STUDY CASE

Study Region

Beijing lies on the northeastern margin of the North China Plain.The spatial organization of this city is characterized by the continu-ous addition of ring roads. Today Beijing has five rings completedfor public use. The first ring circles the Forbidden City (the palaceof the emperors). Because the area has no dense residential areas andfew employees, the ring does not carry heavy regional traffic, andhas remained local. All other ring roads were added to either miti-gate existing traffic congestion or to open suburban land to urbandevelopment. Construction of the second ring began in the 1950sand then completely changed and became an urban expressway in1992. The third ring road was completed in 1994. The fourth andfifth ring roads were added in 2001 and 2003, respectively. Figure 1presents the locations of the five ring roads.

Today Beijing hosts more than 10 million residents at its corebuilt-up area, which is served by the five ring roads and other con-necting roads. The second ring roughly delineates the boundary ofthe “old Beijing,” developed before the communist party took powerin 1949. Development in the 1980s mainly occurred around or withinthe third ring and seldom went beyond the fourth ring. Since the1990s, the fourth and fifth rings have enabled conversion of rural andsemirural areas into urban landscapes, stimulated by a strong demandfor housing and office space.

The expansion of the city has increased trip length as travelerslook for employment and entertainment opportunities throughoutthe ever-enlarging urban region. Relocation from the inner city outinto what were once semirural areas has also increased the spatialseparation between workplaces and residences (7 ). Large volumesof traffic increase road congestion as the capacity limit of existingtransport facilities is exceeded. It has been reported that bus speedis about 10 km/h during peak periods.

Urban rail (surface or subway) in Beijing provides an alterative forthose who want reliable mobility but cannot afford a car. The sub-way system in Beijing opened to the public in 1981 (8). At that time,the length of the system was 27.6 km, and it included only 19 stops.By 2008, this system had eight lines, more than 200 km of tracks, and123 stations. At least seven more lines, with 164 km of track length,are now under construction. Figure 1 shows three rail lines. The otherfive are omitted because they were completed later and are notdirectly relevant to the transportation pattern described in this article.

Data and Method

Research was enabled by the availability of a key household surveyconducted in 2006. The survey’s data were collected by John Logan

Spatial and Social Characteristics of Urban Transportation in Beijing

Jiawen Yang

59

This paper aims to improve knowledge of urban transportation in Chinaby analyzing the spatial characteristics of urban transportation in Beijing,China’s capital. Neighborhood-level indicators—such as automobile own-ership per household, commuting time, commuting mode percentages,and household income—were all extracted from a household survey con-ducted in 2006. Urban transportation was then analyzed spatially with areference to Beijing’s geometric center, job centers, and urban rail sta-tions. The analysis led to several findings. First, the variation in commut-ing time by gender and household composition was similar to what hasbeen observed in American cities. Second, automobile ownership andcommuting modes exhibited no systematic, spatial variation, unlike theirAmerican counterparts. These differences might be explained by a thirdfinding: in China, relatively high-income households preferred centrallylocated housing.

Urban transportation in China has undergone dramatic transition. Anincrease in automobile ownership, an increase in reliance on motor-ization for everyday trips, and an increase in congestion and air pol-lution in China’s megacities are well-discussed topics. A recent studysuggests that China is following the motorization path of developedeconomies (1). The increase in automobile ownership in relation tothe growth in China’s gross domestic product is similar to that in theUnited States, South Korea, Japan, and Germany.

As the macro trend becomes increasingly clear, researchers havebegun to pay attention to the social details of China’s urban trans-portation (2). Little, however, is known about the spatial details ofChina’s urban transportation, partially because of the lack of rele-vant data. How does automobile ownership vary from the centralcity to suburban areas? How does transportation mode vary amongdifferent neighborhoods? How does commuting time vary amongdifferent social groups? Is it similar to what has been observed inWestern cities? A better understanding of these spatial and socialcharacteristics of urban transportation in China can help assess its current transportation and urban development policies (3), par-ticularly since Chinese cities are still experiencing significant spa-tial transformations (4), which could have significant impacts onmobility and accessibility (5, 6). Toward this end, this article pre-sents a spatial analysis of urban transportation in Beijing, China’scapital.

Program of City and Regional Planning, Architecture Building East, Room 204,Georgia Institute of Technology, Atlanta, GA 30332-0155. jiawen.yang@gatech.edu.

Transportation Research Record: Journal of the Transportation Research Board,No. 2193, Transportation Research Board of the National Academies, Washington,D.C., 2010, pp. 59–67.DOI: 10.3141/2193-08

and his research team at Brown University (9). The survey was con-ducted to understand the dynamics of housing markets, householdlocation, and travel behavior. Coded survey results recorded detailedinformation on vehicle ownership, commuting duration and modes,residential location, and workplace location. The 2006 survey covered48 urban neighborhoods, with 25 households surveyed in each neigh-borhood. These neighborhoods were in the central city, the inner sub-urbs, and the outskirts beyond the fifth ring road. Figure 2 shows thelocation of the surveyed neighborhoods; the size of each dot representsthe average household income in each neighborhood. The survey datawere used to develop a group of neighborhood-level indicators. Analy-sis of these indicators can show how car ownership and individualtravel behavior vary from one neighborhood to the other.

Being cognizant of significant suburban development and urbanrail investment in Beijing, the spatial characteristics of urban trans-portation were analyzed from multiple perspectives. Three dis-tance variables were introduced to define the location of eachsurveyed neighborhood: distance to Beijing’s geometric center, dis-tance to the closest job centers, and distance to the closest subwaystation.

Tian’anmen Square, which has always been viewed as the civiccenter of Beijing, was selected as the geometric center for this study.A geographic information system (GIS) was used to calculate thedistance from every surveyed neighborhood to this geometric cen-ter. Examination of the association between urban transportationindicators and the distance to the geometric center revealed to whatextent a monocentric model can help interpret urban transportationpatterns.

60 Transportation Research Record 2193

To test whether the formation of job centers in the suburbsaffects urban transportation, the top 10 subdistricts for job densitywere selected. Figure 3 shows the location of these top 10. Jobcounts in each subdistrict originated from the 2002 work unit cen-sus. The selected job centers were generally located on the second,third, and fourth ring roads. The selected subdistricts includedwell-known job centers, such as Zhongguancun and Jianguomeng.GIS was used to calculate the distance from every surveyed neigh-borhood to the centroid of the nearest job center. Examination ofhow urban transportation indicators varied according to this dis-tance revealed to what extent a multicenter perspective could helpexplain Beijing’s transportation patterns.

The distance was calculated from each surveyed neighborhoodto the closest urban rail station. Examination of how transportationindicators varied according to this distance revealed to what extentaccess to high-end transit might affect the spatial patterns of urbantransportation.

Correlation of population and job density with the three distancevariables indicated that the above distance variables provided reason-able perspectives to interpret the urban spatial structure. As shown inTable 1, all three categories of distance had significant impacts on spa-tial distribution of residences and workplaces: the shorter the distances,the higher the density. A lower distance value, therefore, indicated arelatively central location. Comparable correlation coefficients for theAtlanta, Georgia, metropolitan statistical area in the United States sug-gest a similar pattern (Table 2). Both tables include correlation coeffi-cients for mode split and commuting time, which are explained later inthis paper.

FIGURE 1 Major transportation infrastructure in study region.

Yang 61

FIGURE 2 Location and average household income of surveyed neighborhoods.

FIGURE 3 Ten subdistricts of highest job density.

62 Transportation Research Record 2193

for automobile ownership (11). In Chinese cities, however, it hasnever been clear whether neighborhoods with high automobile own-ership are clustered in a certain urban area. In Beijing, a visual exam-ination of the surveyed neighborhoods suggests a significant spatialvariation of household vehicle ownership among different neighbor-hoods (Figure 5). No systematic spatial variation can be observed,however, and it is difficult to say whether it follows a certain spatialpattern. In a correlation of vehicle ownership with the three distancevariables (Table 1), none of the variables can significantly explain thespatial variation of household vehicle ownership.

This spatial characteristic could be easily misinterpreted. It mightbe assumed that household automobile purchasing decisions are notaffected by land use intensity, which decreases from a central locationto the periphery (Table 1). The spatial pattern, however, might becaused by the spatial characteristics of other relevant factors such ashousehold income. To test this idea, a regression model was devel-oped of vehicle ownership per household. The model used three inde-pendent variables. The first two were average household income (inthousands) of the surveyed neighborhood and the population density(1,000/sq km) of the subdistrict where the neighborhood was located.Because the spatial relationship between workplace and residence

TABLE 1 Correlation Coefficients with Three Distance Variables

Distance

To To UrbanTo Job Geometric Rail

Variable Centers Center Station

DensityPopulation density 1982 −.523** −.637** −.521**Population density 1990 −.639** −.723** −.616**Population density 2000 −.678** −.706** −.633**Job density 2001 −.579** −.599** −.532**

IncomeAverage household income −.346* −.260 −.250

Vehicle ownershipAverage number of motor .188 .128 .060

vehicles per household

Mode splitPercentage of walking .116 .218 −.091Percentage of biking .130 .073 .314*Percentage of transit −.278 −.246 −.332*Percentage of driving −.049 −.031 −.022

DurationCommuting time −.145 −.318* −.124

**Correlation is significant at the .01 level (two-tailed).*Correlation is significant at the .05 level (two-tailed).

TABLE 2 Comparable Correlation Coefficients in the AtlantaMetropolitan Area

Distance

To Job To Geometric To UrbanVariable Centers Center Rail Station

DensityPopulation density −.557** −.510** −.523**Job density −.211** −.236** −.196**

Mode splitDriving alone .453** .575** .497**Carpool −.070 −.112** −.084*Transit −.438** −.595** −.503**Nonmotor −.260** −.323** −.254**

DurationCommuting time .221** .143** .183**

NOTE: Data are from the 2000 U.S. census. The analysis units are census tracts.The geometric center is the census tract with the highest job density indowntown. The job centers include all census tracts with job density higherthan 2,000 jobs per square kilometer.**Correlation is significant at the .01 level (two-tailed).*Correlation is significant at the .05 level (two-tailed).

0

0.5

1

1.5

2

2.5

3

3.5

1940 1950 1960 1970 1980 1990 20102000

Mill

ion

s

Year

FIGURE 4 Registered motor vehicles in Beijing.

AUTOMOBILE OWNERSHIP

The increase in motor vehicles has been the most dynamic aspect ofChina’s urban transportation development. Figure 4 illustrates thegrowth of registered motor vehicles in Beijing. Two distinct phasescharacterize Beijing’s motorization. The pre-1980 period saw almostno increase, even though the city’s urban area had expanded, andurban residents had tripled from 1949. After the mid-1990s, motorvehicles increased at an annual rate of more than 10%. This fastgrowth of automobile ownership can be attributed to many factors. Asin other developing countries, economic development and incomegrowth were influential factors (10). Changing travel demands as aresult of land and housing reforms also played a significant role, asYang reported in 2006. According to 2006 survey data, the averagehousehold owned 0.23 cars that year.

In the United States, it is widely known that automobile ownershipis lower in central cities, because there is access to transit in high-density areas, and a mixed-use, built-environment reduces the demand

was an important factor that affected travel demand (12), the thirdvariable was controlled, namely, the percentage of households thatlived and worked in the same subdistrict. The analysis units were the48 surveyed neighborhoods.

The regression results were no surprise. Higher ownership of motorvehicles is significantly associated with higher household income,lower population density, and reduced spatial balance between work-place and residence. Further examination of the spatial variation ofthe population density and household income can help explain whyautomobile ownership does not show a systematic spatial variation.Population density tends to be higher at central locations (Table 1),which reduces automobile ownership there. Household income, how-ever, tends to increase automobile ownership in central locationsbecause higher income households are more likely to live there, assuggested by the negative correlation coefficient in Table 1. Income

cars household income

2.673

= +

( )0 227 0 005

3 43

. .

.

�

99

0 0059

2 593

0 15

( )−

−( )−

.

.

.

� population_density

44

1 967

2492

� balance_percentage

−( )=

.

.R

Yang 63

and density, therefore, work in opposite directions to shape thespatial pattern of car ownership.

It is widely known that high-income Americans are more likely tolive in the suburbs. Why are relatively high-income Chinese house-holds more likely to live in central locations? The question is beyondthe scope of this paper, but a likely answer has to do with the spatialvariation of the quality of public goods. In China, central locationsare generally equipped with better schools, better access to shoppingand entertainment opportunities, and better urban infrastructure. Thesingle-family house in the outer suburbs is attractive in terms ofhousing characteristics. Its distance and the associated road conges-tion that separate it from the city, however, have disqualified the sub-urban house as a primary residence for high-income households,even for typical automobile owners.

COMMUTING MODE

Motor vehicles are not owned only as a matter of pride but also to usefor everyday trips. The 2006 survey showed a dramatic difference incommuting modes between car owners and nonowners (Figure 6).Households with cars used them to make 48% of their trips. Com-pared with households without cars, the car owners were less likelyto commute by walking, biking, and transit riding. Commuting byurban rail, however, was actually higher among car owners thannonowners, which indicated that car owners could substitute drivingtrips with urban rail.

FIGURE 5 Average household motor vehicle ownership in surveyed neighborhoods.

64 Transportation Research Record 2193

0%

10%

20%

30%

40%

50%

60%

walk bike fixed route bus employer bus car or taxi subway

car owners

non-owners

FIGURE 6 Commuting mode split for car owners and nonowners.

0%

10%

20%

30%

40%

50%

60%

walk bike bus exclusivemotorized modes

others

1996

2006

FIGURE 7 Commuting mode split in Beijing: 1996 and 2006; exclusive motorized modes includeprivate auto, taxi, and motorcycle.

A comparison of the 2006 survey with a similar survey conductedin 1996 (12) indicated that the overall commuting mode pattern hadchanged significantly in the context of motorization (Figure 7). From1996 to 2006, the biking share had decreased more than 10%, partiallybecause of cyclists’ decreased road safety and exposure to pollutedair. Despite the decrease, however, biking remained the primary modeof commuting and accounted for 43% of commuting trips. Driving(including taxi) as a share of the commuting mode was about 10%,much lower than the corresponding numbers in American urbanregions such as Atlanta (95%) and Boston, Massachusetts (84%).(The Boston and Atlanta percentages were calculated on the basis ofthe 2006 Public Use Microdata Samples (PUMS) data set.)

A unique and interesting feature of Beijing’s urban transport is thatwalking is the third primary mode of commuting, with a mode shareof 15%. Its increase from 1996, according to Yang et al., was relatedto the housing location preference of a group of relatively high-income households (13). They chose to live close to their workplacesand to walk to their offices.

Similar to what has already been observed about automobile own-ership, no spatial variation of commuting modes was obvious. Anyobvious spatial pattern of mode split could hardly be detected fromFigure 8. The correlation analysis (Table 1) can confirm this. Theonly two significant coefficients were associated with the distanceto the closest urban rail stations: the closer to the stations, the higher

the percentage by transit and the lower the percentage of biking.This association makes sense, because better access to urban rail ser-vices tends to reduce the chance of commuting by bike and increasethe chance of commuting by transit. Again, this makes Beijing’s urbantransportation different from what has been observed in Atlanta, wherecommuting mode split varied systematically according to distances tocentral locations (Table 2).

No American-style, suburb–central city division exists in Beijing.Given Beijing’s urban development history, the study area was dividedinto two parts—those within the third ring and those beyond. Sincethose residences beyond the third ring were mainly developed duringthe motorization period, this area was treated as equivalent to the sub-urbs in a U.S. metropolitan area. Delineation of the central city fromthe suburbs was different from that of existing studies of populationsuburbanization in Beijing (14), which have treated areas beyond thesecond ring as suburbs. This difference was justified by the purpose ofthe research, which was to study the spatial characteristics of urbantransportation. To delineate the city–suburb boundary on the basis ofthe stage of motorization made sense in this context. In addition, thegrouping split the surveyed neighborhoods almost in half, whichallowed for the largest possible sample size in each subset.

The null hypothesis was then tested that commuting mode selec-tion was statistically the same between the central city and the sub-urbs. The statistics testing did not reject the null hypothesis. The test

result was definitive. A further test was conducted to see whetherneighborhoods that had high or low percentages of motorized tripsclustered together. A spatial analysis conducted with the Moran Index(use of an inverse distance neighborhood) showed the percentage ofdriving was –0.06, or close to random. The percentage of transit rideswas 0.04, which was also likely to be random. The test for the aver-age number of vehicles per household was 0.03, which had a 5%chance of being random.

The same tests were run with Atlanta’s 2000 census data, by usingcensus tracts as the analysis units. The null hypothesis could berejected at a 99% confidence level. That is to say, a statistical differ-ence between driving and transit use by commuters existed betweenthose that resided in the city of Atlanta and those that resided outsideit. The Moran index (use of inverse distance for neighborhoods) fortransit use was 0.35%, with less than a 1% chance of randomness.Similarly, the index for driving alone was 0.31%, with a less than 1%chance of randomness. The above statistical testing suggests no sig-nificant differences between urban transportation among those thatreside in the central part of the city and those that reside in the sur-rounding suburbs. This spatial characteristic makes Beijing differentfrom its Western counterpart.

COMMUTING TIME

Commuting time, which is a measure of access to the workplace, hasalways been an important measure of urban transportation. Table 3presents average commuting time for different social groups. The2006 Beijing survey data can be compared with the 1996 survey and

Yang 65

FIGURE 8 Commuting mode split in surveyed neighborhoods.

TABLE 3 Average Commuting Time by Social Groupsand Mode

Commuting Time (min)

BeijingAtlanta Boston

Group 1996 2006 2006 2006

Average 39.0 36.5 31.5 28.2

Housing ownershipTenant 38.4 35.7 30.8 28.4Owner 39.1 38.7 31.1 27.6

GenderFemale 40.0 35.7 29.7 26.2Male 38.0 37.1 33.1 30.0

College educationNo 38.5 34.6 31.2 25.3Yes or above 40.0 39.0 32.0 31.4

Household compositionOne worker 40.1 38.9 32.8 29.7Biworker 37.3 35.1 31.7 29.2Multiworker 44.6 38.5 29.8 25.2

ModeWalk 13.0 14.6 12.8 12.3Bike 31.1 26.3 19.8 24.5Fixed-route bus 63.9 57.8 56.9 42.4Employer bus 42.4 54.6Driving 33.9 33.1 30.7 26.5Urban rail 50.3 56.7 47.6

NOTE: Atlanta and Boston indicators are extracted from PUMS, U.S. CensusBureau.

with the 2006 Boston and Atlanta PUMS data. Several things can beobserved.

First, the gender difference in average commuting time for Beijingreversed itself from 1996 to 2006, which made it similar to Atlantaand Boston. In 1996, the commuting duration of female workers was2 min longer than for male workers. In 2006, however, the averagecommuting time for male workers was 1.4 min longer than for femaleworkers. This change can be probably explained by the change inhousing location selection. Before 1996, the housing market was notwell developed, and employers still played an important role in hous-ing subsidy and location selection. If a household had a working hus-band and a working wife, the husband was more likely to have ahigher status than the wife and therefore was more likely to receivehousing support from his work unit. The couple’s housing locationthus was typically closer to the husband’s workplace than the wife’s,which led to a shorter commuting time for the husband. As marketpower has penetrated, however, residential selection and housingconsumption have come to depend more on the housing market andhousehold income. Housing location is now more likely to be closerto the wife’s workplace because of her relatively lower physicalstrength and her larger share of responsibilities for childcare andhousework.

In a market economy, housing location decisions become increas-ingly separated from the collective decision of the work unit. House-holds have gained more autonomy in residence selection. Under thisnew circumstance, couples are more likely to select a residence con-venient for the wife so that she can have more time and energy tocook, clean and take care of children. In contrast, the husband isexpected to make a longer commute, given his relatively greaterphysical strength and lesser involvement in childcare and housework.

Second, the number of workers in a household also has effects oncommuting time. A dual-worker household tends to have a shortercommuting time than a single-worker household. Those householdswith three or more workers, however, commute longer, which indi-cates a difficulty in balancing home and work location. The Atlantaand Boston data indicate that the more workers in a household, theshorter the commuting time. The result appears to challenge the tradi-tional assumption that workers from dual-worker households find itmore difficult to achieve a balance between workplace and residenceand thus tend to commute longer distances than those in single-workerhouseholds. Sultana once explained this contradiction with housingaffordability (15). Even though multiworker households tend to havemore difficulty than single-worker households in the achievement of abalance between workplace and residence, the lack of affordable hous-ing could overshadow this difficulty. The more workers a householdhas, the higher the income it can earn, and the more likely it is that itsmembers can afford a better location to live. Given Beijing’s highhousing price relative to income, it is no surprise that this affordabilityproblem is reflected in commuting time.

Third, commuting time in Beijing according to modes has becomesimilar to its American counterparts. The two primary commutingmodes in Beijing are biking and fixed route bus. Average commutingtime by bike decreased from 31.1 min in 1996 to 26.3 min in 2006,which made it more comparable to that in Boston. The shortening ofbiking commuting time could be caused by the replacement of longbiking trips with motorized trips. The average commuting time byfixed-route bus decreased from 63.9 min to 57.8 min, which was closeto the 56.9 min commuting duration in Atlanta.

Despite the observable increase in congestion in Beijing, the surveydata do not reveal any increase in average commuting time. This couldbe caused by the imperfection of the data. The size of the sample

66 Transportation Research Record 2193

may be too small; however, it could also reflect the actual situation.Households could respond to congestion by switching to higher-mobility options, such as driving. In addition, they can select to livecloser to the workplace (13). Furthermore, the continuous addition ofthe ring roads, particularly the fourth and fifth rings, has helped toreduce congestion. As suggested by the correlation coefficients ofcommuting time (Table 1), the commute from a suburban location onaverage tends to be shorter than it is from a central location. This couldbe a Chinese version of the American commuting paradox (16). It alsoconfirms an observation made by Yang (17): the effect of decentral-ized development on commuting depends on the format of suburbandevelopment. Beijing’s high density and clustered development insuburban areas might have helped to reduce commuting time.

Fourth, incremental urbanization has not necessarily lengthenedcommuting time in Beijing. In American metropolitan areas, a closeassociation exists between commute duration and age of neighbor-hood. The newer the neighborhood, the further away it is from thecenter of the city, and the longer the commute (18). In Beijing,neighborhoods developed in the 1970s and thereafter have a reversedpattern of commuting duration: the newer the neighborhood, theshorter the commute time (Table 4). As for mode split, Table 5 indi-cates that the newer the neighborhood, the higher the percentage ofnonmotorized modes (including walking and biking).

Beijing’s difference can be partially attributed to the spatial char-acteristics of its new housing construction. In the United States, rede-velopment and infill development in established urban areas aregenerally limited, compared with opportunities for new developmentin suburban areas. Therefore, new neighborhoods are more likelyto be located in peripheral areas and require longer commutes and ahigher percentage of driving. In China, significant redevelopmentoccurs in the central part of the city. Consequently, a relative higherpercentage of new neighborhoods are actually developed in the cen-tral city. Higher density and mixed use there enable a higher percent-age use of nonmotorized modes and create the convenience of beingable to reach destinations within a short travel time. As for new neigh-borhoods in suburban Beijing, the addition of ring roads enables those

TABLE 4 Average Commuting Time by Age of Neighborhood

Beijing Atlanta Boston2006 2006 2006

Post-2000 29.4 35.5 32.0

1990s 33.1 32.1 30.1

1980s 39.4 31.2 28.9

1970s 39.3 29.4 29.0

NOTE: Atlanta and Boston indicators are extractedfrom PUMS, U.S. Census Bureau. Commuting timeis in minutes.

TABLE 5 Mode Split by Age of Neighborhood in Beijing

Age of Car or Nonmotorized WalkingNeighborhood Taxi (%) (%) (%)

Post-2000 10.0 58.0 18.0

1990s 18.5 49.7 13.0

1980s 13.8 47.4 10.3

Pre-1980 11.6 53.9 8.9

that rely on motorized modes to reach their destinations fast, despitethe possibility of a long commuting distance.

CONCLUSION

Since the late 1980s, land and housing reforms have reshaped urbanspace in China (4, 6). Economic growth has increased motorization.This article presents an updated view of the spatial and social detailsof China’s urban transportation, which stems from the existingknowledge base. Information extracted from a household surveyindicates that Beijing is becoming similar to its Western counter-parts in terms of driving behavior by automobile owners, and accessto workplaces according to different commuting modes and differ-ent social groups. This trend toward similar outcomes is no surprise,as individuals make decisions on vehicle ownership, motorizedtravel, and residential location selection on the basis of a similar setof constraints and objectives.

Despite the trend, China’s urban transportation shows distinctspatial characteristics. The survey data reveal no American style, city–suburb division of urban transportation. They show no significant norsystematic variation in vehicle ownership and commuting modes bydistance to central locations. The regression model of automobileownership suggests that residence location preference of high-incomehouseholds could help explain Beijing’s spatial characteristics ofurban transportation. Relatively high-income households are morelikely to live in central locations, which increase automobile owner-ship there. It is still unknown if this spatial difference will last long. Itcould disappear as China continues to change its planning practicesfor land development, urban transportation, and public goods supply.

ACKNOWLEDGMENTS

This research was supported by the Lincoln Institute of Land Pol-icy. The author thanks John Logan for the survey data, and Ge Songfor valuable research assistance.

REFERENCES

1. Schipper, L., and W. S. Ng. Urban Transport Options in China: TheChallenge to Choose. EMBARQ. World Resources Institute, Center forTransport and Environment. http://www.embarq.org/en/urban-transport-options-china-the-challenge-choose.

Yang 67

2. Zegras, P. C., and S. Srinivasan. Household Income, Travel Behavior,Location, and Accessibility: Sketches from Two Different DevelopingContexts. In Transportation Research Record: Journal of the Trans-portation Research Board, No. 2038, Transportation Research Board ofthe National Academies, Washington, D.C., 2007, pp. 128–138.

3. Gakenheimer, R. Social Effects of Potential Motorization Pathways. InPersonal Cars and China (Chinese Academy of Engineering and NationalResearch Council of the National Academies) National Academies Press,Washington, D.C., 2003, pp. 131–149.

4. Zhou, Y. X., and L. Ma. Economic Restructuring and Suburbanizationin China. Urban Geography, Vol. 21, No. 3, 2000, pp. 205–236.

5. Shen, Q. Urban Transportation Problems in Shanghai, China: Problemsand Planning Implications. International Journal of Urban and RegionalResearch, Vol. 21, No. 4, 1997, pp. 589–606.

6. Yang, J. W., and R. Gakenheimer. Assessing the Mobility and Accessibil-ity Consequences of Land Use Transformation in Urban China. HabitatInternational, Vol. 31, 2007, pp. 345–353.

7. Yang, J. Transportation Implications of Land Development in a Transi-tional Economy: Evidence from Housing Relocation in Beijing. In Trans-portation Research Record: Journal of the Transportation ResearchBoard, No. 1954, Transportation Research Board of the National Acade-mies, Washington, D.C., 2006, pp. 7–14.

8. Major Events for Beijing Subway: 1981–1990. http://www.bjsubway.com/zjdt/dtdsj/2007-05-15/2976.shtml. Accessed July 1, 2009.

9. Urban China Research Network, Multi-City Study of Urban China.http://www.s4.brown.edu/ChinaProject/.

10. Gakenheimer, R. Urban Mobility in the Developing World. Transporta-tion Research Part A, Vol. 33, 1999, pp. 671–689.

11. Newman, P., and J. Kenworthy. Sustainability and Cities: OvercomingAuto Dependence. Island Press, Washington, D.C., 1999.

12. Li, S. M. Life Course and Residential Mobility in Beijing, China. Envi-ronment and Planning A, Vol. 36, No. 1, 2004, pp. 27–43.

13. Yang, J. W., J. Feng, and R. Gakenheimer. The Physical and SocialDimension of Job-Housing Balance in Urban China. In Smart Growthin China, Chapter 11. Lincoln Institute of Land Policy Press, Cambridge,Mass., 2009.

14. Zhou Y. X. On the Suburbanization of Beijing. Chinese GeographicalScience, Vol. 7, No. 3, 1997, pp. 208–219.

15. Sultana, S. Effects of Married-Couple Dual-Earner Households on Met-ropolitan Commuting: Evidence from the Atlanta Metropolitan Area.Vol. 26, No. 4, 2005, pp. 328–352.

16. Gordon, P., H. Richardson, and M.-J. Jun. The Commuting Paradox:Evidence from the Top Twenty. Journal of the American PlanningAssociation, Vol. 57, 1991, pp. 416–420.

17. Yang, J. W. The Commuting Impacts of Spatial Decentralization: AComparison of Boston and Atlanta. Journal of Regional Analysis andPolicy, Vol. 35, 2006, pp. 69–78.

18. Sultana, S., and J. Weber. Journey-to-Work Patterns in the Age of Sprawl:Evidence from Two Midsize Southern Metropolitan Areas. The Profes-sional Geographer, Vol. 59, No. 2, 2006, pp. 193–208.

The Transportation in the Developing Countries Committee peer-reviewed thispaper.