Factor Analysis of Rent Using Space Syntax Measures: Comparative Analysis by Building Use around Shibuya Station

Abstract

The determinants of residential, office and commercial rent are each analyzed by taking the 10 min walking distance area around Shibuya Station in Tokyo as the research target area. Conventional physical and quantitative factors such as distance from the nearest station, the width of the frontal road, age of the building, the number of floors, total floor area, contracted floor area, building structure etc. are included in the multiple-regression analysis as well as qualitative factors such as the visible area representing visibility on a main street and the integration value representing street network centrality—easy accessibility from other locations such as any intersection points—based on the space syntax measures (SS measures) used in the analysis by Jake Desyllas from University College of London. It is shown that both SS measures affect the rent for all the residential, office and commercial uses with the statistically significant level of 5%. It is quantitatively confirmed that the residential rent increases in a secluded location, and the office and commercial rent rise in a location with high street network centrality and good access, which would suggest that the SS measures are crucial important rent factors for residential, offices, and commercial uses in other places in Japan.

1 Introduction

1.1 Research Background and Objectives

Even though the location of a property is an important factor in the real estate business, it is difficult to evaluate location quantitatively. It is commonly considered that the advantages and disadvantages of a location are reflected in the rent and land price of a building; therefore, to better understand the location valuation, it is potentially meaningful to analyze rent factors using the quantified measures of a location.

Candidate factors in previous rent factor analysis have been broadly divided into two types of attribute: location and building. Location attributes have included distance from the nearest station, the width of the frontal road, surrounding facilities, and the like, and building attributes have included age of the building, the number of floors, total floor area, contracted floor area, and the like. As possible location valuation factors, macro-factors include stations, railway lines, the type of district and its public image, etc. (Osaragi and Ogawa 2004), and micro-factors include walking distance from any stations, surrounding facilities, etc.

Regarding the location as micro-factors, generally, the best suited locations for properties differ depending upon the use; for residential property, a secluded and quiet spot is preferred rather than a noisy place on a main street and, conversely, for office and commercial property, a location with high visibility on a main street or with easy accessibility from other locations, such as any intersection points, with a high degree of street network centrality. Therefore, it is known that a variety of micro-location attributes affect the rent; however, they have been regarded as qualitative factors, and little is taken into account as rent factors. Each building use, such as residential, offices, or commercial, has its own best suited location, which is possible to clarify through quantitative location valuation.

As a method to analyze micro-location attributes, a group from University College London proposed space syntax (hereinafter, SS) theory (Hillier and Hanson 1984). This theory is known as a means to establish an urban spatial configuration as a quantitative measure. Through visibility graph analysis, an analytic technique based on SS theory, two measures—the visible area representing visibility and the integration value representing street network centrality (hereinafter, SS measures)—are calculated, and the effect of their addition to rent factors is examined.

Centering on Shibuya Station in Tokyo Metropolis, a circular area with a radius corresponding to a maximum 10-min walk from the station was set as the research target area. To examine the factors for residential, office, or commercial rent, this paper conducted multiple regression analysis by adding two SS measures, visible area and integration value, to the following candidate factor variables:

Building attributes: building age, floor number, number of stories, contracted floor area, and three dummies: ground floor, south-facing, and building structure.

Location attributes: distance from the station and one dummy concerning the accessibility of two stations.

Factor variables found to be significant in a t-test were selected, and while examining the applicability of the models, the factor ranking was considered according to the level of the absolute t value for a factor variable. The obtained multiple regression models for each building use were compared to clarify the rent factors for residential, offices, and commercial; and the influence of location valuation quantified by the SS measures on each building use was explored.

1.2 Prior Research on Rent Factor Analysis

We first examined earlier Japanese reports on rent factor analysis published by the Architectural Institute of Japan, the City Planning Institute of Japan, and Japan Association for Real Estate Sciences and then classified the explanatory variables into location attributes and building attributes (Table 16.1). Takeshita et al. conducted time-series analysis in 2006 for office rents in Osaka City using the distance from the station, road width, and other factors as location attributes and the number of stories, floor area, and other factors as building attributes, and the coefficient of determination was 0.713 (Takeshita and Nakamura 2006). Takizawa et al. conducted analysis by adding sensitivity evaluation of office buildings as a measure to conventional location and building attributes, and the coefficient of determination was 0.89 (Takizawa et al. 2008a). In the analysis for residential rents, by using graph mining, Takizawa et al. added the actual position of any room(s) to building attributes, and the multiple correlation coefficient was 0.876, and the coefficient of determination was 0.767 (Takizawa et al. 2008b). For commercial rents, Ishii et al. conducted analysis by adding the level of recognition of a street (hereinafter, street familiarity) to the location attributes, and the maximum coefficient of determination was 0.88 for Urawa City (Ishii et al. 2015).

Table 16.1 Prior research on rent factor analysis

Japanese research in recent years has reported factor analysis in which sensitivity evaluation of such factors as the exterior or entrance, internal configuration, actual position of room(s), or the like are added to the conventional building attributes; however, regarding location attributes, apart from the street familiarity, conventional factors have been used for analysis, and the visible area and integration value of the SS measures have never been used. The previous factor analysis using the SS measures often targeted pedestrians (e.g., Araya et al. 2005; Ueno and Kishimoto 2008), but analysis has been made using land price measures, which have a correlation with rent, and it was found the SS measures were significant in a statistical test (Ota and Kaneda 2015).

In an overseas research, in 1999, Desyllas reported factor analysis for office rents in which the SS measures were added to the location attributes (Desyllas 1999). In this report, in West Berlin between 1991 and 1997, in addition to the start month/year of the lease, the integration value (t = 6.145, p < 0.0001), which is an SS measure, was also found statistically significant, and the coefficient of determination was 0.663. Similarly, for the whole of Berlin, it was found the next statistically significant factor after the start month/year of the lease was the integration value (t = 9.354, p < 0.0001), with a coefficient of determination of 0.481 (Desyllas 2000). In regard to commercial rents, in 2010 Netzell conducted analysis using the SS measures for the center of a downtown area in Stockholm, and in addition to the intensity of commercial use, and discounted rent and district dummies, the integration value (t = 2.3, p < 0.0005) was found statistically significant, and the coefficient of determination was 0.65 (Netzell 2010).

Overseas, rent factor analysis using SS measures has been reported, demonstrating a certain degree of effectiveness; however, to date, no such analysis has been reported in Japan. In fact, even overseas, no case study exists in which rent factor analysis of residential, office, and commercial uses has been conducted for the same district and factors compared for each use.

2 Rent Data as an Explained Variable

2.1 Collection of Rent Data and Target Area

The residential, office, and commercial rent data (hereinafter, rent data) used in this research were taken from the Real Estate Information Network Systems (hereinafter, REINS), and those cases where a contract was made during the 2 years from June 1, 2014, to May 31, 2016, were extracted and analyzed. In general, real estate agents share real estate information through REINS and provide it to their clients, usually individuals or medium and small enterprises. Among the abovementioned contracts in Tokyo Metropolis, the average lease square measure for office and commercial use is 64 m², and the contracts are for relatively small-size properties. The districts within 10 min of each station on the JR Yamanote Line in the center of Tokyo Metropolis are easily accessible to numerous train passengers. They have a high number of lease contracts, which were classified by use into residential, offices, or commercial (Table 16.2). The primary use of each district was established according to the number of contracts; however, since this research aims to understand rent factors for residential, office, and commercial use, Shibuya Station with a high number of contracts for all uses was selected as the target. The number of contracts was 500 residential, 575 offices, and 358 commercial. Commercial with residential and commercial with office facilities were classified as commercial properties.

Table 16.2 Number of contracts for residential, office, or commercial properties in station districts on the Yamanote Line (June 1, 2014, to May 31, 2016)

2.2 Rent Data as an Explained Variable

This research excluded rental offices under special contract conditions and small-scale offices of about 3 tsubo or 9.9 m² or below that could be shared offices. When several contracts were found in the same building, only the contract with the lowest-priced rent was adopted, the other contracts excluded. Any contracts with imperfect data, such as a non-specific address or inconsistent property name and address, were also excluded, which reduced the number of contract samples to 189 residential, 177 offices, and 143 commercial.

For the rent data as an explained variable, rent unit price/tsubo (approx. 3.3 m²) was used. This unit price was calculated as follows: any service and communal charges were added to the rent, and the total is divided by the contracted floor area (tsubo). In addition, residential rent was tax-free, but office and commercial rent included consumption tax.

3 Candidate Factor Variables

3.1 Space Syntax Measures

SS theory is a general term for the method of the spatial configuration analysis proposed by Hillier et al. of University College London in the early 1980s. This research conducted the visibility graph analysis in the SS theory and used DEPTHMAP software to calculate the SS measures. On a map, grids were laid out at intervals of 4 m, which is the Building Standards Act requirement for road width, and any grid square where more than half of the area is occupied by a street was marked. When no screen was found between the center of a marked grid square and the center of another grid square, it was assumed that both grid squares were visible to each other and the number of visible grid squares was counted. In DEPTHMAP, the total number of grid squares that can be seen from a grid square at a certain point is called “connectivity” and represents the square measure of the visible area (Turner 2004).

In addition, a measure, Integration Value (hereinafter, IV), known as a means of indicating the street network centrality, was also prepared. The IV indicates the intensity of spatial connection in a street network, and a higher value means high street network centrality and inadequate depth. Firstly, a topological distance between a certain two points (grids) is called Depth, and the Depth is calculated as the least number of steps between one point and another point. Then, the total value of each Depth from one point to all other points is leveled off by the number of all points, and this leveled value is called the Mean Depth (hereinafter, MD). Using the MD, Relative Asymmetry (hereinafter, RA) (Eq. 16.1) is found, which is the relative depth of each point (grid) seen from the entire area. When the RA value is higher, it is considered that many spaces must be traversed and the relevant space is positioned at a deep and complex location. However, since the RA is affected by k, the total number of grids, a Real Relative Asymmetry (hereinafter, RRA) is found (Eq. 16.3), which is a value standardized by a correction coefficient Dk determined by only k (Eq. 16.2). The inverse number of the RRA value is an IV (Eq. 16.4) (Sayed et al. 2014):

$$ \mathrm{RA}=\frac{2\left(\mathrm{MD}-1\right)}{k-2} $$

(16.1)

$$ {D}_k=\frac{2\left[k\left\{\mathrm{lo}{\mathrm{g}}_2\left(\frac{k+2}{3}\right)-1\right\}+1\right]}{\left(k-1\right)\left(k-2\right)} $$

(16.2)

$$ \mathrm{RRA}=\frac{\mathrm{RA}}{D_k}=\frac{\left(\mathrm{MD}-1\right)\left(k-1\right)}{\left[k\left\{\mathrm{lo}{\mathrm{g}}_2\left(\frac{k+2}{3}\right)-1\right\}+1\right]} $$

(16.3)

$$ \mathrm{IV}=\frac{1}{\mathrm{RRA}} $$

(16.4)

In regard to IV, a scope for analysis can be specified, and in general, two types of values are used: Local Integration Value (hereinafter, LIV3), an IV limited to Depth 3, and Global Integration Value (hereinafter, GIV), which covers the entire area. In this research, since the center of the Metropolis, an extensive scope, was to be handled, GIV for the streets of the entire area was measured.

3.2 Spatial Distribution of Space Syntax Measures

The two SS measures, visible area and integration value (IV), were found by visibility graph analysis of the streets in the 2-km radius circle centering on Shibuya Station. Those contracts as samples found within a 10-min walk (800-m radius) circle were extracted, but since the SS measures show a tendency for the value to increase toward the center of a target area, a buffer zone of 1200 meters was applied to create the larger 2-km radius target area.

The Shibuya Station 2-km circle covers an area comprising nearly up to Kita-sando Station to the north, Gaien-Nishi-dori Street to the east, Ikejiri-Ōhashi Station to the west, and Meguro-dori Street to the south. Figures 16.1 and 16.2 show spatial distribution maps of the found visible areas and integration values.

Fig. 16.1

Visible area spatial distribution map in the Shibuya Station 2-km radius circle

Fig. 16.2

IV spatial distribution map in the Shibuya Station 2-km radius circle

The spatial distribution map of the visible areas (Fig. 16.1) shows higher values along National Highway Route 246. The highest value was found in the vicinity of the Shibuya Station east side, which has a linear unobstructed view of National Highway Route 246. Higher values also tended to be found on other streets with larger widths.

The spatial distribution map of the IV (Fig. 16.2) shows a similar trend to the visible areas, where generally, any street with a broad width had a high value; however, these values were not conspicuously high, and with the IV, it can be read that the variation among points with high and low values was smaller compared to the visible areas.

3.3 Other Candidate Factor Variables

This section explains other candidate factor variables that were used. As building attributes, the following were included: floor number, number of stories, building age, contracted floor area, and three dummies—ground floor, south-facing, and building structure. The ground floor dummy is a dummy variable of whether or not the floor number is the ground floor, and the south-facing dummy is a dummy variable of whether the balcony faces south, southeast, or southwest. For the building structure dummy, steel-frame structure or RC (reinforced concrete) structure was used as a dummy variable. Since there were no office or commercial contracts with wooden structures, the classification was excluded. For the location attributes, the distance from the station was obtained by measuring the shortest walking distance from Shibuya Station, which is the terminal station. Moreover, in addition to Shibuya Station, the circular area also includes Omotesandō, Harajuku, Meiji-jingumae, Shinsen, Ebisu, and Daikanyama Station, and when a property was also located within a 10-min walk from one of these stations, a dummy variable for the accessibility of two stations was used.

4 Examination of Rent Factors Using Space Syntax Measures

4.1 Correlation Matrix Between Rent Unit Prices and Candidate Factor Variables and VIF

Tables 16.3, 16.4, and 16.5 show the correlation coefficients R between the rent unit prices of residential, offices, and commercial properties and candidate factor variables (lower left half) and variance inflation factors (hereinafter, VIF) (upper right half). The values exceeding the correlation coefficient absolute value 0.4, which is regarded as a slightly high correlation, were shaded. VIF was calculated using Eq. (16.5) (shown below), and it is considered as a measure of suspected multicollinearity (Marquardt 1970).

Table 16.3 Correlation coefficients between residential rent unit price and candidate factor variables and VIF

Table 16.4 Correlation coefficients between office rent unit price and candidate factor variables and VIF

Table 16.5 Correlation coefficients between commercial rent unit price and candidate factor variables and VIF

$$ \mathrm{VIF}=\frac{1}{\left(1-{R}^2\right)} $$

(16.5)

Regarding the correlation between the rent unit price and candidate factor variables, the following factor variables showed a slightly high correlation with the rent unit price: residential, building age (−0.490); offices, contracted floor area (0.459) and building age (−0.424); and commercial, ground floor dummy (0.456).

Regarding the correlation between candidate factor variables, in residential, the floor number and number of stories had a high correlation (0.752), and the visible area and IV showed a slightly high correlation (0.613). In offices, a slightly high correlation was seen between the following pairs: the visible area and IV (0.671), the floor number and number of stories (0.602), the number of stories and RC dummy (−0.502), and the number of stories and IV (0.463). In commercial properties, a slightly high correlation was found between the following pairs: the visible area and IV (0.671), the number of stories and RC dummy (−0.508), the distance from the station and IV (−0.491), the floor number and ground floor dummy (−0.449), and the number of stories and IV (0.433).

4.2 Examination of Models of the Multiple Regression Analysis Results

Similarly, to the prior research, multiple regression analysis was conducted for each use, and factors that were significant at a 5% level in a t-test were selected, and Tables 16.6, 16.7, and 16.8 show the results. In model selection, the stepwise variable increase and decrease method was applied, and any model with the minimum value of Akaike’s Information Criterion (hereinafter, AIC) was selected.

Table 16.6 Multiple regression analysis results: residential rent

Table 16.7 Multiple regression analysis results: office rent

Table 16.8 Multiple regression analysis results: commercial rent

The coefficient of determination of each rent model was as follows: residential, 0.305; offices, 0.475; and commercial, 0.367, and in the variance analysis of the multiple regression equation, all uses showed a significant level of 1%.

In those analysis models excluding the SS measures from the candidate factors, the coefficients of determination of residential, offices, and commercial were 0.288, 0.471, and 0.343, respectively, and AIC were 3425.82, 3354.55, and 2998.11. AIC of the models using the SS measures were lower for all uses, which confirmed that the SS measures improved the multiple regression models. However, our models showed lower values in comparison with the coefficients of determination of the models in earlier overseas research using the SS measures, such as 0.485 for offices in the whole of Berlin and 0.65 for commercial in downtown Stockholm.

Concerning the factor variables, in the residential rent, building age (−8.204), distance from the station (−3.698), and IV (−2.155) were adopted in the order of absolute t value. In the same way, in the office rent, building age (−6.554), contracted floor area (6.277), distance from the station (−5.015), IV (2.284), and steel-frame dummy (2.108) were adopted. In the commercial rent, ground floor dummy (7.396), distance from the station (−3.406), building age (−2.610), and IV (2.254) were adopted.

In addition, by regarding each explanatory variable as an explained variable, and using all other explanatory variables, multiple regression analysis was conducted to calculate the VIF of each explanatory variable; however, it was less than 4 for all explanatory variables, and therefore it was assumed there was no doubt concerning multicollinearity, and all explanatory variables were adopted.

4.3 Comparison of Multiple Regression Models by Building Use

With a focus on the ranking of each factor variable, the models were compared by use. The building age ranked first in residential and offices, but third in commercial. Whereas, in commercial, the ground floor dummy ranked first. The distance from station ranked second in residential and commercial and came third in offices. In offices the contracted floor area came second. The building age and distance from the station ranked relatively high in all three uses, allowing us to read their great influence on the rent. It is possible to consider that since commercial properties are visited by an unspecified large number of customers, the ground floor dummy that is easy to recognize and access took first place. In addition, the reasons why the contracted floor area ranked second for offices could be that there is a certain level of demand to rent a substantial floor area, and a larger lease floor area means a larger company with the means to pay more rent.

Next, IV, an SS measure, took third place in residential and fourth place in offices and commercial; even though it did not rank in any higher position, it was adopted through a statistical test for all uses and showed a certain level of influence. For offices and commercial, a location with high street network centrality and allowing easy access from anywhere is preferred, whereas for residential, a secluded and quiet location with low street network centrality is favored, and this could contribute to the minus coefficient values in the table.

Moreover, for offices, the adopted steel-frame structure ranked fifth, which allows us to read the rent of a steel-frame structure building to be higher. Offices include Small Office Home Office (SOHO) where rooms are used both for residential and office, and SOHO rent tends to be lower than regular offices. Generally, a SOHO property is often a wooden detached house or in an RC structure apartment, not in a steel-frame structure. For this reason, it is possible to consider that the rent increases for a steel-frame structure, which is highly likely to be a regular office and not a SOHO.

4.4 Examination of Rent Factor Analysis Using the Visible Area

Since the IV and visible area had a slightly high correlation, multiple regression analysis was conducted using the visible area as a candidate factor, instead of the IV. As a result, the visible area p values were 0.118 for residential, 0.196 for offices, and 0.606 for commercial, and in all models, no significant difference at a 5% level was found for the visible areas. Particularly, it is possible to think the visible area has hardly any influence on commercial properties.

From the spatial distribution map, it was found that the visible area tended to show a higher value at a point on a street with a large width; however, National Highway Route 246 is partially occupied by an expressway and is not a busy pedestrian street, and therefore, the commercial rent unit price along National Highway Route 246 could not be so high. For this reason, the National Highway Route 246 samples (residential, 6; offices, 11; commercial, 5) were excluded, and the same analysis was conducted. Tables 16.9, 16.10, and 16.11 show the results. Moreover, regarding the spatial distribution map of visible areas, the shading scale was changed so as to allow values other than National Highway Route 246 to be read, as shown in Fig. 16.3.

Table 16.9 Multiple regression analysis using the visible area: residential rent

Table 16.10 Multiple regression analysis using the visible area: office rent

Table 16.11 Multiple regression analysis using the visible area: commercial rent

Fig. 16.3

Visible area spatial distribution map in the Shibuya Station 2-km radius circle (shading changed)

The coefficients of determination of residential, offices, and commercial were 0.298, 0.473, and 0.377, respectively, and the visible area was not adopted for residential, and its p values for offices and commercial were 0.007 and 0.034, thereby confirming a significant difference at a 5% level. These results indicate that in a location with good visibility, the rent unit price for offices and commercial goes up.

5 Conclusion

In regard to the rent factors for residential, office, and commercial properties within the Shibuya Station 10-min walk circle in Tokyo Metropolis, multiple regression analysis was conducted. As candidate factor variables, the following building attributes were used: building age, floor number, number of stories, contracted floor area and three dummies—ground floor, south-facing, and building structure. The following location attributes were also used: distance from the station and one dummy concerning the accessibility of two stations. Two SS measures were also added: visible area representing visibility and integration value representing street network centrality.

Factor analysis for the residential rent found that in the order of absolute t value, the factors were building age, distance from the station, and integration value, and the coefficient of determination was 0.305. The factors for the office rent were building age, contracted floor area, distance from the station, integration value, and steel-frame dummy, and the coefficient of determination was 0.475. The factors for the commercial rent were the ground floor dummy, distance from the station, building age, and integration value, and the coefficient of determination was 0.367.

The integration value, an SS measure, was adopted in all uses with a significance level of 5%. For offices and commercial, a location with high street network centrality and allowing easy access from anywhere is preferred, whereas for residential, a secluded and quiet spot with low street network centrality is favored, and this could contribute to the minus coefficient values.

In another SS measure, visible area, National Highway Route 246 samples, which showed extremely high values, were excluded, and instead of the integration value, the visible area was used for analysis, and the coefficients of determination of residential, offices, and commercial were 0.298, 0.473, and 0.377, respectively, and the visible area p value for offices and commercial were 0.007 and 0.034, respectively, confirming a significant difference at a 5% level. The value of a partial regression coefficient resulted in a plus value, which confirmed that under certain conditions, in a location with good visibility, the office or commercial rent unit price goes up.

This research suggested that in Japan as well, the SS measures were rent factors for residential, offices, and commercial. In addition, based on the value of the partial regression coefficient of the integration value, which is an SS measure, it was quantitatively confirmed that the residential rent increases in a secluded location, and the office or commercial rent rises in a location with high street network centrality and good access.

Micro-locations involve many aspects that have not been quantified yet, and as tasks to be tackled in the future, the authors would like to aim at the further development of location valuation, such as techniques and measures for quantification, other than visibility and street network centrality using SS measures.