Abstract
PART 2 J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u This page intentionally left blank A Study of LEED vs. Non-LEED Office Buildings Spatial & Mass Transit Proximity in Downtown Chicago A u t h o r Sofia Dermisi A b s t r a c t Although the number of Leadership in Energy and Environmental Design (LEED) certified office buildings continues to increase, research on their spatial distributions in comparison to non-LEED buildings and mass transit links need to be explored in depth. This paper focuses on these aspects using all the downtown Chicago Class A office buildings as the study area. The findings show that LEED buildings are 21% closer to each other, indicating possible proximity pressure. LEED-Gold buildings are also 18% closer to each other compared to Silver. Regarding mass transit, LEED compared to non-LEED buildings are on average 14% closer to a metro area commuter rail station (Metra) and 12% closer to a local commuter rail station (CTA). In addition, LEED and non-LEED buildings show some evidence of small group clustering in certain areas, while the econometric results indicate that buildings located along the most prominent office market street (Wacker Drive) achieved 12% higher LEED points compared to other LEED buildings. A similar result was experienced among buildings built after 1979 and those certified under LEED v.2009 (12% and 19%, respectively). Additionally, LEED-Silver buildings achieved a lower number of points compared to other certification levels by 20%. One can argue that the inclusion of a building’s sustainability status [Leadership in Energy and Environmental Design (LEED) and ENERGY STAR] in commercial property databases such as the CoStar Group may pressure building owners to pursue such standards, especially in markets with increased adoption of LEED certification (Dermisi, 2009). The elevation of a building’s performance to a LEED standard usually requires a combination of strategies and measures in order to accomplish long-term energy cost efficiencies / decreases (e.g., energy- efficient lighting, efficient heating and cooling equipment, etc.), as well as improved emissions and an overall healthier work environment. The incurred costs by building owners may vary significantly as can their payback periods (Nils, Miller, and Morris, 2012). However, most owners are aware of the simple check- box on the CoStar Group’s website, which if checked can exclude them from a perspective tenant’s consideration if their building is not sustainable. Researchers have compared the performance of LEED and non-LEED buildings based on their vacancies, rents, sale prices, and valuations (Miller, Spivey, and J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 1 6 D e r m i s i Florance, 2008; Fuerst and McAllister, 2009, 2011; Miller and Pogue, 2009; Eichholtz, Kok, and Quigley, 2010a, 2010b, 2010c; Wiley, Benefield, and Johnson, 2010; Dermisi and McDonald, 2011; University of San Diego, CBRE, and McGraw Hill Construction, 2011). Others have focused on spatial regression modeling of real estate related issues, although LEED spatial patterns have not been analyzed. Specifically, Ayse (1998), Lipscomb (2004), Valente, Wu, Gelfand, and Sirmans (2005), Osland (2010), and Wallner (2012) focused on residential and mortgage spatial allocation patterns. Jennen and Brounen (2009) and Montero and Larraz (2011) focused on commercial property pricing patterns. Clapp and Rodriguez (1998) focused on travel distances calculations while Anselin (1998) and Thrall (1998) focused on the geographic information system (GIS) applications for real estate in a broad context. This paper tries to fill the void in the study of spatial distribution of LEED versus non-LEED buildings in dense urban environments, such as a downtown, and the lessons we can learn by studying their building characteristics. Focusing exclusively on all Class A office buildings (LEED and non-LEED with a limited exploration of ENERGY STAR labeled buildings) in downtown Chicago, the objective is to explore the underlying spatial patterns of these buildings in relation to each other as well as the mass transit rail system, their possible clustering, and the effect of building characteristics on LEED points a building can achieve. D a t a The study of LEED and non-LEED buildings requires the combination of two data sources: one for real estate (CoStar Group) and the other on sustainability [U.S. Green Building Council (USGBC)], allowing for the development of a full profile of a building. Due to the analysis of the spatial dynamics of both types of buildings in downtown Chicago in relation to mass transit stations, a third data source was introduced (City of Chicago database), which provided information on the location of all mass transit rail stations (Chicago metro area commuter: Metra and local commuter: CTA). The data extracted from the CoStar Group database included all Class A office buildings in downtown Chicago (LEED and non-LEED) with their specific characteristics, such as ENERGY STAR and LEED status, year built, rentable building area (RBA), number of stories, and submarket. This dataset was then complemented with more detailed information on the LEED building designation [rating (only buildings that achieved LEED: Existing Buildings Operations & Management (EBOM) or Core & Shell were selected for the study), version, certification level and points] from the USGBC database. Finally, the mass transit rail station locations were extracted from the City of Chicago database, with an exclusive focus on downtown Chicago. The overall dataset consists of 102 Class A office buildings with 71.6% (73 buildings) achieving ENERGY STAR label status at least once and 50.9% (52 buildings) of them being certified as LEED. Assessing the sustainability footprint of only the non-LEED buildings, 46% (23 buildings) have already achieved the ENERGY STAR label at least once. The area of study also includes five Chicago metropolitan area commuter stations, from which two are the main entry points A S t u d y o f L E E D v s . N o n - L E E D O f f i c e B u i l d i n g s 1 1 7 Exhibit 1 u Location of LEED and Non-LEED Buildings with the largest volume of daily commuters (Union and Ogilvie stations–Metra) and 23 local commuter rail stations (CTA). M e t h o d o l o g y Two aspects of the Class A office buildings in downtown Chicago were analyzed. The first is the spatial distribution of LEED versus non-LEED buildings and possible proximity pressure in achieving LEED in areas with significant LEED adoption. The spatial distribution of both groups of buildings is studied with the use of GIS, which allows the visualization of the locations of both groups of buildings (Exhibit 1) and the analysis of their spatial distributions. The availability of the ENERGY STAR status (Exhibit 2) allows a further evaluation of the first step towards sustainability managers of buildings take before pursuing LEED. The second aspect I analyze is the effect of building characteristics and location on the LEED scores achieved. Four research questions were evaluated. The first question is: Are LEED and non- LEED buildings concentrated in the same areas? What is the average distance among buildings in each group and how is it affected by mass transit rail stations? These questions require the spatial analysis of these buildings. Directional distribution (Exhibits 3 and 4) is a first step in answering the first part of the question and determined the dispersion of the buildings within each of the two groups. Utilizing ArcGIS a standard deviation ellipse polygon was generated J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 1 8 D e r m i s i Exhibit 2 u ENERGY STAR and Non-ENERGY STAR Buildings Exhibit 3 u Directional Distribution and Central Features of LEED Buildings A S t u d y o f L E E D v s . N o n - L E E D O f f i c e B u i l d i n g s 1 1 9 Exhibit 4 u Directional Distribution and Central Features of Non-LEED Buildings based on the buildings’ coordinates. The denser patterns of the LEED buildings led to the decision to define the directional distribution ellipse to encompass 68% of the data (1-standard deviation polygon), (Exhibit 3). For consistency purposes, a 1-standard deviation polygon was also used for the non-LEED buildings (Exhibit 4). The identification of the most central point of the spatial distributions (Exhibit 5) of both building groups (LEED and non-LEED) allowed the quantification of any difference between them with the use of ArcGIS. The central feature was also determined for each of the two mass transit rail modes (METRA and CTA) to effectively estimate their overall proximities to the central features of LEED and non-LEED buildings. Identifying the distances between neighboring buildings for each of the two groups (LEED and non-LEED) provided insights on the concentration or dispersion pattern in a quantitative beyond a visual representation (Exhibit 6). Distances were estimated based on clusters of three neighboring buildings. LEED-Silver and LEED-Gold buildings were the most popular certification levels (17 and 25 buildings, respectively) and distances were estimated using the same three- neighbor logic. A one-on-one distance analysis between each of the mass transit stations (Metra and CTA) and each of the LEED and non-LEED buildings was performed to J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 2 0 D e r m i s i Exhibit 5 u Central Feature Comparison of LEED and Non-LEED Buildings Exhibit 6 u Distance between Neighbors Mean Min. Max. All LEED 0.127 0.100 0.261 Non-LEED 0.161 0.067 0.397 Silver 0.218 0.141 0.320 Gold 0.178 0.107 0.390 determine the existing distances and the evolving patterns. Exhibits 7 and 8 present some basic statistics from this comparison by station and type of building. The second research question is: Are LEED and non-LEED buildings randomly located throughout the Chicago downtown area? This question is answered using both a quantitative and a visual approach. Utilizing the average nearest neighbor equations from ArcGIS, an assessment was made regarding the randomness of the distributions (Exhibits 9 and 10). The A S t u d y o f L E E D v s . N o n - L E E D O f f i c e B u i l d i n g s 1 2 1 Exhibit 7 u Distances between Mass Transit METRA Rail Stations and Buildings LEED Non-LEED METRA Station Average Min. Max. Average Min. Max. Ogilvie 0.48 0.05 1.11 0.68 0.21 1.48 Union 0.50 0.07 1.17 0.66 0.10 1.66 Randolph 0.58 0.03 0.98 0.61 0.05 1.13 LaSalle 0.67 0.21 1.19 0.74 0.09 1.81 Van Buren 0.77 0.29 1.15 0.79 0.17 1.63 Note: All values in miles. spatial autocorrelation was then assessed with ArcGIS’s Global Moran’s I, which evaluated autocorrelation based on the buildings’ proximity to others and the assessed characteristic [e.g., year built, rentable building area (RBA), LEED levels, and points achieved]. Under the null hypothesis, various characteristics are randomly distributed among the buildings in the area of study. With the help of ArcGIS, a surface was generated throughout the study area based on the location of each building compared to the others, a building characteristic (e.g., year built, RBA, LEED levels, and points achieved), and the number of cells sharing the building’s characteristic within a defined neighborhood (Exhibits 11– 17). The maps generated calculated a magnitude effect per foot, based on the building’s characteristics, which fall within a neighborhood generated around each cell. Due to the close proximity, a torus neighborhood was defined among LEED buildings with an inner radius of 0.0036 miles and outer of 0.011 miles. In contrast, due to the distribution patterns of non-LEED buildings, the inner radius decreased slightly to 0.004 but the outer increased to 0.013 miles. Another set of density maps was generated utilizing the proximity of each building to each of the mass transit stations (Metra and CTA). Each building was assigned a ranking from 1 to 3 based on the overall average proximity across all of the Metra or CTA stations. Buildings that achieved an overall average distance of less than 0.5 miles across all transit stations were assigned to group 1, those that were between 0.51 and 0.7 miles were assigned to group 2, and those with more than 0.7 miles from a station were assigned to group 3. This grouping was then used to develop the density map based on this feature. A final approach in identifying the statistical significance of clustering (hot spots, cold spots, and outliers) was explored using Anselin’s Local Moran’s I. Each building of both groups (LEED and non-LEED) was assessed for their possibility of clustering based on their characteristics and location, including mass transit (Exhibits 14–17). Four spatial significance groups were generated, which determined the underlying patterns: High-High Clusters (HH), High-Low Outlier (HL), Low-High Outlier (LH), Low-Low Cluster (LL), and for the lack of J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 2 2 D e r m i s i Exhibit 8 u Distances between Mass Transit CTA Rail Stops and Buildings LEED Non-LEED CTA Station Average Min. Max. Average Min. Max. Adams/Wabash 0.56 0.09 0.92 0.61 0.12 1.46 Chicago/Franklin 0.97 0.56 1.30 1.00 0.47 1.43 Chicago/State 0.96 0.37 1.39 0.92 0.22 1.52 Clark/Lake 0.38 0.07 0.66 0.50 0.09 1.08 Clinton/Congress 0.74 0.22 1.43 0.87 0.09 1.94 Clinton/Lake 0.56 0.01 1.09 0.75 0.29 1.37 Grand/State 0.66 0.03 1.08 0.68 0.06 1.21 Jackson/Dearborn 0.54 0.13 0.93 0.60 0.04 1.57 Jackson/State 0.58 0.16 0.91 0.62 0.06 1.56 Lake/State 0.43 0.16 0.76 0.51 0.11 1.11 LaSalle 0.65 0.19 1.14 0.72 0.06 1.77 LaSalle/VanBuren 0.57 0.11 1.07 0.64 0.01 1.69 Library 0.62 0.23 0.99 0.67 0.07 1.65 Madison/Wabash 0.49 0.45 0.85 0.55 0.08 1.29 Merchandise Mart 0.49 0.07 0.79 0.61 0.07 0.95 Monroe/Dearborn 0.44 0.01 0.77 0.51 0.03 1.40 Monroe/State 0.48 0.09 0.81 0.54 0.06 1.39 Quincy/Wells 0.46 0.05 1.02 0.57 0.09 1.59 Randolph/Wabash 0.47 0.14 0.83 0.53 0.08 1.12 State/Lake 0.44 0.14 0.79 0.52 0.09 1.05 Washington/Dearborn 0.39 0.08 0.67 0.48 0.05 1.24 Washington/State 0.43 0.12 0.75 0.50 0.04 1.22 Washington/Wells 0.35 0.04 0.82 0.49 0.06 1.33 Note: All values in miles. Exhibit 9 u Quantifying Patterns: Average Nearest Neighbor Observed Mean Expected Mean Nearest Pattern Distance (miles) Distance (miles) Neighbor Ratio z-score p-value Distribution All LEED 0.069 0.064 1.085 1.170 0.242 Random Non-LEED 0.084 0.093 0.898 21.381 0.167 Random A S t u d y o f L E E D v s . N o n - L E E D O f f i c e B u i l d i n g s 1 2 3 Exhibit 10 u Quantifying Patterns: Moran’s I All LEED Moran’s I z-score p-value Pattern Distribution YB 0.030 0.422 0.673 Random RBA 0.130 1.242 0.214 Random Certification level 20.007 0.107 0.914 Random Certification points 20.099 20.670 0.503 Random Non-LEED YB 0.068 0.675 0.499 Random RBA 0.078 0.803 0.422 Random significance. The existence of HH and LL clusters are indicative of the existence of statistically significant similar values in the surrounding buildings. In contrast, the existence of HL and LH represents statistically significant spatial outliers. The third research question is: What is the level of differentiation among characteristics of LEED versus non-LEED buildings? Initially, building characteristics such as RBA, year built, and number of stories were assessed for their average and standard deviation trends between the two building groups. These same characteristics were also assessed within each of the four different types of LEED certification (Certified, Silver, Gold and Platinum) and between buildings with and without an ENERGY STAR label. Exploring the characteristics of buildings with and without an ENERGY STAR label provides additional insight on each of the two types Exhibits 18 and 19. These ENERGY STAR buildings are potentially more likely to pursue LEED because achieving the ENERGY STAR label indicates an embracing of the sustainability mentality [an ENERGY STAR score of 69 is a prerequisite for LEED under version 2009, but an ENERGY STAR label (75 score) is a prerequisite under version 4]. Three hypotheses are examined between LEED and non-LEED buildings, as well as those with and without an ENERGY STAR label. Hypothesis 1: Larger RBA buildings are on average sustainable (either LEED and / or ENERGY STAR label). The argument behind this assumption is that larger RBA buildings can attain significant operating expense reductions by adopting sustainable practices that reduce energy and water use significantly because of the building volume, allowing it to remain competitive. Hypothesis 2: Newer buildings are on average more sustainable (either LEED and / or ENERGY STAR label) because of the advanced building systems they benefit that allow for a lower retrofit cost. Hypothesis 3: The average number of stories is not differentiated between sustainable (LEED and / or ENERGY STAR label designations) and non- sustainable buildings. The fourth research question is: How do building characteristics affect the LEED points achieved? The two types of econometric models applied were a weighted fixed effects and a weighted least squares regression. The two different approaches J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 2 4 D e r m i s i Exhibit 11 u Density Analysis Based on Year Built Panel A: LEED Buildings Panel B: Non-LEED Buildings A S t u d y o f L E E D v s . N o n - L E E D O f f i c e B u i l d i n g s 1 2 5 J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u Exhibit 12 u Density Analysis Based on RBA Panel A: LEED Buildings Panel B: Non-LEED Buildings 1 2 6 D e r m i s i Exhibit 13 u Density Analysis of LEED Level Panel A: LEED Certification Levels Achieved Panel B: LEED Points Achieved A S t u d y o f L E E D v s . N o n - L E E D O f f i c e B u i l d i n g s 1 2 7 J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u Exhibit 14 u Density and Cluster Analysis Based on CTA Station Panel A: LEED Buildings Panel B: Non-LEED Buildings 1 2 8 D e r m i s i Exhibit 15 u Density and Cluster Analysis Based on Metra Stations Panel A: LEED Buildings Panel B: Non-LEED Buildings A S t u d y o f L E E D v s . N o n - L E E D O f f i c e B u i l d i n g s 1 2 9 J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u Exhibit 16 u Cluster Analysis Based on RBA Panel A: LEED Buildings Panel B: Non-LEED Buildings 1 3 0 D e r m i s i Exhibit 17 u Cluster Analysis of LEED Certification Levels Achieved Exhibit 18 u Descriptive Statistics of LEED and Non-LEED Buildings Mean Std. Dev. N RBA Year Built Stories RBA Year Built Stories All buildings 102 836,441 1979 36 539,351 26 18 All LEED 52 939,456 1983 36 333,741 20 12 Non-LEED 50 729,305 1974 35 678,753 31 23 LEED Certified 3 1,136,288 1964 44 296,018 27 5 LEED Silver 17 927,775 1980 35 359,731 20 13 LEED Gold 25 935,760 1984 36 265,375 20 11 LEED Platinum 4 892,932 2005 34 587,090 4 24 LEED in progress 3 901,645 1987 39 537,929 5 2 were used to analyze the effect of submarkets overall, as well as on an individual basis. The weighted fixed effects model assessed the effect of submarkets (overall dataset), as well as clusters generated with the use of ArcGIS for the most popular LEED certification levels: Silver and Gold (partial dataset). All regression models were weighted by RBA for consistency purposes. A S t u d y o f L E E D v s . N o n - L E E D O f f i c e B u i l d i n g s 1 3 1 Exhibit 19 u Descriptive Statistics of ENERGY STAR Label and Non-ENERGY STAR Buildings Mean Std. Dev. N RBA Year Built Stories RBA Year Built Stories All ENERGY STAR 73 914,657 1983 36 433,612 20 14 All Non-ENERGY STAR 29 639,552 1969 35 713,958 37 27 ENERGY STAR & Non-LEED 23 886,175 1982 36 604,390 18 18 ENERGY STAR & LEED 50 927,758 1983 36 334,232 20 12 Non-ENERGY STAR & Non-LEED 27 595,674 1967 35 720,312 37 28 Non-ENERGY STAR & LEED 2 1,231,902 1988 40 169,672 21 13 Exhibit 20 u T-test Results N Mean Std. Dev. t-Test Panel A: RBA Non-LEED 50 729,305 678,753 21.996 LEED 52 939,456 333,741 Non-ENERGY STAR 29 639,552 713,958 22.377 ENERGY STAR 73 914,657 433,612 Panel B: Year built Non-LEED 50 1974 31 21.799 LEED 52 1983 20 Non-ENERGY STAR 29 1969 37 22.557 ENERGY STAR 73 1983 20 Panel C: Number of stories Non-LEED 50 35 23 20.275 LEED 52 36 12 Non-ENERGY STAR 29 35 27 20.309 ENERGY STAR 73 36 14 The fixed effects models control for two distinct groups. The first group contains the allocation of buildings within one of the five downtown submarkets as defined by the CoStar Group [Exhibit 21, column 1, Equation 1; Exhibit 22 identifies the submarkets]. The submarkets were used to explore differences between them, which are rumored among the real estate professionals to exist. The second group is buildings based on their geocoding and spatial distribution by LEED Gold and Silver certification (Exhibit 21, columns 3 and 4, respectively, and Equation 2). Exhibits 23 and 24 show the spatial distribution of the groups. The groups J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 3 2 D e r m i s i Exhibit 21 u Regression Models Submarket Spatial Group (1) (2) (3) (4) Land size 0.01 0.00 20.02 0.15 (0.2) (0.11) (20.46) (2.66)* Number of stories 20.005 20.005 20.001 20.004 (21.99)* (22.41)* (20.16) (22.35)* Dummy Variables Wacker Address 0.12 0.12 20.01 20.07 (1.82)** (1.71)** (20.13) (20.64) Year built from 1980 and after 0.18 0.18 0.04 0.15 (1.73)** (1.87)** (0.19) (1.89)** Renovated 0.10 0.10 0.02 0.02 (0.99) (1.10) (0.11) (0.41) LEED 2009 version 0.29 0.30 0.29 0.32 (5.75)* (5.30)* (3.93)* (5.86)* LEED Silver Certification 20.19 20.19 (22.53)* (21.84)** LEED Gold Certification 20.10 20.11 (21.39) (20.91) Submarket Dummies Central Loop 20.13 0.13 Dropped (21.87)** (1.08) East Loop 20.22 Dropped 0.17 (21.98)* (1.38) West Loop 20.08 0.14 0.22 (21.23) (0.78) (3.71)* Constant 3.884 3.983 3.765 3.402 N 46 46 23 16 R 62.02% 61.81% 74.66% 96.50% VIF (multicolinearity test) 5.02 2.40 0.21 6.43 F-statistic 1.435 5.91 8.63 Note: t-statistics are in parentheses. *Statistically significant at the 5% level. **Statistically significant at the 10% level. for the two LEED certification levels were determined using ArcGIS based on the coordinates of each building, assuming the existence of at least one natural neighbor in common with another group of buildings (Delaunay triangulation); five groups were generated for Gold buildings and four for Silver buildings. A S t u d y o f L E E D v s . N o n - L E E D O f f i c e B u i l d i n g s 1 3 3 Exhibit 22 u Submarkets and Major Road Names ln(LEED points) 5 a 1 b LS 1 b NS 1 b WA 1 i 2 i 3 i 1 b YB80 1 b R 1 b L09 1 b LS 4 i 5 i 6 i 7 i 1 b LG 1 h 1 « . (1) 8 i i i ln(LEED points) 5 a 1 b LS 1 b NS 1 b WA 1 i 2 i 3 i 1 b YB80 1 b R 1 b L09 1 b CL 4 i 5 i 6 i 7 i 1 b EL 1 b WL 1 k 1 « . (2) 8 i 9 i i i Where: LS 5 Lot size; NS 5 Number of stories; WA 5 A dummy variable that takes the value 1 if the building has a Wacker Drive address and zero otherwise; J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 3 4 D e r m i s i Exhibit 23 u Groups of LEED Gold Buildings in Fixed Effects Regression YB80 5 A dummy variable that takes the value 1 if the building was built more recently than 1979 and zero otherwise. This cutoff year was determined based on the age distribution of the buildings and the data mean, which was 1979. R 5 A dummy variable that takes the value 1 if the building has been renovated and zero otherwise; L09 5 A dummy variable that takes the value 1 if the building received its LEED certification under the current LEED v.2009, and zero otherwise; LS 5 A dummy variable that takes the value 1 if the building is certified at the Silver level and zero otherwise; LG 5 A dummy variable that takes the value 1 if the building is certified at the Gold level and zero otherwise; CL 5 A dummy variable that takes the value 1 if the building is located in the Central Loop submarket and zero otherwise; EL 5 A dummy variable that takes the value 1 if the building is located in the East Loop submarket and zero otherwise; WL 5 A dummy variable that takes the value 1 if the building is located in the West Loop submarket and zero otherwise; h 5 The submarket specific characteristics; k 5 The spatial groupings specific characteristics; and « 5 The error term. A S t u d y o f L E E D v s . N o n - L E E D O f f i c e B u i l d i n g s 1 3 5 Exhibit 24 u Groups of LEED Silver Buildings in Fixed Effects Regression The weighted least square (WLS) model in column 2 of Exhibit 21 assessed the effect of the same variables from equation (1) on LEED points with the only difference being the exclusion of the submarket grouping and the inclusion of dummy variables representing the three submarkets (Central, East, and West Loop) where LEED buildings are mostly present. The inclusion of these variables allows the evaluation of the individual effect experienced by each of these submarkets: ln(LEED points) 5 a 1 b LS 1 b NS 1 b WA 1 i 2 i 3 i 1 b YB80 1 b R 1 b L09 1 b CL 4 i 5 i 6 i 7 i 1 b EL 1 b WL 1 « , (3) 8 i 9 i i where all the variables are as defined under equations (1) and (2). R e s u l t s A first step in assessing the spatial dynamics of LEED and non-LEED buildings was the visual representation of their locations. Exhibit 1 shows a mixed concentration of both building groups in the Loop area of downtown Chicago J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 3 6 D e r m i s i (highlighted in a semi-transparent circle). In contrast, the area west of the Loop, where newly constructed office buildings are located, as well as the two main commuter stations (Union and Ogilvie stations), shows a more pronounced concentration of LEED buildings. The presence of non-LEED buildings is more evident north of the river, especially along North Michigan Avenue above Ohio Street. The comparison of LEED (Exhibit 1) with ENERGY STAR label buildings (Exhibit 2) indicates that throughout the Loop area a significant number of non- LEED buildings have achieved the ENERGY STAR label, which is not the case for buildings north of the river. Achieving an ENERGY STAR label is important because it is a first step in energy conservation with long-term cost benefits for owners and tenants. The ENERGY STAR label can possibly be indicative of a future pursuit of LEED, due to the new requirement under LEED version 4. The first question focused on the spatial distribution of LEED versus non-LEED buildings, their distances and the links with mass transit rail stations. The three approaches applied provide both visual and quantitative evidence of different concentration patterns within each of the two groups, as well as differences in the proximity of buildings in each group (Exhibits 3–8). LEED buildings are mainly agglomerated in the Loop area, with an overspill to the West due to the proximity to the main commuter stations (Union and Ogilvie) and the new office construction activity along Wacker Drive during the last decade. This construction activity prompted a number of building managers to pursue LEED certification to remain competitive to the new stock, leading to this northeast-to-southwest directional distribution (Exhibit 3). In addition, Exhibit 3 includes the mass transit system stations (Metra and CTA) and their central features, allowing the comparison between mass transit and buildings spatial distribution, which is also shown in Exhibits 4 and 5. The Exhibit 3 ellipse characteristics are x-axis standard distance of 0.246 miles and y-axis standard distance 0.473 miles, with a rotation of 54.8 . On the other hand, the significant number of non-LEED buildings inside and north of the Loop creates a north-to-south trend, with a slight shift to the west (Exhibit 4). In Exhibit 4’s case, the x-axis standard distance is 0.314 miles and the y-axis standard distance is 0.704 miles, the rotation is 26.3 . Additional proof of the difference between the spatial concentrations of the two building groups is evident by the 0.226 mile difference between the two central points of each building group (Exhibit 5). The comparison between central feature of LEED buildings and Metra stations suggests a distance differentiation of 0.361 miles, while the distance to the central feature of CTA stations is 0.247 miles. The values for non-LEED buildings show some variation, with the central feature distances reaching 0.492 miles for Metra while the distance to CTA stations is only at 0.085 miles. The spatial dispersion between the two building groups is studied with a third approach, which determines the distance among neighboring buildings within the same group. The results indicate that LEED buildings are on average 20.9% closer to each other compared to non-LEED buildings. The similar result is evident with the maximum distance, indicating a tighter concentration for LEED buildings by 34.16% compared to non-LEED buildings (Exhibit 6). An analogous comparison, among the most popular certifications (LEED Gold and Silver), indicates that Gold buildings are located on average 18.05% closer to each other compared to Silver A S t u d y o f L E E D v s . N o n - L E E D O f f i c e B u i l d i n g s 1 3 7 buildings (Exhibit 6). The comparison among the five Metra station distance results indicates that LEED buildings are on average 13.8% closer to Metra stations compared to non-LEED buildings (Exhibit 7). The overall average distance across all buildings and the five Metra stations is 0.599 miles (Exhibit 7), in contrast to non-LEED buildings, which is 0.696 miles (Exhibit 7). Both LEED and non-LEED buildings achieve the same level of minimum distances of 0.05 miles (LEED buildings to Ogilvie station and non-LEED to Randolph station). Another significant observation is the significantly smaller maximum distances from any of the five Metra stations achieved by LEED versus non-LEED buildings (Exhibit 7). Shifting the focus on the local mass transit rail stations (CTA), the results are similar, with LEED buildings being located 12.02% closer to a CTA station compared to non-LEED buildings (Exhibit 8). The overall average distance across all buildings and the 23 CTA stations is 0.550 miles (Exhibit 8), in contrast to non-LEED buildings, which is 0.625 miles (Exhibit 8). Although in 65.22% of the cases the minimum distances of LEED buildings to CTA stations is larger compared to non-LEED buildings, in 100% of maximum distances LEED buildings are closer compared to non-LEED buildings (Exhibit 8). The second research question, which focuses on the clustering or randomness of the LEED and non-LEED buildings, was also explored through three different methods. Although the overall assessment of both groups of buildings suggests random patterns, the visual representation with density and cluster analysis hints towards potential clustering in some areas (Exhibits 9–17). Comparing the spatial distribution of both building groups with a random one, the results indicate that the patterns among both are random (Exhibit 9). The positive z-score among LEED buildings is indicative of a dispersed pattern, but the score is not sufficiently high to designate the pattern as dispersed when compared to a random distribution (Exhibit 9). In contrast, the negative z-score among the non-LEED buildings indicates possible clustering, but the score is not sufficiently high to designate the pattern as clustered when compared to a random distribution (Exhibit 9). Evaluating the spatial distributions of both building groups, while considering one of the building characteristics (RBA, year built, etc.), further reinforces the initial randomness result (Exhibit 10). The negative Moran’s I in two of the Exhibit 10 results indicates the existence of outliers among both the LEED certification levels and points achieved. Exhibits 11–17 provide a visual representation of the densities experienced throughout the study area, by generating surfaces showing the predicted distribution of a building’s characteristic (e.g., year built, RBA, LEED certification level, and points achieved), based on the value present at each location, as well as those in close proximity. Panels A and B in Exhibit 11 show the spatial distribution patterns of LEED and non-LEED buildings based on their construction completion. The results show the existence of aggregate patterns in close proximity to the two main Metra stations (Ogilvie and Union), although transportation is not taken into account in this case, and certain parts of the Loop area for LEED buildings (Exhibit 11, Panel A). The patterns tend to be very different for non-LEED buildings, indicating increased density of similar buildings J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 3 8 D e r m i s i in the middle of the Loop, as well as north of the river (Exhibit 11, Panel B). The existence of such densities in both cases (Exhibit 11, Panels A and B) shows that buildings of a similar age group in certain areas are aligned with each other in their decisions to pursue or not LEED certification. Shifting the focus to RBA (Exhibit 12, Panels A and B), the densities show the existence of aggregation patterns towards the Metra stations to the west and east for LEED buildings. In contrast, the non-LEED buildings seem to be denser in the Loop. The density analysis of the LEED certification level as well as the points (Exhibit 13, Panels A and B) provides evidence of similarities across certain areas as well (e.g., close to the two main mass transit stations to the west and a third one in the Loop). The existence of such density patterns, along with those seen in Exhibits 11–13, show that in certain areas evidence of peer building pressure may be a reality in order to remain competitive. A further examination of both density patterns and potential clustering helps explore the underlying trends in the dense downtown area. The results in Exhibits 14 and 15, which take into account mass transit, provide clear evidence of clustering when only the location of the buildings is taken into account regardless of their other characteristics. Some evidence of clustering is also seen based on the building’s RBA and the LEED certification achieved (Exhibit 16, Panel A, and Exhibit 17). The last two research questions focus on statistical trends and modeling of the dataset rather than spatial representation. The third question examines the level of differentiation among the characteristics of LEED versus non-LEED buildings. Exhibits 18 and 19 provide an assessment of the average and standard deviation trends experienced by LEED and non-LEED buildings, as well as buildings with and without the ENERGY STAR label. The comparison of average RBA, year built, and number of stories between the two building groups indicates that newer LEED buildings are larger in size, while the number of stories does not seem to differentiate between the two groups (Exhibit 18). The evaluation of building characteristics among LEED certification levels shows that the most frequent certification levels are Silver and Gold. LEED-Gold buildings are also larger on average and newer than Silver, while there was no difference based on the number of stories (Exhibit 18). The comparison of buildings with and without the ENERGY STAR label indicates that ENERGY STAR buildings are on average larger in size and newer compared to the non-ENERGY STAR buildings (Exhibit 19). The comparison of ENERGY STAR buildings which are non-LEED to those that are shows that those with both sustainability designations (LEED and ENERGY STAR) are larger in RBA; however, the average year built and number of stories do not show differentiation between the two (Exhibit 19). The t-test results on the RBA indicate that we can accept the hypothesis that LEED buildings are on average larger (RBA) than non-LEED buildings, with the same result being true for ENERGY STAR label buildings (Exhibit 20). Factors contributing to such an outcome can be the significant long-term operating cost reduction, based on the size of the footprint, and the continuation of the competitiveness of these buildings compared to the newer ones, which are usually smaller in size. The results of the second t-test support the second hypothesis regarding the construction timing of LEED versus non-LEED buildings, due to the newer A S t u d y o f L E E D v s . N o n - L E E D O f f i c e B u i l d i n g s 1 3 9 building systems which require less investment to achieve the sustainability standards (Exhibit 20). The same is true for ENERGY STAR label buildings. The results of the third t-test are also in agreement with the original hypothesis, indicating that the average number of stories does not show any differentiation between sustainable (either LEED and / or ENERGY STAR label designations) or non-sustainable buildings. The final research question assesses the effect of building characteristics on the points a LEED building can achieve. Columns 1 and 2 of Exhibit 21 explore the effect of building and other characteristics on the LEED points achieved with the difference being that column 1 is using a weighted fixed effects model and column 2 a weighted least squares model. The latter model (column 2) also assesses the effect on an individual submarket basis (Exhibit 22). The absence of any statistically significant difference between columns 1 and 2 underscores the stability of the models and their highlighted effects. Specifically, taller buildings achieve lower LEED points, with the results showing that a one-story increase is associated with a 0.5% decrease in the LEED points achieved (Exhibit 21, columns 1 & 2). In contrast, LEED points are higher for buildings with a Wacker Drive address by 12.6%, compared to all other LEED buildings (Exhibit 21, columns 1 & 2). This variable was included because Wacker Drive has seen an office building construction boom the last decades and buildings with this address represent the most prominent office stock in downtown Chicago. Buildings built from 1980 and beyond achieved 19.7% higher points compared to all other buildings, while those built under the current LEED version (v. 2009) experienced a 34.2% increase compared to the previous version of LEED (Exhibit 21, columns 1 & 2). The results in both columns also show that LEED Silver buildings achieve 20.5% less points compared to the other certification levels (Exhibit 21, columns 1 & 2). The F-statistic reported in column 1 shows that the generated submarket dummies were not statistically significant. Focusing on the submarket performance, the results in column 2 show that LEED buildings in the Central and East Loop experience a fewer number of points compared to the other submarkets by 14.3% and 24.4%, respectively. Shifting the focus exclusively on the points achieved among LEED Gold buildings (Exhibit 21, column 3; Exhibit 23 map) and LEED Silver buildings (Exhibit 21, column 4; Exhibit 24 map), the only common effect is the point increases both experienced under the current version of LEED compared to the previous. In both cases, buildings certified under LEED v. 2009 achieved a 34.2% increase in points for Gold buildings and 38.1% for Silver buildings. The F-statistic reported in both columns shows that the generated cluster dummies were statistically significant (Exhibit 21, columns 3 & 4). Other results of column 4 show that taller buildings achieved lower LEED points among LEED Silver buildings. Specifically, a one- story increase is associated with a 0.3% decrease in the LEED points achieved. Newer buildings, as well as buildings located in the West Loop submarket, achieve higher points under the Silver certification level by 16.6% and 25.1%, respectively (Exhibit 21, column 4). J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 4 0 D e r m i s i C o n c l u s i o n Exploring downtown Chicago’s Class A office building sustainability (LEED certification and ENERGY STAR label) adoption patterns, there is evidence of possible proximity pressures in the pursuit of LEED certification and a key link between mass transit rail stations and LEED versus non-LEED buildings. Benefiting from the use of geospatial and econometric modeling, evidence is provided for a number of trends, including the more concentrated pattern of LEED versus non-LEED buildings, especially in the most prominent areas. The proximity experienced among LEED buildings reached 21% compared to non-LEED buildings, while LEED Gold buildings were located 18% closer than Silver buildings. LEED buildings in the study are located 14% closer to metropolitan area commuter rail (Metra) and 12% closer to local commuter rail stations (CTA). Exploring the possible clustering of LEED buildings, the results show a random overall pattern, although small clusters are evident among groups of buildings. The comparison between average size (RBA) and year built for LEED and non- LEED buildings indicates that LEED buildings are larger and constructed more recently. The same is true for ENERGY STAR versus non-ENERGY STAR buildings. Finally, exploring the effect of buildings and other area characteristics on the LEED points achieved, the results show that taller buildings, those with Silver certification, and those in certain submarkets achieve lower points compared to other LEED buildings. In contrast, buildings with a prominent street address, as well as those constructed after 1979 and under the current LEED v. 2009, achieved higher LEED points than other LEED buildings. This trend is also maintained for LEED Silver buildings, with the exception of the prominent address, although one of the prominent submarkets shares the same effect. E n d n o t e s The dataset includes three LEED Core & Shell buildings with one of them already in the process of receiving LEED: Existing Buildings Operations & Management (EBOM). The standard deviation ellipse surface given in ArcGIS is applied in this study as follows: n 2 n 2 SDE 5 Ïo (x 2 X) / n; SDE 5 Ïo ( y 2 Y ) / n, where x and y are coordinates x i51 i y i51 i i i for building i, {X, Y} represent the mean center within each of the two buildings groups, and n equals 52 for the LEED buildings and 50 for the non-LEED buildings. The n 2 n 2 angle rotation is calculated as: u 5 (A 1 B) / C; where A 5 (o x 2 o y ); ˜ ˜ i51 i i51 i n 2 n 2 2 n 2 n 2 2 B 5 ; and C 5 2 , where and are Ï(o x 2 o y ) 1 4(o x y ) o x y x y ˜ ˜ ˜ ˜ ˜ ˜ ˜ ˜ i51 i i51 i i51 l l i51 i i l l the deviations of the xy 2 coordinates from the mean center. Source: http: / / resources.arcgis.com / en / help / main / 10.2 / index.html# / How Directional Distribution Standard Deviational Ellipse works / 005p0000001q000000 / . The average nearest neighbor ratio given in ArcGIS is applied in this study as follows: ANN 5 D / D , where D is the mean distance between each building and its nearest o E o neighbor: D 5 d / n and D is the expected mean distance for the features given in o i51 i E a random pattern D 5 0.5 /Ïn / A, where d is the distance between building i and its E i A S t u d y o f L E E D v s . N o n - L E E D O f f i c e B u i l d i n g s 1 4 1 nearest neighbor, n equals 52 for the LEED buildings and 50 for the non-LEED buildings, and A is the area of a min enclosing rectangle around each of the two building groups. The z-score is calculated as z 5 (D 2 D ) / SE, where SE 5 0.21636 /Ï(n / A). Source: o E http: / / resources.arcgis.com / en / help / main / 10.2 / index.html# / How Average Nearest Neighbor works / 005p0000000p000000 / . The Global Moran’s I given in ArcGIS is applied in this study as follows: I 5 n / S n n n 2 (o o w z z ) /o z , where z is the deviation of an attribute (e.g., year built, RBA, i51 j51 i, j i j i51 i i LEED levels, and points achieved) for building i from its mean (x 2 X), w is the spatial i i j weight between building i and j, w does not take a value, n equals 52 for the LEED buildings and 50 for the non-LEED buildings, S is the aggregate of all the spatial weights n n S 5 o o . The z -score is then computed as z 5 (I 2 (21 / (n 2 1))) / o i51 j51 I I 2 2 ÏE[I ] 2 E[I] . Source: http: / / resources.arcgis.com / en / help / main / 10.2 / index.html# / How Spatial Autocorrelation Global Moran s I works / 005p0000000t000000 / . The map units are based on feet due to the very close proximity of buildings. The Local Moran’s I given in ArcGIS is applied in this study as follows: I 5 (x 2 X) / 2 n S {o w (x 2 X)}, where x is an attribute (e.g., year built, RBA, LEED levels, i j51, j5i i, j j i and points achieved) for building i, X is the mean of the attribute x , w is the mean of i i j 2 n the attribute, w did not take any values, and S 5 [o w (x 2 X) ] / (n 2 1) 2 i, j i j51, jÞi i, j j X with n equal to 52 for the LEED buildings and 50 for the non-LEED buildings, n 2 the z 5 (I 2 E[I ]) /ÏV[I ], where E[I ] 5 2(o w ) / n 2 1 and V[I ] 5 E[I ] 2 I i i i i j51, jÞi i, j i i E[I] . Source: http: / / resources.arcgis.com / en / help / main / 10.2 / index.html# / How Cluster and Outlier Analysis Anselin Local Moran s I works / 005p0000001200 0000 /. A positive z-score is obtained when the observed mean distance is greater than the excepted mean distance. A negative z-score is obtained when the observed mean distance is less than the excepted mean distance. Due to the log regression models used in Exhibit 24, all the dummy variables require an adjustment to [exp(coefficient)-1]% in their explanation. R e f e r e n c e s Anselin, L. GIS Research Infrastructure for Spatial Analysis of Real Estate Markets. Journal of Housing Research, 1998, 9:1, 113–33. ArcGIS. http: / / resources.arcgis.com / en / help / main / 10.2 / Ayse, C. GIS and Spatial Analysis of Housing and Mortgage Markets. Journal of Housing Research, 1998, 9:1, 61–86. Clapp, J. and M. Rodriguez. Using a GIS for Real Estate Market Analysis: The Problem of Spatially Aggregated Data. Journal of Real Estate Research, 1998, 16:1, 35–55. Dermisi, S. Effect of LEED Ratings and Levels on Office Property Assessed and Market Values. Journal of Sustainable Real Estate, 2009, 1:1, 23–47. Dermisi, S. and J. McDonald. Effect of ‘‘Green’’ (LEED and ENERGY STAR) Designation on Prices / sf and Transaction Frequency: The Case of the Chicago Office Market. Journal of Real Estate Portfolio Management, 2011, 17:1, 39–52. Eichholtz, P., N. Kok, and J.M. Quigley. Doing Well by Doing Good? Green Office Buildings. Berkeley Program on Housing and Urban Policy. American Economic Review, 2010a, 2494–2511. J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 4 2 D e r m i s i ——. The Economics of Green Building. Berkeley Program on Housing and Urban Policy. Working Paper Series W10-003, 2010b. ——. Why Do Companies Rent Green? Ecological Responsiveness and Corporate Real Estate. Berkeley Program on Housing and Urban Policy. Working Paper Series W09-004, 2010c. Fuerst, F. and P. McAllister. An Investigation of the Effect of Eco-Labeling on Office Occupancy Rates. Journal of Sustainable Real Estate, 2009, 1:1, 49–64. ——. Green Noise or Green Value? Measuring the Effects of Environmental Certification on Office Values. Real Estate Economics, 2011, 39:1, 45–69. Jennen, M. and D. Brounen. The Effect of Clustering on Office Rents: Evidence from the Amsterdam Market. Real Estate Economics, 2009, 37:2, 185–208. Lipscomb C.A. An Alternative Spatial Hedonic Estimation Approach. Journal of Housing Research, 2004, 15:2, 143–60. Miller, N., J. Spivey, and A. Florance. Does Green Pay Off? Journal of Real Estate Portfolio Management, 2008, 14:4, 385–99. Miller, N. and D. Pogue. Do Green Buildings Make Dollars and Sense?, USD-BMC Working Paper 09-11, 2009. Montero, M.J. and B. Larraz. Interpolation Methods for Geographical Data: Housing and Commercial Establishment Markets. Journal of Real Estate Research, 2011, 33:2, 233–44. Nils, K., N.G. Miller, and P. Morris. The Economics of Green Retrofits. Journal of Sustainable Real Estate, 2012, 4:1, 1–22. Osland, L. An Application of Spatial Econometrics in Relation to Hedonic House Price Modeling. Journal of Real Estate Research, 2010, 32:3, 289–320. Thrall, G.I. GIS Applications in Real Estate and Related Industries. Journal of Housing Research, 1998, 9:1, 33–59. University of San Diego–Burnham-Moores Center for Real Estate, CBRE, and McGraw Hill Construction. Do Green Buildings Make Dollars & Sense? Green Building Study ver. 3.0, 2011. Valente, J., S. Wu, A. Gelfand, and C.F. Sirmans. Apartment Rent Prediction Using Spatial Modeling. Journal of Real Estate Research, 2005, 27:1, 105–36. Wallner, R. GIS Measures of Residential Property Views. Journal of Real Estate Literature, 2012, 20:2, 225–44. Wiley, J.A., J.D. Benefield, and K.H. Johnson. Green Design and the Market for Commercial Office Space. Journal of Real Estate Finance and Economics, 2010, 41, 228– Sofia Dermisi, University of Washington, Seattle, WA 98195 or sdermisi@ uw.edu.
Journal
Journal of Sustainable Real Estate
– Taylor & Francis
Published: Jan 1, 2014
