents, but the SP survey respondents whose recent trip was for business purposes included a smaller proportion of higher-income households ($250,000 or more). When adjusted for the number of air trips in the previous year reported by the SP survey respondents, the proportion of business trips made by SP survey respondents with a household income of $250,000 or more was somewhat higher than found in the LAX survey. 4) The SP survey respondents were found to have a higher proportion of female respondents than the LAX survey respondents, although the difference is not significant in the case of those whose recent trip was for business purposes.24 When adjusted for the number of reported air trips in the previous year, the proportion of both business and leisure trips reported by male respondents was slightly higher than found in the LAX survey. 5) The SP survey respondents were found to be substantially older than the air passenger population in the LAX survey, particularly for male respondents. Some 44% of male respondents in the SP survey were over 64, compared to only about 8% for Southern California visitors in the LAX survey, while 24% of female respondents to the SP survey were over 64, compared to about 9% for Southern California visitors. In summary, the air party size on the reported recent air trips by SP survey respondents had fewer single-person parties than found in the LAX survey, which is consistent with the smaller proportion of business trips, although the difference was less when adjusted for the number of reported air trips in the previous year. The SP survey respondents appear to have generally higher household incomes than air travelers using LAX. Perhaps the most significant difference between the SP survey respondents and air travelers using LAX is the age distribution, with both male and female respondents to the SP survey being significantly older than the LAX survey respondents, particularly male respondents. 24 However, it should be noted that in the case of an intercept survey, such as the LAX air passenger survey, there may be a tendency for a male member of an air party with both genders to respond on behalf of the party, whereas in the SP survey the respondent, male or female, will report the recent trip irrespective of the composition of that air party. Page 85

Further Analysis of Survey Results and Model Estimation The Research Team undertook additional analysis of the survey results and model estimation. Some of the issues were resolved while others will require further research beyond the current project to fully address. The findings of this analysis are summarized as follows: 1) Air passengers’ estimated value of willingness-to-pay (WTP) for savings in airport ground access time, and hence for the various on-airport time components that depend on the WTP value for ground access time, was found to vary with the respondents’ age, although it is unclear how much of this variation is due to differences in respondents’ income levels with age. The estimated WTP values for savings in flight time did not show a similar variation with age, with the values being generally similar across the age groups, with one unexplained exception for business travelers in the age group 55-64. The Research Team was unable to determine why the WTP values for ground access time savings appeared to vary by age group while those for flight time savings did not. The consistency of the WTP values for flight time savings across the age groups means that the age imbalance in the survey sample did not significantly affect the estimated WTP values for flight time. Further research is needed to understand whether the variation in WTP values for ground access time by age were simply a reflection of differences in respondents’ income in each age group or there is an effect of age on WTP after controlling for income, and if so, how the age imbalance in the survey sample may have affected the estimated WTP values for ground access time. 2) Multi-person travel parties were found to have lower per-person WTP values. However, the drop in per-person WTP values follows more or less a linear trend as party size increases. This implies that respondents were actually making trade-offs in the choice experiments as if they were bearing the access cost themselves, not sharing the costs among the other party members, as was assumed in the model estimation described in Section 3.4. While it was intended that the airport ground access and airport time component SP choice experiments were for the travel party in the most recent trip described by the survey respondents, this may not have been fully understood by the survey respondents, who may have assumed that the choice experiment trade-offs applied to a single-person air trip. The implications of this for the estimated WTP values of ground access travel time should be explored in future research. 3) The Research Team undertook further analysis to try to explain the lower WTP values for access time savings compared to the WTP values found for savings of time spent in the airport and flight time, but was unable to come up with a clear explanation. It was felt that these differences might potentially be due to the scale differences in the travel times shown in the choice experiments and the effect of perceived access mode preferences. 4) In order to explore possible non-linearity in traveler sensitivity to time differences, choice models were estimated with non-linear functions of time and cost. However, the Research Team was unable to find any significant and conclusive results from Page 86

this analysis. An attempt to estimate WTP values for flight delay using a non-linear function of expected delay also gave inconclusive results, suggesting that further research with a larger survey sample size may be needed in order to capture any such effects. 5) After testing several continuous income transformations for the cost/fare variable, the research team used a regression-based approach to estimate the relationship between WTP values and income. It was found that using a respondent’s individual income generally provided a better model fit of the relationship between WTP and income than using household income with no adjustment for household size and composition. This is not particularly surprising, since household income does not take account of household size and households with two or more adults will have a higher household income than a single-person household with the same per-person income. Using per-person household income, where each child was counted as 0.5, generally gave similar model fit as individual income. Individual income generally gave a slightly better model fit for business trips, while per-person household income gave a slightly better model fit for non-business trips. This seems reasonable since travelers on business trips are likely to base their value of time on their salary or wage rate, while those making personal trips are likely to base their value of time on their household income and composition. 6) The values of time shown in the Guidebook are expressed in terms of individual income. However, air passenger surveys that ask respondents to indicate their income typically request household income, as it may be perceived to be a less intrusive question than personal income and takes account of the incomes of other household members. The Research Team has included a table in the Guidebook that allows the user to estimate the corresponding individual income for any given household income, based on the responses to the SP survey, so that an airport may use either in determining the values of time for use in analysis. 7) The estimated relationships between WTP values and income obtained from the regression models gave progressively increasing, non-linear relationships between the WTP values for flight time and ground access time and income, with the resulting WTP values expressed as a percentage of the respondent’s hourly income (assuming 2,000 hours worked per year), or the equivalent household income on a per person basis, declining with increasing income levels. Thus although lower income respondents had significantly lower estimated WTP values than higher income respondents, as would be expected, their WTP values as a percentage of their hourly income were found to be much higher. This may partly reflect the fact that higher income households generally pay a higher proportion of their income in taxes, so when expressed as percentage of their after-tax hourly income, the decline in WTP values would not be so great. However, it is unclear how much of the apparent relationship this would account for and the finding is deserving of future research to better understand this relationship. Page 87

8) Pending the results of future research to clarify and extend these findings, the estimated relationships provide a basis for incorporating a continuous function of income into air traveler choice models in order to account for differences in income when exploring the effect of other variables, as well as adjusting WTP values in airport BCA studies to account for changes in income distribution. 9) Not surprisingly, it was found that SP survey respondents making both business and leisure trips who paid for their trip themselves had a significantly lower value of WTP for savings of flight time and expected delay than those for whom someone else paid for the trip. This suggests that future research to confirm and extend the findings of the current study should take into account whether travelers are paying for their trip themselves in estimating WTP for travel time savings so that the estimates of WTP values can take account of any changes in the proportions of travelers paying for their trip themselves, since these are likely to vary by trip purpose and destination. Survey Ground Access Coefficient Comparison with Prior Airport Ground Access Models The estimates of WTP for ground access time savings obtained from the analysis of the SP survey results is critically important to the estimated WTP for time savings in each of the other time components between arriving and the airport and boarding the flight, since those values are estimated as a ratio of the WTP for ground access time savings. Because there is typically no cost charged for the other airport time components, the stated preference experiments only gave times for those components. Therefore, the disutility of time spent in each component was estimated relative to the disutility of time spent in ground access, where a cost is typically incurred, and the implied WTP for time savings in each component depends on the WTP for ground access time savings. Since the cost and time differences for the ground access component in the SP experiments resulted in many cases from differences in the ground access mode assumed for each option in the experiments, in order to estimate the WTP for ground access time savings, it was necessary to control for differences in the perceived utility of the different ground access modes arising from attributes other than travel time and cost. This was done by estimating a constant term for each mode, expressed in equivalent minutes of travel time. Hence the estimated values of these terms influenced the estimated WTP for ground access time savings. In order to determine how consistent the estimated WTP for ground access time savings and the modal constants obtained from the SP survey experiments are with the corresponding values given by airport ground access model choice models developed in prior studies, an analysis was undertaken to compare the values estimated from the SP survey experiments with the corresponding values given by a sample of five prior airport ground access mode choice models. Page 88

Since these models were estimated on data collected at different points in time, adjustments were made so that the resulting estimates of WTP were expressed in consistent dollars to the values estimated from the SP survey data, allowing for both inflation and changes in real household income over time using the national consumer price index published by the Bureau of Labor Statistics. In comparing the WTP values given by the SP survey experiments with those given by the prior studies, it should be borne in mind that the five metropolitan regions covered by the prior studies had an average income in 2012 (weighted by population) that was 12% higher than the U.S. overall, and account for only 10% of the national population (U.S. Bureau of Economic Analysis, Table CA30). The higher average income of residents of the five metropolitan regions compared to the country as a whole may partly account for differences between WTP values estimated on data from those regions and from the SP survey which included respondents from a broader range of communities that may better reflect the nation as a whole. The results of this comparison are summarized as follows: 1) The range of WTP values implied by the models estimated in the five prior studies is very large. The values implied by the coefficients, estimated using the SP data in the current study without taking respondent income into account, correspond to the lowest values in the range established by the five studies. 2) The WTP values for low- and high-income survey respondents from the three prior studies that considered respondent income span a narrower range, since they exclude the two studies that had the lowest and highest values (and did not consider the income of the respondents), although this is still quite wide. The definition of low-income households in the prior studies covers the lowest of the three income ranges used in the SP model, while the definition of high-income households in prior studies covers the two higher-income ranges in the SP model. The WTP values from the SP models for the three individual income ranges lie below the lower end of the range of WTP values given by the three prior studies for the corresponding household income range. 3) Although the difference varies by mode, it appears that the modal constants given by the SP model show a greater disutility relative to drop-off trips for all three modes compared to the values found in prior studies, although the difference is much greater for taxi than for the other two modes (drive and park, and transit). It is unclear to what extent these differences may have arisen from the ground access travel time and cost values used in the choice experiments, and if any distortions in the choice process that this may have generated could have affected the implied WTP for ground access time savings. It is also not clear to what extent these differences are due to the different methodologies of the prior studies, or how much the higher-income levels in the five regions covered by the prior studies compared to the national income distribution may have influenced the implied WTP values given by the prior studies relative to those Page 89

obtained from the SP survey experiments. Lastly, it must be recognized that air passenger surveys and mode choice modeling primarily concerned with airport ground access can address a wider range of factors, including ground access modes not considered in the SP survey experiments, than a survey that spans the range of travel time components associated with an air travel trip. This last point is a primary reason why the Research Team recommends that considerations for future research building on the current project include a larger survey effort, with segments of the surveys concentrated on different aspects of the air trip, including ground access (see Chapter 4 of this report). Survey Ground Access WTP Values Comparison with Current U.S. DOT Guidance In addition to other surveys, the WTP values for airport ground access obtained from the SP survey experiments can be compared to the recent guidance on the value of travel time savings for use in economic analysis issued by the U.S. DOT, which includes both air and surface modes (U.S.DOT, 2014). The U.S. DOT recommended values for surface modes are broadly consistent with the WTP values for airport ground access estimated from the SP survey choice experiments conducted for this research project. Table 26 compares the SP ground access WTP values to the value of travel time savings (VTTS) from the U.S. DOT guidance for local and intercity surface trips. In this comparison, we assume that while “local travel” reflects the typical distances and modes of ground access trips made by air travelers to airports, the “intercity” values may be closer to the perceived VTTS for these trips, which, after all, form part of an intercity journey. Table 26. Comparison of Ground Access WTP Values from SP Survey Experiments to U.S. DOT Surface Transportation Value of Time Guidance Trip Purpose SP Survey (2013 $/hour) U.S. DOT Local (2012 $/hour) U.S. DOT Intercity (2012 $/hour) Business $18.60 $24.10 $24.10 Leisure $16.95 $12.30 $17.20 Note: The SP survey uses the term leisure travel, while the U.S. DOT guidance uses the term personal travel. It is assumed that these are equivalent. The values in the U.S. DOT guidance for business travel are higher than those estimated in the current study, while those for personal local travel are lower than the WTP values estimated from the SP survey for leisure trips and the values for intercity personal travel are similar. Adjusting the U.S. DOT values to 2013 dollars using the approach specified in the U.S. DOT guidance will increase the 2012 values slightly. Comparing 2013 values for personal or leisure travel, the value for local surface travel would remain below the WTP estimates for airport ground access trips from the SP survey experiments, while the value of intercity surface travel would be about 3% above the WTP values from the SP survey. The following points should be noted in the above comparison: • The U.S. DOT values for local travel by surface modes covers all local travel, including commuting, shopping, recreation, etc., by all travelers. To the extent that air Page 90

travelers have a higher average income than travelers in general, it could be expected that the VTTS for airport ground access trips would be higher than for local travel in general. • The values for business travel in the U.S. DOT guidance are based on a percentage of the hourly median gross compensation, including fringe benefits, rather than any behavioral response of travelers making a business trip (as in the current study). While it may be argued that this is the value to the employer of any travel time savings, as the U.S. DOT guidance suggests, this is a different thing from the perceived VTTS to the traveler, as measured in the current study. Therefore, while it is useful to compare the values in the U.S. DOT guidance to those estimated in the current study to see how consistent they are, it is important to bear in mind the differences in methodology used in the deriving the two different sets of values. In particular, the U.S. DOT guidance does not explicitly provide different values for travelers with different income levels, as was done in the current study, although the difference between the VTTS values for intercity travel by surface modes and those by air travel in the U.S. DOT guidance is largely justified on the grounds that air travelers have a higher median household income than households in general. 3.7 Summary and Conclusions The stated preference survey gathered information from 1,171 air travelers who recently made an air trip within the U.S. The questionnaire collected data on current air travel behaviors and engaged the travelers in a series of conjoint choice experiments. Multinomial logit (MNL) choice models were developed using the survey data to produce estimates of value of time (VOT) of air travelers for two market segments: business and leisure travelers. The magnitude and signs of the sensitivity estimates appear reasonable and intuitively correct. The model coefficients were used to estimate values for the willingness to pay for travel time savings, or values of time, for different components of an air trip, including: • Ground access time • Terminal access time • Check-in and security time • Time to reach the gate area • Gate time The values of willingness to pay (WTP) for ground access time savings estimated from the results of the survey can be compared to the WTP values found in the literature on prior studies that estimated the value of airport ground access time using stated preference (SP) Page 91

or revealed preference (RP) methods. The values of time for airport ground access estimated here, at about $19 per hour for business travelers and $17 per hour for leisure travelers, are at the lower end of the range of studies that use SP methods, and less than most studies that use RP methods. However, they are comparable to WTP values found in several prior airport access studies conducted by Resource Systems Group, Inc. (RSG), including a recent airport ground access mode choice study conducted in the greater New York region, which found ground access WTP values of approximately $17 per hour by those making automobile trips (RSG, 2009, RSG 2007 and Wilbur Smith Associates 2008). The WTP values for the time to reach the airport terminal from the location where parked, dropped off, or alighting from public transportation are somewhat higher than those for the access trip itself, at $34 per hour for business travelers and $26 per hour for leisure travelers. This result could be partly due to non-linearity in traveler sensitivities to time savings. Airport ground access journeys are, on average, much longer than the terminal access component of the trip, and the per-minute sensitivity to time savings may, therefore, be lower. The WTP values for the time to reach and pass security are higher still at $37 per hour and $28 per hour for business and leisure travelers, respectively. This makes sense intuitively, as the portion of the trip between arriving at the airport terminal and clearing security may be more stressful due to uncertainty in the wait times to check-in and clear security, as well as unfamiliarity with the physical layout of the terminals. This time is also not likely to be “productive” time, with travelers generally having little or no ability to work, relax, or even make phone calls while walking through the terminal, checking in, or navigating the security line. The WTP values for the time to reach the gate from security ($32 per hour and $23 per hour for business and leisure travelers, respectively) and time spent in the gate area ($20 per hour and $18 per hour for business and leisure travelers, respectively) are somewhat lower than those for the other two airport time components discussed above. This can be attributed to the fact that travelers can spend this time patronizing airport services (such as shopping, eating, etc.), and can be more productive by using mobile communications technology. This is also the time component that travelers have some control over by changing their departure time to the airport from their starting locations. It should be noted that these WTP values are still higher than the ground access WTP (although marginally so, in the case of gate time). This is somewhat unexpected because the ground access trip generally does not provide the same opportunity for the activities that can be accomplished at the gate and can be associated with a high level of anxiety over unexpected delays in reaching the airport. However, as noted above for the time to reach the airport terminal from the location where parked, dropped off, or alighting from transit, these differences could possibly be explained by non-linearity in traveler sensitivity to time differences, with a lower per-minute sensitivity for the longer times typically experienced in ground access travel. This aspect could benefit from further investigation in future research. Page 92

The value for flight time (which includes in-aircraft time and time spent making any flight connections) is estimated to be the highest among all the time components ($51 per hour for business travelers and $35 per hour for leisure travelers). While these values are broadly consistent with the most recent guidance from the U.S. Department of Transportation on the values of travel time to be used in economic analysis (U.S. DOT, 2014), the higher values for flight time relative to the time spent in the airport terminal are somewhat unexpected. However, it should be noted that the values for flight time are based on trade-offs between flight time and airfare, while the values for the airport access and terminal time components are based on trade-offs between these times and airport access cost. The extent to which these differences in the WTP values are due to the different cost basis could benefit from further investigation in future research. Overall, while the magnitudes of the WTP values seem broadly consistent with values found in other studies, some of the relationships between the different values were unexpected. These include the differences noted above between the values for flight time and those for airport ground access time, as well as between the values for ground access time and gate time. Because there are no existing studies that have estimated values of time for all of the on-ground time components identified here, there are no opportunities for a direct comparison of the relationships between the values of airport ground access time and those of other in-airport time components with those found in prior studies. Therefore, these relationships appeared to deserve more study in the additional analysis undertaken during the remainder of the project described below and in Section 3.6, including an examination of non-linear effects in travelers’ time and cost sensitivities. Additional Research Undertaken The comparison of the distribution of airport-pair markets flown by the stated preference (SP) survey respondents in their reported recent trips with the distribution of domestic air passenger travel given by the U.S. DOT 10% Origin & Destination Survey (DB1B data) indicates that the reported trips in the SP survey data provide a representative sample of domestic air passenger travel. However, the composition of those trips in terms of trip purpose may understate the amount of business travel and overstate the proportion of personal trips, although the proportion of business trips in the total reported air trips made by SP survey respondents over the previous year is much closer to the proportion observed for domestic air travelers in an air passenger survey performed in 2011 at Los Angeles International Airport (LAX). To the extent that separate estimates of willingness to pay (WTP) for travel time savings have been made for business and leisure travel, any under- representation of business travel in the recent trips by SP survey respondents does not appear likely to have introduced a significant bias in the resulting WTP values. Compared to air travelers in domestic markets from LAX, the SP survey respondents are older, particularly the male respondents, and tend to have higher household incomes. To the extent that separate WTP values have been estimated for survey respondents in respondents’ different income ranges, any over-sampling of air travelers with household incomes above $50,000 per year in 2012 (assuming that survey respondents reported their Page 93

income in the prior year to the survey) may not have significantly affected the resulting WTP values. However, although the extent of any bias is somewhat unclear due to differences in the way that income was measured and respondents grouped by income. In contrast to most air passenger surveys, which ask respondents to indicate their household income, the WTP values estimated from the SP survey data, and included in the Guidebook, were based on individual income, although the SP survey respondents also reported their household income. Further analysis of the SP survey results and additional model estimation found that the effect of the age of the survey respondent did not have a significant effect on the estimated WTP values, after controlling for other factors. However, multi-person travel parties were found to have a lower per-person WTP for travel time savings compared to those traveling alone, as if the full cost of the air party trip was being paid by the survey respondent. It is unclear whether this truly reflects how air travelers regard the cost of travel or is due to of the way the SP choice experiments were presented to the survey respondents, and, therefore, represents an aspect that is deserving of future research to better understand (see Chapter 4). The additional analysis was not able to quantify any non-linearity in the SP survey respondents’ sensitivities to differences in flight time or airport ground access time, or WTP for reductions in flight delay. However, the additional analysis was able to develop a continuous relationship between WTP and individual or per-person household income to supplement the WTP values estimated for the three individual income groups in the analysis results included in the Guidebook. These aspects appear worthy of future research to confirm or modify the WTP values recommended in the Guidebook, as well as facilitating the process of updating WTP values to reflect changes in real household (or individual) incomes. Finally, a comparison of the estimated model coefficients for the ground access trip component from the SP survey data with the corresponding coefficients from airport ground access mode choice models developed in prior studies for five metropolitan regions suggests that the SP survey model may have yielded a low estimate of the value of ground access time. However, the differences may partly reflect the higher average incomes in the five regions compared to the national average, which make it probable that the value of ground access time of airport travelers would be higher among the five regions than the nation as a whole. While the values of time given by the models developed from the SP survey are comparable to recent U.S. DOT guidance and previous studies conducted by RSG, they are at the low end of the range given by the airport ground access mode choice models developed in prior studies for the five regions, suggesting that this is a subject that should be revisited in future research. Page 94