Int. J. Oil, Gas and Coal Technology, Vol. 5, No.

1, 2012

29

An analysis of inaccuracy in pipeline construction cost estimation Zhenhua Rui*, Paul A. Metz and Gang Chen
Department of Mining and Geological Engineering, University of Alaska Fairbanks, Duckering Building 418, P.O. Box 750708, Fairbanks, Alaska, 99775, USA Fax: +1-907-474-6635 E-mail: zhenhuarui@gmail.com E-mail: pametz@alaska.edu E-mail: gchen@alaska.edu *Corresponding author
Abstract: The aim of this paper is to investigate cost overrun of pipeline projects. A total of 412 pipeline projects between 1992 and 2008 have been collected, including material cost, labour cost, miscellaneous cost, right of way (ROW) cost, total cost, pipeline diameter, pipeline length, pipelines location, and year of completion. Statistical methods are used to identify the distribution of the cost overrun and the causes for overruns. The overall average cost overrun rates of pipeline material, labour, miscellaneous, ROW and total costs are 4.9%, 22.4%, 0.9%, 9.1% and 6.5% respectively. The cost estimation of pipeline cost components are biased except for total cost. In addition, the cost error of underestimated pipeline construction components is generally larger than that of overestimated pipeline construction components except total cost. Results of analysis show that pipeline size, capacity, diameter, length, location, and year of completion have different impacts on cost overrun of construction cost components. [Received: May 26, 2011; Accepted: June 28, 2011] Keywords: pipeline cost; cost overrun; cost estimation. Reference to this paper should be made as follows: Rui, Z., Metz, P.A. and Chen, G. (2012) An analysis of inaccuracy in pipeline construction cost estimation, Int. J. Oil, Gas and Coal Technology, Vol. 5, No. 1, pp.2946. Biographical notes: Zhenhua Rui is a PhD candidate in Energy Engineering Management and MBA student at the University of Alaska Fairbanks. He also received his Masters degree in Petroleum Engineering from the same university, in addition to his Masters degree in Geophysics from China University of Petroleum, Beijing. His current research is the Engineering Economics of the Alaska In-state Natural Gas Pipeline. Paul A. Metz is a Professor of Department of Mining and Geological Engineering at the University of Alaska Fairbanks. He received his PhD from Imperial College of Science Technology and Medicine. He also received his MS in Economic Geology and MBA from the University of Alaska. His research interest include: market and transportation analysis of mineral resources; analysis of transport systems; engineering geological mapping and site investigation; mineral and energy resource evaluation.

Copyright 2012 Inderscience Enterprises Ltd.

30

Z. Rui et al.
Gang Chen is a Professor of Department of Mining and Geological Engineering at the University of Alaska Fairbanks. He received PhD in Mining Engineering from Virginia Polytechnic Institute and State University. He also received his MS in Mining Engineering from the Colorado School of Mines. His research interest include: rock mechanics in mining and civil engineering; mine ground engineering; frozen ground engineering and GIS application in mining industry.

Introduction

Cost error is the tendency for actual costs to deviate from estimated cost. Bias is the tendency for that error to have a non-zero mean (Bertisen and Davis, 2008). Cost error or bias is common and a global phenomenon in cost estimating (Flyvbjerg et al., 2003). Cost estimating error and bias in other types of projects were mentioned and studied in many papers. Pohl and Mihaljek (1992) reviewed 1,015 World Bank projects from 1947 to 1987 with 22% average cost overrun and 50% time overrun. Merrow (1998) found that 47 of 52 megaprojects have an average overrun of 88%, and large projects and megaprojects appear to have more cost growth than smaller projects. Flyvbjerg et al. (2003) examined 258 transport infrastructure projects (rail, bridge and road) with average 28% cost overrun. Bertisen and Davis (2008) reviewed 63 international mining projects with an average of 14% higher than estimated cost in the feasibility study. The literature reviewed also shows that cost overrun exists over time. Overall cost overrun rates of all Indiana departments of transportation projects was 4.5%, and 55% of all projects experienced cost overruns (Bordat et al., 2004). Jacoby (2001) found that 74 projects with a minimum cost of $10 million had 25% cost overruns. Many researchers also try to explain the project cost overrun phenomenon. Some researchers proposed that optimism and deception are major reasons for causing cost overrun (Flyvbjerg et al., 2003). Some researchers believe that engineers and managers have incentive to underestimate cost (Bertisen and Davis, 2008). Flyvbjerg (2007) suggested that cost underestimation and overestimation of transport infrastructure appear to be intentional by project promoters. Information asymmetries were also suggested as a reason for cost overrun (Pindyck and Rubinfeld, 1995). Rowland (1981) mentioned that large projects increase the likelihood of a high number of change orders. Jahren and Ashe (1990) suggested that large projects have large cost overruns due to complexity, but also mentioned that managers of large projects try to keep cost overrun rates from growing excessively large. Large projects can lead to savings in unit cost, but it will limit the number of companies who are able to carry out projects. Therefore, there is a trade-off between economies of scale and competitive bidding practices (Bordat et al., 2004). Odeck (2004) indicated that large projects have better management than small projects. Soil, drainage, climate and weather conditions have an impact on design standard and cost of materials for road and rail projects, and location influences construction and material cost due to varying distance from supplies (RGL Forensics, 2009). An Australian study shows that publicprivate partnership projects perform better than traditionally procured projects, While a European study shows public-private partnerships exhibit higher costs than traditionally procured infrastructure (Infrastructure Partnerships Australia, 2008; RGL Forensics, 2009). Flyvbjerg (2007) suggested that more research on the role of ownership in causing

An analysis of inaccuracy in pipeline construction cost estimation

31

efficiency difference between projects needs to be conducted. He also used technical, psychological and political-economic factors to explain cost overruns. Although many studies have been conducted on projects cost overrun, there are limited available references about pipeline project cost overrun. With available pipeline data, this paper will focus on the cost estimation error of pipeline construction components, and will investigate and identify the frequency of occurrence of cost overruns as well as the magnitude of the differences between estimated cost and actual cost in pipeline projects. In addition, cost overrun in terms of pipeline project size, pipeline capacity, diameter, length, location, and year of completion are also investigated.

Data source

In this study, the pipelines are selected on the basis of data availability. The Oil and Gas Journal pipeline cost data are collected from Federal Energy Regulatory Commission filings from gas transmission companies, which are published by the Oil and Gas Journal annual data book (Penn Well Corp, 19922009). Due to limited offshore pipeline data, the pipeline data set in this paper contains only onshore pipeline data, and the pipeline cost in this paper does not include compressor station cost. The pipeline data set provides location and year, pipeline diameter and length. Pipelines in the data set were distributed in all states in the USA except Alaska and Hawaii. It also contains the cost information of 15 Canadian pipelines. The pipelines were completed between 1992 and 2008. Unfortunately, the data does not show the construction period of the pipelines. Therefore, cost is defined as real, accounted costs determined at the time of completion. The entire data set has 412 onshore pipelines. The data include estimated and actual cost of five cost components. The estimated costs are defined as budget, or forecast, costs at the time of decision to build the pipeline. The actual costs are defined as real accounted costs determined at the time of completing pipelines (Flyvbjerg et al., 2003). The five cost components are material, labour, miscellaneous, right of way (ROW) and total costs. Material cost covers cost of line pipe, pipeline coating and cathodic protection. Labour costs consist of the cost of pipeline construction labour. Miscellaneous cost is a composite of the costs of surveying, engineering, supervision, contingencies, telecommunications equipment, freight, taxes, allowances for funds used during construction, administration and overheads, and regulatory filing fees. ROW cost contains the cost of ROW and allowance for damages. The total cost is the sum of material cost, labour cost, miscellaneous cost and ROW cost (Penn Well Corp, 19922009). The location information for U.S pipelines was provided in a state format. A total of 48 states were referred to, except for Alaska and Hawaii. The US Energy Information Administration (EIA) breaks down the US Natural Gas pipelines network into six regions: Northeast, Southeast, Midwest, Southwest, Central and Western. The state grouping is defined based on 10 federal regions of the US Bureau of Labour Statistics (EIA, 2010). The map of regional definitions is shown in Figure 1. These regional definitions are used to analyse geographic difference. In this paper, the USA pipeline data are divided in to six regions according to the EIA definition. In addition, there are 15 Canadian pipelines, but reports did not show a specific province in Canada.

32
Figure 1

Z. Rui et al.
US natural gas pipeline region map (EIA) (see online version for colours)

Note: Alaska and Hawaii are not included in the US Natural pipeline network map.

In order to take a comparative analysis, all costs are adjusted by the Chemical Engineering Plant Cost Index to 2008 dollars. The Chemical Engineering Plant Cost Index is widely applied on process plants for adjusting construction cost. The Chemical Engineering Plant Cost Index has 11 sub-indexes and a composite index, which is the weight average of the 11 sub-indexes (Chemical Engineering, 2011). Pipeline index and construction labour index is used to adjust pipeline material and labour cost. The Chemical Engineering Plant Index is applied to pipeline miscellaneous and ROW cost.

Performance of individual pipeline construction component cost estimation

This section will evaluate performance of pipeline construction component cost estimation. Several methods may be used to study the difference between the estimated cost and the actual cost. In this study, the estimated cost and the actual cost are used to calculate cost overrun rate as a measurement of cost overrun. The formula for cost overrun rate is:
Cost overrun rate = (actual cost - estimated cost) / estimated cost

If the cost overrun rate is positive, the cost is underestimated, otherwise it is overestimated. In this paper, all cost overrun rates are calculated with the above formula. The histogram of the cost overrun rate for pipeline construction components are shown in Figure 2 to Figure 6. If the cost error is small, the histogram would be narrowly concentrated around zero. If underestimated cost is as common as overestimated cost, the histogram would be symmetrically distributed around zero. It appears that five figures exhibited non-symmetric distributions, and none of them satisfied the above mentioned

An analysis of inaccuracy in pipeline construction cost estimation

33

assumption. For material cost, 172 (42.0% of total) pipelines were underestimated, and 238 (58.0% of total) were overestimated. For labour cost, 273 (66.7% of total) pipelines were underestimated, and 136 (33.3% of total) were overestimated. For miscellaneous cost, 166 (40.8% of total) pipelines were underestimated, and 241 (59.2% of total) were overestimated. For ROW cost, 174 (45.7% of total) pipelines were underestimated, and 207 (54.3% of total) were overestimated. For total cost, 222 (54.0% of total) pipelines were underestimated, and 189 (46.0% of total) were overestimated.
Figure 2 Cost overrun rates of material cost (see online version for colours)

Figure 3

Cost overrun rates of labour cost (see online version for colours)

34
Figure 4

Z. Rui et al.
Cost overrun rates of miscellaneous cost (see online version for colours)

Figure 5

Cost overrun rates of ROW cost (see online version for colours)

In summary, more pipelines were overestimated for material, miscellaneous and ROW costs, while more pipelines were underestimated for labour and total cost. In general, the percentage of overestimated pipelines implies that there are still a fairly good number of pipelines being completed with costs less than the estimated cost. In addition, the majority of pipelines (87.1% for material cost, 72.3% for labour cost, 67.3% for miscellaneous cost, and 89% for total cost) have cost overrun rates between 0.4 and 0.4. However, only 49.0% of the pipelines for ROW cost have cost overrun rates between

An analysis of inaccuracy in pipeline construction cost estimation

35

0.4 and 0.4. It demonstrates that ROW cost overrun is more severe than cost components, which is also indicated by its standard deviation (SD) (Table 1).
Figure 6 Cost overrun rates of total cost (see online version for colours)

Table 1

Summaries of cost overrun of pipeline construction cost components Material Labour 4.83 44.88 0.94 7.04 7.98 0.22 0.62 409 273 136 Miscellaneous 4.77 42.06 0.94 4.56 5.50 0.01 0.56 407 166 241 ROW 3.25 15.77 1.00 4.55 5.55 0.09 0.81 381 174 207 Total 2.20 12.29 0.94 2.12 3.06 0.07 0.34 411 222 189

Skewness Kurtosis Minimum Maximum Range Average Standard deviation Total number of pipelines Number of underestimated pipelines Number of overestimated pipelines

5.77 49.22 0.95 5.67 6.61 0.05 0.55 410 172 238

Furthermore, statistical summaries of cost overrun of individual pipeline construction components are shown in Table 1. Skewness is a quantitative way to measure symmetry of the distribution. Symmetrical distribution has a skewness of 0. Positive skewness means that the right tail is heavier than the left tail. Negative skewness means that the left tail dominates distribution. Kurtosis is a quantitative method to evaluate whether the shape of the data distribution fits the normal distribution. A normal distribution has a kurtosis of 0. Kurtosis of a flatter distribution is negative and that of a more peaked distribution is positive (Hill et al., 2007). Values of skewness and kurtosis in Table 1 show that none of the cost overrun of five components is symmetrical normal distribution, which matches the implication from the histogram graphs. Some

36

Z. Rui et al.

transformation techniques (such as natural log transformation) are applied to cost overrun rate data for fitting them to normal distribution, but the data transformations are unsuccessful. Therefore, the non-parametric statistical test is used in the below sections. P-value will be produced by each test. In statistics, traditionally, p < 0.01 is considered highly significant, p < 0.05 is significant. This criterion is also adopted in this paper. Table 1 shows that the minimum cost overrun rate for individual cost components are between 94% (labour cost) and 100% (ROW cost). The maximum cost overrun rate for individual cost components are between 212% and 704%. The value of minimum and maximum indicates that cost performance for some pipelines is extremely bad. The labour cost overrun has the largest maximum-minimum range of 798%, while total cost overrun has the smallest range of 306%. The SD of individual cost components are fairly significant, between 34% and 81% of the estimated cost. The large maximum-minimum range and SD indicate that performance of pipeline construction cost estimating is very unstable. It is noteworthy that labour cost has the largest maximum-minimum range and second largest SD, and ROW cost has the largest SD. Therefore, it is difficult to estimate labour cost and ROW cost accurately. Total cost overrun has the smallest maximum-minimum range and SD due to its aggregation of other components. Average cost overrun is a key parameter to measure cost estimation performance of individual pipeline construction cost components. The labour cost has the highest average cost overrun of 22%, followed by ROW cost of 9%, total cost of 7%, material cost of 5% and miscellaneous cost of 1%. The material, labour, ROW and total costs show positive average cost overrun, while the miscellaneous cost has negative average cost overrun. This result denotes that, in average, actual cost is larger than estimated cost for all pipeline construction cost components except the miscellaneous cost. As mentioned before, there are more pipelines with overestimated material, miscellaneous and ROW costs than those with underestimated pipelines, and there are more pipelines with underestimated labour and total costs than those with overestimation of these two cost components. However, it is an interesting finding that the average cost overruns of material and ROW cost are still positive, even though there are more pipelines with overestimated material cost, and ROW cost. It appears that cost estimation of pipeline construction cost components is biased, and the underestimating error is generally greater than the overestimating error for some pipeline construction cost components. In this paper, two statistical tests are performed to investigate this inference. A binomial test is conducted to examine if the error of cost overestimating is as common as the error of cost underestimating. As shown in Table 2, the p-value of the binomial test rejects the null hypothesis that the overestimating error is as common as the underestimating error for material, labour, miscellaneous, and ROW cost estimation (p < 0.05, two sided test), but fails to reject that for total cost (p > 0.05, two sided test). Therefore, the cost estimations of all pipeline construction cost components are biased except total cost. Furthermore, the non-parametric Mann-Whitney test is employed to test if the cost underestimating error is the same as the cost overestimating error. The p-value shown in Table 2 shows that the error of underestimated pipelines cost overrun are much larger than the error of overestimated pipelines cost overrun for material, labour, miscellaneous and ROW cost (p < 5%, one sided test), but not for total cost (p > 5%, two sided test). Hence, the underestimating error is significantly more common and larger than the overestimating error for all pipeline cost components, but not for total cost.

An analysis of inaccuracy in pipeline construction cost estimation

Table 2 Statistical tests of cost overrun of pipeline construction cost components Material Binomial test Mann-Whitney test 0.001 0.047 Labour 0 0 Miscellaneous 0 0 ROW 0 0.039

37

Total 0.114 0.082

After analysing overall cost overrun of pipeline projects, it is more important to identify significant factors that influence pipeline projects cost overruns. The analyses of cost overruns in terms of pipeline project size, pipeline capacity, diameter, length, location, and completion time are carried out in the following sections.

Cost overrun in terms of pipeline project size

Here, the project size is measured by the pipelines actual total cost. The pipeline total cost ranges from $33,576 to $1,933,839,076. According to the pipeline actual total cost, the pipeline project size is classified into groups of small, medium and large. 185 pipelines with a total actual cost less than $10,000,000 are classified as small projects, 192 pipelines with a total actual cost between $10,000,000 and $100,000,000 are classified as medium projects, and 33 pipelines with a total actual cost larger than $100,000,000 are classified as large projects.
Table 3 Average cost overrun rate for different project size groups Project size Small Medium Large Small Medium Large Small Medium Large Small Medium Large Small Medium Large Average 0.10 0.01 0.04 0.16 0.28 0.13 0.01 0.01 -0.04 0.18 0.30 0.23 0.04 0.08 0.12 SD 0.70 0.42 0.13 0.47 0.76 0.41 0.46 0.46 0.46 1.20 1.39 0.54 0.36 0.32 0.24 Skewness 4.55 7.34 1.22 1.31 5.19 -0.50 1.08 0.98 1.08 2.87 3.28 1.60 1.89 2.70 2.50 Kurtosis 30.86 80.95 4.19 6.95 40.15 4.03 5.68 4.01 5.68 13.30 15.00 7.12 10.04 15.24 11.58

Components Material

Labour

Miscellaneous

ROW

Total

Descriptive statistical analysis of cost overrun in terms of project size is shown in Table 3. As seen in Table 3, for the material, labour, miscellaneous cost and ROW cost, there is no linear relationship between average cost overrun rates and project sizes. However, for total cost, the average cost overrun rate increases as the projects size increases. For the total cost, large projects have the highest cost overrun rates. A plausible explanation is that a large pipeline project, normally bigger than 1 billion dollars, can cause a huge demand that influences market price, such as steel price, and

38

Z. Rui et al.

further increases the cost of pipeline construction. Expectation of increased pipeline construction costs induced an increase in the current unit construction cost (Rui et al., 2011a). Suppliers would raise prices with expectation for more demand. In addition, a large project limits the numbers of suppliers and contractors, reduces competition and increases the cost (Bordat et al., 2004; RGL Forensics, 2009). However, for the miscellaneous cost, large projects have the lowest cost overrun. It is possible that larger projects have better management systems which coordinate different departments, increase the efficiency of material utilisation and take advantage of economies of scale. In order to determine if there is a strong relationship between project sizes and cost overrun for different pipeline construction components, the non-parametric Kruskal-Wallis (KW) test is used to test the null hypothesis that the project size has no effect on cost overrun of pipelines. The KW test is chosen because the value of skewness and kurtosis shows that the cost overrun of each diameter group is not a normal distribution. Therefore, the KW test will be used in this part and following parts when the data does not produce normal distributions. For total cost, the result of the KW test shows that cost overrun for different project size groups are significantly different (p < 0.05). However, such a significant difference is not found for other cost components (p > 0.05). Therefore, it is concluded that the project size significantly influences cost overrun for total cost, but not for other individual cost components.
Table 4 Average cost overrun rate for different diameter groups Diameter groups 420 inches 2230 inches 3448 inches 420 inches 2230 inches 3448 inches 420 inches 2230 inches 3448 inches 420 inches 2230 inches 3448 inches 420 inches 2230 inches 3448 inches Average 0.13 0.03 0.00 0.39 0.21 0.09 0.17 0.16 0.02 0.43 0.24 0.11 0.17 0.03 0.02 SD 0.68 0.42 0.52 0.95 0.42 0.33 0.99 0.35 0.48 1.57 1.38 0.81 0.48 0.24 0.23 Skewness 3.70 5.62 8.56 3.85 1.14 0.87 4.11 0.93 0.92 2.61 3.31 2.71 1.72 0.65 1.39 Kurtosis 21.37 51.28 92.67 23.97 5.79 7.18 24.64 4.39 4.38 10.01 15.53 15.52 6.96 6.55 9.59 Num. of pipelines 124 131 155 126 131 155 123 131 152 115 122 153 124 131 155

Components Material

Labour

Miscellaneous

ROW

Total

Cost overrun in terms of pipeline diameter

The range in pipeline diameter is between 4 and 48 inches. The pipeline projects are categorised into three diameter groups: 420 inch, 2230 inches and 3448 inches. Pipeline construction component cost overruns for three different pipeline diameter groups are shown in Table 4.

An analysis of inaccuracy in pipeline construction cost estimation

39

For material, labour, ROW and total costs, 420 inch pipelines have the highest average cost overrun rate, followed by 2230 inches pipelines and 3448 inches pipelines. For miscellaneous cost, 420 inches pipelines have the highest average cost overrun, but 2230 inches pipelines have the lowest average cost overrun of 16%. 420 inches groups have the highest average cost overrun rate for all construction components costs. It appears that small diameter pipelines are prone to cost overrun. Therefore, the non-parametric KW test is used to test the null hypothesis that type of pipeline diameter has no effect on cost overrun of pipelines construction component cost. For material, ROW and total cost, the result of the KW test shows the cost overrun is not significantly different for different diameter groups (p > 0.5). For labour and miscellaneous costs, the result of the KW test shows the cost overrun for types of diameter groups is significantly different (p < 0.01). Therefore, it is concluded that types of diameter groups influence cost overrun of pipelines for labour and miscellaneous cost, and not for other components costs.

Cost overrun in terms of pipeline length

The cost overrun in terms of pipeline length is tested in this section. The range of pipeline length is between 0.1 and 713 miles. The length group is divided into two groups: 020 miles and 20713 miles. About 78% of pipelines are shorter than 20 miles, and the rest of the pipelines are between 20 and 713 miles. Pipeline construction component cost overruns for two different pipeline length groups are shown in Table 5.
Table 5 Average cost overrun rate for different length groups Length groups 020 miles 20713 miles 020 miles 20713 miles 020 miles 20713 miles 020 miles 20713 miles 020 miles 20713 miles Average 0.05 0.04 0.21 0.26 0.17 0.03 0.23 0.30 0.06 0.10 SD 0.56 0.51 0.60 0.70 0.72 0.40 1.28 1.18 0.35 0.30 Skewness 5.25 8.14 5.37 3.46 4.71 0.82 3.11 3.89 2.12 2.80 Kurtosis 44.21 73.36 56.39 20.41 38.60 4.74 14.53 21.62 11.85 14.57 Num. of pipelines 321 89 323 89 319 87 303 87 321 89

Components Material Labour Miscellaneous ROW Total

For material, miscellaneous and ROW costs, the 020 miles group has the highest average cost overrun rate, followed by the 20713 miles inches pipelines. For labour cost and total cost, the 20713 miles pipelines have incurred the highest average cost overrun, followed by 020 miles pipelines. It appears that different construction component costs have different cost overrun rate patterns. The KW test is used to test the null hypothesis that type of pipeline length has no effect on cost overrun. For all construction cost components, the results of the KW tests show the cost overrun rate difference between types of length groups are not significant at the 5% significance level (p > 0.1). Therefore, cost overrun rates of all construction cost components are not significantly influenced by types of pipeline length.

40

Z. Rui et al.

Cost overrun in terms of pipeline capacity

In this paper, the pipeline volume (capacity) is calculated with formula (Zhao, 2000):

V = S *L
D where S = , V is the pipeline volume (ft3), S is the pipeline cross-sectional area 2 (ft2), L is the pipeline length (ft), and D is the pipeline diameter (ft). In the data set for this study, the smallest pipeline capacity is 92 ft3, and the largest is 36,220,080 ft3. In this section, all pipelines are divided into three different groups of pipeline capacities to test whether cost overrun rate is significantly different for different pipeline capacity. 135 pipelines with a capacity between less than 75,000 ft3 are classified as small projects, 136 pipelines with a capacity between 75, 000 ft3 and 284,768 ft3 are classified as medium projects, and 139 pipelines with a capacity larger than 284,768 are classified as large projects. Descriptive statistical analysis of cost overrun in terms of pipeline capacity is shown in Table 6. A noticeable observation is that the small pipeline capacity group has the highest average cost overrun rates for all construction cost components. Pipelines with small capacity appear to be particularly prone to cost overrun. Projects with large capacity may take more advantage of economies of scale.
Table 6 Average cost overrun rates for different capacity groups Capacity groups Small Medium Large Small Medium Large Small Medium Large Small Medium Large Small Medium Large Average 0.19 0.03 0.05 0.24 0.25 0.16 0.13 0.08 0.03 0.34 0.19 0.20 0.12 0.03 0.05 SD 0.80 0.24 0.43 0.57 0.84 0.38 0.97 0.43 0.43 1.50 1.19 1.05 0.46 0.29 0.21 Skewness 3.81 0.97 8.75 1.54 5.37 0.47 4.14 1.11 0.94 2.50 4.00 3.54 1.72 2.45 0.95 Kurtosis 22.63 4.47 93.69 7.35 38.99 5.10 25.36 4.25 4.79 9.29 23.95 19.34 7.77 13.17 7.96 Num. of pipelines 135 136 139 135 137 140 133 135 138 128 130 132 135 136 139
2

Components Material

Labour

Miscellaneous

ROW

Total

The KW test is used to verify that the pipeline capacity has no effect on cost overrun for construction cost components. For material cost, the result of KW test rejects the null hypothesis (p < 0.001), indicating that pipeline capacity influences material cost overrun, and projects with small pipeline capacity have large positive cost overrun rates. Pipeline projects with large capacity mean that more material is consumed. Therefore, projects with a large capacity take advantage of economies of scale in purchasing materials, resulting in lower costs for materials as pipeline capacity increases. It may be that the

An analysis of inaccuracy in pipeline construction cost estimation

41

estimator of pipeline projects did not estimate unit price with capacity changing accurately or did not consider economies of scale in material cost. This may result in pipeline projects with small capacity that have large cost overrun. For labour, miscellaneous, ROW and total costs, the results fail to reject the null hypothesis that pipeline capacity has no effects on the cost overrun rates (p > 0.05). Therefore, the cost overrun differences in the labour, miscellaneous, ROW and total costs are not statistically significant for different types of pipeline capacities.
Table 7 Average cost overrun rates for different regions Region groups Midwest Northeast Southwest Canada Central Southeast Western Midwest Northeast Southwest Canada Central Southeast Western Midwest Northeast Southwest Canada Central Southeast Western Midwest Northeast Southwest Canada Central Southeast Western Midwest Northeast Southwest Canada Central Southeast Western Average 0.02 0.02 0.02 0.18 0.06 0.26 0.09 0.12 0.12 0.28 0.02 0.20 0.33 0.55 0.06 0.07 0.05 1.32 0.02 0.04 0.08 0.34 0.10 0.12 1.59 0.26 0.08 1.01 0.01 0.00 0.84 0.14 0.11 0.13 0.19 SD 0.29 0.56 0.37 0.26 0.28 0.92 0.50 0.38 0.34 0.60 0.33 0.49 0.85 1.14 0.43 0.45 0.52 2.25 0.37 0.55 0.54 0.99 0.76 1.14 2.18 1.11 0.65 2.35 0.24 0.26 0.34 0.31 0.29 0.45 0.48 Skewness 0.03 7.33 0.32 0.80 1.58 3.63 3.22 1.19 0.87 1.04 1.04 0.31 3.01 4.20 1.72 1.14 0.62 1.56 0.84 1.60 1.73 3.42 2.31 3.66 0.76 2.74 1.18 1.84 0.77 1.72 0.60 0.86 1.35 1.93 2.76 Kurtosis 4.81 72.15 5.35 2.75 8.24 15.69 16.22 8.36 5.91 3.30 3.95 2.38 14.70 23.28 7.74 5.97 2.62 4.03 3.97 5.93 6.28 20.21 12.10 17.27 2.01 12.51 4.90 5.14 5.72 10.97 3.68 4.11 6.69 6.80 10.77 Num. of pipelines 55 156 30 14 52 55 48 55 157 30 15 52 55 48 54 155 30 14 51 55 47 53 150 27 14 50 52 44 55 155 30 15 52 55 48

Components Material

Labour

Miscellaneous

ROW

Total

42

Z. Rui et al.

Cost overrun in terms of different regions

As we know, pipeline cost is significantly different for different regions (Rui et al., 2011b). This section discusses whether cost overrun of pipeline construction cost components is different for different regions. As seen from Table 7, it is noticeable that cost overrun rate of the pipelines in the Northeast regions is the lowest in the USA as compared to the other regions, even though the Northeast has a relatively high cost of living. In addition, the total cost overrun rate of pipelines in the Northeast regions is a perfect 0. A possible explanation is that 155 out of 412 pipelines in the data set are located in the Northeast regions, which provides more practical experience and historical information for new pipeline cost estimating in this region. A few negative cost overrun rates also appear in some regions for different construction component cost. The result of KW tests shows that the cost overruns of difference for different regions are highly significant for all construction cost components (p < 0.001). Weather condition, soil property, population density, cost of living, terrain condition, and distance from supplies are variable for different regions and make pipeline project cost estimation more difficult (Rui et al., 2011b; Zhao, 2000). More detailed information on pipeline routes is needed to explain cost overrun difference among different regions. Therefore, it is concluded that the cost overrun rates of all cost components show significant difference for different regions and the pipelines location matters to cost overrun in all cost components.

Cost overrun over time

Forty-seven megaprojects between the mid 1960s and 1984 were reported with a cost overrun rate of 88% (Merrow, 1998). Over 1,000 World Bank projects between 1947 and 1987 had cost estimated errors (Pohl and Mihaljek, 1992). 55% of all Indiana Department of Transportations projects between 1996 and 2001 experienced cost overruns (Bordat et al., 2004). Cost overrun is constant for more than a 70-year period between 1910 and 1998 for 208 transportation projects in 14 nations on five continents (Flyvbjerg et al., 2003). All the literature show that the cost estimated error persists over time in many different types of projects. But is there any improvement in pipeline compressor station projects over time? This section tries to discover whether the cost estimating performance has improved over time. Improved performance of cost estimating is normally expected. The average cost overrun rate of compressor station construction components between 1992 and 2008 are displayed in Figure 7. The cost overrun rate of ROW cost fluctuates widely with a declining trend. The cost overrun rate of labour cost shows a decrease before 2004 and then a significant increase after 2004. But cost overrun rate of material, miscellaneous and total costs change more gradually over time. The length of the construction phase influences cost overrun rate. Therefore, it is better to use the year of planning to build as the time measurement (Flyvbjerg et al., 2003), but the available data does not provide the year of building and construction period. Therefore, the year of completion is used to as a measure of the time, which may cause bias. The non-parametric Nptrend test is conducted to test whether there is a trend of cost overrun rate over years. All results of the Nptrend test show only cost overrun rate of ROW decreases over time (p < 0.05). Therefore, based on available data, it is

An analysis of inaccuracy in pipeline construction cost estimation

43

concluded that cost estimating of ROW cost has improved over time, not for other components.
Figure 7 Annual average cost overrun rate of pipeline construction cost components (see online version for colours)

10 Conclusions and future work

This paper statistically analyses the cost estimating performance of individual pipeline construction cost components by using 412 pipeline projects. The trend and distribution of all 412 pipelines construction cost components cost estimation over the 19922008 periods are analysed. Overall average cost overrun rates of the material cost, labour cost, miscellaneous cost, ROW cost and total cost are 4.9% (SD = 54.8%), 22.4% (SD = 61.8%), 0.9% (SD = 56.2%), 9.1% (SD = 80.9%) and 6.5% (SD = 33.5%) respectively. The labour and ROW costs have the largest cost overrun compared to the other cost components. The statistical test results show that cost estimating in all cost components is biased except in the total cost. And the magnitude of cost underestimating error is generally larger than the overestimating error except for total cost. Furthermore, cost overrun rates of pipeline construction cost components are analysed in terms of pipeline project size groups, capacity groups, diameter groups, length groups, location groups, and the year of completion to investigate the relationship between cost overruns and different groups. The cost overrun rate for the total cost shows a significant difference for different project size groups, and the cost overrun rate increases with project size. An expected large demand, limited supplies and contractors for large-size projects cause big cost overruns (Bordat et al., 2004; RGL Forensics, 2009; Rui et al., 2011a). The cost overrun rates of the labour and miscellaneous costs show significant differences for diameter groups, and the small diameter group has the highest average cost overrun rate. The cost overrun rates of all construction cost components are not significantly influenced by pipeline length. The cost overrun rate of the material cost is significant for different pipeline capacity groups, and projects with small pipeline capacities appear to be particularly prone to cost overrun. Large capacity pipeline projects may purchase material at low price due to large scale. The planners or estimator

44

Z. Rui et al.

may not estimate the material unit price changing with scale accurately or may even fail to consider the economies of scale factors. The cost overruns of all construction cost components are significantly different for different regions. The weather, soil, terrain, terrain condition, population density and experience are suggested as causes for making it difficult to estimate cost accurately. Cost estimating accuracy of pipeline projects have not improved over the 19922008 time period. Based on the analysis of historical pipeline cost estimated errors, Table 8 provides some proposed guidelines for project estimators conducting pipeline cost estimation. It is considered that individual cost components should receive varying degree of attention under different conditions in order to make cost estimation efficient and reliable. A four level scale: maximum attention, moderate attention, less attention, and minimum attention, is applied for the amount of attention and effort paid to individual cost component cost analysis, depending on the project size, pipe diameter, pipeline length, pipeline capacity, and region of construction, as given in Table 8.
Table 8 Category Project size Proposed guidelines for pipeline cost estimators Sub-category Small Medium Large 420 inches 2230 inches 3448 inches 020 miles 20713 miles Small Medium Large Midwest Northeast Southwest Canada Central Southeast Western Material C D D B D D D D B D D D D D B D D B Labour B A B A A C A A A A B B B A D B A A Miscellaneous D D D B B D B D B C D C C D A D D C ROW B A A A A B A A A B B A C B A A C A Total D C B B D D C C B D D D D A B B B B

Diameter

Length Capacity

Region

Notes: A = maximum attention; B = moderate attention; C = less attention; D = minimum attention

To the best of the authors knowledge, this paper is the first in-depth analysis of pipeline construction component cost overruns. Suggested future work may include the following list:

Different reasons for cost overrun for other types of projects are proposed by different researchers, and many of the proposed causes did not have hard data support. The same situation applies to pipeline projects; lack of good quality projects data is a major difficulty in investigating more in-depth causes of pipeline cost overrun. Therefore, collecting more accurate information on the pipeline

An analysis of inaccuracy in pipeline construction cost estimation construction period, the ownership of projects, the material of pipeline, and the thickness of pipeline wall is a major part of future work. Furthermore, quantity analysis of collecting data should be conducted to investigate the causes of cost overrun.

45

More application with analysis results will be used in the future pipeline projects, such as pipeline cost overrun distribution and average cost overrun rate. Develop a set of recommendations to help mangers and engineers to better estimate pipeline project costs and minimise the cost estimating errors.

References
Bertisen, J. and Davis, G.A. (2008) Bias and error in mine project capital cost estimation, The Engineering Economist, Vol. 53, No. 2, pp.118139. Bordat, C., McCullouch, B., Sinba, K. and Labi, S. (2004) An Analysis of Cost Overruns and Time Delays of INDOT Projects, Joint Transportation Research Program, Indiana Department of Transportation and Purdue University, West Lafayette, Indiana. Chemical Engineering (2011) Chemical Engineerings Plant Cost Index, available at http://www.che.com/pci (accessed on May 2011). Energy Information Administration (EIA) (2010) U.S. Natural Gas Pipeline Regional Definitions Map, available at http://www.eia.doe.gov (accessed on July 2010). Flyvbjerg, B. (2007) Policy and planning for large infrastructure projects: problems, causes, cures, Environment and Planning B: Planning and Design, Vol. 34, No. 4, pp.578597. Flyvbjerg, B., Skamris, M. and Buhl, S. (2003) How common and how large are cost overrun in transport infrastructure projects?, Transport Reviews, Vol. 23, No. 1, pp.7188. Hill, R.C., Griffiths, W.E. and Lim, G.C. (2007) Principle of Econometrics, 3rd ed., John Wiley and Sons, Inc., New York. Infrastructure Partnerships Australia (2008) Available at http://www.infrastructure.org.au (accessed on May 2011). Jacoby, C. (2001) Report on Supplemental Agreement Reasons, AASHTO-FHWA Project Cost Overrun Study, Federal Highway Administration, US Department of Transportation, Washington, D.C. Jahren, C. and Ashe, A. (1990) Predictors of cost-overrun rates, ASCE Journal of Construction Engineering and Management, Vol. 116, No. 3, pp.548551, American Society of Civil Engineers, Reston, VA. Merrow, E.W. (1998) Understanding the Outcomes of Megaprojects: A Quantitative Analysis of Very Large Civilian Projects, R-3560-PSSP, Rand Corporation, CA. Odeck, J. (2004) Cost overruns in road construction-what are their sizes and determinants?, Transport Policy, Vol. 11, No. 1, pp.4353. Penn Well Corporation (19922009) Oil & Gas Journal Databook, Tulsa, Oklahoma. Pindyck, R.S. and Rubinfeld, D.L. (1995) Microeconomics, 3rd ed., Prentice-Hall, Englewood Cliffs, N.J. Pohl, G. and Mihaljek, D. (1992) Project evaluation and uncertainty in practice: a statistical analysis of rate-of-return divergences of 1015 world bank projects, The World Bank Economic Review, Vol. 6, No. 2, pp.255277. RGL Forensics (2009) Ex Post Evaluation of Cohesion Policy Programmes 20002006, Work Package 10, European Commission, Brussels, BE. Rowland, H. (1981) The Causes and Effects of Change Orders on the Construction Process, PhD Thesis, Georgia Institute of Technology, Atlanta, GA.

46

Z. Rui et al.

Rui, Z., Metz, P.A., Reynolds, D.B., Chen, G. and Zhou, X. (2011a) Historical pipeline construction cost analysis, International Journal of Oil, Gas and Coal Technology, Vol. 4, No. 3, pp.244263. Rui, Z., Metz, A.P., Reynolds, B.D., Chen, G. and Zhou, X. (2011b) Regression models estimate pipeline construction costs, Oil & Gas Journal, Vol. 109, No. 14, pp.120127. Zhao, J. (2000) Diffusion, costs and learning in the development of international gas transmission lines, Working paper IR-00-054, International Institute for Applied Systems Analysis, Austria.