1# RECORD e2459715fae0e20316482cb1778f7b7fac58b6af2# TITLE: Understanding Spatial Variations in the Impact of Accessibility on Land Value Using Geographically Weighted Regression3# YEAR: 2012.0 VENUE: nan DOI: 10.5198/jtlu.v5i2.2254# AUTHORS: H. Du; Corinne Mulley5# PAGES: 14 CANDIDATE TABLE PAGES (1-based): [9, 10]6# ABSTRACT: This paper aims to understand the spatial variability in house prices and accessibility. The motivation for understanding the connection between accessibility and house prices stems from the increasing attention given in recent years to the potential for funding transport infrastructure by land value capture policies. Establishing whether there is identifiable land value uplift, and further quantifying this uplift, is a prerequisite to sensible discussions on the potential for land value capture. Although there has been substantial related research in the United States, not only have there been fewer studies in the United Kingdom, but these have concentrated on London. London, as a capital city, differs in many respects from other cities. Large conurbations such as Manchester, Sheffield, and Tyne and Wear are more typical of British cities. This study focuses on the Tyne and Wear area, which has an extensive public transport system, with a light rail system—the Tyne and Wear Metro—forming the backbone of the public transport system. The investigation reported in this paper is underpinned by the use of Geographically Weighted Regression (GWR) methodology with property prices as the dependent variable, which in turn is explained by independent variables designed to standardize for household features and spatially defined factors including the transport accessibility of the house location. This methodology allows for estimation of the importance of transport accessibility in determining house prices. The empirical results show that, on average, the internal factors of the property and the socio-economic classification of its location are dominant determinants of property prices, but transport accessibility variables are also significant. However, the local model approach of GWR shows a significant spatially varying relationship between property prices and transport accessibility to be identified. This study contributes to a quantification of the impact of accessibility on house prices. Moreover, the paper demonstrates the application of a relatively new methodology in the transport field that takes account of the spatial nature of the data required in this process.789===== PAGE 1 (score 15, numeric-paren 0, table-words 2, var-words 11) =====1011 http://jtlu.org12. 5 . 2 [2012] pp. 46–59 doi: 10.5198/jtlu.v5i2.22513Understanding spatial variations in the impact of accessibility on land value using 14geographically weighted regression15Hongbo Du 16 17 18 19Corinne Mulley20Dongguan Institute of Urban Planning and Construction a 21 22University of Sydney b23Abstract: This paper aims to understand the spatial variability in house prices and accessibility. The motivation for understanding the con24nection between accessibility and house prices stems from the increasing attention given in recent years to the potential for funding transport 25infrastructure by land value capture policies. Establishing whether there is identifiable land value uplift, and further quantifying this uplift, 26is a prerequisite to sensible discussions on the potential for land value capture. Although there has been substantial related research in the 27United States, not only have there been fewer studies in the United Kingdom, but these have concentrated on London. London, as a capital 28city, differs in many respects from other cities. Large conurbations such as Manchester, Sheffield, and Tyne and Wear are more typical of Brit29ish cities. This study focuses on the Tyne and Wear area, which has an extensive public transport system, with a light rail system—the Tyne 30and Wear Metro—forming the backbone of the public transport system. The investigation reported in this paper is underpinned by the use 31of Geographically Weighted Regression (GWR) methodology with property prices as the dependent variable, which in turn is explained by 32independent variables designed to standardize for household features and spatially defined factors including the transport accessibility of the 33house location. This methodology allows for estimation of the importance of transport accessibility in determining house prices. 34The empirical results show that, on average, the internal factors of the property and the socio-economic classification of its location are 35dominant determinants of property prices, but transport accessibility variables are also significant. However, the local model approach of 36GWR shows a significant spatially varying relationship between property prices and transport accessibility to be identified.37This study contributes to a quantification of the impact of accessibility on house prices. Moreover, the paper demonstrates the application 38of a relatively new methodology in the transport field that takes account of the spatial nature of the data required in this process.39Keywords: accessibility, public transport, land value, land value capture40Copyright 2012 Hongbo Du and Corinne Mulley.41Licensed under the Creative Commons Attribution – NonCommercial License 3.0.421 43Introduction44Public transport investments improve accessibility for locations 45where users are served directly by this investment or use this 46investment as part of the public transport network. In general, 47improvements in accessibility are translated into enhanced 48land values; this process is called land value uplift. In recent 49years, land value uplift has attracted increasing attention as a 50potential means of financing transport infrastructure, particu51larly in the United Kingdom, where a swathe of new light rail 52schemes failed to be implemented because of a lack of funds 53(GVA Grimley 2004, 1). However, the empirical evidence to 54date is not sufficiently robust to identify how much transport 55infrastructure and its associated accessibility impacts contrib56ute to land value uplift, and this is a prerequisite to sensible 57discussions on land value capture. In the early stages of seek58ing evidence on this matter, comparison methods were used 59(Pickett and Perrett 1984; Cervero and Landis 1993; Du and 60Mulley 2006). Hedonic price methods have also been widely 61considered as an approach to identify the impact of transport 62investment on land value, and this is particularly so in the 63United States, (Weinberger 2001, Cervero and Duncan 2002a, 642002b). In the United Kingdom, two studies have employed 65a hedonic price approach to evaluate the impact of a light rail 66system on house prices in Manchester (Forrest and Glen 1995), 67and in Sheffield (Henneberry 1998). 68However, to properly understand the relationship between 69transport infrastructure and land value, it is necessary to deal 70with spatial data. Hedonic models, as a subset of traditional 71multiple regression models, assume that the observations in the 72regression are independent of one another. This assumption is 73likely to be broken when data are spatially correlated (correla74tion of a variable with itself through space), as is common with 75geographical data. Geographical data are likely to follow the 76first law of geography, for which “everything is related to every77thing else, but near things are more related than distant things” 78a duhongbo@gmail.com79b corinne.mulley@sydney.edu.au808182===== PAGE 9 (score 40, numeric-paren 0, table-words 7, var-words 24) =====83845485JOURNAL OF TRANSPORT AND LAND USE 5.286Table 2: Results of global regression parameters.87PARAMETER88ESTIMATE89Significance (p-value)90INTERCEPT9111.965920.00093BEDS94-0.074950.00096FLATBED970.070980.00099SEMIBED1000.0821010.000102DETABED1030.0921040.000105SP_AVE106-0.0091070.085108ETHNM1090.0011100.685111HPROF1120.0291130.000114UNEM115-0.0741160.000117CAR_ACC1180.0211190.000120PT_ACC121-0.0121220.003123As this is a semi-log functional form, the interpretation 124of the estimated coefficients relates to their proportional (or 125when multiplied by 100, the percentage) effect on Price. For 126the internal factors of the property, it can be seen that there is 127significant interaction between the size of the property, mea128sured by the number of bedrooms, and the property type as 129well as a significant estimated coefficient for the number of 130bedrooms. The interpretation of these interaction terms is that, 131within a given property type, there is a differential impact of 132the number of bedrooms on the price of property and more133over, this differential impact declines when moving from de134tached houses to the higher density flat type. To understand 135the impact of an additional bedroom, the estimated parameter 136for BEDS needs to be taken into account and thus, the global 137regression suggests that, ceteris paribus, an additional bedroom 138in a detached house will lead to an 1.8 percent (-0.074 + 0.092 139= 0.018) increase in price.140With respect to the socioeconomic factors, the percentage 141of ethnic minority residents and the percentage unemployed 142(ETHNM and UNEM, respectively) might be expected to 143vary with property value, but only the percentage unemployed 144was significant in the global regression, with an increase in 1 145percent in unemployed, ceteris paribus, leading to a 7 percent 146decrease in property price. Having a higher percentage of pro147fessional residents (HPROF) and having a better school nearby 148would be expected to lift property value, and while this is the 149case for HPROF (adding nearly 3 percent to property price), 150the school variable (SP_AVE) is not significantly different from 151zero. However, the factors of high professional residents and 152levels of unemployment reflecting income of residents do, as 153expected, significantly contribute positively and negatively to 154property value respectively.155In terms of car and public transport accessibility, greater 156travel time implies worse accessibility and vice versa. If accessi157bility improves, then everything else being held constant would 158suggest that the estimated parameters should be negative, as 159one additional minute of car/public transport travel time 160(implying less good accessibility) is expected to lead to lower 161property prices. Both estimated parameters are significant, and 162the estimate for the public transport parameter PT_ACC is 163negative, albeit with a small absolute value of 1 percent. At 164the mean, a one-minute saving in public transport travel time 165equates to an approximately 3 percent saving from the average 166time (see Table 1) and generates a house price premium of ap167proximately £1600 (1 percent of the mean price; see Table 1). 168For the car variable, CAR_ACC, the estimated parameter is 169significant but positive, suggesting that longer travel times to 170the city center have higher property prices. In general this ap171plies to houses located further away from the city center and is 172evidence of the general trend of suburbanization in the United 173Kingdom, which is so noticeable in the Tyne and Wear area. 174It should be noted that, while this global regression used car 175accessibility times based on a road network between congested 176and free flow, the modeling was not sensitive to whether this 177accessibility or one based on the full or free-flow network was 178used.179The next section considers the effect of geography, as iden180tified by the local GWR model. 1815.2 182Local model: the GWR estimation183One of the advantages of the GWR model is the ability to 184examine the spatial variability of independent variables includ185ed as explanatory variables. Some independent variables, there186fore, might be non-significant in the global regression model 187but might vary significantly over the geographical area and be 188revealed as significant local parameters by the GWR modeling. 189It is expected that mapping the estimated parameters for public 190transport accessibility to employment, using the results of the 191local estimates from GWR, will show these distributed in ac192cordance with public transport facilities such as bus stops and 193Metro stations, and the estimated parameters for car accessibil194ity to the city center are expected to show a relationship with 195good road access, such as arterial roads.196The GWR software provides diagnostic information to 197assess whether the local model is an improvement over the 198global model described above. In this analysis, the local model 199benefits from a higher adjusted coefficient of determination 200(adjusted R2), from 0.328 to 0.467, and a lower AIC (from 2011721 down to 1392), suggesting that the GWR local model 202gives a significantly better explanation, after taking the degrees 203204205===== PAGE 10 (score 24, numeric-paren 0, table-words 5, var-words 12) =====20620755208Understanding spatial variations in the impact of accessibility on land value using geographically weighted regression209of freedom and complexity into account.210As identified above, one of the advantages of GWR is the 211ability to examine spatial variability that might otherwise be 212hidden in a global regression model. Property prices are very 213likely to vary over the geographical area. Based on a Monte 214Carlo significance test procedure, the GWR software can ex215amine the significance of the spatial variability of parameters 216identified in the local parameter estimates. The results of these 217tests, shown in Table 3, demonstrate that there is highly sig218nificant (at the 5 percent level) variation in the local parameter 219estimates for all variables except the percent ethnic minority. 220The school point score varies significantly over space, although 221this variable was insignificant in the global regression, suggest222ing that this is a factor in some areas but not in others. 223Table 3: The results of Monte Carlo significance test. 224PARAMETER225SIGNIFICANCE 226LEVEL FOR SPA-227TIAL VARIABILITY228(p-value)229SIGNIFICANCE 230LEVEL IN GLOBAL 231REGRESSION232(p-value)233INTERCEPT2340.0002350.000236BEDS2370.0002380.000239FLATBED2400.0002410.000242SEMIBED2430.0002440.000245DETABED2460.0002470.000248SP_AVE2490.0402500.085251ETHNM2520.1002530.685254HPROF2550.0002560.000257UNEM2580.0172590.000260CAR_ACC2610.0002620.000263PT_ACC2640.0002650.003266The results for the two variables—car accessibility (CAR_267ACC) together with the variable looking at public transport 268accessibility (PT_ACC)—are mapped next to examine their 269spatial variation because of this paper’s focus on the impact of 270accessibility on house prices.271In Figures 2 and 3, the background shading refers to the 272significance of the local parameter estimates, which are shown 273as different colored points. Areas shaded grey contain local pa274rameters that are statistically significant, with dark grey areas 275referring to t-values of less than -2 and light grey areas referring 276to t-values greater than +2. The values of the local parameters 277have been colored so that positive values are shades of red while 278negative values are shades of green. It is clear from the maps 279that the parameters demonstrate considerable spatial variation.280Figure 2 shows that there is clear variation in the local param281eter estimates for the variable PT_ACC over different areas. 282Everything else being equal, it is expected that quicker access 283by public transport (thus a decrease in time of access) will 284lead to an increase in house price, suggesting there should be 285a negative relationship between public transport travel time 286and property price. In Figure 2, there is a large area of sig287nificant and negative local parameters that covers the majority 288of the Newcastle upon Tyne District, overlaps into the Gates289head District in the south, and includes the central and most 290westerly parts of the Metro system. Using known information 291about Tyne and Wear, within this area, the effect is stronger in 292the poorer areas where savings in public transport accessibil293ity of one minute give a bigger percentage change to house 294prices (up to 6 percent, as compared to the global average of 2951.2 percent, shown in Table 2). This reflects the way in which 296public transport accessibility is likely to be more important to 297households with lower incomes, especially as car ownership 298is relatively low in Tyne and Wear. Figure 2 shows also that 299within this area there is a strong cluster of significant negative 300local parameters around the Metro link going north and west. 301These areas are regarded as wealthier and include inner suburbs 302that have been gentrified. Although a less strong effect, Figure 3032 shows that the house price premiums associated with these 304locations are equal or greater (between 1 and 2 percent) to that 305identified by the global regression.306Similarly, Figure 3 exhibits clear variation in the local pa307rameter estimates for the variable CAR_ACC over the differ308ent areas, being positive in most areas and negative in some. 309The geographic central core shows significant negative local 310parameters, although the absolute values are close to zero, sug311gesting that accessibility by car does not add a premium to 312house prices here. However, the locations of these properties 313are such that need to travel by car to the center is low, if non314existent. This core is surrounded by an area of insignificant 315local parameters. On the outer area is a large area of significant 316positive local parameters. Positive parameters suggest that an 317extra minute of car travel time adds to house prices. These 318areas with positive relationships between car travel time to 319city center and property price are hypothesized to have other, 320stronger neighborhood features with specific amenities, such 321as the seaside (in the east), countryside (in the north and west), 322or proximity to a major shopping center (in the north part of 323Gateshead District). In the east in particular, it can be seen that 324there is a concentration of positive local parameters, suggesting 325one additional minute travel time by car increases house prices 326by more than 7 percent, as compared to the average of 2 per327cent identified by the global regression. The results suggest that 328329330===== PAGE 11 (score 15, numeric-paren 0, table-words 2, var-words 11) =====33133256333JOURNAL OF TRANSPORT AND LAND USE 5.2334occupants of properties in these locations are prepared, ceteris 335paribus, to pay for suburban living with additional minutes 336travelling by car.337The way in which house prices vary over geographical 338space could be mapped for each of the independent variables 339identified in the model, as reported in Table 3, giving a set of 340maps which together would provide a fuller picture of the driv341ers of house price over geographical space. Indeed, while more 342difficult to present in two-dimensional space, the building of a 343multilayered map may be an alternative to simply considering 344a number of maps side by side.345As a way of modeling data with a spatial content, GWR 346provides an elegant way of estimation with outcomes that can 347be interpreted over geographical space. However, the literature 348identifies a significant theoretical issue arising from the addi349tional complexities of the GWR estimation, potentially giv350ing rise to multicollinearity or interrelationships among the 351local estimates even when there are no such relationships at 352the global model level. Checks for linear dependency at the 353global level were carried out by an examination of the correla354tion matrix and with regard to the stability of the parameter 355estimates and the associated standard errors. At the local level, 356checks for multicollinearity followed Wheeler and Teifelsdorf 357(2005), who propose a number of different tests for identifying 358the presence of multicollinearity; this work has followed two 359of these tests, which follow directly from GWR output (scatter 360plots between the local parameter estimates, supplemented by 361histograms of local parameter correlations) with little evidence 362of multicollinearity in this dataset.363As in all analysis, missing variables will lead to model 364specification problems. In this analysis, improvements in the 365model specification need to be considered in the context of 366the results discussed above. One category of missing variables 367is those variables which might better distinguish a property’s 368internal factors—for example, the number of bathrooms or the 369existence of a garage. As these were unavailable, their absence 370is likely to have resulted in a depressed R2. A second category 371of missing variables, such as proximity to amenities, proximity 372to stations, parking availability, crime, and other social factors, 373also creates limitations in terms of model specification. But be374Figure 2: Map of the local parameter estimates associated with public transport accessibility (variable PT_ACC).375