A review of the impact of parking policy measures on travel demand

l/TIrart

308A

1935

A Review of the Impact of

Parking Palicy Measures on

Travel Demand

Bernard P. Feeney I

(db

t Swedish Road and Traffic Research Institute 0 3-58 1 0 1 Linkc'ping Sweden

V rar;

308 A

1.986

A Review of the Impact of

Parking Polioy Measures on

Travel Demand

Bernard F. Feeney

Vag- 00/)

Statens va'g- och trafikinstitut (VT/l - 581 0 1 Linkb'ping

IllStltUtBt Swedish Road and Traffic Research Institute - 8-58 1 01 Lin/«Spiny Sweden

[\

WN

N

w.

The Personal Travel Demand Response Constraints on the Demand Response Empirical Sources

THE EVIDENCE OF MODAL CHOICE STUDIES Mode Availability

Multi-Trip Decision Making

Variable Specification and Modal Choice Elasticity Measurements

Empirical Results

THE EVIDENCE OF PARKING LOCATION STUDIES Aggregate Parking Location Studies

Disaggregate Parking Location Studies

THE EVIDENCE OF SITE SPECIFIC STUDIES Commercial Parking Charges

Charging for Worksite Parking

Interpretation of Elasticity Measurements CONCLUSIONS

REFERENCES

VTI RAPPORT 308A

II \ I U I U J

12

13

18

18

23

25

3o

30

31

34

34

A Review of the Impact of Parking Policy Measures on Travel Demand

by Bernard P. Feeney

Swedish Road and Traffic Research Institute(VTI)

5-581 01 Linkoping Sweden

ABSTRACT

This report reviews the empirical evidence relating to the impact of parking policy measures on the demand for parking and for travel. Disaggregate modal choice models, parking location models and site-specific studies of parking behaviour are examined. With regard to modal choice models, it is concluded that few studies deal adequately with parking factors, but that there is some support for the view that parking policy measures have only a moderate influence on modal choice. When parking location models are examined parking policy variables are shown to have a substantial impact on choice of parking location. With regard to site-specific studies, the report concludes that there is great variation in the parking price elasticities quoted, which reflects partly the method-ological problems associated with such studies.

II

A Review of the Impact of Parking Policy Measures on Travel Demand

by Bernard P Feeney

Swedish Road and Traffic Research Institute(VTI)

5-581 01 Linkoping Sweden

SUMMARY

This paper reviews the empirical evidence relating to the impact of parking policy measures on the demand for parking and for travel. The primary concern is with parking and the journey to work, an approach which is justified by the fact that the negative effects of urban travel are largely the result of the travel peaks created by the journey to work; additionally, the bulk of the empirical evidence on the effects of parking measures relates to the commuting journey.

The structure of the demand for personal travel is examined and the possible responses to parking measures are identified. Of these, changes in modal split and parking location are considered to be the most important. A central theme of the paper is that there is a range of constraints on commuter's travel decision, which limits his freedom to change his travel behaviour, and renders inadequate the analysis of the commuting travel decision in terms of the time and money costs of the journey to work alone. Among such constraints are the need to undertake business trips during the working day, for which a car is mandatory or optional; the requirement on the car driver to transport other individuals to other than his work location; and the desire to make after work shopping, social or recreational car trips. The evidence presented in the paper points to considerable constraints on individual choice arising from these considera

tions.

The impact of parking measures is examined through a review of disaggregate modal choice models, parking location models, and site-specific studies of parking behaviour. With regard to modal choice, it is noted that while there are general problems associated with such models, arising from the failure to link the commuting mode choice decision with

other travel decisions (most notably, the decision to own, or have

III

available, a car), particular problems arise in that most studies fail to specify parking time and money costs as variables separate to other journey times and costs. Thus, it is assumed that, say, marginal changes in parking and other motoring costs have the same impact on mode choice. Only three studies are identified which assess parking variables separately and two of these indicate that parking has considerably more effect than other aspects of vehicle cost. For the remaining studies, some indirect supporting evidence for this view may be deduced from the fact that, invariably, out-of-vehicle travel costs are found to play a significantly larger role than in-vehicle costs. These results support the view that parking policy measures have only a moderate influence on modal choice but, nevertheless, are likely to be relatively more important than many other traffic management measures.

One reason for the small impact of parking policy measures on modal choice, lies in the ability to offset the consequent time and money penalties by a change in parking location. This paper presents evidence, derived from disaggregate models of minimum price and time elasticities of the choice of parking location of - 0.3. This represents a substantial effect as it is measured over all car commuters and not just those with a

mode choice.

However, the paper finds that, in general, disaggregate models, whether

of mode choice or parking location are ill adapted to the analysis of parking measures. Because they have tended to consider travel in a piecemeal fashion, they give the policy-maker little guidance as to the total impact of a parking measure. Another area of concern is that because the data sets used are not usually collected with the analysis of parking measures as a priority objective, the data are often inadequate for parking analyses.

Turning to site-specific studies, it is concluded that they exhibit great

variation in the price elasticities measured, with a tendency for those at commercial car parks to exceed those for commuter parking at worksites. A number of explanations are offered for this result, and it is concluded that the elasticities derived are virtually useless for application outside the particular situation from which they are drawn. Recommendations are made with a view to improving the methodology of such studies.

1. INTRODUCTION

The purpose of this paper is to review the impact of parking policy measures on the demand for parking, and, by implication, on travel, in generaL

Parking policies have played a central role in urban traffic management over the last forty years. Initially, the concern with planners was to remove parked vehicles from urban streets, so as to improve road capacity and traffic speeds. The combination of parking prohibitions and charges, which were introduced to effect this, was, after a very short time, accepted by the motoring public as a necessary evil. It was natural, then, when planning objectives turned to the need for traffic restraint, that parking policies would play a substantial role. The impracticability of urban road pricing, the failure of public transport subsidies to produce major changes in modal shares, and the political repercussions of more draconian restrictive measures, all served to place emphasis on parking policy.

Assessing the contribution of parking measures to urban policy objectives is a two-step process. There is a need to consider both the effect on traffic levels and patterns and the subsequent impact of these on broad policy objectives. For example, the contribution of a given set of parking measures to the goal of energy conservation, depends on the degree to which travel demand reacts to such measures, and the implications of the altered travel situation for energy consumption. This paper is concerned solely with the first step in this process. The emphasis is largely but not exclusively on the effect of parking policy measures on commuting travel. This is justified by the fact that the negative effects of urban travel -congestion, environmental pollution, and waste of energy spring primarily from the travel peaks created by the journey-to-work. Other objectives are indirectly affected by this phenomenon: the problems of the decline of the inner city business economy or the destruction of residential neighbourhoods have been attributed in part to the parking problems created by commuting traffic. The concern of the paper is to review the quantitative rather than qualitative effects of parking, and where possible to summarise these effects as demand elasticities.

Section 2 discusses the structure of the travel demand response to parking policy measures, some methodological problems in measuring such re-sponses, and the empirical sources which are available. Section 3 evaluates the evidence of modal choice studies, while section ll- considers analyses of the determinants of parking location. Section 5 looks at site specific policy measures and their impacts. Conclusions and recommen-dations are presented in section 6.

2. GENERAL

Before turning to an examination of the empirical evidence relating to the impact of parking measures, it is necessary to consider the structure of the travel demand response, the methodological problems associated with measuring that response, and the range of empirical evidence available. This section examines the structure of demand and the probable demand responses to car parking measures. Possible data sources such as the Urban Transport Planning Model (UTPM), disaggregated travel demand analysis, and before and after studies of parking behaviour are then

outlined.

2.1 The Personal Travel Demand Response

Parking policy measures operate in two ways: firstly, by changing the level or structure of parking charges facing the motorist; and secondly, by changing the supply of parking spaces. The latter should be interpreted as referring not only to the physical supply of parking, but also to the restriction of access to parking, at certain times of the day, for certain

durations, or to certain classes of user.

It should be noted, however, that other journey characteristics not related to parking, have, in theory, at least an impact on parking behaviour. The time and money cost components of a trip may be conveniently

cate-gorised into in-vehicle and out of vehicle components. Thus, the

generalised cost can be written as

C = + + + mov

where t and m refer to time and money costs respectively, and the iv and ov refer to in-vehicle and out-of-vehicle components.

Trade-offs exist between each of the four components of generalised cost, so that, for example, if the costs of parking increase by A mov, then some of this increase may be absorbed by taking action which changes the

other components. Increased parking charges may, therefore, induce a change of location which increases tov. It is also possible that changes in the in vehicle variables will induce changes in parking behaviour. One such trade-off is between in vehicle journey time and money costs, on one hand, and out-of vehicle time costs on the other. If tiv or miv are increased (through fuel taxation or increased congestion, for example) then motorists might be prepared to reduce their in vehicle journey length by parking further from their ultimate destination. While this sort of trade-off may seem unlikely, few car commuters have not experienced occasions when congestion becomes so severe, that they leave their vehicle at a more distant parking location and complete their journey on foot. A second set of trade offs occurs because increases in generalised cost can also be offset by substituting a different journey or by changing the time frame for the journey.

In the light of these considerations, it is possible to identify a range of travel responses, which are influencedby the parking variables (tov and

mov) and, in theory at least non parking variables (tiV and miv):

* a change in parking location

* a change in the starting time for the trip * a change in the mode used

* a change in trip destination * abandonment of the trip

These are essentially short term behavioural responses. Longer term consequences, such as the effect of parking supply on urban decentralisa tion, are important, but outside the scope of this paper.

In examining empirical sources of travel the response to parking policy variables, it is important which or how many of the travel responses identified above are under analysis. This is particularly relevant, when an attempt is made to summarize the demand response in the form of elasticities. For example, empirical results relating to the parking price elasticity of demand for car use need have no similarity tothese for use of particular parking locations. In reviewing and comparing the data available from various empirical sources, it is important to bear this in

mind.

The extent to which any of these responses occurs is determined, in part, by the trip purpose. For the journey to work which is the main concern

here there is little or no prospect, in the short term, of either changing

the trip destination or of abandoning the trip. The more likely responses are therefore a change in parking location, a rescheduling of the trip or a change in the mode used. With regard to journey rescheduling the impact of parking policy measures is probably limited. Such measures are usually applied when there is pressure on parking spaces. In such a situation, although the individual response to parking supply constraints may be to start the journey to work earlier, this implies displacement of existing parkers. Therefore, the aggregate effect is likely to be manifest in parking location and mode split changes. The bulk of the analyses in this paper is concerned with estimating the effect of parking measures on these two quantities*.

2.2 Constraints on the Demand Response

Which of these two responses is adopted depends on many factors. However, most empirical analyses and particularly those which use a disaggregated modelling approach implicitly or explicitly make the assumption that the motorists' commuting travel decisions are based on the journey to-work time and money costs facing the car driver. In fact, the car driver's freedom to choose may be constrained when his commuting travel decision affects his ability to undertake other non-commuting trips, whether on behalf of himself or others. In this situation, his decision framework may encompass the characteristics of a number of trips, for a variety of purposes: that is, he becomes involved in multi-trip decision making. One can envisage a number of situations in which this concept becomes

relevant:-* In any event, journey scheduling has received little attention in the literature. The available evidence does not consider the role of parking measures separately (see Abkowitz, 1981, for example).

* when the individual undertakes business trips during the working day, for which a car is mandatory or optional

* when the individual is required to transport other individuals to other than his work location; the transport of family members to school is the most obvious example.

* when the individual wishes to make trips which have his work place as origin after-work shOpping and social or recreational visits. How important is multi-trip decision making? With regard to the interac-tion between commuting and business trips, some crude indicainterac-tion may be obtained from the proportion of company cars in the fleet. A recent

OECD report (OECD, 1982) has indicated that the company car

propor-tion lies between 2 and 13 per cent for various countries. In the case of Sweden, official statistics indicate 7 per cent of cars are company cars while a further 7.6 per cent are personal business cars. Algers et al. (1975) report that 14 per cent of car commuters in the Greater Stockholm area use their car for business purposes on a daily basis, while a further 20 per cent use it at least once a week. Gantvoort (19840 reports four studies, relating to experiences in Holland, West Germany and the UK, which contend that between ll and 26 per cent of motoring commuters "need" their car for some non commuting purpose. While the concept of need is a very subjective and, therefore, variable one, the figures do suggest a substantial restriction in commuting choice. Two further German studies, quoted in the same source, indicate that between 37 and 50 per cent of commuters consider themselves captive to the car.

Hansson (1980) has investigated the travel undertaken for non-work

purposes in connection with the journey-to-work in Uppsala, Sweden. She found that the multiple-purpose work trip was more common than the single purpose work trip, with shopping, serving passengers, banking and restaurant visiting the most frequent activities.

The implications of the above are, firstly, that the constraints put upon the individual choice process by business and other uses of the car and by the need to transport passengers to non work destinations are substantial. This limits the modal split response to parking policy measures. Secondly,

these constraints should be treated explicitly when modal choice studies are being carried out: we will return to this point in Section 3 below. Finally, comparative studies of constrained commuters would be useful in isolating how the degree of constraint is related to personal household, land use, and transport characteristics.

The arguments, presented above, point to the conclusion that the impact of parking policy measures is more likely to be felt in a change in parking location. Again, however, as with the consideration of journey scheduling effects, a difference may arise between the individual and aggregate effects. If parking supply is constrained, then a diversion of a target group of parkers to other locations may take place only at the expense of those using these facilities. These supplanted parkers then have to consider

changing mode or parking location. The result is that while the primary

effect of parking measures may be on parking location there may be secondary effects on the mode choice of the non-target group. Such effects, which are likely to be felt by parkers on the periphery of the traffic restraint area, may be extremely important, if reduction of congestion on radial routes is the objective. However, it appears that this topic is addressed not at all in the literature.

2.3 Empirical Sources

The previous section has discussed the structure of the travel demand response to parking measures. This section considers alternative sources of empirical information on such responses. There are two types of information available: cross-section and time series. The former compares the travel demand response of different individuals or groups of individual at one point in time in order to make inferences about behaviour over

time, while the latter uses direct observations of such behaviour.

One cross sectional empirical model, which seeks to be comprehensive in its view of travel demand structures, is the Urban Transport Planning Model (UTPM), which was widely usedduring the 19605 and 19705. This model considers trip making in terms of a four-fold structure trip

generation, distribution, modal split, and assignment. However, the

analysis of parking effects within this model has been somewhat restricted, for a number of reasons:

(i) the UTPM is normally estimated at a highly aggregated level. This means that variables describing parking characteristics are represented by zonal averages, and the full range of parking policy measures cannot be given expression in a realistic fashion.

(ii) the use of the notion of generalised costs, especially at the distribu tion and assignment stages means that the impacts of policy changes such as increased parking charges, or restrictions which impose greater walking times, are inadequately modelled. This arises in part because some impacts are exogenously specified, through the use of given not estimated values of time, for example, and also because parking factors are aggregated with other journey attributes, through the use of a common calibrated parameter.

(iii) the full effect of parking policies, on all stages of the modelling process, is rarely considered. In particular, the feedback impacts of parking capacity constraints on trip distribution, assignment and

genera-tion are inadequately handled (if, at all), with little explicit modelling of

the allocation of constrained trips to other modes or locations.

The result has been that parking demand has tended to be considered more as an output of the UTPM; rather than as a vital part of it, interacting with the other demand elements. Thus, the parking requirements con sequent to the predicted traffic flows are normally established, and parking supply is viewed as a means of constraining traffic flows to available road capacity. However, the behavioural response to an excess of parking demand, in terms of the effect on mode choice is not treated, while the effect on parking location, if discussed at all, is treated by the development of a parking location sub model, which ignores feed-back effects on other parts of the model structure. Some of these models are reviewed in section 4; however, in general, the UTPM studies are not a fruitful source of information on the general effects of parking measures. A second source of information on parking measures is the range of

disaggregate models of personal travel behaviour, which have undergone

rapid development in the last twenty years. In contrast to the aggregate approach, these models have several advantages. Firstly by modelling individual or household behaviour, they make fuller use of the information

contained in travel data sets. Secondly, because the observations are not averaged over individuals from different socio economic types or areas, the models may be transferable between areas. Thirdly, smaller data sets with reduced data collection costs may be sufficient. As parking supply is often extremely variable within areas and parking opportunities may be very much related to personal circumstances (access to employer-provided or subsidised parking), disaggregate models would seem to be especially relevant to the analysis of the impacts of parking measures. Initially, such models were applied separately to elements of the travel

response, with modal choice receiving the bulk of attention. However, this

approach raises fundamental questions concerning the nature of personal travel decision making. If decisions on whether to travel and if so, on which destination to travel to, which mode to use, when to travel and where to park are made independently, then separate single equation modelling of each of these components of travel behaviour is valid. If,

however, these decisions are related, or indeed inseparable, then the

modelling procedure should reflect this. A number of possibilities present themselves. Firstly, a sequential modelling procedure may be adOpted with the outcome of each stage of the sequence representing an input to the next stage a recursive system. Secondly, the decision mechanism could be viewed as multidimensional in nature, so that, in fact, one decision with a number of vectors of characteristics is being made. The problem then is one of deciding the nature of these vectors: for example, is the scheduling aspect of the decision simply perceived as a peak or off-peak choice or is a finer gradation of starting times appropriate? A third possibility, worthy of consideration, is that a number of grouped decisions are made, with multidimensional decision making within each group but sequential decision making between groups a block recursive system. This raises the question of the identification of blocks and sequences. If any of these approaches is used when another is appropriate, then biassed estimates may result. While disaggregate models incorporating multi dimensional decision making have been estimated, most notably by Ben

Akiva and Atherton (1977), Lerman and Ben Akiva (1976), Train (1980) and

Thobani (1984), the degree of experience with such models is slight. Yet

there is a widespread view that single equation estimation of discrete

decisions may be seriously misleading, and that considerable research is required into the nature and structure of personal travel decision making and the statistical analysis of multidimensional choices. This concern will

10

be discussed further in section 3 below when the role of car availability in

mode choice is examined.

Most disaggregate modelling has been carried out at the level of the individual. However, there is no certainty that all travel decisions are

individual ones. For other than single-person households, it is quite

possible that the major travel decisions are joint ones* among several household members. For example, where such householdshave only one car, the allocation of that car over individuals and journeys may be undertaken jointly. This view has spawned a body of research into

interactive decision-making referred to above (Clarke et al., 1981). The

implication is that if travel decisions are joint, then it is misleading to model, say, the journey-to work in terms of the time and money costs incurred by an individual commuter for that journey: time and money costs facing other individuals for non-commuting journeys may also be

relevant.

While research into the joint and multidimensional aspects of disaggregate modelling will continue, the implication for the short term assessment of parking policy measures is that the degree of constraint, which they place upon the individual response to such measures, should be taken into

account.

There have been few attempts to study the effects of a general urban parking policy. A major problem apart from data availability, lies in the fact that, in practice, parking measures go hand in-hand with other urban transport policy initiatives, and occur simultaneously with socio-economic changes so that it is difficult both to separate parking from, say, public transport improvement effects and to isolate the effect of transport policy changes in general from the impact of underlying changes in the social and economic structure (Bailey and Layzell, 1982). The result has been that studies of parking demand over time have tended to consider isolated parking measures and their effects over short term, when socio-economic changes are few. Because, it is difficult to characterise non * In the literature of disaggregate demand modelling, there is a

tendency to use the term "joint" to refer both to co-operative decision making among individuals and decisions extending over a range of travel responses. Here, the latter is referred to as "multidimensional"

11

price measures such as parking bans, these studies have tended to emphazise analyses of price measures. This is an obvious deficiency in such studies, to which further reference will be made in Section 5, where they are reviewed. The time series studies fall into two groups: the first is concerned with parking price changes at commercial garages while the second relates to the imposition of charges at work sites. In each case the price effect is measured by before and after counts of vehicles or commuters. This raises the important question of whether a pure demand function is identifiable by this procedure: if supply is constrained either before or after the price change, then the resulting demand change may

be over or understated. This is further discussed in Section 5 below.

Section 2 has examined different views of how the demand for personal travel is structured, the means by which parking policies can'affect that demand, the assumptions underlying the empirical measurement of models of demand, and some of the problems of applying such models to the assessment of parking policy measures. The next section, examines the empirical evidence on the role of parking policy in influencing one of the more desirable changes in travel patterns an alteration in the modal split.

12

3 THE EVIDENCE OF MODAL CHOICE STUDIES

As mentioned above, one of the primary objectives of car parking policy is to induce changes of mode. Car sharing, public transport, cycle, and walk modes are all seen as advantageous in this respect. Since the early nineteen sixties, a large number of studies have beenundertaken with a view to determining the factors which influence mode changes especially between car and public transport modes. While a number of studies have examined modal choice at a zonal, route, or similarly aggregated level

(for a recent example, see Lioukas(l982)), the bulk of them have been

disaggregate in nature, and it is these studies which are reviewed here. They have a number of features:

(1) they analyse modal choice at the level of the individual. They,

therefore, assume that the modal choice decision is not a joint one.

(ii) they are calibrated on commuting trip data, either as perceived by the respondent* or based on indirect or engineering estimates

(iii) Generally, they view the modal choice decision as independant of

other travel decisions.

(iv) they express the disutility of travel by a particular mode as a function

of both "system" and socio-economic variables. The former comprise variables which describe the transport network facing the individual in terms of time and money costs. The latter are variables describing the characteristics of the respondent and those of the household in which he

resides.

Before turning to the results of the studies themselves, it is necessary to consider a number of constraints imposed on them by the features

outlined above.

* Two types of data have been used. The first relies on revealed data concerning the respondents existing mode and his perceived

alterna-tive mode(s). The second determines changes in system costs which

would induce the respondent to change mode the transfer pricing method. The bulk of available studies are of the former type, although there has been a recent resurgence of interest in the latter. See Bonsall (1983) for a brief review of transfer pricing and related methods.

13

3.1 Mode availability

As the car is the predominant mode in most urban areas, it is normally part of the choice set considered by commuters and, therefore, by modal choice analysis. However, commuters cannot choose a mode which is unavailable to them, and this raises the question of which commuters have a genuine car choice. Obviously, car availability is only loosely related to car ownership: membership of a car owning household does not guarantee car availability. Conversely, membership of a non-car owning household does not preclude car use, especially as a car passenger.

How can car availability be measured? Gwilliam and Banister (1977)

propose an objective measure, deeming a trip to be car available, when the car is not in use and located at the point from which that trip started. This definition can be criticised on a number of grounds, most notably, that it ignores car passenger availability, and that, fundamentally, the presence of a car at the origin of a trip dOes not guarantee the release of that car for the trip in question. The car is available only at that point in time, and not necessarily for the duration of the trip. An alternative procedure considers the car to be available only if it remains unused for the duration of the trip. However, as Bailey (1984) points out, even in this situation, the trip-maker may not perceive that he or she has a right to the car; even though unused, it may be unavailable. He suggests that there are four categories of trip-maker, each of which requires a different approach:

* _{licence holders in car owning households who have exclusive use of a}

car

* _{licence holders in car owning households where there is competition}

for the use of the car

* non licence holders in car-owning households * people in non-car owning households

14

Of these, only the first category may be regarded a priori as having a mode choice, which can be characterised in terms of the commuting journey system costs. Bailey reports one study which shows that only 15 per cent of individuals fall into the first category. For the remaining categories, car availability (either as a driver or passenger) depends on a complex set of non-objective factors.

The alternative approach is the subjective one viz. that of relying on the individual's perception of availability. Lucarotti (1977) reports on a study of #57 households, in which availability was established by asking the respondent whether the trip could have been made as a car driver or car passenger. While a substantial variation in car availability (both as drivers and passengers) as between men and women, and car and non-car owning households, was apparent, little variation in the use of cars (where they were available) occurred. Between 82 and 92 per cent of those who had cars available, did, in fact, use them. Thus, nearly all of the variation in modal shares observed was due to car availability and not utilisation rates. Another study in the same vein is that of Donald (1979) for whom a car is available, if either one or more of the following situations occur:

(1) the person makes the trip by car

(ii) the person uses another mode, but states that he could have travelled as a car passenger without altering the driver s trip pattern

(iii) the person uses another mode, but states that he could have undertaken the journey as a car driver, without denying its use to another

member of the household who wished to use it.

In a survey of 1,732 households Donald found that whereas 833 trips or 12 per cent of trips from car owning households used public transport, this fell to 78 trips or 1 per cent, when car available trips were considered. The remaining 755 trips did not have a car available for use and could be considered captive to public transport. The most common causes of their captive status were lack of a driving licence and use of car by other household members. These subjective evaluations of car availability are also open to critisism: respondents may perceive that they have a right to use the household car, even though they never in fact choose to exercise

15

that right. However, this does not obviate the major conclusion emanating from these studies: in certain urban configurations at least, where trip makers have a car choice they are extremely unlikely to choose public

transport.

Consideration of the problem of car availability indicates that the modal choice decision often may not be an individual one. If a household contains more licence holders than cars, then an element of competition for car use will occur, which can only be resolved by joint decision making. The decision process may then be dependent not on the characteristics of the commuting journey for a single individual, but rather on the daily demands for car use emanating from competing household members, arising from their travel needs and the generalised costs of different trips and modes. Modelling a decision process which is based on travel demand throughout the day, and across different individuals, is a difficult task. It may be

more fruitful to use the concept of car availability (with all its faults)

together with mode choice in a multidimensional decision framework. So far the discussion has been in terms of car rather than mode availability. However, the availability of public transport may also be important. The issue here is a different one, as ownership of a vehicle is not in question. Availability of public transport is then more likely to be a reflection of its system costs. One American study (Tardiff, 1977) found that of a sample of Los Angeles suburban dwellers only a negligible number considered that they did not have a car available. The remainder could be divided into two groups those who were car captive and those who perceived the existence of a choice for public transport. In the case of commuting trips, 56 per cent of respondents considered themselves captive to the car; for non-commuting trips the equivalent figure was 69 per cent. The study finds that the perception of public transport avail-ability is very much related to system characteristics. If this is true, then unlike car availability, public transport availability may not be a con straint on modal choice analyses.

How do modal choice studies deal with the problem of car availability? Table 1 gives some indication of this. The most notable feature of the table is the number of studies which do not make mention of the problem. While this does mean that the problem was ignored, the lack of reporting

16

indicates a low level of concern. For the remaining studies, two approaches are discernible. The first approach attempts to limit the data set to those who may be regarded as having a genuine choice. The limitation is usually based on one of the

following:-* respondents who have a car available * respondents who hold driving licences * respondents from car owning households.

While the first method, that of subjective measurement of car availability has the potential, at least, of excluding those for whom car is not a real choice, the other methods have substantial drawbacks: the presence of a car in the household, or the holding of a driving licence in no way guarantees access to a car. Studies which use either of these methods, inevitably include respondents for whom a real choice is not available. The implications for the model co efficient values are unknown, but confidence in the results as a reflection of a trade off process is not

enhanced.

The second approach which is used does not place any limitation on the data set, but rather includes an explanatory variable to reflect car availability. The explanatory variables used

are:-* a subjective measure of car availability * ownership of a driving licence

* individual or household car ownership * the number of household cars per driver

These variables are normally included as additive dummies so that the co efficient values associated with the variables are (or should be) un affected. They are, therefore, invariant as between those who are exercising a real choice and those who are not. While such models may give a good fit to the sample data, their use for predictive purposes is highly dubious. They are considerably inferior to the former approach of limiting the data set to those who have a choice. As is evident from the above, modal choice studies have tended to treat car availability (or car ownership) as an exogenous variable. However, for many, the decision to own a car may be closely linked to the commuting mode choice decision.

17

It may, therefore, be more appropriate to view such decisions in a multidimensional framework. This has been done by Lerman and Ben

Akiva (1976), for example who consider that a maximum of five alterna

tive mode and ownership choices are available. The probability of a household choosing a given level of car ownership and of the primary work travelling by a given mode to work is estimated using a multinominal logit approach. However, the fact that only the primary worker is considered, means that the problems of competing demands for car use within a household are not considered in a simultaneous framework. Similar

remarks apply to a later study by Ben Akiva and Atherton (1977). One of

the problems with the multinominal logit approach is that one of the properties of this model the independence from irrelevant alternatives property becomes increasingly untenable as the choice set expands. An alternative approach, the structured logit model has been used as a means

of avoiding this problem (Train, 1980 and Thobani, 1984). In these models,

the effect of the work trip on car ownership is modelled through the inclusion of an aggregate work trip utility term in the ownership model. Because of data limitations, Train measures this .term for only one worker in the household, but finds that the resulting model parameters do not differ significantly between households with one and more than one worker. Thobani tests aggregate work trip utility terms based on one and more than one worker, but finds that only the former is significant. These results suggest that it is the primary worker in the household who has the-major or exclusive access to the car, and that other workers cannot be

considered to have a car available to them.

What are the implications of the foregoing for car parking policy? Firstly, very few trip makers, in the towns surveyed, who perceived a car choice, did in fact use public transport. Whether this remains true for the larger urban conurbations or across trip types is uncertain. There is a substantial need for the systematic reporting of car availability, on both objective and subjective bases, when data collection for travel demand research is being undertaken. While the relatively low use of public transport by car-available travellers suggests that there is a large potential for transfers to public transport, it also indicates that such transfers may be difficult to achieve: the range of system cost differences, which currently exists between modes, has not induced such achange. Secondly, the credibility of modal choice studies as a source of empirical information on the

18

impacts of parking policy measures, is, inter alia, related to the way in which car availability is handled in such models.

3.2 Multi-Trip Decision Making

In the section above, consideration was given to the situation in which the commuting decision is not an individual one, but requires co operative action on the part of household members. Thus, mode choice decisions often cannot be related to system costs facing an individual commuter. Another situation in which the characteristics of the commuting trip are less relevant is when the mode choice decision affects the commuter's ability to undertake other non commuting trips during the day. The decision is then likely to be based on the characteristics of a number of trips, for a variety of purposes. This was discussed in Section 2.4 above, when it was concluded that the constraints put upon the individual choice process by business and other uses of the car is substantial. Perusal of the modal choice studies of Table 1 indicates that few deal with these con-straints in an adequate fashion. Thus, for many studies the choice population includes many who either have no real choice, or one con strained by considerations other than the money and time costs of the alternative journey to work modes. This has unpredictable effects on the

co-efficient values derived.

3.3 Variable Specification and Modal Choice

It is obvious that if modal choice studies are to provide useful information on the effectiveness of car parking policy measures, then they must contain relevant parking variables. The variables relevant to car parking are, firstly, the money cost of parking, and secondly, the time costs associated with searching for a parking place and travelling between parking and work 10cations. The usefulness of modal choice studies depends on the manner in which these variables are incorporated in the model. In relation to parking costs, it can be seen from Table i that five of the nineteen studies specify parking costs as a separate variable, thirteen include it in the total money costs of the trip, while one study omits it altogether. Elasticities of mode choice with respect to total

VTI RAPPORT 308A Ta bl e 1 A Sur ve y of Mo de Ch oi ce Mo de ls fo r the Jo ur ne y to Wo rk AU TH OR (S ) PU BL IC AT IO N LO CATI ON SA MP LE 5 DATE Mc Gi ll ivr ay Br own Mc Gi ll ivr ay Ta lvi tie Wi gn er Mc Fa dd en Al ge rs et al. 1970 1972 1972 1972 1973 1974 1975 SI ZE Sa n Fr an cisc o -U. S. A. Va nc ouve r -Ca na da Sa n Fr an ci sc o3, 00 3 U. S. A. Ch ic ag o 15 9 U. S. A. Ch ic ag o 1, 31 4 U. S. A. to 2,17 9 Sa n Fr an ci sc o 16 0 U. S. A. St oc kh olm, 1, 05 0~ Swe de n DA TA ' YE AR 19 67 19 65 19 64 19 56 19 72 19 68 MO DE L TY PE Di sc ri mi na nt Disc ri mi na nt Di sc ri mi na nt Di sc rimi na nt Lo gi t Pr ob it Lo gi t Pr ob it Lo gi t Lo gi t CH OI CE CO NS TR AINT S ON TH E SE T CH OI CE PR OCES S Ca r: Bus Car: Bus Ca r: Ra il Ca r: Tr an sit Ca r: Tran si t Ca rzB us Ca rzR ai l Ca rzB us Ca r: Tr ansi t Ca r ava il ab ilit y in cl ud ed as in de -pe nden t va ri ab le Ca r own er sh ip us ed as in depe nd en t va ri ab le Re sp on de nt s mus t ha ve "a cc es s to both mo de s" ; "R eS po nd ent do es no tdr ive " in -cl ud ed as in de -pe nd en t vari ab le Da ta se tl im it ed to th os ewh o sh ow "c om pl et e" mo de ava ilab il it y; th os ewh o us e ca r for bus in es s pur -po se s exc lud ed SP ECIF IC AT IO N OF PA RK ING CO ST S In cl ud ed in to ta l mo ne y co sts Da il y pa rk in g char ge In cl ud ed in to ta l mo ne y co sts Da il y pa rk in g ch ar ge In cl ud ed in to ta l mo ne y cost s In cl ud ed in to ta l mo ne y co st s Da ily pa rk in g ch ar ge SPEC IF IC AT IO N OF EXCE SS TI ME In cl uded in to ta l tr ave l ti me In cl ud ed in to ta l tr ave l cost s As se pa ra te va ri ab le 19

VTI RAPPORT 308A AU TH OR(S ) PU BL IC AT IO N DA TE LO CA TI ON SA MP LE SI ZE DA TA YE AR MO DE L TY PE CH OI CE CO NS TR AI NT S ON TH E SP EC IF IC AT IO N OF SE T CH OI CE PR OC ES S PA RK IN G CO ST S SP EC IF IC AT IO N 0F EX CE SS TI ME Le rm an an d Be n Ak iva 19 76 Wa sh in gt on DC U. S. A. 19 68 Lo gi t Ca r: A mul ti -d im en si o-In cl ud ed in Tr an si t na l ca r own er sh ip -to ta l mo ne y mo de ch oi ce mo de l co st s is ca li br at ed . Ca rs pe r dr ive r in cl . as in de pe n-de nt va ri ab le As se pa ra te va ri ab le but di vi de d by di st an ce Be n-Ak iva an d Ri ch ar ds 19 76 Ei nd ho ve n Ho ll an d 390 19 70 Lo gi t wa ik ; Da ta se tl im it ed to Cyc le : ho me -wo rk h om e tr ip Mo pe d; ch ai ns ,a nd to in di -Ca r: vi dua ls wi th a li ce nc e Bus ; re si di ng in ho us e-Tr ai n; ho ul ds wi th a ca r As se pa ra te va ri ab le Ga ne k an d Sa ul in o 19 76 Pi tt sb ur gh U. S. A. 74 0 19 75 Lo gi t Dr ive Ca rs pe r dr ive r In dl ud ed in al on e: in cl ud ed as to ta l mo ne y Ca rs ha re zi nd ep en de nt co st s Tr an si t: va ri ab le ' Eg re ss ti me us ed as se pa ra te va ri ab le Be n-Ak iva an d At he rt on 19 77 Wash in gt on DC U. S. A. 1975 Lo gi t Dr ive A mul ti -d im en si o-In cl ud ed in al on e: na l ca r own er sh ip -to ta l mo ne y Ca rs ha re zm od e ch oi ce mo de l co st s Tr an si t: is ca li br at ed fo r pr im ar y wo rk er s; Ca r own er sh ip pe r li ce ns ed dr ive r in cl ud ed as in de -pe nd en t va ri ab le wh en mo de ll in g mo de ch oi ce fo r se co nd ar y wo rk er s As se pa ra te va ri ab le Gi ll en 19 77 To ro nt o Ca na da 51 5 19 64 Lo gi t Ca r: Da ta se t li mi te d Pa rk in g ch ar ge Tr an si t: to ho us eh ol ds wi th pe r jo ur ne y at le as t on e ca r an d ha vi ng at le as t one li ce nc e-ho lder ! In cl ud ed wi th pa rk in g-ch ar ge in co mp os it e ti me an d mo ne y va ri ab le 20

VTI RAPPORT 308A AU TH OR(S ) Pa ro dy Tr ain Ga lb ra it h an dHe ns he r K0 pp el ma n an d Wi lm ot Ba ji c Dun ne Th ob an i PU BLIC AT IO N DA TE 19 77 19 80 19 82 19 82 19 84 19 84 19 84 LO CA TI ON Am hers t, U. S. A. SA MP LE SIZE 16 4 Sa n Fr an ci sc o 63 5 U. S. A. Syd ne y Aus tr al ia Wa shin gt on DC 2, 08 8 U. S. A. To ro nt o Ca na da Li vi ng st on U. K. Ka ra ch i In dia 38 2 24 3 38 5 33 0 DA TA YE AR 19 72 -3 19 75 19 71 19 75 19 79 MO DE L TY PE Lo gi t Lo git Logi t Lo gi t Lo gi t Lo gi t Lo gi t CH OI CE SE T Ca r: Bus : Ca r: Bus : Ra il : Ca rp oo l: Ca r: Ra il : Dr ive al on e: CO NS TR AI NT S ON TH E SP EC IF IC ATIO N 0F SP EC IF IC ATIO N 0F CH OI CE PR OC ES S Th os e wh odi d no t In cl uded in ha ve a car ava il ? to ta lmo ne y able or we re no t co st s near a bus se rvi ce we re exc lud ed A mul ti di me ns iona l In cl ud ed in ca r own er ship -m od e to ta l mo ney ch oi ce mo de lis co st s ca li br at ed On ly li ce nc eh olde rs Pa rk in g co st s from ho us eh ol ds an d pa rk in g wi th ca rs we re in -co st s di vi ded cl ud ed ; Th os ewh o by in co me ha d a co mpan y ca r or re qui re d a ca r fo rbus in es s us e we re exc lud ed ; a va -ri ab le re pr esen ti ng co mp et it ion fo r ca rs was in cl ud ed Ho us ehol d ca rs pe r Incl ud ed in dr ive r in cl ud ed as to ta l mo ne y PA RK IN G CO STS Ca rs ha re zexp la na to ry va ri ab -co st s Ca r: Tr an si t: Ca r: Bus : Ca r: Ta xi Mul ti di mens io na l Ri ck sh awzm odel of ca r Mi ni bus : Wa lk : Bus : le -In cl ud ed in to ta l mo ne y cost s A num be r of car In cl ud ed in ava il ab il it y to ta l mo ney va ri able s te st ed co sts In cl uded in to ta l mo ne y own er sh ip an d co st s mo de ch oi ce wa s te st ed EX CE SS TI ME On ly bus exc es s ti me in cl ud ed as se pa ra te va ri ab le In cl ud ed as se pa ra te va ri ab le In cl ud ed as a se pa ra te va ri ab le Incl ud ed as ase pa ra te va ri ab le Exc ess ti me di vi -de d by di st ance incl ud ed as se pa -ra te vari ab le Incl ud ed as se pa ra te va ri ab le 21

22

money costs are useful for parking policy assessments, only if parking costs and other motoring costs are always in fixed proportions, or the responses to changes in these two components of money cost are identical. The fixed proportionality of parking and other motoring costs is obviously an unrealistic assumption to make. The hypothesis of equal responses is also unlikely to hold true as parking costs are fixed or time related payments, while other motoring costs (principally, fuel costs) are distance related. In addition, parking charges are often incurred on a trip by trip basis in the sense that a separate parking money transaction must be undertaken with each trip; fuel purchase on the other hand is made periodically and the fuel consumption characteristic of commuting trips, as Opposed to trips for other purposes, may not be known to the motorist.

Gillen (1977) has tested the hypothesis of equal co-efficients relating to

parking and other money costs on commuting trip data relating to 515 individuals working in Toronto C.B.D. He found that the null hypothesis of equal co efficients was rejected. When couched in terms of elasticities, the parking charge was found to have five times the effect of relative

fare costs (mileage related car costs relative to the public transport fare).

This result implies that the equality of parking and other money costs cannot be assumed and should be put to empirical test. A large number of mode choice studies may be, therefore, of little value for the evaluation of parking policy.

For the home to work journey, total car travel time may be broken up into three components access time from the home to the car, in-vehicle travel time, and egress time from the parking to work locations. The first and last of these components are often called "excess time". Again from the point of view of estimating the effects of different parking policies, it would seem important to treat in vehicle and excess time differently. As

excess time is made up of walking and (in the case of public transport)

waiting time, it is likely to be valued differently than time spent in the

vehicle. Of the nineteen studies examined, ten assess excess time

separately, while one includes it in a composite variable, representing time and money costs (Gillen, 1977). The failure to identify parking costs and excess time as separate variables, represents a source of potentials

misspecification in mode choice modelling. It is noteworthy that of the nineteen studies examined only two (Talvitie (1972) and Galbraith and

Hensher (1982) incorporate both variables. Moreover, in-vehicle time

23

comprises two components: one is travel to the environs of the ultimate destination, the other is parking search time, which may involve looking

for parking of the appropriate type (cost, duration etc.). It is reasonable

to suppose that the disutility associated with search time differs from that of other in vehicle time, and merits separate specification in the

models.

3.4 Elasticity Measurements

In making comparisons between different studies, knowledge of the co efficient values is insufficient, as these are related to the units in which the explanatory variables are measured. Elasticity values which relate the modal choice to each of the policy variables of interest avoid these problems. In some instances, it is possible to measure arc elasticities from the presentation of the response of the model to discrete changes in the variables (an increase of ten minutes in excess time, for example). The

log arc elasticity of demand for travel (Q) with respect to an explanatory variable (X') may be defined as

A(log Q)

Tig og-ar C) =

Alternatively, the linear arc elasticity is defined as

AQ - l(X1 + X2)

l<Qi+Q2)

AX

Q - -

=

nX(11n arc)

24

However, the logit model which is the mathematical specification em ployed by nearly all the relevant studies, presents an opportunity of

estimating point elasticities, directly from the modelform. In logit

models, the probability of choosing a particular mode j (in a two mode

choice situation) is given by:

Pj : 2 b1jX1j / 1 Z bij1j

where Xij are the independent variables, and bij are

the co-efficient values.

The elasticity of mode choice with respect to the variable Xij is given by

P.

J ~ ..

n = 6 PJ - zlJ-= (1 - Pj) bjj ij

ij

6 xii

Pi

This represents the percentage change in the proportion of trips by, say, car, arising from a unit percentage change in the independent variable Xij. In order to estimate the elasticity appropriate values for Pj and X must be input. An obvious procedure is to input the mean values of these variables the "representative individual approach". However, as the function is non-linear, aggregation problems arise, in that the elasticity nggj evaluated at the means of the explanatory variables is not the same as the mean of the individual elasticities. Dunne (1984) has shown that the representative individual approach tends to overestimate the impact of the independent variable changes, the values derived sometimes being double that of the corresponding weighted elasticity. Thus, the manner in which the elasticity estimates have been derived is of importance in determining their usefulness. Where elasticities are not reported, it may be possible to derive them from the co-efficient values, although only representative individual elasticities will normally be possible.

25

Where elasticity estimates are not available and cannot be calculated, information on the effectiveness of parking policies relative to other policy measures, may be obtained by comparing the co-efficient values relating to the in vehicle and out-of vehicle components of the system costs. If the co efficient values relating to the out of vehicle elements are relatively large, then parking policy measures are likely to be

relatively more effective.

In comparing elasticity values acrossstudies, a final point which must be considered is the form of the system variables. Three alternative specifications have been used: mode specific variables for time and money costs the absolute variable form; time and money cost differences between modes the difference variable form; and time and money cost ratios the ratio variable form. These alternative specifications in corporate very different behavioural assumptions and the elasticities derived from them are not comparable. In recent studies, the most widely used has been the difference variable form, which assumes an identical response to system variables across modes.

3.5 Empirical Results

With these points in mind, it is possible to discuss the results from the twelve studies, which specify parking costs or excess times as separate

variables. McFadden (1974) examined the car-bus choice for 160 commuters in San Francisco. While he reports detailed results relating to

total travel time and total cost variables, he also alludes to additional

analyses which assessed parking costs separately. These suggest, but not with sufficient statistical precision, thatparking costs are weighted more heavily than mileage related or car maintenance costs.

The study by Brown (1972) of car-bus and car rail choice commuters in

Vancouver departs from the usual procedure adopted in mode choice studies in that stated rather than revealed consumer preferences are used. That is the commuter is asked to state how his behaviour would change in the face of hypothetical changes in policy variables. One of the policy

26

changes considered by Brown is a virtual doubling of the daily parking charge. It is possible from the data presented in the study to estimate the log-arc price elasticity of demand, resulting from this change, at 0.32 that is a 3 per cent shift to public transport for every 10 per cent increase in parking charges. Algers et al (1975) considered the car transit choice situation for 1,050 commuters in Stockholm, again using logit analysis. While car parking costs are specified separately as an explanatory variable, the analysis of car excess time is restricted (somewhat strange-ly) to access time to the vehicle. It is not possible to estimate the elasticity with respect to parking costs from the results presented, as the mean values of the variables, which are required to calculate representa-tive individual elasticities, are not given. It is interesting to note, however, that the size of the co-efficient for daily parking charges is somewhat greater than that for the difference variable representing in

vehicle costs. Gillen (1977) also presents an analysis of the effect of

parking charges on car-transit choice. Using the method proposed by Westin (1974), a weighted parking price elasticity of the probability of car use is estimated at 0.31. It should be noted that this elasticity is with respect to a ratio variable parking costs divided by public transport fares. The impact of car excess time is not assessed separately, but included in an alternative specification of a composite variable compris ing parking fees plus excess time costs. Thus, not only is it impossible to estimate the elasticity with respect to excess time separately, but it is ' also unclear as to the effect this has on the estimated parking price elasticity.

Talvitie (1972) does not present the results of his analysis of car-public transport choices in elasticity form. However, he finds that in vehicle travel time is not significant, but excess time is. The co efficient with respect to out of-vehicle costs is lessthan half that of in-vehicle costs.

Ben Akiva and Richards (1976) do not include parking charges even as part

of a composite money variable. Because of the absence of information relating to the means of the explanatory variables, it is not possible to calculate elasticities from the logit functions which are calculated;

however, the co-efficient values associated with excess time are substan

tially greater than those for in vehicle time: in particular, walking time has twice the disutility of in vehicle time. Ben Akiva and Atherton (1977)

27

do include parking costs, but as part of a composite variable. Excess time is also assessed but as a ratio of distance travelled. It is, therefore, not possible to compare the relative values of in-vehicle and excess time.

Bajic (1984) examined the modal choices of 385 Toronto car or transit

users using a binary logit formulation. Whereas parking costs were included in total money costs, excess time was evaluated separately. For a 10 minute time change, it was found that the excess time elasticity fell in the range 0.30 to 0.35, depending on the income level of the commuter. It was also found to exceed that of in-vehicle time by a substantial margin: for the highest income groups a ratio of 6:1 was

established. Galbraith and Hensher (1982) calibrated two models relating

to the Sydney region (of Australia) in order to examine the geographical transferability of modal choice models. No elasticities are presented and, it is not possible to estimate even the representative individual values.

However, excess time was ag relatively more important than in vehicle

time, although parking costs were not significant. Ganek and Saulino

(1976) provide an analysis of 740 commuters with a car, car pooling or

public transport choice. Parking charges are not assessed separately, but excess time has, again, a larger co-efficient value than in vehicle time.

Train (1980) and Thobani (1984) both present structured models of car

ownership and mode choice. The former finds that the disutility associ ated with walking is almost double in-vehicle time and four times that of transit-in vehicle time. For the latter, the general result holds but the weighting of car and bus in vehicle time is reversed. In both studies

journey costs are significant.

As may be seen, the concept of an individual-level modal choice function calibrated on commuting system costs is open to criticism on a number of counts. Firstly, the mode choice decision may be a joint one between household members, and determined by car allocation decisions. Thus, many commuters from car-owning households may not have a car avail able to them and may not therefore, be exercising a real choice. Secondly, modal choice and car ownership decisions may be closely linked in a simultaneous process: only a small number of studies have examined this possibility. Finally, mode choice decisions may depend not only on the characteristics of the journey to work, but also on the characteristics of

28

non work trips. In the present state of knowledge, it is not possible to say

to what extent these factors lead to biassed modal choice co efficients.

Even apart from these concerns modal choice studies are a limited source of information on the effects of parking measures. While most recent studies distinguish between in-vehicle and out of-vehicle journey times, none have considered the valuation of the search costs associated with parking. More importantly, the failure to specify parking costs as a variable separate to other journey costs. is a major drawback.

Few elasticity measurements are available, and these represent different variable specifications. However, a number of common observations emerge. Firstly, in general, travel time is found to be an important explanator of modal choice. When the separate components of travel time are analysed, excess time co-efficients are found to have lower standard errors and higher co-efficient and elasticity values (where estimated) than in-vehicle time. With regard to travel costs, the position is less clear. However, many studies find that total travel costs are either not statistically significant or are subject to large standard errors (Koppelman and Wilmot (1982), and Ben-Akiva and Richards (1976), for example). Where travel costs are found to be significant, the relevant variable is usually specified to include parking charges. The evidence presented by

Gillen (1977) and Algers et al. (1975) indicates that when in-vehicle costs

and parking charges are separately assessed, the latter is found to be a considerably more important explanator of modal choice; however, this

result is not supported by Galbraith and Hensher (1982).

In general, the results indicate that out-of vehicle costs, whether of time or money, are substantially more important in determining modal choice. This supports the view that parking policy measures are likely to be relatively more important than many other traffic management measures in influencing mode choice. Distortions in parking pricing policy are therefore more serious in their implications for modal choice, than distortions in public transport fares or car fuel prices. However, these general conclusions must be tempered by the knowledge that the number of commuters who have an individual commuting choice may be very

small.

29

As may be seen, for a variety of reasons the number of modal choice studies which are of use for the study of parking policy are few. In order to make modal choice studies more relevant to car parking policy evaluation, a number of conditions must

apply:-* the system variables should be more carefully specified; in particular, time and money cost variables should be broken down into their component parts - walking, waiting, and in-vehicle time * the implications of the alternative specifications of the system

variables in terms of absolute, difference or ratios should be

examined and their relative efficiency assessed

* the results should be couched in the form of elasticities, calculated using an explicit and accurate aggregation procedure.

30

THE EVIDENCE OF PARKING LOCATION STUDIES

As explained in Section 2.3, from the point of view of congestion alleviation, the impact of parking policies on mode choice are of great interest. However, changes in parking location may also be desirable, if they help match demand and supply, thereby reducing search times, if they result in diversions to less congested routes or if they indirectly cause changes in the modal split. This section reviews studies which model the distribution of parking demand over the urban area. Two types of parking location study have been undertaken. The first type arose as an addendum to UTPM studies*, and seeks to establish parking patterns at an aggregate level - a zonal level, for example. The second is disaggregate in nature and models parking location decisions at the level of the individual.

4.1 Aggregate Parking Location Studies

At its simplest level, there has always been a need for general parking planning purposes to develOp relationships between land use characteris-tics and parking supply characterischaracteris-tics. Levinson and Pratt (198M make use of the relationship between short term parking and retail and shopping activities, and long term parking and employment to model parking location. Firstly, the long and short term peak accumulation of parkers is calculated. Then, the ratio of zonal employment to total CBD employ-ment is used to model the allocation of long term demand, and the ratio of the zonal retail and service floor space to the similar total CBD floor space to model short term allocations. This approach ignores the cost of parking as a determinant of location and fails to deal with situations where parking demand exceeds supply. Bates (1972) has proposed (but not

* The deficienCies in the way in which UTPM studies deal with parking have also given rise to a number of studies aimed at establishing relationships between individual parking characteristics (time of arrival, duration of stay etc) and aggregate measures of parking performance (accumulation ratios, the relationship of trip ends to capacity require ments). These studies range from simple numerical accumulation methods

(Parker, 1973) to more 50phisticated analyses based on queueing theory applications (Wigan and Broughton, 1980).

31

calibrated) a model for the allocation of excess zonal demand using the gravity model analogy. Austin (1973) uses diversion curve analysis to allocate zonal parking demand between available parking facilities. The disutility of using each parking facility is expressed as the sum of money and time costs. Excess demand is eliminated by an iterative procedure which alters the values of the exponent in the diversion relationship. Ellis and Rassam (1984) adopt a linear programming approach to the problem, whereby the total disutility of all parkers is minimised subject to supply and demand constraints. The disutility of a given parking location is characterised by combining cost and distance from parking location to destination in a single relationship. Gur and Beimborn (1984) model parking choice as an equilibrium network assignment problem based on finding minimum impedance paths for each parker. Impedance is cal culated as a function of walking time, parking fees, search time and the

probability of incurring a fine (for illegal parking).

Many of these models can be calibrated to provide a reasonably close fit to reality, and are therefore supportive of the view that parking behaviour can be explained in terms of money costs and walking times. However, they have a number of drawbacks: firstly, some of the parameter values which are of interest to the development of policy (the value of walking

time, for example) are often exogenously specified, and are not derived

from the model calibration process itself; secondly, the models are not supported by a theory of individual parking behaviour, and the calibration constants derived are not amenable to economic interpretation.

4.2 Disaggregate Parking Location Studies

The use of disaggregate data to model mode choice, has prompted similar analyses of parking location. One of the first of these is due to Ergun (1971), who related the proximity of parking location (to ultimate destination) to parking charges and a number of socio economic variables. The model, which is estimated on data relating to Chicago, examines the

individual's choice of whether to park within one, two, three etc. blocks of

his parking location. The effect of parking cost differences on the choice

32

of block is incorporated through the use of a variable representing the rate of change in parking costs by block. Both parking cost variables were found to influence parking location; a 50 per cent increase in parking costs would give rise to a decrease in the percentage parking in the first block from 60 to 45 per cent implying an arc price elasticity of - 0.43. The percentage in the second block would be unchanged, while that in other blocks would increase. A study with a similar methodology was undertaken by Gillen (1978) using data relating to Toronto. He calculates a point elasticity of - 0.33 with respect to daily parking charges at the first block; moreover, the elasticity rises for blocks of greater distance, indicating that peripheral parkers are more price sensitive. Gillen also estimates elasticities with respect to time costs, which is the same as that for price for the first block, but declines sharply at successive blocks. Both these studies indicate a substantial tendency for relocation in the face of parking policy measures.

In the European context the differentiation of choices in terms of blocks is not valid. The approach adopted by Van Der Goot (1982) with respect to an analysis of parking choice in Haarlem (Netherlands), is to identify certain groups of parking locations and to model the decision to choose between such groups. The latter were defined in such a way that parking charges and restrictions were uniform within each group. Initially, twenty-two alternative groups were identified, although these were subsequently reduced to six. The study considers a wide range of factors influencing parking choice walking time, parking charges, duration restrictions and occupation rates. The use of the latter variable is in contrast to the other studies which assume that market prices adjust to maintain an occupancy rate close to unity. For the journey to work, the study finds that walking times and duration restrictions are statistically significant, although parking charges are not. The study results also indicate consumer preferences for off street carparks and parking

garages.

All of these studies support the view that parking supply restrictions, which increase walking times, have a substantial effect on parking location behaviour. There is considerable, although not universal support for a similar role for car parking charges. If relocation is substantial, will