XI Riunione Scientifica Annuale - ?Societ

1
XI Riunione Scientifica Annuale - ?Società
Italiana di Economia dei Trasporti e della
Logistica Trasporti, logistica e reti di
imprese competitività del sistema e ricadute sui
territori locali, Trieste, 15-18 giugno 2009

• Individual and triadic preferences in a choice
  experiment on housing location preference
  heterogeneity and relative power

Edoardo Marcucci, Università di Roma Tre Amanda
Blomberg Stathopoulos, Università di Trieste
2
Outline

• Study Context
• Research questions
• Related literature
• Methodology Data description
• Econometric results
• Conclusions Future research

3
Study context

• Standard welfare and demand theory is based on
  individual preferences, and modern theoretical
  analysis of household behaviour is based on the
  rejection of the notion that households may be
  regarded as unitary decision makers rather than
  groups of individuals (Becker).
• Quiggin. J., (1998) Individual and Household
  Willingness to Pay for Public Goods, American
  Journal of Agricultural Economics, Vol. 80, No.
  1, pp. 58-63

4
Research questions

• Given that household location choices are taken
  jointly we control for
• attribute-specific preference heterogeneity among
  three members
• (if relevant heterogeneity exists) who influences
  the family choices the most (at the attribute
  level)
• potential polarization in collective choices
• This leads us to estimate the potential bias
  compared to using the conventional (unitary)
  approach.

5
Related literature

• Since the 1980s, the shortcomings of a black
  box approach where the household is the basic
  unit of analysis have been exposed.
• Joint and Individual preferences fail to
  coincide in numerous empirical tests regarding
  risk avversion, financial allocation,
  environmental WTP, labour choices, consumption of
  durables (car, vacation, housing) and activity
  patterns.
• A growing body of research is dedicated to
• 1) finding the appropriate level of analysis to
  understand household behaviour,
• 2) explore data collection methods,
• 3) quantify power of influence and
• 4) consider preference and IPS heterogeneity
  between members of a decision making unit.

Arora Allenby 1999, Corfman 1991, Dalleart
1998, Bateman Munro 2003, Dosman Adamowicz
2006, Hensher et al 2008, Beharry-Borg et al
2009, Marcucci et al (in press).
6
Main contributions of current study

• Adopting a triadic approach as opposed to the
  universally used dyadic one (i.e. couple based
  analysis),
• Considering the child/adolescent as a decision
  maker in the household choice,
• Focusing on hypotheses testing rather than a
  definition of a GUF,
• Concentrating on attribute level influence
  patterns,
• Controlling for polarization in household choice
  of residential location.

7
Methodology

• We study household interaction via stated choice
  experiments (single vs. joint interviews), Katz
  (1997), Manski (2000).
• Household members were first asked to perform the
  choice experiments singularly and were
  stimulated to choose according to their personal
  preferences
• Subsequently, after grouping together three
  family members, encouraging them to discuss and,
  then, choose a collectively acceptable housing
  alternative.

8
Methodology (cont.d)

• Stated choice experiments
• Two stage
• Conjoint
• Design
• 4 attributes (31 42 51)
• Orthogonal
• Full profile
• Fractional factorial (240 sets 16 rept. ? 15
  blocks)
• 4 holdout questions (2 monotonicity / 2
  stability)
• Model specifications
• MNL, MMNL, Individual-specific MMNL

9
Attributes
10
Data description

• Sample 53 Italian families (53 adolescents, 53
  mothers, 53 fathers 53 joint interviews)

Variables Unit of measurement Sample value
Age µ years Mother (50), Father (54), Adolescent (22)
Family size µ (min-max) 3,6 (3-6)
Travel by car full sample 50
Travel time µ time in minutes Mother (19), Father (23), Adolescent (20)
Sex female 48
Income in income bracket 30.000 - 60.000 59
11

• Estimation Results

12
Econometric results MNL

• All var.s for each model have expected signs and
  are highly significant

13
Econometric results NL (cont.d)

• Test for scale differences among
  membertype-models,
• Scale corrected with nested logit trick

14
Econometric results MMNL (cont.d)

• Rent Noise non random variables
• SQ, Access, Air all random variables, normal
  dist significant variance
• Significant improvement compared to MNL
  specification

15
Econometric results daily WTP WTA (cont.d)

• Similarity in results between model
  specifications
• Coefficients have expected signs
• Extremely high WTP for accessibility for the son
  (walking mode?)

16
Test of representative member model (pooled vs.
segment)
LR pooled vs. segment 2 ? LL (pooled) - ?
LL(single) ?2df pooled - single
Pooled model ? ?single models
17
Individual heterogeneity? - MMNL Kernels
18
Kernel densities for ßs WTP Family

• 16

(Beta SQ) only 16 ? 0 85 gt0 This prevail for
all member types!
(Beta Air Acc) are all ? 0 100 lt 0




19
Kernel densities for ßs WTP Son



(Beta ACC) 50 ? 0 of which 100 lt0 WTP (ACC)
extremely high


(Beta Air) 39 ? 0 of which 98 lt0






20
Kernel densities for ßs WTP Mother
















21
Kernel densities for ßs WTP Father






?




?
?












22

• Individual vs. Group
• Polarization Analysis

23
Polarization Status Quo
24
Concentration Rent
25
Polarization Accessibility
26
Concentration Air Pollution
27
No Difference Noise
Unitary model would produce unbiased estimates
only for this attribute (!)
28
Polarization Concentration Overview
Rent Concentrated towards mother
Access Polarized towards son
SQ Polarized towards son
Air Concentrated towards mother
29
CONCLUSIONS

• At the individual level
• We have detected relevant attribute-specific
  heterogeneity among members thus casting doubt on
  the representative member hypothesis (e.g. air
  pollution is considered differently by all
  members).
• Comparing individual to household choices
• We have shown that different members have
  varying degree of influence in joint decisions
  for housing, (e.g. mother heavily influences for
  rent son dominates accessibility)
• we have discovered statistically significant
  polarization in collective choices (Status quo
  and accessibility)

30
FUTURE RESEARCH will focus on

• Capturing heterogeneity in its various forms
  through advanced model specifications, such as
  ML with heteroschedasticity in the variance of
  the parameters Error components creating
  correlations among utilities of different
  alternatives,
• The decision making process including different
  strategies for information processing (IPS) among
  members/groups,
• Comparing the relative explanatory power of
  continuous (MMNL) or discrete (LC) mixing
  functions to discover latent groups once choice
  invariant variables (eg. Socio-economics and IPS)
  are introduced in group based models,
• Explore cost-efficient and simplified
  data-collection methods to study group choices
  and test their robustness.

31
FINE

• Grazie per la vostra attenzione!
• Domande?

32
Research question (general)

• Is there empirical evidence to question the
  unitary decision model?
• If so, what can we do to avoid biased estimates?
• How can we model interaction within groups?
• Especially, how do we measure relative power
  among members.

33
Methodology (cont.d)

• Discrete choice models
• RUM framework
• Different model specification
• MNL
• MMNL
• Individual-specific MMNL
• Estimates produced
• Attribute coefficients and WTP
• Individual specific attribute coefficients and WTP

34
Test of representative member model (Mixed vs.
Multinominal)
LR of MMNL vs. MNL 2 ? LL (r) - ? LL(u)
?2df u - re
ML improves MNL for all members