The Built Environment, Car Ownership, and Travel Behavior in Seoul Sang-Kyu Cho, Ph D. Candidate So-Ra Baek, Master Course Student Seoul National University Abstract Although the idea of integrating land use and transportation policy has gained popularity among many urban planners and designers, the studies into land use and travel behavior in high-density Asian cities have been scarce. This study presents empirical evidences of the association between land use and travel behavior in Seoul considering various dimension of urban space. The results of analysis showed significant association between the built environment, car ownership and daily car time in Seoul. Introduction Recently, many urban planning and design strategies have sought a normative basis from the relationship between built environment and travel behavior. For instance, the popular land use policies including the transitoriented development (TOD), the neo-traditional development (TND), the compact city and the health city are supported for their potential association with more desirable trip demand characteristics. A number of empirical studies including Cervero and Kockelman(1997), Handy(1996) and McNally and Kulkami(1997) have argued that denser and more mixed land use pattern would reduce automobile use, thus increasing in walking trips. The argument might be relevant in the low density suburban area in the United States; however, the effects of these strategies in a high density metropolis have received relatively less attention from academia. Although Seoul has been developed as a high-density metropolis, it has been also experiencing the rapid growth of automobile use which has led to the various problems: congestion, air-pollution and decrease in transit use. Obviously, the increase in automobile traffic might be attributed largely to the increase in household income along with the growth of the national - 1 -
economics. However, the influence of land use factors in Seoul needs more researches. This paper presents empirical findings of the association between land use and travel behavior in Seoul using archival data of the Seoul Metropolitan Government. The findings of this paper would provide better understanding of land use and travel behavior in a high density metropolis. Literature Review The relationship between land use and travel behavior has been a major theme of a great amount of recent empirical researches in urban planning and design field. One of the reasons for this intense interest is the potential of the theme that it might provide a scientific foundation for the various policies involving the idea of sustainable development: TOD, TND and the Health City. In particular, majority of these empirical studies are focusing on the relationship between microscopic land use and travel behavior. The term microscopic means that the basic unit of quantifying land use is as small as neighborhood scale within walking distance. Researches into the microscopic land use and travel behavior can be grouped by research designs: the group comparison and the multivariate analysis. The group comparison approach involves comparison between the auto-oriented suburban neighborhoods and the transit-oriented or walkable suburban neighborhoods. Cervero and Gorham(1995), Handy(1996) and Lund(2003) employed this type of research design and argued that more walkable and transit-oriented suburban neighborhoods can decrease automobile trip. Although results from these researches are easy to be interpreted and effectively capture the total effects of neighborhood design upon travel behavior, the influence of individual land use characteristics cannot be identified through this approach; therefore, the majority of related studies have employed multivariate analysis. The multivariate analysis approach involves conceptual classification and selection of numerous land use variables. Previous literature has proposed various conceptual frameworks for classifying land use variables including - 2 -
3Ds(Cervero and Kockelman, 1997), 3Ds+R(Lee and Moudon, 2006). In particular, the density and the degree of land use mix have been considered to be the key variables to be tested. Although empirical findings varies across researchers, many studies suggest that density and land use mix are significantly associated with less automobile trips and more walking, but their effects are marginal in magnitude. For example, Cervero and Kockelman(1997) suggested that the density variable and neighborhood accessibility are significantly associated with less automobile trips. Similarly, Shin(2004) found significant association between the density of commercial use and automobile dependence, and Rhim(2006) showed that the density factors and land use mix are associated with the transit ridership. Despite the fact that previous researches have made great contribution in clarifying the association between the microscopic land use and travel behavior, these researches have two shortcomings when dealing with the case of Seoul. Firstly, previous researches have paid relatively less attention to the land use factors related to parking which might have strong association with travel behavior in Seoul. The second shortcoming is that most studies except Giuliano and Dargay(2006) have not distinguished the influence of land use on car ownership and travel behavior. The car ownership rate deserves special treatment because it directly confines possible set of travel options. These shortcomings might be attributed to the fact that the major study areas of related literature have been suburban neighborhoods in the United States where parking capacity factors and car ownership rate show little differences across neighborhoods. Theoretical Framework The common theoretical framework employed by the majority of previous literature is that travel behavior variable would be a function of the land use variables and the household income factors. Figure 1 illustrates the theoretical framework employed by many related empirical researches: density, land use mix and urban form factors change the neighborhood accessibility (Krizek, 2003), thus leading to changes in travel behavior along with the household income factors. Although some recent studies including Bagley and Mokhtarian(2002), Giuliano and Dargay(2006) and Rhim(2006) used more complex structural approaches, the concept depicted in Figure 1-3 -
might be helpful for understanding empirical researches into the microscopic land use and travel behavior. Figure 1. The common theoretical framework of previous literature This study tries to expand this framework in two ways. Firstly, it introduces the concept of scale for categorizing the land use variables. Due to the fact that a metropolis such as Seoul covers quite large urban area, treating it as single city might be misleading. Rather, it would be better treating a metropolis as an agglomeration of cities consisting of smaller neighborhoods. For this reason, quantifying the land use characteristics of a metropolis would require explicit consideration for scale of urban space: otherwise, results from analyses would be difficult to be interpreted. Therefore, this study categorizes land use variables into three categories: housing level, neighborhood level and regional level land use characteristics. Among the three categories, the housing level variables have received little attention from previous literature; however, physical characteristics related to housing types are important due to the fact that they might be closely associated with parking capacity and convenience of parking in a high density metropolis. Variables captured at the neighborhood level would include most of land use variables referred in previous literature including development density and degree of land use mix. Finally, the regional level variable would be added to measure the centrality - 4 -
of given location in the entire metropolitan region. Secondly, this study assumes that the land use characteristics influence on car ownership rate and travel behavior respectively. This is a similar approach employed by Giuliano and Dargay(2006). Currently, analytic results of this paper are limited to the respective effects of the land use variables on car ownership and travel behavior, and results for the indirect effects via car ownership variable are not presented. Taking daily automobile time for the representative travel behavior variable, the travel behavior model is specified as C T = f( DL,, ε ) Similarly, the car ownership model is specified as where C N = g( DL,, ε ) C T average car time of households of each neighborhood (Dong) N D L ε C average car ownership rate of households of each neighborhood (Dong) the demographic variables of the neighborhood (Dong) the land use variables of the neighborhood (Dong) unobserved factors or error term In our model specification, the neighborhood average values are used for the car time variable, the car ownership rate and household characteristics variables due to the fact that the land use variables are measured at the neighborhood level. Many previous researches have used the household level model or the individual level model; however, in our view, the neighborhood level model would be relevant when the exact location of each household cannot be specified due to the limitation of data accuracy. Study Variables The variables for the empirical analysis can be classified into five groups: travel behavior variables, the household attributes, the housing attributes, the neighborhood land use attributes and the regional accessibility. The travel behavior variables and the household attributes were constructed upon the - 5 -
Household Daily Trip Survey conducted by the Seoul Metropolitan Government in 2002. The built environment variables involve various sources: The Property Tax Ledger of the Seoul Metropolitan Government (2004), the Seoul Business Survey Data (2003) and the Seoul Land Use Survey (2000). Detailed specifications of variables are presented in Table 1. Table 1. Study variables 1. Travel behavior (*) average household daily car time: derived by averaging daily car time of the surveyed households within a Dong (*) average household car ownership rate: average number of vehicles per the surveyed household within a Dong 2. Household (*) average household income: average monthly income for the surveyed household within a Dong; (*) average household size: average number of household members within a Dong (*) average home ownership rate: proportion of home owners within a Dong 3. Regional accessibility regional accessibility: an accessibility measured in a gravity type model; for zone i, Regional Accessibility = ( A ) exp( βd ), where i =index of origin zone, j =index of destination zone, j j ij A j =number of employees in zone j, d ij =distance between zone i and j (measured in Kilometer), and β =distance friction factor; the distance friction factor estimated = -0.289 4. Neighborhood land use development density: total floor area divided by developed land area within a Dong (*) walking accessibility to activities: measured in a gravity type model similar to the regional accessibility index above; distance friction factor for walking accessibility = -0.421; derived by averaging accessibility to activities within 500m from each land parcel proportion of the commercial floor area: total commercial floor area divided by total floor area within a Dong commercial use density: total commercial floor area divided by developed land area within a Dong commercial to residential floor area ratio: total commercial floor area divided by total residential floor area within a Dong 5. Housing housing type: a dummy variable which distinguishes data from the apartment households and the non-apartment households (*) average parking capacity: average parking capacity per household within a Dong; (*) average housing size: average residential floor area per household within a Dong (*) indicates that the variable is measured for the apartment households and the nonapartment households respectively. - 6 -
Analyses and Results Descriptive Statistics Table 2 provides the descriptive statistics for the analysis data. As can be seen in the table, the mean value of average car time per household is 50 minutes, and the mean value of car ownership rate is approximately 0.6. Table 2. Descriptive statistics for study variables Variable Min. Max. Mean Std. Error average household daily car time (minutes) 3.3333 116.5833 49.5351 19.4743 average household car ownership rate 0.0833 1.1842 0.5943 0.1866 average household income (won/month) 117.6163 301.7727 188.6441 33.7889 average household size 2.0408 4.8462 3.5848 0.3279 average home ownership rate 0.0204 1.0000 0.5928 0.1720 regional accessibility 86189.6018 736920.7236 371873.9995 151622.0593 development density 0.0793 3.8480 0.9581 0.3293 walking accessibility to activities 115.4279 17500.8069 3060.6161 2009.7033 proportion of commercial floor area 0.0300 0.9902 0.2969 0.1606 commercial use density 0.0142 3.8102 0.2934 0.2617 commercial to residential floor area ratio 0.0309 100.6666 0.7985 4.4605 housing type (0=non apartment; 1= apartment) 0.0000 1.0000 0.3131 0.4642 average parking capacity 0.0000 1.8333 0.4628 0.3817 average housing size 7.6250 65.7692 23.1520 5.9968 Valid N (listwise) 557 Factor Analysis of Explanatory Variables The variables describing the built environment are known to be correlated with each other. For example, variables related to the density of the neighborhoods tend to be correlated with variables related to the land use mix. This paper introduced factor analysis method for identifying underlying structure among the built environment variables. Table 3 shows the results of factor analysis for the built environment variables. Three factors with eigenvalue over 1 were extracted, and the extracted factors explain approximately 70% of total variations. The variables loaded in factor 1 represent the density dimension of the built environment; factor 2 represents the accessibility dimension and factor 3 includes housing - 7 -
type variables affecting parking capacity. This result suggests that the initial categorization including regional, neighborhood and housing dimension might be replaced by density, accessibility and parking capacity dimension; however, distinction between housing factors and neighborhood factors still remains valid. Table 3. Factor analysis for the built environment variables Factor1 Density Factor2 Accessibility Factor3 Parking capacity development density 0.8644 commercial to residential floor area ratio 0.7711 development density 0.7645 regional accessibility 0.7947 proportion of the commercial floor area 0.7158 walking accessibility to activities 0.6298 average housing size 0.7520 housing type(0=non apartment; 1=apartment) 0.8538 average parking capacity 0.6945 Eigenvalue 2.5219 1.9075 1.8495 % of Variance 28.0210 21.1943 20.5500 Cumulative % 28.0210 49.2152 69.7653 Extraction Method: Principal Component Analysis Rotation Method: Varimax with Kaiser Normalization Land Use Influence on Car Ownership The influence of land use variables on travel behavior was estimated using OLS regressions. Although we have results from the factor analysis procedure, raw variables were used in the analysis for ensuring legibility of results; therefore, possible inter-correlation among variables should be considered when interpreting the results. Firstly, the car ownership rate was estimated using the household and the built environment variables. Table 4 presents the estimation results. As can be seen in the table, the model is well-fitted to data: the model explains approximately 60% of total variance of the car ownership variable. All three groups of explanatory variables are significantly associated with the car ownership rate. The average income variable and the average housing size have strong positive association; in contrast, the commercial use density and the regional accessibility have negative coefficients. The results suggest that the land use mix decreases car ownership rate; however, the association - 8 -
between car ownership rate and the density related variables is not significant in case of Seoul. Table 4. OLS regression for average car ownership rate beta std.err. std. beta t-statistics sig. (intercept) -0.0201 0.0685-0.2941 0.7688 average income ** 0.0024 0.0002 0.4407 10.6017 0.0000 average household size 0.0079 0.0169 0.0139 0.4703 0.6383 average home ownership rate * 0.0626 0.0373 0.0577 1.6779 0.0939 regional accessibility ** 0.0000 0.0000-0.1408-4.6593 0.0000 development density * -0.0600 0.0320-0.1054-1.8719 0.0618 walking accessibility to activities 0.0000 0.0000 0.0485 0.9920 0.3216 proportion of commercial floor area * -0.1718 0.0990-0.1468-1.7359 0.0831 commercial use density * 0.1640 0.0976 0.2270 1.6795 0.0936 ratio of commercial to residential floor area * -0.0044 0.0025-0.1053-1.7863 0.0746 housing type ** 0.0417 0.0165 0.1037 2.5332 0.0116 average parking capacity ** 0.0479 0.0157 0.0962 3.0464 0.0024 average housing size ** 0.0073 0.0011 0.2360 6.5783 0.0000 R 0.7839 R Square 0.6145 Adjusted R Square 0.6060 * significant at 0.1 level ** significant at 0.05 level The OLS Regression for Daily Car Time As noted in the theoretical framework section, car ownership rate directly confines the travel option for a household. However, considering the OLS regression results for the car ownership rate, using the car ownership rate as an explanatory variable for estimating daily car time might involve more complex model specification. The study simply hedges this complexity by dropping the car ownership variable when modeling the daily car time. Using this reduced model specification, the results from this simplified estimation strategy would hold the required validity for discussing the influence of land use variables on car time. Table 5 provides the results of this simplified model. As can be seen in the table, the estimation result suggests that the income factors, the parking related factors and regional accessibility influence the time of car travel of a household, but the influence of land use variables in neighborhood level is not significant in Seoul. - 9 -
Table 5. OLS regression for average daily car time beta std.err. std. beta t-statistics sig. (Constant) -3.3794 9.7327-0.3472 0.7286 average income** 0.1064 0.0326 0.1847 3.2637 0.0012 average household size** 8.3348 2.3997 0.1402 3.4732 0.0006 average home ownership rate 2.7728 5.3034 0.0245 0.5228 0.6013 regional accessibility** 0.0000 0.0000-0.1844-4.4848 0.0000 development density -5.5901 4.5520-0.0941-1.2281 0.2200 walking accessibility to activities -0.0009 0.0006-0.0962-1.4440 0.1493 proportion of commercial floor area -0.2690 14.0678-0.0022-0.0191 0.9848 commercial use density 10.2138 13.8766 0.1354 0.7360 0.4620 ratio of commercial to residential floor -0.2060 area 0.3501-0.0472-0.5884 0.5565 housing type 1.2233 2.3409 0.0291 0.5226 0.6015 average parking capacity** 6.9008 2.2340 0.1327 3.0890 0.0021 average housing size** 0.5226 0.1585 0.1610 3.2983 0.0010 R 0.534447 R Square 0.285633 Adjusted R Square 0.269875 * significant at 0.1 level ** significant at 0.05 level Conclusions The conceptual framework of this study which explicitly groups the built environment variables according to the scale proved to be relevant for analyzing the association between the built environment and travel behavior: the results of the factor analysis procedure suggest that the housing type dimension and the neighborhood dimension can be statistically distinguished. Apart from the income effect, the built environment variables are significantly associated with car ownership and car time. The car ownership variable and the car time variable differ from each other in the way they are associated with the built environment variables. In detail, the car ownership variable is associated with the built environment at various level including housing type, neighborhood configuration and regional accessibility. In contrast, the car time variable showed significant association only with the housing and regional accessibility variables. In particular, the built environment variables related to the land use mix and parking capacity proved to be influential on travel behavior in Seoul. - 10 -
These results provide a consistent explanation for the land use and travel behavior in Seoul, however drawing specific land use policy from these results might be impetuous due to several limitations of this study. In particular, the structural modeling approach would be helpful for more seamless modeling results, and, more importantly, a possible association between the built environment and the household income should be investigated. Despite of these limitations, findings from this study would be useful for better understanding of the land use and travel behavior in a high density metropolis, and this understanding would be valuable for better urban planning and design in Asian mega-cities. References Cervero, R. & Kockelman, K. (1997). "Travel demand and the 3Ds: Density, diversity, and design". Transportation Research Part D: Transport and Environment, 2, pp. 199-219 Handy, S. L. (1996). "Understanding the Link between Urban Form and Nonwork Travel Behavior". Journal of Planning Education and Research, 15, pp. 183-198 McNally, M. G. & Kulkarni, A. (1997). "Assessment of influence of land use-transportation system on travel behavior". Transportation Research Record, pp. 105-115 Cervero, R. & Gorham, R. (1995). "Commuting in transit versus automobile neighborhoods". Journal of the American Planning Association, 61, p. 210 Lund, H. (2003). "Testing the Claims of New Urbanism". Journal of the American Planning Association, 69, p. 414 Lee, C. & Moudon, A. V. (2006). "The 3Ds + R: Quantifying land use and urban form correlates of walking". Transportation Research Part D: Transport and Environment, 11, pp. 204-215 Shin, S. Y. (2004). "The Relationship between Land Use and Automobile Dependence: The Case of Seoul". Seoul Studies, 5, pp. 71-93 Rhim, J. H. (2006). The Land Use Characteristics of Rail Transit Station Area Influencing Transit Demand: A Case Study of Seoul. Doctoral Dissertation. School of Civil, Urban and Geo-systems Engineering. Seoul National University Giuliano, G. & Dargay, J. (2006). "Car ownership, travel and land use: a comparison of the US and Great Britain". Transportation Research Part A: Policy and Practice, 40, pp. 106-124 - 11 -
Krizek, K. J. (2003). "Operationalizing Neighborhood Accessibility for Land Use-Travel Behavior Research and Regional Modeling". Journal of Planning Education and Research, 22, pp. 270-287 Bagley, M. N. & Mokhtarian, P. L. (2002). "The impact of residential neighborhood type on travel behavior: A structural equations modeling approach". Annals of Regional Science, 36, pp. 279-297 - 12 -