Figure 2.2 The context of the Construction Industry in the National Economy
STAGE 3: STATISTICAL SELECTION STAGE 4: USAGE <N^
No On O Y es e s R e m o v e from further co n sid eratio n Is it s e le c te d b y th e m o re re stric tiv e p r o c e d u r e ? Is th e indicator s e le c te d by th e p ro c e d u re ? C a n its statistical insignificance b e justified ? U se a s a m o d ellin g v a ria b le for co n stru c tio n d e m a n d Apply se le c tio n p ro c e d u re s to list of ind icato rs U se a s a n ad d itio n al m odelling v a ria b le b a s e d o n e c o n o m ic sig n ifican c e
U se to co m p ile a n e x h a u s tiv e list of indicators th a t h a s m e t e c o n o m ic sig n ifican c e a n d statistica l a d e q u a c y C h o o s e a restrictive a n d a le s s restrictive v a ria b le se le c tio n m e th o d to g e th e r with a p p ro p ria te le v e ls of sig n ifican c e
of demand for construction in order to identify potential influencing factors. The idea is to consider
all possible factors at an early stage, before eliminating those that do not conform to economic theory
in the next step. It implies that a thorough understanding of the nature o f demand for construction
and its associated economic and social factors is crucial. Hence, a good comprehension o f the theory
o f construction economics is required in order to undertake the stage o f theoretical identification
effectively. There is a direct implication for modellers to be proficient in the statistical aspects o f
modelling, as well as, knowledgeable o f economic theory.
The second imphcation is that modellers would not be able to pre-determine the number o f variables
to be used, which has been the case in past studies (Tang et al., 1990; and Akintoye and Skitmore, 1994). Without this restriction, it means that more potential indicators can be identified. This is in
support o f the second objective o f demand modelling which is to build more realistic explanatory
models. On a broader perspective, it also helps to promote an in-depth understanding o f the intricate
relationship between the general economy and the construction industry.
Following the second implication, less subjectivity involved in the theoretical identification of
indicators would imply the building o f better predictive models. A thorough examination o f all
possible influencing factors increases the likelihood o f identifying more significant modelling
variables. Hence, the third implication is that demand models with higher forecasting abilities can
be expected as a result of using more significant modelling variables. This, in effect, fulfils the third objective o f construction demand modelling.
Increased complexity in future modelling tasks is the final implication. Apparently, the consideration
o f more influencing factors increases the number of modelling variables. Hence, modellers must be
prepared for greater complexities in the modelling process. As complex models do not necessarily mean better models, a general observation made by some writers (Jenkins, 1979; and Makridakis
et a l , 1982), expert judgement needs to be applied in order to keep a balance between complexity and efficiency. Better knowledge of the data and subject matter would help to ensure that a
parsimonious set o f variables is used. The proposed approach allows for judgemental input in Stage
Three where the degree of complexity can be regulated by the type o f variable selection method and
the level o f significance chosen. Modellers would have to be familiar with the use o f alternative
selection methods and the relevance o f different levels o f significance in order to play a more active role in the future.
7.5.2 Problems anticipated
From the discussion of the implications o f the proposed stage-by-stage approach, several problems
can be anticipated. It is necessary to examine these potential problems and recommend courses o f
action to resolve them.
The first problem relates to the availability o f data. As the approach advocates a thorough
examination o f all possible factors o f construction demand, it raises the expectations o f the
availabihty, accessibility and quality of statistical data. Without a broad base o f economic and social
statistics, the statistical significance of suitable indicators cannot be tested owing to the
unavailability o f data. To help resolve this problem, the resourcefulness o f the researcher is
important. Often, many do not go beyond extracting information fi'om national data sources. There
are also external research organisations compiling vital national statistics o f countries worldwide,
with the aim o f providing potential foreign investors an independent account o f a country's past and
present economic position. International institutions such as. Organisation for Economic Co
operation and Development (OECD), International Monetary Fund (IMF) and the United Nations,
also compile and publish information on national economic indicators o f many countries. In short, it is recommended that for studies relying on data fi'om secondary sources, a detailed search o f all
available published sources is warranted. It increases the likelihood o f obtaining more data which may be crucial to the expansion o f the scope o f the research.
Following the difficulty in data collection, the second problem is associated with the time factor. It
is apparent that the approach will be more time consuming as more factors have to be considered and
more data to be collected. However, if a more accurate and realistic model is compensated for the
longer time spent, the trade-off may prove to be worthwhile. Nevertheless, it is recommended that
initial studies adopting the approach should carry out comparative studies to justify the proposition
that the consideration of more associated indicators indeed produces better forecasting models.
Otherwise, the mere fact that the approach eliminates subjectivity in the choice o f suitable modelling variables may alone be a sufficient basis for its adoption.
In a statistical sense, the use o f more modelling variables increases the chance o f encountering
problems such as, overfitting and multicollineahty. Therefore, it is recommended that extra care is given during the theoretical identification stage. Indicators which measure similar economic or social
m ultiple correlation among them, giving rise to problems o f multicollinearity. Over-enthusiasm during the identification stage causes too many factors to be considered. As a result, time is wasted
on data collection and the risk o f overfitting the model increases. Hence, ample knowledgable o f the theory of construction economics is required in order to be more selective at this early stage. Better
modelling skills are also necessary to detect and resolve the potential problem o f multicollinearity
at a later stage.
It is clear that the proposed approach imposes on the modeller the requirement to familiarise himself
with alternative variable selection methods. Without acquiring better technical skills, the chances
o f deriving spurious and unparsimonious models are high. More advanced methods, such as
principal component regression and biased estimation, may also be usefiil. They can serve as viable
alternatives to variable selection methods especially if the primary objective is to study the structure
o f the regression relationship and not only to predict values o f future demand.
7.6 Chapter summary
This chapter has proposed a stage-by-stage approach to the theoretical identification and statistical
selection o f economic indicators for construction demand modelling. The systematic approach has
been rigorously discussed which included a detailed description o f each stage, a schematic
representation, implications of its use and the potential problems.
In Section 7.2, the steps involved in the first stage o f theoretical identification were elaborated. The
first step entails a detailed examination of the characteristics and determinants o f the type o f demand
for construction concerned. This is followed by a thorough literature search to obtain a list of
economic and social factors that has been associated with this type o f demand. The final step
involves the selection of one or more economic indicators that represent each o f these factors.
The discussion in Section 7.3 dealt chiefly with the statistical selection stage. It began with a
description o f the steps involved in the data collection and preprocessing stage. It comprised the re
classification of economic indicators using statistical terminology, the collection o f economic time-
series data from published sources and data transformation to remove the unwanted statistical
properties. The objective is to collect and prepare the data for the statistical selection stage. In this
of variable selection methods and levels of significance to be applied. Generally, by determining the
usage o f the indicators, the choice of appropriate selection methods and levels o f significance could
be imphed. A less restrictive method would produce an exhaustive list o f indicators which satisfies
both economic significance and statistical adequacy. A more restrictive method would derive a
shorter list o f indicators that is suitable for modelling purposes. A list o f additional modelling
variables can also be obtained.
In Section 7.4, a schematic representation o f the four-stage approach was given in Figure 7.1. The
section also explained the need for a systematic approach and outlined the different steps involved
in each stage.
Finally, the benefits and shortcomings o f the proposed approach were discussed in Section 7.5. In
particular, four imphcations o f the use o f the approach were highlighted. They relate to the main
objectives o f construction demand modelling. The first implication was that the approach would
encourage a more thorough examination o f the different economic and social factors that influence
construction demand, especially with a better understanding o f the characteristics and determinants
of demand. It also meant that modellers would have to be more knowledgeable o f economic theory
in order to fulfil their future role. The second implication was that the approach would allow for
more potential indicators to be considered, increasing the likelihood o f developing better explanatory
models which can reveal more meaningfiil relationships between the influencing factors and demand
for construction. With less subjectivity in the choice o f suitable indicators, the third implication was that better predictive models should emerge as a result o f using more relevant variables after taking
into consideration all possible ones. Increased complexity o f fiiture modelling tasks was the final
imphcation. This is the result o f using more modelling variables. It was emphasised that care has
to be taken to keep a balance between complexity and efficiency. In this respect, the choice of
appropriate variable selection methods and levels o f significance plays an important part in
regulating the complexity: of the modelling process. Several problems were also anticipated and they
are; non-availability of data, more time-consuming, more prone to problems such as
multicollineahty and overfitting, and a higher chance o f deriving spurious models owing to a lack of understanding o f different statistical selection methods. Recommendation o f ways o f resolving
CHAPTERS
THE REGRESSION APPROACH TO FORECASTING: