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.
111 Final Perspectives and Future Research Opportunities Research for this project has examined a wide range of existing studies of air passenger demand and air passenger activity by many types of researchers, as well as the types of data typically used to design and conduct these studies, along with the sources of these data. This effort has focused both on understanding how and why these studies were conducted and the extent to which the studies made use of disaggregated socioeconomic data. The research team used what was learned about the state of practice in modeling air passenger demand as a starting point for new research into how disaggregated socioÂ economic data could be used to improve the modeling of passenger demand and observed enplanements. This new research proceeded along several paths. One of these research paths investigated what could be learned from air passenger, household travel, and consumer survey data about how air passenger travel demand varies with the disaggregated demoÂ graphic and socioeconomic information about air travelers that is typically collected in these surveys. A second research path investigated the effects of including a disaggregated socioeconomic variable in traditional models of air passenger demand at individual airports (which already make use of aggregated regional socioeconomic variables). This research took a case study approach, evaluating the results from including a disaggregated socioeconomic variable in traditionally structured regression models of annual O&D enplanements for eight U.S. airports or airport systems. A third path examined alternative ways new specifications and equation forms could incorÂ porate disaggregated socioeconomic variables or relationships in models of air passenger activity at an airport or in an airport system. This effort used insights from the analysis of past air pasÂ senger surveys and the differences in the propensities to travel by air of different socioeconomic cohorts revealed by those analyses. The fourth research path evaluated the usefulness of new forms of disaggregated socioÂ economic data about air passengers. These data were developed from deÂidentified financial transaction records of individuals who made use of personal financial management software. While only a small fraction of such consumer spending is directly related to air travel choices and behavior, these digital forms of data collection represent a rapidly evolving frontier for the development of information about consumer purchase choices in a wide range of markets, including passenger aviation. These four research avenues pursued in this project together comprise a multifaceted approach to developing an improved understanding of the potential value of disaggregated socioeconomic data in the analysis of air passenger demand for airports. Aspects of this overall C H A P T E R 6
112 Using Disaggregated Socioeconomic Data in Air Passenger Demand Studies approach directly address the four forwardÂlooking questions identified in the introduction to this report: â¢ How does air passenger behavior differ across socioeconomic cohorts or subgroups, especially with respect to individual propensities to choose to travel by air? â¢ Can disaggregated socioeconomic factors be introduced into traditional approaches to modeling air passenger demand at individual airports or systems of airports, and do those variables provide new information compared to a baseline of traditional air passenger demand modeling using aggregate socioeconomic variables? â¢ Can new approaches to structuring econometric models or other approaches be developed and used to realize the value of incorporating disaggregated socioeconomic data in underÂ standing or modeling air passenger demand? â¢ Are there new forms of disaggregated socioeconomic data, or new ways of collecting such data, that can be used to model or study air passenger demand at airports and in regions? In this report we have summarized how we conducted project research to address these questions, and also reported the results from these research efforts. In the remainder of this chapter, we present the findings from our research and the research opportunities that we believe these findings open for future researchers. Summary of Research Approach This research project has investigated the effectiveness of incorporating the use of disaggreÂ gated socioeconomic variables in studies of air passenger demand and air passenger activity at airports. The research was organized to first identify how such studies have been conducted and reported in the past. The review of these past examples extended across research conducted by a wide range of individuals and organizations, including academics, governmental and industry organizations, and airport practitioners and consultants. The research next covered the types of data that have been used in studies of air passenger demand as well as the types and sources of disaggregated socioeconomic that could also be used in such studies. This research identiÂ fied trends in the evolution of age and household income cohorts in the United States. The research then extended to an analysis of air passenger and consumer surveys to identify ways the propensity of individuals to travel by air vary by their demographic and socioeconomic charÂ acteristics, such as age, gender, or household income. The research then used a case study approach to examine the effects of adding a disaggreÂ gated household income variable as an independent regressor to more traditional models using aggregate socioeconomic variables. This model performance comparison was done for eight case study airports using observations of annual O&D enplanements between 1990 and 2010 for seven U.S. airports and one U.S. airport system. This approach permitted the assessment along several dimensions of the impact of incorporating this type of data in air passenger demand modeling. A second case study analysis was conducted using data from the BaltimoreâWashington airport system to examine the effectiveness of more sophisticated model formulations, again comparing model performance without and with the inclusion of a form of disaggregated socioÂ economic variable among the independent variables of the regressions. The research then examined the potential for using new forms of disaggregated (or more individualized) socioeconomic data for air passenger demand modeling, based on individual financial transactions data from a group of individuals.
Final Perspectives and Future Research Opportunities 113 Summary of Findings from Research The primary goal of the research project was to investigate the extent emerging or ongoing socioeconomic changes in the population (such as the age structure of society, increased immiÂ gration, wealth concentration, geographic redistribution of the population, and changing views on the use of disposable income) are likely to influence the future demand for air travel and may not be well captured in current approaches to air passenger demand modeling. More specifiÂ cally the research explored whether the inclusion of disaggregated socioeconomic data, such as regional distributions of age, gender, or household incomes, in air passenger demand models can improve the ability of those models to anticipate future changes in the overall demand for air travel and composition of the air traveling public. The research found that the use of air travel by different subsets of the population (in terms of household income, age, race and ethnicity) varies widely. It is clear that changes in the disÂ tribution of these characteristics across the population are likely to have a significant impact on future air travel demand. In particular, an increasing concentration of wealth and income in the wealthiest segment of society appears likely to reduce the amount of air travel compared to a less concentrated distribution for any given level of total income, excluding the effect of other factors, for reasons discussed earlier in this report. At the same time, increasing concentration of wealth and income may continue to change air travel in more qualitative ways, with airlines charging separately for service amenities that some segments of the traveling public are willing to forego in order to obtain a lower airfare. Similarly, an aging population will move an increasing proportion of the population into age ranges that make fewer air trips per year than those in the age range from 45 to 65 that make the most air trips per year on average. The research explored a number of different ways of reflecting these trends in air passenger demand studies and models, although an attempt to develop models of enplaned air passenger traffic for eight case studies of individual airports and one regional airport system that incorÂ porated disaggregated variables that reflected the distribution of household incomes met with mixed success. Adding a disaggregated household income variable to relatively simple models that used only two aggregate socioeconomic variables did not noticeably improve the ability of the models to either explain past air passenger enplanements or predict future passenger trafÂ fic levels in a simulated forecasting exercise. However, a more complex model using a different disaggregated household income variable did show improvement in either its ability to explain past levels of air passenger enplanements or its ability to predict future air passenger traffic. While this improvement was not large, the implications for future levels of air passenger demand of the change in the implied demand elasticity (model coefficients) between the model without the disaggregated income variable and the model with the variable were significant. This finding has broader implications beyond the specific models estimated in the research. Elasticity values obtained from air passenger demand models are sometimes used or quoted in other studies. If these values are biased due to the omission of disaggregated socioeconomic variables in the models from which they were obtained, this could distort the results of these other studies. Future Research Opportunities The analysis of disaggregated response data from air passenger, household travel, and conÂ sumer expenditure surveys raised a number of issues that appear deserving of further research. These are discussed in more detail in the description of the analysis findings in Chapter 3, but can be summarized as follows: â¢ The analysis of the three broad categories of survey (airport intercept, household travel, and consumer expenditure) gave different estimates of air travel propensity for given population
114 Using Disaggregated Socioeconomic Data in Air Passenger Demand Studies subgroups. In particular, the California Household Travel Survey appeared to undercount air trips by California residents. Further work to resolve these differences in findings from different surveys is needed. â¢ The analysis of the survey results looked at different socioeconomic factors (income, age, etc.) separately, but in reality they are most likely correlated. Further research to explore the impliÂ cations of this correlation and develop techniques to account for different socioeconomic factors simultaneously would be valuable. â¢ The demand for air travel depends not only on socioeconomic factors of the population but also on the level of airfares experienced by travelers from a given region. It would be valuable to develop techniques to account for differences in airfare levels between airports or regions in order to be better able to compare findings on air travel propensity by household or air traveler socioeconomic characteristics across different airports or regions. The case study regression analysis relied on a relatively simple model specification to investiÂ gate the effects of incorporating a specific example of a disaggregated socioeconomic variable on the performance of baseline models relying on aggregate socioeconomic variables (and another independent variable unrelated to socioeconomic factors). While the specific dis aggregated socioeconomic variable chosen for these comparisonsâone reflecting regional household income distributionsâis a natural candidate for consideration in a model of air passenger demand, there may be other disaggregated socioeconomic variables that could be analyzed. In addition, it may be valuable to analyze other ways of introducing the information from the disÂ aggregated variable into the regression, since the comparisons of baseline and alternative models were affected by the strong correlation between the aggregate and disaggregated variables used in the case study regressions. The more detailed analysis of the passenger enplanements in the BaltimoreâWashington regional airport system performed as part of the overall case study analysis developed a model that provided an excellent fit to the historical data using a range of independent variables that make intuitive sense and had estimated coefficients that had values that appeared reasonable and were estimated with a high level of statistical significance. As noted above, including a disaggreÂ gated variable for household income distribution not only improved the fit of the model to the data and its predictive ability but also changed the estimated values and increased the statistical significance of the other variables, suggesting that omitting such variables from air passenger demand models could lead to biased estimates of the model coefficients. However, there were aspects to this model that are deserving of further research, in particular: â¢ In addition to the continuous variables the model made use of yearÂspecific dummy variables that accounted for factors not reflected in the continuous variables. These played a major role in the model fit. It would be valuable to explore other continuous variables that could account for these effects without relying on yearÂspecific dummy variables (which are problematic in using models for forecasting). â¢ The model used average airline yield at a national level as the airfare price variable. However, this may not be a good reflection of airfares for any particular airport. It would be valuable to explore the use of a variable that better reflects the average airfares at the airport(s) in questions. â¢ The model used a disaggregated variable for household income distribution but did not conÂ sider changes in the distribution of other socioeconomic factors, such as the age of air travelers. It would be valuable to explore how to incorporate changes in the distribution of other socioÂ economic factors in similar models. â¢ The model variables reflected socioeconomic factors for residents of the BaltimoreÂ Washington region, but of course a large proportion of the passengers at the regionâs airports (over half in the case of the BaltimoreÂWashington region) are visitors. It would be valuable to explore how to account for socioeconomic factors of visitors as well as residents.
Final Perspectives and Future Research Opportunities 115 More broadly, the approach followed in developing the more detailed model of the BaltimoreÂWashington region needs to be applied to other airports and regions to determine whether the resulting models for those airports or regions also show a similar improvement or whether the results found for the BaltimoreÂWashington region are somehow unique to that region or maybe even an artifact of the data used in the analysis. In addition, the BaltimoreÂ Washington region is one of the largest multiÂairport regions in the country and the airports serving the region are much larger than most other individual airports to which the approach could be applied. It would be valuable to explore whether such differences in scale or the extent of the airport system being modeled also affect the applicability or performance of this modeling approach. There are several research opportunities related to using the new data sources for airport demand modeling studies discussed in Chapter 5. We found that many of the most successÂ ful applications of using cell phone or locationÂbased data for airports were for revenueÂ generating and costÂreduction purposes. In the context of airport demand studies, we found successful applications using cell phone and ticketing data to analyze airport catchment areas. In terms of future research, cell phone usage could be a source of new data on passenger choices and passenger movements through terminals. Future research could explore linking cell phone tracking data to smallÂarea demographic and socioeconomic data to explore air travel propensity by household characteristics without the need for air passenger surveys (or to track changes between surveys). Cell phones could also be used as a tool to gather data simiÂ lar to the types of data collected in traditional airport survey efforts. For example, cell phone beacons and airportÂspecific phone apps could invite air travelers to participate in an online air passenger survey. Travelers could download a survey form that they could complete at leisure (for example, on their flights) and then submit online when they next have an internet connection. Both of these future research opportunities help address one of the key challenges to using new sources of disaggregated data based on cell phone, GPS, or financial transactions, namely that due to the need to protect individualsâ privacy there is limited socioeconomic information available from these sources directly. Chapter 5 also discussed alternative potential approaches to incorporating disaggregated socioeconomic factors in air passenger demand studies and models. However, the resources of the project only allowed one of these approaches to be explored and then only with a very limited number of variables and for only one socioeconomic factor (household income). It would be valuable to develop techniques for applying the other approaches or including a broader range of factors and to explore their relative feasibility and effectiveness. Because the sources of data on air passenger enplanements do not typically distinguish between travel by residents of a region and visitors to a region or between travel for personal and business purposes, models of air travel demand have generally attempted to model total air travel. Since the survey analysis undertaken in the course of the project shows, not surprisingly, that the average number of annual air trips made for business and personal purposes differs considerably by household or individual characteristics, while it can also be expected that the factors influencing air travel by residents of a region and visitors to the region are different, it would be valuable to explore the feasibility and effectiveness of developing air passenger demand models that can distinguish between travel by residents of a region and visitors to the region or predict travel by trip purpose. The foregoing potential research opportunities represent a wide range of possible research projects that would build on the work performed in the current project and that could be purÂ sued through future research. The description of the research performed for the current project and the discussion of potential research opportunities in this report provide a fairly clear indication of how potential future research projects could be undertaken. Nonetheless, more detailed research problem statements could be developed following publication of this report.
116 Using Disaggregated Socioeconomic Data in Air Passenger Demand Studies This would allow other interested researchers to contribute suggestions and comments on the scope of future research. Although each of the potential research opportunities would contribute to a better underÂ standing of how air passenger demand is influenced by disaggregated socioeconomic factors and how to best reflect this in air passenger demand models, and hence there is no obvious priority between the various research needs, it would seem reasonable to focus initially on two issues: explore applying the modeling approach used for the more detailed analysis of the BaltimoreâWashington region to other airports and regions in order to determine how transÂ ferable these findings are and investigate the reasons for the differences in air travel propensity between air passenger intercept surveys and household travel and expenditure surveys. A clearer understanding of the reasons for these differences would allow future research on air passenger demand to draw on a broader range of disaggregated socioeconomic data with greater confiÂ dence, as well as potentially provide useful guidance for the design of such surveys in the future. Conclusions Passenger aviation is often described as an important catalyst for broad economic growth and is often depicted, qualitatively and in formal models of air passenger demand, as being driven by aggregate measures of regional economic activity and growth. Yet it is also underÂ stood that air transportation is not used to the same extent by all segments of society. This project has examined whether this growing understanding of the differences in propensities to fly that depend on individual and household characteristics such as household income, age, or educational status can be used to improve models and forecasts of air passenger activity. These differences were consistently seen in analyses of a number of surveys of air passengers and consumers. While data about distinct subgroups of individuals and householdsâdisaggregated socioÂ economic dataâand about how their distributions of characteristics are changing are availÂ able, these data are sometimes correlated with the aggregate regional socioeconomic data used in the past for the airport passenger demand models used in a variety of airport studies and planning efforts. This correlation can raise statistical challenges for simple approaches to incorporating some forms of disaggregated socioeconomic data into existing models of air passenger demand. The projectâs analysis of the effect of incorporating a particular disaggreÂ gated socioeconomic variable into fairly simple regression models of annual enplanements in case studies of seven individual U.S. airports and one regional airport system ran into such challenges, while resulting in only modest improvements in the performance of the models that included the disaggregated socioeconomic variable. Therefore, care must be taken in including disaggregated socioeconomic variables in air passenger demand models to avoid or control for potential correlation between aggregate and disaggregated socioeconomic variÂ ables to the extent possible. It should also be recognized that there may be tradeoffs in model performance between omittedÂvariable bias in models that only include aggregate socioÂ economic variables and a loss of statistical significance and potential bias in coefficient estiÂ mates if disaggregated socioeconomic variables are included that are partially correlated with the aggregate socio economic variables. As more experience is gained in using disaggregated socioeconomic variables in air passenger demand models, it should become clearer how best to resolve such tradeÂoffs. Although including a disaggregated household income variable in the simple case study regression models did not show a significant improvement in the predictive ability of the models, including a different disaggregated household income variable in a more complex model speciÂ fication for the case study of the BaltimoreâWashington regional airport system did provide an
Final Perspectives and Future Research Opportunities 117 improvement in both model fit to the historical data and predictive ability. Furthermore, includÂ ing the disaggregated income variable resulted in a significant change to the demand elasticities implied by the estimated model coefficients. Three important conclusions can be drawn from this experience: â¢ The way in which disaggregated socioeconomic variables are defined is important. If they are defined in a way that partially reflects the factors measured by the aggregate socioeconomic variables in a model, their addition to the model may not result in any improvement in model fit and may reduce the statistical significance of the estimated coefficients of the aggregate variables in the model. â¢ Simple models with relatively few independent variables that do not fit the historical data very closely are not likely to show much improvement by including variables that reflect the distribution of aggregate socioeconomic factors included in the model. This is because there are clearly factors influencing the dependent variable that are not well represented by the independent variables. It is not likely that including a variable that reflects the disÂ tribution of one of the factors measured by an aggregate variable will rectify this problem and the model estimation process may use the additional variable to account for factors that happen to be correlated with it, distorting not only the estimated coefficient of that variable but those of other variables as well. â¢ For use in air travel demand forecasting what matters most in a model is that the demand elasticities implied by the model coefficients accurately reflect the likely effect of the indeÂ pendent variables in question. Good fit to the historical data is desirable and provides some assurance that the model is reasonable, but if this is achieved through biased coefficients of the independent variables the ability of the model to correctly anticipate the effect of any given scenario for future values of the independent variables will be compromised. An alternative approach to the modeling undertaken in the case study analysis, given initial investigation in this project, would be to develop new models or new forms of disÂ aggregated data that may avoid some of the difficulties encountered in the case study analyÂ sis. The modest but real improvements in model performance observed in the project case studies and the consistent results from survey analysis linking respondent socioeconomic characteristics and differences in household propensity to travel by air suggest that continÂ ued research would be worthwhile in this area. Further investigation and exposition of this alternative approach to air passenger demand modeling could be the basis of additional research. In addition, a second project could focus on further development of the model specifications developed for the BaltimoreÂWashington region, applying them to other airÂ ports and airport systems, and explaining the modeling results and their implications. Such research may bring new insights to airport managers and decision makers facing a changing air travel marketplace. It was noted that airports use air passenger demand studies for many purposes, from analyÂ sis of behaviors and choices of the passengers they currently serve or hope to serve to the supÂ port of an airportâs planning and preparation for future passengers and services. The research conducted in this project can provide value for those who will prepare these studies for airport managers as well as for the airport managers, planners, and decision makers who will use the studies. â¢ Airport staff and airport consultants who prepare models and reports on air passenger demand for airport managers may find new insights in the reportâs analysis of air passenger surveys and the variability in the propensity to travel by air that is revealed across different demographic and socioeconomic subgroups. These new insights may influence the formulaÂ tion of new models for air passenger demand or may contribute to the interpretation of results from more familiar modeling specification.
118 Using Disaggregated Socioeconomic Data in Air Passenger Demand Studies â¢ Those engaged in air passenger modeling may also find the case study analysis and especially the investigation of new modeling specifications for air passenger demand at the BaltimoreÂ Washington airport system suggestive of new approaches that could be taken in their own assignments. â¢ Airport planners and managers (and other users of air passenger demand studies) may find the reportâs presentations of demographic and socioeconomic trends, air passenger choices and behaviors, and the shortcomings and opportunities that are identified for formal passenger activity modeling valuable to their own needs to interpret air passenger demand studies and act on them. â¢ Finally, air passenger demand models are also developed and used by other aviation indusÂ try participants, such as government agencies, aircraft manufacturers, airlines, and academic researchers. Although these parties may look to models at different levels of generality or complexity, they may also find that the cautions and innovations for air passenger modeling techniques developed for this report can also contribute to their own modeling efforts. The current research project represents an initial effort to both understand how the distriÂ bution of socioeconomic factors across the population affects the demand for air travel and to explore how to incorporate these effects in air passenger demand studies and models. It has generated a large amount of relevant information from a detailed analysis of air passenger, household travel, and other surveys, the full analysis of which will require much further work. It has also demonstrated that incorporating these effects in air travel demand models not only can improve the predictive ability of those models, but indeed is essential for them to have good predictive ability, given the underlying changes in the socioeconomic composition of society. However, much work remains to be done to develop robust models of air passenger demand that can be applied in a range of settings. This will require a sustained research effort over many years. As the literature review undertaken in the current project has demonstrated, the state of practice of air travel demand analysis has hardly evolved at all over the past 40 years. However, the current project has laid the foundations upon which a significantly improved understandÂ ing or air travel demand and an evolving state of practice can be built through continuing future research efforts.