An Analysis Of Larry Beck's Passenger Flows

Passenger Flows

1

COMPARATIVE PASSENGER FLOW
IN US AIRPORTS by Larry Beck
November 25, 2014
For Dr. Bruce Gockerman
Robert Morris University – Fall 2014
Prd 590 – Downtown Chicago – Tuesdays
[e-mail]
[phone] cell

Larry Beck

Passenger Flows

2

Larry Beck

Contents
Executive Summary......................................................................................................................... 4
Introduction .................................................................................................................................... 5
Objective ......................................................................................................................................... 8
Brief Literature Review
................................................................................................................... 9
(1)
(2)

Box & Jenkins.................................................................................................................... 9
Geoff Cumming ................................................................................................................ 9

Data Sources ................................................................................................................................. 10
Time Series Analysis ...................................................................................................................... 13
(1) Regression Betas ............................................................................................................ 13
Standardized Beta Analysis .................................................................................................. 15
Trend Assessment ................................................................................................................ 17
(2) Seasonal Factors ............................................................................................................. 17
SF Hypothesis........................................................................................................................ 17
SF Assessment ...................................................................................................................... 18
(3) ARIMA Models................................................................................................................ 19
Conclusion ..................................................................................................................................... 21
References .................................................................................................................................... 22
Appendices.................................................................................................................................... 23
Appendix 1: Airports In Study............................................................................................... 23
.............................................................................................................................................. 23
Appendix 2: SPSS Algorithm for Seasonal Decompostion ................................................... 24
Appendix 3: Descriptive Statistics ....................................................................................... 25
Appendix 4: Seasonal Factor Summary ............................................................................... 26
Appendix 5: “Expert” Model Parameters ............................................................................ 27
Appendix 6: Online Resources & Data................................................................................. 29

Passenger Flows

3

Larry Beck

Table of Figures
Figure 1: Net Profit Margins 2014.................................................................................................. 6
Figure 2: Breakeven Load Factor by Decade.................................................................................. 7
Figure 3: Passenger Counts O 'Hare Airport ................................................................................. 10
Figure 4: Example Charts of Passenger Flow ............................................................................... 11
Figure 5: Regression Trend Lines (unstandardized beta) ............................................................ 14
Figure 6: Example of Visual Significance Test .............................................................................. 16
Figure 7: Comparing Standardized Betas ..................................................................................... 16
Figure 8: Model Comparison Total USA ....................................................................................... 19
Figure 9: Map of Airports in Study ............................................................................................... 23

Passenger Flows

4

Larry Beck

COMPARATIVE PASSENGER FLOW IN US AIRPORTS
Executive Summary
“Competition” is the daily war fought by airline companies in the past three+ decades.
Since deregulation of the industry in 1978 there has been non-stop disruption to the old guard air carriers. In fact, of the four carriers operating under Federal regulations pre-1978 only one survives to this day (American Airlines). In the last ten years the industry’s net profit margin 1 has varied from -4.6% to +5% (in 1988 and 2014 respectively). Competition among carriers drives down the revenue stream and prevents any consistent growth pattern.
Competitive information is vital in such antagonistic environments. Each airline is looking for the patterns and habits within its source of revenue, passengers. The question then arises “is each airport unique in local passenger flows, or do they all follow a similar pattern?”
This is the context surrounding this report: Intense competition, external forces applying pressure, rising performance expectations, significant regulation, and a limited resource – passengers. Hence, it is vital for an airline to know the actual trend of passenger flows as they plan (long range) future operations, as well as finding strategic advantage in the marketplace.
Each airline is vying for the same pot of passengers, which is limited, so they must have a clear idea of how big that pot will be.
This research yields three clear facts for each airport:
1. Has unique passenger flow trends,
2. Exhibits distinct seasonality (if any), and
3. Must employ different models to forecast future passenger volumes.

1

Airline profit margins at: http://www.iata.org/pressroom/facts_figures/fact_sheets/Documents/industryfacts.pdf

Passenger Flows

5

Larry Beck

Introduction
The airline industry is recognized as a primary component of the US economy (587,676 employees)2, as reflected by inclusion in a separate economic index called the Dow Jones
Transportation Average (DJTA). As of September 2014 there are five passenger-mode airline companies (DJTA, 2012) 3 included in the twenty transportation companies followed by DJTA
(plus another three delivery companies which also utilize air transport). The DJTA was dominated by railroad companies from 1896 to January 1970, when six airlines were added to the DJTA index at the expense of six railroad companies.
The federal government regulated airlines (prices, routes, safety, mergers) from 1938 until 1978. In 1930 a “spoils of war” 4 conference divided the U.S. airline market into four segments. Four major airlines were awarded exclusive routes for US mail delivery service: TWA was the central region; American Airlines was the southern region; United Airlines was the northern region; and Eastern Airlines along the East coast. Forty years of regulation allowed the industry to grow and mature, with no competitive disruption. All of this changed in 1978 when the airline industry was deregulated. It is interesting to note that all four of the airline companies listed above filed for bankruptcy 5; with only American Airlines still operating today.
After deregulation newer and smaller airlines began encroaching upon the “turf” of the older and mature carriers. Price wars began as the new airlines took away routes via “low cost” tactics, as well as creating new routes for underserved markets. The older carriers took little notice of this competition at first, which was a big mistake in hindsight.
A prime example is seen with Southwest Airlines (SWA), incorporated in March 1967 in
Texas. Yet it took four years for SWA to begin operations due to lawsuits filed by three other area carriers6. Ultimately it required the Texas Supreme Court to rule in favor of SWA as a viable airline. Those other three airlines took their case to the US Supreme Court, who denied
2

Current airline employee data is at http://www.transtats.bts.gov/Employment/Employment.aspx
The current five DJTA airlines indexed are: Alaska Air Group, Delta Air Lines, JetBlue Airways, Southwest Airlines, and United Continental. Alaska Air Group replaced American Airlines due to their bankruptcy in 2011.
4 “Spoils of war” reference is from (Sullivan, 2014, p. 7).
5 Airline bankruptcy listing at http://en.wikipedia.org/wiki/Airline_bankruptcies_in_the_United_States
6 Southwest’s history is at http://www.swamedia.com/channels/By-Date/pages/1966-to-1971
3

Passenger Flows

6

Larry Beck

their appeal. Not until June 1971 did SWA begin service (due to further local lawsuits). Their initial routes were between 3 cities, with a one-way airfare of $20 (about $115 today). From this humble (and battered) position, today finds SWA is the second largest passenger carrier
(US Bureau of Transportation Statistics, 2014) in the United States (less than 1% behind number one Delta Air Lines). However, Southwest is the number one carrier of domestic US passengers7. Airlines today operate on very slim margins. The industry as a whole sees only 5% net profit margin as of first half 2014 (Airlines for America, 2014). For every dollar earned, they keep a nickel! It is interesting to see how this compares other businesses, see Figure 1 (page 6).

Figure 1: Net Profit Margins 2014

As seen in Figure 1, the airline margin of 5% resides in the lower end of company margins, where 9.3% is the national average 8. Note that in the first quarter 2014 they had only a 1.5% net profit margin. The competitive nature of air travel has taken a toll on the profits available, not to mention the external influences of fuel prices, taxes, and compliance. It seems almost criminal to note that the tobacco industry (not shown above) realizes an 18% profit margin. Another way to visualize the fiscal pressure on modern aviation is to see how passengerload-per-flight impacts the breakeven point with time, see Figure 2 (page 7). Competition is the driving force behind the increasing need for efficiency in passenger traffic. This has also forced
7

As of July 2014, airline passenger data at http://www.rita.dot.gov/bts/press_releases/bts047_14
Tobacco and National average profit margins from “Motley Fool” at http://www.fool.com/investing/general/2014/04/08/tobacco-is-still-an-extremely-profitable-busines-2.aspx 8

Passenger Flows

7

Larry Beck

airlines to be highly cost conscious in regards to fuel prices and aircraft design; newer aircraft are being ordered to reduce fuel costs and increase passenger counts. Since 2013 US airlines have spent $1 billion per month in capital expenditures (Airlines for America, 2014).

Figure 2: Breakeven Load Factor by Decade

In the current decade (2011-2013) the breakeven point occurs when about 80% of capacity is enplaned; any less means the airline will lose money. Notice how the non-regulated decade, 1971-1980, had a comfortable cushion of passenger counts near a 57% load factor.
With time new entrants arose and the ensuing competition disrupted the status quo forcing the whole industry to be aware of pricing. Net profits dropped which put the old guard carriers in jeopardy. The major airlines have enormous capitalization, high fixed labor costs, large marketing outlays, and operate from old business models. The newer carriers are cost effective, relatively low capitalization, and are highly flexible. This is the text-book example of how a mature industry changes; new ideas enter the mix, forcing adaptation or closure from the old guard.
This then is the context surrounding this report: Intense competition, external forces applying pressure, rising performance expectations, significant regulation, and a limited resource – customers. Hence, it is vital for an airline to know the actual trend of passenger flows as they plan (long range) future operations, as well as finding strategic advantage in the marketplace. Each airline is vying for the same pot of passengers, which is limited, so they must have a clear idea of how big that pot will be.

Passenger Flows

8

Larry Beck

Objective
This report will focus on two important factors in a time series; the trend and the seasonality, with the goal of creating a good forecasting model. The end-use is to determine how passenger counts will change in the future, allowing better planning for airline operations, staffing, resource allocation, and equipment acquisitions. The research question will focus solely on domestic passenger data, as recent international events (war, disease, terrorism) is impacting travel abroad. Modeling international passenger flow needs to wait for less turbulent times.
A national model for all domestic passengers (all carriers) will be constructed. This national model will then be compared to the top 15 airports (by passenger counts) to discern if they differ significantly from the national count. One use of this information is to be cognizant of local trends, if any, affecting the planning process.
The 15 airports in this study will be ARIMA modeled for trend and seasonality with a
“goodness of fit” determination. The trend lines will be constructed from a smoothed seasonally adjusted series, and checked for significant difference from the national trend. In addition, the calculated seasonality factors will be examined for any significant departure from the national seasonality factors.
The end result is to determine if passenger enplanements, at all airports in this study, follow the national model; or if each airport has a unique pattern of passenger flow. Three tests will be constructed: (1) Calculate the best trend line (seasonally adjusted series) for enplanements over time; do the trend beta coefficients differ significantly from the national model; (2) Determine if the Seasonality Factors at each airport are statistically the same as the national values; and (3) Compare the ARIMA models generated for each airport, examine them for similarity or difference.

Passenger Flows

9

Larry Beck

Brief Literature Review
(1) BOX & JENKINS
This classic textbook on Time Series Analysis (Box, Jenkins, & Reinsel, 2008), first published in 1970, was updated to Edition 4 in year 2008. This work is often cited within the statistics community, and a scan of Google Scholar shows it has been cited 33,585 times 9 by authors (as of November 19, 2014).
Within this text, the authors describe a case pertinent to this report (Box, Jenkins, &
Reinsel, 2008, p. 388), a look at TSA for airline passenger counts. Their data was international enplanements from 1949 to 1960 by month. Their analysis concluded that the data was best modeled by an ARIMA (0,1,1)x(0,1,1) process.
This report will compare the Box-Jenkins suggested ARIMA model, to models produced by SPSS “Expert Modeler.” Using goodness of fit descriptive measures a preferred model will be constructed for each airport.
(2) GEOFF CUMMING
An interesting tool for quick assessment of statistical hypothesis testing has been formulated by Geoff Cumming (Cumming, 2009). He postulates, and then confirms, that a visual chart of 95% confidence intervals (CI) of two populations is sufficient to determine statistical difference between them. The primary take-away is that when the two CIs overlap by up to half the length (or less) of one CI arm, then it signifies statistical difference at the p=.05 level or better. The research abstract follows:
When 95 per cent confidence intervals (CIs) on independent means do not overlap, the two-tailed p-value is less than 0.05 and there is a statistically significant difference between the means. However, p for non-overlapping 95 per cent CIs is actually considerably smaller than 0.05: If the two CIs just touch, p is about 0.01, and the intervals can overlap by as much as about half the length of one CI arm before p becomes as large as 0.05. Keeping in mind this rule—that overlap of half the length of one arm corresponds approximately to statistical significance at p = 0.05—can be helpful for a quick appreciation of figures that display CIs, especially if precise pvalues are not reported. The author investigated the robustness of this and similar rules, and found them sufficiently accurate when sample sizes are at least 10, and the two intervals do not differ in width by more than a factor of 2. The author reviewed previous discussions of CI overlap and extended the investigation to p-values other than 0.05 and 0.01.

This technique will be used within this paper to compare standardized regression betas for enplanements by airport.

9

Google search for Box & Jenkins: http://scholar.google.com/scholar?hl=en&q=time+series+analysis+box+and+jenkins&btnG=&as_sdt=1%2C14 Passenger Flows

10

Larry Beck

Data Sources
The data used shall be truncated to the post-recession time frame, 2009 and later, for domestic passengers only; constituting 66 months of observation for each airport. Although there is substantially more data available, it seems prudent to focus on the near-term data that is free from the instability of the Great Recession 10. This will provide a more accurate model specification reflecting the current state of economic factors. An example is seen in Figure 3 for
O’Hare airport in Chicago.

Figure 3: Passenger Counts O 'Hare Airport

We see from Figure 3 that in 2009 and later (green line) the passenger counts appear to be a stationary process, with the expected seasonality. Does this pattern hold for each airport?
Is this pattern significantly different from the national passenger counts? Strategic planning – such as routes, acquisitions, pricing, underserved markets – all rely upon knowing the likely passenger counts in specific locations, along with national trends. The answers are important for an airline company and will affect how they conduct existing operations and prepare for future opportunities.

10

The official group to decide US economic cycles is the National Bureau of Economic Research. They defined the
“Great Recession” as lasting from Dec 2007 through June 2009. http://www.nber.org/cycles/sept2010.html

Passenger Flows

11

Larry Beck

For this study, the top 15 airports were investigated (see Appendix 1), a little over half of the 25 “Category X” (largest) airports as defined by the Transportation Security
Administration (those which enplane at least 1% of total passengers). These airports represent
48.9% of all passenger traffic in the US (as of 2013), yet only comprise 3% of the 506 airports in the US. Three examples of domestic passenger counts are shown in Figure 4; the total US market, Atlanta market (largest), and the Seattle market (smallest in study). All 15 such charts are available at the Tableau site, follow this link: https:/public.tableausoftware.com/views/Top15Airports-LarryBeck/Graph_Passengers Figure 4: Example Charts of Passenger Flow

Note: X-axis time index, 1 = Jan 2009

A single data source was found, US Bureau of Transportation Statistics, which provides detailed information about all transportation modes in the US. Specifically, this study used the

Passenger Flows

12

Larry Beck

national data from (US Bureau of Transportation Statistics, 2014) “TranStats” system 11 which provides enplanements for each airport (or a specific carrier) by month. Values were extracted for 2009 through June 2014, sixty-six months of data. The data files derived from this source are available online:
•
•

Excel: https://drive.google.com/file/d/0BxuZ9YCOBeHQmNPdm45US1DQ1k/view?usp=sharing
SPSS: https://drive.google.com/file/d/0BxuZ9YCOBeHb2FfV3lKam1VdFU/view?usp=sharing

The primary data elements from TranStats are (by airport/carrier):
Year
Month
Domestic
International
Total

4-digit year
Month number: 1 - 12
Total passengers enplaned for domestic flights.
Total passengers enplaned for international flights (not used)
Sum of Domestic and International

Additional data clean-up included: adding the airport code, record numbers, case numbers, date, and a calculation of %_of_USA for each record. This value is simply the percent of enplanements for a given airport/month versus the national total. In addition, SPSS added seasonal dates for analysis along with seasonally adjusted values, as well as splitting the file into
“cases” – by airport code for separate inquiry. All of these items appear in the SPSS data file referenced in the link above.

11

Monthly passenger enplanements at http://www.transtats.bts.gov/Data_Elements.aspx?Data=1

Passenger Flows

13

Larry Beck

Time Series Analysis
Broadly, the research question asks if all airports exhibit the same trend and pattern for passenger enplanements. Or, should each airport be considered unique and examined separately? Put another way, can a single “national” model be used for planning passenger flows, or not. Three possible answers to this question are generated: Regression βs of smoothed seasonally adjusted series; comparison of Seasonal Factors; and components of best
ARIMA models. Each of these will be tested for statistical significance where each airport shall be compared to the national model.
(1) REGRESSION BETAS
The primary data of passenger flows was analyzed in SPSS. This data was transformed for seasonality and trend factors; including a seasonally adjusted series (SAS). The SAS was further processed with a 3x3 moving average (see Appendix 2), known as a “Smoothed SAS.”
The following comparison of regressed beta coefficients was generated:

The betas are keyed to a time index, where 1 = January 2009 and 66 = June 2014. Of course, a visual version of the above data is preferred. The following chart (Figure 5) shows the regression trend lines for all 15 airports (unstandardized B). We see a high value at LAX of 6681 and a low value of IAH at -1358, clearly all of these trend lines have different growth patterns.
However, we shall postpone a final judgment regarding significance until further analysis.
Two curious items, negative slopes, are seen in this data for MCO and IAH:

Passenger Flows

14

Larry Beck

(1) MCO (Orlando), the home of Disneyland, a premier tourist destination for domestic travel, appears to have no trend at all. In fact, the estimated slope value of -194 is not even significant (p=.28). Hence, we can say that the trend for Orlando is not different from zero. The passenger counts at Orlando maintain a seasonally adjusted pace of 1.27 million enplanements per month (year 2009 and beyond). Of course, there is seasonality, which is discussed later.
(2) IAH (Houston) has a significant trend line of negative -1358 (p<.0001). We cannot find a documented reason for this decline. One possibility is that IAH is a hub for Spirit Airlines, an ultra-low cost carrier, providing only domestic travel. The key element here is that Spirit
Airlines has been tagged as the eleventh worst airline in the world 12 (in 2013). It could simply be that the public goes elsewhere for domestic travel. This is pure conjecture.

Figure 5: Regression Trend Lines (unstandardized beta)

12

Article on Spirit Airlines’ rating as “11th worst” airline is at http://www.businessinsider.com/worst-airlines-tofly-economy-2013-5#11-spirit-10

Passenger Flows

15

Larry Beck

STANDARDIZED BETA ANALYSIS
To compare the different beta values found in the above regression, normalized the data and then again regress upon the newly transformed data. This process creates standardized data points (z score) for regression. The benefit lies in the removal of scaling effects, allowing a direct comparison of values. The following table shows the result of this process to provide standardized beta coefficients and confidence intervals. Note that “B” is the beta coefficient of the regression on standardized data. SPSS bootstrapping was used to generate the confidence intervals. As before, the entry for MCO is not significant (p=.264) and both MCO and IAH still exhibit a negative slope. These calculations provide a means to compare beta values of each airport versus the USA total beta.

Recall that the research question is to discern a statistically significant difference in the beta values, or
Ho: BetaUSA = BetaAirport
Here is an opportunity to use the Cumming technique (see Literature Review) as a visual assessment of the hypothesis. Briefly, this method compares overlap of confidence intervals; when the CI of one sample enters less than half-way into the tail of another CI, then we reject

Passenger Flows

16

Larry Beck

Ho and declare that the two samples are significantly different. The p-level for this is 0.05 or less. The following illustration (Figure 6) helps to illustrate the process.
The Cumming visual test of significant differences is illustrated here. Actual beta points are shown as red diamonds, the 95%
CIs are the green bars. We are comparing the “USA” beta coefficient versus two other airports, SFO and LAS. The red band centered on USA, “Accept Ho: No Sig
Difference,” is the critical point to accept/reject the hypothesis. It covers 50% of the CI around the USA point estimate. We see that SFO’s CI extends into the red band, hence, we accept Ho and assert that SFO is not different from USA. The other airport,
Figure 6: Example of Visual Significance Test

LAS, falls just a bit short of the red band, hence, we reject Ho and assert that LAS is significantly different from USA.
Having seen the Cumming method in detail, we now look at the entire set of all 15 airports, comparing them to the USA point estimate. Determining if the airport betas are not different from USA is very easy; if any part of an airport’s CI enters the red band, then there is no statistical difference in betas.

Figure 7: Comparing Standardized Betas

Passenger Flows

17

Larry Beck

TREND ASSESSMENT
Figure 7 allows the following assessment: Not all airports follow the same domestic passenger trend pattern as the USA total. In this study’s sample of 15 airports, 9 of them exhibited no significant difference from USA (p>.05). Of the remaining 6 airports, all were lower than the USA total with p<.05.
Result: Each airport requires an independent analysis of projected trend lines.

(2) SEASONAL FACTORS
We now look at the seasonal factors (SF) for each airport then compare them to the USA total domestic SF.
The SFs are the average value for each month’s ratio of (actual values / moving average). The MA is a 12-month centered average of the actual values, as follows:
MAt = [ Average(Data t-5 to t+6) + Average(Data t-4 to t+7) ] / 2
The SFs are then calculated as follows: SFt = Actualt / MAt – which yields a decimal percent of deviation from the moving average. These values are then averaged by month for a single number to represent the month.
Appendix 4 displays the result of these calculations for each airport by month. The actual data tables are available in the Excel file online 13.
SF HYPOTHESIS
Per the research objective, we now test for statistical significance (α=.05) on the difference between the USA total SF and each airport’s SF. The hypothesis is stated as:
Ho: SFusa = SFairport
A t-test between means was then conducted on all 15 airports over 12 months (180 comparisons). The significance level was calculated for each comparison. A visual

13

Excel data file, see tab {Seasonal Factors}: https://drive.google.com/file/d/0BxuZ9YCOBeH-UTdPcVpKalM4WDg/view?usp=sharing Passenger Flows
18

Larry Beck

representation is provided here for the month of January (all 12 charts are available on the
Tableau site 14):
The horizontal axis, Seasonal Factors, are calculated as described in this report. The value SF=1 is the mark where an actual data point is equal to the 12-month moving average. Clearly, there is less domestic travel in the month of January, except for Florida
(MCO and MIA). It appears that cold weather motivates people to visit a warmer state.

SF ASSESSMENT
The finding gives a split decision – 47% of the Airport-Months accept Ho, where 53% reject Ho. This violates the research question which requires that all airports match the USA total SF. (The calculations are available in the online Excel data file 15.)
Result: We conclude that individual airports exhibit different seasonality factors versus the USA total.

14

Open Tableau site for SF charts: https://public.tableausoftware.com/views/Top15Airports-LarryBeck/SF_Chart
Excel data file, see tab {SF T-Tests }: https://drive.google.com/file/d/0BxuZ9YCOBeH-UTdPcVpKalM4WDg/view?usp=sharing 15

Passenger Flows

19

Larry Beck

(3) ARIMA MODELS
The final assessment of model fitting is to examine the output of two time series modes:
ARIMA and SPSS’s “Expert” Modeler. Using ARIMA modes offers many possible parameter combinations, too many to try them all. Hence, this report will use the ARIMA model suggested by Box & Jenkins in their classic book (Box, Jenkins, & Reinsel, 2008, p. 388); they prescribe an
ARIMA (0,1,1)x(0,1,1) for international passenger enplanements (using a monthly data set from
1949 to 1960).
For comparison to ARIMA, another model will be suggested by SPSS software, using the
“expert” modeler. That system performs hundreds of combinations using several major model types. All of these different models are compared for “goodness of fit;” the best estimate, by airport, is presented in this report.
A visual representation of the different models is shown in Figure 8; for USA total domestic enplanements from 2012 and forecast through June 2015. Similar charts for individual airports are presented online using Tableau 16.
The two models are shown versus actual enplanements (blue);
ARIMA is green, Expert is red. The seasonality is matched fairly by both models. However, the
Expert model appears to fit the Actual data more consistently. The ARIMA model shows lower values across all dates.
Hence, the “Expert” model is preferred.
Figure 8: Model Comparison Total USA
16

Open Tableau site for model comparisons: https://public.tableausoftware.com/views/Top15AirportsLarryBeck/Model_Comparison

Passenger Flows

20

Larry Beck

The best models for each airport are derived from the SPSS Expert modeler. The exact model type and parameters for each airport can be found in Appendix 5. Briefly, the following two forms of model were presented by SPSS 17:
Simply Seasonal
This model is appropriate for series with no trend and a seasonal effect that is constant over time. Its smoothing parameters are level and season. Simple seasonal exponential smoothing is most similar to an ARIMA model with zero orders of autoregression, one order of differencing, one order of seasonal differencing, and orders 1, p, and p+1 of moving average, where p is the number of periods in a seasonal interval (for monthly data, p = 12).
Winters’ Additive
This model is appropriate for series with a linear trend and a seasonal effect that does not depend on the level of the series. Its smoothing parameters are level, trend, and season. Winters ' additive exponential smoothing is most similar to an ARIMA model with zero orders of autoregression, one order of differencing, one order of seasonal differencing, andp+1orders of moving average, where p is the number of periods in a seasonal interval (for monthly data, p = 12).

17

See SPSS overview of forecasting: http://public.dhe.ibm.com/software/analytics/spss/documentation/statistics/22.0/en/client/Manuals/IBM_SPSS_ Forecasting.pdf

Passenger Flows

21

Larry Beck

Conclusion
The research question posed was “Do all airports in the US exhibit the same trend and seasonality for domestic passenger enplanements?” Three tests were constructed; Trend lines,
Seasonal Factors, and model comparisons. In each case the conclusions confirmed this answer:
Airports exhibits significantly different trends, seasonal factors, and “best fit” models.
All airports exhibit a unique process for domestic passenger enplanements, separate analytics must be derived based upon their specific data flows.

Passenger Flows

22

Larry Beck

References
Airlines for America. (2014, Aug 27). Annual Results U.S. Airlines. Retrieved Oct 26, 2014, from
Airlines.org: http://www.airlines.org/data/annual-results-u-s-airlines-2/
Box, G. E., Jenkins, G. M., & Reinsel, G. C. (2008). Time Series Analysis: Forecasting and Control
(4th ed.). Hoboken, New Jersey, USA: John Wiley & Sons, Inc.
Cumming, G. (2009). Inference by eye: Reading the overlap of independent confidence intervals. Statistics in Medicine, 28(2), 205-220. doi:10.1002/sim.3471
DJTA. (2012, Nov 1). Historical Components. Retrieved Oct 26, 2014, from DJaverages.com: https://www.djaverages.com/docsprivate/level2/Dow_Jones_Transportation_Average_Historical_Components_Report.pdf Sullivan, S. W. (2014). Institutional Setting and Carrier Viability in the Airline Industry: A
Continuing Review of the Post-Deregulation Experience. Knoxville: University of
Tennesee Honors Thesis Projects. Retrieved Oct 25, 2014, from http://trace.tennessee.edu/utk_chanhonoproj/1715 US Bureau of Transportation Statistics. (2014, April 30). T-100 Market All Carriers. Retrieved Oct
25, 2014, from Research and Innovative Technology Administration: http://www.transtats.bts.gov/Oneway.asp Wikipedia. (2014, Oct 24). Dow Jones Transportation Average. Retrieved Oct 24, 2014, from
WikiPedia.org: http://en.wikipedia.org/wiki/Dow_Jones_Transportation_Average

Passenger Flows

23

Larry Beck

Appendices
APPENDIX 1: AIRPORTS IN STUDY
This study includes the top 15 airports by enplanements in 2013. These represent 3% of the 506 airports in the US, and 48.9% of all passenger traffic.

Figure 8 shows the above data on a US map, with the size of the circle proportional to the number of enplanements at that airport in 2013 18.

Figure 9: Map of Airports in Study

18

Interactive map at: https://public.tableausoftware.com/profile/larry.beck#!/vizhome/Top15AirportsLarryBeck/Map

Passenger Flows

24

Larry Beck

APPENDIX 2: SPSS ALGORITHM FOR SEASONAL DECOMPOSTION
The SPSS software uses a 3x3 moving average 19 for a smoothed trend line. This moving average was used to construct the regression trend lines for passenger counts.

19

SPSS algorithms are at: http://public.dhe.ibm.com/software/analytics/spss/documentation/statistics/22.0/en/client/Manuals/IBM_SPSS_ Statistics_Algorithms.pdf

Passenger Flows

25

Larry Beck

APPENDIX 3: DESCRIPTIVE STATISTICS
Sixty-six monthly observations of domestic passenger enplanements by airport: January
2009 to June 2014. These data are not seasonally adjusted.

Passenger Flows

26

Larry Beck

APPENDIX 4: SEASONAL FACTOR SUMMARY
Seasonal Factors (SF) were calculated as Mean = (Actual Data / Moving Average) as per a multiplicative model. These values were not computed by SPSS due to its “medial average” technique which eliminates the low and high values prior to calculation; this was inappropriate for a series with only 4 or 5 values. The moving average used here is a 12-month centered average.

Passenger Flows

27

Larry Beck

APPENDIX 5: “EXPERT” MODEL PARAMETERS
SPSS “Expert” modeler evaluates many different models to arrive at a “best fit.” Here is the output for the best model of each airport.

Passenger Flows

28

Larry Beck

The specific algorithms20 for the two models shown above are:

20

SPSS algorithms are at: http://public.dhe.ibm.com/software/analytics/spss/documentation/statistics/22.0/en/client/Manuals/IBM_SPSS_ Statistics_Algorithms.pdf

Passenger Flows

29

Larry Beck

APPENDIX 6: ONLINE RESOURCES & DATA
Data Files
This report was constructed using Excel and SPSS. The actual data file for each application is available online as follows:
Excel: https://drive.google.com/file/d/0BxuZ9YCOBeHQmNPdm45US1DQ1k/view?usp=sharing
SPSS: https://drive.google.com/file/d/0BxuZ9YCOBeH-b2FfV3lKam1VdFU/view?usp=sharing
One technical note regarding the SPSS file. To properly run the analysis, the first step is to
“split” the file, using variable Airport. This will force all calculations to draw only airport specific data, and group the results for comparison. The SPSS screen needed is:

Tableau Public Graphics
This report compares 15 airports and is heavy in the use of graphics. To save paper (trees) we placed many of the figures onto a public website using Tableau. Each such use is indicated in the text. I repeat here the list of those charts appearing online:
Tableau: https://public.tableausoftware.com/views/Top15Airports-LarryBeck/Map
Tabs:
{Map}

Map of US airports studied in this report.

{Graph_Passengers} Raw data for domestic passenger enplanements by airport.
{SF_Chart}

Compares seasonal factors for each airport by month.

{Model_Comparison} For each airport shows three lines (2012 through June 2015):
Actual enplanements, ARIMA model, and “Expert” SPSS model.