Cover Image

Not for Sale

View/Hide Left Panel
Click for next page ( 102

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 101
Air Passenger Surveys 101 The most important distinction to make is between vehicles parked by air passengers for the duration of their air trip and those parked by well-wishers or greeters. This distinction is com- plicated by the frequent practice of airports calling different parking lots or facilities short-term or long-term parking based on pricing and distance to the terminal, rather than the amount of time drivers are allowed to park. Long-term parking facilities may be located some distance from the terminal, and users may think of these as being off-airport, even though they are operated by the airport authority. Then there are privately operated parking lots in the vicinity, which are usually considered off-airport parking. In addition, some airport area hotels may offer parking at competitive rates with the airport. Although most passengers parking a vehicle for the duration of their air trip will use the park- ing facilities designated for daily or longer parking, rather than those for hourly or short-term parking, some passengers making a one-day or overnight trip may choose to use the closest park- ing to the terminal, and pay the higher rate. Finally, some survey respondents may think of a private vehicle standing at the terminal curb for a few minutes while the passengers and their baggage are unloaded as being "parked" for a short while. It is therefore desirable to ask survey respondents to identify the parking facility where the vehicle is parked and the duration that it was (or will be) parked, rather than rely on vague cat- egories such as "parked short term." These questions should provide response options that use the formal designation of different parking facilities (e.g., hourly parking, economy lot, termi- nal garage), but allow respondents to write in or state other locations if they do not recognize the correct names of the facilities. This will often be necessary anyway in the case of off-airport park- ing, where there may be a large number of different providers. 5.5 Weighting Survey Responses In spite of the survey team's best efforts to design the sampling plan to obtain a representative sample of air passenger trips, it is very unlikely that the responses will fully reflect the composi- tion of the target population, because of unavoidable consequences of the survey methodology as well as varying response rates by different categories of travelers. Therefore it will be necessary to weight the survey responses in order to improve the accuracy of the resulting data. Because the exact composition of the air passenger population is generally unknown (this is why the survey is being performed), a variety of other types of data is needed to calculate the sur- vey response weights. These data should be assembled at the time the survey is performed--or as soon as possible afterwards--and should include the following: Enplaned passengers on each flight (if available)20 or enplaned passengers by flight destina- tion for the month in question.21 Number of connecting passengers on each flight (if available) or estimated from available data.22 Parking lot exits from each airport parking facility by parking duration and hour for each day of the survey period. 20 Some airports collect these data routinely. Where this is not the case, the airport may be able to obtain this information from the airlines for the period of the survey with an assurance that it will not be made public and only used to help analyze the sur- vey results. 21 These data are available from the U.S. Department of Transportation Bureau of Transportation Statistics T-100 database. 22 Quarterly data on connecting passengers by airline and flight sector can be estimated from the U.S. Department of Trans- portation Bureau of Transportation Statistics Airline Origin and Destination Survey database.

OCR for page 101
102 Guidebook for Conducting Airport User Surveys Automated vehicle information system counts (where available) by class of vehicle (taxis, lim- ousines, shared-ride vans, shuttle buses, etc.) and by hour for each day of the survey period. Ridership statistics on airport-operated shuttle buses to remote rental car facilities or rail sta- tions by hour for each day of the survey period. Where these are not routinely recorded, it will be necessary to arrange for their collection for the survey period. Ridership statistics on scheduled airport bus services by run (or hour) for each day of the sur- vey period. Where these are not routinely reported to the airport authority, the airport may be able to obtain them from the operators with an assurance that they will not be made pub- lic and only used to help analyze the survey results. Terminal roadway traffic counts (where available) by hour for each day of the survey period. Where these data are not routinely collected, consideration should be given to placing traffic counters on the terminal roadways for the duration of the survey. The process of calculating survey response weights consists of two steps: 1. Calculation of weights to correct for known bias in the survey sampling methodology. 2. Calculation of weights to correct for differences between the survey results and external data on traffic composition. Each survey response should include the size of the air party. If the survey responses reflect air passengers (i.e., there are multiple survey responses for parties with more than one passenger), then counts obtained from the survey responses should be divided by the air party size in order to express the traffic composition in terms of air parties. Conversely, if the survey responses reflect air parties (i.e., there is only one survey response for each air travel party), then counts obtained from the survey responses should be multiplied by the air party size in order to express the traffic composition in terms of air passengers. Statistical computer software packages can perform these adjustments very easily in tabulating survey results. The difference between expressing survey results in terms of air passengers or air parties is critically important to the correct interpretation of the survey results and should be clearly understood. 5.5.1 Proportional Weighting Proportional weighting uses weights that adjust the proportions of the survey response data to reflect the proportions of the control data without changing the total number of responses. In general this will result in non-integer counts for many reported survey responses when expressed using weighted data. Proportional weights can only adjust survey response data to correspond to the proportions of a single characteristic of the control data. Separate weights can be deter- mined for different characteristics, but in general it is not possible to determine response weights that adjust survey response data to correspond to the proportions of multiple characteristics of the control data. If there are N total survey responses and ni of those responses reported some characteristic i that composes a proportion pi of the population in the control data, then the proportional weight wi that should be assigned to each of the n responses is given by: pi N wi = ni Since all N of the survey responses must have reported some value for characteristic i (even if this was only "Don't know" or "Refused"), a weight wi will be assigned to each survey response. Statistical analysis software can generally be set to optionally exclude missing data cases--such as "Don't know"--from the tabulated results. However, some users of the results may be interested

OCR for page 101
Air Passenger Surveys 103 in knowing the extent of missing data in the survey responses. Therefore, it is better to set the weight for missing data responses to one rather than zero, and adjust the other weights so that the weighted total of the non-missing cases corresponds to the unweighted total of the non-missing cases (i.e., replace N by N minus the number of missing data responses in the above equation). 5.5.2 Correcting for Known Bias in the Sampling Methodology The sampling methodology adopted for the survey may introduce some bias into the response data that can be calculated and corrected. The most obvious example occurs with self-completed surveys handed out to all adult passengers in an airline gate lounge where fewer responses are received from a given air party than the number of passengers in the party. This case will always occur where there are children in the air party (who do not complete the survey). Some passen- ger surveys have asked the respondents to indicate how many members of their air party have completed a survey form and have then used this information to weight the results. However, experience indicates that these statements are often unreliable. Some respondents may misunder- stand the meaning of the term "air party," while others may not realize that another member of their party is also completing a form. Or they may think that another member of their air party is completing the form, but in fact that form is not turned in. It is therefore preferable (although more time consuming) to examine the survey responses; identify responses from the same party based on the party characteristics, such as their trip origin address or other information; and revise the reported survey completion information before calculating weights to correct for under- reporting of air party members. It is also quite common to apply weights to self-completed survey responses to factor up the responses to the number of passengers boarding the flight. There are two problems with this approach: It can give a misleading impression of the number of survey responses, as discussed in Sec- tion 5.5.4, unless the resulting weights are normalized to ensure that the total of the weighted responses equals the actual number of survey responses. It will over-weight responses from under-sampled flights. For example, if generally 50% of passengers on sampled flights are surveyed, but on a particular flight only 10% of passengers are surveyed for some reason, the responses from passengers on the under-sampled flight will be weighted by a factor of 10, rather than the factor of 2 used on other flights. Differences between the distributions of characteristics on a particular flight with only a few respondents compared to other flights in the same market are most likely due to the higher vari- ance that occurs with small samples, not because the characteristics of all the passengers on that flight are different. Scaling up the responses to the total number of passengers on the flight implicitly assumes that all the passengers on the flight have the same distribution of character- istics as the respondents. For example, if only four respondents are surveyed on a particular flight and one of these is leaving on a 10-week trip to Japan, it would be incorrect to infer from this that 25% of the passengers on the flight are leaving on 10-week trips to Japan, but that would be the effect of scaling up the responses. There is of course no way to tell from the results of a survey whether differences in the char- acteristics of the respondents on different flights are due to a true difference or are simply a result of the sampling variance. It is possible to perform statistical tests to determine whether the hypothesis that the results are drawn from the same distribution can be rejected at some level of confidence, but that is not the same thing as knowing that they are different. With a small sam- ple size, the variance in any particular characteristic is likely to be so high that it is unlikely to be possible to reject the hypothesis that the results are from the same distribution as that for other flights in the market at any reasonable level of confidence. Therefore, one is left to make the not

OCR for page 101
104 Guidebook for Conducting Airport User Surveys unreasonable assumption that the underlying distributions of the characteristics of passengers in a particular market are the same, and that differences across flights in that market are due to sampling variance. Of course, the distribution of a particular characteristic within each market may vary by other dimensions, such as the time of day or day of the week, further complicating the analysis and reducing the ability to determine whether any apparent differences across flights in the market are simply due to chance. Because passengers on under-sampled flights are likely to be less representative of the character- istics of passengers on other flights in that market, factoring up responses to the total number of pas- sengers on a flight will over-weight those passenger responses that are less representative of the characteristics of the market in question, potentially biasing the results of the survey. Therefore, it is better to consider those passengers who did complete the survey forms as a representative sample of air passengers and make any required adjustment to correct for differences between the survey results and the distribution of air traffic across different markets, as discussed in the next section. In the case of interview surveys, there is likely to be sampling bias that results from the sampling protocol: A sequential sampling strategy in an airline gate lounge will miss any passengers who arrive in the lounge after surveying has started and sit in areas that have already been sampled. Thus passengers arriving closer to the time boarding begins have a lower chance of being sampled. If the survey team performs approximately the same number of interviews in each gate lounge, irrespective of the number of passengers on the particular flight, the probability of a given air party being surveyed is lower on flights with more passengers. If the distributions of characteristics of respondents who are over-sampled are the same as those who are under-sampled, the difference in sampling rate will not affect the survey results. How- ever, if the distribution of some characteristic is different, the results will be biased. For example, it is likely that some passenger characteristics, such as trip purpose or air party size, will differ between those arriving in an airline gate lounge well before boarding begins and those arriving shortly before boarding begins. Similarly, if interviews of passengers exiting security screening are performed at approximately the same rate--as is likely with a survey team of a constant size--the result will be a lower sampling rate during busy periods. If passenger characteristics are different between busy periods and slow periods (as is quite likely), the results will be biased. Weights can be calculated to adjust for these sources of bias by examining the results for dif- ferent periods or subgroups of respondents to see if there are any differences in the distribution of characteristics that might vary by period or subgroup. If such differences are found, the sur- vey responses can be weighted by the ratio of the number of air parties or air passengers in each period or subgroup to the number of responses obtained for that period or subgroup. Such weights should be normalized so that the total number of weighted responses is the same as the number of actual responses. 5.5.3 Correcting for Differences Between the Survey Results and External Data Once a set of weights has been determined to correct for known bias in the sampling method- ology, an additional set of weighting factors can be calculated, using the weighted results to cor- rect for differences between the weighted results and external data on the composition of the passenger traffic using the airport. The most obvious potential difference between the survey results and external data on the com- position of the passenger traffic at the airport is if the percentage of passengers in each flight des- tination market given by the survey responses does not agree with the passenger traffic reported

OCR for page 101
Air Passenger Surveys 105 by the airlines. Because connecting passengers may have been sampled at a different rate from originating passengers, it will generally be advisable to consider connecting passengers boarding a flight as a separate market from originating passengers and calculate separate weights for each. Other characteristics of the survey respondents for which it may be worth calculating weight- ing factors include the following: Airline. Time of day and day of week of flight. Ground access mode use by originating passengers. Where several different weighting factors have been calculated for different survey response characteristics, it will generally be advisable to compare the results for each characteristic using the appropriate weighting factor with the corresponding results using each of the other weight- ing factors, in order to determine the sensitivity of the results to the choice of weighting factor. 5.5.4 Weighting for Total Traffic It is common for survey responses to be assigned weights that convert the total number of sur- vey responses to the corresponding count of annual passenger traffic. While this process allows the survey results to be directly expressed in terms of the corresponding annual passenger char- acteristics, two important caveats should be borne in mind before doing this: The characteristics of the air passenger market at a given airport will vary throughout the year, while the survey data will generally have been obtained at one or two discrete points in time. Thus the resulting data may be quite misleading. For example, if a survey is performed dur- ing August, it will reflect a high proportion of vacation travel. This result is unlikely to corre- spond to the characteristics of the air passenger population during the rest of the year. Expressing the results of the survey in terms of annual passengers conceals the true size of the survey sample and may give a completely false impression of the accuracy of the results. For example, for a survey with 1,200 responses at an airport handling 12 million annual passen- gers, each survey response is equivalent to 10,000 annual passengers. Thus if the estimated number of annual passengers with some characteristic was given as 22,400 (after weighting for other considerations), it might easily be overlooked that this represents only two survey responses and is likely to be highly inaccurate. Therefore, it is recommended that survey results not be expressed as annual traffic, but rather that weights be calculated so that the resulting totals of weighted responses equal the size of the actual survey sample. It is easy enough for users of the survey results to factor the results up to the level of annual traffic if they so desire, but they will then be fully aware that they have done this and should recognize the accuracy limitations that this implies. One advantage of expressing the survey results in terms of the sample size is that it allows users to easily distinguish whether the results reflect the distribution of air passenger characteristics or air party characteristics, because the response totals will be quite different in each case. For example, if the average number of air passengers per air party is 1.4, a survey with 5,000 air party responses will show results summing to 5,000 when showing the distribution of air party char- acteristics and 7,000 when showing the distribution of air passenger characteristics. If the results are weighted so that the total is equal to the annual passenger traffic, it may be unclear whether they are showing the distribution of passenger characteristics or air party characteristics (which will typically be different).23 23 If the survey results will be expressed in terms of annual traffic, it is important that results showing air party characteristics be weighted to give the total number of annual air parties, not air passengers.