# RECORD 75ba68147bf6e34f9b631b70a8cea636eda11e65
# TITLE: Use of Geographically Weighted Regression (GWR) Method to Estimate the Effects of Location Attributes on the Residential Property Values
# YEAR: 2017.0  VENUE: nan  DOI: 10.22146/ijg.27036
# AUTHORS: M. Dziauddin; Z. Idris
# PAGES: 14  CANDIDATE TABLE PAGES (1-based): [6, 7]
# ABSTRACT: This study estimates the effect of locational attributes on residential property values in Kuala Lumpur, Malaysia. Geographically weighted regression (GWR) enables the use of the local parameter rather than the global parameter to be estimated, with the results presented in map form. The results of this study reveal that residential property values are mainly determined by the property’s physical (structural) attributes, but proximity to locational attributes also contributes marginally. The use of GWR in this study is considered a better approach than other methods to examine the effect of locational attributes on residential property values. GWR has the capability to produce meaningful results in which different locational attributes have differential spatial effects across a geographical area on residential property values. This method has the ability to determine the factors on which premiums depend, and in turn it can assist the government in taxation matters.


===== PAGE 1 (score 13, numeric-paren 0, table-words 0, var-words 13) =====

Use of Geographically Weighted Regression (GWR) Method to Estimate the 
Effects of Location Attributes on the Residential Property Values
Mohd Faris Dziauddin and Zulkefli Idris
Received: 08 04 2016 / Accepted:  24 08 2016 / Published online: 30  06 2017
© 2017  Faculty of Geography UGM and The Indonesian Geographers Association
Abstract This study estimates the effect of locational attributes on residential property values in Kuala Lumpur, Malay­
sia. Geographically weighted regression (GWR) enables the use of the local parameter rather than the global parameter 
to be estimated, with the results presented in map form. The results of this study reveal that residential property values 
are mainly determined by the property’s physical (structural) attributes, but proximity to locational attributes also con­
tributes marginally. The use of GWR in this study is considered a better approach than other methods to examine the 
effect of locational attributes on residential property values. GWR has the capability to produce meaningful results in 
which different locational attributes have differential spatial effects across a geographical area on residential property 
values. This method has the ability to determine the factors on which premiums depend, and in turn it can assist the 
government in taxation matters.
Abstrak Studi ini memperkirakan pengaruh atribut lokasi terhadap nilai properti residensial di Kuala Lumpur, Malaysia. 
Regresi berbobot secara geografis (GWR) diperkirakan memungkinkan penggunaan parameter lokal daripada parameter 
global , dengan hasil disajikan dalam bentuk peta. Hasil penelitian ini mengungkapkan bahwa nilai properti residensial 
terutama ditentukan oleh atribut fisik (struktural) properti, namun kedekatan dengan atribut lokasional juga berkontribusi 
sedikit. Penggunaan GWR dalam penelitian ini dianggap sebagai pendekatan yang lebih baik daripada metode lain untuk 
menguji pengaruh atribut lokasi terhadap nilai properti residensial. GWR memiliki kemampuan untuk menghasilkan hasil 
yang berarti dimana atribut lokasi yang berbeda memiliki efek spasial yang berbeda di wilayah geografis mengenai nilai 
properti residensial. Metode ini memiliki kemampuan untuk menentukan faktor-faktor dimana premi bergantung, dan 
pada gilirannya dapat membantu pemerintah dalam hal perpajakan.
Keywords:  Geographically weighted regression; Kuala Lumpur ; Location attributes; Residential property values
Kata kunci: Atribut lokasi; Kuala Lumpur ; Nilai properti perumahan; Regresi tertimbang secara geografis. 
1.Introduction
Location is acknowledged as an important 
attribute of immovable objects such as land and 
improvements to land. In the last five decades or so, 
many studies have been conducted to investigate the 
effect of location on land values and subsequently 
residential property values. In the standard practice 
of property valuation, property location is considered 
the most important factor in determining property 
value because the physical structure of the property 
depreciates over time. The land underneath the physical 
structure is what increases in value, and this is why 
the most common mantra we usually hear from real 
estate practitioners is ‘location, location and location’. 
The theory of property value formation examines 
location as a composite effect of a set of locational 
Mohd Faris Dziauddin
Department of Geography & Environment Faculty of Human 
Sciences, Sultan Idris Education University, 35900 Tanjong Malim, 
Perak, Malaysia.
Zulkefli Idris
UMW Technology Sdn. Bhd.16-1, Level 16, Menara Prestige No 1, 
Jalan Pinang, 50450 Kuala Lumpur. Malaysia.
Correspondent E-mail: faris@fsk.upsi.edu.my
attributes [Kauko, 2003]. In other words, attributes such 
as the central business district (CBD), parks, shopping 
malls, schools, rail transit systems, highway, etc. 
may contribute positively towards the formation of a 
residential property value, particularly in urban areas. It 
has to be noted that, the benefits of locational attributes 
are realised mainly in the form of externalities, hence 
they are collectively shared by a large number of people 
and houses [Kauko, 2003; Orford, 1999]. Externalities 
can be positive (beneficial) and negative (harmful). 
For example, rail transit systems can generate 
both positive (improve accessibility) and negative 
externality (traffic congestion and noise pollution) 
effects. Since an assessment of the relationship 
between residential property values and a locational 
externality effect will benefits the government, 
housing developers and home buyers, it is important 
to evaluate their relationship in monetary terms. 
The most commonly used method to estimate the 
value formation of a residential property in monetary 
terms is hedonic pricing model (HPM). HPM is a well-
established technique used for analysing a market for 
a single commodity with many attributes, in particular 
ISSN 2354-9114 (online), ISSN 0024-9521 (print)
Indonesian Journal of Geography Vol. 49, No.1, June 2017 (97 - 110)
DOI: http://dx.doi.org/10.22146/ijg.2323, website: https://jurnal.ugm.ac.id/ijg
© 2017 Faculty of Geography UGM and The Indonesian Geographers Association


===== PAGE 6 (score 40, numeric-paren 0, table-words 7, var-words 17) =====

USE OF GEOGRAPHICALLY WEIGHTED REGRESSION 
Mohd Faris Dziauddin, et all
102
from the Environmental Systems Research Institute 
support centre. The network–distance measurement 
using this programme requires three layers of spatial 
data: points of origin (observations), points of 
destinations (locational attributes) and road network 
data. This measurement allows the shortest route 
from each observation to the locational attributes 
to be calculated. Moreover, the Multiple Origins to 
Multiple Destinations programme enables more than 
one destination to be selected at any one time. Thus, 
proximity to the locational attributes can be calculated 
simultaneously for each observation. Table 1 presents 
a list of explanatory variables considered for inclusion 
in the HPM together with their descriptive statistics.
In all regression-based analysis, some explanatory 
variables are usually multicollinear. To handle this 
problem, the correlations among the explanatory 
variables used for the inclusion in the final models 
were detected by using Pearson’s correlation coefficient 
and variance inflation factors (VIFs). Following 
Orford [1999], a Pearson’s correlation coefficient 
above 0.8 and a VIF above 10 indicate harmful 
collinearity. This rule was applied in this study.
3. Results and Discussion
The results of the HPM and GWR models 
are presented in two stages: the first part shows 
the results from the HPM and the second 
part shows the results from the GWR model.
The HPM Estimation
According to the seminal works in the area of 
hedonic pricing of Rosen [1974], Freeman [1979b], 
Halvorsen and Pollawski [1981] and Palmquist [1984], 
four common functional forms are used to measure 
the relationship between the dependent and the 
explanatory variables in HPM: linear, semi-log, double-
log and Box-Cox transformations. Unfortunately, 
no studies have been conducted on how to choose a 
particular functional form in HPM except for the work 
of Cropper et al. [1988], which extensively reviewed 
the choice of functional form for HPM by previous 
researchers. This study concluded that linear, semi-
log, double-log and Box-Cox linear perform best with 
quadratic forms, with the quadratic Box-Cox faring 
relatively poorly. After performing and comparing 
these functional forms for the data in this study, we 
found that the semi-log functional form performed 
the best, and thus we used it estimate the relationship 
being studied. The semi-log regression equation 
is regressed on a set of determinants as follows:
lnPi = β0 + β1FREEHOLDi + β2FLOORSIZEi 
           + β3LOTSIZEi + β4BEDROOMSi + β5TYPE 
           TERRACEDi + β6TYPE_TOWNHOUSEi 
           + β7TYPE_SEMIDETACHEDi + β8TYPE 
           BUNGALOWi + β9PRIMARY_SCHOOLi + 
           β10SECONDARY_SCHOOLi + 
           β11SHOPMALLi + β12FORESTi + β13RAIL 
           STATIONi + β14MALAYi + β15CHINESEi + εi.......(4)  
Table 1. A list of explanatory variables and descriptive statistics of the model’s variables
Variable
Hedonic price model (HPM)
GWR model
Coefficients 
(β)
Standardised 
coefficients (β)
t-ratio
Lower 
quartile
Median
Upper 
quartile
Intercept
11.647
0.159
183.944
11.080
11.985
12.328
FREEHOLD
0.294
0.542
18.733
0.084
0.155
0.232
FLOORSIZE
0.005
0.086
49.578
0.004
0.005
0.005
LOTSIZE
1.860E-4
0.108
9.593
4.800E-4
4.030E-4
0.001
BEDROOMS
0.190
-0.043
11.497
0.076
0.132
0.173
TYPE_TERRACED
-0.084
-0.062
-2.841
0.038
0.290
0.391
TYPE_TOWNHOUSE
-0.223
0.124
-5.645
-0.210
0.024
0.179
TYPE_SEMIDETACHED
0.357
0.096
10.454
0.338
0.571
0.635
TYPE_BUNGALOW
0.442
0.031
9.869
0.182
0.361
0.497
PRIMARY_SCHOOL
2.832E-5
0.070
2.783
2.700E-5
1.400E-5
8.500E-5
SECONDARY_SCHOOL
5.244E-5
-0.112
6.881
9.000E-6
3.900E-5
1.070E-4
SHOPMALL
-8.112E-5
0.052
-11.385
-8.000E-5
-5.600E-5
-4.000E-6
FOREST
-2.973E-5
-0.018
-5.934
-2.200E-5
-4.000E-6
8.000E-6
RAIL_STATION
-0.036
0.117
-2.132
-0.073
-0.019
0.030
MALAY
0.041
0.067
12.629
0.006
0.033
0.070
CHINESE
0.025
7.416
-0.037
-0.002
0.025
Notes: Goodness of fit: Adjusted R2=0.73 (HPM); adjusted R2=0.85 (GWR). AICc=5663.63 (HPM); 
AICc=3185.55 (GWR)


===== PAGE 7 (score 43, numeric-paren 0, table-words 7, var-words 20) =====

Indonesian Journal of  Geography, Vol. 49 No. 1, June2017: 97- 109
103
where i is the subscript denoting each property, Pi is 
the price of property i in Malaysian Ringgit (MYR), ln 
is the natural logarithm; FREEHOLD is the dummy 
variable that indicates the status of ownership of 
the property, FLOORSIZE is the floor area of the 
property measured in square metre, LOTSIZE is the 
lot size of the property measured in square metre, 
BEDROOMS is the dummy variable that indicates 
a property with four or more bedrooms, TYP xxx 
is the set of dummy variables that illustrates the 
type of house, and it is further described as follows:
TYPE_TERRACED is 1 if the property is cluster, 0 
otherwise;
TYPE_TOWNHOUSE is 1 if the property is town house, 0 
otherwise;
TYPE_SEMIDETACHED is 1 if the property is semi-
detached, 0 otherwise;
TYPE_BUNGALOW is 1 if the property is bungalow, 0 
otherwise.
PRIMARY_SCHOOL, 
SECONDARY_SCHOOL, 
SHOPMALL and FOREST are the network–distance 
variables measured in metres from the property to 
primary schools, secondary schools, shopping malls 
and forests, respectively. RAIL_STATION is the 
dummy variable that indicates properties within a 2 km 
radius to the nearest station. The 2 km radius is chosen 
in accordance with the theory of the rail transit system–
residential property value relationship. MALAY and 
CHINESE are the percentages of residents from the 
Malay and Chinese ethnic groups. β0,...,β15 denotes a 
set of parameters to be estimated on the explanatory 
variables 
(including 
the 
intercept 
term), 
and 
εi is the standard error of the estimation assumed 
to be independently and identically distributed.
The parameter estimates for HPM together with 
the GWR model are shown in Table 2. In general, 
HPM explains approximately 73% of the variation 
in the dependent variable, consistent with other 
published empirical results. In the final model, all of 
the explanatory variables that influenced residential 
property values were significant at the 1% level.
Note that explanatory variables were incorporated 
in the final model based on significant coefficient 
values, and they alleviated the potential issues of 
multicollinearity. As the model is a semi-log functional 
form, the interpretation of the parameter estimates 
relates to their proportional (or their percentage 
when multiplied by 100) effect on price. Among the 
structural attributes, property size, which is measured 
by floor size (FLOORSIZE), is the most significant 
contribution to the value formation of a residential 
property. Specifically, for every square metre increase 
in floor size, the expected selling price of a residential 
Table 2. Estimation results of the global (HPM) and local (GWR) models (n=4393)
Variable
Hedonic price model (HPM)
GWR model
Coefficients 
(β)
Standardised 
coefficients 
(β)
t-ratio
Lower 
quartile
Median
Upper 
quartile
Intercept
11.647
183.944
11.080
11.985
12.328
FREEHOLD
0.294
0.159
18.733
0.084
0.155
0.232
FLOORSIZE
0.005
0.542
49.578
0.004
0.005
0.005
LOTSIZE
1.860E-4
0.086
9.593
4.800E-4
4.030E-4
0.001
BEDROOMS
0.190
0.108
11.497
0.076
0.132
0.173
TYPE_TERRACED
-0.084
-0.043
-2.841
0.038
0.290
0.391
TYPE_TOWNHOUSE
-0.223
-0.062
-5.645
-0.210
0.024
0.179
TYPE_SEMIDETACHED
0.357
0.124
10.454
0.338
0.571
0.635
TYPE_BUNGALOW
0.442
0.096
9.869
0.182
0.361
0.497
PRIMARY_SCHOOL
2.832E-5
0.031
2.783
2.700E-5
1.400E-5
8.500E-5
SECONDARY_SCHOOL
5.244E-5
0.070
6.881
9.000E-6
3.900E-5
1.070E-4
SHOPMALL
-8.112E-5
-0.112
-11.385
-8.000E-5
-5.600E-5
-4.000E-6
FOREST
-2.973E-5
0.052
-5.934
-2.200E-5
-4.000E-6
8.000E-6
RAIL_STATION
-0.036
-0.018
-2.132
-0.073
-0.019
0.030
MALAY
0.041
0.117
12.629
0.006
0.033
0.070
CHINESE
0.025
0.067
7.416
-0.037
-0.002
0.025
   Notes: Goodness of fit: Adjusted R2=0.73 (HPM); adjusted R2=0.85 (GWR). AICc=5663.63 (HPM);    
   AICc=3185.55 (GWR)


===== PAGE 8 (score 14, numeric-paren 0, table-words 4, var-words 6) =====

USE OF GEOGRAPHICALLY WEIGHTED REGRESSION 
Mohd Faris Dziauddin, et all
104
property increases by 0.5% (at the mean value, this 
equates to a premium of MYR 4331 or USD 1379, with 
the FOREX rate at MYR 3.14 or USD 1.00 in December 
2010). Among the locational attributes, distance to a 
local shopping mall is the most statistically significant. 
The model suggests that for every metre away from a 
local shopping mall, residential property values are 
likely to decrease at approximately 0.008% (at the 
mean value, this equates to MYR 70 or USD 22), thus 
signifying a strong appreciation of house buyers for this 
locational attribute category. The distance to rail transit 
stations (RAIL_STATION) is the least statistically 
significant locational attribute, and it has a counter-
intuitive sign. Residential properties are estimated to 
be sold at discounts of up to 3.6%, which is equal to 
MYR 31,181 (USD 9930) at the mean value, in a radius 
of 2 km to the nearest station. Among neighbourhood 
variables, the percentage of residents from the Malay 
ethnic group contributes most significantly to the 
value formation of a residential property. For every 
1% increase in ethnic Malay (MALAY), residential 
property values are likely to increase by about 4.1 
%, which is equal to MYR 35,512 (USD 11,309).
In all regression-based analyses, the independent 
variables possess various units of measurements, and 
thus their relative importance to the dependent variable 
is difficult to assess. To help overcome this difficulty, 
the computer-intensive LMG method developed by 
Lindeman et al. [1980] can be used to decompose 
R2 [Grömping, 2006]. The LMG method can be 
written in the following form [Christensen, 1992]:
where n(S) is the number of variables included in 
the set of regressors S. Therefore, LMG is defined 
as the average contributions in models of different 
sizes. The order of the regressors in any model 
is a permutation of available regressors x1, …, 
xp, which is denoted by the tuple of indices r.
When the unit of measurement for independent 
variables is standardised, the relative importance of these 
variables on the dependent variable can be determined. 
The results from the LMG method indicate that 
structural variables are more influential than locational 
and neighbourhood variables in estimating residential 
property values. Specifically, the structural attributes 
Figure 2. Shares of relative importance (LMG) for residential property 
values, OLS-Model (Note: The R2 with significant variables is 73%)