Beruflich Dokumente
Kultur Dokumente
$
%
&
'()*
+
,
-
.
.
+
/!+
'
/!+*
0
#
+
'0#+*
/
!
+
'/!+*
-
.
!
"-
(
%
.
%
% % % $
$$ -
-
- -
by
2017
DEDICATION & ACKNOWLEDGEMENT
The completion of this book has been possible with the assistance that I
have received from many people. Primarily, I must thank my supervisor,
Associate Professor Dr. Narimah Samat for her continuing support,
motivation, commitment and dedication in supervising my studies.
Without my supervisor, I would not have been able to reach this stage of
completion. I also would like to extend my thanks to Prof. Dr. Ruslan
Rainis, who had helped me provide solution to some of the main
problems I faced during the early and end stage of this study. Thanks are
also due to the chairperson of Geography section and Dean of School of
Humanities for all assistances while pursuing PhD at Universiti Sains
Malaysia (USM). My heartfelt thanks go to all the public and private
organisations that were willing to provide data for this research and their
staff who have been helpful in dealing with my enquiries regarding the
data. The DBKK had given permission for me to have their data through
Mr. Edmund Oktoberian Sepikit, Mr. Peter Liow and Stanley David
from Earth Info Sdn Bhd who, then, was willing to share some of the
data they had acquired and cleaned. To Mr. Lifred Wong, director of
valuation department in DBKK, who has my indebt gratitude by helping
me answered my queries about the valuation methods conducted in
DBKK including verified the results produced from this study.Beyond
that, I must thank Ms. Lauren Rosenshein, Geoprocessing Product
Engineer in ESRI, for her guidance regarding spatial statistics
communicated through emails. Thank you also to Rosmiyati Hasni and
Azizul Ahmad, postgraduate students from Geography Dept that assists
me a lot with the daily activities in USM. Denis Lajium from UMS has
been extremely helpful in providing me journal papers from Waikato
University in New Zealand.A debt of gratitude is also due to Universiti
Malaysia Sabah (UMS), my employer, which has been sponsoring my
studies and supporting me financially since October 2011. To all my
friends and colleagues, near and far, wherever they are now, I thank
them sincerely especially for their moral and technical support. Finally,
yet most importantly, I owe a great appreciation to my family especially
my parents, and my parents-in-law, for their prayers and moral support.
To my wife, Patriecia, and our three children, Joanne, Jojo and Jonjon,
their love, understanding and sacrifices are just invaluable to me. Above
all, I thank GOD for the knowledge and experiences that I had gained
along this journey.
ii
TABLE OF CONTENTS
CHAPTER 1 - INTRODUCTION
1.1 Preliminary ..................................................................................... 1
1.2 Research background...................................................................... 1
1.3 Problem statement .......................................................................... 6
1.4 Aim of the study ............................................................................. 9
1.5 Objectives of the study ................................................................... 9
1.6 Scope of study ................................................................................ 9
1.7 Summary of research methodology............................................... 10
1.8 Contribution of the study .............................................................. 11
1.9 Book structure .............................................................................. 12
v
6.4 Model comparison of OLS and SRM for the study area.............. 142
6.5 Visualizing and interpretation of property rating value map ....... 143
6.6 Conclusion.................................................................................. 151
APPENDICES........................................................................................ 188
vi
LIST OF ABBREVIATIONS
vii
ABSTRACT
According to Local Authorities Act 1976 (Act 171), property tax rates
are required to be valued every five years to accommodate the present
market value. Usually, the revaluation activity involves exhaustive, time
consuming and costly processes because it involves large areas and
many properties that are needed to be covered. There are various
property valuation models being developed using spatial statistics
method to estimate property values of large quantities in a short time
involving small manpower and at low cost. However, it has been
difficult to produce either one property valuation model that is suitable
for the study area or subdivides it to many models in order to ensure
accurate of the model produced. Furthermore, the type of variables used
in model development and which variables have the most influence in
determining the property rating also are difficult to be examined.
Therefore, the aim of this study is to develop property rating valuation
model in Kota Kinabalu City Hall (DBKK) using Ordinary Least
Squares (OLS) and Spatial Regression Model (SRM). The model
developed could accurately estimate the property rating and eliminate the
error. These spatial models are developed based on the 1997 residential
property valuation data obtained from DBKK and subsequently tested to
measure its accuracy and reliability. The study found that using
segmentation approach of the data based on different building types are
suitable to be represented with separate models. However, the overall
study area is best represented by SRM as OLS model contain spatial
autocorrelation error. Similarly, the model based on building type in this
study is also suitable to be represented by SRM. The findings shows that
data segmentation based on building type model performed better and
each building type model produced different number of significant
variables that influence the property rating value. This was proven based
on the big marginal accuracy differences of R2 achieved by the highest
building type model of intermediate terrace with 84.3% compared to the
overall model with 59.2%. The study also found out that each building
type has it own significant variables that influence the most of its value.
Spatial statistics can be used to produce residential property rating
valuation model. This approach is also suitable for large valuation
dataset with fast processing and low in cost. The model developed is
suitable to be used by DBKK or local authorities in management of the
property rating valuation data.
viii
CHAPTER 1
INTRODUCTION
1.1 Preliminary
This chapter discusses the introduction of the study that includes the main
elements of the book. It comprises the research background, problem
statement, the aim and objectives of the study and the general methodology
to be used. This chapter concludes with the contribution of this study and
finally, the structure of the book.
1
Table 1.1: Global proportion of the urban population increase
Urban Population
Year Proportion
(million)
1900 220 13%
1950 732 29%
2005 3200 49%
2030 4900 60%
Source: UN Population Division (2013)
2
Price
of
houses
P2
P1
DD Quantity of
houses
Q2 Q1
The value of a property or house price usually reflects the local economic
growth (Lin, 2010). In Malaysia, the residential property transaction values
gradually increased in parallel with the number of units sold as shown in
Figure 1.3 (CIMB, 2012). This figure illustrates property transaction from
1990 to 2012, which reflects rapid development of Malaysian economy
during this period.
3
Source: CIMB Property Market Report 2012
Figure 1.3: Residential transaction values and volumes 1990 - 2012
The properties values also have become the basis of property rating in
which the local authorities used to impose tax to the property owners
(Buang Alias, 2000). As stated in the Local Government Act 1976 Act 171
(2006), one of the tasks of the local authority in Malaysia is to collect
property tax annually, where the tax imposed is important for the relevant
authorities to generate revenues and continue to provide goods and services
for local consumption. Moreover, the property rating needs to be reviewed
every five (5) years such that the property will reflect the current property
market value. Therefore, a property valuation or revaluation is required to
accommodate the current value for the whole property in the area under the
jurisdiction of the local authority.
Valuation has been defined as the art and/or science of estimating values
(Millington, 2001:4). Property valuation is conducted by professional
valuer who will inspect the said property and assess the property value
based on the property characteristics such as its physical structure factor,
geography factor, location factor, economic factor of demand and supply,
4
and government policies (Ismail Omar, 1992). However, property location
which involves the accessibility and surrounding neighbourhood is the
most influential factor to the property valuation (Suriatini Ismail, 2005;
Kim, 2005; Theriault et. al., 2003; Wyatt, 1997; Goodall, 1972).
Various methods have been used to conduct property valuation but many
studies consistently mentioned five methods namely comparable method,
cost method, residual method, investment method and income method
(Scarrett, 2008; Richmond, 1985; Ismail Omar, 1992; Appraisal Institute,
1992). Another valuation method, however, has gaineds popularity
especially during this age of technology is the regression method (Brown,
1974; Gloudemans and Miller, 1978; Mark and Goldberg, 1988; Cannaday,
1989; Ismail Omar, 1992). Regression analysis is capable to model,
examine, explore spatial relationships, to better explain the factors behind
observed spatial patterns, and predict outcomes based on that
understanding. Thus, the regression was used to develop a mass appraisal
model.
5
(Dzurllkanian et al., 2006:2). Therefore, new method and new study is
needed to convince the local authority in Malaysia to adopt this approach.
6
In order to shorten the time taken to conduct property valuation for large
area, property value model with mass appraisal capabilities could
potentially be used as it has ability to peform valuation faster. This can be
done as mass appraisal technique provides uniformity and consistency in ad
valorem valuations particularly when revaluations of large number of
parcels at the same time (Deddis, 2002:4). Mass appraisal is suitable for
revaluation purpose and there are various methods that capable to perform
it (Löchl, 2010). For example, Ordinary Least Squares (OLS),
Geographical Weighted Regression (GWR) and Spatial Regression Model
(SRM) were some of the spatial statistics techniques commonly used to
conduct mass appraisal (Borst and McCluskey, 2009). The study by
Renshaw (1958), Brown (1974), Gloudemans and Miller (1978), Mark and
Goldberg (1988) and Cannaday (1989), for example conducted mass
appraisal using OLS approach. Ordinary OLS is one of the regression
techniques that provide a global model to understand and make prediction
of the variables in the study (Scott & Janikas, 2010). However, the OLS as
a global statistics model, only capable of providing a single regression
equation to represent the whole data in the region. Consequently, it
produced limited result (Fotheringham et al., 2002:6).
7
Therefore, another modelling method namely Spatial Regression Modelling
(SRM) specifically used to address the spatial autocorrelation error was
used (Suriatini Ismail, 2005; Löchl and Axhausen, 2010). The SRM also
called spatial econometrics was originated since early 1970s as a set of
technique to deal with spatial autocorrelation (Anselin, 2001). It has the
capability to detect the spatial autocorrelation in two different forms
namely, spatial error model and spatial lag model using the Lagrange
Multiplier (LM) test (Wilhelmsson, 2002). Thus, SRM managed to provide
good estimation in some property value model studies (Suriatini Ismail,
2005; Löchl and Axhausen, 2010) and potentially eliminate the model
error.
Based on the previous studies, it shows that different dataset or area can be
explained with different type of model. The OLS represents the global
model that normally used as a benchmark for comparison with other
models. Although it usually produced an undesired model but it could still
perform better if the data is significantly stationary. The GWR, being the
8
local model, was currently used by many studies in property valuation
modelling. It capable in dealing with spatial heterogeneity and has been
proven to be the best model in most property studies. In addition, SRM, as
an alternative model to accommodate spatial dependence is useful when
spatial autocorrelation error exists in the data. Based on different
capabilities highlighted in the model, there is no unique model that can be
used for all dataset. The selection of the model will depends on the nature
of the dataset undertaken.
Therefore, based on the issue above, some research questions for this study
were summarized as follows:
i. What is the suitable property rating model that should be used for the
local authorities of Malaysia based on the nature of the data?
ii. Is the model suitable as single model or should be segmented in
multiple models such as building type?
iii. What are the model’s variables that influence the most in determining
the property rating?
9
involves different parameters or data input which is difficult to make
comparison. Based on availability of the data, only 14 zones within DBKK
selected for this study. This study used ArcGIS 9.3 as data preparation,
database development and for OLS analysis. While GeoDa software was
used as an additional spatial statistics tool for spatial regression model
(SRM) and SPSS/PASW version 22 as a tool for statistical analysis of the
property valuation data. The spatial statistics tools of OLS and SRM were
used as the main tools provided by ArcGIS 9.3 and GeoDa. The tool in
SPSS/PASW and was also used specifically for additional model testing.
The second stage of the research is empirical. This stage involved model
development and addressed the use of spatial statistics to cover the issue of
multicollinearity, spatial autocorrelation and spatial heterogeneity in a case
study of Kota Kinabalu property value. GIS analysis was used to assists in
preparing spatial data, distance measurement for location factor variable to
be used in the model development and display the data pattern. The
procedures followed were guided by the framework identified in the
theoretical stage. Empirical data were collected and prepared before
property valuation models for ratings are estimated. Depending on the data
pattern, model estimation made use of three types of specifications, namely
OLS and SRM. The OLS models undergo specific testing for
heteroscedasticity and spatial autocorrelation. However, if the spatial
autocorrelation persists, it was then be addressed by SRM modelling
(Anselin, 2001:316; Suriatini Ismail, 2005:260; Löchl and Axhausen,
2010:42).
10
performance of the model. Then the model was validated and implemented.
Finally, at this stage, the important results were pulled together to draw
more general conclusions about spatial statistics in property valuation
modelling.
THEORITICAL
IDENTIFYING
RESEARCH ISSUE & AIM
LITERATURE &
THEORITICAL STUDY
EMPIRICAL
IDENTIFYING DATA &
GATHERING
GIS ANALYSIS
CONSTRUCT PROPERTY
VALUATION MODELS USING
SPATIAL STATISTICS
MODEL DIAGNOSTICS
EVALUATION
MODEL COMPARISON
MODEL IMPLEMENTATION
11
statistics methods suitable for mass property rating valuation model for
local authorities and subsequently analyse which variables that plays an
important roles in influencing the property value. Finally, it exemplifies the
usage of spatial statistics for mass residential property valuation model for
rating purpose in Malaysia for not only to Kota Kinabalu City Hall
(DBKK) but potentially for other local authorities in Malaysia.
Chapter 4 provide description of the study area and preparation of the data.
This involves the introduction of the geographical study area of the Kota
Kinabalu under the jurisdiction of Kota Kinabalu City Hall (DBKK) and
description of the modelling data in terms of its gathering, preparation and
quality. The studies expected to gather over 5000 dataset of various
property types in Kota Kinabalu using the available dataset used by DBKK.
The use of GIS at this stage is to map the data pattern of the study area and
it will also be conducted to accommodate the distance and accessibility
measurement for location variables.
12
also be discussed to identify the error and accuracy of the OLS model for
the whole area and each building type of the study.
13
CHAPTER 2
LITERATURE REVIEW
2.1 Introduction
This chapter reviews the main definitions, methods and previous research
regarding the study of property valuation modelling that needs to be
examined and subsequently produce the basis of the study and provided
information for the subsequent stages of empirical and evaluation activities.
Theoretical overview and literature review of property market, property
valuation, property tax, rating, GIS and spatial statistics including its
connection to each other were focused in this chapter. It highlights the
importance of spatial elements in property valuation analysis that were
undertaken in the past studies of property valuation modelling.
14
price of their business dealing (Smith and Corgel, 1987; Abdul Hamid,
2002).
15
The property valuation defined above applied for all property types
including residential, commercial and industrial properties (IAAO,
2013:10). The property valuation was used to estimate the value of the
property for various purposes such as for sale, purchase, mortgage, rental,
insurance, inheritance tax, stamp duty and rating for property tax (Shapiro
et al., 2009:5). Since this study was undertaken to estimate property value
for tax purpose, the following section will discuss property tax.
Generally, property tax was not based on personality (in personam) but in
kind (ad valorem) (Musgrave and Musgrave, 1984:331). This type of tax
could be recognised based on its characteristics. Among the characteristics
of property tax were stated as follows:
i. One of the main revenue sources of local authority. For example,
rates contributed 66.9% to the revenue of Majlis Perbandaran
Petaling Jaya in 1996 (Phang, 1997);
ii. The revenue and burden of this type of tax vary considerable from
one locality to another. Different local authorities may impose
different rates for similar type of property (Buang Alias, 2000);
iii. This type of tax applies to land as well as man-made capital
(buildings, machinery, and other items) (Buang Alias, 2000);
iv. Tax applied on real property which includes real estate (Santoso
Makmur Palal, 1989)
The property tax has become means to raise revenue or for other purposes
and had been exists and utilised the revenue system in about 130 countries
(Eckert et al., 1990).
16
provided in Act 171 of Local Government Act 1976 (2006). Under section
39(a) of the Act 171, properties located within the local authorities
boundary limits are subjected to this type of tax. As stated under section 2
in Act 171 of Local Government Act 1976 (2006), this tax is based on the
annual value which is the estimated gross annual rent at which the holding
might reasonably be expected to let from year to year the landlord paying
the expenses of repair, insurance, maintenance or upkeep and all public
rates and taxes. It also stated in Article 156 of the Federal Constitutions
that all lands, buildings or hereditaments occupied for public purposes by
or on behalf of the Federal Government State or public authority are
required to make contributions in aid of rates to the Local Authorities.
17
2.6.2 Basis of Assessment
Rates imposed by all local authorities are based on the annual value of the
property with the exception of the State of Johor which adopts improved
value (Ahmad Ariffian dan Hasmah Abu Zarin, 2001; Kwong and Mani,
1997).
Section 2 of Local Government Act 1976 (Act 171) (2006) defined the
annual value as “the estimated gross annual rent at which the holding might
reasonably be expected to let from year to year with the landlord paying the
expenses of repair, insurance, maintenance or upkeep and all public rates
and taxes”. While the improved value defined by the same act as “the price
that an owner willing and not obliged to sell might reasonably expect to
obtain from a willing purchaser with whom he was bargaining for the sale
and purchase of the holding”. Additionally, the act also stated that the
revaluation for these rates of all holdings was done once every 5 years or
within such extended period as determined by the State Authority.
For example, to calculate the rate of annual value, the equation 2.1 is used
and is stated as follows;
ൌͳʹ͵Ψ (2.1)
Where;
R = rates of annual value;
RV = the monthly rental value
18
The Local Government Act 1976 also provides for the use of different rates
based on the location or use of the property. The responsibility for the
determination of Annual Value or Improved Value, percentage rate and the
collection of rates lies with the respective Local Authority.
Once the rates have been estimated, the valuation list could then be
prepared before it is implemented by the local authority. This would be
discussed in the following section.
The new valuation list has to be prepared once every five years. However,
the state authority has the discretion to extend the time interval between the
preparations of valuation lists. In practice, due to the shortage of resources,
the old valuation list together with amendments which were made from
time to time would be generally adopted as the new valuation list in many
local authorities (Kwong and Mani, 1997).
Apparently, the inability of the local authority to prepare the valuation list
was also due to the method of assessment conducted in Malaysia
(Dzurllkanian Daud, 2006) in which traditional method was still being
used. The following section discussed the traditional and modern methods
of assessment used for residential property valuation.
19
2.7 Methods of Assessment
In order to estimate an appropriate property value, the right choice of
property valuation method was needed. Currently, there were various
valuation methods used in conducting property valuation where some are
from traditional methods and others are new or recent methods.
Another downside of this method was that the estimation of the property
value needed to be conducted manually for each property. Estimating a
large scale of property area, therefore, would require a lot of time and cost
(Tretton, 2007). Thus, an alternative method was needed to overcome this
problem.
20
2.7.2 Other Method in Residential Property Valuation
A different type of valuation method that gains momentum especially
during this age of technology was the regression method (Brown, 1974;
Gloudemans and Miller, 1978; Mark and Goldberg, 1988; Cannaday, 1989;
Ismail Omar, 1992). Although the regression method had been mentioned
in numerous property valuation studies (Brown, 1974; Gloudemans and
Miller, 1978; Mark and Goldberg, 1988; Cannaday, 1989; Ismail Omar,
1992), it has been seriously used after the advancement of computer and
statistics programs. This method started with a simple regression that
describes the relationship between one factor or variable in relation to
another in statistic term (McCluskey et al., 1997a). One of the factors,
however, must be a dependent variable (Y) and the other one was the
independent (X) variable. This method can be described in the equation 2.2
below (Ismail Omar, 1992:96):
ൌ ܽ (2.2)
Where;
Y is the dependent variable
X is the independent variable
a is the constant
b is the coefficient or a weight for the independent
21
shown in the diagram in Figure 2.1, most studies focused on these main
influence factors to the property value which consists of physical structure,
geography, economic, government policies and location factors which
would be described in the following sub topics.
WƌŽƉĞƌƚLJ/ŶĨůƵĞŶĐĞ
&ĂĐƚŽƌ
22
2.8.3 Economic Factor
General economic activity and economic well-being of the country as a
whole would also influenced the property values (Ring, 1970). Investment
in property in an undeveloped area would be highly unlikely to be as
attractive as in thriving area, unless the undeveloped area offers investment
possibilities at bargain-basement prices with a future prospect (Millington,
2001). Moreover, the fundamental relationship between current and
anticipated supply and demand and the economic ability of the population
must be analyzed to satisfy its wants, needs and demands through its
purchasing power (Appraisal Institute, 1992). Other than that, it has been
identified that the Gross Domestic Product (GDP) and the house price
index were the major macroeconomic determinants of commercial property
supply in which a unit increase in the house price index has caused supply
of shop lots to drop while a unit increase in the GDP increased the supply
of the said property (Abdul Hamid, 2002).
23
The importance of location towards property value was stressed out since
long time ago with Thunen (1826) produced a rental value model which
showed that rental value influenced by distance to the city or market centre.
Thus, land located near to the city was likely to produce high rental value.
While low rental value was obtained for the land located far from the city.
The usage of location factor also exemplified from a city location theory
model, in which rental value would decrease based on the increase of
handling cost and low production (Alonso, 1964). This study added that
different land use produced different type of rental value. In a recent note,
location factor had been observed as an important concept to any market
study that implies it as a position within clusters of properties with similar
neighbourhood criteria (Pearson, 1991) and this would be more relevance if
it consistently attract high or low value especially with the same existence
neighbourhood influence (Scarret, 2008).
24
Aspect 1 1 1
Aesthetic appearance 1 2 3
Well built 1 0 2
Fire resistance 1 0 0
Privacy 0 2 0
Detached residence 0 2 4
Parking 0 0 3
Source: Adapted from Daly et al. (2003)
Thus, location needed to be chosen properly as any bad location occurred
would be difficult to be rectified unlike bad building design which was
easy to be corrected (Abdul Hamid, 2002). However, choosing a good
location was a complex subject and difficult to be analyse (Abdul Hamid,
2006). Ring (1970) observed some of the location analysis that can be
considered in a property valuation;
i. Protection against inharmonious land uses
ii. Physical and social attractiveness
iii. Adequacy of civic, social and commercial centres
iv. Adequacy of transportation
v. Sufficiency of utilities and services
vi. Levels of taxes and special assessments
25
location of neighbourhood profiles and accessibility to service to interact
with the housing attributes to enhance property value assessment.
The discussion above shows that the main factors of property value
influence are required to obtain an accurate property valuation model.
However, the modelling process could only be conducted if variation
distribution occurred in the dataset (Osborne and Waters, 2002). The
discussion also highlighted the location factor as an important influence in
determining the property values. To provide an accurate measurement of
influence distance towards the property value, a more quantitative approach
can conducted by using GIS tools (Hening Widi Oetomo, 2003). This
would be discussed further in the following section.
Many studies had shown that GIS can effectively be used in describing
location attributes in property valuation (Fung et al., 1995; Wyatt, 1996a;
Plaut and Plaut, 1998; Din et al., 2001; &LFKRFLĔVNL and Parzych, 2006;
Abdul Hamid, 2007). For, example, GIS can effectively be used to produce
a value map, which is a spatial representation of statistical data that reflects
the value of property (Howes, 1980:7). Moreover at present, GIS was able
to address some of the problems inherent in traditional value maps by
producing them efficiently and as part of a wider suite of data analysis
techniques. By using spatial analysis in GIS, Wyatt (1996a) was able to
produce a better value map and at the same time enhances the valuer’s
understanding of locational influences on value. Besides that, the overlay
26
operation has also been used in the mapping of property values of Memphis
and Shelby County in 1980 and 1990 to describe the pattern of change
(Fung et al., 1995).
In the application of GIS in local authority, some case studies shows the
advantages of GIS to assists the local council in terms of facilities
management, tax assessment, asset management, population analysis and
environment monitoring and management (Taher Buyong, 1995; Wyatt and
Ralphs, 2003; Han and Yu, 2001; Gilfoyle and Thorpe, 2004). The usage of
GIS for tax assessment, in particular, greatly helped the local authority to
reduce cost and gain more collections for the purpose of development.
27
Source: &LFKRFLĔVNL and Parzych (2006)
Figure 2.2: Classification operation informs the property belong to certain
landuse type
Besides that, another GIS analysis, in the form of overlay operations had
also been used in the mapping of property values such as in Memphis and
Shelby County in 1980 and 1990; in which it described the pattern of
change (Fung et al., 1995). In addition to that, the study also presented
other GIS analysis such as address matching, data geocoding and data
query of the property.
28
Parzych, 2006; Zeng and Zhou, 2001; Lake et al., 1998; Des Rosiers et al.,
2000; Henneberry, 1998; and Chen, 1994). The ability to identify what
kind of objects and in what radius should be looked for could be applied
using this analysis to measure the property value. For example, in Figure
2.4, if the parcel was located inside any of the radius zone of 100m, 200m
and 500m, the neighboring parcels received similar influenced from the
noisy factory. This would decrease the property value with properties in
100m radius received the highest negative effect.
29
Source: &LFKRFLĔVNL and Parzych (2006)
Figure 2.5: Zones of given travel time from Bus Stop
Lastly, another popular GIS function applied in property study was the
interpolation analysis. This analysis could predict the unknown property
values from observed data of known property values especially when
involves wider scale of property. It has been empirically tested in the
western countries for property value prediction (Dubin, 1998; Martinez and
Lorenzo, 2000; Chica-Olmo, 2007) and also in Malaysia (Abdul Hamid,
2007; Taher Buyong, 2008b). For example, one of the studies applied a
kriging interpolation surface for residential property valuation in which an
output of prediction residuals (errors of prediction) across the study area
was produced as shown in Figure 2.6 (Abdul Hamid, 2007). The display
clearly showed the “bumps” and “potholes” in the map which represent
under valued and over valued areas of the sampled properties. However,
the continous surface value generated by the the interpolation analysis must
be used with cautious as not all surface has land value (Stylianidis et al.,
2008:320). This required someone with experience or familiar with the
place as for example, some “potholes” might represent lakes or hills.
30
Source: Abdul Hamid (2007)
Figure 2.6. Kriging interpolation surface of residual value of residential
property in Johor, Malaysia
However, these analyses are incapable to check any error within the data or
the validity of the output undertaken by each operation on the data. These
would result a wrong interpretation and incorrect solution in decision
making. It is only useful to provide distance measurement that would be
used to generate one of the location factor variables to be included in the
model. As a solution, spatial statistics was subsequently used by some
researchers to accommodate the limitations produced from the GIS analysis
31
(Krivoruchkoa and Gotway, 2003:2). The reason to that, the spatial
statistics has the ability to conduct estimation, prediction and hypothesis
testing. This method would be discussed intensively in the following topic.
The spatial statistics is the process of learning from data and answers the
questions that arise in spatial analysis which involves statistics (Sherman,
2011:1) such as;
i. What are the phenomena under study?
ii. What are the relevant data and how should it be collected?
iii. How should we analyze the data after it is collected?
iv. How we can draw inferences from the data collected to the
phenomena under study?
There are various spatial statistics methods can be used in order to answer
those questions above. This would be discussed in the following section.
32
could be used in trying to understand or predict by creating a single
regression equation to represent that process.
The OLS approach had widely been used for property valuation modelling
(Renshaw, 1958; Brown, 1974; Gloudemans and Miller, 1978; Mark and
Goldberg, 1988; and Cannaday, 1989) specifically for mass appraisal.
Recently, the OLS was enhanced by incorporating spatial element through
GIS to provide a better model especially when location is concerned
(Gallimore et al., 1996; Wyatt, 1996b; McCluskey et al., 1997b; Deddis,
2002; Gonzalez et al., 2002; and Suriatini Ismail, 2005). The usage of OLS
was still widely used even after the emerging of other spatial statistics
modelling such as SRM and GWR. However, it normally used for
comparison (Hernandez et al., 2003; Kim and Zhang, 2005; Lehner, 2011;
Taher Buyong, 2011).
Where,
y = Dependent Variable
ȕ0 = Constant
ȕ1x1 ȕnxn = Independent Variable Component
İ YHFWRURIHUURUWHUPV
Based on the equation 2.3, the predictions of the dependent variable are
obtained through a linear combination of the independent variables. This
basic regression of OLS was also adopted by GWR with some additional
element as described in the following topic.
33
using GWR (Bitter et al., 2006; Long et al., 2007; McCluskey and Borst,
2011; Taher Buyong et al., 2008a; Taher Buyong, 2011).
ݕ ሺ࢛ሻ ൌ ߚ ሺ࢛ሻ ߚଵ ሺ࢛ሻݔଵ ߚଶ ሺ࢛ሻݔଶ ڮ ߚ ሺ࢛ሻݔ
for i = 1 … n (2.4)
Where,
y = Dependent Variable
ȕ0 = Constant
ȕ1x1 ȕnxn = Independent Variable Component
u = vector of coordinates
The notation ߚ ሺ࢛ሻ indicates that the parameter describes a relationship
around location u and is specific to that location. A prediction may be
made for the dependent variable if measurements for the independent
variables are also available at the location u.
34
Lagrange Multiplier (LM) diagnostic was used to assess the spatial
dependencies. There are two types of diagnostic tests for LM, namely LM
(error) for spatial error dependence and LM (lag) for substantive spatial
dependence (Bell and Bockstael, 2000). Whichever test that produced
significant result, the LM model for that test will be used for the study as it
indicates the correct form of dependence. However, if both tests are
significant and have high values, the one with the highest value will prevail
(Anselin, 2005:200). The details of the two spatial models would be
explained as follow.
ൌ ɏ Ⱦ ɂ (2.5)
Where,
y = Dependent Variable
ȡ VSDWLDOFRHIILFLHQW
Wy = weight matrix for dependent variable
x = matrix of observations on the independent variables
İ YHFWRURIHUURUWHUPV
35
interaction (Suriatini Ismail, 2005). A spatial error model can be written in
2.6 as follows (Lehner, 2011:5):
Where,
y = vector of dependent variable
ȕ0 = Constant term
ȕ1x1 ȕnxn = Independent Variable Component
u= vector of spatially correlated error
Ȝ VSDWLDODXWRUHJUHVVLYHFRHIILFLHQW
W = spatial weight matrix
İ UDQGRPHUURU
The spatial error model in SRM was regarded as the most popular model
and widely used in real estate economics (Willhemsson, 2002; Suriatini
Ismail, 2005) compared to spatial lag model. Although SRM suitable to be
used if spatial dependence exists in the data, it however was unable to
accommodate spatial heterogeneity (Taher Buyong, 2011).
36
2.11.1 Multicollinearity
Multicollinearity exists whenever two or more of the predictors in a
regression model are moderately or highly correlated (Allen, 1997:176). In
statistical models, multicollinearity leads to large standard error of
estimators (Gujarati, 1999, 327). Thus, its presence may produce unstable,
misleading and incorrectly signed coefficients, the common symptoms of it
(Leishman, 2001). This error can occur when variable in the model facing
any situations as follow (Williams, 2012:2);
a. Improper use of dummy variables (e.g. failure to exclude one
category)
b. Include a variable that is computed from other variables in the
equation
c. Include the same or almost the same variable twice (height in feet
and height in inches; or, more commonly, two different
operationalizations of the same identical concept)
37
Positive autocorrelation is said to occur when high or low values for a
random variable tend to cluster in space. While a negative autocorrelation,
occurs when locations tend to be surrounded by neighbours with very
dissimilar values (Lee and Wong, 2001; Taher Buyong, 2006). However, to
attain good result when dealing with the model residual, a random pattern
must be achieved as positive or negative autocorrelation (low and high
residual) indicates a key variable is missing from the model
(misspecification) (Rosenshein et al., 2011; Suriatini Ismail, 2005; Orford,
1999). The tools normally used to detect spatial dependence or spatial
autocorrelation is Moran’s I (Moran, 1948).
Both spatial error and spatial lag problems can be identified and rectified
using SRM method. By using the SRM, the spatial dependant of the
property value would be specified and subsequently, an interdependent link
would be produced (Figure 2.7). This method of maximum likelihood
estimation is carried out in which the spatial dependent parameters are
estimated along with the coefficients (Anselin, 1988; Taher Buyong, 2011).
38
Source: Taher Buyong (2011)
Figure 2.7: Estimation of SRM with specified spatial dependance
39
There are two types of spatial heterogeneity that have been identified and
categorized as discrete spatial heterogeneity and continuous spatial
heterogeneity. While the discrete spatial heterogeneity generated from
different relationship of different sub-region or sub-market, the continuous
spatial heterogeneity take place when different relationship occur vary
smoothly across a region (Fotheringham et al., 2002). The tools normally
used to detect spatial non-stationarity or heteroscedasticity is Breusch-
Pagan test which would be explained in a later topic. GWR was a powerful
tool in exploring spatial heterogeneity (Taher Buyong 2011; Yu, 2010;
Chasco et al., 2007). Being the local model, GWR addresses spatial
heterogeneity by regressing in smaller portion across a study region, thus, it
permits this effect to be captured (Figure 2.8). As each regression portion
produces a set of estimated coefficients and variation in their values
between different portions, it enables the spatial heterogeneity in the region
to be evaluated (Zhang et al., 2009). Therefore, this method would be
adopted for this study.
40
2.12 Main statistical procedure for property valuation
modelling
This topic focuses on the main statistics procedures that are relevant with
the property value modeling. The modelling procedures usually based on
the modelling requirement. By fulfilling the modelling requirements, it
would surely help in obtaining an accurate property rating model with
correct variables chosen from the correct data and correct assumption with
valid interpretation of the estimated regression (Gujarati, 1995: 66). This
process would be discussed in the following section.
This was supported by Theriault et al. (2003), Ting Xu (2008) and Lin
(2010) in which their studies also focused on data selection in the first
stage. Some studies would scrutinize or categorize the data in the first stage
before the modelling were initiated (Tretton, 2007; Theriault et al., 2003).
This was conducted to ensure the quality of the data was not compromised.
Valuation accuracy and the quality of the result in the modelling depended
heavily of the data obtained as it was the foundation of the model.
Based on the discussion above, this study would adopt the ‘data selection’
as the first step of modelling and provide suitable measurement value or
dummy variable if necessary before the second stage begin.
41
Source: Tretton (2007:504)
This study, therefore, would use this approach as the second stage of the
modelling procedure and adopt the various testing as suggested to identify
suitable variables for the model.
42
Starting point Property attributes
Stepwise multiple
regression
procedure
Factor analysis of
interrelated independent
variables
Improve market
segmentation or use binary
no interactions with property
specifics
no
43
Source: Lin (2010:44)
44
There are three types most common use of functional form in multiple
regression models which are the linear, semi-log and log-log. The formula
of linear function of OLS, GWR, and SRM of Spatial Error and Spatial Lag
was shown in equation 2.3, 2.4, 2.5 and 2.6 from the previous section
respectively. While the semi-log (Malpezzi, 2002) and log-log (Azhari
Husin, 1990) is stated in equation 2.7 and 2.8 respectively as follow:
Where;
Ln P = a vector of log 2 of housing prices
Ln Sk = a matrix of log of structural characteristics,
k=1,…,K
Ln Lj = a matrix of log of locational characteristics,
j=1,…,J
Įȕk DQGȖj = corresponding parameters
İ = a vector of random error terms
Where,
Ln P = a vector of log of housing value or rent for the unit
X = a list of housing and neighborhood characteristics
Ti = series of dummy variables representing the time periods
ȕi = corresponding parameters
It was initially addressed by Can (1990) which pointed out that the property
value modelling was usually specified in a linear form. This was supported
by Azhari Husin (1990) that linear form was widely used compared to log-
log and semi-log.
4
“Ln” represents the normal log of base e (e= 2.718). “Log” represents the common log of base 10. “Ln”
is more common in real estate studies. All the log values in this study are based on the natural logs.
45
value as per transacted value which is straightforward to interpret. Adair et
al. (1996) highlighted that, although a linear model is convenient, it
imposes constraints on value response to change in attributes levels, that it
does not capture the non-linearity characteristic of housing. Nevertheless,
Watkins (1998) argues that property specific dummies that included in
linear model, such as different dummy variables for different number of
rooms, capable to capture the non-linear effect of housing characteristics on
prices. This implies that linear functional form is still feasible to be use for
property value modelling. Therefore, Based on the discussion above, this
study adopted the linear functional form. The equation stated in 2.3, 2.4,
2.5 and 2.6 would be applied in the study to develop the property valuation
modelling for rating purpose.
46
relationship or tell the same story towards the dependant variable. As
mentioned in the previous topic, multicollinearity will cause incorrect
coefficient of the model. To detect this error, stepwise regression, variance
inflation factors (VIF) and scatter plot can be used.
Previous studies by Theriault et al. (2003), Suriatini Ismail (2005) and Lin
(2010) show a preference for stepwise regression as a variable selection
procedure. The stepwise procedure is regarded as the best safeguard
measure against multicollinearity in hedonic modelling. This study adopts
the stepwise regression procedure and the “dropping variables” measure in
dealing with signal of multicollinearity in the variable selection stage
presented in Chapter 5.
VIF was used as conducted by Des Rosiers and Thériault (2008), Suriatini
Ismail (2005) and Orford (1999). VIF calculation is given in equation 2.9
(Gujarati, 1999:325) as follow;
ଵ
ܸ ܨܫൌ మ (2.9)
ଵିோ
Where;
R2i = the coefficient of determination of the regression equation for the
sample i
Scatter plot are other statistical methods that were used to measure the
degree of relationship between two independent variables in a model
(Taher Buyong, 2006:101) in which it provides graphic presentation of the
pairs (Wheeler and Tiefelsdorf, 2005:163). The usage of scatter plot for
47
multicollinearity was straight forward as straight line shape of the dots in
the graph would indicates high multicollinearity while scattered dots would
show no or low multicollinearity between two independent variables (Taher
Buyong, 2006:102).
As there are many method of multicollinearity test given, not all would be
apply for each model. This study would use two or three multicollinearity
tests for certain model as it consider sufficient to show the model was free
from this error.
The test statistic JB (Jarque and Bera, 1980) is defined in equation 2.10
with its sub equation in 2.11 and 2.12.
ଵ
ܤܬൌ ሺܵ ଶ ሺ ܭെ ͵ሻଶ ሻ (2.10)
ସ
Where;
n is the number of observations (or degrees of freedom in general);
S is the sample skewness, and
K is the sample kurtosis:
భ
ෝయ
ఓ σ ሺ௫ ௫ҧ ሻయ
సభ ష
ܵൌ ෝయ
ൌ యȀమ (2.11)
ఙ భ
ሺ σసభሺ௫ష ௫ҧ ሻమ ሻ
భ
ෝ
ఓ σ ሺ௫ ௫ҧ ሻర
సభ ష
ܭൌ ෝ రర ൌ భ మ (2.12)
ఙ ሺ σసభሺ௫ష ௫ҧ ሻమ ሻ
48
Where;
and = the estimates of third and fourth central moments,
respectively,
= the sample mean, and
= the estimate of the second central moment, the variance.
σ
సభ σೕసభ ௪ೕ ሺ௬ ିఓሻሺ௬ ିఓሻ
ܫൌ݊ (2.13)
ሺ σ మ
సభሺ௬ ିఓሻ ሻሺσ σಯೕ ௪ೕ ሻ
Where;
n = the number of observations,
wij = the spatial weights between observations,
yi = the value at location i, and
= the mean value of y
This test will describe the spatial pattern of the residual whether it is
clustered, random or dispersed. The residual must exhibit random pattern to
ensure that there is no spatial autocorrelation present.
49
value of the coefficient R2, known as the coefficient of determination, has
two (2) important interpretations (Taher Buyong, 2006; 105): i) It serves as
a strength measurement of the relationship between the variables; ii) It
explains the proportion of the variance in one variable accounted for by the
other variable. The R2 is calculated from the equation in 2.14 (Taher
Buyong, 2006: 168) as follow:
σሺ௬ෝഢ ି௬തሻమ
ܴଶ ൌ (2.14)
σሺ௬ ି௬തሻమ
In which;
yi is the data y at sample i
ݕത is the sample mean of data y
ݕො is the estimated value of yi with residual (error)
50
is deemed to be the best fit to the data (Schabenberger and Gotway, 2005).
However, as a rule of thumb, in cases where the differences between AIC is
less that around 3, the comparison is considered very close and therefore
there is no clear evidence as to which of the two models is better
(Fotheringham et al., 2002). The AIC calculation takes the following
formula in 2.16 (Akaike, 1974) as follow:
ା௧ሺௌሻ
ܥܫܣ ൌ ʹ݊ ሺߪො ሻ ݊ ሺʹߨሻ ݊ ቀ ቁ (2.16)
ିଶି௧ሺௌሻ
Where;
n = the number of observations in the dataset,
ߪො = the estimate of the standard deviation of the residuals, and
tr(S) = the trace of the hat matrix.
The t-test will assists in identifying which variable is significant that should
be at the level of 0.05 or less. If level 0.05 was chosen, the variable must be
statistically significant at 95% confidence level. This task usually called as
critical value of t or probability of t. This statistical test computes the
probability that the coefficient is actually zero. If a coefficient is zero (or
very near zero), the associated explanatory variable has very little impact
on the model or in other words, the variable unable to contribute
(Rosenshein et al., 2011; Taher Buyong, 2006).
The calculation for the t statistic for regression model is given in equation
2.17 as follow (Taher Buyong, 2006:193):
ିఉ
ݐൌ (2.17)
௦್
Where;
bi is the determined coefficient
ȕi is the corresponding population parameter which is equal to 0
51
ݏ is the std. deviation of bi
While the calculation for the critical value of t for regression model is
given in equation 2.18 as follow (Taher Buyong, 2006:193):
ݐ ൌ േݐഀǡିିଵ (2.18)
మ
Where;
Į WKHOHYHORIVLJQLILFDQFHUHTXLUHG
n-m-1 = degree of freedom
However, the above method could only be used if the model indicated
stationary relationship as it was not suitable for non-stationary relationship
(Rosenshein et al., 2011). Breusch-Pagan (BP) (Breusch and Pagan, 1979)
test is used to identify the stationarity of the model (De Graaf et al., 2001).
If the residuals follow a normal distribution (Fletcher et al., 2000), its use is
sufficient in testing for the presence of significant non-stationarity in a
model. Moreover, it can determine if the explanatory variable in the model
have consistent relationship (same story happen to all places) to the
dependent variable both in geographic and data space. Generally, the BP
test formula is given in 2.19 and 2.20 as (Breusch and Pagan, 1979);
Where;
ݑො ଶ = residual variance
ߚ = constant
ߚଵ = coefficient parameter
x = the independent variable
(2.20)
52
approach was normally used for non-stationary relationship to test the
individual parameters (Hope, 1968; Fotheringham et al., 2002; Brunsdon et
al., 1998a). Monte Carlo test is capable for not only to explore the variation
of the parameters, but can also test the significance of the variation (Leung
et al., 2000). This method creates random numbers observing how the
fraction reacts in every case. In effect, it is about the transformation of non
random problems in random form so as to facilitate finding a solution to the
problem via statistical sampling (Kyratso and Yiorgos, 2004). However,
Matthews and Yang (2012) pointed out that the mapping of Monte Carlo
test results does not provide the map reader with sufficient information to
be able to discern the areas where local parameter estimates have
significant local t-values. Ultimately, the study suggests the local t-value
needs to be visualized separately or overlay it with other map to better
illustrate the study outcome. The study by Bitter et al. (2007) adopted
Monte Carlo test for property modeling studies. That study managed to
detect spatial variation with 0.01 significant using Monte Carlo tests from
its seven parameters of dwelling area, lot size, number of storey, built prior
to 1940, structural quality, factor1 and factor2 that influence the house
price. The factor1 and factor2 parameters were derived using principal
component analysis (PCA) from 8 orginal factors.
53
ȕˆ ( X' X ) 1 ( X' X )Y
(2.21)
Where;
Y = the vector of dependent variable,
Ⱦ= the vector of coefficients,
X = the matrix of independent variables
54
model was first introduced (Lancaster, 1966). The linear model which was
later called OLS was one of the earliest approaches used in spatial statistics
to produce property value model. The development during that time,
however, faced difficulties in expanding the spatial statistics approach
because of lack of technology which limited its scope and appeal (Pace et
al., 1998). Nevertheless, since the 1990s, the usage of spatial statistics for
property value modelling was accelerated due to the rapidly growing GIS
software (Kulczycki and Ligas, 2007).
Other study which was also conducted in Toronto, Canada examined the
spatial effect of house price estimation when using GWR, Moving Window
Regression (MWR) and Moving Window Kriging (MWK) undertaken by
Long et al. (2007). The result showed that the GWR managed to achieve
55
slight advantage in accuracy estimate the house price than the other
modelling approaches. However, since spatial dependency was detected in
the model, it failed to provide test to eliminate the error such as using SRM
to probably achieved better result.
The study by McCluskey & Borst (2011) on property rating valuation
system for the residential properties was conducted in USA using GWR.
That study also found that spatial autocorrelation present in the residual for
all three counties (Catawba, Sarasota, Fairfax) in the study area after GWR
analysis was conducted. However, by using the segmentation or submarket
approach, the spatial autocorrelation error was gradually decreased when
the number of segmentation increased. That study showed that
segmentation or submarket approach was able to improve prediction
accuracy (R2) and reduce spatial autocorrelation (Morans I) in residual
errors. Eventhough the approach able to decrease the spatial autocorrelation
but it still failed to eliminate the error completely in the model.
The study undertaken by Löchl and Axhausen (2010) that developed the
hedonic property rents model undertaken in Zurich, Switzerland which
involved 8592 dataset showed that the GWR model was unable to eliminate
the spatial autocorrelation error. This prompts the study to use SRM, where
it managed to produce good accuracy of the predicted values. Its data,
however, were taken from public web site from December 2004 until
October 2005 based on asking rent and does not reflect paid market price.
Moreover, since that study used old data, it also did not inflate the rental
value to accommodate it to the present market value. Consequently, the
accuracy of the estimated value from the model is questionable.
56
location and time that influenced land value. The study showed that the
usage of GIS to assists the four models were beneficial. Both of these
studies however does not perform any model testing for error and also does
not compare it with other model such as GWR to observe the result in local
model.
57
visualised the estimated value in a GIS map. This system however, required
two separate softwares namely SPSS and Arcview GIS to obtain the
desirable result and did not perform other model such as GWR or SRM for
comparison. Moreover, model error testing was also not conducted in this
study.
Recently, few studies had been conducted for property value model using
spatial statistics particularly in GWR. One of the studies by Taher Buyong
(2008a) was conducted property values prediction on 196 single storey
terrace house for rating purposes in Kulai, Johor. The study compared
GWR with OLS in which the former standout to be a better model. The
GWR achieved high accuracy for adjusted R2 and was capable to capture
spatial variation of the data. That study however, did not perform any
model error testing and only small dataset was used. Therefore, the
accuracy of the estimated model is questionable.
Ibrahim Sipan (2009) applied and compared OLS, GWR, Kriging and
Spatial Hedonic Model (SHM) to produce an Automated Valuation System
for property rating in Kulai, Johor which involved 1500 transaction data.
The study shows that the OLS was the suitable model to be used in the
study area while the SHM, although is the best model, has calibration
problems to perform mass appraisal. Other study conducted by Ibrahim
Sipan et al. (2012) which used GWR method for mass property rating
assessment on 178282 properties in Selayang, Selangor. The result showed
that the GWR fared slightly better than the OLS model. That study
however, did not perform any test for spatial autocorrelation and
heteroscedasticity to verify the validity of the model.
58
In addition to the above studies, the study conducted by Taher Buyong
(2011) compared various spatial statistics models for mass appraisal on
residential properties in Johor Bahru, Johor based on 463 transaction
records and in Kajang, Selangor based on 1399 rental records. All records
acquired in year 2008. That study used OLS, GWR and SRM approaches.
The three models were tested for spatial autocorrelation and spatial
heterogeneity (heteroscedasticity). The result showed that GWR model
proved to be a better model for both study areas. SRM loses out to GWR
because spatial heteroscedasticity was more dominant compared to spatial
autocorrelation in the dataset (Taher Buyong, 2011). This study provides a
detail solution in property valuation modelling. However, the study was
conducted using small dataset compared to the vast size of Johor Bahru and
Kajang.
Previous studies had shown that OLS, SRM and GWR were preferred in
developing property valuation model. The selection of the model to be
used, however, could be undertaken based on few considerations. Firstly,
the selection was based on the present of spatial autocorrelation of the
model. For example, if spatial autocorrelation was detected, then SRM
shall be selected. Otherwise, OLS and GWR would be selected. Secondly,
the model accuracy can be increased by applying area segmentation or
submarket approach. As stated by McCluskey and Borst (2011), smaller
area would encourage GWR approach and produce a better accuracy of the
model. However, none of the previous study has tested building type as
their data segmentation as model accuracy could also be improved by using
this approach. Finally, the accuracy of the model was based on the
independent variables selected. The influence of the independent variables
or property valuation factors towards property rating need to be identified
and measured to find out the number of significant factors and which of it
that plays an important role to determine the property rating.
2.15 Conclusion
A review has been performed in this chapter, in which the property
valuation modelling was highlighted as the current popular method to
assists in property valuation work especially in rating purpose. The
importance of value influence factor including model testing was stressed
out in order to develop an accurate property valuation modelling. Other
than that, various spatial statistics modelling of OLS, GWR and SRM were
discussed as one of the approach to model the residential property
valuation. As stated from the previous studies, the OLS was suitable to
59
model the property value if the dataset is global or stationary in nature.
While the GWR model was best used if the dataset indicates local or non-
stationary pattern. In the event of when spatial autcorrelation unable to be
removed from the model, the SRM would help produced an alternative
model for the dataset. Some studies also highlighted the usage of
segmentation or subdivided approach to increase accuracy and reduced the
error of the model. This approach however, usually been conducted to
subdivided the area but segmentation based on building type has never
been tested yet. Another question of whether each segmented model use the
same variables or not and which variables play an important role, need to
be examined. Finally, the main important part in developing a model is to
have a good modelling process. The modelling process not only constitutes
the step-by-step process to produce a good model but also includes various
modelling tests that need to be performed. As a result from the outcome of
this chapter, it sufficiently can help this study to develop an accurate and
acceptable property valuation model for property rating valuation.
60
CHAPTER 3
RESEARCH METHODOLOGY
3.1 Introduction
Based on the discussion on the issues of property valuation methods and
spatial statistics in Chapter 2, it was found that there is a growing
recognition on the importance of spatial elements in property value
modelling studies. This study would then embark on the empirical
investigation. In this chapter, the conceptual framework, methodology of
the modelling process, model performance, the assessment and
visualization of the model output were discussed.
61
Figure 3.1: Conceptual framework of this study
As shown in Figure 3.1, the first stage namely input involved theoretical
study and data collection. It aimed to find evidence to substantiate the need
for the research and to obtain the data required for this study. At this stage
the property market value including its contributing factors and the spatial
elements in property valuation that needed for valuation were acquired. In
addition, attributes consists of physical building, geographical aspect,
neighbourhood, external facilities and legality represent the non-spatial
data were also compiled. Expert judgement was required to identify the
62
data scaling for some of these attributes. This was conducted using
questionnaire to obtain feedback from the valuers to ensure data
measurement used for the modelling are valid. Next, the location factors
were derived using GIS in which distance from each property location for
the spatial data was measured. The selected relevant data were then
gathered and examined by going through steps such as verification,
cleaning and conversion to prepare database suitable for analysis. The
discussion of the input stage was elaborated further in chapter 4.
Data gathered during input stage was used in the processing stage, where
analysis was performed using spatial statistic method to produce spatial
model based on the data acquired. Spatial pattern of the data need to be
identified first to choose which modelling method need to be applied. Once
the type of model was chosen and developed, it would run through a series
of tests and assessment to obtain a suitable property rating value model for
the residential properties in the study area. Detail discussion of these
models would be explained in section 3.3 below.
63
2011:45). The processing of OLS models were conducted using ArcGIS
software version 10. Figure 3.2 shows the Data Flow Diagram (DFD) to
perform the process of the global OLS model. Based on that figure, the
process of OLS analysis began firstly by scrutinizing the list of property
attributes which were obtained from DBKK database. Only the relevant
data used for property valuation was extracted from the list. Secondly, the
relevant data was set as the independent and dependent variables which
were recoded based on suitability of the residential property valuation
modelling process. Thirdly, the OLS were then run and the model output
was generated. Fourth, using the output run by the OLS analysis, model
tests were conducted to determine the reliability and acceptability of the
model. Finally, the tested model was then prepared for performance
assessment to be compared with SRM or GWR model.
64
Source: Adapted from Rosenshein et al. (2011)
65
Referring to the Figure 3.3 above, once the OLS analysis was conducted
using the selected property variables, multicollinearity test would be the
first test to be carried out. Pairwise correlation and variation inflation factor
(VIF) were used to identify the multicollinearity error. The pairwise
correlation was used to detect high correlation among the independent
variables, while the VIF measured redundancy among independent
variables. Any independent variable that produced correlation value of 0.4
or more (Lin, 2010:35) and VIF value larger than 7.5 (Rosenshein et al.,
2011:42) should be removed or modified.
Secondly, the spatial autocorrelation test was conducted towards the data’s
estimated residual. Spatial Autocorrelation (Moran's I) tool is used on the
regression residuals to ensure they are spatially random. Statistically
significant clustering of high or low residuals (model under and over
predictions) indicates a key variable is missing from the model
(misspecification) (Rosenshein et al., 2011). Thus, OLS results cannot be
trusted when the model is misspecified.
For the third test, the Jarque-Bera statistics test was conducted to detect
model bias. The Jarque-Bera statistic indicates whether or not the residuals
are normally distributed (Jarque and Bera, 1980) as model bias occurred if
the residuals distribution was not normal. The histogram would show that
the classic bell curve or Gaussian distribution was produced if the residuals
are normally distributed. When the p-value (probability) for this test is
small (is smaller than 0.05 for a 95% confidence level, for example), the
residuals are not normally distributed, indicating model misspecification (a
key variable is missing from the model). Results from a misspecified OLS
model are not trustworthy (Rosenshein et al., 2011).
Once the tests above were performed and any error were rectified or
minimized, than the fourth test for stationarity would be conducted. The
Koenker (BP) Statistic (Koenker's studentized Bruesch-Pagan statistic was
used as a test to detect stationarity (Breusch and Pagan, 1979). When
results from this test are statistically significant to non-stationarity, it
proceed to the fifth test of Joint Wald Statistics to estimate the overall
model accuracy and the robust coefficient standard errors and probabilities
were consulted to assess the effectiveness of each explanatory variable.
Regression models with statistically significant for non-stationarity is
especially good candidates for GWR analysis (Rosenshein et al., 2011:45).
However, if the Koenker test is statistically non-significant (stationary), the
Joint-F statistics or F-Test was used to measure the overall model accuracy
66
and subsequently leads to the usage of t-test (probability) to assess the
explanatory variable statistical significance.
The outcome at the end of these series of test, an error free and acceptable
global model of property value would be produced.
The GWR model procedure began firstly by acquiring the same variables
produced from the OLS model analysis. Since the variables were tested
from the OLS diagnostic tools, most of the error were reduced or
67
eliminated to before it can be used local model analysis. In the second step,
the GWR analysis was processed and the output was generated. Thirdly,
the model was tested using some proposed local model tests that had been
experimented by Rosenshein (2011), Rosiers and Theriault (2008), Bitter et
al. (2006), Wheeler and Tiefelsdorf (2005) and Hernandez et al. (2003).
Finally, the tested GWR model can be proceed to the assessment process
stage.
The GWR model need to be tested as some modelling error still existed
from local model which was detected from the previous study and
currently, the local model specifically GWR, has no established diagnostic
tool (Rosenshein, 2011; Rosiers and Theriault, 2008; Bitter et al., 2006;
Wheeler and Tiefelsdorf, 2005; Hernandez et al., 2003). However, this
study tested GWR model with the usage of scatterplot as proposed by
Wheeler and Tiefelsdorf (2005).
This study performed the tests as shown in Figure 3.5. This figure
illustrates the steps taken to test the GWR model. Firstly, the GWR model
testing began with spatial autocorrelation test using Moran’s I tool to
identify the residual pattern of the value in the study area (Charlton and
Fotheringham, 2009; Leung et al., 2000). Key variable need to be added if
spatial autocorrelation exists with residual patterns formed a non-random
pattern.
68
Source: Adapted from Wheeler and Tiefelsdorf (2005) and Charlton & Fotheringham,
2009
Figure 3.5: DFD of GWR model testing
69
2003; Rosenshein et al., 2011). Therefore a spatial regression model was
needed to rectify it (Anselin, 2001:316; Suriatini Ismail, 2005:260; Löchl
and Axhausen, 2010:42). Figure 3.6 below show how the spatial regression
process was conducted to identify the correct model that need to be
selected for the study. This procedure was conducted using GeoDa 3
software tool. It began by running the OLS regression. As shown in Figure
3.6 below, firstly, Lagrange Multiplier (LM) diagnostics was examined
once the OLS regression was initiated using the GeoDa software. If no
significant exist to any LM-Error or LM-Lag test estimation, this study
would proceed with the OLS result. Otherwise, if any of its value was
significant, then the designated spatial error model or spatial lag model
would be processed (Anselin, 2005). However, if both tests were
significant, then it would proceed to the next step of using robust LM
diagnostics. Secondly, similarly to the above steps, if any of the robust LM-
Error or robust LM-Lag was significant, then the designated spatial error
model or spatial lag model would be initiated. Thirdly, there would be a
rare case in which both robust LM tests were significant and if this happen,
the study would choose the model with the largest value for the test statistic
(Anselin, 2005).
3
GeoDa software was developed by GeoDa centre for Geospatial Analysis and Computation which was
founded by the new School of Geographical Sciences and Urban Planning Director Luc Anselin while at
the University of Illinois
70
Source: Adapted from Anselin (2005)
Figure 3.6: DFD of spatial regression decision process
71
3.4 Model Performance Assessment
After the models were tested, the performance of the models must be
assessed. Assessments were conducted for OLS, GWR and if necessary the
spatial regression model. This is to ensure the models were able to
represent the property value in the area compared with the present or actual
value.
72
3.5 Implementing and visualizing the findings
After the assessment and comparison, the estimated property rating could
now be displayed in a form of value map. This would assist the valuer
greatly in decision-making especially when involved many properties in
certain area. At the same time, it would showed the clustered area of high
and low values of the property rating in that area which would also benefits
to the valuer in their daily work by providing some rough estimation of the
property rating value. Point map would be utilized for vizualization and
would involve plenty of property value maps depending on the number of
models produced and chosen.
3.6 Conclusion
This chapter described the modelling methodology to be adopted in
developing the residential property valuation model using the spatial
statistics modelling of global and local model. By producing both global
and local models, the performances were compared to determine which
model would be better suited to be applied in the study area. In both models
however, a series of steps in variables selection and diagnostic tests were
conducted. Diagnostic tests in spatial statistics were conducted to detect
modelling problems such as spatial autocorrelation, multicollinearity and
heteroscedasticity which can lead to model misspecification, model bias
and measurement error. This problem needs to be rectified to obtain a
reliable and error free model. This chapter also elaborated some ways to
perform model assessment and compared it with the current data in the
study area to measure the general performance of the model. In the end, a
reliable and error free models were produced that could be implemented
and visualised for the authority to use.
73
CHAPTER 4
4.1 Introduction
Having discussed the theoretical part, the study continued with the
empirical investigation. This chapter consist of two stages of the study and
is organised as follows. The first stage discusses the study area in which
several theoretical and practical issues were considered before the
particular area was chosen. The area selected was also justified based on
the overview of the Malaysian housing market. After that, the data
gathering of the study area could then be conducted which involved
identification of the data requirements, investigation of the data availability
and data quality checking. The second stage involved data preparations that
consist of data verification, data cleaning and data format conversion. In
addition to that, the preparation of location factor variable that shows the
distance from the property location to the location factor was prepared
using GIS tool.
74
The city covers approximately 349.65 km2, with population made up of
Malay, Chinese, Indian and other 31 local ethnic groups including
Kadazan, Murut, Bajau, Kedayan, Sulu, Bisaya, Rumanau, Minokok and
Rungus. Modern and resorts Kota Kinabalu City with the multiple mix of
customs and cultures and its fast expanding cityscape, it is strikingly up to
date, hastening and happening here. Post-war reconstruction has altered the
city coastline, land area and skyline. Indeed, most of Kota Kinabalu city is
built on reclaimed land (Wah, 2010). Kota Kinabalu also has a population
of 462,963 while the larger urban area has an estimated population of
900,000. It is the largest urban centre in Sabah and the sixth largest in
Malaysia.
75
of 2012. The property price index showed rapid changes in the prices and
measured the pulse of the property market which was very useful for
property investors and potential home buyers. As shown in Table 4.1,
throughout the year between 2011 and 2012, the property price index in
Sabah showed a rapid increase when compare to the rest of the states in
Malaysia and consistently among the highest.
Table 4.1: Property price index in Malaysia by State
76
rating collection had shown steady increase of property value between
1998 and 2010 especially for residential properties. This value called the
rateable property and it was obtained based on the market rental value and
also using the valuation methods (Wong, 2011). Figure 4.2 and Figure 4.3,
for example, shows the rateable property value for the urban and sub-urban
area of Kota Kinabalu respectively. These figures illustrate that the area
experienced rapid increase of the rateable property value within short
period of time.
77
Figure 4.3: Total rateable properties in sub-urban area of Kota Kinabalu
from 1998 to 2010
The rating area in Kota Kinabalu is categorised by the DBKK into three
types of area which are the City, Urban and Sub-Urban. The rest of the area
78
is considered non-rating area. Each of the rating zones consists of different
way of rating calculating rate (DBKK, 2012). As stated by Local
Government Ordinance of Sabah 1961 (1997) under the provision of
section 72, DBKK is using gross annual rental value as its rateable value as
the rent at which any property might reasonably be expected to be at the
time of valuation. In addition to that, as to obtain the rateable value, be the
rent as herein before determined or a sum of five or ten percentum of the
market value of the house or building and land at the time of valuation. The
DBKK usually takes up the five percentum (5%) approach as its estimation
of the rateable value. The new rateable value that had been recognised and
approved by the parliament will then be used to impose tax to the property
owners for the next five years or until the next revaluation is done. This
approved rateable value list is called the tone of list or valuation list
(DBKK, 2012).
79
Source: Adapted from DBKK (2012)
Figure 4.4: Rating zone area of Kota Kinabalu District
80
Source: DBKK (2012)
Figure 4.5: Valuation list of DBKK based on year of implementation
The valuation lists for old rating areas are implemented in 1969 and 1975
while the valuation list for the new rating area was conducted in 1995.
However, the rateable values in these two valuation lists in the old rating
area were arbitrarily increased by 41% in 1981 to reflect the current market
value at that time (DBKK, 2012). Furthermore, another adjustment had
been undertaken based on the valuation made by an outsource company,
Smiths and Gore, in year 1997.
This study used the adjustment rateable values of 1997 because it showed a
much more appropriate value to reflect the current market value. This
adjustment value had also currently been used by DBKK for taxation
purpose and thus, it became the sample for the property value modeling
used in this study (DBKK, 2012).
There are two types of rates applied in DBKK for the Kota Kinabalu
District which are the General Rate and the Sewerage Rate. Based on the
DBKK annual rates order, each area of the Kota Kinabalu district has been
imposed with different rate (percentage from the rating value) which were
described in Table 4.3 below.
81
commercial and public sewerage)
residential 18% (Not connected
with public sewerage)
Commercial & 26% (Connected with
Residential public sewerage)
20% (Not connected
with public sewerage)
Urban and Commercial 21% (Connected with 1%
Sub-urban public sewerage)
15% (Not connected
with public sewerage)
Residential 13% (Connected with
public sewerage)
7% (Not connected
with public sewerage)
All undeveloped 5%
land ownership
except Native Land
that has been used
for agriculture
purpose
Source: DBKK (2012)
Due to data constraint, only the selected zones in city and urban area were
used for modelling purpose which includes Kota Kinabalu, Luyang,
Luyang Timur, Teluk Likas, Sembulan, Tanjung Aru, Damai, Kolam,
Ridge, Kepayan, Dah Yeh and Signal Hill as shown in Figure 4.6 below.
As stated in the scope of the study, this study only covered the residential
properties involves terrace, semi-detached, detached, and townhouse to
ensure manageable size of area to finish this study.
82
Figure 4.6: Selected zones for the study area in Kota Kinabalu
The process of valuation involved various stages before the rating value
can be endorsed to be implemented to the public (as shown in Figure 4.7).
The process begins by identifying the valuation case before the said
property was inspected and its value analyzed. The activities of searching
and selecting the relevant influence factors were conducted at this stage to
come out with the best value for the property. Afterwards, the value would
be certified and preceded to the billing stage in which the property owner is
required to pay the rating amount stated in the bill (DBKK, 2013).
However, the value can still be changed as any objection from the property
83
owner will be given a chance to appeal the rating amount as stipulated in
section 142 and 145 of Local Government Act 1976 (Act 171) (Malaysia,
2006). The final value can only be endorsed after the appeal period is over.
As there are many data involved in property rating in DBKK with more
than 30,000 residential properties in the area, human error or the nature of
the data itself could cause some problem to the valuation data. The
following section discussed the cause of this issue.
84
4.2.3.1 Record problems
Despite of the different in the type of building structure, the property was
identified under the same category. For example, the structure of a
detached house can be temporary, semi-permanent or permanent but the
category of that house was still taken as mere detached house.
Inconsistency in the recording of the type of structure would contribute to
the error in the valuation data.
In other case, some residential properties were used for personal business
purpose has made the model confused as although the size of the area is big
but low in value or small area size but high in value. These residential
properties were mainly used either as play school or showroom cum office.
Although this was legal but it was not recorded properly in the valuation
database. Therefore, it was difficult to be detected and cleaned.
In some cases, in order to keep up with the current value, inflation rate
factor was also taken into account by DBKK. Too depending with the old
tone of list would make the rateable value of property far outdated.
Other than that, quality neighbourhood, newer design and major renovation
have led to a higher rateable value of property that has been re-built.
Therefore, any incoming refurbished houses in the same housing estate will
tend to be assessed with higher rateable value as the comparable used will
be among those properties recently assessed. As a result, spatial
dependence exists.
Based on the discussion on the background of the study area, the following
section focused on data gathering undertake in this study.
85
the identification of data requirement, investigation of data availability and
data quality. The following sub-sections describe identification of data
requirement, data availability and data quality investigation carried out in
this study.
Since location factor becomes the main factor in this study, GIS analytical
operations played important role to generate the attributes of the variables
especially those involved distance or time. The location factors in this
study consist of accessibility measure from the said property to any
important landmark and neighbourhood characteristics that include the
socioeconomic characteristics as well as the environmental characteristics
of the area.
86
dLJƉĞŽĨĂƚĂ
WƌŽƉĞƌƚLJ
^ƉĂƚŝĂů
ƚƚƌŝďƵƚĞ
The spatial data consists of zone boundary, district boundary and cadastral
map which were represented in a form of polygon. While the location
factor and residential location was represented as point. Location factors
involved popular or hotspot location such as tourism centres, public
institutions, public recreations, public facilities, commercial areas,
government offices and religious centres.
In this study, most of the spatial data (obtained from DBKK) had a scale of
1:500 which include the layers of Zone, Location Factors, Road, Property
Points, Housing Park and Cadastral. Except the data of cadastre which was
updated in year 2007, the rest of the layers were updated in year 2012. All
spatial data obtained covers the area of DBKK in urban area.
87
The attribute valuation data of DBKK were based on the year 1997 when
the last revaluation exercise was conducted by DBKK. Most of the main
data were incomplete therefore it was gradually updated using JPPH data in
year 2012 to fulfil the modelling requirement. Although the property rating
value of 1997 data were maintained, other important variable or factor were
updated to complete the missing value such as the main floor area,
neighbourhood quality and building quality.
88
Table 4.4: Property Information (Attribute)
Variable Name Type Measurement Description
PID Number Scale Property ID
Postcode Text Nominal Postcode Number
House No Text Nominal House Number
Jln_Lorong Text Nominal Name of road or alley
Tmn_Kg Text Nominal Name of housing area or village
Lot No Text Nominal Lot Number
Blok No Text Nominal Block Number
Floor No Text Nominal The house unit floor number
Unit No Text Nominal Property unit number
Title No Text Nominal Property title number
Gazette No Text Nominal Property gazette number
Zone Text Nominal Zone name
Subzone Text Nominal Subzone name
Prop_type Text Nominal Property type
Bld_type Text Nominal Building type
Bldg_grp Text Nominal Building group
No_storey Number Scale Number of storey in the building
No_room Number Scale Number of room in the building
Floor_Level Number Scale Level of floor the property locate
Mezzne Text Nominal Mezzanine area
Mfa Number Scale Main floor area
Afa Number Scale Ancillary floor area
RCA Number Scale Reduced Covered Area Estimation
Ext_build Text Nominal Availability of building extension
Manval Number Scale Manual Valuation
Rateable_v Number Scale Rateable value
Oldrat Number Scale Old rating value
Transact Number Scale Property Transaction Price
Appeal Text Nominal Appeal received
Appdat Text Nominal Appeal Date
Cvalue Number Scale Compute Value
Planno Text Nominal Plan Number
Division Text Nominal Division Name
Source: DBKK (2012)
89
Table 4.4: (Continued)
90
Based on the recommendation from the previous studies (Scarrett, 2008;
Appraisal Institute, 1992; and Millington, 2001), 21 variables from the list
in Table 4.4 were suitable to be included in the model. In which, there were
19 structural features namely, number of rooms, number of storey, floor
level, availability of special facilities, availability of mezzanine, availability
of sewerage, main floor area, ancillary floor area, reduced coverage area,
land area, building quality, building type, property type, topography, access
to main road, site preparation, lot shape, sanitation type and property
frontage. While two variables reflected the surrounding features which
were neighbourhood quality and flood prone.
For the location factors, the variables were derived using GIS analysis
based on the location of the location factors. These analysis involved
Euclidean distance from the property to the location factors, which would
be described in section 4.5 below.
GIS was used to perform the analysis and visualized the results in maps.
There were 10 spatial data layers obtained from DBKK at 1:500 scales as
shown in Table 4.5. Data obtained included zones, sub-zone, roads,
cadastral, housing park, property, building outline, landuse and location
factors. The location factors consists of public institutions, public facilities,
main office building, cemetery and public hotspots such as shopping
complex, famous restaurants, sport complex, sport club and entertainment
club.
91
The graphical outputs of the spatial data mentioned above were also shown
in the Figures 4.9, 4.10, 4.11, 4.12, 4.13, 4.14 and 4.15 below. Figure 4.9
illustrates the residential property location, housing park and cadastral map.
While Figure 4.10 displays the data for DBKK zone and district boundary.
Other than that, the road and building outline are shown in Figure 4.11. The
location factors of religious centre, commercial centre and tourism are
displayed in Figure 4.12 and 4.13 respectively. Other location factor of
public facilities, public recreations and public institutions are shown in
Figure 4.14 and 4.15 respectively.
92
Figure 4.10: DBKK zone and district boundary
93
Figure 4.12: Location factors of religious centre
94
Figure 4.14: Location factors of public facilities
Some of the data listed in Table 4.4 were not included for modelling
process, but the land area, floor area and building quality are compulsory in
95
the valuation model (Goodman and Thibodeau, 1998; Fletcher et al., 2000;
Berry et al., 2003; and Stevenson, 2004). The floor area could not be used
since it has missing data and the available records were not sufficient to
perform the modelling process. However, the Reduced Covered Area
(RCA) had sufficient records to replace the floor area. The RCA which is
also called as gross external area, is the basis measurement for council tax
to all houses and bungalows. It measured the external area of the building
at each floor level (RICS, 2007). The following Table 4.6 illustrates the
areas that are included and excluded in RCA calculation.
However, the final selection of variables for modelling purpose would only
be known once the modelling had been processed as it is subjected to the
model’s test result and this would be explained in chapter 5. The following
section discussed data cleaning.
96
interpretations or findings. Thus, the first part of data cleaning was to
remove non-year 1997 data, non-residential use, and suspected error in data
entry or odd residential property information.
Originally, the study collected 14520 observations for the whole area of
DBKK through selection of residential property valuation data excluding
apartments, flats and condominiums within the urban area. Further
cleanings finally retained the data of 5625 records for the analysis after
removing of invalid residential property data. This data was then included
in the modeling process. However, as the study unable to cover the whole
data in the area, the usage of GWR was not possible as it only capable to
estimate area with complete data.
The details of the data availability based on zones are summarised in Table
4.7, which represented about 38.74% of the residential property data
recorded for 1997 by the DBKK (2012). This in line with the requirement
needed for sampling to be valid for statistics analysis as stated by Krejcie
and Morgan (1970:607-610). The following section discussed data
recoding process.
97
Table 4.7: Data availability based on zones in Kota Kinabalu
Zones No. of Data Availability %
BUKIT PADANG 838 402 47.97
DAH YEH 1628 814 50.00
DAMAI 836 299 35.77
FUNG YEE TING 251 72 28.69
KEPAYAN 1181 273 23.12
KOLAM 2159 937 43.40
LIKAS 509 81 15.91
LUYANG 1178 310 26.32
LUYANG TIMUR 1560 576 36.92
RIDGE 2930 1271 43.38
SEMBULAN 354 205 57.91
SIGNAL HILL 87 1 1.15
TANJUNG ARU 168 62 36.90
TELUK LIKAS 841 322 38.29
ALL ZONES 14520 5625 38.74
For each variable mentioned above, data scaling was conducted using
questionnaire which was distributed to the valuers in DBKK which consists
of Valuation Director, Valuation Officer and Valuation Assisstant. The
example of this questionnaire was shown in Appendix A. There were 5
respondents obtained for this study and each respondent required to provide
98
input based on scale 1 to 5 for each attribute of the variable. This scale is
adopted based from Likert Scale. The scale 1 to 5 was described as in Table
4.8 as follows:
Low scale was given for low value while high scale was given for high
value. Once all the respondents had provide the scale input for each of the
variable’s value, the mean score of the scale 1 to 5 can then be estimated to
identify which value is the lowest or the highest for the variable. For
example, the ‘neighbourhood quality’ variable with the value of ‘poor’ was
given the lowest scale of ‘1’as it has the lowest mean score of 1.4. While its
value of ‘exclusive’ was given the highest scale of ‘4’ as it has the highest
mean score of 4.6.
The scaling for the relevant variables was shown in Table 4.9. Based on
that table, low ranking was given for small mean score. While high ranking
was given for big mean score. Once data scaling was undertaken, then
format conversion activity was conducted and explained in the following
topic.
99
No (1) 1.2
Site Preparation
Yes (2) 3.6
Irregular (1) 1.8
Lot Shape Eccentric (2) 2
Compact (3) 3.6
Nil (1) 1.4
Uniformed road reserved (2) 3.4
Frontage
Main Road (4) 3.6
Secondary Road (3) 3.8
Poor (1) 1.4
Average (2) 2.8
Neighbourhood Quality
Good (3) 4
Exclusive (4) 4.6
Yes (2) 1.4
Flood Prone
No (1) 4
No Access (1) 1.4
Bridge (2) 1.4
Road Access Earth (3) 3.2
Gravelled (4) 3.4
Sealed (5) 4.8
None (1) 1.2
Septic Tank (2) 3.6
Sanitation Treatment Plant Private (3) 4
Treatment Plant DBKK (4) 4
Public Sewer (5) 4.2
100
4.4.4 Data format conversion
Integration or joining of the data was required to enable it to be used by the
selected tools to carry out the property modelling tasks. Various softwares
were used to perform modelling function. SPSS 16.0 was used to undertake
classical regressions, ArcGIS 9.3 was used to perform OLS and to measure
distances as well as for general data management, GWR 3.0 software was
used to run the GWR local analysis and as an additional tool for GWR
model testing and Geoda was used to run the SRM. This means that the
5625 observations data need to be in the .sav, the .shp and the .txt formats
respectively. The workflow diagram for the format conversion activity is
shown in Figure 4.16. Apparently the data were ready for analysis, but
those data were separated and in different format. The majority of the
attribute data was from DBKK in MS Access database .mdb format and
some from JPPH in MS Excel .xls spreadsheet, while the spatial data is in
ArcGIS .shp format.
Based on Figure 4.16, the main source of database for this study was the
DBKK property attribute which was stored in MS Access (.mdb) format.
Most of the variables used for property valuation were included. The other
property attribute obtained from JPPH which was stored in MS Excel.(.xls)
format was used as an additional source to update any missing data or old
data in the main database. Once the data was updated, verified, cleaned and
recoded, the data were integrated with the spatial data that provide the
location of each residential property involved. The data were joined with
shapefile (.shp) format to be prepared for GIS analysis which involves
measuring the distance from the residential property location to the nearest
location factor. After the GIS analysis process, the data were also converted
to .sav and .txt to perform the property valuation modelling.
101
Figure 4.16: Workflow for the data format conversion process
The distance measurement using GIS analysis was the last step for data
pre-processing before the modelling was performed. This was explained
further in the next section.
The location factors obtained from this study are categorized as tourism
centres, public institutions, public recreations, public facilities, commercial
102
areas, government offices and religious centres. The individual location
factor that included in each category mentioned above is shown in Table
4.10 below.
Table 4.10: Location factor in this study based on category and individual
factor
Category of Location Factor Individual Location Factor
Tourism centres Market
Public institutions School
Public recreations Sports area, Field, Park
Public facilities Bank, Hall, Hospital, Post Office, Police Centre,
Cemetery
Commercial areas Shopping Mall, Supermarket, Commercial Centre
Government offices Government administration office
Religious centres Church, Mosque, Chinese temple, Hindu Temple
103
Source: Adapted from Hening (2003)
Figure 4.17: Workflow diagram of the location factor distance analysis
104
Figure 4.18: Distribution of property rateable value in RM for the 5625 data
105
Figure 4.19: Morans I output for the 5625 data
106
Figure 4.20: Distribution of property rateable value in RM for intermediate terrace building type
107
Figure 4.21: Distribution of property rateable value in RM for semi-detached building type
108
Figure 4.22: Distribution of property rateable value in RM for corner terrace building type
109
Figure 4.23: Distribution of property rateable value in RM for detached building type
110
Figure 4.24: Morans I output for the intermediate terrace building type
Figure 4.25: Morans I output for the semi detached building type
111
Figure 4.26: Morans I output for the corner terrace building type
Therefore, since the nature of the data shows that spatial autocorrelation
error was presents, the study able to decide that the usage of GWR was
112
not feasible as it unable to eliminate the error. Therefore, only the OLS
and SRM method were conducted for this study. The OLS was used as
the bases of comparison while the SRM as the solution to overcome the
spatial autocorrelation error.
4.7 Conclusion
This chapter described the study area of Kota Kinabalu City Hall
(DBKK) and data preparation for the property value model for rating
purpose. Data gathering and preparation procedures were undertaken to
produce datasets suitable for modelling purpose. This chapter also
discussed the generation of factors influencing property values which
included the distance analysis to measure the distance from each
residential location to the nearby location factor using GIS. Lastly, the
distribution of the property rateable value was mapped to identify the
data pattern for spatial autocorrelation error. This procedure was
undertaken using Moran I tool and it indicated that spatial
autocorrelation exists in the study area which prompts the study to use
the OLS and SRM method for modelling. The OLS was needed for
comparison and benchmarking purpose. However, the GWR was not
used for this study as it is unable to estimate an incomplete data with no
dependent value and incapable to eliminate spatial autocorrelation error.
The modelling of OLS and SRM would then be conducted and explained
in chapter 5 and chapter 6.
113
CHAPTER 5
5.1 Introduction
This chapter discusses the empirical stage of the study by presenting the
development of the global OLS model based on 5625 observation data of
1997. It started with the OLS model developed for this study and
addressed the variable selection by investigating the multicollinearity
that could contribute to the modelling error. Then diagnostic test of the
OLS model would be discussed to identify the error and accuracy of the
OLS model for the overall and based on building type of the study.
114
5.3 Variable Selection
In order to ensure suitable variables were chosen for the study, the
regression model developed for this study was ensure to be free from any
model error in order to prevent any model bias and mislead or incorrect
interpretation. Therefore, this study adopted the pairwise correlation
matrix and variance inflation factor (VIF) to conduct this task. This
processed was conducted continuously by experimented the variables to
ensure that no modelling problems presented in the model. These were
conducted by adding, removing or adjusting the available variables from
the model. The workflow of this task could be visualized in Figure 5.1
below in which, firstly, it started by conducted manual selection from the
availability of the variables that relevant for the model (Millington,
2001; Appraisal Institute, 1992; Scarrett, 2008). Secondly, stepwise
procedure was then applied for the chosen variables in which any non-
significant variable of level higher than 0.1 was removed to eliminate
multicollinearity. Thirdly, another multicollinearity test was conducted
using VIF in which any multicollinearity error were also removed or
adjusted before the final selection of variables with no multicollinearity
error obtained to proceed to the next task of modelling.
115
As discussed in Chapter 4, variables for the model could be divided into
property characteristics and location factors. Property characteristics
were divided into 21 factors. However, due to a large number of missing
data, some of the variables were removed. Finally, there were 12
variables describing property characteristics used (Table 5.1). In
addition, location factors involved 19 variables that measures distances
in unit meter (Table 5.2) were included. All these variables were then
combined and processed for multicollinearity checking as explained in
sub-topic 5.3.1 and 5.3.2.
116
Table 5.1: Property Characteristics Variables
Variable Name Measurement Description
Bld_type* Scale Building type (e.g. Terrace, Detached)
No_storey Scale Number of storey in the building
No_room Scale Number of room in the building
Floor_Level Scale Level of floor the property locate
Mfa Scale Main floor area
Afa Scale Ancillary floor area
RCA* Scale Reduced Covered Area Estimation
Ext_build Scale Availability of building extension
Extuse Scale External Use
Subuse Scale Sub use
Bldq* Scale Building Quality
Spfaci Scale Special Facilities
Land_area* Scale Land area estimation
Topo* Scale Land Topography
Sitepr* Scale Site Preparation
Access* Scale Access to main road
Flood* Scale Flood Prone Area
Ltshp* Scale Lot Shape
Frontage* Scale Property frontage to road
Sanitn* Scale Type of Sanitation
NhoodQ* Scale Neighbourhood quality
Note: ‘*’ Indicate the variables chosen without the problem of missing data
117
Table 5.2: Location factor variables
Variable Name Measurement Description
Market Scale Place for buying and selling of goods
with offered price
School Scale Education centre
Sports area Scale Place for sports activities
Bank Scale Centre for finance activities
Shopping Mall Scale Large place for various stores, business
and restaurant
Government Office Scale Various Government activities
Church Scale Place of worship for Christians
Mosque Scale Place of worship for Muslims
Hindu Temple Scale Place of worship for Hindus
Chinese Temple Scale Place of worship for Chinese
Commercial Centre Scale Place of various commerce activities
Police Centre Scale Police station including
Post Office Scale Provide postal service and bill payment
Hospital Scale Large medical centre
Hall Scale Public hall for gathering, sports or
entertainment
Park Scale Recreation area
Field Scale Open space area
Supermarket Scale Large stores that sells variety of food
and household items
Office Centre Scale Place for business activities
118
Based on the output of the stepwise regression as shown in Appendix B,
the best result would be the model with the highest adjusted R2 which
consists of 18 independent variables. Nonetheless, this study also relies
on the VIF values of the independent variables as a check in selecting the
final variables as now discussed.
Apparently, some data from the 18 variables produced more than 7.5
VIF value. Therefore, certain variables were dropped and re-analyze was
conducted. Finally, the study manages to obtain a model which was free
form multicollinearity error with 15 variables as stated in Table 5.3. The
result shows that all the following independent variables consist of
property physical characteristics and location factors that passed the test
by producing VIF value of less than the cut-off point of 7.5.
119
5.3.3 Final selection of model variables
As previously stated, initially there were 40 variables or valuation factors
that involved 21 property characteristics and 19 location factors used.
However, after removing some variables in which had missing data and
multicollinearity error, a total of 15 independent variables had been
selected to estimate the dependent variable (rateable value). The
measurement values for each variable were based on questionnaire’s
mean score, metric unit and currency as shown in Table 5.4.
The measurement for the rateable value was based on currency scale in
Ringgit Malaysia (RM). The property rating value was obtained from the
1997 DBKK mass revaluation exercise. Although the property
revaluation was conducted quite long ago, the value was still valid and
currently applied in DBKK (DBKK, 2012). It was used to estimate and
produce an updated value for the current year using inflation rate
calculation. For the Reduced Coverage Area (RCA) and Land Area, the
value was also obtained directly from the revaluation exercise taken in
year 1997. Both variables were measured using square feet unit. As
explained in chapter 3, RCA which was considered as an important
120
element for property valuation in DBKK, represents the main floor area
of the property but was recalculated to be better suited for valuation
purpose. Next, the location factor variable consists of Hospital, Office
Centre, Shopping Mall, Post Office, Hindu Temple and Field was
obtained based on the GIS analysis conducted as mentioned in chapter 4
and the measurement was based on meter unit from the property location
to the nearest location factor. As for the building type, building quality,
sanitation, frontage, flood prone and topography, the value was also
obtained from the revaluation exercise but it was change from
categorised to scale measurement using mean score based from the
questionnaire’s feedback.
Once the OLS model was produced, the study continued with the OLS
diagnostic or model error detection for the overall or based on building
types as explained in the following section.
121
5.4 OLS model diagnostic
In conducting OLS model, diagnostic tests need to be conducted to
check whether or not the model’s residuals were normally distributed,
identify spatial autocorrelation, examine the model non-stationarity
pattern and assess the significance of the model. All these tests were
needed to ensure the OLS model fulfilled the criteria to become an
accurate and acceptable property valuation model in estimating the
residential property rating value in the study area.
Table 5.6 below shows the diagnostic output produced from the OLS
model. The value from the Jarque-Bera statistics (JB) was used to
identify the model’s normal distribution, while the Moran’s I statistic or
Z-score examined the spatial autocorrelation pattern of the model. Next,
the Koenker (BP) statistic was used to examine non-stationarity pattern
and lastly, based on the result of the Koenker (BP) statistic, the Joint-
Wald or Joint-F statistic would be chosen to measures the overall model
significance. The discussion of these tests would be explained in the
following sections 5.4.1 to 5.4.4
122
pattern. Figure 5.2 shows the locations where both positive and negative
residuals occurred.
To test the spatial autocorrelation formally, this study adopted the spatial
statistics of Moran’s I to determine the existence of significant spatial
autocorrelation. Figure 5.3 and the previous Table 5.6 shows the results
from the spatial autocorrelation tests of Moran’s I undertaken on the
OLS model developed in this study. This test enables identification of
the two forms of spatial autocorrelation, of positive or negative. Moran’s
I value from the Table 5.6 and Figure 5.3 indicates positive spatial
autocorrelation (Z score = 45.857, p-value = 0.00) meaning that similar
residuals cluster together. This means that it is more likely for the spatial
autocorrelation detected to occur out of missing variables for important
property characteristics. The Moran Index value for overall OLS data
residual of 0.062 however, was smaller compared to Moran Index value
for overall current value with 0.179. This shows that the spatial
autocorrelation was reduced when using OLS.
123
Figure 5.2: Geographical distributions of the positive and the negative residuals of the OLS’s overall model
124
Figure 5.3: Spatial autocorrelation for OLS residual model indicate
clustered pattern
Similar outcome also obtained from previous studies that showed the
presence of significant spatial autocorrelation despite using rich dataset
(Des Rosiers et al., 2000; Theriault et al., 2003; and Suriatini Ismail,
2005). Most studies do not address the problem. However, as suggested
by Theriault et al. (2003), spatial autocorrelation can be more severe
because it could cause heteroscedasticity. Suriatini Ismail, (2005) for
example conducted further analysis and stated that there was no
straightforward relationship between spatial autocorrelation and number
of observations but it provides obvious association with geographical
aggregation, flat properties and density of dwellings.
This study was then proceeding to the next test to determine the
stationarity pattern of the model which was conducted in the following
section.
125
When the p-value (probability) for the JB test was high in which it
achieved more than 95% confidence level, it indicated normal
distribution among the residuals and the model was usable. However, the
JB test was mentioned to have size distortion problem (Chen and Kuan,
2003) and furthermore it has zero breakdown of value as even a single
outlier can make the test worthless (Brys et al., 2004).
Based on the result presented in Table 5.6, the value of the JB test
indicated that the residuals appeared to be a non-normal distribution as
the p-value produced smaller than 0.05 confidence level. Although,
reselection from the available variables was conducted, the error was still
there. The study conducted by Charlton and Fotheringham (2009) also
had similar problem where it failed the JB test. This could probably
occur because of non-stationarity pattern exists in the residual.
Therefore, stationarity test was conducted as discussed in the following
sub-topic.
Based on the output in Table 5.6, the probability for Koenker (BP)
Statistics obtained for this test was smaller than 0.05 indicates
statistically significant heteroscedasticity or non-stationarity for this
model. This shows that a key variable is missing from the model or not
suitable for global model. This shows that the study need to use the
robust coefficient and robust probabilities estimation.
128
Table 5.7: Summary of the OLS Model Output for DBKK Rating
ĞƚĂ
KǀĞƌĂůů sĂƌŝĂďůĞ ŽĞĨĨŝĐŝĞŶƚ ^ƚĚƌƌŽƌ ƚͲ^ƚĂƚŝƐƚŝĐ WƌŽďĂďŝůŝƚLJ ZŽďƵƐƚͺ^ ZŽďƵƐƚͺƚ ZŽďƵƐƚͺWƌ ŽĞĨĨŝĐŝĞŶƚ
ZϮсϱϬ͘ϱй ŽŶƐƚĂŶƚ ͲϵϲϬ͘ϱϳϰϰϳϮ ϱϬϴ͘ϲϬϱϳϲϱ Ͳϭ͘ϴϴϴϲϰϯ Ϭ͘Ϭϱϴϵϴϵ ϰϲϴ͘Ϯϯϰϭϱϯ ͲϮ͘Ϭϱϭϰϴϯ Ϭ͘ϬϰϬϮϱϱΎ
Ěũ͘ZϮсϱϬ͘ϰй ^,KWD>> ͲϬ͘ϳϴϬϴϰϳ Ϭ͘ϬϰϲϬϮϮ Ͳϭϲ͘ϵϲϲϵϭϲ Ϭ͘ϬϬϬϬϬϬΎ Ϭ͘Ϭϱϲϵϱϯ Ͳϭϯ͘ϳϭϬϰϮϰ Ϭ͘ϬϬϬϬϬϬΎ ͲϬ͘Ϯϰ
WK>/dZ Ͳϭ͘Ϯϯϳϵϱϰ Ϭ͘Ϭϱϰϲϰϱ ͲϮϮ͘ϲϱϰϮϴϮ Ϭ͘ϬϬϬϬϬϬΎ Ϭ͘Ϭϲϱϱϲ Ͳϭϴ͘ϴϴϮϴϲϮ Ϭ͘ϬϬϬϬϬϬΎ ͲϬ͘ϱϮϭ
WK^dK&&/ ϭ͘ϭϲϰϳϰϲ Ϭ͘ϬϲϬϰϬϯ ϭϵ͘ϮϴϯϬϮϭ Ϭ͘ϬϬϬϬϬϬΎ Ϭ͘Ϭϲϱϭϯϴ ϭϳ͘ϴϴϭϮϵϲ Ϭ͘ϬϬϬϬϬϬΎ Ϭ͘ϰϵϯ
K&&/dZ ͲϬ͘ϯϯϴϬϱϵ Ϭ͘ϬϱϭϳϳϮ Ͳϲ͘ϱϮϵϴϬϯ Ϭ͘ϬϬϬϬϬϬΎ Ϭ͘Ϭϱϱϲϱϱ Ͳϲ͘Ϭϳϰϭϳϲ Ϭ͘ϬϬϬϬϬϬΎ ͲϬ͘ϭϮϴ
&/> ϭ͘ϭϬϰϴϵϵ Ϭ͘Ϭϯϴϴϱϴ Ϯϴ͘ϰϯϯϵϳϰ Ϭ͘ϬϬϬϬϬϬΎ Ϭ͘Ϭϱϵϱϱϯ ϭϴ͘ϱϱϯϬϳϭ Ϭ͘ϬϬϬϬϬϬΎ Ϭ͘ϱϳϵ
E< Ϭ͘ϮϮϭϴϵ Ϭ͘Ϭϰϲϭϰϵ ϰ͘ϴϬϴϬϳϯ Ϭ͘ϬϬϬϬϬϯΎ Ϭ͘Ϭϲϱϴϵϳ ϯ͘ϯϲϳϮϮϮ Ϭ͘ϬϬϬϳϴϭΎ Ϭ͘Ϭϴϵ
,/EdDW> Ϭ͘ϭϵϱϱϳϴ Ϭ͘ϬϱϮϱϭϵ ϯ͘ϳϮϯϵϲϳ Ϭ͘ϬϬϬϮϭϭΎ Ϭ͘ϬϱϮϲ ϯ͘ϳϭϴϭϲϴ Ϭ͘ϬϬϬϮϭϱΎ Ϭ͘Ϭϳϳ
,/EdDW> ͲϬ͘ϯϮϲϭϰϱ Ϭ͘ϬϮϯϲϱϯ Ͳϭϯ͘ϳϴϴϳϮϴ Ϭ͘ϬϬϬϬϬϬΎ Ϭ͘ϬϲϯϯϮϵ Ͳϱ͘ϭϰϵϵϳϴ Ϭ͘ϬϬϬϬϬϭΎ ͲϬ͘ϭϰϴ
Zͺ&d Ϭ͘ϮϮϬϴϰϳ Ϭ͘Ϭϭϭϰϳϴ ϭϵ͘ϮϰϬϲϲϯ Ϭ͘ϬϬϬϬϬϬΎ Ϭ͘ϬϳϴϬϯϰ Ϯ͘ϴϯϬϭϮϭ Ϭ͘ϬϬϰϲϳϰΎ Ϭ͘ϮϬϯ
>EZͺ&d Ϭ͘ϮϬϮϴϴ Ϭ͘ϬϬϱϵϮϵ ϯϰ͘ϮϮϬϵϯϭ Ϭ͘ϬϬϬϬϬϬΎ Ϭ͘Ϭϰϲϭϱϰ ϰ͘ϯϵϱϳϯϭ Ϭ͘ϬϬϬϬϭϰΎ Ϭ͘ϯϰϳ
dKWK' ͲϮϰϲ͘ϱϰϬϯϭϲ ϰϬ͘ϬϲϭϮϬϰ Ͳϲ͘ϭϱϰϬϵϭ Ϭ͘ϬϬϬϬϬϬΎ ϰϲ͘ϱϱϵϵϯϮ Ͳϱ͘Ϯϵϱϭϭϴ Ϭ͘ϬϬϬϬϬϬΎ ͲϬ͘Ϭϲ
h/>Y ϲϬϮ͘ϴϳϳϬϴϯ Ϯϳ͘ϴϬϮϲϰϭ Ϯϭ͘ϲϴϰϭϲϲ Ϭ͘ϬϬϬϬϬϬΎ ϰϯ͘ϯϴϰϮϭϲ ϭϯ͘ϴϵϲϮϯϭ Ϭ͘ϬϬϬϬϬϬΎ Ϭ͘ϮϮϴ
&>KKW ϳϳ͘ϲϵϵϴϬϴ ϭϱ͘Ϯϴϵϳϭϳ ϱ͘Ϭϴϭϴϯϰ Ϭ͘ϬϬϬϬϬϭΎ ϭϵ͘ϲϳϯϭϵϴ ϯ͘ϵϰϵϱϮϲ Ϭ͘ϬϬϬϬϴϴΎ Ϭ͘Ϭϱ
^E/dd/KE ͲϮϬϴ͘ϬϴϯϮϮϳ ϱϲ͘ϱϱϮϯϮϵ Ͳϯ͘ϲϳϵϰϴϭ Ϭ͘ϬϬϬϮϰϵΎ ϲϳ͘ϰϭϬϮϭϯ Ͳϯ͘ϬϴϲϴϮϭ Ϭ͘ϬϬϮϬϰϳΎ ͲϬ͘Ϭϯϳ
129
Based on table 5.7 also, all variables were statistically significant at 95%
confidence level based on the Robust_Probability measurement except
the building type. The non-significant of building type was probably due
to inconsistency of certain building type with the rating values.
However, this shows that the building type might be suitable to be used
as model segmentation or multiple OLS model that would represent the
property rating better. This would be discussed in the following section.
There were 8 types of building type listed in the DBKK database for
rating purpose in the area that consists of Kampung House, Walk-Up
Apartment, Intermediate Terrace, Semi-Detached, Corner Terrace, Town
House and Detached. However, the data in the study is limited to 5 types
of building type as it only focused on housing in urban area and non-
apartment/flat type. The 5 types of building that were used to categorise
the property rating model are Intermediate Terrace, Semi-Detached,
Corner Terrace, Town House and Detached. The total number of
property for each building type is shown in Table 5.8 as follows:
Table 5.8: Number of record for each building type in the study
Building Type Value
Intermediate Terrace 2160
Semi-Detached 2150
Corner Terrace 773
Town House 5
Detached 537
Total 5625
130
Table 5.9 summarises the estimated parameters for each property rating
valuation in DBKK based on building type of Intermediate Terrace,
Semi-Detached, Corner Terrace and Detached. Table 5.9 shows that for
the four building types, adjusted R2 ranges from 49.3% to 58.1% with
intermediate terrace obtained the lowest and semi-detached as the
highest. The difference of 7.7 percentage between the semi-detached and
the overall OLS model indicates a marginal improvement in the
explanatory power brought about by the OLS modelling based on
building type.
Each building type also have different significant variables that influence
its value. Although intermediate terrace has all 14 independent variables
significant, the semi detached and corner terrace has 13 significant.
While the detached building has only 9 variables that significant to it.
Based on the table, the beta coefficient for POLICE_CTR, FIELD and
POSTOFFICE gave high value for all three building type of intermediate
terrace, semi detached and corner terrace. The POLICE_CTR, FIELD
and POSTOFFICE produce magnitude range from 0.398 to 0.636, 0.475
to 0.915 and 0.449 to 0.854 respectively to the three building type. The
detached produce slightly different outcome as the variables of Land
Area (LANDAR_FT), POLICE_CTR and POSTOFFICE gives the
highest beta coefficient value with 0.384, 0.381 and 0.325 respectively.
131
Table 5.9: Summary OLS model output based on building type
/ŶƚĞƌ ZŽďƵƐƚͺ ZŽďƵƐƚͺ ZŽďƵƐƚͺ ĞƚĂ
dĞƌƌĂĐĞ sĂƌŝĂďůĞ ŽĞĨĨŝĐŝĞŶƚ ^ƚĚƌƌŽƌ ƚͲ^ƚĂƚƐ WƌŽď ^ ƚ Wƌ ŽĞĨĨ
ZϮс
ϰϵ͘ϳй /ŶƚĞƌĐĞƉƚ Ͳϳϴϳ͘ϲϳϵ ϯϭϱ͘Ϯϲϰ ͲϮ͘ϰϵϴϰ Ϭ͘ϬϭϮϱΎ ϯϴϳ͘ϯϮϲ ͲϮ͘Ϭϯϯϲ Ϭ͘ϬϰϮϭΎ
Ěũ͘ZϮс
ϰϵ͘ϯй ^,KWD>> ͲϬ͘ϯϯϯϬϳ Ϭ͘ϬϯϳϬϰ Ͳϴ͘ϵϵϬϳ Ϭ͘ϬϬϬϬΎ Ϭ͘ϬϰϭϬϯ Ͳϴ͘ϭϭϲϱ Ϭ͘ϬϬϬϬΎ ͲϬ͘Ϯϲ
WK>/dZ ͲϬ͘ϲϮϭϵϴ Ϭ͘Ϭϰϳϵϲ ͲϭϮ͘ϵϲϳ Ϭ͘ϬϬϬϬΎ Ϭ͘ϬϱϬϭϬ ͲϭϮ͘ϰϭϯ Ϭ͘ϬϬϬϬΎ ͲϬ͘ϲϯϲ
WK^dK&&/ Ϭ͘ϴϯϳϰϴϰ Ϭ͘ϬϱϮϳϱ ϭϱ͘ϴϳϱϭ Ϭ͘ϬϬϬϬΎ Ϭ͘Ϭϱϱϲϲ ϭϱ͘ϬϰϱϮ Ϭ͘ϬϬϬϬΎ Ϭ͘ϴϱϰ
K&&/dZ ͲϬ͘ϰϬϭϭϮ Ϭ͘ϬϰϮϬϯ Ͳϵ͘ϱϰϯϬ Ϭ͘ϬϬϬϬΎ Ϭ͘Ϭϰϳϭϰ Ͳϴ͘ϱϬϴϴ Ϭ͘ϬϬϬϬΎ ͲϬ͘ϯϱϵ
&/> Ϭ͘ϱϬϰϬϵϵ Ϭ͘Ϭϯϰϱϱ ϭϰ͘ϱϴϴϴ Ϭ͘ϬϬϬϬΎ Ϭ͘ϬϯϰϲϮ ϭϰ͘ϱϱϵϳ Ϭ͘ϬϬϬϬΎ Ϭ͘ϲϱϴ
E< Ϭ͘ϮϬϮϵϬϲ Ϭ͘Ϭϯϲϳϯ ϱ͘ϱϮϯϴϯ Ϭ͘ϬϬϬϬΎ Ϭ͘Ϭϯϳϱϭ ϱ͘ϰϬϵϬϰ Ϭ͘ϬϬϬϬΎ Ϭ͘ϭϴϱ
,/EdDW> Ϭ͘ϭϯϰϬϬϴ Ϭ͘ϬϰϱϬϳ Ϯ͘ϵϳϮϴ Ϭ͘ϬϬϮϵΎ Ϭ͘ϬϱϱϬϭ Ϯ͘ϰϯϱϲϱ Ϭ͘ϬϭϰϵΎ Ϭ͘ϭϯ
,/EdDW> ͲϬ͘ϭϰϬϭϲ Ϭ͘ϬϮϭϮϲ Ͳϲ͘ϱϵϭϴ Ϭ͘ϬϬϬϬΎ Ϭ͘ϬϮϲϯϭ Ͳϱ͘ϯϮϲϳ Ϭ͘ϬϬϬϬΎ ͲϬ͘ϭϯ
Zͺ&d Ϭ͘ϬϮϳϬϰϴ Ϭ͘ϬϬϳϴϴ ϯ͘ϰϯϭϭϭ Ϭ͘ϬϬϬϲΎ Ϭ͘ϬϬϴϵϴ ϯ͘ϬϬϴϵϰ Ϭ͘ϬϬϮϳΎ Ϭ͘Ϭϱϵ
>EZͺ&d Ϭ͘ϬϯϴϮϱ Ϭ͘ϬϬϳϱϴ ϱ͘Ϭϰϯϯϱ Ϭ͘ϬϬϬϬΎ Ϭ͘Ϭϭϳϵϳ Ϯ͘ϭϮϳϵϰ Ϭ͘ϬϯϯϰΎ Ϭ͘Ϭϳϴ
dKWK' ͲϭϮϬ͘ϴϱϮ ϯϰ͘ϯϵϳϴ Ͳϯ͘ϱϭϯϯ Ϭ͘ϬϬϬϰΎ Ϯϯ͘ϵϭϲϴ Ͳϱ͘Ϭϱϯ Ϭ͘ϬϬϬϬΎ ͲϬ͘Ϭϱϴ
h/>Y ϮϮϯ͘ϭϱϴϳ Ϯϯ͘ϵϴϬϳ ϵ͘ϯϬϱϳ Ϭ͘ϬϬϬϬΎ Ϯϯ͘ϬϮϳϭ ϵ͘ϲϵϭϭϮ Ϭ͘ϬϬϬϬΎ Ϭ͘ϭϱϲ
&>KKW ϲϯ͘ϵϬϬϲϭ ϭϭ͘Ϭϭϭϵ ϱ͘ϴϬϮϴϰ Ϭ͘ϬϬϬϬΎ ϭϭ͘ϰϱϯϭ ϱ͘ϱϳϵϯϮ Ϭ͘ϬϬϬϬΎ Ϭ͘Ϭϵϳ
^E/dd/KE ϯϬϱ͘ϯϯϭϯ ϲϴ͘ϰϭϮϯ ϰ͘ϰϲϯϭϬ Ϭ͘ϬϬϬϬΎ ϵϱ͘Ϯϰϯϰ ϯ͘ϮϬϱϳϵ Ϭ͘ϬϬϭϰΎ Ϭ͘Ϭϵϯ
^Ğŵŝ ZŽďƵƐƚͺ ZŽďƵƐƚͺ ZŽďƵƐƚͺ ĞƚĂ
ĞƚĂĐŚĞĚ sĂƌŝĂďůĞ ŽĞĨĨŝĐŝĞŶƚ ^ƚĚƌƌŽƌ ƚͲ^ƚĂƚƐ WƌŽď ^ ƚ Wƌ ŽĞĨĨ
ZϮ
сϱϴ͘ϰй /ŶƚĞƌĐĞƉƚ ϮϱϱϬ͘ϬϴϯϬ ϯϲϭ͘ϴϴϮ ϳ͘ϬϰϲϳϮ Ϭ͘ϬϬϬϬΎ ϯϱϳ͘ϵϭϮ ϳ͘ϭϮϰϴϳ Ϭ͘ϬϬϬϬΎ
Ěũ͘ZϮс
ϱϴ͘ϭй ^,KWD>> ͲϬ͘ϳϯϱϱϬϰ Ϭ͘ϬϲϬϮϲ ͲϭϮ͘ϮϬϰ Ϭ͘ϬϬϬϬΎ Ϭ͘ϬϲϰϮϯ Ͳϭϭ͘ϰϱϬ Ϭ͘ϬϬϬϬΎ ͲϬ͘ϮϵϮ
WK>/dZ Ͳϭ͘Ϭϭϯϱϵϵ Ϭ͘ϬϱϳϰϬ Ͳϭϳ͘ϲϱϳ Ϭ͘ϬϬϬϬΎ Ϭ͘ϬϲϮϵϴ Ͳϭϲ͘Ϭϵϭ Ϭ͘ϬϬϬϬΎ ͲϬ͘ϱϰϯ
WK^dK&&/ ϭ͘ϰϳϮϭϮϴ Ϭ͘Ϭϲϯϳϰ Ϯϯ͘Ϭϵϱϲ Ϭ͘ϬϬϬϬΎ Ϭ͘Ϭϲϵϰϱ Ϯϭ͘ϭϵϰϯ Ϭ͘ϬϬϬϬΎ Ϭ͘ϴϰϳ
K&&/dZ ͲϬ͘ϰϱϱϬϭϮ Ϭ͘Ϭϱϴϱϯ Ͳϳ͘ϳϳϯϭ Ϭ͘ϬϬϬϬΎ Ϭ͘Ϭϲϭϲϰ Ͳϳ͘ϯϴϭϲ Ϭ͘ϬϬϬϬΎ ͲϬ͘Ϯϰϱ
&/> ϭ͘ϯϲϵϳϲϰ Ϭ͘Ϭϰϯϵϱ ϯϭ͘ϭϲϭϮ Ϭ͘ϬϬϬϬΎ Ϭ͘ϬϰϱϮϬ ϯϬ͘ϯϬϬϭ Ϭ͘ϬϬϬϬΎ Ϭ͘ϵϭϱ
E< ͲϬ͘ϮϰϴϭϮϲ Ϭ͘ϬϱϰϴϬ Ͳϰ͘ϱϮϳϳ Ϭ͘ϬϬϬϬΎ Ϭ͘Ϭϱϵϭϱ Ͳϰ͘ϭϵϰϴ Ϭ͘ϬϬϬϬΎ ͲϬ͘ϭϯϰ
,/EdDW> ͲϬ͘ϮϳϯϰϬϵ Ϭ͘ϬϯϬϴϬ Ͳϴ͘ϴϳϰϳ Ϭ͘ϬϬϬϬΎ Ϭ͘ϬϰϮϭϰ Ͳϲ͘ϰϴϳϱ Ϭ͘ϬϬϬϬΎ ͲϬ͘ϭϲϵ
Zͺ&d Ϭ͘ϭϬϯϵϱϰ Ϭ͘Ϭϭϰϰϲ ϳ͘ϭϴϱϵ Ϭ͘ϬϬϬϬΎ Ϭ͘ϬϮϬϲϯ ϱ͘Ϭϯϴϱϰ Ϭ͘ϬϬϬϬΎ Ϭ͘ϭϭϮ
>EZͺ&d Ϭ͘ϬϮϴϲϮϰ Ϭ͘ϬϬϳϬϯ ϰ͘ϬϳϬϲ Ϭ͘ϬϬϬϬΎ Ϭ͘ϬϭϭϳϬ Ϯ͘ϰϰϱϲϵ Ϭ͘ϬϭϰϱΎ Ϭ͘Ϭϱϴ
dKWK' ͲϱϮϲ͘ϬϲϬϱϯ ϰϱ͘Ϭϭϱϳ Ͳϭϭ͘ϲϴϲ Ϭ͘ϬϬϬϬΎ ϱϰ͘Ϯϵϲϱ Ͳϵ͘ϲϴϴϮ Ϭ͘ϬϬϬϬΎ ͲϬ͘ϭϴϯ
h/>Y ϰϳϵ͘ϴϲϱϰϴ Ϯϴ͘ϰϱϳϱ ϭϲ͘ϴϲϮϱ Ϭ͘ϬϬϬϬΎ Ϯϵ͘ϯϬϱϳ ϭϲ͘ϯϳϰϰ Ϭ͘ϬϬϬϬΎ Ϭ͘Ϯϳϯ
&>KKW ϲϵ͘ϭϰϯϬϲϳ ϭϱ͘ϲϯϮϯ ϰ͘ϰϮϯϬ Ϭ͘ϬϬϬϬΎ ϭϱ͘ϱϰϳϯ ϰ͘ϰϰϳϮϳ Ϭ͘ϬϬϬϬΎ Ϭ͘Ϭϲϲ
^E/dd/KE Ͳϭϲϰ͘ϴϳϭϴϯ ϱϮ͘ϱϬϮϭ Ͳϯ͘ϭϰϬϮ Ϭ͘ϬϬϭϳΎ ϯϰ͘ϯϴϴϬ Ͳϰ͘ϳϵϰϰ Ϭ͘ϬϬϬϬΎ ͲϬ͘Ϭϰϳ
132
Table 5.9: (Continued)
133
5.6 Conclusion
This chapter explained the global modelling process of Ordinary Least
Square (OLS). The process started with the variable selection by
removing variables which were affected by multicollinearity error. The
study managed to obtain 15 independent variables that were cleaned
from the error. Unfortunately, the OLS model failed in the next test of
normality and the spatial residual distribution in which the study
detected non-normal distribution and spatial correlated clustered residual
pattern. This depicts the model was misspecified and some variables
were missing from the OLS model. Since it has been tested significantly
for spatial autocorrelation error, therefore, the data need to be analysed
using SRM as an alternative solution. The performance of the OLS
model was average with only 50.4% adjusted R-Square value achieved.
The performance of OLS model based on building type manages to
improve the model with the highest from semi-detached with 58.1%. A
bit improvement also comes from Corner Terrace and Detached with
51.9% and 51.3% respectively. This shows that model segmentation
capable to improve the model’s accuracy. In the explanatory power of
the independent variables, POLICE_CTR and FIELD consistently
provide as one of the most influence towards property value for all
models. Ultimately, average performance in adjusted R-square value and
failure in the normality and spatial autocorrelation test indicated that
there is a need to conduct the modelling approach using SRM as an
alternative solution for this problem.
134
CHAPTER 6
6.1 Introduction
The existence of spatial autocorrelation in OLS model as discussed in
chapter 5 ultimately lead to the development of spatial regression model.
As discussed earlier, the spatial autocorrelation error was unable to be
removed even though other measures such as work around the model
variables or subdivided the data into building type models were
performed. Although spatial heteroscedasticity also exists in the previous
model, addressing the spatial autocorrelation error, also can overcome
the heteroscedasticity problem (Theriault et al., 2003). Therefore, the
following section discusses the development of the spatial regression
model, where similar variables as those used in developing OLS model
were used to maintain consistency. Spatial Lag or Spatial Error of the
spatial regression model would be identified which best represent the
study area. Finally, the model comparison would then be conducted to
assess the model performance among OLS and Spatial Regression in
estimate property rating value in the study.
Table 6.1: Output from the LM spatial autocorrelation test on the spatial
model
Lagrange Lagrange Robust Lagrange Robust Lagrange
Multiplier Multiplier Multiplier Multiplier
(Error) ( Lag) (Error) ( Lag)
Value 1135.3641 1138.3406 99.06282 102.0393
Probability 0.0000 0.0000 0.0000 0.0000
136
The spatial regression model was identified for each building type by
following the same procedure with the overall SRM modelling. The
result of this analysis is shown in table 6.2 below. The table shows that
the intermediate terrace, corner terrace and detached exhibits LM (lag) or
spatial lag model while semi-detached produced LM (error) or spatial
error model. The spatial lag model was selected as it produced higher
value of 67.99 and 24.9 in robust LM Lag for intermediate terrace and
corner terrace compared to robust LM error with value of 24.55 and
13.92 respectively. The spatial lag model also selected for detached as it
was significant with lower than 0.05 level compared to spatial error.
Meanwhile, the spatial error model was selected for semi-detached as it
produced higher value of 142.35 in robust LM error compared to 62.75
in robust LM lag.
CORNER TERRACE
Value 498.0653 509.0468 13.9192 24.9006982
Probability 0.0000 0.0000 0.0000 0.0000
SEMI DETACHED
Value 2179.0477 2099.4485 142.3479 62.7487
Probability 0.0000 0.0000 0.0000 0.0000
DETACHED
Value 2.8334 4.4201 0.0048 1.5915
Probability 0.0923 0.0355 0.9446 0.2071
137
6.3 The performance of SRM’s property rating
As the suitable type of spatial regression models were known to
represent the overall and building type based model, the performance of
the models can then be examined. Table 6.3 shows the summary of SRM
output which based on the overall and each building type within the
DBKK area. The output was produced from the Geoda software in which
the coefficient, standard error, z-value and probability were used as the
variable’s performance measurement. However, only the coefficient’s
value was used to measure the strength and type of the relationship
between the independent variable to the estimated property rating value.
If a coefficient is zero (or very near zero), the associated explanatory
variable has very little impact on the model or in other words, the
variable unable to contribute (Rosenshein et al., 2011; Taher Buyong,
2006). For example, if the coefficient value within range of -100 to 100,
it shows that the variable strength is low. If the coefficient’s value is
more than 100, it shows strong positive influence while coefficient’s
value less than -100 shows strong negative influence. The standard error
of z-value and probability were used to determine the usability of the
variables. If the probability indicates within 0.00 to 0.05, it shows that
the variable’s coefficient can be used (Charlton and Fotheringham,
2009).
Based on table 6.3, the SRM for overall model manage to achieve fair
result with R2 of 59.2%. Although 15 independent variables were used to
estimate the property rating value in overall SRM model, only 12 were
significant as the variables of FLOODP, SANITATION and
BUILD_TYPE was not valid for interpretation. The removal of
BUILD_TYPE was similar with the OLS model and it proves that the
usage of this as independent variables was not suitable.
Table 6.3 also summarises the estimated parameters for property rating
valuation in DBKK based on building type of Intermediate Terrace,
Semi-Detached, Corner Terrace and Detached. The four building types
produced R2 ranges from 52.9% to 84.3% with detached obtained the
lowest and semi-detached as the highest. The difference of 25.1
percentage between the semi-detached and the overall SRM model
138
indicates a big marginal improvement in the explanatory power brought
about by the SRM modelling based on building type. Each building type
model also have different significant variables that influence its property
rating value. Intermediate terrace has 9 independent variables significant,
semi detached has 10 significant, corner terrace has 8, while the detached
building has 10 variables that significant to it. POLICECTR produced
the highest influence for the property rating value in intermediate terrace
and corner terrace building type with 0.104 and 0.150 respectively.
While FIELD provide the highest influence for semi detached with 0.804
and LANDAR_FT for detached with 0.374. In contrast, HINDTEMPLE
produce the least influence for intermediate terrace building type with
beta coefficient value of 0.035. TOPOG for corner terrace with 0.041
and LANDAR_FT for semi detached with 0.033. While CHINTEMPLE
and BUILDQ produce the least influence with similar value of 0.116
towards detached building type.
139
Table 6.3: Summary of SRM’s overall and each building type output
KǀĞƌĂůů sĂƌŝĂďůĞ ŽĞĨĨŝĐŝĞŶƚ ^ƚĚƌƌŽƌ njͲǀĂůƵĞ WƌŽďĂďŝůŝƚLJ ĞƚĂŽĞĨĨ
ZϮсϱϵ͘Ϯй KE^dEd ͲϭϮϰϬ͘ϬϬϲ ϰϲϭ͘ϭϭϱϯ ͲϮ͘ϲϴϵϭϰϰ Ϭ͘ϬϬϳϭϲϯϲ
^,KWD>> ͲϬ͘ϯϯϮϳϳϵϰ Ϭ͘ϬϰϯϵϱϵϱϮ Ͳϳ͘ϱϳϬϭϯϯ Ϭ ͲϬ͘ϭϬϮ
WK>/dZ ͲϬ͘ϲϵϮϵϯϲϵ Ϭ͘ϬϱϮϱϭϱϱϱ Ͳϭϯ͘ϭϵϰϴϵ Ϭ ͲϬ͘ϮϵϮ
WK^dK&&/ Ϭ͘ϲϭϴϯϰϴϱ Ϭ͘ϬϱϳϲϰϰϱϮ ϭϬ͘ϳϮϲϵϯ Ϭ Ϭ͘ϮϲϮ
K&&/dZ ͲϬ͘ϭϰϵϴϬϰ Ϭ͘Ϭϰϳϯϱϭϯϳ Ͳϯ͘ϭϲϯϲϲϵ Ϭ͘ϬϬϭϱϱϴϭ ͲϬ͘Ϭϱϳ
&/> Ϭ͘ϱϯϱϰϰϴϭ Ϭ͘ϬϯϵϱϮϮϵϲ ϭϯ͘ϱϰϳϳϳ Ϭ Ϭ͘Ϯϴϭ
E< Ϭ͘ϭϰϭϬϯϮϳ Ϭ͘Ϭϰϭϴϵϭϵϭ ϯ͘ϯϲϲϱϴϱ Ϭ͘ϬϬϬϳϲϭϮ Ϭ͘Ϭϱϳ
,/EdDW> Ϭ͘ϬϵϳϭϰϬϲϮ Ϭ͘Ϭϰϳϲϴϵϳϰ Ϯ͘ϬϯϲϵϮϵ Ϭ͘Ϭϰϭϲϱϳ Ϭ͘Ϭϯϴ
,/EdDW> ͲϬ͘ϮϯϲϰϭϯϮ Ϭ͘ϬϮϭϲϱϲϳϰ ͲϭϬ͘ϵϭϲϯϴ Ϭ ͲϬ͘ϭϬϳ
Zͺ&d Ϭ͘ϭϴϳϲϱϳϯ Ϭ͘ϬϭϬϰϵϵϳϴ ϭϳ͘ϴϳϮϱ Ϭ Ϭ͘ϭϳϯ
>EZͺ&d Ϭ͘ϭϲϮϳϭϱϵ Ϭ͘ϬϬϱϱϯϴϵϭ Ϯϵ͘ϯϳϲϴϴ Ϭ Ϭ͘Ϯϳϴ
dKWK' Ͳϭϲϴ͘ϯϮϮϱ ϯϲ͘ϰϭϭϬϵ Ͳϰ͘ϲϮϮϴϯϱ Ϭ͘ϬϬϬϬϬϯϴ ͲϬ͘Ϭϰϭ
h/>Y ϯϵϳ͘ϲϯϯϴ Ϯϲ͘ϭϲϭϴϭ ϭϱ͘ϭϵϵϬϮ Ϭ Ϭ͘ϭϱϬ
/ŶƚĞƌŵĞĚŝĂƚĞ
dĞƌƌĂĐĞ sĂƌŝĂďůĞ ŽĞĨĨŝĐŝĞŶƚ ^ƚĚƌƌŽƌ njͲǀĂůƵĞ WƌŽďĂďŝůŝƚLJ ĞƚĂŽĞĨĨ
ZϮсϴϯ͘ϴй KE^dEd Ͳϲϯϵ͘ϵϲϴϵ ϭϳϴ͘ϵϳϭϵ Ͳϯ͘ϱϳϱϴϬϳ Ϭ͘ϬϬϬϯϰϵϮ
WK>/dZ ͲϬ͘ϭϬϭϴϱϳϵ Ϭ͘ϬϮϳϴϭϬϰϳ Ͳϯ͘ϲϲϮϱϳϮ Ϭ͘ϬϬϬϮϰϵϴ ͲϬ͘ϭϬϰ
WK^dK&&/ Ϭ͘Ϭϵϰϰϯϱϲϱ Ϭ͘ϬϯϬϴϮϲϬϲ ϯ͘Ϭϲϯϱ Ϭ͘ϬϬϮϭϴϳϴ Ϭ͘Ϭϵϲ
&/> Ϭ͘ϬϱϮϳϯϰϭϰ Ϭ͘ϬϮϬϬϱϭ Ϯ͘ϲϯ Ϭ͘ϬϬϴϱϯϴϲ Ϭ͘Ϭϲϵ
E< Ϭ͘ϬϰϭϳϬϰϮϯ Ϭ͘ϬϮϭϬϮϳϬϳ ϭ͘ϵϴϯϯϱϵ Ϭ͘ϬϰϳϯϮϳϮ Ϭ͘Ϭϯϴ
,/EdDW> ͲϬ͘Ϭϯϴϭϲϯϴϰ Ϭ͘ϬϭϮϮϭϰϬϴ Ͳϯ͘ϭϮϰϱϳϲ Ϭ͘ϬϬϭϳϴϬϳ ͲϬ͘Ϭϯϱ
>EZͺ&d Ϭ͘ϬϭϴϴϵϳϬϭ Ϭ͘ϬϬϰϮϴϲϴϱϱ ϰ͘ϰϬϴϭϮϵ Ϭ͘ϬϬϬϬϭϬϰ Ϭ͘Ϭϯϴ
h/>Y ϱϳ͘ϳϬϲϲϵ ϭϯ͘ϲϲϯϯϰ ϰ͘ϮϮϯϰϲϵ Ϭ͘ϬϬϬϬϮϰϭ Ϭ͘ϬϰϬ
&>KKW Ϯϰ͘ϬϬϳϴϯ ϲ͘ϯϯϵϳϰϱ ϯ͘ϳϴϲϴϳϳ Ϭ͘ϬϬϬϭϱϮϲ Ϭ͘Ϭϯϳ
^E/dd/KE ϭϱϲ͘ϵϰϯϰ ϯϵ͘ϭϲϴϵϴ ϰ͘ϬϬϲϴϯ Ϭ͘ϬϬϬϬϲϭϲ Ϭ͘Ϭϰϴ
ŽƌŶĞƌ
dĞƌƌĂĐĞ sĂƌŝĂďůĞ ŽĞĨĨŝĐŝĞŶƚ ^ƚĚƌƌŽƌ njͲǀĂůƵĞ WƌŽďĂďŝůŝƚLJ ĞƚĂŽĞĨĨ
ZϮсϳϱ͘ϳй KE^dEd ͲϭϮϮϬ͘ϵϱϲ ϯϴϲ͘ϳϱϵϱ Ͳϯ͘ϭϱϲϴϴϳ Ϭ͘ϬϬϭϱϵϰϴ
WK>/dZ ͲϬ͘ϭϱϳϵϱϲϰ Ϭ͘Ϭϲϱϭϵϱϴϳ ͲϮ͘ϰϮϮϳϵϳ Ϭ͘ϬϭϱϰϬϭϱ ͲϬ͘ϭϱϬ
E< Ϭ͘ϭϳϬϲϱϬϲ Ϭ͘ϬϱϬϵϰϱϬϮ ϯ͘ϯϰϵϳϬϭ Ϭ͘ϬϬϬϴϬϵϭ Ϭ͘ϭϯϱ
,/EdDW> ͲϬ͘ϬϵϭϴϲϮϳϮ Ϭ͘ϬϮϵϮϵϲϮϳ Ͳϯ͘ϭϯϱϲϰϱ Ϭ͘ϬϬϭϳϭϰϵ ͲϬ͘ϬϳϮ
Zͺ&d Ϭ͘ϬϱϯϴϮϱϬϳ Ϭ͘Ϭϭϯϵϱϭϰϴ ϯ͘ϴϱϴϬϮ Ϭ͘ϬϬϬϭϭϰϰ Ϭ͘Ϭϳϱ
>EZͺ&d Ϭ͘ϬϰϴϯϯϯϮϲ Ϭ͘ϬϭϭϬϮϵϱϲ ϰ͘ϯϴϮϭϱϵ Ϭ͘ϬϬϬϬϭϭϴ Ϭ͘Ϭϳϵ
dKWK' ͲϭϬϬ͘ϵϲϭ ϰϲ͘ϮϯϬϵϳ ͲϮ͘ϭϴϯϴϯϵ Ϭ͘ϬϮϴϵϳϰ ͲϬ͘Ϭϰϭ
h/>Y ϭϳϴ͘ϳϯϮ ϯϮ͘ϲϳϮϬϭ ϱ͘ϰϳϬϰϵϯ Ϭ Ϭ͘ϭϭϮ
^E/dd/KE Ϯϲϱ͘ϯϱϰϰ ϴϬ͘Ϯϴϯϭϲ ϯ͘ϯϬϱϮϯϭ Ϭ͘ϬϬϬϵϰϵϭ Ϭ͘Ϭϳϲ
140
Table 6.3: (Continued)
^ĞŵŝͲ
ĞƚĂĐŚĞĚ sĂƌŝĂďůĞ ŽĞĨĨŝĐŝĞŶƚ ^ƚĚƌƌŽƌ njͲǀĂůƵĞ WƌŽďĂďŝůŝƚLJ ĞƚĂŽĞĨĨ
ZϮсϴϰ͘ϯй KE^dEd ϱϰϭ͘ϲϰϲϭ ϱϭϱ͘ϱϲϴϲ ϭ͘ϬϱϬϱϴ Ϭ͘Ϯϵϯϰϱϭϲ
^,KWD>> ͲϬ͘ϵϴϬϴϴϬϰ Ϭ͘ϭϯϴϰϳϬϯ Ͳϳ͘Ϭϴϯϲϴϲ Ϭ ͲϬ͘ϯϵϬ
WK>/dZ ͲϬ͘ϱϱϳϭϴϵϳ Ϭ͘ϭϱϯϲϬϮϱ Ͳϯ͘ϲϮϳϰϳϵ Ϭ͘ϬϬϬϮϴϲϯ ͲϬ͘Ϯϵϴ
WK^dK&&/ ϭ͘ϮϳϲϬϭϳ Ϭ͘ϭϱϴϭϰϬϯ ϴ͘Ϭϲϴϴϵϯ Ϭ Ϭ͘ϳϯϱ
K&&/dZ ͲϬ͘ϱϵϳϮϬϱϱ Ϭ͘ϭϱϱϯϱϮϮ Ͳϯ͘ϴϰϰϮϬϰ Ϭ͘ϬϬϬϭϮϭ ͲϬ͘ϯϮϭ
&/> ϭ͘ϮϬϯϭϴϳ Ϭ͘ϭϯϮϵϬϵϰ ϵ͘ϬϱϮϲϴϲ Ϭ Ϭ͘ϴϬϰ
,/EdDW> ͲϬ͘ϯϬϰϱϮϵ Ϭ͘ϭϯϳϬϳϲϲ ͲϮ͘ϮϮϭϱϵϳ Ϭ͘ϬϮϲϯϭϬϰ ͲϬ͘ϭϱϱ
Zͺ&d Ϭ͘Ϭϲϯϳϯϳϲϯ Ϭ͘ϬϬϵϳϲϵϯϰϮ ϲ͘ϱϮϰϮϱ Ϭ Ϭ͘Ϭϲϵ
>EZͺ&d Ϭ͘Ϭϭϲϰϳϵϯϯ Ϭ͘ϬϬϰϴϳϭϲϵϱ ϯ͘ϯϴϮϲϲϵ Ϭ͘ϬϬϬϳϭϴ Ϭ͘Ϭϯϯ
h/>Y ϯϬϵ͘ϮϮϴϲ Ϯϱ͘Ϭϲϴϱϭ ϭϮ͘ϯϯϱϯϰ Ϭ Ϭ͘ϭϳϲ
&>KKW ϭϯϭ͘ϯϴϴϵ ϭϳ͘ϰϯϳϱ ϳ͘ϱϯϰϴϰϴ Ϭ Ϭ͘ϭϮϲ
ĞƚĂĐŚĞĚ sĂƌŝĂďůĞ ŽĞĨĨŝĐŝĞŶƚ ^ƚĚƌƌŽƌ njͲǀĂůƵĞ WƌŽďĂďŝůŝƚLJ ĞƚĂŽĞĨĨ
ZϮсϱϮ͘ϵй KE^dEd ͲϱϬϰϴ͘Ϭϰϳ ϭϴϴϰ͘Ϯϱϭ ͲϮ͘ϲϳϵϬϳϰ Ϭ͘ϬϬϳϯϴϮϳ
^,KWD>> Ͳϭ͘ϱϲϭϬϭϴ Ϭ͘ϯϬϴϱϱϭϮ Ͳϱ͘Ϭϱϵϭϴϴ Ϭ͘ϬϬϬϬϬϬϰ ͲϬ͘ϮϬϰ
WK>/dZ Ͳϭ͘ϳϭϳϬϲϯ Ϭ͘ϯϭϲϴϴϳϮ Ͳϱ͘ϰϭϴϱϮϵ Ϭ͘ϬϬϬϬϬϬϭ ͲϬ͘ϯϱϱ
WK^dK&&/ ϭ͘ϲϴϭϬϰ Ϭ͘ϯϲϱϮϲϳϯ ϰ͘ϲϬϮϮϭϵ Ϭ͘ϬϬϬϬϬϰϮ Ϭ͘ϯϬϭ
&/> ϭ͘ϮϮϲϵϴϭ Ϭ͘ϮϯϯϲϭϮϲ ϱ͘ϮϱϮϮϬϰ Ϭ͘ϬϬϬϬϬϬϮ Ϭ͘Ϯϳϭ
E< Ϭ͘ϲϵϵϳϳϯϲ Ϭ͘Ϯϱϳϴϳϭϰ Ϯ͘ϳϭϯϲϱϰ Ϭ͘ϬϬϲϲϱϰϳ Ϭ͘ϭϱϵ
,/EdDW> Ϭ͘ϲϵϯϴϳϬϮ Ϭ͘ϮϴϱϵϲϮϮ Ϯ͘ϰϮϲϰϰ Ϭ͘ϬϭϱϮϰϳϴ Ϭ͘ϭϭϲ
,/EdDW> ͲϬ͘ϰϯϵϰϳϱϴ Ϭ͘ϭϭϲϱϳϵϯ Ͳϯ͘ϳϲϵϳϱϵ Ϭ͘ϬϬϬϭϲϯϱ ͲϬ͘ϭϯϱ
Zͺ&d Ϭ͘ϱϭϬϬϴϰϱ Ϭ͘ϬϱϳϮϭϰϰϱ ϴ͘ϵϭϱϯϬϵ Ϭ Ϭ͘Ϯϴϱ
>EZͺ&d Ϭ͘ϯϭϲϲϬϵϭ Ϭ͘ϬϮϲϯϳϵ ϭϮ͘ϬϬϮϯϮ Ϭ Ϭ͘ϯϳϰ
h/>Y ϲϭϱ͘ϭϮϰϴ ϭϳϲ͘ϭϵϯϭ ϯ͘ϰϵϭϭϵϳ Ϭ͘ϬϬϬϰϴϭ Ϭ͘ϭϭϲ
141
6.4 Model comparison of OLS and SRM for the study area
Based on the results obtained from the two different modelling
procedures namely OLS and SRM used, the study then, compared its
performance in estimate the property rating values in DBKK. Although
the SRM was an obvious choice as all the OLS model exhibit spatial
autocorrelation error, but the performance of both methods still need to
be measured. The comparison was conducted based on the models
developed for the overall and based on building type in the DBKK area.
Morans I tool was used again to determine the presence of the spatial
autocorrelation error in the residual of the OLS model. The Akaike
Information Criterion (AIC) was used to compare the performance of
OLS and SRM model. Other tools such as the R2, was used for
measurement of goodness-of-fit of the model. It measures the accuracy
of the model’s estimation.
Based on the output, the spatial regression in the form of spatial lag
model was used as property rating value model to represent the overall
which covers the whole study area in DBKK. This model also managed
to produce fair estimation by obtaining 0.592 or 59.2% in R2 value but it
still higher than the R2 of OLS with only 0.505 or 50.5% achieved. The
SRM also have lower AIC value with 94707 compared with OLS
(95578).
Similar result also applied for the property rating value based on building
type. As shown in table 6.4, the building type of intermediate terrace,
corner terrace and semi-detached indicated spatial autocorrelation for
OLS models. This was based on the Z-score that produced values which
were out of the range of ±1.96 in critical value of the Morans I statistics.
Therefore, the spatial regression model was suitable to represent these
models. The goodness of fit based on the R2 value shown by the SRM
was also higher than the OLS model for all building type. The SRM
produced the R2 for intermediate terrace, corner terrace, semi-detached
and detached with 0.838, 0.757, 0.843 and 0.529 respectively compared
to OLS with 0.497, 0.528, 0.584 and 0.526 respectively. The same goes
142
with the AIC performance as the SRM produced lower value for
intermediate terrace, corner terrace, semi-detached and detached building
type with 30962, 11586, 33075 and 10000 respectively compared to
OLS with 32989, 11994, 34768 and 10001 respectively. Unfortunately
for town house building type, the analysis using both modelling type of
OLS and SRM were unable to be conducted due to data limitation as
there was only five usable data for it.
Table 6.4: Summary output for model comparison between OLS and
SRM
Model Available OLS SRM Model
Data R2 AIC Moran’s I (Z- Residual R2 AIC Type
score, p-value) Pattern
OVERALL 5625 0.505 95578 45.857, p=0.000 Clustered 0.592 94707 Spatial
Lag
Intermediate 2160 0.497 32989 77.075, p=0.000 Clustered 0.838 30962 Spatial
Terrace Lag
Corner 773 0.528 11994 24.632, p=0.000 Clustered 0.757 11586 Spatial
Terrace Lag
Semi 2150 0.584 34768 84.987, p=0.000 Clustered 0.843 33075 Spatial
Detached Error
Detached 537 0.526 10001 3.596, p=0.000 Clustered 0.529 10000 Spatial
Lag
Townhouse 5 Unable to do to analysis because lack of data
Once the new property value was obtained based on the SRM model,
property rating value map was then produced to visualize the pattern in
the study area. This was discussed in the following section.
143
Based on the beta coefficient result in table 6.3 for SRM (spatial lag)
model that represented the overall area, the police centre, post office and
open space (field) provide the significantly major influence to the
property rating value. This shows that most of the residence in Kota
Kinabalu area chooses public facilities as their main consideration in
owning a residential property especially that can provide safe
neighbourhood. Topography factor and Chinese temple depicted as the
least influence factor for this model. The former even have a negative
magnitude in the unstandardized coefficient with -168.32 which
indicates that it reduced property value as steep landscape and swampy
area were definitely put off the interest to most of the potential owner of
the residential property.
144
Figure 6.1: New property rating map applied for overall model
145
This study then produced individual maps based on each building type
model to show property rating values in greater detail. All the maps
which represent the 4 building type models were shown in figure 6.2 to
6.5. In figure 6.2, the Intermediate Terrace model indicates that this
building type covered high values mostly in Bukit Padang area followed
by Kolam, Luyang Timur and Ridge at the bottom part of the study area.
The intermediate terrace property value in this area was much influenced
also due to police centre, post office and open space (field) as indicated
by SRM’s spatial lag in the output of table 6.3. However, the upper side
of the map displayed mostly low values for area such Luyang, Likas and
Fung Yee Ting.
The corner terrace model as shown in figure 6.3 displayed more or less
similar to intermediate terrace model. The concentration of high values
also focused at the bottom part of the study area especially in Dah Yeh,
Bukit Padang and Kolam. Based on SRM’s spatial lag in table 6.3, police
centre, bank and building quality were the main significant influence of
this building type. While the topography variable, was the least influence
to the property value.
146
Figure 6.2: New property rating map for intermediate terrace building type model
147
Figure 6.3: New property rating map for corner terrace building type model
148
Figure 6.4: New property rating map for semi detached building type model
149
Figure 6.5: New property rating map for detached building type model
150
6.6 Conclusion
By using the spatial regression modelling method, this study was able to
produce an error free property rating value model. The results obtained
for this study shows that most of the models for the overall and based on
building type models were best represented by SRM method. This was
due to the present of spatial autocorrelation error in all of the models. In
SRM, both LM error and LM lag model were present. The LM lag was
significant for the overall model. While for building type modelling, the
LM lag represent in three (3) models compared to LM error which
represent one (1) model. In comparison with the OLS model, the SRM
achieved better performance by producing higher R2 and lower AIC for
all models. Lastly, the models were applied to the study area and results
were presented as property rating value map. The maps were produced
for the overall and each building type of the study area. The maps were
useful for the local authorities to identify which value should be applied
for the property rating. Although the values were presented in ranges, it
was sufficient to provide rough estimation of the property value.
151
CHAPTER 7
7.1 Introduction
In this study, two types of property rating value models were developed
using spatial statistics methods namely SRM. The models were
developed to estimate the new residential property rating value of Kota
Kinabalu under the jurisdiction of Kota Kinabalu City Hall (DBKK).
The model was produced for the overall study area and based on
building types within the study area. These models were assessed to
identify whether the OLS or SRM is suitable to represent the study area.
Furthermore, it will be able to assist the valuer to understand the strength
and limitations of each model and therefore, make the right choice. The
following section discusses the findings of the study and followed by the
discussion of the originality of the research and subsequently the
limitations of the research which lead to the potential areas for future
research.
152
developed for overall residential property and based on building type in
DBKK. The model is based on 15 independent variables to determine the
rating value as dependent variable. Furthermore, these models were
tested for model errors namely multicollinearity, heterosdasticity and
spatial heterogeneity or spatial autocorrelation.
Once the model was tested, it was then used to estimate property value
for the overall and based on building type in DBKK area. The result
indicated that OLS model only managed to produce 50.5% for the
goodness-of-fit in R2 for the overall model. The OLS model however,
indicated the presents of spatial autocorrelation. Therefore, SRM’s
Spatial Lag model was used in which it managed to obtain 59.2% in R2.
For model performance comparison, the AIC value obtained in SRM was
94707 compared to OLS (95578). This shows that the SRM performs
better as it obtained the lowest value in AIC measurement. The study
also produced model based on building type for intermediate terrace,
corner terrace, semi-detached and detached. The town house building
type was unable to be analysed due to data limitation as there was only
five usable data for it. For the model based on building type, SRM’s
spatial lag model was much more suitable to represent the intermediate
terrace, corner terrace and detached building type, while SRM’s spatial
error model best represent the semi-detached building type. It was noted
that all the SRM models based on building type performed better than
the OLS building type models. The SRM produced high R2 for
intermediate terrace, corner terrace, semi-detached and detached with
0.838, 0.757, 0.843 and 0.529 respectively compared to OLS with 0.497,
0.528, 0.584 and 0.526 respectively. The SRM also performed better
with lowest AIC value for intermediate terrace, corner terrace, semi-
detached and detached building type with 30962, 11586, 33075 and
10000 respectively compared to OLS with 32989, 11994, 34768 and
10001 respectively.
The study also found out that with data segmentation based on building
type, it was able to improve the model performance. This was evidence
based on the big marginal differences of R2 achieved by the highest
SRM buiding type model with 0.843 compared to the SRM overall
model with 0.592. Therefore, the segmentation of the data based on
building type was capable to improve the model’s performance.
154
Therefore, although the modeling of OLS and SRM had been used in
other countries, but it was not been officially implemented here by local
authorities in Malaysia. Eventhough Taher Buyong (2011) conducted
studies using all three of the models done in this study, but only small
dataset was used in which it eventually lead towards GWR model. The
differences can be clearly seen from this study as large dataset would
bring the modeling to SRM. Thus, this study best represents the scenario
of property rating valuation in Malaysia as it also involves large dataset
with 30% of the total data in the study area and this is valid based on
Krejcie and Morgan (1970:607-610). This is important as the local
authorities need to cover the valuation process for the whole property in
the area under its jurisdiction.
Previous study that uses these models also did not take into account to
incorporate or integrate all the tests as conducted in this study. With all
the error identified and rectified, this study was able to obtain the most
suitable model for the study area and at the same time be able to address
the problem regarding the valuation process conducted in DBKK.
Lastly, by having a reliable and valid model, this study produces a major
contribution towards improving revaluation exercise in such accurate
property rating could be obtained and at the same time cost, time and
manpower can be minimized even for large dataset.
However, the suitability of the model to represent the area involved can
still be improved if not because of the limitation faced by the study. This
limitation would be discussed in the following section.
In the second issue, missing variables were unavoidable in this study due
to data limitation. Apart from the incomplete records from the existing
variables that needed to be filled-in (as mentioned above), other
unavailable but relevant variables were not included in the DBKK
database. These data were not obtainable for this study as it involves
volume of work which is not practical given the time constraint. The
unavailable variables that might be important are surrounding viewing
factor, depreciation of property structure, crime level, noise pollution
effect and valuation approach. Viewing factor is to measure the
surrounding scenery aspect from the property (Yu et al., 2005). In
addition, property structure depreciation would enable the study to
measure depreciation of the property value (Suriatini Ismail, 2005).
Crime level data can be used to measure the safety of the property
surrounding area (Lynch and Rasmussen, 2001; Tita et al., 2006). While
the noise pollution effect is to measure the loudness or deafening sound
of the area that can affect to those who intends to live in quiet
neighborhood. This is related to property located in nearby airport (Pitt
and Jones, 2000). Last but not least, the valuation approach applied by
DBKK that causes most of the spatial autocorrelation error in the study.
These data could reflect the property rating value of the area and would
probably be able to explain the model better if obtained and included in
the model.
Finally, the third issue regarding the location factors. The distance
measurement from the location factors to the nearby properties was
applied in this study using NEAR function in ArcGIS 9.3 sofware in
which 19 location factors were identified and used. Euclidiean distance
was calculated between the two points of the location factor to the
nearby properties using NEAR function. Unfortunately, the approached
used to calculate location factor did not take into consideration specific
travel elements such as mode of travel, time of travel, maximum speed
limit and traffic flow density. This information might be useful in
determining the accurate distance measurement of location factor that
could influence the property value.
156
7.5 Recommendation and direction of future study
Further research is needed to address issues mention in the previous
section. Spatial statistics model developed for property rating valuation
needs to take into account the data from DBKK where identified missing
variables detected by this study should be included in the future studies.
The study suggests that the variables which have incomplete data are
required to be filled-in. However, this requires massive digital data input
as most of the data are still on papers. The incomplete data variables
consist of main floor area, ancillary floor area, number of room, number
of storey, building age, building extension and external usage. These
variables are the main factor of influence to the property value (Taher
Buyong, 2011; Wong, 2011) and some of it could probably be the
missing variables that needed to overcome the model error. Other than
that, the measurement of available variables which were included in the
model may require to be updated. For example, the property type that
indicates detached house need to be updated as permanent structure
detached house and semi-permanent structure detached house. This
would probably solve the problem of spatial autocorrelation as depicted
in the study. Apart from that, other methods to test the error can be
explored such as by using Luc Anselin’s Local Morans I for local spatial
autocorrelation error (Anselin, 1995). This method was not conducted in
this study because of its complexity that may cause problems especially
when comparing with global model (Lloyd, 2007:3) but it may produce
different outcome if the study focused on local model only.
Next, in response to the second issue, there are various ways to acquire
the unavailability or missing data mentioned. Planning authority or the
Quantitity Surveyor (QS) is one of the sources that can provide the
information for property structure depreciation data. Based on its record,
it showed the year the buildings of the area were developed or obtained
the certificate of fitness (CF) and subsequently, the depreciation of the
building can be estimated to the present time with the assistance from the
QS. For the other variables, GIS analysis could be used to construct the
measurement influence of viewing factor, crime rate and noise level
index. However, this can only be done if sample data across the study
can be obtained. For example, the viewing factor could be obtained from
the groundwork of the area, crime level from the police authority and
noise pollution from the environmental department. For the
socioeconomic data, it can be acquired from the Department of
Malaysian Statistics. Those data might be costly and time consuming to
collect. Lastly and probably the most important, is the different valuation
approach by DBKK which needed to be included in the model. This is
critical as some area use different approach than the other area.
157
Therefore, a detailed personal discussion with the DBKK personnel is
needed to find ways on how it can be converted into suitable variables
which the model can represent the area better. This will help to reduce or
eliminate the spatial autocorrelation error thus avoid spatial lag and
spatial error to occur as the usage of other model such as GWR can
produce better estimation.
Finally, the recommendation for the third issue from this study is by
providing an accurate distance measurement of location factor by using
the road network in the area. Network analysis from GIS can be adopted
to estimate the distance based on the road network as it also can include
the time of travel, speed limit, road traffic and mode of travel in the area.
This information can be acquired from the Public Works Department
(JKR) or from any independent study. Although the data preparation for
this factor is very time consuming, but it has the potential to produce an
accurate distance measurement for the property rating model. Thus, this
would be a good approach for future studies to focus on.
158
BIBLIOGRAPHY
Abdul Hamid Mar Iman. (1999). GIS-MRA Techniques in Property
Valuation: A Framework for Implementation. In Buletin
Geoinformasi, pp. 63-79.
Almy, R. (2010, June). IAAO at 75: Reminiscences About The Past And
Speculation About The Future. Fair & Equitable, 8.
159
Anselin, L. (1992a). Spacestat Tutorial: A Workbook for Using
SpaceStat in the Analysis of Spatial Data. Santa Barbara, CA:
NCGIA, University of California.
160
Bagdonavicius, A., & Deveikis, S. (2005). Individual and Mass
Valuation – Present and Future Practices. Paper presented at the
FIG Working Week 2005 and GSDI-8.
Bateman, I. J., Day, B., Lake, I., & Lovett, A. (2001). The Effect of Road
Traffic on Residential Property Values: A Literature Review and
Hedonic Pricing Study. ESRC, UCL: UEA.
Berry, J., McGreal, W. S. J., Stevenson, S., Young, J., & Webb, J. R.
(2003). Estimation of Apartment Submarkets in Dublin, Ireland.
Journal of Real Estate Research, 25(2), pp. 159-170.
Brace, N., Kemp, R., & Snelgar, R. (2002). SPSS for Pyschologist: A
Guide to Data Analysis Using Spss for Windows (5th ed.).
Hampshire: Palgrave Macmillan.
162
Brunsdon, C., Fotheringham, S., & Charlton, M. (1996). Geographically
weighted regression: a method for exploring spatial
nonstationarity. Geographical Analysis, 28, pp. 281-298.
Can, A., & Megbolugbe, I. (1997). Spatial Dependence and House Price
Index Construction. Journal of Real Estate Finance and
Economics, 14, pp. 203-222.
Carbone, R., & Longini, R. (1977). A feedback model for automated real
estate assessment. Management Science, 24(3).
163
Charlton, M., & Fotheringham, A. S. (2009). Geographically Weighted
Regression. Maynooth, Co Kildare, Ireland: National University
of Ireland Maynooth.
Chasco, C., Garcia, I., & Vicens, J. (2007). Modeling spatial variations
in household disposable income with Geographically Weighted
Regression. Munich Personal RePEc Archive (MPRA).
Clapp, J., & Rodriguez, M. (1998). Using a GIS for Real Estate Market
Analysis: The Problem of Spatially Aggregated Data. Journal of
Real Estate Research, 16(1), pp. 35-55.
164
Coley, M. C., Florkowski, W. J., & Bowker, J. M. (2006). Valuing
House and Landscape Attributes: Application of the Hedonic
Pricing Technique Investigating Effects of Lawn Area on House
Selling Price. Paper presented at the Southern Agricultural
Economics Association Annual Meetings.
Daly, J., Gronow, S., Jenkins, D., & Plimmer, F. (2003). Consumer
behaviour in the valuation of residential property: A comparative
study in the UK, Ireland and Australia. Property Management,
21(5), pp. 295-314.
Des Rosiers, F., Lagana, A., & Theriault, M. (2001). Size and Proximity
Effects of Primary Schools on Surrounding House Values. Journal
of Property Research, 18(2), pp. 149-168.
Des Rosiers, F., & Thériault, M. (2008). Mass Appraisal, Hedonic Price
Modelling and Urban Externalities: Understanding Property Value
Shaping Processes. In T. Kauko & M. d’Amato (Eds.), Mass
Appraisal Methods: An International Pers pective for Property
Valuers (pp. 111-147). United Kingdom: Blackwell Publishing
Ltd.
Des Rosiers F., Theriault, M., & Villineuve, P. (2000). Sorting out
Access and Neighbourhood Factors in Hedonic Price Modelling.
Journal of Property Investment & Finance, 18(3), pp. 149-168.
Din, A., Hoesli, M., & Bender, A. (2001). Environmental Variables and
Real Estate Prices. Urban Studies, 38(11), pp. 1989-2000.
166
Dzurllkanian Daud, Buang Alias, & Chitrakala Muthuveerappan. (2008).
The Needs For Capacity Building In Local Government In
Malaysia (With Regards To Property Taxation Administration).
Paper presented at the International Real Estate Research
Symposium (IRERS) 28 - 30th April 2008, PWTC Kuala Lumpur,
Malaysia.
Dzurllkanian Daud, Eboy, O. V., Buang Alias, Mohd Safie Mohd, &
Mustafa Omar. (2006, August 25-26). The Prospect of Applying
CAMA System for Rating Valuation in Malaysia. Paper presented
at the 1st Symposium of the World's Best Practices in Mass
Appraisal for Asia, Bangkok, Thailand.
Eckert, J. K., Robert, J. G., & Almy, R. (1990). Property Appraisal and
Assessment Administration. Chicago: International Association of
Assessing Officers.
167
ESRI. (2013). Geographically Weighted Regression (Spatial Statistics).
Retrieved 1 May 2013, from
http://resources.esri.com/help/9.3/arcgisdesktop/com/gp_toolref/sp
atial_statistics_tools/geographically_weighted_regression_spatial_
statistics_.htm
Evans, A., James, H., & Collins, A. (1992). Artificial neural networks:
an application to residential valuation in the UK. Journal of
Property Valuation and Investment, 11(2), pp. 195-203.
168
Fotheringham, A. S., & O’Kelly, M. E. (1989). Spatial Interaction
Models: Formulations and Applications. Kluwer, North Holland.
Fung, D. S., Kung, H.-t., & Barber, M. C. (1995). The application of GIS
to mapping real estate values. The Appraisal Journal, 63(4), pp.
445 - 450.
Gallimore, P., Fletcher, P., & Carter, M. (1996). Modeling the influence
of location on value. Journal of Property Valuation and
Investment, 14(1), pp. 16-19.
169
Goodman, A. C., & Thibodeau, T. G. (2003). Housing Market
Segmentation and Hedonic Prediction Accuracy. Journal of
Housing Economics, 12, pp. 181-201.
Hening Widi Oetomo, & Ruslan Rainis. (2002). Peranan Faktor Struktur
Dalam Model Nilai Tanah. Jurnal Teknologi, 37, pp. 27-42.
170
Heywood, I., Cornelius, S., & Carver, S. (2002). An Introduction to
Geographic Information System. Harlow, England: Pearson
Education Ltd.
171
Ireland, M. W., & O'Connor, P. M. (2002). Location Analysis of
Commercial Properties. Assessment Journal, 9(6), pp. 21-26.
Johnes, G., & Hyclak, T. (1999). House Prices and Regional Labor
Markets. Annals of Regional Science, 33(1), pp. 33-49.
Kahr, J., & Thomsett, M. C. (2005). Real Estate Market Valuation and
Analysis. Hoboken, NJ: John Wiley & Sons Inc.
172
Kestens, Y., Theriault, M., & Rosiers, F. D. (2006). Heterogeneity in
Hedonic Modelling of House Prices: Looking at Buyer’
Household profiles. Journal Geography System, 8, pp. 61-96.
Kim, C. W., Phipps, T. T., & Anselin, L. (2003). Measuring the Benefit
of Air Quality Improvement: A Spatial Hedonic Approach.
Journal of Environmental Economics & Management, 45(1), pp.
24-39.
Kong, F., Yin, H., & Nakagoshi, N. (2007). Using GIS and landscape
metrics in the hedonic price modeling of the amenity value of
urban green space: A case study in Jinan City, China. Landscape
and Urban Planning, 79, pp. 240-252.
173
Kutner, M., Nachtsheim, C., Neter, J., & Li, W. (2004). Applied Linear
Regression Models (4th ed.). Burr Ridge, Illinois, U.S.A:
McGraw-Hill Irwin.
Lee, J., & Wong, D. W. S. (2001). Statistical Analysis with ArcView GIS.
New York: John Wiley & Sons Inc.
Lentz, G. H., & Wang, K. (2003). Residential Appraisal and the Lending
Process: A Survey of Issues. Journal of Real Estate Research,
15(1/2), pp. 11-39.
174
LeSage, J., & Pace, R. K. (2009). Introduction to Spatial Econometrics.
Boca Raton: CRC Press.
Leung, Y., Mei, C.-L., & Zhang, W.-X. (2000a). Statistical tests for
spatial nonstationarity based on the geographically weighted
regression model. Environment and Planning A, 32, pp. 9-32.
DOI:10.1068/a3162
Leung, Y., Mei, C.-L., & Zhang, W.-X. (2000b). Testing for spatial
autocorrelation among the residuals of the geographically
weighted regression. Environment and Planning A, 32, 871-890.
DOI:10.1068/a32117
Leung, Y., Mei, C.-L., & Zhang, W.-X. (2003). Statistical test for local
patterns of spatial association. Environment and Planning A, 35,
pp. 725-744. DOI:10.1068/a3550
Liao, W. C., & Wang, X. (2011). Hedonic house prices and spatial
quantile regression. Journal of Housing Economics, 21(1), pp. 16-
27.
Long, F., Paez, A., & Farber, S. (2007). Spatial Effects in Hedonic Price
Estimation: A Case Study in the City of Toronto. Retrieved from
http://sciwebserver.science.mcmaster.ca
Longley, P., Higgs, G., & Martin, D. (1994). The predictive use of GIS
to model property valuations. Journal of Geographical
Information Science, 8(2), pp. 217-235.
DOI:10.1080/02693799408901995
Malpezzi, S., 2002. Hedonic Pricing Models: a Selective and Applied Review. In: T.
O'Sullivan and K. Gibb, eds, Housing Economics & Public Policy. Oxford:
Blackwell Publishing.
176
Matthews, S. A., & Yang, T.-C. (2012). Mapping the results of local
statistics : Using geographically weighted regression.
Demographic Research, 26(6), pp. 151-156.
DOI:10.4054/DemRes.2012.26.6
McCluskey, W., Deddis, W., Mannis, A., McBurney, D., & Borst, R. A.
(1997b). Interactive Application of Computer Assisted Mass
Appraisal and Geographic Information System. Journal of
Property Valuation and Investment, 15(5), pp. 448-465.
McCluskey, W. J., Dyson, K., McFall, D., & Anand, S. (1996). Mass
Appraisal for Property Taxation: An Artificial Intelligence
Approach. Land Economics Review, 2(1), pp. 25-32.
177
Meacham, A. (1988, Summer). Applying regression analysis to real
estate appraisals. The Real Estate Appraisers & Analyst, pp. 23-
27.
Mustafa Omar, Rosdi Ab. Rahman, Dzurllkanian Daud, & Buang Alias.
(2004). Pelaksanaan Teknik Mass Appraisal Bagi Tujuan
Penilaian Kadaran Di Pihak Berkuasa Tempatan Di Malaysia.
Paper presented at the 3rd Management in Construction
Researcher Association (MICRA) Conference, AB Motel,
Langkawi.
178
Noordin Ahmad. (1997). Implikasi Pembangunan Ke Atas Model
Penilaian Barta Tanah Dengan Menggunakan Sistem Maklumat
Geografi. Kuala Lumpur: Universiti Teknologi MARA.
Orford, S. (1999). Valuing The Built Environment: GIS and house price
analysis. Hampshire: Ashgate Publishing Ltd.
Pace, R. K., Barry, R., & Sirmans, C. F. (1998). Spatial Statistics and
Real Estate. The Journal of Real Estate Finance and Economics,
17(1), pp. 5-13.
Pace, R. K., & LeSage, J. P. (2004). Spatial Statistics and Real Estate.
The Journal of Real Estate Finance and Economics, 29(2), pp.
147-148.
179
Pagourtzi, E., Assimakopoulos, V., Hatzzichristos, T., & French, N.
(2003). Real Estate Appraisal: A Review of Valuation Methods.
Journal of Property Investment & Finance, 21(4), pp. 383-401.
DOI:10.1108/14635780310483656
Pitt, M., & Jones, M. (2000). Modelling the effect of airport noise on
residential property values: an examination of the Manchester
Airport second runway. Facilities, 18(13/14), pp. 497-501.
Ralphs, M., & Wyatt, P. (1998). The application of geographic and land
information systems to the management of local authority
property. Property Management, 16(2), pp.83-91.
180
Renshaw, E. F. (1958). Scientific appraisal. National Tax Journal.
Scott, I., & Gronow, S. (1989). Valuation Expertise: Its Nature and
Application. Journal of Valuation, 8(4), pp. 362-375.
181
Scott, L., & Pratt, M. (2009). Answering Why Questions: An
introduction to using regression analysis with spatial data.
ArcUser, pp. 40-43.
Taher Buyong. (1995). GIS For Local Authorities. Buletin Ukur, 6, pp.
61-68.
183
Taher Buyong, Suriatini Ismail, Ibrahim Sipan @ Atan, Mohd Ghazali
Hashim, & Mohd Nurul Fakhri. (2008b, 28th - 30th April).
Kriging In Mass Appraisal For Rating. Paper presented at the
Real Estate Research Symposium (IRERS), PWTC Kuala
Lumpur, Malaysia.
Waller, L., Zhu, L., Gotway, C., Gorman, D., & Gruenewald, P. (2007).
Quantifying geographic variations in associations between alcohol
distribution and violence: A comparison of geographically
weighted regression and spatially varying coefficient models.
Stochastic Environmental Research and Risk Assessment, 21(5),
pp. 573–588.
185
Wheeler, D. C. (2010). Visualizing and Diagnosing Coefficients from
Geographically Weighted Regression Models. In B. Jiang & X.
Yao (Eds.), Geospatial Analysis and Modelling of Urban
Structure and Dynamics, pp. 415-436: Springer Science+Business
Media B.V. .
Wyatt, P., & Ralphs, M. (2003). GIS in Land and Property Management.
London: Spon Press.
186
Wyatt, P. J. (1996b). The development of a property information system
for valuation using a geographical information system (GIS).
Journal of Property Research, 13(4), pp. 317-336.
Yu, D., Wei, Y. D., & Wu, C. (2007). Modeling spatial dimensions of
housing prices in Milwaukee. Environment and Planning B:
Planning and Design, 34(6), pp. 1085-1102.
Yu, S.-M., HAN, S.-S., & Chai, C.-H. (2005). Modeling the Value of
View in Real Estate Valuation: A 3-D GIS Approach. Journal of
the Pacific Rim Real Estate Society, 11(3), pp. 1-22.
Zeng, T. Q., & Zhou, Q. (2001). Optimal Spatial Decision Making Using
GIS: A Prototype of a Real Estate Geographical Information
System (REGIS). International Journal of Geographical
Information Science, 15(4), pp. 307-321.
187
Appendix A
Designation: _________________________________________________
Please tick ( ¥ ) in one of the column for each variable based on your opinion that could increase the property value with the
following scale: 1 – No increase; 2 – Slightly Increase; 3 – Somewhat Increase; 4 – Very Much Increase; 5 – Extremely
Increase
Variable Value 1 2 3 4 5
Topography Swamp
Steep
Undulating
Level
Building Quality Low Cost
Basic
Average
Good
Excellent
Superior
Site Preparation No
Yes
Lot Shape Irregular
Eccentric
Compact
Frontage Nil
Uniformed road reserved
Secondary Road
Main Road
Neighbourhood Quality Poor
Average
Good
Exclusive
Flood Prone No
Yes
Road Access No Access
188
Bridge
Earth
Gravelled
Sealed
Variable Value 1 2 3 4 5
Sanitation None
Septic Tank
Treatment Plant Private
Treatment Plant DBKK
Public Sewer
Building Type Kampung House
Walk-Up Apartment
Intermediate Terrace
Semi-Detached
Corner Terrace
Town House
Detached
Building Structure Permanent
Semi-Permanent
Temporary
Building Position Corner
Intermediate
End Lot
Not Applicable
Property Category Residential
Commercial
Industrial
Vacant Land
Agriculture
Educational
Religious
Sport & Recreation
Government
Cemetery
189
Appendix B
190
Regression 5630883005.602 3 1876961001.867 1031.489 .000d
3 Residual 10228316361.543 5621 1819661.335
Total 15859199367.145 5624
e
Regression 6087763423.141 4 1521940855.785 875.338 .000
4 Residual 9771435944.005 5620 1738689.670
Total 15859199367.145 5624
f
Regression 6563164727.956 5 1312632945.591 793.423 .000
5 Residual 9296034639.189 5619 1654393.066
Total 15859199367.145 5624
g
Regression 6979925189.527 6 1163320864.921 736.044 .000
6 Residual 8879274177.618 5618 1580504.482
Total 15859199367.145 5624
h
Regression 7157970356.387 7 1022567193.770 660.109 .000
7 Residual 8701229010.759 5617 1549088.305
Total 15859199367.145 5624
i
Regression 7247983499.156 8 905997937.395 590.867 .000
8 Residual 8611215867.989 5616 1533336.159
Total 15859199367.145 5624
j
Regression 7380063080.763 9 820007008.974 543.020 .000
9 Residual 8479136286.383 5615 1510086.605
Total 15859199367.145 5624
Regression 7525641386.062 10 752564138.606 506.974 .000k
10 Residual 8333557981.084 5614 1484424.293
Total 15859199367.145 5624
Regression 7674954885.404 11 697723171.400 478.519 .000l
11 Residual 8184244481.741 5613 1458087.383
Total 15859199367.145 5624
m
Regression 7757450880.098 12 646454240.008 447.792 .000
12 Residual 8101748487.047 5612 1443647.271
Total 15859199367.145 5624
n
Regression 7756203664.803 11 705109424.073 488.434 .000
13 Residual 8102995702.342 5613 1443612.275
Total 15859199367.145 5624
o
Regression 7849943655.475 12 654161971.290 458.364 .000
14 Residual 8009255711.671 5612 1427166.021
Total 15859199367.145 5624
p
Regression 7849906887.565 11 713627898.870 500.118 .000
15 Residual 8009292479.581 5613 1426918.311
Total 15859199367.145 5624
q
Regression 7849617630.080 10 784961763.008 550.188 .000
16
Residual 8009581737.066 5614 1426715.664
191
Total 15859199367.145 5624
r
Regression 7847955063.229 9 871995007.025 611.172 .000
17 Residual 8011244303.916 5615 1426757.668
Total 15859199367.145 5624
s
Regression 7909496196.046 10 790949619.605 558.561 .000
18 Residual 7949703171.100 5614 1416049.728
Total 15859199367.145 5624
t
Regression 7944195015.827 11 722199546.893 512.155 .000
19 Residual 7915004351.318 5613 1410120.141
Total 15859199367.145 5624
u
Regression 7968092457.312 12 664007704.776 472.229 .000
20 Residual 7891106909.833 5612 1406113.134
Total 15859199367.145 5624
v
Regression 7986296882.536 13 614330529.426 437.832 .000
21 Residual 7872902484.609 5611 1403119.316
Total 15859199367.145 5624
w
Regression 8004333420.706 14 571738101.479 408.339 .000
22 Residual 7854865946.439 5610 1400154.358
Total 15859199367.145 5624
x
Regression 8018047818.882 15 534536521.259 382.369 .000
23 Residual 7841151548.263 5609 1397958.914
Total 15859199367.145 5624
y
Regression 8029906367.764 16 501869147.985 359.481 .000
24 Residual 7829292999.382 5608 1396093.616
Total 15859199367.145 5624
z
Regression 8039496449.603 17 472911555.859 339.094 .000
25 Residual 7819702917.543 5607 1394632.231
Total 15859199367.145 5624
aa
Regression 8047567439.576 18 447087079.976 320.851 .000
26 Residual 7811631927.569 5606 1393441.300
Total 15859199367.145 5624
a
Coefficients
Model Unstandardized Coefficients Standardized t Sig.
Coefficients
B Std. Error Beta
(Constant) 1877.359 31.752 59.127 .000
1
LandAr_Ft .274 .007 .468 39.736 .000
(Constant) -67.447 85.636 -.788 .431
2 LandAr_Ft .232 .007 .397 34.291 .000
BuildQ 743.606 30.638 .281 24.271 .000
192
(Constant) -1706.337 108.088 -15.787 .000
LandAr_Ft .246 .007 .422 37.911 .000
3
BuildQ 708.520 29.310 .268 24.173 .000
Bank .620 .027 .250 23.203 .000
(Constant) -1832.536 105.942 -17.298 .000
LandAr_Ft .227 .006 .388 35.128 .000
4 BuildQ 613.166 29.248 .232 20.964 .000
Bank .656 .026 .264 25.032 .000
RCA_Ft .195 .012 .179 16.210 .000
(Constant) -1249.858 108.908 -11.476 .000
LandAr_Ft .226 .006 .386 35.748 .000
BuildQ 622.761 28.536 .236 21.824 .000
5
Bank .628 .026 .253 24.494 .000
RCA_Ft .214 .012 .197 18.164 .000
Spermarket -.651 .038 -.175 -16.952 .000
(Constant) -969.352 107.841 -8.989 .000
LandAr_Ft .226 .006 .387 36.676 .000
BuildQ 575.686 28.042 .218 20.530 .000
6 Bank .886 .030 .357 29.860 .000
RCA_Ft .236 .012 .217 20.354 .000
Spermarket -.832 .039 -.223 -21.260 .000
Hospital -.591 .036 -.200 -16.238 .000
(Constant) -943.906 106.790 -8.839 .000
LandAr_Ft .225 .006 .384 36.812 .000
BuildQ 632.471 28.263 .239 22.378 .000
Bank .888 .029 .358 30.220 .000
7
RCA_Ft .247 .012 .228 21.476 .000
Spermarket -.820 .039 -.220 -21.140 .000
Hospital -.528 .036 -.179 -14.464 .000
ChinTemple -.281 .026 -.111 -10.721 .000
(Constant) -881.790 106.555 -8.275 .000
LandAr_Ft .219 .006 .375 35.889 .000
BuildQ 608.851 28.287 .230 21.524 .000
Bank .725 .036 .292 20.076 .000
8 RCA_Ft .235 .012 .216 20.304 .000
Spermarket -.749 .040 -.201 -18.862 .000
Hospital -.410 .039 -.139 -10.379 .000
ChinTemple -.438 .033 -.173 -13.207 .000
Field .221 .029 .116 7.662 .000
(Constant) -890.175 105.748 -8.418 .000
9
LandAr_Ft .214 .006 .366 35.063 .000
193
BuildQ 615.324 28.080 .233 21.913 .000
Bank .717 .036 .289 20.013 .000
RCA_Ft .216 .012 .199 18.569 .000
Spermarket -.680 .040 -.182 -16.982 .000
Hospital -.163 .047 -.055 -3.442 .001
ChinTemple -.487 .033 -.192 -14.625 .000
Field .341 .031 .179 10.870 .000
ShopMall -.430 .046 -.133 -9.352 .000
(Constant) -1212.670 109.786 -11.046 .000
LandAr_Ft .215 .006 .368 35.592 .000
BuildQ 612.381 27.842 .232 21.995 .000
Bank .457 .044 .184 10.322 .000
RCA_Ft .220 .012 .203 19.068 .000
10 Spermarket -.489 .044 -.131 -11.072 .000
Hospital -.130 .047 -.044 -2.765 .006
ChinTemple -.481 .033 -.190 -14.559 .000
Field .521 .036 .273 14.457 .000
ShopMall -.501 .046 -.154 -10.855 .000
SportArea .348 .035 .152 9.903 .000
(Constant) -2075.378 138.229 -15.014 .000
LandAr_Ft .208 .006 .356 34.515 .000
BuildQ 674.651 28.272 .255 23.863 .000
Bank .461 .044 .186 10.506 .000
RCA_Ft .227 .011 .209 19.806 .000
Spermarket -.176 .054 -.047 -3.275 .001
11
Hospital -.083 .047 -.028 -1.772 .076
ChinTemple -.188 .044 -.074 -4.294 .000
Field .723 .041 .379 17.670 .000
ShopMall -.544 .046 -.168 -11.839 .000
SportArea .515 .039 .226 13.363 .000
PoliceCtr -.523 .052 -.220 -10.119 .000
(Constant) -2071.357 137.543 -15.060 .000
LandAr_Ft .208 .006 .356 34.696 .000
BuildQ 653.470 28.271 .247 23.115 .000
Bank .369 .045 .149 8.161 .000
RCA_Ft .225 .011 .207 19.712 .000
12
Spermarket -.239 .054 -.064 -4.419 .000
Hospital -.220 .050 -.075 -4.409 .000
ChinTemple -.044 .047 -.017 -.929 .353
Field .843 .044 .442 19.292 .000
ShopMall -.555 .046 -.171 -12.129 .000
194
SportArea .342 .045 .150 7.661 .000
PoliceCtr -.774 .061 -.326 -12.644 .000
PostOffice .423 .056 .179 7.559 .000
(Constant) -2123.546 125.559 -16.913 .000
LandAr_Ft .208 .006 .356 34.697 .000
BuildQ 654.445 28.251 .247 23.165 .000
Bank .373 .045 .150 8.264 .000
RCA_Ft .226 .011 .208 19.791 .000
Spermarket -.229 .053 -.061 -4.321 .000
13
Hospital -.236 .047 -.080 -4.998 .000
Field .848 .043 .445 19.570 .000
ShopMall -.553 .046 -.170 -12.103 .000
SportArea .343 .045 .150 7.691 .000
PoliceCtr -.815 .042 -.343 -19.409 .000
PostOffice .443 .051 .188 8.655 .000
(Constant) -1622.520 139.310 -11.647 .000
LandAr_Ft .206 .006 .353 34.543 .000
BuildQ 649.577 28.096 .246 23.120 .000
Bank .169 .051 .068 3.295 .001
RCA_Ft .238 .011 .219 20.787 .000
Spermarket -.010 .059 -.003 -.161 .872
14 Hospital .014 .056 .005 .256 .798
Field 1.059 .050 .555 21.038 .000
ShopMall -.618 .046 -.190 -13.388 .000
SportArea .051 .057 .022 .890 .374
PoliceCtr -1.060 .052 -.446 -20.559 .000
PostOffice .790 .066 .334 11.877 .000
HindTemple -.340 .042 -.154 -8.104 .000
(Constant) -1628.442 134.324 -12.123 .000
LandAr_Ft .206 .006 .353 34.677 .000
BuildQ 650.042 27.944 .246 23.262 .000
Bank .166 .046 .067 3.621 .000
RCA_Ft .238 .011 .219 20.789 .000
Hospital .020 .044 .007 .450 .653
15
Field 1.064 .040 .558 26.477 .000
ShopMall -.620 .044 -.191 -14.052 .000
SportArea .052 .057 .023 .911 .362
PoliceCtr -1.065 .043 -.448 -24.544 .000
PostOffice .791 .066 .335 11.980 .000
HindTemple -.343 .037 -.156 -9.195 .000
16 (Constant) -1641.727 131.034 -12.529 .000
195
LandAr_Ft .206 .006 .353 34.796 .000
BuildQ 649.300 27.894 .246 23.278 .000
Bank .172 .044 .069 3.945 .000
RCA_Ft .238 .011 .219 20.886 .000
Field 1.059 .039 .556 27.229 .000
ShopMall -.611 .039 -.188 -15.678 .000
SportArea .059 .055 .026 1.079 .280
PoliceCtr -1.061 .043 -.447 -24.850 .000
PostOffice .787 .065 .333 12.037 .000
HindTemple -.337 .034 -.153 -9.767 .000
(Constant) -1563.442 109.138 -14.325 .000
LandAr_Ft .206 .006 .353 34.799 .000
BuildQ 645.705 27.694 .244 23.315 .000
Bank .168 .043 .068 3.865 .000
RCA_Ft .239 .011 .220 20.981 .000
17
Field 1.067 .038 .560 27.908 .000
ShopMall -.606 .039 -.187 -15.656 .000
PoliceCtr -1.083 .038 -.456 -28.836 .000
PostOffice .841 .042 .356 19.989 .000
HindTemple -.364 .023 -.165 -15.819 .000
(Constant) -393.698 208.101 -1.892 .059
LandAr_Ft .205 .006 .350 34.595 .000
BuildQ 635.077 27.637 .240 22.979 .000
Bank .131 .044 .053 3.004 .003
RCA_Ft .239 .011 .220 21.105 .000
18 Field 1.090 .038 .572 28.497 .000
ShopMall -.598 .039 -.184 -15.500 .000
PoliceCtr -1.101 .038 -.464 -29.345 .000
PostOffice .889 .043 .376 20.894 .000
HindTemple -.369 .023 -.167 -16.076 .000
TopoG -263.558 39.979 -.064 -6.592 .000
(Constant) -388.931 207.667 -1.873 .061
LandAr_Ft .208 .006 .355 34.986 .000
BuildQ 624.809 27.657 .236 22.591 .000
Bank .260 .051 .105 5.130 .000
RCA_Ft .239 .011 .220 21.132 .000
19
Field 1.101 .038 .577 28.791 .000
ShopMall -.729 .047 -.225 -15.611 .000
PoliceCtr -.966 .046 -.407 -20.863 .000
PostOffice .809 .045 .343 17.834 .000
HindTemple -.368 .023 -.167 -16.052 .000
196
TopoG -248.350 40.013 -.060 -6.207 .000
GovtOffice -.260 .052 -.089 -4.961 .000
(Constant) -71.070 221.242 -.321 .748
LandAr_Ft .207 .006 .354 34.942 .000
BuildQ 631.285 27.662 .239 22.821 .000
Bank .170 .055 .068 3.076 .002
RCA_Ft .229 .012 .210 19.724 .000
Field 1.247 .052 .654 23.916 .000
20 ShopMall -.818 .051 -.252 -15.926 .000
PoliceCtr -.874 .051 -.368 -16.996 .000
PostOffice .808 .045 .342 17.833 .000
HindTemple -.407 .025 -.185 -16.414 .000
TopoG -261.220 40.078 -.063 -6.518 .000
GovtOffice -.226 .053 -.077 -4.272 .000
Market -.181 .044 -.107 -4.123 .000
(Constant) -348.783 234.069 -1.490 .136
LandAr_Ft .206 .006 .353 34.814 .000
BuildQ 624.886 27.690 .236 22.567 .000
Bank .196 .056 .079 3.519 .000
RCA_Ft .225 .012 .207 19.417 .000
Field 1.242 .052 .652 23.843 .000
ShopMall -.833 .051 -.256 -16.179 .000
21
PoliceCtr -.888 .052 -.374 -17.244 .000
PostOffice .800 .045 .339 17.662 .000
HindTemple -.396 .025 -.180 -15.864 .000
TopoG -243.887 40.323 -.059 -6.048 .000
GovtOffice -.246 .053 -.084 -4.624 .000
Market -.162 .044 -.095 -3.662 .000
FloodP 54.634 15.168 .035 3.602 .000
(Constant) -474.857 236.445 -2.008 .045
LandAr_Ft .207 .006 .354 34.934 .000
BuildQ 626.284 27.663 .237 22.639 .000
Bank .251 .058 .101 4.360 .000
RCA_Ft .225 .012 .207 19.404 .000
ChinTemple .171 .048 .067 3.589 .000
22
Field 1.268 .053 .665 24.135 .000
ShopMall -.917 .057 -.283 -16.214 .000
PoliceCtr -.977 .057 -.411 -17.112 .000
PostOffice .835 .046 .354 18.040 .000
HindTemple -.397 .025 -.180 -15.921 .000
TopoG -248.268 40.299 -.060 -6.161 .000
197
GovtOffice -.355 .061 -.121 -5.799 .000
Market -.166 .044 -.098 -3.766 .000
FloodP 61.292 15.265 .040 4.015 .000
(Constant) -717.019 248.589 -2.884 .004
LandAr_Ft .207 .006 .354 34.987 .000
BuildQ 619.693 27.722 .234 22.354 .000
Bank .271 .058 .109 4.683 .000
RCA_Ft .221 .012 .204 19.024 .000
ChinTemple .244 .053 .097 4.610 .000
Field 1.218 .055 .639 22.152 .000
ShopMall -.924 .057 -.285 -16.334 .000
23
PoliceCtr -1.082 .066 -.456 -16.338 .000
PostOffice .999 .070 .423 14.307 .000
HindTemple -.367 .027 -.167 -13.710 .000
TopoG -245.068 40.281 -.059 -6.084 .000
GovtOffice -.254 .069 -.087 -3.686 .000
Market -.125 .046 -.073 -2.705 .007
FloodP 68.489 15.425 .044 4.440 .000
OfficeCtr -.193 .061 -.073 -3.132 .002
(Constant) -68.560 333.495 -.206 .837
LandAr_Ft .205 .006 .351 34.554 .000
BuildQ 609.107 27.940 .230 21.800 .000
Bank .283 .058 .114 4.882 .000
RCA_Ft .220 .012 .203 18.928 .000
ChinTemple .221 .054 .087 4.116 .000
Field 1.205 .055 .632 21.876 .000
ShopMall -.915 .057 -.282 -16.159 .000
24 PoliceCtr -1.109 .067 -.467 -16.595 .000
PostOffice 1.019 .070 .432 14.536 .000
HindTemple -.357 .027 -.162 -13.250 .000
TopoG -245.831 40.254 -.059 -6.107 .000
GovtOffice -.241 .069 -.082 -3.482 .001
Market -.091 .047 -.054 -1.917 .055
FloodP 73.489 15.510 .048 4.738 .000
OfficeCtr -.206 .062 -.078 -3.339 .001
Sanitation -170.081 58.358 -.030 -2.914 .004
(Constant) -970.301 478.907 -2.026 .043
LandAr_Ft .207 .006 .353 34.669 .000
25 BuildQ 607.995 27.929 .230 21.769 .000
Bank .282 .058 .114 4.869 .000
RCA_Ft .221 .012 .203 19.001 .000
198
ChinTemple .219 .054 .086 4.086 .000
Field 1.205 .055 .632 21.891 .000
ShopMall -.910 .057 -.280 -16.071 .000
PoliceCtr -1.113 .067 -.468 -16.650 .000
PostOffice 1.022 .070 .433 14.585 .000
HindTemple -.359 .027 -.163 -13.313 .000
TopoG -252.395 40.311 -.061 -6.261 .000
GovtOffice -.237 .069 -.081 -3.432 .001
Market -.088 .047 -.052 -1.855 .064
FloodP 74.193 15.504 .048 4.785 .000
OfficeCtr -.206 .062 -.078 -3.342 .001
Sanitation -172.293 58.333 -.031 -2.954 .003
LotShp 258.113 98.430 .025 2.622 .009
(Constant) -1901.244 615.453 -3.089 .002
LandAr_Ft .206 .006 .352 34.571 .000
BuildQ 605.332 27.939 .229 21.666 .000
Bank .272 .058 .109 4.671 .000
RCA_Ft .221 .012 .203 18.980 .000
ChinTemple .226 .054 .089 4.207 .000
Field 1.203 .055 .631 21.849 .000
ShopMall -.908 .057 -.280 -16.038 .000
PoliceCtr -1.102 .067 -.464 -16.460 .000
26 PostOffice 1.024 .070 .434 14.609 .000
HindTemple -.362 .027 -.164 -13.429 .000
TopoG -252.271 40.294 -.061 -6.261 .000
GovtOffice -.248 .069 -.085 -3.584 .000
Market -.088 .047 -.052 -1.859 .063
FloodP 76.762 15.534 .050 4.942 .000
OfficeCtr -.203 .062 -.077 -3.299 .001
Sanitation -170.802 58.312 -.030 -2.929 .003
LotShp 257.654 98.388 .025 2.619 .009
Build_Type 201.058 83.542 .023 2.407 .016
a. Dependent Variable: Rate_Value
199