Transport demand elasticities estimated by discrete choice models

1
Transport demand elasticities estimated by
discrete choice models

• Brett SmithPhD StudentUWA
• PATREC Research Forum
• 19 September 2006

2
Income
3
Mode choice elasticities

• it is important to distinguish between
  mode-choice and regular demand elasticities.
• Mode-choice elasticities express the change in
  demand given a fixed demand of traffic for all
  modes.
• They do not take into account the change in price
  on the aggregated volume of traffic.
• Mode choice elasticities are therefore a lower
  limit to regular demand elasticities.
• Nijkamp and Pepping 1998

4
Subject Review

• Quandt 1968 (bi-modal)
• Taplin 1982 (general case)
• Oum et al (1992)
• Acutt and Dodgson (1996)
• Nijkamp et al (1998, 2000, 2002)
• Wardman (1997, 2002)
• Grahem and Glaister (2004)
• Dargy and Hanly (2000, 2004)
• Govt reports -- TRACE

5
Literature Review Demand elasticities

• Reviews (fuel demand)
• Dargy and Hanly, Grahem and Glaister (2004)
• (fuel consumption) Long run -0.6 , short run -0.3
• Goodwin 1992
• (Fuel consumption) Long run -0.71, Short run
  -0.27
• Disaggregate data and selection models
• Archibald and Gillingham (1981) Demand system
• -025 to -0.37
• Kayser (2000) single equation
• -0.24
• Public Transport
• Nijkamp and Pepping -0.4 to -0.6 (country of
  study is a significant factor)
• Wardman -0.4 to -0.6 (after 1990)
• Wardman and Shires 2003, Demand elasticities are
  about 36 more elastic than choice elasticities

6
Outline

• Motivation
• Context for theory
• Theory
• Mode choice studies as conditional demand systems
• Conditional demand systems from micro-economics
• Contrast between DCM and CDS
• Application
• Sydney Discrete choice model
• Estimating income elasticities from HHES
• Applying theory
• Discussion

7
Response to pricing

• Long term
• Activity locations (travel patterns)
• Modal choices
• Change of vehicle
• Work choice
• Residential location
• Short term
• Mode switching
• Frequency of travel (add or remove activities)
• Trip chaining
• Activity locations
• Time of day

8
Rational for activity based models
9
Mode choice studies

• Policy directed
• Inclusion of a new service or mode
• Pricing policy
• Value of travel time savings
• Negotiations between Govt and private operators
• Behavioural outputs
• Welfare measurements
• Value of travel time savings
• Valuation of other modal attributes (willingness
  to pay)
• Elasticities
• Elasticities are used to determine the changes in
  usage for each mode
• Mode switching
• Other short term travel arrangements
• Summary elasticities from mode choice studies
  are used to examine demand responses to price
  movements in the transport system

10
Demand What? and How Much?
Level of demand
Mode choice elasticities provide a description of
What? These models implicitly assume that the
other travel decisions are left unchanged. In a
sense the demand models are conditional
11
Where are we now

• Motivation
• Context for theory
• Theory
• Mode choice studies as conditional demand systems
• Conditional demand systems from micro-economics
• Contrast between DCM and CDS
• Application
• Sydney Discrete choice model
• Estimating income elasticities from HHES
• Applying theory
• Discussion

12
Multi Stage Budgeting
Utility tree diagram
Weak separability
13
Utility Tree Development

• Leontief (1947) and Sono (1945)
• Cross price effects
• Strotz (1957)
• Two stage budgeting
• Gorman (1959) and Strotzs reply
• weak and strong (additive) separability
• Frisch (1959)
• Inferring price elasticities using want
  independence (block additivity)
• Pollak (1969, 1971)
• Conditional demand systems
• Substitution given a sub-budget
• Change in budget allocation

14
Where are we now

• Motivation
• Context for theory
• Theory
• Mode choice studies as conditional demand systems
• Conditional demand systems from micro-economics
• Contrast between DCM and CDS
• Application
• Sydney Discrete choice model
• Estimating income elasticities from HHES
• Applying theory
• Discussion

15
The difference between DCM and CDS

• DCM
• Derive utility from a trip (activity)
• If the price increases the utility derived from
  the trip (activity) does not change, but you have
  less money to spend elsewhere
• McFaddens social welfare function (extend to
  generalised costs)
• CDS
• Derive utility from consumption of a bundle of
  goods
• If a price increases then the consumer needs to
  manage with the same budget, all the loss in
  utility is experienced in the consumption
  activity (travel).
• The utility of the pre-allocated good does not
  change
• Butclearly the consumer need not behave
  according to either model they are free to
  balance the changes in activity (travel) patterns
  with the money left over for other things.

16
Demand implications DCM and CDS

• DCM
• Demand for transport is fixed
• Implied elasticity of transport demand with
  respect to price is zero
• Implied income elasticity is zero
• Marginal share (percentage of additional dollar
  allocated to transport) is zero
• CDS
• Demand for all other expenditure is fixed
• Implied elasticity of transport demand with
  respect to price is minus one
• Implied marginal share is one

17
Traditional conditional demand function

• conditional demand elasticity
• ordinary demand elasticity
• expenditure generation elasticity

18
DCM conditional demand function

• conditional demand elasticity
• ordinary demand elasticity
• demand generation elasticity

19
The second stage elasticity

• Frish price flexibility

20
Second stage elasticity in words

• Separability is about the structure we are to
  impose on our model what to investigate in
  detail and what can be sketched in with broad
  strokes without violation to the facts Gorman
• The broad strokes
• If income is a constraint on choice then the
  response to price will include an income effect
• The magnitude of the price response depends on
  the magnitude of the of the income elasticity
  (eTm)

21
Where are we now

• Motivation
• Context for theory
• Theory
• Mode choice studies as conditional demand systems
• Conditional demand systems from micro-economics
• Contrast between DCM and CDS
• Application
• Sydney Discrete choice model
• Estimating income elasticities from HHES
• Applying theory
• Discussion

22
The Empirical Setting

• Combined RP/SP survey for Independent Pricing and
  Regularity Tribunal (IPART) NSW to investigate
  elasticities (Hensher and Raimond 1996)
• Quota based sample to obtain reasonable numbers
  of public transport users
• On board sample for ferry users
• Collect data about current trip (RP)
• Present respondents with experimentally designed
  pricing scenarios (SP)

23
Pricing show card
24
Household income distribution
Low lt30k Moderate 30k 90k High gt 90k
25
Discrete choice model

• Segment the market in to low, middle and high
  income
• Include a cost squared term to detect a
  decreasing marginal utility of money (M.U.M.)
  Jara-diaz and Videla
• Lower income are more price responsive (cost)
• Within income class, those who choose higher cost
  alternatives have a lower M.U.M. (positive and
  significant cost squared)
• This effect is less apparent at higher income
  classes (cost squared)
• Heteroscedasity may confound these interpretations

26
Results
27
Results Discussion

• The marginal utility of income is lower for
  higher income groups (cost term)
• For each income class those who choose a higher
  cost alternative have a lower MUM (cost squared
  term).
• Testing for heteroscedasity allowing for
  increased variance of the unobserved utility did
  not account for all the variation between income
  classes (c23df 26, p-value lt0.001)
• Correlation between unobserved component of
  utility are fairly similar for each income group
• Non-linear in income deterministic utility
  functions improve fit (marginally) over linear
  specification
• The income effect is identified in models which
  treat the 5 responses of each individual as a
  single observation (mixed logit models).
• Non-linear in time resource improves fit
  remarkably

28
Elasticities
29
Where are we now

• Motivation
• Context for theory
• Theory
• Mode choice studies as conditional demand systems
• Conditional demand systems from micro-economics
• Contrast between DCM and CDS
• Application
• Sydney Discrete choice model
• Estimating income elasticities from HHES
• Applying theory
• Discussion

30
Income
31
(No Transcript)
32
Literature Review

• Reviews (fuel demand)
• Dargy and Hanly, Grahem and Glaister (2004)
• Long run 0.8 , short run 0.4
• Disaggregate data and selection models
• Archibald and Gillingham (1981)
• 0.3 to 0.6
• Kayser (2000) selection model
• Fuel demand 0.5
• Nolan (2002) tobit regression
• Fuel 0.5, bus fare (0.7 no car, 0.1 with car)
• HHES Transport sector
• Bergantino (1998)
• Lower incomes gt 1, higher income 0.9
• Haque (2005) Aus. HHES 1998-1999
• Lowest income 1.6, highest income 0.7
• Demand Systems Aggregate data
• Selvanathan (1991)
• 1.4 for Australia and 1.5 NSW

33
Transport out-of-pocket expenditure

• The model is a linear regression on a truncated
  dependent variable
• Travel expenditure gt 0
• P(Travel gt 0)
• Exp(Travel travel gt 0)
• It is clear that the variance is a function of
  income so a generalised approach is adopted
• Tobit regression with heteroscedacity (Greene
  2004)

34
Income Elasticity Engel Curves
35
(No Transcript)
36
(No Transcript)
37
Final model for income elasticity
38
Where are we now

• Motivation
• Context for theory
• Theory
• Mode choice studies as conditional demand systems
• Conditional demand systems from micro-economics
• Contrast between DCM and CDS
• Application
• Sydney Discrete choice model
• Estimating income elasticities from HHES
• Applying theory
• Discussion

39
(No Transcript)
40

• Discussion
• Ordinary demand elasticities are more elastic
  than choice elasticities
• An aggregate travel demand response represents
  the difference

41

• Discussion
• The generation effect represents the other travel
  changes made by individuals
• Mode choice studies do not collect this data
• The method presented here considers why
  individuals make these choices budgeting with
  non-travel expenditure

42

• Discussion
• The generation effect is greatest for car use
  because the change in fuel price affects most
  people
• Bus and rail generation elasticities are
  smallbut this does not mean people using those
  modes are not adjusting travel demand any less
• The method can be improved by taking into account
• Conditional income elasticities
• Vehicle ownership