NAS Performance and Passenger Delay
Michael Ball
NEXTOR
University of California, Berkeley & University of Maryland
Coauthors: Andy Churchill, Bargava Subramanian, Alex Tien
On-Time Performance
On-Time Performance for 35 OEP Airports (Delay < 15min)
90%
.
85%
Percentage
80%
75%
70%
65%
60%
1998
1999
2000
2001
2002
2003
YYYY
On-time percentage is decreasing.
Data Source: ASPM Analysis Database
2004
2005
2006
2007
Flight Delay Trend
100%
90%
80%
Percentage
70%
Avg Fl Del > 45 mins.
60%
50%
15< Avg Fl Del <=45
40%
0< Avg Fl Del <=15
30%
Avg Fl Del <= 0 min
20%
10%
0%
2000
2001
2002
2003
2004
2005
2006
2007(Jun)
Percentage of flights with early arrival and delay less than 15 min is decreasing.
Percentage of flights with long delay is increasing.
Data Source: BTS On-Time Performance Database
Flight Cancellation Trend
4.0%
911 effect
Flight Cnx Rate (%)
3.5%
3.0%
2.5%
2.0%
1.5%
1.0%
0.5%
0.0%
2000
2001
2002
2003
2004
2005
2006
2007toJun
YYYYMM
Cancellation rate decreased in 2006 but has jumped up in 2007.
Data Source: BTS On-Time Performance Database
Cnx Rate vs Ave Delay
8%
7%
.
Y2000
6%
Y2001
Flight Cnx Rate (%)
Y2002
5%
Y2003
Y2004
4%
Y2005
Y2006
3%
Y2007(to May)
2%
1%
0%
0
2
4
6
8
10
12
Avg. Flight Delay (Mins)
14
16
18
Delay Statistics and Passenger “Pain”
The most widely quoted performance statistic is on-time
performance. Yet, customer dissatisfaction is principally driven
by the occurrence of very large delays. These are most often
associated with the: disrupted passenger
long
delay
cancellation
missed
connection
late arrival
A disrupted passenger is a customer who must
use a flight other than the one on which the
customer was originally scheduled due to a
missed connection or flight cancellation.
– The average delay for a disrupted passenger has
been estimated to be 7 hours.
– Cancelled flights are not accounted for in delay
statistics nor is the true delay associated with
passengers who miss a connection.
Passenger Delay Model
Disrupted passenger
f1 canceled
Passenger delay =
flight delay
direct
trip:
Requires distribution of
flight delays conditioned
on delay > 15 min
f1 not
canceled
f1 delay
> thresh
f1 not
canceled
f2 canceled
f1 delay ≤
thresh
2-leg
trip
f1canceled
f2 not
canceled
Parameters:
% Direct: from coupon data.
Cancel Prob: flight cancel prob.
Disrupted Pax Delay: from MIT
simulations (7 hrs).
Prob Miss. Connect: complex
appx model.
Another View
Average passenger delay =
A1 (average flight delay)
+ A2 (average flight delay)(1 + e)
+ A3 (flight cancellation probability)
+ f (load factor)
[future improvement]
Flight Delay vs. Passenger Delay (I)
4.5
Avg. Flight Delay
Avg. Pax Delay
Ratio
40
Minutes
35
4
3.5
30
3
25
2.5
20
2
15
1.5
10
1
5
0.5
0
0
2000
2001
2002
2003
2004
2005
Avg. pax delay is almost three times of avg. flight delay.
2006
2007toMay
R atio
45
Flight Delay vs. Passenger Delay (II)
120min (Sept,2001)
80
70
.
Y2000
Y2001
60
Avg. Pax Delay (Mins)
Y2002
Y2003
50
Y2004
Y2005
40
Y2006
Y2007(to May)
30
20
10
0
0
2
4
6
8
10
12
Avg. Flight Delay (Mins)
14
16
18
Demand vs. Delay
(35 OEP Airports)
(35 OEP Airports)
1,400,000
80
1,200,000
70
60
50
Avg. Flight Delay (Mins.)
800,000
Avg. Pax Delay (Mins.)
40
600,000
30
400,000
20
200,000
10
200705
200701
200609
200605
200601
200509
200505
200501
200409
200405
200401
200309
200305
200301
200209
200205
200201
200109
200105
200101
200009
0
200005
0
YYYYMM
The fluctuation of pax delay is more significant than that of flight delay.
Minutes
Flight Demand (Ops.)
200001
Operations
1,000,000
Load Factor vs. Cancellation Rate
8.0%
7.0%
Flight Cnx Rate (%)
.
Y2004
6.0%
Y2005
5.0%
Y2006
4.0%
Y2007(to
May)
3.0%
Linear
Regression
2.0%
1.0%
0.0%
0.5
0.55
0.6
0.65
0.7
Load Factor
Cnx% = 0.0741 – 0.0856*LoadFactor
0.75
Generally, there is a negative
correlation between load
factor and cancellation rate:
airlines are reluctant to
cancel flights when there
are fewer options for
accommodating disrupted
passengers.
8%
0.7
7%
0.6
6%
Load Factor
Cnx Rate
12 per. Mov. Avg. (Cnx Rate)
12 per. Mov. Avg. (Load Factor)
0.5
0.4
5%
4%
0.3
3%
0.2
2%
0.1
1%
YYYYMM
Data Source: BTS On-Time Performance Database
200705
200701
200609
200605
200601
200509
It was well-publicized
0% that the airlines
cancelled fewer flights in 2006 (this is
good for passenger delay)
This trend seems to have been reversed
in 2007
200505
200501
200409
200405
200401
200309
200305
200301
200209
200205
200201
200109
200105
200101
200009
200005
0
Flight Cnx Rate (%)
0.8
200001
Load Factor
Trend of Load Factor vs.
Flight Cancellation Rate
Some Final Thoughts
High load factors greater delays when disruptions do
occur
– Future analysis will replace constant disruption delay with delay
function that depends on load factor and possibly other factors – most
likely will use George Mason models.
High load factor + high cancellation rates is a
particularly disturbing trend
– Question: are airlines thinking strategically about what an “ideal”
load factor should be??
Question: should on-time performance metric be replaced with
more passenger oriented metric??