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 20

21
Exhibit I-23. Possible explanatory variables.
Expected Impact
Type Variable Measure on Seat Offers
Total Cost Real (adjusted for inflation) annual ATA Composite Cost Index Negative
Jet Fuel Cost Real (adjusted for inflation) annual ATA Jet Fuel Cost Index Negative
ATA Jet Fuel Cost coefficient of variation (monthly variation
Macro Jet Fuel Cost Volatility ?
around annual mean)
ATA Oil Price coefficient of variation (monthly variation around
Oil Price Volatility ?
annual mean)
9-11 Shock Separate dummy variables for 2002 and 2003 Negative
Population in the Census metropolitan or micropolitan area
Population Positive
where airport is located
Per capita real income in the Census metropolitan or
Income Positive
micropolitan area where airport is located
Changes in Technology Average seat size at airport (larger aircraft have lower costs per
Positive
and Fleet Mix seat)
100 - Percent O-D passengers originating at airport calculated
Leisure Destination Index Positive
from DOT ticket sample
Airport-Specific Demand/Supply Balance Airport load factor calculated from FAA T-100 reports Positive
Domestic seat-departures at large or medium hubs within 50
Inter-Airport Competition Negative
miles of airport
Low Cost Carrier (LCC)
Percentage of seats flown by LCCs at airport Positive
Presence
HHI (sum of squared market shares) at airport calculated
Airline Concentration Negative
from OAG seats
Average O-D yield at airport from DOT ticket sample (high fares
Pricing Strategy ?
could reflect high service levels or weak competition)
For the large connecting hub group, two separate equations ing combinations of the variables (which would allow the
were estimated--one for local traffic and one for connecting effects of one variable to change depending on the magnitude
traffic. The observed seat levels at each connecting hub were of another), as well as other categorizations of the airports.
broken into local and connecting categories based on observed An analysis was undertaken to assess whether airports with
local passenger shares on flight segments from the Data Bank access to only a small number of major carrier hubs may be
1B (DB1B) ticket sample published each year by U.S. DOT. affected differently by fuel price spikes (e.g., down-gauging
A total of five panel equations were estimated--two for the vs. flight reductions). This effort did not result in any signif-
connecting hub group and one each for the remaining large/ icant findings, other than the revelation that even very small
medium hub group, the small hub group, and the non-hub airports typically have service to several hubs. For example,
group.5 For all but the non-hub group, a so-called "one-way among airports with an average of at least 100 daily seats over
fixed effects" model with airport-specific effects was estimated.6 the past 20 years, there are only nine that have an average of
In addition to directly testing the variables listed in Exhibit three or fewer hub connections over the same time period.
I-23, an effort was made to consider interaction terms involv- While overall service from hubs indeed has declined over time
since the 1990s for many smaller airports, many still have ser-
vice to multiple connecting locations.
5 From a technical standpoint, an important consideration is that within each
category there is much more seat variation between airports at any given time
than there is variation at a given airport over time. Thus it would not be prudent 3.2 Statistical Results
to expect that changes in the level of a given explanatory variable would have the
same impact on the level of seats at a small airport as at a larger one. Conse- The regression analysis for the 271 airports included in the
quently our regression models utilize log values of the dependent and indepen- database led to statistical models that explain between 86 and
dent (explanatory) variables, which is equivalent to modeling percentage changes
98 percent of the variation in seat offers over 20 years. Sum-
rather than raw differences. This ties in directly with the plan to apply percent-
age changes from the model predictions to the TAF or user-supplied baseline mary results for the five models are shown in Exhibit I-24.
forecasts. Among the potential macro variables, jet fuel cost (lagged by
6 An airport-specific fixed effects specification would have been preferred for the
one year) and the 9-11 dummy variables for 2002 and 2003 have
non-hub group as well, but given the focus in this study on fuel prices and
income effects, more reasonable results were obtained using simple ordinary statistically significant negative impacts on observed seat offers.
least squares in this case. The oil price/fuel cost volatility variables did not show to be

OCR for page 20

22
Exhibit I-24. Equation estimates for daily domestic seat departures.
Coefficients
(t-statistics)
Model: Connecting Hubs Connecting Hubs Other Large-
Small Hubs Non-Hubs
Local Traffic Connecting Traffic Medium Hubs
Explanatory Variable
0.75240 0.06815 0.66652 0.54409 0.74530
Daily Seat-Departures[-1]
(123.76***) (4.15***) (25.08***) (35.57***) (97.99***)
-0.09112 -0.09876 -0.09863 -0.08764 -0.06185
Real Jet Fuel Cost[-1]
(-8.58***) (-3.79***) (-9.75***) (-6.44***) (-4.52***)
0.34308 0.75304 0.39448 0.05269 0.13843
Real Per Capita Local Income[-1]
(7.38***) (8.06***) (10.56***) (1.05) (5.23***)
0.14261 0.18217
Average Seat Size[-1]
(2.82***) (4.45***)
-0.12085 -0.06322 -0.08060 -0.30011
HHI Index
(-5.03***) (-3.68***) (-3.66***) (-20.96***)
Seat-Departures at Lrg/Med Hubs -0.04466 -0.28266 -0.10383 -0.32717
within 50 miles (-1.36) (-3.87***) (-2.81***) (-3.31***)
-0.14957 -0.06643 -0.12500 -0.12362 -0.14252
9-11 Dummy for 2002
(-8.86***) (-1.62) (-8.32***) (-5.91***) (-5.95***)
-0.10640 -0.08462 -0.09004 -0.10150 -0.06242
9-11 Dummy for 2003
(-6.38***) (-2.09**) (-6.02***) (-4.93***) (-2.54**)
Adjusted R2 0.98206 0.94698 0.97599 0.93836 0.86654
Note: [-1] indicates one-year lag
***Significant at 99% level
**Significant at 95% level
The numbers in parentheses of Exhibit I-24 are "t-statistics," which relate directly to the degree of statistical significance indicated in the exhibit. In the
current context, a variable that is "statistically significant" means that the researchers are confident that the impact of the variable is not zero; the higher the
t-statistic (in absolute value), the more confident the researchers are that the effect is not zero. A t-statistic of around 1.65 in absolute value correlates to a
90 percent confidence level; a t-statistic of around 1.96 in absolute value correlates to 95 percent confidence. Note that in a few instances in Exhibit I-24,
the estimated significance level is less than 90 percent (indicated by no asterisk next to the t-statistic). These variables were nevertheless kept in the analysis
so that the equations are relatively parsimonious with each other.
It is important to understand that just because an explanatory variable is statistically significant does not necessarily mean that it is "important" in the sense
that a given change in the variable will lead to a large change in projected seat departures. The impact could be small, but from a statistical standpoint it is
"significantly" different from zero. A measure of the relative impact of an explanatory variable is given by its elasticity, which is briefly discussed in
Section 3.2.
significant in any of the model specifications tested and so are to a 0.091 percent decrease in the number of seat departures
not included in the equation estimates shown in Exhibit I-24. offered at a given airport (holding all else constant).
As for airport-specific effects, variables measuring local It is interesting to compare the results across the five differ-
income, average seat size, airport concentration [Herfindahl ent airport groupings. Not surprisingly, the trend component
Hirschman index (HHI)], and inter-airport competition (seat measured by the lagged value of daily seat-departures is much
departures at neighboring airports) all showed statistically sig- smaller for the connecting hubs' connecting traffic relative to
nificant impacts with the expected signs in most of the models.7 their local traffic; this is consistent with the notion that there
Given the functional form used, the coefficients can be inter- is significant random year-to-year variation in how traffic
preted as elasticities, meaning that a 1 percent change in the flows over carrier hubs.8 The impact of jet fuel costs and the
variable indicated would lead to a percentage change in airport 9-11 dummies are fairly consistent across airports, while local
seat departures equal to the coefficient value. For example, the income effects are smaller at the small hub and non-hub air-
model representing local traffic at connecting hubs projects ports. In addition, the effect of airline concentration (mea-
that a 1 percent increase in the real price of jet fuel would lead
8 But some of this apparent random variation may simply reflect data sampling
7As noted earlier, except for the non-hub model, the equations also include a variation from the DB1B data, which by its design does not accurately depict
separate constant term estimated for each airport (not shown in Exhibit I-24). through routings.