Bond University

Research Repository

Forecast models for actual construction time and cost

Skitmore, R. Martin; Ng, S. Thomas

Published in:
Building and Environment

DOI:
10.1016/S0360-1323(03)00067-2

Licence:
CC BY-NC-ND

Link to output in Bond University research repository.

Recommended citation(APA):
Skitmore, R. M., & Ng, S. T. (2003). Forecast models for actual construction time and cost. Building and
Environment, 38(8), 1075-1083. https://doi.org/10.1016/S0360-1323(03)00067-2

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

For more information, or if you believe that this document breaches copyright, please contact the Bond University research repository
coordinator.

Download date: 29 Jun 2023

 Paper for

Building and Environment

Forecast Models for Actual Construction Time and Cost

R. Martin Skitmore 1
S. Thomas Ng 2
1
School of Construction Management and Property, Queensland University of Technology,
GPO Box 2434, Brisbane Q4001, Australia.

2
Department of Civil Engineering, The University of Hong Kong,
Pokfulam Road, Hong Kong.

Please contact:

Dr. S Thomas Ng
Department of Civil Engineering
University of Hong Kong
Pokfulam Road, Hong Kong

Tel: Int+ (852) 2857 8556

Fax: Int+ (852) 2559 5337
Email: [email protected]

Version 9a (19th March 2003)

1
FORECAST MODELS FOR ACTUAL CONSTRUCTION TIME AND COST

Abstract:

The actual construction time and cost of construction projects may be affected by the client,

project and contractual characteristics and in many cases can be very different from the

contract time and cost. In this paper, details of 93 Australian construction projects are used to

develop several models for actual construction time and cost prediction. A forward

crossvalidation regression analysis is used for the development of the model for actual

construction time forecast when client sector, contractor selection method, contractual

arrangement, project type, contract period and contract sum are known. The standard

deviation of the deleted residual indicates the best model for actual construction time

prediction to comprise the independent variables log contract time, lump sum procurement

and non-standard contractor selection. Regression models are also developed for forecasting

the actual construction time and cost when client sector, contractor selection method,

contractual arrangement and project type are known while contract period and contract sum

are estimated. Different forms of regression analyses, including the standard regression and

the crossvalidation regression, are used and the crossvalidation regression model with the

smallest deleted residual sum of squares is selected.

Since these models for time and cost are dependent on the contract period and contract sum

being known, it is necessary to investigate the effects in situations where these have to be

estimated. The results of the sensitivity analyses show that the errors in predicted actual

construction time become smaller as the contract period increases. In contrast, the errors in

predicted actual construction cost are virtually the same for large and small projects.

2
The effects of different project type, contractor selection method and contractual arrangement

are also examined. The results indicate that the actual construction time for industrial project

is the longest when compared with residential, educational and recreational projects and that

significant savings in actual construction time can be achieved when negotiated tender and

design and build contract are used instead of the traditional open tendering and lump sum

contract approaches.

Finally, some practical applications of the models are illustrated for predicting the actual

construction time and cost based on the risks and uncertainties of different client sector,

contractor selection method, contractual arrangement and project type.

Keywords: Construction, time, cost, forecasting, regression, crossvalidation.

3
INTRODUCTION

An accurate forecast of construction time and cost is crucial to contract administration as the

predicted duration and cost forms a basis for budgeting, planning, monitoring and even

litigation purposes. In practice, there are two common methods for estimating construction

time and cost: (1) according to the client’s available budget and time constraints, i.e.

occupancy need, or (2) through a detailed analysis of work to be done and resources

available, using estimates of the time and cost requirements for each specific activity

(Telford, 1994). The detailed estimation of construction activities usually relies on the

estimators’ experience and judgement to correctly interpret project and site information and

make the best possible decisions (Alfred, 1988).

To reduce the subjectivity of time-cost estimation, Bromilow (1969) developed a regression

model for predicting contract duration based on the estimated final cost of construction

project. The model, commonly known as Bromilow’s Time-Cost (BTC) model1, is used by

clients and contractors for estimating and benchmarking the contract period of construction

projects. Various research studies in Australia (Bromilow and Henderson, 1976; Bromilow et

al, 1980, 1988; Ireland, 1983; Mak, 1991; Ng et al, 2000; RAIA, 1989; Sidwell, 1984;

Walker, 1994, 1995), United Kingdom (Kaka and Price, 1991), Hong Kong (Chan, 1999;

Kumaraswamy and Chan, 1995) and Malaysia (Yeong, 1994) revealed that the time taken to

construct a project is highly correlated with the construction size as measured by cost.

Ireland (1983:137) concluded that the BTC model is “the best predictor of construction time”.

1
Detailed discussion of the BTC model can be found in Ng et al (2001)

4
Despite the development and use of the BTC model for construction time-cost estimation

(developed at the pre-contract stage), the actual construction time and cost (at the post-

contract stage) might be influenced by risks and uncertainties emanated from different client

sector, project type, contractor selection method and contract arrangement approach.

According to Walker (1994), one shortcoming of the BTC model is that it fails to consider

factors other than cost when establishing the contract duration.

Several research studies (Ireland, 1983; Laptali et al, 1996) have been carried out to improve

the accuracy of the BTC model. Ireland (1983) developed a multiple regression model based

on the construction time, cost, area and number of storeys. Progress was halted by the

occurrence of unreasonably high standard errors. Walker (1994) also measured construction

time performance in terms of the gross floor area of a building. However, problems occurred

as the construction cost included a significant external works component, which present

difficulties in measuring construction scope per unit of construction time.

This paper examines the relationships among the actual construction time/cost and other

contract details, such as the estimated construction time and cost, client sector, project type,

contractor selection method and contractual arrangement, through 93 recently completed

construction projects in Australia. The results indicate that relationships exist between the

actual construction time/cost and other contract details as mentioned. This leads to the

development of several regression models for actual construction time and cost prediction.

Analyses were conducted to examine the sensitivity of the regression models considering the

likely inaccuracy in the estimated contract time and cost. The results indicate that the

regression models could assist clients and contractors to predict the actual construction time

5
and cost pertinent to a particular client sector, project type, contractor selection method and

contractual arrangement when estimated contract time and cost become available.

DATA

Useful data of recently completed construction projects were collected by postal

questionnaire survey. Bromilow et al (1988) argued that attempts to obtain too much

information would be counter-productive, and consequently only the essential variables in

which researchers (e.g. Bromilow, 1969; Bromilow and Henderson, 1976; Bromilow et al,

1980; Ireland, 1983; Laptali et al, 1991; Yeong, 1994; Mak, 1996; Walker, 1995; Chan,

1999; Ng et al, 2001) believed to have an influence on a project’s time-cost performance

were sought. The data requested included the company name, project name, project location,

client sector, project type, contractor selection method, contractual arrangement, original

contract period, actual contract period, original contract sum, and final contract sum.

The population was confined to projects having a contract value more than AUS$0.5 million

completed in the past eight years. Projects below AUS$0.5 million were considered to have

limited scope and complexity. For logistical reasons, the survey was limited to projects

completed between 1991 and 1998.

Construction companies from two biggest cities of New South Wales, Australia, i.e. Sydney

and Newcastle, were considered in this study. Names and addresses of 100 construction

companies were obtained by random selection from the telephone directories under the

classification of “Building Contractors”. Telephone interviews were conducted with the

6
companies and 44 indicated that they were interested in the study and could provide the

required data.

A survey package containing a covering letter, survey instructions, six separate sets of survey

questionnaires and stamped self-addressed envelopes was distributed to each company. The

companies were asked to provide the details of up to six projects for analysis. Due to the

sensitivity of the data required, 12 companies dropped out from the study at this stage. The

32 remaining companies provided 93 completed project surveys. This represents a

reasonable response rate of 35% (based on 264 project surveys distributed).

The average time for construction was 237 working days, the longest and shortest times being

864 and 60 working days respectively. All costs were rebased to March 1998 prices using

the Building Price Index (BPI) in the price book (Rawlinsons, 1998). The average rebased

cost of projects in the sample was AUS$21.4 million, the lowest and highest costs being

AUS$0.5 million and AUS$619 million respectively.

ANALYSIS

The first analysis aims to develop a model for forecasting actual construction time when

client sector, contractor selection method, contractual arrangements, project type, contract

period and contract sum are known. This model is more relevant to the post-contract stage as

the contract period, contract sum, and other contract details are readily available during this

stage. The model will assist the client and contractor predicting the actual construction time

required to complete a project.

7
Forward Crossvalidation Regression

In this analysis, the dependent variable is the actual construction time (ATIME) to practical

completion. Various forms of ATIME were considered, including log and power

transformations. The choice of dependent variable is very much restricted due to the

difficulties in comparing models with different forms of dependent variables. Bromilow

(1969) and others (e.g. Chan, 1999; Ng et al, 2000) have shown log construction time to be

the generally most appropriate variable to analyse. This (LATIME) was then used as the

dependent variable. The form of the independent variables is less of a problem and various

forms can be examined and compared for their contribution. Dummy variables may be

created for the categorical factors of client sector, project type, contractor selection method

and contractual arrangements but, with only 93 cases to analyse, this cannot be done in any

nonarbitrary way. The method of analysis eventually chosen was to concentrate solely on the

various forms of contract time (CTIME) and contract cost (CCOST) as independent

variables, with the intention of checking the residuals for any significant sub-group (within

client sector, project type, contractor selection method and contractual arrangements) effects.

The regression model was built by a forward crossvalidation procedure. Each independent

variable was tested separately for inclusion by the standard deviation of the deleted residual.

As a result, the log contract time (LCTIME), with a deleted residual standard deviation of

0.179, was selected for inclusion in the model. The next stage was to repeat the test for

inclusion of further form of cost and time variables (Table 1). No improvement to the deleted

residual standard deviation was found and therefore the procedure stopped.

8
 < Table 1 >

The deleted residuals of the final model were analysed for differences between the sub-

groups of client sector (SECTOR), contractor selection method (SELECTION), contractual

arrangements (CONTRACT) and project type (TYPE). Table 2 shows the results of the

ANOVA means tests and Levine variance homogeneity tests. The SELECTION and

CONTRACT groups produced significantly different means, with selective tendering

residuals being higher than ‘other’ forms of contractor selection (mainly negotiated) and

lump sum being higher than design and construct contractual arrangements. This indicated

that at least one variable was missing from the regression equation. Dummy variables were

therefore created for the SELECTION sub-groups of selective tendering and ‘other’ and the

CONTRACT sub-groups of lump sum and D&C and entered into a forward stepwise

crossvalidation regression on LATIME along with LCTIME (Table 1). This showed the best

model to comprise the independent variables LCTIME and the ‘other’ and ‘lump sum’

dummies. Analysis of the subgroups showed that the significant differences had disappeared

(Table 2).

< Table 2 >

Although the procedure is essentially nonparametric, it is of interest to check the Gauss

assumptions. The fitted model to all the data is:

LATIME = 0.207638 + 0.966737(LCTIME) + 0.097269(LS) – 0.083980(OT) … (1)

9
(R=0.96984455, R²=0.94059846, Adjusted R²=0.93859616, F(3,89)=469.76, p<0.0000, Std.

Error of Estimate=0.16382, see Table 3) where LATIME, LCTIME, LS and OT denote the

log-actual time, log-contract time, lump sum dummy variable and ‘other’ selection dummy

variable. D-W’s d=1.592, K-S d=0.104 (p>0.2) and the ANOVA on categorised LCOST is

F(3,89)=1.262 (p=0.292), suggesting the regression assumptions are not unduly violated.

< Table 3 >

‘Standard’ Crossvalidation Regression

To avoid the arbitrary nature of the forward crossvalidation regression, a ‘standard’

crossvalidation was conducted. This involves simultaneously entering all the independent

variables into the equation. The model used was:

y = α0jβ0+ α1jβ1x1j + α2jβ2x2j + α3jβ3x3j + ... + αnjβnxnj (β≥0) … (2)

where α0j , α1j, α2j, α3j , ..., αnj are the crossvalidation regression coefficients and x1j, x2j, x3j,

..., xnj are the values of the independent variables for the j=1, 2, ... , n cases (projects). The

β0, β1, β2, β3, ..., βn coefficients are computed to minimise the deleted residual sum of

squares (ssq) by means of a quasi-Newton algorithm. The α coefficients are then calculated

for all the (noncrossvalidation) data and the αβ products obtained. This method enables the

best crossvalidation model to be found, the β coefficients, once computed, being simply

multiplied by the α coefficients. Table 4 summarises the results (the columns ‘alpha’, ‘beta1’

and ‘ab1’ give the α , β and αβ coefficients and the last two rows give the residual ssq and

10
deleted residual ssq)2. This shows that the standard regression model (‘alpha’ column), as

expected produced a better residual ssq than the crossvalidation regression (‘ab1’ column)

with 2.11909 compared with 2.15984 but that the crossvalidation regression produced a better

deleted residual ssq than the standard regression with 2.43525 compared with 3.12966

(ssq=2.43525 = 0.162 deleted residual standard deviation – representing a marginal

improvement on the 0.168 of the forward method). As the ‘ab1’ coefficients show, the 16

independent variables plus the constant show seven to be redundant for crossvalidation

prediction, with zero β coefficients.

< Table 4 >

Actual Construction Time

The column ‘beta2’ in Table 4 gives the results when β is unconstrained and ‘ab2’ provide

the consequent αβ coefficients. The ssq results show that the residual ssq values to be worse

than the ‘ab1’ model but the deleted residual ssq value is better at 1.77817. The last two

models, ‘beta3’ and ‘beta4’, are obtained by setting all α coefficients to unity, with the

‘beta3’ model for β≥0 and the ‘beta4’ model for unconstrained β. The ‘beta3’ model, with

10 zero coefficients is close to the model obtained earlier by forward crossvalidation, whilst

the ‘beta4’ model has a reduced deleted residual ssq of 2.09846 against 2.37386. However,

the ‘ab2’ model is preferred because of its smallest deleted residual ssq of 1.77817.

Actual Construction Cost

2
The raw CCOST, CCOST2 and CTIME2 could not be handled by the procedure, presumably due to scaling problems.

11
Table 4 also gives the results of the various models when applied to forecasting the actual

construction cost (forward crossvalidation regression was not used). This shows a similar

pattern of results to the time forecasts, with the ‘alpha’ (ordinary regression) model having a

good residual ssq (1.08674) but relatively poor deleted residual ssq (1.77806). The

crossvalidation model ab1 again has a worse residual ssq (1.15303) and better deleted

residual ssq (1.02433). Again, the ‘ab2’ model produces the best forecasts, with a deleted

residual ssq of only 0.56923, against 1.17982 and 1.08395 for the ‘beta3’ and ‘beta4’ models

respectively, making the ‘ab2’ the preferred model once again.

SENSITIVITY ANALYSIS

All the models described to this point assume that the contract period and contract sum are

known. In practice, this is rarely the case, and contract period and contract sum are simply

estimated from whatever information is available at time of estimation. Therefore, since the

prediction of actual construction time and cost is based on the estimated contract period and

contract sum, it is necessary to examine how sensitive the prediction models are when the

contract period and contract sum deviate from their estimates.

Assume a contract sum of $10M and the other project particulars are public sector client,

residential building, open tendering and lump sum contract. The actual construction time for

a range of projects with an estimated contract period of 100 to 1000 days could be derived by

substituting all data into the prediction models ‘ab2’ for time. The actual construction time as

predicted by the model ranges from 122.51 days (CTIME=100) to 1056.93 days

(CTIME=1000) with zero percent estimate error (Table 5). This indicates that the actual

12
construction time is, on average, longer than the contract period by +22.51% for 100 day

contract period reducing to +5.693% for a 1000 day contract period.

< Table 5>

The contract period estimate error was altered at ±5% intervals to examine the effects on the

actual construction time prediction. In the prediction model, the contract period is expressed

in several different forms, including CTIME, LCTIME, RCTIME, and INVCTIME. The

deviations (ie. +5% of the contract period) were applied to all different forms of contract

period when the calculating the revised actual construction time. Figure 1 shows the results

of the sensitivity analysis for contract period. At ±5% deviation level, the variations to the

actual construction time were between ±3.5% (CTIME=1000) to ±5.2% and (CTIME=100)

(Table 5). These indicate that the ‘ab2’ model for time is more sensitive at 100 days than

1000 days contract period.

< Figure 1 >

Similarly, the actual construction costs for projects between $1M to $10M were calculated

using the ‘ab2’ model for cost. The actual costs derived from the model ranged from $0.92M

(CCOST=$1M) to $9.47M (CCOST=$10M) with a zero percentage error (Table 6).

< Table 6 >

Assuming a contract period of 100 days and the contract sum ranging from $1M to $10M, the

effects on the actual construction costs due to the deviations from the estimated contract sum

13
were examined by altering the contract sum at ±5% intervals. The deviations were reflected

in different forms of contract sum, including LCCOST, RCCOST, and INVCCOST. The

results of sensitivity analysis for ‘ab2’ model for cost is presented in Figure 2. The variations

of the ‘ab2’ model for cost are very constant throughout the $1M to $10M range. At ±5%

levels of deviation, the variations were approximately ±5% throughout the analysed contract

cost range.

< Figure 2 >

The sensitivity of other factors, including project type, contractor selection method and

contractual arrangement approach, was also examined. Figure 3 depicts the sensitivity of the

‘ab2’ model for time to different project types, assuming that open tendering and lump sum

contract are used. The results as also shown in Table 5 indicate that the average actual

construction time for industrial, educational and recreational projects will take much longer

than that of the residential type. The differences in actual construction time between the

residential building and each of the industrial, educational and recreational buildings were

+11.06%, +4.73% and +5.73% respectively.

< Figure 3 >

Figure 4 shows the sensitivity of the ‘ab2’ model for time to different contractor selection

methods and contractual arrangements, assuming a residential building. The results as shown

in Table 5 indicate that the actual construction time for negotiated tender and design and

construct contract is much less than open tender and lump sum contract. A saving of the

actual construction time of 16.67% can be achieved if negotiated tender instead of lump sum

14
tender is used. When a design and construct contract is used, a saving of 12.26% in the

actual construction time can be achieved over the lump sum contract (Table 5).

< Figure 4 >

PRACTICAL APPLICATIONS

With the ‘ab2’ models, contract administrators could predict the actual time and cost for a

given project type, contractor selection method and contractual arrangement based on the

estimated contract time and cost. However, since the risk attitudes of clients may vary, it is

necessary to be able to cater for different risk perceptions. The effects on actual time and

cost due to the changes in risk attitudes can be represented diagrammatically by the actual

time and cost curves.

Figures 5 and 6 illustrates the actual time and cost curves of a typical project with an

estimated construction period of 237 days and contract sum of AUS$21.4 million. The

differences in risk attitudes are presented as the percentage deviations from the estimated

contract time and cost (±20% on x-axis), and the effects to the predictions are shown in the

actual time and cost curves. The actual time and cost curves facilitate a client to predict the

appropriate actual time and cost of a project based on his risk attitude. For instance, a risk-

seeking client is optimistic with the actual construction time and cost, and the predictions

along the left of the x-axis (-5% to -20%) could be referred to. A risk-averse client is

pessimistic with the actual construction time and cost. This type of client could refer to the

predictions along the right of the x-axis (+5% to +20%).

15
 < Figure 5 >

< Figure 6 >

The actual time and cost curves for different types of project, contractor selection and

contractual arrangement are presented in Figures 5 and 6. These curves can assist the client

in selecting an appropriate contractor selection and contractual arrangement approaches

during the pre-contract stage. Once the project type is identified, a client can compare the

actual time and cost curves for different contractor selection method and contractual

arrangement to determine which is the best option for a particular project. The contractor

selection method and contractual arrangement that would result in the shortest actual time

and least actual cost are preferred. In addition, the effects of risks on different contractor

selection and contractual arrangement options could be compared through the actual time and

cost curves.

CONCLUSIONS

A set of 93 Australian construction projects was used to develop several models for actual

construction time and cost prediction. A forward crossvalidation regression analysis was

adopted for the development of the model for actual construction time forecast when client

sector, contractor selection method, contractual arrangement, project type, contract period

and contract sum are known. The standard deviation of the deleted residual revealed that the

best model for actual construction time prediction comprises the independent variables ‘log

contract time’ (LCTIME), ‘lump sum’ (LS) in the CONTRACT sub-group, and ‘other’ (OT)

in the SELECTION sub-group.

16
Regression models were also developed for forecasting the actual construction time and cost

when client sector, contractor selection method, contractual arrangement and project type are

known while contract period and contract sum are estimated. Different forms of regression

analyses, including the standard regression and the crossvalidation regression, were adopted.

However, the results of analyses show that the crossvalidation regression models ‘ab2’, with

the smallest deleted residual ssq of 1.77817 for time and 0.56923 for cost, are preferred.

Since the ‘ab2’ models for time and cost are dependent on the contract period and contract

sum being known, it is necessary to investigate the effects in situations where these have to

be estimated. The results of the sensitivity analyses on the ‘ab2’ models show that the errors

in predicted actual construction time become smaller as the contract period increases. In

contrast, the errors in predicted actual construction cost are virtually the same for large and

and small projects. The reliability of the actual construction time and cost prediction based

on ‘ab2’ models thus depends on the accuracy of the estimated contract period and contract

sum.

The effects of different project type, contractor selection method and contractual arrangement

were also examined. The results indicate that the actual construction time for industrial

project is the longest when compared with residential, educational and recreational projects.

A significant saving in actual construction time can be achieved when negotiated tender and

design and build contract are used instead of the traditional open tendering and lump sum

contract approaches.

17
Despite the ‘ab2’ models for time and cost being influenced by inaccurate contract period and

contract sum, the models provide practical tools for clients and contractors to predict the

actual construction time and cost based on the risks and uncertainties of different client

sector, contractor selection method, contractual arrangement and project type.

ACKNOWLEDGMENT

The authors would like to acknowledge the construction companies who participated in the

survey for providing the necessary project information to facilitate this research study, and

Michael Mak and Mark Varnam for collecting the project data.

18
REFERENCES

Alfred L.E. (1988) Construction Productivity – On Site Measurement and Management,

McGraw-Hill Book Company, Sydney, Australia.

Bromilow, F.J, (1969) Contract Time Performance Expectations and the Reality, Building

Forum, Vol. 1, No. 3, pp. 70-80.

Bromilow, F.J. and Henderson, J.A. (1976) Procedures for Reckoning and Valuing the

Performance of Building Contracts, The Chartered Builder, Vol. 10, No. 9, pp. 57.

Bromilow, F.J., Hinds M.F. and Moody, N.F. (1980) AIQS Survey of Building Contract

Time Performance, Building Economist, Vol. 19, No. 2, pp. 79-82.

Bromilow, F.J., Hinds M.F. and Moody, N.F. (1988) The Time and Cost Performance of

Building Contracts 1976-1986, The Australian Institute of Quantity Surveyors, Sydney,

Australia.

Chan, A.P.C. (1999) Modelling Building Durations in Hong Kong, Construction

Management and Economics, Vol. 17, No. 2, pp. 189-196.

Chan, D.W.M. and Kumaraswamy, M.M. (1995) A Study of the Factors Affecting

Construction Durations in Hong Kong, Construction Management and Economics, Vol. 13,

No. 4, pp. 319-333.

19
Ireland, V. (1983) The Role of Managerial Actions in the Cost, Time and Quality

Performance of High Rise Commercial Building Projects, Unpublished Thesis Submitted for

the Fulfilment of Degree of Doctor of Philosophy, University of Sydney, Australia.

Kaka, A.P. and Price, A.D.F. (1991) Relationship between value and duration of construction

projects, Construction Management and Economics, Vol. 9, No. 4, pp. 383-400.

Laptali E., Wild S. and Bouchlaghem N.M. (1996) Linear Programming for Time and Cost

Optimisation: Optimisation of Time and Cost, Proceedings: CIB W65 Commission on

Organisation and Management of Construction, 1996, D.A. Langford & A. Retik (eds),

E&FN SPON, London, pp. 124-137.

Mak, M.Y. (1991) The measurement of time performance in major building construction

projects, Unpublished Thesis Submitted for the Fulfilment of Degree of Master of Project

Management, University of New South Wales, Australia.

Ng, S.T., Mak, M.M.Y., Skitmore, R.M., Lam, K.C. & Varnam, M. (2001) The predictive

ability of Bromilow’s time-cost model, Construction Management and Economics, Vol. 19,

No. 2, pp. 165-173.

Rawlinsons (1998) Australian Construction Handbook (1998) – Quarterly Update,

Rawlhouse Publishing Pty. Ltd., Australia.

RIRA (1989) Pre-Assessing Using Contract/Time Related Formula, The Royal Australian

Institute of Architects, Australia.

20
Sidwell, A.C. (1984) The Time Performance of Construction Projects, Architectural Science

Review, Vol. 27, pp. 85-91.

Walker, D.H.T. (1994) An Investigation into Factors that Determine Building Construction

Time Performance, Unpublished Thesis Submitted for the Fulfilment of Doctor of

Philosophy, Royal Melbourne Institute of Technology, Australia.

Walker, D.H.T. (1995) An Investigation into Construction Time Performance, Construction

Management and Economics, Vol. 13, No. 3, pp. 263-274.

Yeong, C.M. (1994) Time and cost performance of building contracts in Australian and

Malaysia, Unpublished Thesis Submitted for the Fulfilment of MSc in Construction

Management, University of South Australia, Australia.

21
LIST OF CAPTIONS

Figure 1: Sensitivity analysis on contract time

Figure 2: Sensitivity analysis on contract cost

Figure 3: Sensitivity analysis on project type

Figure 4: Sensitivity analysis on contractor selection method & contractual arrangement

Figure 5: Actual construction time curves

Figure 6: Actual construction cost curves

Table 1: Results of forward crossvalidation procedure

Table 2: Deleted residuals by sub-groups

Table 3: Regression Summary for Dependent Variable – LTIME

Table 4: LATIME and LACOST forecasting model results (X years)

Table 5: Sensitivity of contract time, project type, contractor selection and contractual

arrangements

Table 6: Sensitivity of contract cost

22
 1200

T -20 (ROL)

1000 T -15 (ROL)

Actual Time (Day)

T -10 (ROL)
800 T -5 (ROL)

T 0 (ROL)
600
T +5 (ROL)

T +10 (ROL)
400
T +15 (ROL)

T +20 (ROL)
200

0
100 200 300 400 500 600 700 800 900 1000

Contract Time (Day)

(Based on a contract cost of AUS$10 million)

Note: T -20 = -20% of contract time, ROL = residential + open tendering + lump sum

Figure 1: Sensitivity analysis on contract time

23
 12

11

10 C -20 (ROL)

9 C -15 (ROL)
Actual Cost (AUS$)

8 C -10 (ROL)

7 C -5 (ROL)

C 0 (ROL)
6
C +5 (ROL)
5
C +10 (ROL)
4
C +15 (ROL)
3
C +20 (ROL)
2

0
1 2 3 4 5 6 7 8 9 10

Contract Cost (AUS$)

(Based on a contract time of 100 days)

Note: C -20 = -20% of contract cost, ROL = residential + open tendering + lump sum

Figure 2: Sensitivity analysis on contract cost

24
 1200
T -5 (ROL)

T 0 (ROL)
1000
T +5 (ROL)
Actual Time (Day)

800 T -5 (IOL)

T 0 (IOL)

600 T +5 (IOL)

T -5 (EOL)

400 T 0 (EOL)

T +5 (EOL)

200 T -5 (COL)

T 0 (COL)
0
T +5 (COL)
100 200 300 400 500 600 700 800 900 1000

Contract Time (Day)

(Based on a contract cost of AUS$10 million)

Note: ROL = residential + open tendering + lump sum, IOL = industrial + open tendering + lump sum,
EOL = educational + open tendering + lump sum, COL = recreational + open tendering + lump sum

Figure 3: Sensitivity analysis on project type

25
 1200

1000 T -5 (ROL)
Actual Time (Day)

T 0 (ROL)
800
T +5 (ROL)

T -5 (RNL)
600
T 0 (RNL)

T +5 (RNL)
400
T -5 (ROD)

T 0 (ROD)
200
T +5 (ROD)

0
100 200 300 400 500 600 700 800 900 1000

Contract Time (Day)

(Based on a contract cost of AUS$10 million)

Note: ROL = residential + open tendering + lump sum, RNL = residential + negotiated tender + lump sum,
ROD = residential + open tendering + design & contract

Figure 4: Sensitivity analysis on contractor selection method & contractual arrangement

26
 T-ROL
400
T-RNL

T-ROD

350 T-RND

T-IOL
Actual Time (Day)

T-INL
300
T-IOD

T-IND

250 T-EOL

T-ENL

T-EOD
200
T-END

T-COL

150 T-CNL
-20 -15 -10 -5 0 5 10 15 20
T-COD
Deviation From Contract Time (%) T-CND
(Contract time = 237 days, contract cost = AUS$21.4 million)

Note: T = time,
1st letter after hyphen represents project type: R = residential, I = industrial, E = educational, C = recreational
2nd letter after hyphen represents contractor selection method: O = open tendering, N = negotiated tender
3rd letter after hyphen represents contractual arrangement: L = lump sum, D = design & construct

Figure 5: Actual construction time curves

27
 C-ROL
26
C-RNL

C-ROD
24
Actual Cost (AUS$ million)

C-RND

C-IOL
22
C-INL

C-IOD
20
C-IND

C-EOL
18
C-ENL

C-EOD
16 C-END

C-COL

14 C-CNL
-20 -15 -10 -5 0 5 10 15 20
C-COD
Deviation From Contract Cost (%) C-CND
(Contract time = 237 days, contract cost = AUS$21.4 million)

Note: C = cost,
1st letter after hyphen represents project type: R = residential, I = industrial, E = educational, C = recreational
2nd letter after hyphen represents contractor selection method: O = open tendering, N = negotiated tender
3rd letter after hyphen represents contractual arrangement: L = lump sum, D = design & construct

Figure 6: Actual construction cost curves

28
 Standard deviation
Variable
Step 1 Step 2 Step 3 Step 4
CTIME 0.332 0.184
CTIME2 0.735 0.454
√CTIME 0.231 0.180
lnCTIME 0.179 -
1/CTIME 0.270 0.179
Model 1
CCOST 0.701 0.186
CCOST2 0.759 1.264
√CCOST 0.619 0.193
lnCCOST 0.448 0.179
1/CCOST 0.518 0.181
lnCTIME 0.179 - -
Select - 0.173 0.169 na
Other - 0.172 0.168 -
Model 2
Lump sum - 0.170 - -
D&C - 0.175 0.171 0.169
Table 1: Results of forward crossvalidation procedure

29
 Deleted residuals
Group Sub-group n ANOVA Levine’s
Mean Variance
(p) test (p)
public 31 0.022 0.031
Sector 0.404 0.830
private 62 -0.011 0.031
recreational 9 -0.048 0.032
industrial 26 0.010 0.030
Project type educational 15 0.001 0.722 0.018 0.651
residential 11 0.058 0.039
other 32 -0.015 0.039
Model 1 selective 59 0.035 0.035
Contractor
open 15 -0.010 0.012 0.021 0.064
selection
other 19 -0.102 0.018
lump sum 61 0.043 0.032
Contractual
D&C 16 -0.083 0.005 0.011 0.201
arrangements
other 16 -0.081 0.036
Total 93 0.000 0.032

Sector public 31 -0.004 0.872 0.030 0.591

private 62 0.002 0.028
recreational 9 -0.007 0.039
industrial 26 0.022 0.027
Project type educational 15 -0.024 0.900 0.017 0.669
residential 11 0.020 0.032
other 32 -0.012 0.033
Model 2 selective 59 0.005 0.033
Contractor
open 15 -0.023 0.852 0.025 0.174
selection
other 19 0.000 0.018
lump sum 61 -0.001 0.031
Contractual
D&C 16 -0.004 0.991 0.013 0.352
arrangements
other 16 0.004 0.035
Total 93 0.000 0.028
Table 2: Deleted residuals by sub-groups

30
 Std. Err. Std. Err.
BETA B t(89) p-level
of BETA of B
Intercept .207638 .144462 1.43732 .154133
LCTIME .970507 .026231 .966737 .026129 36.99849 0.000000
Lc .070275 .028096 .097269 .038889 2.50122 .014206
OT -.051494 .027914 -.083980 .045524 -1.84476 068399

Note: R=0.96984455 R²=0.94059846 Adjusted R²=0.93859616 F(3,89)=469.76 p<0.0000

Std. error of estimate=0.16382

Table 3: Regression Summary for Dependent Variable – LATIME

31
 i Label alpha beta1 ab1 beta2 ab2 beta3 beta4
1 Constant -1.25943 0.85662 -1.07885 1.26350 -1.59130 0.06807 -11.44751
2 CTIME 0.00106 0.30771 0.00033 -0.00067 0.00000 0.00000 0.00752
3 LCTIME 1.48502 0.93565 1.38947 0.95108 1.41237 0.98254 4.50414
4 RCTIME -0.09797 0.67958 -0.06658 0.45174 -0.04426 0.00000 -0.64119
5 INVCTIME -0.18694 0.00000 0.00000 -83.29847 15.57190 0.00000 92.78079
6 LCCOST -0.00275 0.89772 -0.00247 1.05852 -0.00291 0.00000 0.00652
7 RCCOST -0.00001 0.00000 0.00000 0.16468 0.00000 0.00000 -0.00001
LATIME results

8 INVCCOST -0.00004 1.00001 -0.00004 1.00003 -0.00004 0.00000 -209.57518

9 SECTOR -0.03344 0.00000 0.00000 -2.12731 0.07113 0.00000 -0.02638
10 REC 0.00325 0.00000 0.00000 -7.27903 -0.02364 0.00000 0.00855
11 IND 0.07385 0.59430 0.04389 0.34626 0.02557 0.03747 0.07104
12 EDUC -0.05669 0.53680 -0.03043 0.58582 -0.03321 0.00000 -0.04831
13 RESID 0.01259 0.00000 0.00000 -6.30386 -0.07936 0.01927 0.00871
14 OPEN -0.00986 0.00000 0.00000 -6.67807 0.06582 0.00823 -0.00826
15 NEG -0.10604 1.04052 -0.11033 1.11113 -0.11782 0.00000 -0.10949
16 LUMP 0.11396 0.89200 0.10165 1.15205 0.13128 0.14686 0.10860
17 D&C 0.00956 0.00000 0.00000 0.04803 0.00046 0.01991 0.00523
resid ssq 2.11909 2.15984 2.37572
dresid ssq 3.12966 2.43525 1.77817 2.37386 2.09846
0 Constant 0.13130 0.00000 0.00000 1.00432 0.13187 0.14177 3.84230
1 CTIME -0.00031 0.00000 0.00000 -1.63171 0.00051 0.00000 -0.00267
2 LCTIME 0.04819 0.58187 0.02804 -0.19688 -0.00949 0.08086 -1.05199
3 RCTIME 0.01609 0.45427 0.00731 -1.19396 -0.01921 0.00000 0.21409
4 INVCTIME 0.01284 244.04398 3.13318 -394.38433 -5.06333 0.00000 -33.86345
5 LCCOST 0.97442 1.01515 0.98919 1.03972 1.01312 0.96680 0.97112
6 RCCOST -0.00001 1.20251 -0.00001 1.23071 -0.00001 0.00000 -0.00001
LACOST results

7 INVCCOST 0.00000 1.00000 0.00000 1.00000 0.00000 0.00000 236.25736

8 SECTOR -0.03424 0.00000 0.00000 -0.19823 0.00679 0.00000 -0.03680
9 REC -0.05940 0.87335 -0.05188 0.34267 -0.02036 0.00000 -0.06142
10 IND 0.00655 0.00000 0.00000 -0.99174 -0.00649 0.01986 0.00755
11 EDUC -0.07906 0.35664 -0.02820 0.07325 -0.00579 0.00000 -0.08208
12 RESID -0.00394 0.00000 0.00000 -6.62376 0.02609 0.00930 -0.00258
13 OPEN 0.01746 0.00000 0.00000 -2.09804 -0.03663 0.02385 0.01688
14 NEG -0.02534 0.24503 -0.00621 1.88626 -0.04781 0.00000 -0.02405
15 LUMP -0.02753 0.00000 0.00000 -0.81953 0.02256 0.00000 -0.02555
16 D&C -0.00221 0.00000 0.00000 -17.40347 0.03842 0.00809 -0.00055
resid ssq 1.08674 1.15303 1.29445
dresid ssq 1.77806 1.02433 0.56923 1.17982 1.08395

Table 4: LATIME and LACOST forecasting model results (X years)

32
 Contract time (days)
100 200 300 400 500 600 700 800 900 1000
-20% 97.39 199.61 302.10 401.33 496.09 585.98 670.94 751.06 826.51 897.50
-15% 103.61 212.52 321.01 425.46 524.71 618.42 706.58 789.35 866.98 939.70
-10% 109.88 225.41 339.79 449.31 552.85 650.16 741.30 826.51 906.07 980.31
-5% 116.18 238.29 358.42 472.85 580.51 681.22 775.13 862.55 943.84 1019.36
Accuracy 0% 122.51 251.13 376.91 496.09 607.68 711.59 808.07 897.50 980.31 1056.93
+5% 128.86 263.94 395.25 519.03 634.37 741.30 840.15 931.39 1015.53 1093.05
+10% 135.24 276.71 413.43 541.65 660.59 770.35 871.39 964.25 1049.53 1127.78
+15% 141.63 289.43 431.45 563.97 686.32 798.75 901.79 996.11 1082.36 1161.16
+20% 148.04 302.10 449.31 585.98 711.59 826.51 931.39 1026.99 1114.05 1193.26
Residential 122.51 251.13 376.91 496.09 607.68 711.59 808.07 897.50 980.31 1056.93
Industrial 136.06 278.92 418.61 550.98 674.91 790.32 897.47 996.79 1088.76 1173.86
Project type
Educational 128.30 262.99 394.72 519.52 636.38 745.20 846.24 939.89 1026.61 1106.85
Recreational 129.53 265.52 398.51 524.52 642.50 752.37 854.38 948.93 1036.48 1117.49
Contractor Open 122.51 251.13 376.91 496.09 607.68 711.59 808.07 897.50 980.31 1056.93
selection Negotiate 101.96 209.00 313.68 412.86 505.73 592.21 672.51 746.93 815.85 879.61
Lump sum 122.51 251.13 376.91 496.09 607.68 711.59 808.07 897.50 980.31 1056.93
Contractual
Design &
arrangement 107.49 220.34 330.70 435.26 533.17 624.33 708.98 787.45 860.10 927.32
build
Note: Base Contract cost = $10M, project type = residential, contractor selection method = open, contractual arrangement
= lump sum

Table 5: Sensitivity of contract time, project type, contractor selection and contractual
arrangements

33
 Contract cost ($ million)
%
1 2 3 4 5 6 7 8 9 10
-20 0.73 1.48 2.23 2.99 3.74 4.50 5.26 6.02 6.79 7.55
-15 0.78 1.57 2.37 3.17 3.98 4.79 5.60 6.41 7.22 8.03
-10 0.83 1.67 2.51 3.36 4.22 5.07 5.93 6.79 7.65 8.51
-5 0.87 1.76 2.65 3.55 4.45 5.36 6.26 7.17 8.08 8.99
0 0.92 1.85 2.80 3.74 4.69 5.64 6.60 7.55 8.51 9.47
+5 0.97 1.95 2.94 3.93 4.93 5.93 6.93 7.94 8.94 9.95
+10 1.01 2.04 3.08 4.12 5.17 6.22 7.27 8.32 9.37 10.43
+15 1.06 2.14 3.22 4.31 5.41 6.50 7.60 8.70 9.80 10.91
+20 1.11 2.23 3.36 4.50 5.64 6.79 7.94 9.09 10.24 11.39

Note: Contract time = 100 days, project type = residential, contractor selection method = open, contractual arrangement =
lump sum

Table 6: Sensitivity of contract cost

34