quantifying and decomposing the uncertainty in appraisal value of travel time savings
DESCRIPTION
Presentation by Dr Phill Wheat and Dr Richard Batley 06/06/2014. www.its.leeds.ac.uk/people/p.wheat www.its.leeds.ac.uk/people/r.batleyTRANSCRIPT
Institute for Transport StudiesFACULTY OF ENVIRONMENT
Quantifying and decomposing the uncertainty in
appraisal value of travel time savings
Phill Wheat, Senior Research Fellow
and Richard Batley
06/06/2014
Highlights
• Work to quantify uncertainty in appraisal Values of Travel
Time Savings (VTTS) (non-work)
• Important as Travel Time Savings are often major benefits in
transport projects
– Uncertainty in VTTS implies uncertainty in CBAs which could impact
on rankings of projects – at least under sensitivity scenarios
• Statistical exercise, initially to motivate a new VTTS study in
Great Britain
• However in doing the analysis, some wider policy
implications for the best use of scarce research funds have
emerged:
– Do moderate sized VTTS studies often as this minimises uncertainty
in appraisal VTTS
Background – Appraisal VTTS
• Current non-work values in Britain (and general approach
taken in other countries e.g. Switzerland and Netherlands)
are estimated as follows:
– In 1994, Stated Preference data used to form a model for VTTS
Separate models were estimated for Commuting and “Other” leisure
travel.
Base VTTS = [/c].
CInc
C
C
Inc
Inc
00
.. ,
Background – Appraisal VTTS
– An overall distance-weighted average was obtained by weighting the
combinations according to the distribution (for all mechanised modes)
in the NTS 1995-2000 data defined in income and distance bands.
– The base VTTS was then up rated by applying the income elasticity
from a separate meta analysis model – GDP elasticity of 0.8
V =
Incy Dd
dyd
Incy Dd
dydyd
DN
DNV
.
..
8.0
1994
GDP
GDPVVTTSAppraisal t
t
Thus Appraisal VTTS are much more than just the base VTTS – multiple
sources of uncertainty
Research approach
• Construct a confidence interval around the VTTS
estimates
– Interval estimation
– Gives a lower and upper bound estimate for the Appraisal
VTTS for a given statistical confidence level (typically
95%)
• Two stage process utilising both asymptotic
simulation (Krinsky and Robb, 1986) and the delta
method
– Quantifies uncertainty arising from the base VTTS and from the use
of an estimated GDP uprating factor
Results – Commuting All Modes
0.00
5.00
10.00
15.00
20.00
25.00
30.00
35.00
40.00
45.00
50.00
1980 1990 2000 2010 2020 2030 2040 2050 2060 2070 2080
Valu
e o
f T
ravel T
ime S
avin
g (
pence p
er
min
ute
)
Year
Central VOTT estimate (p/min) Lower 95% CI Bound (p/min) Upper 95% CI Bound (p/min)
0.00
5.00
10.00
15.00
20.00
25.00
30.00
35.00
40.00
45.00
50.00
1980 1990 2000 2010 2020 2030 2040 2050 2060 2070 2080
Valu
e o
f T
ravel T
ime S
avin
g (
pence p
er
min
ute
)
Year
Central VOTT estimate (p/min) Lower 95% CI Bound (p/min) Upper 95% CI Bound (p/min)
Results – Commuting All Modes
Base VoTT (1994)
has relatively little
uncertainty
associated with it
Given the functional form,
uncertainty becomes
much larger once GDP
moves away from the
base level
NOTE: the larger intervals for later years does not reflect uncertainty in GDP
forecasts, merely the effect of uncertainty in the GDP elasticity estimate
Improving the model
• Two questions:
– What would be the implication for uncertainty in Appraisal VTTS of a
new (base) VTTS study if that study was of a similar accuracy of the
previous study?
– What if such a study resulted in much greater precision (3 times more
precise base VTTS)
• Trade-off:
– More costly one-off study yielding greater accuracy
– Or Greater frequency of smaller scale studies
• Which of the above to go for in terms of spending finite research
funds?
Resampling in 2015
39%
narrower
in 2075
Resampling in 2015 – Improved
Precision
Only 3.4%
narrower
in 2075
Don’t ignore the GDP elasticity in
research…
0.00
10.00
20.00
30.00
40.00
50.00
60.00
1980 1990 2000 2010 2020 2030 2040 2050 2060 2070 2080
Valu
e o
f T
ravel T
ime (
pence p
er
min
ute
)
Year
Central VoTT Updated Income Elasticity Lower CI Updated Upper CI Updated
Central VOTT Existing Income Elasticity Lower CI Existing Upper CI Existing
20%
narrower
in 2075
Updated GDP elasticity of 0.9 (from 0.8) (Abrantes and Wardman, 2011) (SE
reduced circa 33%)
Summary
• Scheme time saving benefits often arise five or even ten
years after a project begins
• Thus the necessary extrapolation of the base year VTTS to
Appraisal values adds a large degree of uncertainty (over
and above the uncertainty in the original VTTS modelling)
• Resampling is important, but not to get more precise
estimates, more to minimise the extent of extrapolation to
form Appraisal VTTS
– However estimates of base VTTS need to be unbiased
• The uncertainty in the uprating process is important – here
the GDP elasticity
Policy Recommendations
• When faced with a constrained set of research funds:
– Do moderate size resampling exercises frequently
• As opposed to very large size resampling exercises less
frequently
– Continue to review and update the uprating parameters
• Improvements to the precision of these can yield large reductions
in uncertainty of Appraisal VTTS