Cover Image

Not for Sale



View/Hide Left Panel
Click for next page ( 91


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 90
90 Guidebook for Conducting Airport User Surveys a flight sampling plan that provides reasonable coverage of different flight destinations and airlines--while utilizing survey teams efficiently and allowing staff to take required breaks at appropriate intervals--is a major challenge. 5.3 Sample Size, Survey Coverage, and Timing Once the details of the survey methodology have been determined, the next steps are to decide on the required sample size, develop a sampling plan to provide adequate coverage of the varia- tion in air passenger characteristics over time, and decide when to perform the survey. 5.3.1 Sample Size The details of calculating sample sizes for different sampling methods are discussed in Section 3.4. In air passenger surveys, the most common type of sampling method uses cluster sampling to select a random sample of flights as clusters and then attempts to survey either all the passengers on each selected flight (with a self-completed survey) or a sequential sample of passengers on each flight (with an interview survey). The accuracy of a cluster sample depends on both the variance of the characteristic of interest within each cluster and the variance between clusters. Where each cluster is a flight, whether the mean of a given characteristic varies significantly between flights will depend on the characteristic. However, many characteristics--such as airfares, ground access mode use, air trip duration, and air travel party size--are likely to vary significantly by destina- tion and hence their sample mean will vary between different flights. Of course, these character- istics vary widely within a given flight as well. Therefore, it will generally be necessary to use a significantly larger sample size with flight-based cluster sampling than with random or sequen- tial sampling of the air passenger population in order to achieve a similar level of accuracy. As a practical matter, the only way to perform random or sequential sampling of the air pas- senger population without resorting to a flight-based cluster sample is to interview passengers at a location where all passengers can be intercepted, such as the exit from security screening. The extent to which a flight-based cluster sample should increase the sample size to provide an equivalent level of accuracy to that calculated for a random sample is dependent on the spe- cific traffic composition at each airport, the characteristics of interest, and how these vary between passengers on particular flights and on average between flights. This issue is not at all well understood and is deserving of further research. In the absence of more specific analysis, it would be prudent to increase the sample size by a factor of two for well-designed interview sur- veys of air parties, with appropriate stratification of flights16 and approximately 20 air passengers interviewed on average per flight. This factor increases as the number of passengers interviewed per flight increases. In the above case, a factor of 1.5 would be more appropriate where an aver- age of 10 passengers per flight is interviewed, but a factor of 6 would be more appropriate where an average of 90 passengers per flight is interviewed. Sample size with cluster sampling is dis- cussed in more detail in Appendix B. 5.3.2 Estimating the Population of Airport Users on a Given Day To develop a sampling plan for a given day, it is desirable to have an estimate of the number of air passengers (and possibly the associated greeters and well-wishers) using the airport on that day. If the sampling plan is being developed a fairly short time before the survey, it may be pos- 16 Assuming an intra-class correlation coefficient of 0.05.

OCR for page 90
Air Passenger Surveys 91 sible to obtain expected passenger loads on each flight from the airlines. If the sampling plan is being prepared well before the start of data collection or the airlines are not willing to provide this information, the population of these categories of airport users per day can be estimated as follows: Passengers--Obtain a schedule of departing flights (and arriving flights if interviewing arriv- ing passengers) for the survey period from the airport operator or sources such as the Official Airline Guide (OAG). The schedule should indicate the flight departure (or arrival) time, air- line, flight destination (or origin), terminal, aircraft type and seats, and whether it is a through flight or originating (or terminating) at the airport. This information can be determined from the full flight itinerary or routing. For each flight, estimate the load factor using past data for the airline and city pair.17 This estimate could be further refined, if necessary, to allow for variation in load factor by month, day of week, and time of day where data for this level of detail is available.18 For through flights, estimate the proportion of passengers who will continue on the flight and not leave the aircraft.19 The proportion will vary by routing, airline, and airport and may need to be estimated based on limited qualitative inputs (e.g., comments by gate staff or other knowledgeable people). Multiply the seats, load factor, and proportion of passengers not continuing on the flight to estimate the enplaning (or deplan- ing) passengers. The daily or hourly passengers are then found by summing the estimated passengers over the flights in that time period. Connecting passengers do not make use of the groundside facilities or transportation system and can be excluded, where necessary, by multiplying by the ratio of origin/destination (O/D) passengers to enplaned/deplaned pas- sengers for that airport. This approach can be extended to estimate the proportion of the passenger population in different market segments (e.g., domestic versus international pas- sengers, or passengers using a specific terminal), based on the flight destination (or origin in the case of arriving flights). Greeters and well-wishers--Numbers vary greatly with airport size, types of service available, proportions of business and visiting (non-local) passengers, and public transportation ser- vices to the airport, among other factors. The numbers of well-wishers and greeters that come into the terminal can be estimated based on the numbers of O/D passengers. While these vary significantly by airport, month, day of week, and time of day, a rough guide to the average number of well-wishers per originating passenger based on recent airport surveys is shown in Table 5-3. The number of well-wishers can be estimated by multiplying the number of originating pas- sengers by an appropriate factor based on the data in Table 5-3. Note that many of these well- wishers will be in groups and may be seeing off several passengers. Numbers of greeters are typically similar to the numbers of well-wishers. 5.3.3 Sampling Strategy The design of the sampling strategy is key to obtaining reliable results in an air passenger sur- vey. A poorly designed sampling strategy can exclude certain subgroups from the sample entirely and lead to biased results. As discussed in Section 3.3, a controlled sample attempts to design the sampling strategy so that the composition of the sample reflects the underlying distribution of the population characteristics fairly accurately. In the case of air passenger surveys, this means 17 Monthly data on passenger loads by airline and flight segment are available from the U.S. Department of Transporta- tion Bureau of Transportation Statistics T-100 database. 18 Some airports require airlines to report passenger loads by flight. 19 The difference between arriving/departing and enplaned/deplaned passengers.

OCR for page 90
92 Guidebook for Conducting Airport User Surveys Table 5-3. Representative ratios of well-wishers to originating passengers. Well-wishers per Airport Size Terminal Type Airport Year Originating Passenger Large hub International New York JFK Terminal 4 2003 0.29 International San Francisco, California 2006 0.17 Domestic San Francisco, California 2006 0.14 Medium hub Domestic San Jose, California 2003 0.16 Domestic Oakland, California 2006 0.18 Domestic Winnipeg, Manitoba 2007 0.25 Small hub Domestic Birmingham, Alabama 2005 0.09 Domestic Quebec City, Quebec 2006 0.19 Domestic Victoria, British Columbia 2006 0.35 Source: Airport surveys conducted by Jacobs Consultancy that the sample has the same (or very similar) proportions as the population for such character- istics as the following: Airline. Flight departure time. Day of the week. Destination. Originating versus connecting traffic. Domestic versus international trips. In practice this is very difficult to achieve. Each of these characteristics varies independently of the others, resulting in a huge number of potential combinations of characteristics, and the logistics of performing the survey limit the ability to vary the sampling rate to match the underlying distribution of characteristics. In particular, intercept interview surveys tend to under-sample during busy periods and over-sample during less busy periods, because it is difficult to schedule the number of interviewers to match the changes in passenger flow. When the sample is not controlled to ensure that the composition of the sample corresponds to that of the population, the results of the survey can be weighted so that the reported results reflect the composition of the population rather than that of the sample. The process for deter- mining the weights to be used is discussed in Section 3.5. Sampling Passengers The first consideration in sampling air passengers is to decide whether to sample individual pas- sengers or air travel parties. As discussed earlier in this chapter, this decision is largely a function of the survey method adopted. Self-completed survey forms are typically distributed to every adult passenger in the group of interest, while intercept interview surveys are usually designed to collect information on air parties. There is obviously no difference in the case of one-person parties. The objective in defining a sampling strategy to survey air passengers is to obtain a random sample of respondents. Where the flow of passengers or the sequence of passengers in a queue is already random with respect to the characteristics of interest, selecting every nth passenger will generate a random sample of respondents. When a respondent is in an air travel party of more than one person, the count to determine the next nth passenger in a sequential sampling strat- egy should start again if the next nth passenger is a member of the air party that has just been interviewed. This method will avoid interviewing multiple passengers from the same air party. Members of the same air party will typically be standing or seated together, and it will usually be

OCR for page 90
Air Passenger Surveys 93 obvious who is in the party. Where it is not obvious, it may be necessary for the interviewer to ask the respondent who has just been interviewed to indicate the members of the air party. In the case of large air parties, such as tour groups, members of the party may not remain together in the terminal, and thus the sampling strategy may happen to interview several members of the same party. Large air parties call for particular consideration in analyzing survey results, as dis- cussed in Section 5.2.1, to account for the possibility that the survey may interview several mem- bers of the same party or that different members of the party may have traveled to the airport separately. With large air parties (more than about six people), it may be helpful to record the name of the group organizing the trip in order to identify other respondents from the same group when analyzing the results. Where passengers are moving, such as at an exit from security screening or an entrance to the terminal building, the duration of each interview will generally be sufficiently long that by the time the interview is completed, the next passenger in the stream will be effectively chosen at random. Also the variation in the duration of each interview will help ensure the randomness of the sample. Where passengers are grouped by some characteristic, such as passengers in a check-in queue at an airline counter, the sampling strategy should attempt to distribute the interviews across the different groups in proportion to the passengers in each group, such as the number checking in on each airline. However, this is difficult to achieve in practice, particularly because those pas- sengers in short queues may not be in the queue long enough to complete an interview. For these reasons, queues are generally not a good location to perform intercept interview surveys. In locations where passengers are not in an ordered sequence, such as in airline gate lounges or baggage claim areas, it will be necessary to define a starting point for the sequence of passengers in the sample and a rule for the direction to move to determine the next pas- senger to interview. One such rule for airline gate lounges was suggested in Section 5.2.2. A similar rule for baggage claim areas might be: Start with the person standing closest to the airline baggage office door, then select every sixth person, proceeding clockwise around the baggage claim device. The exact starting point is not particularly important, and interviewers can use their judgment as to which person is closest to a defined point in ambiguous cases. The point is to ensure that the interviewers follow a defined rule to select the passengers to interview rather than selecting them on the basis of whether they appear likely to be cooperative or some other criterion that might bias the sample. Similarly, the interval between sampled passengers is not particularly crit- ical. The objective is to ensure that interviews take place throughout the area in question and so should be chosen in the light of the number of interviews anticipated to be performed and the number of passengers expected to be present in the area. In the case of baggage claim areas, allowance should be made for greeters waiting with the passengers. When there is more than one interviewer performing interviews in a given area, the sampling rule will need to ensure that they each start at a different location and the sequences of selected passengers do not overlap. Because the number of interviews that can be performed while the passengers are present in an airline gate lounge or baggage claim area is fairly limited, overlap is not usually a problem. For example, if each interview takes an average of four minutes and the passengers in a gate lounge are surveyed over a 40-minute period, each interviewer will only be able to complete 10 interviews. One issue that arises with airline gate lounge surveys is that the seats may all be taken before boarding of the flight begins and thus passengers arriving after this point will either stand in the gate area or find a seat in nearby lounges. A sampling protocol that includes only passengers seated in the gate lounge will systematically exclude passengers arriving close to flight departure

OCR for page 90
94 Guidebook for Conducting Airport User Surveys time. It may therefore be desirable to modify the sampling rule to include standing passengers when most of the seats are taken. In practice, where passengers sit (or stand) in a gate lounge or stand in a baggage claim area may be influenced by characteristics such as the time they reach the gate lounge or the order in which they arrive at the baggage claim device (which is influenced by the order in which they deplane and their walking speed through the terminal). However, as long as the sam- pling protocol includes all occupants of the gate lounge or claim area and samples people in proportion to the number occupying each part of the area, these differences will be reflected in the resulting sample. Sampling Flights In the case of airline gate lounge surveys, whether interview or self-completed, it is necessary to select a sample of flights to survey. This selection is a particularly challenging problem, because of the need to ensure appropriate coverage of airlines, destinations, and times of day, while utilizing the survey personnel efficiently. Flights depart at irregular intervals throughout the day. There will be periods when a large number of flights will depart around the same time and periods with relatively few (or even no) flight departures. To complicate matters further, different airlines serving the same market will often have their flights depart around the same time, partly for competitive reasons and partly due to time zone differences. Although the characteristics of passengers on different airlines in the same market may not necessarily be significantly different, this is not always the case. Factors that can cause passenger characteristics to vary by airline include the following: Nationality of airline (flag of carrier) in international markets. Type of carrier: low-fare versus network. Market share and flight frequency. If the destination is a hub for one of the airlines. Flight departure times. Business travelers will tend to favor airlines that provide flight departures that better match the business day and those that offer higher frequency in case their travel plans change, while leisure travelers may be more willing to use flights at less convenient times and with fewer alternative departure times in order to obtain a lower fare. Passengers who are residents of cities that are airline hubs are more likely to be in the frequent flier program of the hub air- line, while residents of spoke cities are more likely to be in the frequent flier program of another airline, particularly if that airline has a stronger presence in the spoke city. Therefore, a fully stratified sampling plan would group flights by the following: Airline. Destination (possibly grouped by region). Time of day. In practice, at many airports this approach would generate a very large number of separate groups, and it would be impossible to develop a viable sampling plan by selecting a sample of flights from each combination of these factors. Instead, the usual approach is to define a sam- pling strategy that ensures that the selected flights provide a reasonable sample of each of these three factors. One such approach is to list flights throughout the period of the survey in order of scheduled flight departure time and calculate the cumulative number of departing seats for each flight. If information on average load factor by market and time of day is known at the level of individ- ual flights, then the cumulative number of expected passengers can be calculated in place of seats.

OCR for page 90
Air Passenger Surveys 95 Flights are then selected by finding every mth seat in the cumulative seat count, where m is the ratio of the total number of seats to the number of flights to be sampled, considering the expected number of completed interviews per sampled flight. This process ensures that larger aircraft have a higher chance of being selected than smaller aircraft, which offsets the smaller likelihood of a passenger on a large aircraft being interviewed if approximately the same number of interviews is performed for each sampled flight. If more interviews are performed for flights using larger aircraft, as could occur if more inter- viewers are assigned to those flights or more time is available for the interviews, or if all adult passengers are sampled using self-completed survey forms, then either the flight sampling process needs to be modified or the results weighted accordingly. Where all or a constant pro- portion of passengers are sampled, the procedure is simpler. After arranging all flights in order of scheduled departure time, select every mth flight where m is the ratio of total flights to the number of flights to be surveyed. The latter is calculated by the total number of passengers to be surveyed divided by the average passengers per flight times the proportion of passengers on each flight to be sampled. If the survey will only be conducted for certain periods each day, the cumulative list of seats is only determined for flights scheduled to depart during the times when the survey will be per- formed. After identifying an initial list of flights to sample, the characteristics of those flights in terms of the above criteria can be compared to the corresponding proportions across all flights, and any necessary adjustments made to the sample by dropping some of the flights with char- acteristics that are over-sampled and replacing each of them with the next flight in the list of flights with the characteristics that are under-sampled. Further adjustments may be necessary to ensure that the survey field teams have a fairly steady workload throughout their shift. A flight that has a scheduled departure time too close to that of other flights in the sample could be dropped and replaced with another flight with the same characteristics but a scheduled depar- ture time during a period when the number of selected flights is not enough to keep the survey field teams occupied. 5.3.4 Survey Timing At most airports, air passenger characteristics vary seasonally and it is therefore important to determine whether information on these characteristics are required for a specific period (e.g., the peak month) or on an annual basis. If information on air passenger characteristics is required on an annual basis, it will be necessary to perform the survey in two or more periods, reflecting the seasonal pattern of traffic. Depending on the information of interest, it would be desirable to select a peak and off-peak period that represent the highest and lowest levels of the relevant characteristics (e.g., private vehicle trips to and from the airport, or proportion of business travel). This method should allow the corresponding characteristics to be estimated for other months that are not surveyed, although this may not always be possible, depending on how the different characteristics vary during the year. The survey planning team may need to undertake some analysis of monthly variation in those characteristics in order to select appropriate months for the survey. Where seasonal variation in travel characteristics is more complex than can be expressed in terms of peak and off-peak conditions (e.g., winter travel patterns are very differ- ent from summer travel patterns as well as from travel patterns in the spring and fall), it may be necessary to divide the survey into three or four phases or conduct the surveys continuously throughout the year. In addition to seasonal variation, at almost all airports air passenger characteristics vary by time of day and day of the week. This variation can be addressed by ensuring that the survey pro- vides reasonable coverage of different times of day and days of the week. Ideally, the survey data

OCR for page 90
96 Guidebook for Conducting Airport User Surveys collection would take place throughout the day for at least a full week. At many airports, passen- gers begin arriving for early morning flights by 5 a.m. and the last flight does not depart until midnight or later; therefore, the survey would need to cover about a 19-hour day. However, the staffing levels required to achieve this coverage are often impractical, and spreading the desired sample size across every time period may unduly constrain the sampling in each period. There- fore, it is common to survey for a limited period each day (typically at least eight hours) but vary the timing of these periods from day to day. Where traffic peaks occur at certain times on par- ticular days (e.g., Monday morning and Friday and Sunday evenings), the survey shifts should be scheduled to provide coverage for these periods. While some flights may depart early in the morning or late in the evening, typically these are fairly few and the sampling strategy should reflect the relatively small proportion of passengers at these times. It may be adequate to have only one or two survey staff during these times and only survey on a few days. However, at airports with international service, departures for certain markets may take place late at night because of time zone differences or to allow early morning arrivals at the destination. Similarly, some eastbound flights from the West Coast depart late at night in order to arrive at mid-continent hubs or East Coast destinations early the next morn- ing. This may require additional survey staffing at these times to make sure that those markets are adequately surveyed. At other times of day, there may be very few departures for periods of several hours at a time. Such flight schedules can present a challenge to survey staffing, because in general staff will want to work a full shift. It may be possible to schedule staff so that meal breaks occur during periods of low activity, although there are constraints on how long staff can work between breaks, and obviously meal breaks need to occur around the middle of the shift. With careful scheduling, it may be possible to provide increased staffing levels during busy periods by scheduling shifts to overlap during these periods. Figures 5-6 and 5-7 illustrate how traffic levels and the composition of the traffic can vary over the day or the days of the week. The figures show the distribution of departing seats from Logan International Airport in Boston for different markets for the week of the Thanksgiving holiday Canada Cen.Am & Caribb Europe US < 500 miles US 500-999 miles US 1,000+ miles 5,000 4,500 4,000 3,500 Departing Seats 3,000 2,500 2,000 1,500 1,000 500 0 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 Hour of Day Source: Official Airline Guide for November 19, 2007 (excluding flights to Africa, 185 seats Tuesday and Friday). Figure 5-6. Numbers of departing seats from Boston Logan International Airport by time of day and destination region.

OCR for page 90
Air Passenger Surveys 97 Canada Cen.Am & Caribb Europe US < 500 miles US 500-999 miles US 1,000+ miles 60,000 50,000 Departing Seats 40,000 30,000 20,000 10,000 0 Monday Tuesday Wednesday Thursday Friday Saturday Sunday Thanksgiving Holiday Day of Week Source: Official Airline Guide for November 19, 2007 (excluding flights to Africa, 185 seats Tuesday and Friday). Figure 5-7. Numbers of departing seats from Boston Logan International Airport by day of week and destination region. in 2007. The traffic level varies widely over the day, with early morning and late afternoon/ evening peaks and a sharp drop in long-haul domestic departures (stage length of 1,000 miles or more) between 12:00 and 1:00 p.m. (12 and 13). Very few flights depart after 9:00 p.m. (21), most of them to Europe. As shown in Figure 5-6, departures for different international markets vary considerable through the day, with those to Central America and the Caribbean in the morning and those to Europe mostly in the evening. To adequately represent these markets in the sample, the sampling plan would need to ensure that the flights selected to be surveyed at those times include an appro- priate number of international flights. Figure 5-7 shows a reduction in the number of flights on the Thursday of the Thanksgiving holiday, with the number of flights gradually returning to nor- mal levels over the subsequent days. One approach to ensuring that the survey responses are appropriately distributed by time of day is to schedule interviewers so that the distribution of total interviewer time by time of day and day of week is similar to that of the departing seat capacity of the airport. For smaller total sample sizes (e.g., under 1,000 responses), interviewers are assigned to cover a larger number of flights by sampling only a small number of passengers from each flight (e.g., four to twelve depending of the size of the aircraft). Because connecting passengers tend to arrive in airline gate lounges earlier than originating passengers, starting to survey flights too long before flight depar- ture time will tend to over-sample connecting passengers. The interviews for a given flight should generally start no earlier than an hour before flight departure (90 minutes in the case of interna- tional flights) and be evenly distributed up to the time that boarding begins. This approach can work well at small to medium-sized airports. At larger airports, interviewer times would be deter- mined for each terminal in proportion to the numbers of passengers or departing seats from those terminals. Responses are then weighted to match departing passengers by time of day, flight destination, and airline using flight schedules and estimated load factors. As a practical matter it is difficult to schedule interviewers efficiently and match the inter- viewer time to the variation in departing seats on a daily basis. Instead, an attempt can be made to match interviewer time to departing seats in each hour separately for weekdays in total and weekend days in total. The start times and lengths of shifts and break times can be varied so there