M.Alqhatania，S.Bajwab，S.Setungec

(School of Civil，Environmental and Chemical Engineering，RMIT University，Melbourne 3001，Australia)

0 Introduction

Urban structure，urban sprawl，and disconnection between local activities and public transport increase the demand for car ownership.Car ownership is a very important determinant of individual and household travel behavior(Bhat and Pulugurta 1998)［1］.The socioeconomic and land use factors are also essential to enable urban planners and decision makers to understand how households choose the number of cars to own(Potoglou and Kanaroglou 2008)［2］.

The impact of car ownership is seen in its influence on trip frequency choice，destination choice for non-work and mode choice to work and non-work activity destinations，and the capacity for organizing chain activities in a tour(Bhat and Pulugurta 1998)［1］.Moreover，the car ownership variable can be applied as proxy in travel demand such as a cross-classification variable in trip generation modes，in order to better identify the variances in household trip generation.As such，it is understood as a major socioeconomic factor affecting trip distribution and mode choice models，reflective of the relationship between increased car owner-ship and availability and variation in destination or mode choice(Chu 2002)［3］.

To date，few studies have examined the effect of car ownership on travel behavior in Riyadh and Melbourne.Both Riyadh and Melbourne have high levels of car ownership and dominance.Currie and Senbergs(2007)［4］found that 23 percent of households in suburban Melbourne areas have little or no access to local activities，and limited access to public transport.They also noted that，despite their low weekly income，residents were running two or more cars.In the Saudi Arabian context，Koushki(1987)［5］found that the higher car ownership rate may be the result of several factors:high household income，inexpensive cars，tax-free gasoline and the absence of public transport coverage around the city.

In this paper，the aim is to model car ownership using the multinomial logit(MNL)model in a comparison of Riyadh，Saudi Arabia and Melbourne，Australia.The main contribution of this work is to find the reasons for the high levels of car ownership in two different cities:Riyadh in a developing country with limited public transport and Melbourne in developed country with good public transport.The remainder of this paper is organized as follows:part one provides a brief introduction to the MNL model for car ownership，and part two details the methodology used in this study，including the model structure，data collection and definition.Part three shows the results of the estimated models，and the final part concludes the paper with a summary of the findings.

1 Literature Review

Globally，most of people preferred using private car，because that it facilitates the movement easily any time with different weather condition.Reasons associated with an increase in the level of car ownership include，for example，socioeconomic and land use factors.These are the most important factors affecting travel patterns and increasing levels of car ownership.

Previously，a range of car ownership models have been applied to assess high levels of car ownership by households for various purposes(Chu 2002)［3］.The primary modeling approach to car ownership is best understood at two levels:the aggregate level，including regional，zones and national;and the disaggregate level of the household.Structurally speaking，the disaggregate level is more important，as it is more effective in gathering data on the causal correlation between car ownership determinants and car ownership levels than aggregate level analyses (Bhat and Pulugurta 1998)［1］.

Disaggregate models explain the probability of households owning zero，one or more cars.For maximum utility a household selects one of alternative specifications(0，1，2，3 or more).Chu(2002)［3］stated that individual traveller decisions on travel at the household level should be taken into account as the decision making level for car ownership，because household activities(e.g.work，shopping)affect household decisions on car ownership.

Disaggregate analysis has become the preferred approach for car ownership modeling，including MNL(Potoglou and Kanaroglou)［2］，the ordered response mechanism and the unordered response mechanism(Bhat and Pulugurta 1998)［1］.

In terms of explanatory variables，many researchers have studied the socioeconomic and land use characteristics of car ownership.Ryan and Han(1999)［6］and Li，et al(2010)［7］have pointed out that population and residential density have a negative effect on household decisions to own cars.Additionally，Chu(2002)［3］，Pologlou and Kanaroglou(2008)［2］found that car ownership decreases when mixed land use increases.Furthermore，Bento，et al(2005)［8］，Prevedouros and Schofer(1992)［9］and Li，et al(2010)［7］have demonstrated that households located close to central business districts(CBDs)have fewer cars than households located in outer areas，defined by the distance from the CBDs.Ryan and Han(1999)6］，Pologlou and Kanaroglou (2008)［2］， Prevedouros and Schofer(1992)［9］and Kim and Kim(2004)［10］have determined that transit access and services have a significantly negative effect on the probability of households owning more cars.A comparison study by Guiliano，et al(2006)［11］found that in the United Kingdom land use patterns in most areas decrease possible car use because of the existence of local shops and services in many neighborhoods in Britain compared to the U.S.which has disconnection between local shops and services and neighborhood areas.

On the other hand，socioeconomic and demographic characteristics that may affect the likelihood of household car ownership.Chu(2002)［3］，Pologlou and Kanaroglou(2008)［2］and Li，et al(2010)［7］have found that income and the number of licensed drivers increase the likelihood of households owning cars.In addition， Chu(2002)［3］and Li， et al(2010)［7］have determined that income and the variable of number of licensed drivers have more effect than household size and the number of workers on household decisions on car ownership.

2 Methodology

Discretechoice modelsused to reflectthe choices made by travellers from the number of alternatives which establishes the choice set response to availability.The methodology was used to identify and estimate the MNL model for auto ownership.These sections are divided into two parts，with the first representing the model structure，and the second describing the data definition and choice explanatory variables.

More recently，the MNL model has been applied to car ownership modeling.This model allows for the exploitation of the depth of household information that is available at a disaggregate level，while still ensuring compatibility with utility maximization.As such，household decisions to own cars are associated with the value of utility of choice among alternatives which have maximum utility(Bhat and Pulugurta 1998)［1］.

2.1 Model Structure

Where，Uijis the utility of individual j for alternative i;Vijis the deterministic part of the utility of the alternative i for individual j;εijis the random component of the utility of the alternative i for individual j.

Where，B is a vector to be estimated from data;Xijis the attributes of an alternative j and the socioeconomic characteristics of the decision maker i.

Equation(2)assumes that the exogenous variables in the car ownership model correlate with household attributes.An advantage of the MNL model is that it allows for the effect of exogenous variables，such as income，to be negative for the utility of zero car ownership，and positive for the ownership of one and two cars.

The distribution function of an independency and identically distributed(IID)error term(εij)，with a gumbel distribution，was assumed to identify the MNL model.Based on this assumption，the probability of a household choosing car ownership level J(Bhat and Pulugurta 1998)［1］is as follows:

Null log-likelihood is the log-likelihood function of the observed sample for an MNL model where all B vectors equal zero.The log-likelihood function is given as(Bierlaire 2008)［12］:

Where，n is the number of samples，n=1，2，3…，n;W is the weight of sample n;A is the number of alternatives associated with individual n.

The high values of rho-square(ρ2)and adjusted ρ2determine the performance of the model.The value of rho-square is closer to one it reflects the better fit of the model.To measure the strength and accuracy of fit ρ2can be tested as(Bierlaire 2008)［12］:

Where，the adjusted rho-square is:

Where，∂*is the log-likelihood for the final values of the parameters;∂0is the log-likelihood for the initial values of the parameters;K is the number of parameters estimated in the model.

2.2 Data Definitions

The explanatory variables used in the model were based on the data availability.We determined variables based on the socioeconomic attributes of households，demographic characteristics，land use and urban form(see Table 1).The Riyadh data were collected in 2008 and sourced from MOR［13］at the individual，household and journey-to-work levels.Socioeconomic and land use data were also obtained from MOR for Riyadh.

The Melbourne household and demographic and journey-to-work data were obtained from Vista 07(2007)［14］.However，the land use data were obtained from ABS(2006)［15］.

Table 1 Definition of explanatory variables表1 变量定义

3 Factors Affecting Car Ownership and Model Estimation

3.1 Socioeconomic and Demographic Attributes

Household components are divided into household type(single，couple with no children，couple with children，and single parent families)，and household income(low income，medium income，and high income).Household type variables were used because families with children are different to families with no children，and family size is the primary factor affecting household decisions to own a car(Chu 2002)［3］.In addition，in Riyadh and Melbourne，activity areas such as sport，entertainment and school do not connect with home location by public transport.Household size(number of members)was used to identify the correlation between car ownership and the number of cars owned corresponding to the number of household members.Income variables are different between Riyadh and Melbourne，but we converted the Saudi Riyal into Australian dollars(2012)using parity exchange rates.Low income is expected to affect household decisions on car ownership and number of trips made.In addition to the topology we used individual and personal data includes age，nationality and type of car license.Nationality is used for Riyadh to explain the effect of being a Saudi national or a non-Saudi national on household decisions to own a car，because non-Saudi nationals are laborers who represent 30 percent of the population(MOR 2008)［13］.The car license variable was categorized into four types in Melbourne(full license，probationary license，learners’permit，and no car license).We considered learners’permits and probationary licenses as full car licenses and no car license.

The employment variable was determined by full time work，part time work and casual time work.The age variable identified the age of a person as，for example，greater than 16 years of age.

3.2 Land Use Attributes

The second section is the land use and urban form characteristics.The household type variable was divid-ed into traditional，villa，floor villa，apartment and other in Riyadh.However，in Melbourne the household type was divided into separate，house，house/town and other.In both cities，the names of house types are different but they are quite similar in terms of position.Home type is very important because we can identify that houses with a garage or free parking increase the probability of a household decision to have a car，compared to the kind of accommodation available in highrise buildings，where households are less likely to own a car.

To explore the relationship between urban form and car ownership，we used two methods.The first was developed by Potoglou et al.(2008)［2］and Pervedours and Schofer(1992)［9］，which is the density of employment and population divided by area in square kilometers.The second was developed by Potoglou，et al.(2008)［2］and Chu(2002)［3］，which is:

Where，EDiis an employment density;PDiis a population density.

The MDIvariable explains self-containment which provides the short distance and decreases the likelihood of household car ownership because work and home are close together.The distance variable was divided into two parts:firstly，the distance between the origin and destination centroid point using MapInfo software，and secondly，the distance variable was used to measure the proximity of home location to work location.In Riyadh，we calculated the distance based on the origin and destination in traffic area zones(TAZs)，while in Melbourne in suburbs because Melbourne has no TAZs.We calculated the mean，minimum and maximum for all categories of the areas for both Riyadh and Melbourne to select compatible areas which were the TAZ areas for Riyadh and suburb for Melbourne.

We divided the two cities(Riyadh and Melbourne)into large zones which were categorized as the CBD，Inner and Outer areas［see Figures 1(a)and(b)］.Households living in the CBD are less likely to own a car，while households living in the outer areas are more likely to own more than one car.

Fig.1 Melbourne ＆ Riyadh’s car ownership density and road network with large zones(CBD，inner and outer)图1 利雅得和墨尔本家庭平均汽车拥有量分布情况

3.3 Model Estimation Result

Maximum likelihood is used for estimating the parameters of the closed-form GEV.With this method we tried to maximize the log-likelihood function.All estimations were performed using BIOGEME software ver.18(Bierlaire，2008)［12］.

The MNL model is estimated for car ownership.The utility specifications of four alternatives(0，1，2，3 or more cars)were estimated to obtain the best possible utility function，explaining the maximum variance in the data.

The model uses the personal，household and population，employment density index(MDI)for both Riyadh and Melbourne.Table 2 shows the final results of the estimated MNL model with t-statistic shown in brackets.All parameters should be found significant at levels greater than 95%.

Table 2 Multinomial logit estimated parameters of household car ownership表2 家庭汽车拥有量的多项式逻辑参数估计

We identified a full set of alternative specific constants(ASCs)for 1，2，and 4 alternative options.All alternative specific constants appear statistically significant.In Riyadh，a positive alternative specific constant for car ownership indicates households owning more than one car.It also indicates that there is insufficient alternative transport，such as public transport.However，in Melbourne，alternative specific constants were positive，except for alternative specific constant four which was a negative sign.These indicate that Melbourne households are expected to own 1 and 2 cars，with less expectation to own more than 2 cars.Also，Melbourne’s households have different mode options，such as public transport.

In Riyadh，we considered one parameter，Saudi and non-Saudi，because most non-Saudis do not own a car.The alternative specific coefficient of this parameter is not expected to own a car，however，a positive Saudi variable with two and three alternatives(ASCs)indicates that Saudi nationals are expected to own one or more than two cars.

In Melbourne，a positive apartment variable with alternative one indicates that those households living in apartments may not own a car because this type of house exists in and around the CBD area.In Riyadh，villa vs.house and separate house was a positive sign with alternative four，which indicates households living in villas may own more than 3 cars.

In Melbourne，casual work was a negative sign with alternative two，which means that casual workers are less likely to own one car because of their potential income constraints.

In relation to the car license variable，the number of households with driving licenses is positively affected and significant statistically.In Riyadh and Melbourne，the alternative specific coefficient of the car license parameter explains that it affects household utility，which is higher when owning more than one car.

Household size variables were used with different types of household size，such as single，couple with no children，couple with children and single-parent households for both Riyadh and Melbourne，but these variables were not significant in Riyadh compared to Melbourne.A single person with no children with alternative one was positively affected by household utility，and is not expected to own a car.We used the person variable(the total members for each household)due to the number of cars being controlled by the household members.However，in Riyadh，the person variable was not significant.In Melbourne，the person variable was positive with alternative four.Their effect on household utility is that they are expected to own more than two cars.

Household income variables are used because it is an important variable associated with owning a car.It was divided into three categories(low，medium and high)，and the Riyal was converted into Australian dollars(2012)using parity exchange rates.In Riyadh，medium income had a positive effect on household decisions to own more than two cars.Also，low income had a positive sign with alternative one.However，in Melbourne，high income had a positive effect on household decisions to own two and more than three cars.The high and medium income variables indicate that households may buy expensive cars and more than one car.Pervedours and Schofar(1998)［9］stated that high household income is associated with buying an expensive car，rather than a greater number of them.

Mixed density index(MDI)was used to test the density of employment and residence within an area.MDI means that the probability of a household to own one or more cars decreases.In Riyadh，MDI has not been used because it had a weak positive effect on household decisions to own a car，and they had less expectation to own a car.In Melbourne，the MDI variable also had a positive effect on household decisions to own a car.Centre business district(CBD)was used to examine the importance of residents living close to their works and activities，which is expected non-car ownership.In Riyadh and Melbourne，have high indicate that non-car ownership，however，in Melbourne，had a positive sign to own one car.

4 Conclusions

This paper has studied the influence of land use and socioeconomic factors on travel behavior in Riyadh and Melbourne.The main method was used to conduct the modelling car ownership to estimate the MNL model.The data were based on the disaggregation(household level)，aggregation level findings from MOR(2008)［13］for Riyadh and Vista 07(2007)［14］for Melbourne.The results of the MNL model help us to identify the effect of socioeconomic and land use parameters on household decisions to own 0，1，2 and 3 or more cars.The results are compatible with the current literature reviewed in Part 2.We found that proximity of work place to home place by a distance greater than 6 kilometers does not result in household decisions to own cars.The CBD variable was a strong positive effect on household decisions to walk or bicycle rather than own a car.In Melbourne，the disconnection between work or activity location and home increases the probability of households owning one car.

Melbourne and Riyadh are similar to most other developed cities with a heavy dependence on cars and a decentralized structure.In particular，low density produces the highest proportion of car trips.High and medium income has a positive effect on household decisions to own more than three cars in both cities.This result is compatible with Chu(2002)［3］，Pologlou and Kanaroglou(2008)［2］and Li，et al(2010)［7］，who found that income and the number of licensed drivers variables increased the likelihood for household to own cars.

The MNL model can readily be used to identify the effect of socioeconomic and land use characteristics on the utility of household decisions to own car.In Riyadh and Melbourne，the CBD variable is the highest indicator of non-ownership of cars because of the balance between land use developments.Urban sprawl is the main factor that affects household decisions to own more than two cars in Riyadh and one car in Melbourne.Non-motorized modes and mass transit use should be promoted［Jieping，et al(2010)］［7］.

The outcomes of the study provide useful implica-tion for Riyadh and Melbourne，which represented developing and developed countries.Both cities have shifted from monocentric areas with high population and employment densities to sprawling areas with low population densities and scattered employment.A polycentric structure with a good public transport system and high employment and population density can be obtained using a sustainable urban plan(self-containment)，as this reduces car ownership and encourages sustainable transport(walking and bicycling).

［1］ Bhat C R，Pulugurt V.A comparison of two alternative behavioural choice mechanisms for household auto ownership decisions［J］.Transportation Research Part B:Methodological，1998，32(1):61－75.

［2］ Potoglou D，Kanaroglou P S.Modelling car ownership in urban areas:a case study of hamilton in canada［J］.Journal of Transport Geography，2008，16(1):42－54.

［3］ Chu Y L.Automobile ownership analysis using ordered probit models［J］.Transportation Research Record，2002(1805):60 －67.

［4］ Currie G，Senbergs Z.Exploring Forced Car ownership in Metropolitan Melbourne［C］//30th Australian Transport Research Forum 2007.Melbourne，Victoria，Australia:ETM Group，2007.

［5］ Koushki P A.The effect of socio-economic development on household travel behaviour in Saudia Arabia［J］.Socio-economic Planning Sciences，1988，22(3):131 －136.

［6］ Ryan J M，Han G.Vechicle-ownership model using family atructure and accessbality application to Honolulu，Hawaii［J］.Transportation Research Record，1999(1676):1－10.

［7］ Li Jieping，Walker J L，Srinivasan S，et al.Modelling private car ownership in china:investigating of urban form impact across megacities［J］.Transportation Research Record，2010(2193):76－84.

［8］ Bento A M，Cropper M A，Moborak A M，et al.The impact of urban spatial structure on travel demand in the United State［J］.Reivew of Economics and Statistics，2005，87(3):466 －478.

［9］ Prevedouros P D，Schofar J L.Factors affecting automobile ownership and use［J］.Transportation Research Record，1992(1364):152－160.

［10］ Kim H S，Kim E.Effects of public transit on automobile ownership and use in households of the USA［J］.Review of Urban＆ Regional Development Studies，2004，16(3):245－262.

［11］ Giuliano G，Dargay J.Car ownership travel and land use:A comparison of the US and Great Britain［J］.Transportation Research Part A:Policy and Practice，2006，40(2):106 －124.

［12］ Bierlaire M.Estimation of Discrete Choice Models with BIOGEME 1.8［M/OL］.Ascona，Switzerland:Ecole Polytechnique Fédérale de Lausanne，Transport and Mobility Laboratory，2009［2012 －03－ 21］.http://biogeme.epfl.ch/.

［13］ MOR.Demographic，Socioeconomic and Trip Data［R］.Riyadh，Saudi Arabia:Municipality of Riyadh，2008.

［14］ Vista 07.Victorian Integrated Survey of Travel and Activity［G］.Victoria，Australia:State Government of Victoria，Department of Transport，2007.

［15］ ABS.Census of Population and Employment［R］.VIC:Australia Bureau of Statistics，2006.

［16］ Hess D B，Ong P M.Traditional Neighborhood and Automobile Ownership［J］.Transportation Research Record，2002(1805):35－44.

［17］ Lerman S T，Ben-Akiva M.Disaggregate behavioral model of automobile ownership［J］.Transportation Research Record，1976(369):34－51.