Passenger leaks and the fate of small community air service.

Phillips, Owen R. ; Weatherford, Larry R. ; Mason, Charles F. 等

I. INTRODUCTION

Small community air service has undergone a significant transformation since deregulation in 1978. Free to establish prices and routes, most major air carriers have moved to a "hub and spoke" system, in which the carriers feed flights back and forth from relatively small airport spokes to larger hubs. (1) As this transport system has evolved, many smaller communities have experienced a loss of service or have had unstable service. Recently, the General Accounting Office (2003) (GAO) reported, "daily departures from 202 small communities had decreased by 19 percent between October 2000 and October 2001" (p. 12). The U.S. Department of Transportation (2002) reported similar declines for a two-year period beginning in September 2000. Of the 202 community airports tracked by the GAO, the number served by just one commuter airline increased from 83 to 95, and 30 airports were notified of discontinued service, of which 15 had just one carrier. Although these community airports account for less than 5% of all air passenger traffic in the United States, they represent nearly three-fourths of the number of U.S. airports.

For many travelers, driving to a hub airport has become an attractive substitute to flying from the local airport to the hub. The viability of driving depends on numerous factors related to the frequency of flights and fares from the local airport to the hub, the frequency of flights and fares from the hub, and the direct and indirect costs of driving. Passenger leaks are broadly defined as people who could fly to a hub or destination from a locality but instead choose to drive. (2) An obvious way for small community air service to grow is to reduce leaks. More frequent flights and/or larger aircraft flying to and from the local airport can work to further increase local demand for air service. Many communities view growth in local air service as essential to economic development. It is generally believed that businesses require good air connections as part of a region's infrastructure. (3) In light of this importance, little empirical evidence is available on the determinants of local passenger leaks.

This article develops an empirical case study on causes of passenger leaks using data on departing travel originating in the state of Wyoming. As we will discuss, this case has important similarities to a significant subset of local airports across the country. We believe our estimates are likely to be relevant for many other rural areas of the United States. The analysis makes use of Ticket Control Number (TCN) data provided by Sabre. The data describe all ticket coupons sold by Sabre through associated travel agents in Wyoming from September 2000 through August 2002. A record in the TCN data set contains information about a leg of a passenger's airline trip. An individual record describes the point of purchase, itinerary, fare amounts, purchase data, and travel class. We use these data and other airport-specific information to empirically explain the factors that cause leaks from local Wyoming airports. Results from this analysis show that fare differences, distance between the local airport and hub, and the amount of regional jet service are important determinants. Short-run elasticities are estimated that measure the sensitivity of local air travelers to these factors.

II. BACKGROUND INFORMATION

The GAO (2002, appendix III, pp. 47-53) presents departure information for 202 non-hub airports. It tabulates daily departures of turboprops and jets, for calendar year 2001, as well as population for the relevant catchment area (from the 2000 census). (4) These summary statistics tend to be bimodal. Larger communities have more service, but the majority of airports have fewer than 10 flights per day. Table 1 summarizes the information from the GAO report, splitting the sample by frequency of flights. We report a variety of statistics for each of the two groups (9 or fewer flights per day, 10 or more flights per day): mean, median, 10th percentile, and 25th percentile. The first section of Table 1 presents these statistics for the entire sample of the 202 local airports. Because the sample of airports includes some very large communities, we also report the results for the subset of airports with populations no larger than 200,000; these are presented in the middle panel. Last, we report the corresponding statistics for Wyoming airports in the lower panel of Table 1.

The data in Table 1 illustrate some clear patterns. Communities with fewer than 10 daily departures tend to be about half the size of those communities with 10 or more daily departures. The threshold population for service with 10 or more daily departures is about 200,000. The fraction of communities with populations over 200,000 is significantly larger for airports with 10 or more daily flights (33 of 88, as compared to 20 of 114). Both the 10th percentile and 25th percentile populations are about half for communities with fewer than 10 daily flights as compared to communities with 10 or more daily flights.

Smaller communities are served mostly by turboprop airplanes. Both mean and median percentage of prop flights is larger for those communities with fewer than 10 flights per day. Indeed, over 90% of these airports have at least 75% of all flights with turboprops, and 75% of these local airports are served by only turboprops. In contrast, local airports with 10 or more daily flights are more likely to have jet service. The median observation shows nearly 20% of the flights flown on jets, whereas 25% of these airports have at least 50% of their flights on jets. These conclusions are robust when restricting the sample to communities with populations no larger than 200,000. On balance, airports with relatively fewer daily flights tend to serve smaller population bases and to be mostly served by turboprop airplanes. These are the majority of local airports in the complete GAO data set, these local airports represent 56.4% of the observations. Local airports with nine or fewer daily flights represent a significant component of small community airports.

The lower part of Table 1 shows that the subset of Wyoming airports in the GAO data set are broadly similar to the cohort of local airports with fewer than 10 daily flights. Although the typical community is somewhat smaller in Wyoming than in the GAO data set, the tendency to use turboprops and the average number of fights per day is quite similar to the sample of community airports for the Western United States. The GAO (2002, p. 15, fig. 3) report reveals an important distinction for local airports between the Eastern United States versus the Western United States. The majority of local airports east of the Mississippi River appear to be within 100 miles of a hub airport. In contrast, most local airports in the West appear to be considerably farther from hub airports, therefore increasing the importance of local air service. (5) Local airports in Wyoming share this feature--only one airport (Cheyenne) is within 100 miles of a hub.

Before turning to a formal analysis of the TCN data set, we briefly discuss leakage patterns for Wyoming airports. Table 2 presents estimated leakage rates for each Wyoming airport for the first year of the data set (9/2000 through 8/2001), the second year (9/2001 through 8/2002), and the entire sample. A more detailed description of all data subsets is provided in section III. These ratios represent outbound leakage rates for the ten local airports. (6) Overall, the average leakage rate is 40.5%, though there is considerable variance across cities. Leakage rates are highest for those airports closest to hubs (Cheyenne, Cody, Sheridan, and Laramie). Moreover, average leakage rates increase from 37.5% in period 1 to 43.4% in period 2; after the terrorist attacks of September 11, 2001. The most straightforward reason for this increase is that travelers decided to drive to a hub to reduce overall flight time.

III. CAUSES OF LEAKS

The goal of this section is to generally explain the tendency of travelers to use hubs rather than local airports as their point of departure. The TCN data details individual legs of passenger trips. Though the data cover an individual's flight itinerary, many individual specific attributes, for example, location indicators such as address or ZIP code, were suppressed by Sabre due to confidentiality agreements. With key individual level details censored, we aggregated observations to city/ destination pairs. A city/destination pair represents a single Wyoming airport city and a particular end destination (e.g., Cheyenne to Chicago). A city/destination pair was included in the final data set if at least 25 passengers, covering the two-year period (9/2000 to 8/2002), flew from a particular Wyoming airport to the specific destination. This was done to provide ample degrees of freedom for estimates of the determinants of city/destination pair leakage rates. The resulting data set includes 21,374 passenger trips aggregated to 201 city/destination pairs. (7)

Within each city/destination pair, the data were divided into two different time periods as mentioned: trips during the 12-month period from 9/2000 to 8/2001 and during the 12-month period from 9/2001 to 8/2002. Having an adequate number of passenger trips in each city/destination pair hindered any further disaggregation of time periods. Therefore, the underlying data represent an individual choice, but the observed dependent variable in this "grouped" linear probability construct is an aggregated proportion. Suppose the leakage rate for passengers leaving Cheyenne going to Chicago is of interest. A passenger is considered a leak from the Cheyenne-Chicago city pair if the passenger bought the ticket from a Cheyenne travel agent but chose to fly to Chicago from one of the following cities: Denver, Colorado Springs, Salt Lake City, Billings, Rapid City, Idaho Falls, Boise, Bozeman, Scottsbluff, and Pocatello. (8) In all but a very few cases, leaks were to Denver or Salt Lake City.

Information on three other variables was extracted from the TCN data. The first is the average total fare for flights leaving a Wyoming airport and going to specific destinations corresponding to the city pairs. (9) A second observation is the average fare for a ticket leaving the nearest out-of-state hub and going to the same destination corresponding to the city pair. The third variable--average layover times--was also calculated for the passengers flying from Wyoming airports. This was done by averaging the amount of time all outbound passengers waited at a hub airport for a connecting flight when flying from a Wyoming airport to a specific destination. Average layover times fall as the frequency of flights from a local airport increases. This variable measures how well local flight schedules match with connecting flight schedules.

A two-step regression process is used to explain city-pair leakage rates. The first step investigates the influence of those city/ destination characteristics that differ between the two periods. The output from the first stage is then used to explain the relation between leftover, or residual, effects and other variables that do not change over time. The regression model analyzed in the first step of this approach is

(1) [Y.sub.it] = [mu] + [beta][X.sub.it] + [[lambda].sub.t] + [[eta].sub.it].

where [Y.sub.it] measures the leakage proportion for a city/destination pair i in period t, [mu] is the common constant term, [X.su.it] are explanatory variables that vary over both city/destination and period, [[lambda].sub.t] are the period effects, and [[eta].sub.it] is the leftover, or residual, term. It is hypothesized that leakage depends on wedge fare, hub fare, and layover times.

In the next step of the regression, the residual term ([[eta].sub.it]) is regarded as being determined by a fixed effect variable that is related only to the city/destination pair and not time. Along with the fixed effect there is an additional component ([v.sub.it]) that may vary with both city/ destination pair and time. In this second step of the regression model, the fixed effects are then related to time-invariant variables that are linked to the city/destination pair. Notationally, this part of our model can be described as

(2) [[eta].sub.it] : [gamma][Z.sub.i] + [v.sub.it],

where [[eta].sub.it] is the remaining residual or error term from equation (1), [Z.sub.i] are explanatory variables that vary across city/destination pairs but not from period to period, and [v.sub.it] is the error term for equation (2). The goal is to use the available data to obtain estimates of the coefficients [beta] and [gamma]].

Equations (1) and (2) divide explanatory variables into two types. More complete definitions and means of these variables are presented in Table 3. Regressors that vary with both city/destination pair and time, [X.sub.it], include "Wyo fare," "Hub fare," "Wedge fare," and "Layover." As explained in Table 3, wedge fare is the difference between the Wyoming ticket fare and hub fare; because of this identity only two of the first three variables in any particular regression can be used. It seems natural to base the decision on flying directly from a Wyoming airport or driving to the hub airport (at least in part) on the wedge fare. Therefore, hub fare and wedge fare are used in the regression analysis discussed below.

Those regressors that vary only across city/ destination pair, [Z.sub.i], include distance (in road miles) to the hub airport and the percent of seats available on propeller planes (as opposed to jets) per day. Distance to the hub is a measure of convenience to hub travel; leaks tend to be greater the closer the local airport is to a hub. With respect to jet versus propeller travel, surveys in Wyoming and other states have shown that commuters strongly prefer a regional jet to a propeller plane. (10) Jet service is an important quality factor, if not the most important, and produces quality function to the commuter.

Although airlines use several different advance purchase dates, a particularly large difference in airfares is typically seen between tickets purchased 14 or more days in advance as compared to tickets purchased less than 14 days in advance. For a ticket purchased less than 14 days in advance, the mean real price of a ticket in 2000 dollars leaving a Wyoming airport from September 1, 2000, through August 2002 was $643.57 with a standard deviation of $270.44. An equivalent ticket bought 14 or more days in advance had a mean real price of $394.74 with a standard deviation of $110.12. In light of this difference, we investigate the model described by equations (1) and (2) using three samples, (a) all tickets, (b) tickets purchased 14 or more days in advance, and (c) tickets purchased less than 14 days in advance.

Results from the first of our two-step procedure are shown in Table 4. The second column shows the regression results from equation (1) for the entire sample. In this regression, we see that the wedge fare exerts an important effect: a $100 decrease in the wedge fare lowers the leakage rate for a given city/destination pair by about 4.8 percentage points. Neither of the other time-varying parameters (e.g., layover, hub fare) is statistically significant therefore no inference is drawn from the coefficient's magnitude or sign. (11) Observe that the "period effect" is both positive and statistically important. This last observation means that there is an important increase in leakage rates in the second year. Although it is possible that a number of events might explain this difference, one stands out as the most memorable: the terrorist attacks of September 11, 2001. Though it is generally accepted that the demand for air travel was adversely impacted by these events, it is less obvious why individuals who did choose to fly would be relatively less likely to use commuter air service. As suggested already, if travelers generally desired to fly less after 9/11, then when they did fly, it is consistent that they reduce time spent in the air by driving to a hub. Whatever the reason, there is a statistically important effect on leakage between the two years of our sample: After controlling for other time-varying effects, the average leakage rate was approximately 4.6 percentage points higher in the year after 9/11 than in the year before 9/11.

Results for advance purchase passengers are shown in the third column. These results are similar to those for the entire sample; in particular, wedge fare is the only time-varying variable that exerts a statistically important influence on the rate at which passengers leak. Here the economic impact of wedge fare is even more pronounced than for the entire sample: a $100 decrease in this fare would reduce the leakage rate by about 6.2 percentage points. As with the regressions based on all observations, the period effect is both positive and statistically important for the subsample representing advance purchase passengers. For this group of passengers the average leakage rate was approximately 7.2 percentage points higher in the second period.

Results for nonadvance purchase travelers are presented in the fourth column. These results are different from the previous results in that none of the time-varying variables appear to exert a statistically important effect on leakage rates. (12) One reason for these findings is that nondvance purchase travelers do not put much weight on the factors included in equation (1) when they decide on the airport from which they choose to fly. The results from this subset also differ from the other regressions in that the period effect was not significant.

The second step in the analysis is to investigate determinants of city/destination fixedeffects ([[eta].sub.i]). To recover these important estimates, the estimated vector [??] from the first step was used to calculate

(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]

where the dependent variable is the average residual for each city/destination pair across the two periods from the fixed-effects estimation of equation (1), the [[bar.Y].sub.i] are the leakage rates for a city/destination pair averaged across time periods, and the [[bar.X].sub.i] are the independent variables (wedge fare, hub fare, layover time) for a city/destination pair averaged across time periods. To estimate the [gamma] coefficients, the time mean residuals are regressed on the time-invariant regressors [Z.sub.i] (percent of seats on propeller planes, distance to hub) and a single constant term c. The remaining term at; is treated as a composite error term. Because these errors probably would not have the same structure across all observations, an allowance was made for heteroskedasticity (where the variance can be different for different city/destination pairs). To this end, the standard correction for heteroskedasticity was employed using the method proposed by White (1980).

Table 5 contains estimation results from equation (2) using the same set of three samples employed in Table 4. Overall, parameter estimates suggest the percentage of seats available on commercial propeller planes flown from Wyoming airports exert a statistically important effect on leakage rates for all three samples. In particular, the results show that a reduction in the percentage of propeller plane seats would decrease leakage rates. Distance to the hub airport outside of Wyoming also exerts a significant effect on leakage rates for advance purchase passengers, though not for the entire sample or for nonadvance purchase passengers. Apparently, the time invariant variables included do a relatively poor job in explaining the fixed effects for passengers who purchase tickets less than 14 days in advance: Only the "percentage of seats available on propeller planes" coefficient in the last column is statistically important, though the [R.sup.2] value is quite small.

These differences between advance ticketed and nonadvance ticketed passengers are sensible. Advance purchase passengers tend to have more flexibility in arranging travel plans. They can decide on different dates of travel given the different prices of tickets. It appears that if they must fly on propeller planes to get to the hub airport, they are considerably more likely to drive. Distance to the hub also is a significant determinant as to whether potential passengers leak, and being closer to the hub airport makes them more willing to drive. Each increment of a hundred miles closer increases the leak rate by about 3 percentage points. Nonadvance purchase passengers tend not to be sensitive to price, perhaps because an employer typically pays for these tickets. These travelers also do not appear to base their travel decisions on how far away the hub airport is, but they do care about the type of plane.

Previous studies show that the short-run price elasticity of demand for general airline travel is -0.1, but rises to -2.4 in the long run. (13) From this model, several elasticities related to the leakage rate can be derived. (14) Table 6 shows the short-run elasticity measures for the statistically significant variables. In all cohorts, propeller planes have the largest impact on leakage rates. As shown for the "all tickets" purchased sample, the elasticity of propeller planes is roughly twice the estimate for the impact of wedge fare changes.

IV. IMPLICATIONS AND CONCLUDING REMARKS

Our findings fall into two categories: those related to time-varying explanatory variables, and those based on time-invariant variables. In the former group, our results show that leakage rates are positively related to wedge fares--the additional fare one must pay if one uses a connecting flight from the local airport to the hub airport. However, this effect is only statistically important for those passengers who purchase their tickets more than two weeks in advance. Because advance purchase passengers tend to have more flexibility in arranging their travel plans and travel in a group, one would expect the relevant price (here, the wedge fare) to exert more of an influence on their behavior. We also found a small but statistically important effect due to period. The 9/11 terrorist attacks had a measurable influence on travelers' decisions to drive or fly from their home community to the hub airport. People drove more.

With respect to the time-invariant explanatory variables, we found that driving distance to the hub airport exerted a negative influence on leakage rates, but again this effect is only statistically significant for advance purchase travel. Driving distance is a key ingredient in the opportunity cost of driving to the hub, as opposed to flying from the local airport. Driving costs and wedge fare seem most likely to matter for those who plan well in advance; travelers who purchase their tickets at the last minute are less likely to be influenced by local ticket prices or the opportunity costs of driving. The last-minute traveler cares about neither; in this case the opportunity cost of driving may be well above the wedge fare. A second time-invariant variable exerted a statistically important on leakage rates, namely, the proportion of propeller planes providing service from the local airport to the hub. This effect is important for both those passengers who buy tickets well in advance, as well as those who make last-minute purchases. In both cases, the effect is positive. Leaks are higher, the more prop planes are used. The estimated elasticities reported in Table 6 suggest that this effect is more important than the effect of high wedge fares.

We believe these results have obvious policy relevance. In the past few decades, a number of smaller communities have used various financial incentives in an attempt to induce increased demand, and thereby improve local air service. The GAO (2003) report identified 98 communities that tried to stem declining air service with public support. Most tried a temporary subsidy to attract additional fliers; none of the programs were successful in permanently boosting traffic. (15) Our results are consistent with the relative impotence of these policies, in that wedge fares do not exert a statistically important effect on one cohort of travelers, and they do not seem to exert an economically large effect on travelers altogether.

Estimates point to using policies that induce a switch from turboprop planes to small jet planes or regional jets. One can envision local airports encouraging an evolution of a carrier's fleets toward jets, for example, by enhancing runways to better accommodate smaller jet planes. Or even more aggressively, a community could partner with a carrier to lease and maintain small jets. Carriers balk at making the substitution because they are unable to estimate the amount of increased revenue with added seats and then weigh this against the increased incremental cost of operating a small jet versus propeller plane. For many connecting flights the variable costs of a small jet versus a turboprop plane are virtually identical (Phillips et al. 2003, pp. 98-99). The advent of more fuel-efficient smaller jets should mitigate cost concerns over switching to small jet fleets. To the extent that such switches are made, our work indicates that lower leakage rates and hence greater demand for local air service will result.

ABBREVIATIONS

GAO: General Accounting Office

TCN: Ticket Control Number

REFERENCES

Ballard, Laurel A. "The Role of Leakages and Reliability in Regional Airline Markets." Ph.D. dissertation, August 2005.

Borenstein, Severin. "Hubs and High Fares: Dominance and Market Power in the U.S. Airline Industry." RAND Journal of Economics, 20(3), 1989, 344-65.

Brock, James W. "Industry Update: Airlines." Review of Industrial Organization, 16(1), 2000, 39-51.

Brown, John H. "Airline Fleet Composition and Deregulation." Review of Industrial Organization, 8(4), 1993, 435-49.

Chow, Gregory C. "Tests of Equality between Sets of Coefficients in Two Linear Regression." Econometrica, 28(3), 1960, 591 605.

General Accounting Office. Commercial Aviation: Air Service Trends at Small Communities Since October 2000. Report #02-432, March 2002.

--. Commercial Aviation. Factors Affecting Efforts to Improve Air Service at Small Community Airports. Report #03-330, January 2003.

Gillen, David W., William Morrison, and Christopher Stewart. "Air Travel Demand Elasticities: Concepts, Issues and Measurements." Department of Finance, Canada, 2003.

Houthakker, Hendrick S., and Lester D. Taylor. Consumer Demand in the United States, 1929-1970. Cambridge, MA: Harvard University Press, 1970.

Kahn, Alfred E. "The Competitive Consequences of Hub Dominance: A Case Study." Review of Industrial Organization, 8(4), 1993, 381-406.

Phillips, Owen R., Larry R. Weatherford, and Kenyon N. Griffin. "Paths toward Improving Wyoming Passenger Airline Service." College of Business, University of Wyoming, Laramie, Wyoming, 2003.

U.S. Department of Transportation, Office of the Inspector General. Airline Industry Metrics: Trends on Demand and Capacity, Aviation System Performance, Airline Finances, and Service to Small Airports. Number CC-2003-001, Washington, DC, 2002.

U.S. House of Representatives, Subcommittee on Aviation. "Funding and Infrastructure Issues at Small and Medium-Sized Airports." February 22, 1999 and March 2, 1999 testimony transcripts. Available online at: http://commdocs.house.gov/committees/ trans/hpw106-11.000/hpw106-11_2.htm.

White, Halbert. "A Heteroscedasticity-Consistent Covariance Matrix Estimator and a Direct Test of Heteroscedasticity." Econometrica, 48(4), 1980, 817-38.

(1.) For studies describing the evaluation of the hub and spoke system, and industry developments, especially related to the major carriers, since deregulation see Borenstein (1989), Brock (2000), Brown (1993), and Kahn (1993).

(2.) Driving includes all road travel to a hub airport. Travelers who use bus service to a hub represent a leak. This study does not consider "internal" leaks, for example, a passenger who drives from Laramie, Wyoming, to Cheyenne, Wyoming, and then flies to Denver is not a leak.

(3.) See, for example, U.S. House of Representatives, Subcommittee on Aviation (1999).

(4.) A catchment area is that geographic region closest to a commercial airport. Depending on the context, there are small catchment areas for community airports and larger catchment areas for hub airports of different sizes. A household may be in multiple overlapping catchment areas. See GAO (2002, pp. 9-11) for more details.

(5.) Notable exceptions to this pattern are local airports near major urban areas in Texas, California, and Washington state.

(6.) This calculation and definition of leakage exclude travelers who book travel directly with the airline or book through a Web site, such as Orbitz, Expedia, or Travelocity. Point-of-origin information about these people is not contained in the Sabre database. Our definition of leakage likely underestimates total leakage, and hence is a conservative estimate of the actual market of potential passengers that is available to initiate flights out of Wyoming airports. Also, these travelers are likely to be more sensitive to the price and quality factors described in the model, so econometric estimates are conservative. For more detail on alternate leakage definitions, see Phillips et al. (2003, appendix 5.)

(7.) See Phillips et al. (2003, appendix 7) for a comprehensive description of the data. The Sabre TCN database covers about 30% of all commercial air travel in Wyoming and contains 467,972 observations for the two-year period.

(8.) Because not all passengers buy their ticket in the same city from which they fly, there may be some measurement error in the leakage rate. For example if a person in Fort Collins, Colorado, buys a ticket in Cheyenne but flies from Denver, this person would be counted as a leak. It is believed, though, that the extent of this measurement error is minimal.

(9.) The average Wyoming fare for each city pair in the second time period (post 9/11/01) was then adjusted for inflation to allow a real dollar comparison. Ticket prices from 2001 and 2002 were divided by the Consumer Price Index deflator to put the fare in 2000 dollars.

(10.) See survey results in Phillips et al. (2003). See generally discussions in GAO (2002).

(11.) It is interesting that the insignificant layover coefficient has a negative sign. Longer layover times could have the effect of reducing leaks. The negative sign may be caused by the unreliability of local departure and arrival times. Longer layover times allow passengers to make connections in spite of delays. Detailed data on actual arrival and departure times are available for two Wyoming airports and this issue is being studied further. See Ballard (2005).

(12.) To specifically test for a statistical structural change in the subsets of data, a Chow (1960) test was used. The hypothesis that the coefficient vectors are the same in the two sample subsets is analyzed using a Chow test. We reject this hypothesis at the 1% level of significance.

(13.) See Houthakker and Taylor (1970). This study may be relatively dated. Because of online booking services, elasticities could arguably be larger in both the short and long run. For a recent survey of air transportation elasticities see Gillen et al. (2003).

(14.) Elasticity = [??][bar.X]/[bar.Y] where [??] is the estimated coefficient on X, [bar.X] is the mean of the variable averaged across all city/destination pairs and across all time periods, and [bar.Y] is the mean of the leakage rate averaged across all city/destination pairs and across all time periods.

(15.) As ineffective as these policies seem, one version has been successful in a Wyoming community. For 15 years, the Jackson community has privately subsidized airline service. See Phillips et al. (2003) for more details. During this time, enplanements increased from 50,000 in 1985 to 200,000 in 2003. Though the sum of annual payments by the Jackson community have exceeded $16 million over this 15-year period, current subsidies have dwindled to about half of the average. The airlines and community leaders expect the subsidy to disappear altogether in the next two to three years. Each year community leaders meet with two or more airlines to work out connecting flights schedules, aircraft types, and fares to large hubs, mainly Denver and Salt Lake City. The subsidies depend on the package provided by the airlines for the year. During the year an airline board, composed of community leaders, stays in contact with the airlines to discuss changes in connections and fares. Adjustments are mutually agreed on to accommodate traffic and seasonal changes in business. Traffic has grown because good connections exist between large hubs and Jackson. Rather than vacationers flying to Denver or Salt Lake City and then renting a car at these hubs, travelers fly to Jackson and then rent a car or become part of a tour group in the immediate area. The Jackson Hole area is a vacation destination, and tourism has grown substantially. It has both winter skiing attractions and summer National Park attractions. Estimates are that winter enplanements alone generate $75 to $100 million in sales tax revenue for the area. Businesses have a clear incentive to support good air service for the long run. They voluntarily contribute to a subsidy fund each year. Few communities, however, have such a transparent connection between enplanements and benefits.

OWEN R. PHILLIPS, LARRY R. WEATHERFORD, CHARLES F. MASON, and MITCH KUNCE *

* The research assistance of Laurel Ballard, Dustin Carruthers, and Laura Yetter was indispensable. Peter Berdy and Shelly Reams offered important perspectives on small community air service and valuable comments on earlier drafts. Also, thanks to Shenandoah Nieuwsma, Wendy Hunsaker, Leilani Price, and Jane Scheurman for patient technical support. The Aeronautics Division, Wyoming Department of Transportation, and Sabre provided the data sets used in the analysis and modeling. The Governor's Office, State of Wyoming, provided research funding for this project. The findings and opinions expressed in this report are those of the authors and do not necessarily reflect the views of the University of Wyoming or the Wyoming Department of Transportation.

Phillips: Professor of Economics, University of Wyoming, Laramie, WY 82071. Phone 1-307-766-2195, Fax 1-307-766-5090, E-mail [email protected]

Weatherford: Associate Dean/Professor, University of Wyoming, Laramie, WY 82071. Phone 1-307-766-3639, Fax 1-307-766-4028, E-mail [email protected]

Mason: Professor of Economics, University of Wyoming, Laramie, WY 82071. Phone 1-307-5336, Fax 1-3075090, E-mail [email protected]

Kunce: Adjunct Professor of Economics, University of Wyoming, WY 82071. Phone 1-307-766-2543, Fax 1-307-766-5090, E-mail [email protected]

TABLE 1
Frequency of Flights and Proportion of Propeller
Planes at Small Community Airports

 Airports with
 9 or Fewer
 Flights per Day

Statistic Population % Turbopopp

Summary statistics, total GAO
sample of local airports
Mean 123,089 93.4
Median 68,465 100.0
10th percentile 26,560 75.0
25th percentile 39,650 100.0
Average number of flights per day 5.4
Observations 114
Summary statistics, GAO communities
with populations of 200,000 or less
Mean 73,188 93.5
Median 63,376 100.0
10th percentile 25,786 75.0
25th percentile 36,776 100.0
Average number of flights per day 5.4
Observations 94
Summary statistics, Wyoming airports
Mean 37,676 100.0
Median 35,804 100.0
10th percentile 25,786 100.0
25th percentile 26,560 100.0
Average number of flights per day 5.6
Observations 5

 Airports with
 10 or More
 Flights per Day

Statistic Population % Turbopop

Summary statistics, total GAO
sample of local airports
Mean 200,034 72.9
Median 155,996 81.3
10th percentile 61,586 31.3
25th percentile 95,260 50.0
Average number of flights per day 14.6
Observations 88
Summary statistics, GAO communities
with populations of 200,000 or less
Mean 110,568 77.8
Median 120,937 91.7
10th percentile 43,941 31.3
25th percentile 78,161 57.1
Average number of flights per day 13.5
Observations 55
Summary statistics, Wyoming airports
Mean 42,392 90.8
Median 42,392 90.8
10th percentile 18,251 90.0
25th percentile 18,251 90.0
Average number of flights per day 11.0
Observations 2

Source: GAO (2002; appendix 111, tables 7 and 8).

TABLE 2
Estimate of Outbound Leakage at Wyoming
Airports

Airport 9/2000 9/2001
Catchment to to
Area 8/2001 8/2002 Total

Casper 17.6% 21.9% 19.7%
Cheyenne 62.5% 60.7% 61.1%
Cody 59.4% 66.5% 63.0%
Gillette 38.8% 46.3% 42.6%
Jackson 37.4% 32.7% 35.1%
Laramie 58.6% 69.3% 64.0%
Riverton 39.0% 41.9% 40.4%
Rock Springs 14.6% 15.1% 14.9%
Sheridan 47.3% 64.8% 56.0%
Worland 32.5% 45.8% 39.2%
Total 37.5% 43.4% 40.5%

Source: Sabre (see section III for details).

TABLE 3
Variable Descriptions in Model

Variable Description

[Leakage.sub.i,t] Percent of passengers in Wyoming choosing
 to drive to a hub airport outside of
 Wyoming, i.e., 0.40 is 40%

Wyo [Fare.sub.i,t] Average real, round trip airfare for a
 city-destination airline ticket leaving
 from Wyoming airport in 2000 dollars using
 the CPI deflator

Hub [Fare.sub.i,t] Average real, round trip airfare for a
 city-destination airline ticket leaving
 from a hub city in 2000 dollars using the
 CPI deflator

Wedge [Fare.sub.i,t] Average real, round trip airfare
 difference between Wyo Fare and Hub Fare

[Layover.sub.i,t] Average layover time (in minutes) in a hub
 airport when flying from a Wyoming airport
 to a particular destination

[Distance.sub.i] Average driving distance (in miles) from a
 Wyoming airport to the hub airport

Propeller [Planes.sub.i] Percentage of seats available per day from
 a Wyoming airport on propeller planes,
 i.e., 0.81 is 81%

 Tickets Tickets
 Purchased Purchased
 14 or More Less Than
 Days in 14 Days in
Variable All Tickets Advance Advance

[Leakage.sub.i,t] 0.40 0.41 0.37
Wyo [Fare.sub.i,t] $470.23 $394.74 $643.57
Hub [Fare.sub.i,t] $316.28 $264.34 $396.01
Wedge [Fare.sub.i,t] $153.95 $130.40 $247.56
[Layover.sub.i,t] 177.19 175.00 176.71
[Distance.sub.i] 261.29 270.77 255.35
Propeller [Planes.sub.i] 0.81 0.80 0.82

Notes. The subscripts are: i = city pair, t = time period.

TABLE 4
Period-Variant Determinants of Leakage Fixed-Effects Estimates

 Tickets Purchased
 14 or More Days
 All Tickets in Advance
 Coefficient Coefficient
Explanatory Variable (t-statistic) (t-statistic)

Constant 0.2640 (2.952) 0.2116 (2.288)
Wedge fare 4.776E-04 (4.206) 6.219E-04 (4.733)
Hub fare 3.048E-04 (1.322) 3.148E-04 (1.106)
Layover 1.649E-04 (-0.619) 1.694E-04 (0.720)
Period effect 0.0457 (2.873) 0.0715 (4.159)
Summary statistics
No. of observations (a) 402 370
[R.sup.2] 0.85 0.86

 Tickets Purchased Less
 Than 14 Days in Advance
Explanatory Variable Coefficient (t-statistic)

Constant 0.4147 (5.143)
Wedge fare 9.266E-05 (1.290)
Hub fare -1.260E-04 (-0.855)
Layover -1.066E-04 (-0.400)
Period effect 0.0112 (0.537)
Summary statistics
No. of observations (a) 316
[R.sup.2] 0.80

(a) This is the number of valid city-destination pairs
(i.e., > 25 tickets) available for analysis. Thus when we
split the total sample into pre- and post-9/11 pools,
several of the city-destination pairs no longer have
sufficient number of tickets to be considered valid.

TABLE 5
Period-Invariant Determinants of Leakage

 All Tickets Tickets Purchased 14 or
 Coefficient More Days in Advance
Explanatory Variable (t-statistic) Coefficient (t-statistic)

Constant -0.1424 (-2.868) -0.0549 (-1.060)
Distance to hub 4.152E-05 (0.031) 3.075E-04 (-2.155)
Propeller planes 0.1749 (4.450) 0.1735 (4.227)
 (% of seats)
Summary statistics
No. of observations 201 185
[R.sup.2] 0.06 0.09

 Tickets Purchased Less
 than 14 Days in Advance
Explanatory Variable Coefficient (t-statistic)

Constant -0.0782 (-1.148)
Distance to hub -7.856E-05 (-0.441)
Propeller planes 0.1191 (2.259)
 (% of seats)
Summary statistics
No. of observations 158
[R.sup.2] 0.03

TABLE 6
Short-Run Elasticity Measures

 Variable Elasticity

All tickets Wedge fare 0.18
 Prop planes 0.35

Advance purchase Wedge fare 0.20
 Distance -0.20
 Prop planes 0.34

Nonadvance purchase Prop planes 0.26