-48-
A Dynamic Approach to Investigate
Household Car Ownership and Usage
Mark A. Bradley
Transport Studies Unit
Oxford University
ABSTRACT
Many features are necessary in a behavioural model of household car ownership
and usage patterns. A description is given of the features of conventional
equilibrium-based models, followed by a discussion of the most important
dynamic issues underlying travel choice. These issues include household
travel and activity budgets; state-dependent factors such as information
search, cognitive processes, habits, attitudes, and inertia; and the role of
the household lifecycle as a choice catalyst. Recent dynamic modelling ap-
proaches are described, followed by a description of a system dynamics model-
ling approach which incorporates the dynamic hypotheses discussed throughout.
Finally, a direction of research is laid out, in which the model can be used
to simulate household panel data as a basis for hypothesis testing.
INTRODUCTION
A detailed representation of the factors and mechanisms in household decisions
to own and use automobiles is necessary in order to understand ways in which
households react to transport policies or changing travel environments. for
example, one might wish to design policies to reduce fuel consumption or
increase the use of public transport in a particular area. In such cases, it
is essential to know which types of households will adjust their behaviour,
and why their reactions differ from those of other households.
The use of a car comprises several types of decisions, adjusted continuously
as a household attempts to coordinate its travel patterns so that each member
can participate in their desired activities within the options available.
Each member will try to keep the amount of time, cost, and stress undergone in
travel within a reasonable limit, but this limit may be influenced by the
lifestyle, motivations, and past experiences of the individual. Within these
limitations, once acceptable travel patterns are established, car use
decisions appear to be largely a matter of habit.
A change in the number of cars owned by a household, on the other hand, is
best thought of as a major lifestyle decision, where the purchase or sale of a
car enables the members of the household to participate in their current
activities more easily, or to participate in additional activities. This
decision then, in turn, influences the travel patterns of the household, as
each member may adjust their habits and expectations to the new situation and
possibilities. This decision feedback, which is the focus of this paper, can
be best represented with a dynamic, state-dependent modelling approach.
~49-
In the past, for several reasons, most analyses of car ownership and use have
used an equilibrium microeconomic framework to model isolated components of
the household decision process. Over time, these methods have evolved to
include more and more behavioural and structural detail, as described in the
following section. Over the same period, a great deal of research has evolved
in areas generally overlooked in the equilibrium framework: travel and activ-
ity budgets; habit and inertia effects; and the role of the household life-
cycle. Research in these areas has, for the most part, lacked a coherent
framework which can easily be adapted into general models of travel and car
ownership decisions. Accordingly, each of the three areas is discussed
separately below.
More recently, dynamic modelling approaches have been evolved in the transport
field to address issues such as car ownership and mobility. Many of these
studies have arisen from the increasing availability of panel data, which
allows one to test a wider range of hypotheses about the behaviour of indi-
viduals over time as a function of their past behaviour and changes in their
surroundings.
After briefly describing such approaches below, I lay out the proposed struc~
ture of an alternative type of model which incorporates the dynamic issues
raised throughout the paper. This model, which is based on the system
dynamics approach, is essentially a descriptive model. As such, it isa
departure from the usual type of model in transport research, which is based
on a statistical method whereby a set of hypotheses can be tested using
appropriate data. To address this difference, I describe in the final section
ways in which the proposed model could be used to simulate panel data, which
could then be compared to actual data sets and used to test hypotheses with
various statistical methods.
EQUILIBRIUM APPROACHES
Most conventional transport model systems are based on cross-sectional data
which include observations on the number of cars and licenses in particular
households and/or the number, destination, travel mode, and route used for
trips made by each member of the household during a specified period. These
observed choices are statistically related to household profile data, general
socio-economic data, land use data, and/or network travel conditions, with the
aim of producing madels that can be used to predict similar choices in other
households, and perhaps other periods or locations. Such models are usually
compensatory, assuming that individuals will attempt to maximise their
expected utility by trading off among the characteristics that make up each
choice alternative and will choose the one which has the optimum mix.
Generally, no account is given to previous choices made by each individual, as
data is only available for a single period.
One of the more representative and advanced examples of such systems is the
nested logit model currently used for planning in the Netherlands (Sobel
1980). Given the shortest route available using each possible travel mode to
each possible destination zone for a given trip purpose, the individual is
assumed to trade off the times and costs across all possibilities. Based on
the utility of each mode/destination pair, a logistic model is used to calcu-
late the probability that the person will choose that particular pair. Then,
the utilities across all possible pairs are weighted by the choice probabili-
-50-
ties to give a general measure of accessibility for the individual. This
measure is then used, along with various household and personal characteris—
tics, to predict the frequency with which the individual will travel for the
purpose in question.
It has often been noted that car availability within the household must be
represented adequately in such models if they are to give reasonable forecasts
in response to policies which affect car use (Bailey 1984). In the system
described above, the number of cars and licenses in the household are predic-
ted mainly as a function of income and household makeup, but are also
influenced by the differences in the measured "accessibility" that would be
present at various levels of car ownership, in the following manner.
An additional car in the household may increase the utility of the driver and
passenger modes to each destination for each member of the household. This,
in turn, increases the measure of "accessibility" for each member. These
changes in "accessibility" are then added across household members for
alternative car ownership levels, and input to the car ownership model. Thus,
if the household is in a situation where an additional car will allow some or
all of its members to reach attractive destinations or avoid greatly inferior
alternative modes, the household will be treated as more likely to make this
addition. The predicted car ownership is then used in the lower level choice
models. In particular, the car driver mode is not possible if the household
has no car or the person has no license. If this is not the case, competition
within the household is represented by increasing the utility of driver versus
other modes as a function of the number of cars per licensed driver.
In the preceding example, the feedback between car ownership, availability,
and use has been incorporated into a quite rigid statistical framework. As a
consequence, the actual feedback and adjustment mechanisms cannot be treated
in much detail. So, while one might infer from such models which population
segments seem to be most sensitive to certain choice factors, one cannot infer
why this sensitivity might be present. In addition, since the models are
based on data from a single time period, one cannot be certain if these
sensitivities are completely due to mechanisms of individual behaviour, or if
they arise partly out of correlations which happened to exist at the time of
the survey. Concerns such as these have led to further investigations into
the choice processes underlying travel behaviour, several of which are
discussed in the following sections.
TRAVEL AND ACTIVITY BUDGETS
In the equilibrium economics of consumer behaviour, people are assumed to
operate under given budget constraints, and to choose the mix of goods and
services which will maximise their utility across the possible consequences of
their set of choices (Deaton and Meullbauer 1980). Travel behaviour has often
been seen in this light, with people operating under rigid time and cost
budgets. Supporting evidence is given by the fact that people in several
parts of the world are observed to spend quite similar amounts of time travel-
ling each day. There remains, however, a question as to whether these aggre-
gate patterns typify individual behaviour, and thus should be viewed as fixed
constraints on choice (Gunn 1981).
-5l-
In alternative economic paradigms, budgeting decisions are often seen in a
different light. In one example (Earl 1984), a person is seen to budget as
much time or money to a set of related activities as is required to meet their
goals and aspirations while avoiding extreme risk. Thus, if the "cost" of an
activity changes, people may not always change their consumption of that
activity, but may remain satisfied with their present consumption and instead
budget more or less to meet lower priority goals.
This type of behaviour can be seen in models which relate travel directly to
the activities which it enables. Activity models may be based on direct
observation of household trip patterns (Recker, et.al. 1983), or on house-
holds' stated travel patterns in hypothetical situations (Clarke 1984). In
such models, the ability to reliably carry out the desired household activity
patterns serves, effectively, as the constraint. If the time or cost of
travel changes, but the desired activity pattern can still be carried out
without infringing on higher priority activities (i.e. spending time with the
rest of the household), then there will be little motivation to search for
other alternatives. On the other hand, if the "cost" increases enough to
inhibit the desired activities, or decreases enough to allow additional acti-
vities, it may become worthwhile to search out alternative activity patterns.
The purchase of a car is a qualitatively different decision than those
involved in daily travel, but is subject to similar budgeting arguments.
Mogridge (1983) found that the annual fixed costs of owning a car were
inversely related to the amount people spent on operating the car, indicating
a fairly stable total budget. A car, however, can be purchased to meet goals
other than mobility, such as social status, investment value, or (to some) the
sheer pleasure of driving. If mobility is of higher priority than these other
aspirations, then people may buy less expensive cars or replace their current
vehicles less often to save on fixed costs in times when operating costs are
high, without compromising mobility. In the remaining discussion, cars are
treated only in terms of their mobility value.
Given that car purchase and daily travel expectations involve such different
budgeting decisions, but both, at least in part, with the goal of giving
adequate mobility, their interrelationships do not fit easily into most
analytic frameworks. For an additional car to seem necessary, households must
perceive a benefit in terms of the additional activities they could undertake
if the car were available. Attempts have been made to incorporate this
tradeoff explicitly into equilibrium-based models (Burns, et.al. 1976).
Clearly, budget constraints are important in car ownership decisions, as some
households cannot purchase a car, even when the need is perceived clearly. In
other cases, however, there may be additional factors which affect peoples'
perceptions of the desirability of changing their car ownership or travel
patterns. These are discussed below.
HABIT AND INERTIA EFFECTS
It has been found that the most accurate way to predict a person's choice of
travel mode in the future is often to observe which mode they are using today.
When a person persists in making a certain choice, even after it becomes
clearly inferior from an economic standpoint, the economist is left to look
for explanations. Indeed, much attention in equilibrium economics has been
-52-
given to ways of incorporating uncertainty, learning, and inertial and
habitual behaviour into classical choice models (Deaton and Meullbauer 1980).
In transport modelling, there has been a great deal of effort to find evidence
of these effects, and to hypothesise their influence on travel patterns.
One of the most important issues is one of imperfect information, or bounded
rationality. People may not recognise that they are making economically
irrational choices if they do not know about the other alternatives. Since
searching for such information is a cost in itself, and can force one to delve
into an unfamiliar world, dissatisfaction with the current choice may have to
become very high to warrant a search for alternatives. For example, people
might endure long peak hour traffic jams before they will take it upon them-
selves to learn the intricacies and risk the experience of a quicker public
transport System: They may not even be aware that the alternative mode is
quicker.
Compounding this effect are psychological processes such as cognitive
dissonance (Golob, et.al. 1979). This is the self-reassurance that one is
making the right decision by subconsciously blocking out any inferiorities in
one's choices. For example, if the drivers above had heard that the public
transport system was quicker, they might "automatically" underestimate the
amount of time they would save by using it.
A second issue involves-the decision rule people use in certain contexts. For
very regular decisions, such as daily travel, people may conform most to the
"cybernetic" decision maker (Steinbrunner 1974), who evolves a decision
pattern, monitors the outcomes of that pattern, and does not adjust the
pattern unless the outcomes become noticably out of line with expectations.
The longer such a pattern is continued satisfactorily, the less likely one is
to learn about new alternative possibilities. If, for instance, the public
transport system was improved, our drivers above may not be quick to learn
about it unless a much further decline in traffic conditions causes them to
look for alternatives.
Ways of incorporating these issues in conventional utility models have been
hypothesised. Goodwin (1977) showed the implications of using a "habit" term
for the disutility of shifting from current behaviour. Thus, an identical
shift in conditions may show differing effects across individuals depending on
their past conditions and behaviour. Incorporation of the information
diffusion dynamics mentioned above (Lerman and Manski 1982) would magnify
this state-dependence further, making the habit effect even stronger when
changes only affect the utilities of alternatives not currently used. In the
same line, a model based on non-compensatory behaviour rather than tradeoffs
would suggest even further state-dependence, as any shift in behaviour would
depend somewhat on each household's order of priority for travel
characteristics, which could shift in turn as new travel patterns evolve.
Car ownership decisions may be influenced by habitual travel behaviour, but
cannot be thought of as such in themselves. The perceived need for an
additional car could well build up as a household searches for alternatives
to bring their mobility to a satisfactory level. In such a case, the term
"inertia" is often applied. The decision to buy (or sell) a car may require
large changes in household budgets and priorities, and may require a search
for information about cars on the market, financing arrangements, and, for a
first car, about the availability of uncongested routes and parking. All else
-53-
equal, it appears that these types of inertia will be less for households
which already have car-owning experience. This implies a positive feedback
cycle for the level of car ownership within a household- one which is
constrained a great deal by external factors. :
An "inertia" disutility term in car ownership utility models has been hypothe-
sised (Goodwin and Mogridge 1981), with similar implications as for the
"habit" term described above. In many ways, such a treatment is analogous to
that of "transaction" costs in standard market models. A more detailed
treatment of the underlying household travel patterns and perceptions,
incorporating dynamic feedback relationships may go further in explaining
behaviour over time.
Although the discussion above has focussed on the time and cost of travel,
there may be many other factors which influence travel choice. The general
term "attitudes" is often used to describe peoples' perceptions and beliefs
about the more qualitative aspects of travel, such as comfort, convenience,
safety, social acceptability, etc. Such subjective attributes are yet more
succeptable to perception biases such as cognitive dissonance. With this in
mind, much research has been done into the direction of causality between
attitudes toward travel alternatives and the choice between those alterna-
* tives. Though results have varied, the relationship has often been found to
be one of mutual causality, or positive feedback (Tischer and Phillips 1979).
If such is the case, and if qualitative perceptions are of great importance in
travel behaviour, then one wonders what might cause changes in attitudes and
motivations to occur outside of this feedback loop. Major changes in
household circumstances may be one such cause, as described below.
THE ROLE OF THE HOUSEHOLD LIFECYCLE
The household lifecycle can be thought of as a time frame in which the dynamic
processes of travel behaviour unfold. The concept of "lifecycle" involves
categorising household states according to the number of adults and their age
and employment status, and to the number of very young and not-so-young
children. Many of the priorities, motivations, and activities of the
household will change as it moves from one category to another. Accordingly,
many recent dynamic and activity-based modelling approaches have taken
advantage of this concept as a way of grouping behaviourally similar
households, generally using about ten distinct lifecycle groups.
Much of the recent research at the Transport Studies Unit has involved in-
depth household interviews to probe for the reasons behind changing travel
patterns and car ownership (Goodwin, Dix, and Layzell 1985). The occurance of
"life shocks", or discontinuous changes in household characteristics, have
shown particular importance. Such events might be marriage, divorce, birth of
a child, a change in income or the number of workers, or a change in workplace
or residence location. Such an event often disrupts habitual patterns and
perceptions, and necessitates the search for information about new circum-
stances~ helping to overcome inertia. Household goals and priorities may also
be changed or clarified in response to the event.
; -54~
CIRCUMSTANCES ;
ENTRENCHMENT » DISSONANCE
FORMATION }
A EVENT OR SHOCK
BEHAVIOUR
TIME
Figure 1- REPRESENTATION OF PHASES OF HABIT
(from Goodwin, Dix, and Layzell 1985).
The various phases of habitual travel behaviour can be seen-in the lifecycle
framework, as shown in Figure'1. During the habit formation. stage, a
household or individual adjusts to circumstances by gaining experience with
certain alternatives within a limited range until a pattern which satisfies
the priorities of the members is set. In this phase, the representation of
the individual as trading off characteristics of similar options may be
accurate.
In the entrenchment phase, the pattern grows to be the norm. Here, the model
of the cybernetic decision maker is most applicable. If, during this phase,
external circumstances, and thus the "optimal" behaviour, begin to diverge
from the habitual patterns, there are several possibilities. If the change in
circumstances is great enough to cause immediate dissatisfaction, the
household may begin to search for alternatives. In the case depicted above,
however, there is a dissonance-type response, as the household attempts,
consciously or unconsciously, to rationalise the continued habitual behaviour.
Information delays, budget: constraints, or inertia might lead to this type of
response rather than to immediate action.
Finally, a major event, or life-shock, prompts the household to adjust
behaviour patterns toward current circumstances, for possible reasons listed
above. In support of this interpretation, changes in car ownership often
appear to accompany such events. On purely economic grounds, however, these
changes may be just as rational in the period before the shock as afterwards:
it is perceptions, priorities or constraints which have changed. After the
adjustment, the next likely phase is "habit formation", as the household
develops new travel patterns with new constraints and information.
=55~
RECENT DYNAMIC MODELLING APPROACHES
A model which could identify the relative and combined influences of each of
the dynamic processes described above from observed data would be quite
spectacular. Yet, the interest and possibilities in dynamic statistical
models of travel behaviour have increased with the availability of panel data-
multiple observations of the same households over time. Two very recent and
relevant approaches are discussed here.
One major study into car purchase and usage in Australia (Hensher and Wrigley
1985) is based on the nested logit modelling approach mentioned earlier.
Here, the models cascade not only through a decision hierarchy, but also
through time. This means that the characteristics and past choices of each
individual are taken account of, both in estimating and applying the models.
The models can be used to predict the number and type of cars owned, and the
total car travel distance by separate household lifecycle groups over time.
In this and other panel studies, a key factor is the periodicity of data
collection relative to those of the dynamic processes being modelled.
A panel to study more general and continuous changes in mobility is being
carried out in the Netherlands (Golob, et.al. 1985). This study uses
statistical techniques such as factor analysis to define measures of household
mobility, and discriminant analysis to group households according to these
measures. Then, using techniques such as linear structural relations models
or cross-lagged regression, changes in mobility can be related to past
behaviour, as well as to changes in household or external circumstances.
Mobility, in this case, is measured by the distance and frequency of travel
for various purposes using various modes, and the perceived costs to the
household of maintaining this level of travel.
Both of these studies represent advances in the treatment of dynamic behaviour
in a statistical framework. A useful complement is a model which does not
begin from the data and attempt to discern dynamic processes, but, conversely,
provides a flexible framework for simulating desired combinations of these
processes and studying the types of behaviour patterns which arise. Such a
model and its possible uses are described in the final sections.
A PROPOSED MODELLING APPROACH
In the past, two major transport studies have used the system dynamics
methodology for simulating interrelated feedback processes. One (Adler et.al.
1980a) studied the effects of fuel price on aggregate car purchase and travel
behaviour. The other (Adler et.al. 1980b), represented the aggregate
feedbacks between the supply and demand for public transport. Although such
aggregate models can be built on hypotheses of individual behaviour, none to
date have included the explicit representation of household decision processes
that is needed in this case.
The framework for this model is formed by simulating the evolution of a single
household over time. A number of exogenous variables are specified which
define the lifecycle stage of the household, and the general spatial
characteristics of residence, employment, and the most important non-work
destinations. These variable are listed in Table 1. Changes. in these
variables are then simulated to form a basis for car-related decisions.
-56-
Table 1
PARTIAL LIST OF HOUSEHOLD VARIABLES INCLUDED
EXOGENOUS ENDOGENOUS
Number of adults, by: Number of cars owned
- age category Number af licensed drivers
- employment/income group Weekly number of trips and related
Number of children, by: distance, time, and cost, by:
- age category - mode (car, public transport, etc)
Distance from workplace(s) - purpose (work, other)
Distance from non-work attractions Perceptions of the travel variables
Type and cost of housing above
Simulation of the exogenous variables is done using conditional probabilities
of the shifts from one state to another (Clarke and Dix 1981), which can be
specified from standard demographic data. Aging of the household members is
fairly straightforward for the likely simulation period of five to ten years.
There are two related differences between this and standard system dynamics
approaches. First, the model will be used to look at behaviour patterns of
isolated types of households. Therefore, at the beginning of the simulation,
a random assignment will be done based on the probabilities associated with
the exogenous factors. As stochastic transitions are used, "different"
households can be simulated each time by varying the random numbers used.
Alternatively, the evolution of a representative household can be used
repeatedly for comparative policy testing.
Because of this single household approach, the transitions are not smoothed by
aggregation. As a result, several "events" in the model, such as marriage,
birth, or changes in income, location, etc., are treated as instantaneous.
This treatment seems essential in view of the "life-shock" hypotheses above.
It may also be possible to incorporate feedback between car ownership and
the probability of certain household events. For example, a second income or
a move to the suburbs might be more likely for households with more cars.
Using the circumstances of the household and the objective attributes of the
travel network as an exogenous base, the dynamics of the endogenous variables
in Table 1 are represented. Figure 2 depicts the general hypotheses regarding
car ownership decisions in causal loop form. These relationships start from
the assumption that the car is a superior alternative for increasing mobility.
Competition from other modes is shown in Figure 3. In both diagrams, the term
"mobility" is used as a shorthand for the satisfaction achieved from activity
patterns. This satisfaction may be achieved at work and non-work
destinations, or from returning and being at home. Travel is the means of
undertaking these activities, and may detract from satisfaction as the time
and cost required increases.
~57-
Figure 2
1 aN
q
> Househeid
car ouners se (+) Ability to
increase/maintain
Perceived + ier ownership
potential -
to increas\ Nw
REBITEEY A
a Perceived” Changes in
actual ~«—----____-------- household
mobility circumstances
+ d
Perceived a a . Changes in
possible external
mobility 4 Possible 4 ~~~" circumstances
+ f+ mobility
The innermost loop of Figure 2 depicts a goal-seeking negative feedback. A
greater perceived potential to increase mobility tends to increase household
car ownership (the plus sign indicates that both quantities are likely to move
in the same direction, with the arrow giving the direction of causality).
Higher car ownership then increases actual and perceived mobility, which, in
turn, decreases the perceived potential to change mobility, all else equal.
There are several complicating factors represented. The delay between
perceiving a potential for greater mobility and actually doing something about
it is represented by the crossed arrow in Figure 2. Such a delay will
generally be due to inertia or budget constraints, which are both influenced
by household circumstances.
There are additional delays which may be important. If, over time, the goal-
seeking loop does not operate, due to the constraints mentioned, then the
perception of possible mobility may be pulled back into line with the
perception of the current state. This is an example of the eroding goal
structure, which can represent behavioural processes such as cognitive
dissonance (Richmond 1981). The delay indicated between possibilities to
increase mobility and the perception of those possibilities also represents
habitual behaviour and associated information delays.
The idea of the life-sheck is also represented. Changes in external factors
such as travel times and costs only affect the possible mobility of the
household, while changes in the household itself also affect the ability to
perceive and react to these possibilities. Whether or not car ownership
changes, the event will have an effect on actual mobility as the household
adjusts its habitual patterns.
Finally, there can be a positive feedback between the level of car ownership
and the ability to maintain and increase this level, all else equal. The
delay represents the build-up of experience in the household which will reduce
the inertia involved in car ownership decisions, as hypothesised earlier.
-58-
Figure 3
+... ree! a
Actual + - Actual
car _ 4 transit
use use
\ 1 Household 2 i
+ By (-) car (-) { +
Perceived iS ie Perceived
possible N + - possible
mobility 3 ' / mobility
using car (+) \ \ (+) ae en
Perceived } ih Perceived
+ potential / potential +
to increase to increase
Changes in mobility mobility shryes in
car travel using car using transit transit travel
characteristics characteristics
In Figure 2, the car is the only mobility option considered. Figure 3
represents the longer term dynamics of choice between the car and other
alternatives such as public transport. For simplicity, household
circumstances are assumed to remain fairly stable.
In Figure 3, loops 1 and 2 again show the shorter term dynamics of car
ownership adjusting to the potential for increased mobility. There is an
imbalance, however, in that the fixed cost nature of cars makes the adjustment
of car travel qualitatively different than travel by public transport. Even
within the car loop, a sale may be less attractive than a purchase in terms of
the recovery of fixed costs. The number of cars is also a factor, as the sale
of a first car involves higher risk of immobility than the sale of a second
car. These factors tend to make the loops stronger in the positive direction.
In the longer term, loops 3 and 4 show that decisions to increase mobility
either through car ownership or public transport use tend to reinforce future
decisions in the same direction. If conditions for one option improve, users
of that option are more likely to perceive possibilities for increasing
mobility than non-users. The feedback is less likely to take hold in the
other direction if the current option deteriorates, because of the habit and
inertia involved in changing behaviour. If changes are severe enough to
trigger such a shift, however, further shifts in that direction become likely.
Major household events may also trigger such shifts.
The reinforcing mechanisms just described bring out the importance of
decisions made in times of household transition. Within the dynamic framework
outlined above, assumptions made about choice rules, information diffusion,
perception biases and inertial effects are the vital parameters for such
decisions. The model is left flexible in this regard so that different
dynamic hypotheses can be easily input to monitor the influence on behaviour
patterns of various household types. More detail is provided in the final
section.
-59-
A RESEARCH AGENDA
With appropriate tests, the model described above can provide a new means of
testing dynamic hypotheses of car-related behaviour. First, the standard
system dynamics procedure of robustness and sensitivity testing must be
applied to ensure reasonable behaviour (Richardson and Pugh 1981). Then, the
sensitivity to the various dynamic issues above can be tested, separately and
in combination. In review, these issues include:
- Alternative choice rules, such as utility maximisation, priority-based
elimination, and cybernetic patterns; and their role in various phases of
evolving travel patterns;
- Information and perception delays and biases, and their variation with
respect to current and past travel patterns;
- Inertial effects and budgeting constraints, particularly with regard to
changing car ownership levels;
- The role of household lifecycle transitions in triggering behavioural
changes; and
~ Variations in simulated behaviour across different household configurations.
These sensitivity tests can be done by giving an adequate "start up" period
for behaviour to reach an equilibrium, and then introducing one or more
exogenous shocks or policies. In addition to household events, such inputs
could represent changes in travel times, costs, or distances; or possibly
economic factors such car prices or interest rates.
Given the range of behaviour considered, some external validation would
clearly be useful. One method is to use the model to simulate household panel
data which can then be compared to actual panel data sets. The way to carry
out the simulation and the statistical methods for comparison will require a
great deal of thought, but some of the most important issues are raised below.
The first issue is the grouping of households. Many panel data analyses are
based on the household lifecycle concept. In the model, the lifecycle
characteristics are mainly exogenous, and transition rates and state
probabilities would have to be parameterised to match those of the panel data
set.
A second issue is the initial conditions used. Actual surveys cannot be
assumed to start from equilibrium, so a different simulation approach must be
used. It might be possible to specify initial probabilities for the length of
experience with initial travel patterns, car ownership, etc. A more practical
approach is to specify exogenous inputs based on historical data for a period
five to ten years prior to the survey period. This then corresponds to the
start up period for the simulation, ‘with simulated data being recorded only
for the actual survey period.
To test the similarity between actual and simulated data, one is mainly
interested in changes in behavidur. By comparing behavioural changes among
groups. of households who undergo similar lifecycle transitions, one can gain
confidence in the sensitivity of the model to household circumstances. The
-60-
same can be done for external circumstances by simulating economic or policy
changes which actually took place during the survey period. The statistical
methods used could be fairly simple. For example, t-tests could be used to
compare the actual versus simulated percentages of various groups who make
similar changes in car ownership or travel distances over specific periods.
More involved methods such as cross-lagged regression could also be used.
With adequate validation, the model can be used as a tool for looking at the
success of various statistical modelling approaches in capturing dynamic
feedback processes in travel behaviour. In a method similar to that above,
panel data can be simulated using varying assumptions regarding the parameters
used in choice rules, perception biases, information on alternatives, survey
response errors, etc. Then, a range of estimation techniques can be applied
to the data to guage their relative success at reproducing the model
parameters. Andersen and others (1984) used a similar approach to evaluate
regression models applied in education policy. The range of techniques which
could be tested could include cross-section regression or maximum-likelihood
techniques to the more recent dynamic approaches described earlier.
Another application, which could also benefit model development, would be to
use the simulation framework for interactive gaming. The same framework of
exogenous household and external inputs would be used, but with the players
deciding when to buy or sell cars or shift their travel patterns and
activities. As simulated households expand, teams could split up to represent
different household members. Such an approach could give more insight into
the effects of household car competition on ownership patterns. Insight into
habitual behaviour and information search could be gained by only giving
minimal information on alternative options, and then recording which :
additional information people ask for and when. It would be difficult to
incorporate many inertial effects, however, as there is little cost or
difficulty in changing behaviour in imaginary situations. The way in which
household teams are evaluated could also be important, as it would have on
impact on peoples' choice rules.
In summary, several types of research could be based on a descriptive dynamic
model of household car travel. The benefits would be in ways of representing
and studying existing dynamic hypotheses, and in generating new ones. Further
effort could also provide tests or ideas of ways to take account of dynamic
processes in statistical travel demand models.
-61-
REFERENCES
Adler, T.J., J.W. Ison, and J.C. Geinzer. Interactions Between Energy Supply
and Transportation-Related Energy Use. Report to the U.S DOT. Wash., Jan.1980.
Adler, T.J., S.R. Stearns, and Y.J. Stephanedes. Techniques for Analyzing the
Performance of Rural Transit Systems. Report to the U.S.DOT. Wash., Oct.1980.
Andersen, D.F. and others. "Regression and Case Studies of Public Programs:
Discrepant Findings and a Suggested Bridge". Proceedings of the 1984
International System Dynamics Conference. Oslo, 1984, pp. 179-199.
Bailey, J. "The Meaning of Car Availability in Mode Choice Decisions".
Transportation Planning and Technology, 1984. Volume 9. pp. 125-134.
Burns, L.D., T.F. Golob, and G.C. Nicolaidis. "Theory of Urban Household
Automobile Ownership Decisions". Transportation Research Record 569. 1976.
pp. 56-73.
Clarke, M.I. and M.C. Dix. "Simulating Demographic and Behavioural Changes in
a Population with the Progression of Time". Oxford: Transport Studies Unit,
WP 142. 1981.
Clarke, M.I. "An Activity Scheduling Model of Travel Behaviour". Oxford:
Transport Studies Unit, WP 263. 1984.
Deaton, A. and J. Meullbauer. Economics and Consumer Behavior. Cambridge:
Cambridge University Press, 1980.
Earl, P.E. The Economic Imagination: Towards a Behavioural Analysis of Choice.
Brighton: Wheatsheaf Press, 1984.
Golob, J., L. Schreurs, and J, Smit. "The Design and Policy Applications of a
Panel for Studying Changes in Mobility over Time". in Behavioural Research for
Transport Policy. The Hague: Ministry of Transport and Public Works, 1985.
Golob, 1.F., A.D. Horowitz and M. Wachs. "Attitude-Behavior Relationships in
Travel ‘Demand Modelling". In Behavioural Travel Modelling. London: Croon-Helm,
1979.
Goodwin, P.B. "Habit and Hysteresis in Mode Choice". Urban Studies. Volume 14.
1977. pp. 95-98.
Goodwin, P.B. and M.J.H. Mogridge. "Hypotheses for a Fully Dynamic Model of
Car Ownership". International Journal of Transport Economics. Volume 8, No.3.
1981. pp. 313-326.
Goodwin, P.B., M.C. Dix, and A.D. Layzell. "Alternative Methods for Dynamic
Analysis of Travel Demand". Transportation Research (forthcoming). 1985.
Gunn, H.F. "Travel Budgets: A Review of Evidence and Modelling Implications".
Transportation Research. Volume 15A. 1981, pp. 7-24.
GP
Hensher, D. and N. Wrigley. "Panel Data and Econometric Models: An Overview
and Application in Modelling Household Automobile Possession and Usage". in
Behavioural Research for Transport Policy. The Hague: Ministry of Transport
and Public Works, 1985.
Lerman, S.R. and C.F. Manski. "A Model of the Effect of Information Diffusion
on Travel". Transportation Science. Volume 16 No. 2. 1982, pp. 171-191.
Mogridge, M.J.H. The Car Market. London: Pion Limited, 1983.
Recker, W.W., and others. An Empirical Analysis of Household Activity
Patterns. Final Report to U.S. DOT. Washington. , 1983.
Richardson, G.P. and A.L. Pugh. Introduction to System Dynamics Modelling with
DYNAMO. Cambridge, MA: MIT Press, 1981.
Richmond, B. "The Feedback Interplay Between the Physical and Metaphysical
Dimensions of.an Urban System: A Focus on Public Housing". In Essays in
Societal System Dynamics and Transportation, D. Kahn, ed. Final Report to the
U.S. DOT. Washington, March 1981.
Sobel, K.L. "Travel Demand Forecasting with the Nested Multinomial Logit
Model". 59th Annual Meeting of the Transportation Research Board. Wash., 1980.
Steinbrunner, J.D. The Cybernetic Theory of Decision. Princeton: Princeton
University Press, 1974.
Tischer, M.L. and R.V. Phillips. "The Relationship Between Transportation
Perceptions and Behavior Over Time". Transportation. Volume 8. 1979.pp.21-35.
