Case Studies of Cost Overrun

There are plenty of case studies of cost overrun and cost growth in scientific journals and daily newspapers. One project that reached iconic recognition due not only to its architecture but also to its cost overrun is the Sydney Opera House. The project started in 1956 with the announcement of an international architectural competition for a multipurpose opera, concert, and theatre building. The competition received 233 entries and in 1957 an international committee selected an innovative proposal from the Danish architect Jørn Utzon, which the assessors deemed had the potential to put Sydney on the world map.¹

Construction started two years later, with estimated completion in 1963. By then the estimated cost had risen from $A 7.2 million to $A 9.76 million and the well-known London based engineering firm Ove Arup and Partners had been chosen to construct the building.

Contracts were signed before Utzon’s innovative proposal had been transferred into drawings that could be implemented with existing building techniques.² The proposal did not have room for enough seats, and new building technics had to be developed as the design could not be built using pre-stressed concrete. This led to conflicts between the architect’s vision and user requirements, and Utzon left the project in 1966, at a time when there still was no complete architectural drawing of the interior of the building, and the project was delayed as a result. The Sydney Opera House was completed in 1974 at a cost of $A 103 million -- 14 times more than the original estimate made 16 years earlier. Costly, yes, but all cost increases since 1959 were covered by popular lotteries.

The Sydney Opera House exhibits many traits in common with other projects where estimated costs have increased dramatically in that the investment decision was based on (a) poor or non-existing cost estimates, (b) a partly unspecified design with unsolved technical problems, leading to subsequent development through trial and error with substantial re-work and modification throughout the construction process, (c) iterations between the wishes of users, architects, and construction firms, and (d) a long construction time which inevitably makes estimates outdated and increases cost due to inflation, all factors which we see recurring in projects with large cost overruns.

From a project planning point of view, the project was a failure, but its promoters wanted a building that would put Sydney on the map, and that they got. It is one of the most popular tourist attractions in Australia, visited by more than seven million people each year. In 2003, Jørn Utzon received the prestigious Pritzker Prize³ for his design of the Sydney Opera House, and in 2007 UNESCO made the building a World Heritage Site. It was yet further recognition that the initiators of the project had indeed achieved their dream of a world-renowned iconic building that would become a symbol of Sydney.

One might wonder if those involved would have dared to choose Utzon’s innovative sketch had they imagined all the technical problems that had to be overcome. However, had they settled for a traditional building, for which there was a prototype, the Sydney Opera House would not have appeared on the World Heritage list, and might not have fulfilled its purpose of putting Sydney on the map. And by the time the building was finished, financed by lotteries, its large cost overrun contributed to rather than marred its iconic status.

There are many books and articles written about problem-ridden projects with large cost overruns like the Sydney Opera House, usually about public projects as freedom of information acts give the public access to documentation on such projects. The question is whether experiences from these well-known projects are project-specific or if one can draw any general conclusions from their stories. Do any aspects of these projects show common traits or have common denominators? Let us try to paint a brief picture.

It is doubtful if a private company with liability for payment would have taken the risks such public projects represented, which raises the question of how similar these well-known and often very politicized public projects are to private sector projects, and also to public projects that generate less political and national attention.

One can of course look for similarities in private industrial projects. More thorough follow-ups are sometimes made for major private sector investments with large budget overruns, as well as for strategies when management feels targets may not have been achieved. We can learn quite a lot from such reviews, but in most cases it is difficult to generalize as the reasons suggested for the problems are case-specific. Reviews can find that cost overrun was due to, for example: it taking longer than planned to make the plant control equipment work; supplies not being received in time; suppliers having financial problems; unanticipated price increases; a lack of skilled labour causing local inflation; poor cost control; or problems of seasonality, i.e. an unusually cold winter. Such explanations can be correct and relevant for the project in question, but many are difficult to generalize to other projects. However, three overarching explanations repeatedly turn up when going through reviews of a number of large projects, namely:

1. - Changes in design

2. - Increase in scope

3. - Inflation and price increases

These three causes interact. Changes in design can be partly due to inflation and price increases, and lead to an increase in scope. When a project shows a large cost overrun, it is almost always not exactly the same project as was originally costed because the original design has been revised and the project re-costed. Some project managers might object by claiming that the final cost of the project should not be compared with the budget, as the changes mean that we now have a much better building, plant, or machine than what was originally approved. That might be correct, but, nevertheless, if the final cost is higher than the approved budget, then we have a cost overrun.

In the next three sections of Appendix A, we shall review what studies of groups of similar projects have to show us about the occurrence of cost overrun and growth. Building mainly on these studies, in Appendix B we will generate some general conclusions on cost overrun and growth, which we will use in Chapter 4 to build a model of cost overrun.

Construction Projects

To be able to generalize about cost overrun, we either need a theory that helps us to understand our observations of cost overrun, or data on a large enough number of projects to identify statistically significant differences between projects of different types. In the former case, we rely on theories applied earlier to other areas of social science, and explanations at the individual and organizational levels, to explain cost overrun in case studies of individual projects; in the latter, we rely on descriptive explanations based on the similarities and dissimilarities identified in the dataset.

We will draw on both kinds of research, and we will start by reviewing statistical studies of deviations in cost estimates for groups of projects, as such studies provide a good point of departure for theorizing about cost overrun. First, we explore studies of construction projects, then research and development projects, and, finally, software development projects. Some of the studies of cost overruns in construction projects are listed in Table A.1. Figures on average cost overruns in parentheses are in constant money value. It is difficult to find a study showing that construction projects taken as a group become on average less expensive than anticipated; cost overrun is the norm.

Although the number of studies and projects is high, we still have to be careful about drawing overconfident conclusions from them as the quality of the data analysed is sometimes questionable, and this makes it difficult to compare data from one study with another. Admittedly, investigators have considered inflation in several of the studies, but the way this has been done is not always described, and in many of the studies project cost data is compiled from different sources, estimators, organizations, and decades. Considering how common and unwanted cost overrun is, there are relatively few studies of cost overrun of sufficient size and quality that they allow us to draw statistically significant conclusions about cost overruns. We are forced to assume that the errors in project cost data balance each other out when we pool projects in groups of similar projects. Moreover, most of the studies in the social science literature concern major public sector projects. Studies of private sector projects are far less common, and to what extent ownership influences project outcome is poorly researched.

Product Development Projects

Studies of groups of product development projects show great similarities to construction projects. Not only is cost overrun the norm, but also the causes of cost overrun are very similar to those found in construction projects.

Military product development projects tend to demonstrate cost growth. The US authorities commissioned the RAND Corporation, a policy think tank, to study the causes of cost overrun in weapon development programmes in the 1950s. Two of these early studies are listed in Table A.2. Figures in parentheses are adjusted for inflation. The first study was conducted by Andrew W. Marshall and William H. Meckling in 1962²² and the second one by Robert Summers in 1967.²³ They both used regression analyses to study 68 estimates in various phases of 22 military aircraft and missile programmes undertaken between 1945 and 1958 in the USA. In doing so, they corrected their figures for inflation and the number of airplanes produced.

One major reason for cost overrun in military aircraft projects has been that the estimated system cost has been based on the assumption that the cost of development would be paid for by a larger number of aircraft being built than were actually constructed. In this context, one must also mention Merton J. Peck and Frederic M. Scherer’s study²⁸ of the weapon acquisition process. Since these studies, there has been a constant flow of analyses of cost growth in US military and other public sector projects, but none of the studies have solved the problem.

Building on this earlier work at RAND, Edwin Mansfield²⁹ led a research team that analysed product development in two US pharmaceutical companies. They had data from 75 projects in one of the companies and 49 from the other, all carried out between 1950 and 1967. The data from these two different companies is not directly comparable, but the researchers did correct their data for inflation. Their study showed, among other things, that one of the companies was better at estimating costs, and had improved its cost estimates, but not the other. Organizational factors matter.

A later study of development projects in Swedish engineering firms shows similar figures. As is evident from Table A.2, one of the companies showed cost growth four times as large as the other, or 120 per cent. Project cost not only grew more in one of the companies, but also the outcome varied more. The standard deviation was twice as large. Three causes of cost overrun were stressed. The most important was that the level of ambition had increased, i.e. the final product had more functionality than originally planned; the other two were uncertainty and lack of resources due to development resources being shifted to competing products.

As can be seen by comparing Tables A.1 and A.2, cost overrun is, in general, somewhat larger for product development projects than for construction projects, as demonstrated, for instance, by the American study from 1983 of 244 military and 200 civilian projects. In fact, one of the most important conclusions from studies of cost overrun in product development projects is that the larger the advance in new knowledge that is made, the larger the cost overrun tends to be. This connection, which was formulated by Marshall and Meckling in 1962, has been confirmed in many later studies, and has also been advanced as one of the reasons behind military projects being more prone to cost growth than civilian ones.

Cost growth in weapon development projects continues to be scrutinized. Studies at RAND³³, in the UK,³⁴ Norway,³⁵ and Sweden³⁶ have focused on how the cost per unit produced for each new generation of weapon system increases faster than can be explained by the rate of inflation alone. This type of cost growth is called intergenerational cost growth, as opposed to intragenerational cost growth. The latter can decrease as cost per unit may lower due to the learning curve that exists in all repetitive operations.

The cost of naval ships and aircraft has increased faster than the rate of inflation for a very long time, and, because of that, the number of fighting ships and aircraft a country can afford to acquire and maintain has gradually decreased. Three such studies are summarized in Table A.3, and show that the unit cost of a specific piece of military equipment in constant monetary value has increased by 0.7 to 7.6 per cent per year, or more than 4 per cent on average. Unit cost has increased faster for high-tech weapon systems produced in low quantities than for more standardized weapon systems. With each new technology generation, navies can acquire increasingly fewer ships, air forces fewer aircraft, and armies fewer tanks.

Computer Software Projects

Another type of project that is often affected by cost overrun is computer software projects. A study that aroused a lot of public attention is the Standish Group’s CHAOS Report, published in 1995.³⁷ This report was based on a survey answered by 365 IT managers representing 8,380 projects. Thirty-one percent of the projects had been discontinued, and the average cost overrun for these, and the other 52.7 per cent in which costs had increased, was 189 per cent. Schedule overrun was even larger, or 222 per cent; only 61 per cent of the 52.7 per cent of projects with cost overruns had achieved the specifications originally set out. See Table A.4.

The CHAOS Report has been criticized³⁸ on the grounds that the results of the study do not correspond with other American, Canadian, and Dutch studies showing an average cost overrun of 33 to 50 per cent. One factor that can distort the figure is whether discontinued projects are included or not. Many Software and R&D projects are abandoned as they do not meet their targets. This separates software and R&D projects from construction projects. It is unusual to start but not finish the construction of a house, bridge, road, or industrial plant. This is the difference between what can be termed implementation projects and development projects, i.e. projects that can be implemented based on existing knowledge, and projects that require the development of new knowledge to be completed.

Software projects and many other development projects have high complexity because they are made up of many sub-activities that have to be delivered on time for the next phase to begin. In the case of software, there are also external dependencies as customers often cannot specify exactly what they want the programme to be able to do. Project requirements are vague, which is why it often becomes necessary to change them during the course of the project.

The pioneer in the field of software cost estimation, Barry W. Boehm,⁴⁷ estimated the cost of making a change in the design phase to be 5 times higher than in the requirement phase, 10 times higher in coding, and 50 in testing. Later studies⁴⁸ have arrived at even higher costs in later phases. The cost of making changes increases dramatically the nearer the project is to completion. Computer software firms try to reduce the risk of costly late changes by involving customers in all phases of the project.⁴⁹ The anchoring process becomes very important and one can speculate whether the fact that many software projects are abandoned, some are finished close to target, and some exhibit large cost overruns is in part a reflection of the outcome of this anchoring process.

More recently, a Dutch study⁵⁰ showed that the organizations examined had on average no cost overrun on their software projects. The cost of some projects increased and some decreased, but on average organizations Y and Z had no cost overrun. In organization X, many projects did not use up what was budgeted. The reason was that estimators aimed at the maximum cost, while appropriation allocators made cuts and seldom approved additional funding. In a study of software development in one American firm,⁵¹ the main causes of cost overrun were identified as underestimation, changes, and a corporate culture seeing estimates as targets and giving priority to customer satisfaction.

Cost overruns vary by organization, cost estimating, appropriation, and control practice. Results are conflicting and one reason can be that, unlike, say, a bridge, software programs can become operational even when all specifications have not been met. This makes it possible to meet the budget despite not meeting goals and specifications.

Uncertainty Decreases over Time but Cost Tends to Increase

Projects can become less expensive than anticipated, but the opposite is far more common. Peter Morris and George Hugh¹ reviewed 35 studies of groups of major projects and did not find a single study in which the projects studied had become less expensive on average than anticipated. In fact, only three studies listed in Tables A.1–A.3 showed cost underrun, and then only for a specific type of investment project in a single firm. Cost overrun is the norm regardless of whether it is a construction, research and development, or computer software project.

The estimate on which the budget or the decision to invest is based is often preceded by at least one earlier cost estimate. Three or more estimates and appraisals are not uncommon in major capital investments before the board feels certain enough to give the investment its final approval. As well as a tendency for costs to rise after the decision to invest, there is also a tendency for costs to rise between the very first estimate and the one on which the decision is based. The largest increases tend to occur between the first and the second estimates. This tendency for costs to rise over time has been observed in US military aircraft,² all types of US defence projects,³ and pharmaceutical ventures,⁴ as well as in smaller projects.⁵ Early estimates tend to be more biased towards cost increase than later ones, and there are several possible reasons⁶ for this.

Firstly, many estimators do not want to put their name on pure guess work. Therefore, when there is great uncertainty about whether an item or work is really needed, it is often omitted. Estimators want to avoid the risk of being wrong, but as the omission of uncertain future cost is not balanced by some equally large omitted revenue stream, this is a bias that tends to drive costs up when the uncertainty surrounding the omitted items has been resolved.

Secondly, there is a tendency for project planners to choose simple and cheap solutions in the hope that they will work, although they might not. By doing so, they hope to keep cost down and avoid the uncertainty associated with more advanced solutions. Although a means of uncertainty avoidance, this will drive cost up if more expensive solutions prove necessary later on.

Thirdly, there are instances in which the initial estimate has been deliberately set too low, in order to allow a study of the proposed idea to go ahead, or to help the external contractors to gain a contract. The former does not necessarily mean that the estimator wants to deceive; in most cases the estimator is probably fully convinced of the merits of the idea but unable to prove it.

Fourthly, costing is a learning process. We can only cost those items we are aware of, and, as uncertainty is resolved and our knowledge of the final product becomes more detailed and realistic, we will discover that additional items and works are needed to achieve what we intend to do.

And finally, early in this learning process expectations and estimates are based more on emotions than in later phases, when the control system forces project proponents to prove that the investment is the right thing to do. Also, investments in new areas and technologies are based more on hope than investments in existing areas and technologies. Emotions get more space to influence expectations when cost data is lacking and control systems weak.

Thus, the first estimate tends to be an underestimation, and the minimum cost always more reliable than the maximum if such a cost span is given. And as long as we only consider the cost side, there is no way to balance these tendencies out by increasing revenues. Moreover, starting the process by omitting uncertain future cost, choosing solutions based on hope, or setting the budget to a level that allows the project idea to be studied can make the initial budget unrealistically low.

However, early estimates are not only lower than later ones, they also vary more and are less accurate than later ones. The variance in cost estimates decreases over time.⁷ Estimates become better as the uncertainties in the project design are resolved, but more reliable information comes only at a certain cost. Donald S. Remer and Harry R. Buchanan⁸ estimate that it costs 1 per cent of the investment outlay to achieve an accuracy of −5 to +15 percent; more in smaller projects, and less so in larger projects. In practice this seems like a rather low figure for major investments, as they are often preceded by several appraisals of different design alternatives, and evidence from case studies of such major projects suggests otherwise; Om Prakash Kharbanda and Ernest A. Stallworthy⁹ conclude that some 3–5 per cent of total investment is needed to develop a realistic estimate.

In a study by the author¹⁰ of 13 major ventures in new areas, the companies that implemented these successfully spent about 10 per cent of total investment before the board felt certain enough to give the venture its final approval. These were major investments in new areas in the order of almost $500 million in today’s money value. When making such an investment in a market and a technology new to the firm, it is not surprising that the firm first spends a tenth of the budget gaining expertise in the new technology and market, e.g. by studying the market and technology, recruiting managers with experience of the new business, acquiring smaller companies in the new business, test-selling the new product, and signing contracts for parts of the production. If the investment had failed, it could have forced some of these firms into bankruptcy. To spend 10 per cent of the investment outlay on reducing that risk is not a high cost considering that these projects were all ventures into new areas where the parent company lacked previous experience, and that most of the pre-investment could be recovered if the new venture succeeded. This is the cost of learning a new technology and market – the ticket to the industry. For major industrial projects in general, the 3–5 per cent Kharbanda and Stallworthy recommended might very well suffice.

An interesting observation from this study of 13 new ventures was that there seemed to be a relationship between success and the amount of funds spent prior to the decision by the board to invest. Those ventures that had been successful had all been carefully evaluated and planned before the board finally approved the venture, while commitment had preceded careful investigations in many of the ventures that had failed. Strong early commitment to implement the venture is a trait shared by many project failures¹¹ in both the private and public sectors. Positive but critical feedback from a well-developed pre-approval control system forcing project proponents to try to foresee what might lie ahead before they are allowed to implement their idea seems to be a good form of insurance.

Cost growth can be explained as the outcome of a learning process which is not totally predictable. To make a cost estimate, estimators need a design to cost. A cost estimate takes its point of departure in an image of a design and path to realize a design, and as we learn more about the design we often tend to discover that additional work and changes in the design are necessary, making earlier estimates outdated and increasing the final cost. This bias towards cost growth is compounded by the fact that estimators tend to exclude items and work which they are unsure will be necessary. As uncertainty is resolved, changes in the design become smaller and additional costs appear more infrequently. When the design is fixed, work becomes more foreseeable and cost estimates more reliable. Thus, later estimates might be higher, but they are also more reliable. The variance in estimates decreases as more is learnt about the project that is to be implemented.

But this increased precision comes at a cost. The situation then requires a decision as to how accurate the estimate needs to be to approve a budget and give the decision to go ahead. This also means that the size of a cost overrun depends on what we compare it with: an estimate in the idea phase, or in the planning and design phase, or a preliminary budget when approval to ask for tenders are given, or when tenders exist and the final budget is approved. Many studies of cost overrun do not state what the final cost is being compared against.

Decision makers have to have an estimate upon which they can justify their decision to appropriate funds to further investigate a project idea. They cannot allocate funds to unspecified projects. Formal procedures prohibit that. Experienced managers know that these early estimates are uncertain and probably too low. This is generally not a problem in companies, but it can create problems in the public sector as early estimates and commitments become public and estimates can be interpreted as a correct figure. Such a misconception can later make it difficult to explain why project cost has increased due to inflation and changes in the proposed project.

Small Projects Vary More than Large Ones

Calculated as a percentage of total cost, the deviation in cost between approved budget and final cost varies more for smaller than for larger projects, i.e. smaller projects are more often hit by large cost overruns, and also large cost underruns. This tendency for smaller projects to vary more has been observed in power generating,¹² infrastructure,¹³ road,¹⁴ and pharmaceutical development projects.¹⁵ For projects of the same type of technology, smaller projects account for both the largest underruns and the largest overruns; larger projects are usually found closer to the average. However, it should be said that this applies to, or is at least more evident in, homogenous groups of projects – projects representing similar technologies and implemented in the same organization.

This proposition is illustrated by Figures B.1 and B.2, which show the deviation in cost as a function of project size for 67 electrical transformers and 37 power lines.¹⁶ The figures derive from a study of the 115 investments reported to the board of the Swedish State Power Board: 11 hydroelectric power stations, 37 power lines, and 67 transformers. On the vertical axis we have approved project budget and on the horizontal axis deviation per month. Deviation in per cent per month was the measure preferred by the Power Board. Exchanging this measure with total deviation during implementation makes the smallest projects vary less but does not radically change the shape of the distribution; smaller projects vary more than larger ones.

Figure B.1 The size and frequency of cost overrun in 67 transformer projects

Figure B.2 The size and frequency of cost overrun in 37 power line projects

Several explanations were advanced at meetings with project managers at the Power Board. One was that the larger the project, the higher the financial and personal risks, and hence it becomes more important to avoid cost overrun. Larger projects are therefore given more attention and more experienced project management. In fact, many firms, not just the Power Board, require follow-ups for investment projects above a certain expenditure limit.¹⁷ In absolute terms, more resources are spent on planning and reviewing large projects, and this may in part explain why larger projects are generally affected by cost overrun more often.

Secondly, although more resources are spent on large projects in absolute terms, the percentage of total investment outlay needed to estimate the correct cost is higher for smaller than for larger projects, and, furthermore, smaller investments are more often replacement or additional investments to larger ones. Replacement and additional investments tend to deviate more than new investments, as it is difficult to know exactly what is needed, or it would cost too much to find out beforehand. When disassembling a part of an old plant, one often finds additional parts that need to be replaced. Thus, in relative terms, smaller projects, and especially replacement and additional investments, are costlier to estimate correctly.

Thirdly, it was pointed out that a larger share of the total cost of a transformer project was procured. Power lines were built and maintained by the permanent workforce of the Power Board, and wages make up a larger part of such projects. Therefore, Power Board engineers had knowledge and control of a larger share of the total cost of a power line than of a transformer.

Fourthly, changes in investment plans have more impact on smaller projects. Allan and Norris¹⁸ looked for explanations for the deviations between estimate and outcome for 84 smaller research and development projects but could not identify one single explanatory factor causing cost overrun that was valid for all projects. The average size of these projects in current money value would be in the range of £80–£90,000, and the development time 28 months. One possible explanation is that it might be more difficult to identify explanations of a more general kind in small projects as the percentage-wise outcome of such projects are more easily affected when they are postponed or accelerated due to company politics, terms of payments, changes in R&D priorities, and events outside the control of project management.

A fifth explanation builds on the observation that the performance of individual budget items in budgets for major projects can vary quite a lot even in projects that have been implemented to cost, as was illustrated by Table 2.1. Large projects can usually be divided up into several smaller projects and the variance in outcomes cancelled out when smaller projects are brought together. Although the paper board machine project described in Chapter 2 came in close to budget, some of the sub-projects showed cost overruns of almost 100 per cent. The outcome of smaller projects is more easily affected by occasional events such as delays, budget cuts, payment conditions, and project- and company-specific events. Larger projects vary less as deviations in sub-projects tend to cancel each other out. This is an example of the law of large numbers.

The More Unique a Project Is to the Estimator, the More Cost Tends to Grow

Deviations vary not only by size, but also by type of project. Figure B.3 shows the distribution in cost for 191 major public sector infrastructure projects distinguished as highway, water resource, rapid rail, building, and ad hoc projects. Data was collected by Leonard Merewitz,¹⁹ who showed that cost overrun was on average lower for water resource and road projects than for what he terms ‘ad hoc projects’, i.e. projects which are unique, one-of-a-kind to the investor making the investment, such as research and new sports facilities.

Figure B.3 Cost overrun in 191 major infrastructure projects

The tendency for certain kinds of projects and technologies to exhibit larger cost overruns has been shown in a number of studies. Tables A.1 to A.3 give studies showing that military projects exhibited larger cost overruns than civilian projects,²⁰ pipeline larger than liquefied natural gas projects,²¹ nuclear reactors much larger than solar farms,²² highway larger than railway projects,²³ and railway larger than highway projects.²⁴ The results can be contradictory, but study after study shows that some types of projects are more prone to increase in cost.

Figure B.3 shows that the distribution of cost for the 191 major public infrastructure projects is positively skewed. This pattern of projects exhibiting major cost overruns is not balanced by an equal number of projects showing cost underruns, as confirmed in most studies of cost overruns. Such a positive skewness exists in, e.g. studies of major public infrastructure projects,²⁵ roads,²⁶ mining,²⁷ weapons development,²⁸ and computer software projects.²⁹ The distribution of the deviation of cost for a group of projects is in general positively skewed. Exceptions are few, power lines in Figure B.2 being one.

If we compare Figures B.1 and B.2, we find that the average deviation in cost and the distribution of deviation in cost differ between transformers and power lines. One explanation for this given by the project managers interviewed was that labour makes up a higher proportion of the cost of power lines than that of transformers. As the Swedish State Power Board had people permanently employed to build and repair their facilities, they required a high and steady capacity utilization. To achieve this, they sometimes used smaller projects to regulate the workload of their employees. The effect of this should be more evident for power lines than for transformer stations, as the latter are more driven by equipment costs than labour.

One simple explanation for positive skewness is that resources spent cannot be fully recovered when changes are made and extra work added. As it is much more likely that changes and additional work will increase rather than decrease cost, changes and additional work will create a positive skewness.

Comparing Figures B.1 and B.2, one can observe that investments in new transformers and power lines vary less than reinvestments and additional investments. One reason for this is that new needs are often discovered when work begins on old machines, needs that were not visible when the machine was operating. Another reason is that renovation is often regarded as a forced investment. When an old machine is in bad condition there is no alternative but to repair and reinvest, so costing the investment accurately is felt to be less important than for a new investment. The fact that transformers show a pronounced positive skewness but power lines do not could thus be explained by the fact that new power lines are more similar to earlier ones than transformers. If building power lines is a more familiar technology with fewer surprises when implemented, the positive skewness could be explained by unanticipated time delays, additional work, and redesigns causing additional costs.

As was shown in Figure B.3, ad hoc projects in particular often experience cost overrun. Examples of such projects in the Merewitz sample were new research facilities, sports arenas, and other unique public buildings. This is a pattern recognized in many studies. In Swedish municipalities, it is typically fire stations, town halls, stadiums and other buildings that are one-of-a-kind that are most often hit by cost overrun. Such projects are unique and seldom planned and implemented by a municipality. The project does not need to be new to the world, but it is new to those planning and estimating it. Ordinary housing and building projects are less likely to be affected by cost overrun because they are familiar to planners.

It was observed in the 1930s in the US aircraft industry that the man hours it took to produce a unit of production decreased by a constant as the number of units produced was doubled. The cost of producing a unit decreases as the total production increases, and this is deemed to be due to the experience gained as more units are produced. This has since been confirmed in all parts of the value chain. However, the learning curve eventually flattens out, and so this phenomenon is steeper at the beginning, but later slows down.

An important precondition for the cost–benefits of learning in this way is standardization. Profiting from the learning effect presupposes production in a long series, impossible unless the parts to be assembled are standardized. The production of parts to small tolerances was essential to reduce re-work and make the assembly line an economic means of production. Standardization of products and work processes also speeds up construction processes.

A lack of standardization, the potential for cost reduction through learning, and little or no historical data help to explain why unique, one-of-a-kind constructions are much more prone to increases in cost than, for instance, apartment blocks. Standardized items and work processes are easier to cost as there is data from earlier, similar projects. Specially produced items do not offer the same possibility for updating and improving cost data.

The Longer the Time Between Estimate and Follow Up, the More Changes There Will Be

The longer the time between an estimate and the follow-up, the larger the cost growth tends to become. This was true for Merewitz’s water resources, roads, and rapid transit projects, as well as later infrastructure projects in Nigeria,³⁰ Saudi Arabia,³¹ worldwide,³² Malaysian³³ construction projects, and the hydroelectric power stations and power lines built by the Swedish State Power Board. It has also been found to apply to both military³⁴ and civilian³⁵ research and development projects, and to computer software projects.³⁶

One might suspect that cost growth correlates with the length of time between estimate and follow-up due to unanticipated inflation, but this cannot be the whole explanation as the tendency for costs to rise still remains after figures have been adjusted for inflation.³⁷ In fact, the tendency for cost to rise over time can be even stronger, as unanticipated inflation might have been overestimated in studies where researchers did not have knowledge of price clauses or when payments were made – information that is necessary to estimate the effects of inflation correctly. Thus, the longer the time period between estimate and final cost, the larger the cost growth tends to become, even when figures have been adjusted for inflation.

The Swedish State Power Board had two large hydroelectric power station projects during the 1970s which encountered rather large cost overruns. As larger stations take longer to construct, one could interpret Figure B.4 as proof that larger projects are more prone to overruns. However, the four power stations that showed the highest cost overruns in Figure B.4 were all designed before the first oil crisis in 1973–4. Cost overruns for the other seven, constructed after the first oil crisis, were very modest. Three of them showed a smaller cost overrun of about 10 per cent, and the four others almost no deviation at all.

Figure B.4 Time of construction and cost overrun for 11 hydroelectric power stations

Note: Figures such as 72 stands for the year construction started. In this case 1972. Cost overrun in per percent and real monetary value.

The cost overruns on the two largest stations made some managers in the Power Board express thoughts that their organization was perhaps not that well suited to carry out large-scale projects. However, a closer look at these projects revealed that it was not the size of the projects as such that had caused their cost overrun, but their redesign and the long time it took to implement them. As energy prices increased, the new power stations were redesigned to utilize the water better. Sharpened environmental and construction standards also contributed to cost increases, as the 1970s was a decade when environmental and safety standards became more stringent.

The economics of a hydroelectric power station is determined by, among other things, the percentage of water utilized and the height of the fall. It is not economically justified to make use of all the water when the flow varies a lot during the year and across different years, because it would require larger reservoirs. Therefore, project planners put a value on the water utilized and discount it, in a similar way as is done in discounted cash flow analyses of capital investments, to determine the optimal design of the power plant.

In this case, the value of the water increased due to the increased price of competing energy sources, such as fossil power. This made it justifiable to redesign dams and channels to make better use of available water. This increased the cost and construction time of the plants, but also their profitability. Thus, although the plants become more expensive than anticipated when the board approved them, these changes also made them more profitable.

The two hydroelectric power stations that took longest to construct were commenced in 1970 and 1972 and finished in 1978 and 1979, respectively. The power stations were designed in the late 1960s, which means that quite a lot of work and funds had been spent by the time the oil price increased three to four times in 1973–4, and then doubled once more in 1978–80. Construction on the other projects with cost overruns began in 1974–7, i.e. between these two sharp price increases, and those without any significant cost overruns after the second oil crisis in 1980–1. The high cost overruns of the earliest projects can be explained by the fact that the cost of making changes in the design increases sharply the closer they are to the end of the project. Making changes on the drawing board does not necessarily make a project more costly, but when construction starts changes become increasingly more expensive. What has been spent cannot be recovered and, in addition, changes contribute to lengthening the time of construction and thus interest expenses.

When a project takes many years to implement, there tend to be changes as project managers want to incorporate new technology, or adjust the project to changing market demand and also competitors’ new products. When, for instance, a Swedish heavy truck maker developed a new engine for their heaviest truck, they received the news that their main competitors had developed a slightly stronger engine. They therefore redesigned the engine to have the extra horse power necessary to be able to claim that they offered the strongest engine on the market. Such adjustments in response to competitors and changing market demand will of course lengthen development or construction time and increase the final cost of the project, even though it sometimes can lead to the development of better technical solutions.

Cost overrun has a negative connotation; project managers for the hydroelectric power stations argued that additional investments had increased the capacity of the plants and made them more profitable, and that the projects as implemented could not be compared with the ones approved. That is of course correct, and is almost always the case in projects showing large cost overruns.

What one can claim is that these two power stations should have been re-appraised and a new budget approved before they were redesigned. However, that might have led to discussions on whether and how to make the changes in a project that only became more profitable as the value of the electrical power it would produce increased, which would have caused time delays that would have postponed production. Perhaps this was why it was solved in a more informal way.

A large number of studies have found that the longer the time period between approved budget and follow up, the larger the cost overrun tends to become. The longer the construction period, the larger is the probability that there will be changes in the project; changes add to time and cost partly because accrued expenses cannot be fully recovered, and partly because accrued interest increases.

Deviations in Cost Co-vary with Other Measures of Deviation

The US Government Accountability Office has reviewed government-funded projects for many decades. Cost overruns have decreased since the 1960s but have not gone away, despite stricter methods of control and project management.³⁸ The reasons are stated to be the same as always – projects are still affected by ‘funding instability’ (i.e. funds are not allocated as planned) and ‘design instability’ (i.e. changes in the design of the projects caused by funding instability or other changes external to the project). This lack of continuity and predictability forces project management to make changes in their plans for projects which in the end lead to higher costs.

Such instability is not unique to US government-funded projects. An example is the stopping of the Stockholm bypass project. When the Left won the national election³⁹ in 2014, they halted the implementation of the project for a couple of months. The bypass project had been estimated to cost SEK 28 billion in 2009 monetary value, and stopping construction certainly added to its final cost. Such ‘funding instability’ has also been observed elsewhere, for instance in Iranian public investments⁴⁰ and development projects in the Swedish manufacturing industry.⁴¹

Changes in funding, technical standards, and market demand explain why there is a positive covariance between cost overrun and factors such as construction time, volumes of concrete and steel used, man hours, estimated sales volume and price, product mix, and planned capacity. External and internal changes drive costs.

The covariance between deviations of different kinds can be further illustrated by Table B.1 which shows deviations between plan and outcome for 35 new plants.⁴² The sample is small but is still of interest as it illustrates the covariance between cost overrun and deviations in other measures of outcome.

About a quarter of the projects reported deviations larger than 5 per cent for the variables listed in Table B.1. Five percent has been chosen as companies’ financial protocols typically require project management to go to the board to appropriate new funds when cost exceeds 5–10 per cent of the approved budget. This gives project management a strong incentive to make savings in the budget in order not to have to request more funds. Table B.1 shows that deviations that negatively affected the profitability of a project were more than twice as common as the opposite. That applied not only to estimates of the investment outlay, but also to the length of construction period, technical ambition, plant capacity, operating cost, sales volume, and sales price.

Cost overrun is often associated with deviations in other output measures that indicate changes in plans and the implementation of the project. The project that has been implemented is never exactly the same as that which was originally planned and approved. Still, this survey probably underestimates the true occurrence of deviations, as we can suspect that project managers will tend to play down deviations.

Internal and External Complexity Drives Cost Growth

Major projects can be split up into a number of sub-projects and activities which have to be completed in time so as not to delay other activities. When a time delay large enough to delay another activity happens, it can seldom be cancelled out by an activity being performed in a shorter time than planned. It can propagate and delay other activities, with knock-on effects that add up to a total delay, and thereby higher cost. Richard J. Schonberger⁴³ claims that this is why projects are ‘always’ delivered late.

Assume, for instance, a task that has been estimated to take 100 hours to complete. If those performing the task know it well, they might be able to complete it in 80 hours. However, if they do not know exactly what to do or how to perform the task, the time taken can easily slip to more than 120 hours, 150 hours, or even more if they get stuck or make mistakes. But doing it in less than 80 hours is probably very difficult. There is a limit for how fast a task can be performed, but not for how slowly, which explains why projects with many dependencies are more prone to time and cost overrun. It can also explain positive skewness.

This is not a new observation. Raymond Giguet and G. Morlat⁴⁴ argued in 1952 that projects tend to become more expensive than estimated due to the fact that the events that cause deviations are not independent. Engineers determine the probability of various possible events that may occur and try to minimize the risk of undesirable outcomes. However, whenever an unexpected event occurs, this necessitates adjustments and changes that will push the final cost above the previously determined most probable average final cost.

The number of dependencies, their interrelatedness, and their negotiability is by definition a measure of complexity.⁴⁵ Observe, however, that a large project is not necessarily more complex than a small one. It is the number of dependencies and the amount of slack in time and trade-offs between requirements that define complexity, not the size of a project. These dependencies can be both internal and external to the project.

High-tech projects are especially susceptible to cost growth due to delays in interrelated activities, as high-performance products are more tightly optimized. Thomas K. Glennen, Jr⁴⁶ points out that it can be more difficult to substitute a part in a high-performance product than in an ordinary one, as parts are more tightly optimized in space and functionality, and less standardized.

Using standardized components reduces the risk of having to adjust components to each other, and hence reduces the risk of delays and cost growth. This applies not only to hardware components, but also to software components and suboperations performed by humans. The solutions advanced by Glennen and his contemporaries⁴⁷ were parallel product development and more loosely coupled systems. The latter can be achieved by giving components a standardized interface so that they can be easily substituted and rearranged. This reduces the risk of failure⁴⁸ and is a principle widely used in complex development projects and complex production systems.⁴⁹

When cost increases, project management can often counteract this to some extent by reducing the size of investment, choosing less costly solutions, or outsourcing certain parts of the investment, as was illustrated by the Frövifors case in Chapter 2. This is not possible in a highly regulated technology such as nuclear energy. If the construction site is short of nuts and bolts of the steel quality specified in approved drawings, work has to be postponed until the correct nuts and bolts arrive. Strict enforcement of safety requirements reduces the options for project management to solving upcoming problems and keep costs down by improvising and choosing second-best solutions.

The cost of production can be reduced if there are economies of scale, i.e. if the cost of producing additional units declines as fixed cost is distributed over more units. The nuclear industry has tried this avenue by building still larger plants to counteract cost growth, but it has not worked either in France,⁵⁰ or the US.⁵¹ Operating cost has been relatively unaffected by the increasing size of more recent plants.

Additional safety regulations have made nuclear plants increasingly more complex projects. This, together with increased size, has lengthened construction time, which means more accrued interest. Increased site-specific requirements have also made each new plant a unique project: no two plants are exactly the same. Interrelated factors are many, and many parts are designed for a specific plant, which is why they cannot be substituted with another part filling the same function. This lack of standardization has made it impossible to utilize the learning curve effect to reduce cost. The effects of complexity, size, and construction time have been so great in France and the USA that it has totally offset any economies of scale.

Lower complexity and a higher degree of standardization is one possible explanation as to why the solar and wind farms in Table A.1 have much lower cost growth than nuclear reactors.⁵² A solar or wind farm consists of a large number of similar wind generators or solar panels.

Another category of complex projects are large computer software projects. There are internal dependencies between different pieces of software, and there are dependencies between the programme and its user. Early programmers applied the waterfall model to develop software by going from one stage to another, just like a product on an assembly line. This required the previous stage to have been completed before the next could start, which was inappropriate when developing large software systems. The linear model could cause delays, and unsolved problems and user dependencies could cause problems and re-work later on.

A common problem is that customers often are unable to clearly visualize and specify their requirements. This inability causes an instability which programmers try to solve by involving customers in all phases of the development process.⁵³ Furthermore, major software programmes are composed by a very large number of subcomponents and diverse contributors. The solution to this is to create a modular product architecture that allows several project teams to work in parallel, to avoid blind alleys by attending to problems as fast as they appear through iterative work cycles, and by continuously testing smaller modules and sub-parts of the larger programme in a user environment. A variety of such practices has been developed and goes under the term ‘agile programming’, often involving measures such as regular and frequent meetings to adapt activities and different parts of the program to each other, and solving problems immediately when they appear, and has proven to reduce both cost and time to market.

Advances in Knowledge Co-vary with Cost Growth

As we saw in Table A.2, research and development projects have a tendency to increase in time and cost, not least for new military weapon systems, which is why US researchers begun to study deviations in military aircraft and missile projects in the 1950s.

One factor which has contributed greatly to making many military aircraft projects more expensive than anticipated, both back in the 1950s and today, has been when fewer planes have been manufactured than originally planned. As the development cost is a large part of the total system cost, reducing the number of aircraft produced significantly increases the cost per aircraft.

Another factor is inflation. As development times are long for new aircraft projects, high inflation can substantially add to development cost. However, after correcting the figures for the number of aircraft manufactured and inflation, the fact remains that development projects are more prone to cost growth than construction projects.⁵⁴ This can be seen by comparing Tables A.1 and A.2. Observe that the figures for military aircraft projects in Table A.2 have been adjusted for inflation and the number of aircraft produced. More recent studies of military weapon development⁵⁵ have shown that cost overrun is still common.

A third result from early studies of military development projects was that it became clear that cost growth correlates with the degree of technical advance.⁵⁶ The larger the advance in technical knowledge, the more the cost tends to increase. This is a maxim that has subsequently been confirmed to apply to development projects in the pharmaceutical industry⁵⁷ and in the Swedish manufacturing industry.⁵⁸

The factors causing cost growth in development projects are partly the same as in construction projects. The estimator takes his or her point of departure from the project design available at the moment of the estimate. They estimate the cost of the design and add on a lump sum for contingencies. Then, as the development project progresses, unforeseen technical problems may appear which make it very difficult to achieve the design criteria set up, project requirements may have to be altered as market demand changes, or the design may have to be updated to incorporate new technology.

Such changes lead inevitably to changes in the original design, leading to a new estimate based on the new design. In research and development projects this is natural, as the aim of such a project is to develop new knowledge and specify something. This is very different from a construction project, which takes its point of departure in known technology and thus can be specified before it is implemented.

Burton H. Klein⁵⁹ compares building a bridge with a military aircraft project. If the bridge is well designed, the probability is very high that it can be built without making changes in the original design during construction. Military aircraft projects, on the other hand, are seldom implemented as originally planned. Regardless of how much attention had been given to the original design in the cases he studied, there were always surprises later on in the project, and what started with one configuration came out of the process with a different engine, electronic systems, airframe, or tactical role. The essential difference between bridges and aircraft, Klein explains, is that the construction of the former is a question of making best use of existing knowledge while the latter requires a lot of learning.

What Klein described is the distinction between implementation and development projects – that is, projects which can be specified in advance and projects for which the goal partly is to specify something new. Let us assume that we are going to build a new apartment block, a routine project. It should then be possible to estimate the cost of such a project relatively accurately as there are many other similar projects from which cost data can be obtained. We can, if we are thorough, calculate the amount of building material and labour hours needed to build such a block. It is mostly a matter of how much work we decide to spend on planning prior to the construction. If we spend enough resources on planning a standard apartment block then we should be able to estimate the final cost well.

To study whether this was the case also for development projects, Klein used 71 estimates of 22 military aircraft and missile projects. His study found that for some of the projects cost increases were discovered early in the development process, for others late. This differed from project to project, which is why the risk of cost growth in research and development projects cannot be totally eliminated by spending more resources on planning ex ante in an attempt to develop better cost estimates.

Thus, estimating the final cost is partly a question of how much work and money should be spent on project planning prior to implementing the investment. If large resources are invested prior to selecting the final design, then project management has to make early commitments to certain technical solutions. But if they do so, they lose the flexibility that a gradual commitment to technical solutions can offer, perhaps necessitating expensive redesign and causing the final cost of the project to become higher than if certain decisions had been postponed. Some project managers argue that it is easier to keep the final cost low if the project is not specified in too much detail at an early stage because of the high costs of going back and re-specifying should the chosen design not work properly.

Thomas Marschak⁶⁰ asked himself why development times for new weapons systems were shorter during World War II than after. Having studied a number of projects, he concluded that the short development time during the war was because project management did not commit themselves to strict design specifications at an early stage, but were allowed to experiment. If they developed a new radar, or a component in a radar or in an engine, it could be immediately installed in an aircraft and tested in action. Sometimes it worked, sometimes it did not. New design ideas were tested in action before excessive resources had been spent on developing them.

In peacetime conditions, this method of development came into conflict with both safety requirements, and civilian and military authorities’ need to control their fixed budget. This suggests the conflict between the roles of entrepreneur and controller – the one allocating funds wants to know exactly what the project will cost and how funding will be used, while the entrepreneur relies on their ability to meet new challenges and solve problems as they arise.

Cost Overruns Vary by Organization

The planning process matters.

Figure B.5 shows the deviation in estimated investment outlay for the 20 largest investments in Table B.1 for which data on first estimate, budget, and follow-up exist. As can be seen, many of the projects that encountered cost overruns also showed cost increases between the first estimate and the approved budget.

Figure B.5 Cost overrun before and after the formal decision to invest for some major investments in industrial plants

One reason why projects are affected by cost overrun is, as was pointed out earlier, that the decision to go ahead is taken at an early stage when important technical design decisions have not yet been made and any cost estimates are therefore very uncertain. If that was the sole reason, one would expect that the projects in Figure B.5 would show large deviations either between the first estimate and the approved budget, or between the approved budget and follow up. As we can see, this is not always the case. Cost estimates for one of the plants have decreased due to that the firm re-evaluating the market and the strategy for the plant. Moreover, a relatively large number of the projects which encountered large cost overruns did so both between the first estimate and budget and between budget and follow up. This indicates that there must have been changes made in the projects not only after the decision to invest, but also prior to that decision, indicating that the project management team was not in control of the project. The general advice in such situations is not to make additional changes in the project to cut costs, but to go back, specify, and freeze a new design, and to make a new detailed cost estimate and market analysis based on that new design.⁶⁴ This is needed to avoid further unanticipated changes that would otherwise drive up costs.

Another pattern sometimes appears when plotting public sector projects. Swedish local governments previously had to report their planned volume of investment during the forthcoming five years to Statistic Sweden. This created a pattern in which projects tended to increase either before the final decision to invest or after. The first hurdle was for a project to be included in the long-term plan, and to be included a cost estimate was required. Then it might have to wait for many years to be properly planned and appraised before it was implemented, as other competing projects were given higher priority. Sometimes the delay was so long that the plan became obsolete and the project was abandoned. This caused public sector projects to increase either between first estimate and approved budget, or between approved budget and follow up.

An extreme example of cost growth in the long-term plan is provided by a project for a new ice rink in Malmö, a town in the south of Sweden opposite Copenhagen. After some discussion, the town council decided to include building a new changing-room at one of their ice rinks in their long-term plan, but after many years of planning and discussions it was found that building a completely new ice rink was a better solution. The project grew from a new changing-room at an open air ice rink to a new indoor ice rink.⁶⁵ This example is extreme, but it illustrates how the planning process can sometimes take an unanticipated direction. The first hurdle was to have the project included in the long-term plan, the second was to have the project designed the way one really wanted to see it implemented.

This tendency for municipal projects to increase either before or after their final approval is also visible in major Swedish road and railroad projects⁶⁷. Figure B.6 shows deviations in cost both when the projects have been in the long-term plan, and after a budget for the project has been approved and the project implemented. Budget here stands for the budget approved before procurement. The project that encountered a cost overrun of 112 per cent between approved budget and follow up was a road project that was upgraded from being an expressway to a freeway after construction work had commenced. This project was, however, an exception. The largest deviations appeared when the projects in the long-term plan were waiting to be implemented. Long waiting times, inflation, and increased construction standards could have contributed to cost growth.

Figure B.6 Deviation in cost estimates before and after the final decision to invest for some road and railroad projects⁶⁶

The private sector abolished the long-term plans that we still find in the public sector a long time ago. The market changes so fast in a market-based economy that it is not feasible to make long-term investment plans. Comparing public and private sector projects, one cannot disregard the fact that public sector projects sometimes have to wait for five or ten years, or even longer, to be implemented. Many factors can make an old estimate outdated, including inflation, price changes, new standards, requirements, technical development, and market demand – changes that might make radical changes in the original project plan necessary.

Cost Overrun and Decision Biases

Cost Overrun and Lack of Information and Information Asymmetry

Investments have to satisfy the needs and requirements of many actors within and outside the organization to be approved and receive funding. Marketing people have the best knowledge of the market and engineers of the technical choices made, but to have their project approved they must be able to sell their idea, they must convince top management that their investment is of greater importance to the company than competing projects in search of funding.

One problem in this selling and commitment process is that those designing the projects have more and better information about project details, making it difficult for those approving the funding to question technical choices. We refer to this as an information asymmetry between those selling and those approving the project, and many (if not most) of the explanations for intentional underestimation in the literature are based on the presence of information asymmetry between two or more actors.

Raymond Giguet and G. Morlat,⁴¹ who postulated that the events causing cost overruns are not independent (see Appendix B), also postulated that when two investment alternatives filling the same purpose are competing for limited funds, the one underestimating the investment outlay stands a better chance of getting funded than the one that correctly estimates the investment outlay. From this, they concluded that underestimated investments will be over-represented in the capital budget, and the total investment budget thereby exceeded.

Based on statistics, one can also postulate that the more bids there are, the lower the lowest bid will be, and the higher the highest bid will be. The one who wins a popular auction will pay too much, and the one winning a much-wanted assignment tends to have underestimated the cost of delivering that assignment.

Not only does the lowest offer stand a better chance of being funded, there is also an incentive to accept the lowest offer when projects compete for discretionary grants from a limited budget. This has been pointed out in studies of both military⁴² and civilian public sector procurement.⁴³ It is especially a problem in public organizations as these are often obliged to accept the lowest offer. They have no choice unless they can argue that a higher offer fulfils their specifications better, meaning that they have to specify quality requirements clearly.

Underbidding is not a new problem. A 100 years ago it was perceived as a major problem causing unfair competition, especially in the new electro-mechanical engineering industry. Costing was more difficult in this new industry, which appeared in the 1880s, as products were assembled by parts manufactured at different manufacturing units. As is usually the case in a new industry, many new and relatively small firms were soon competing for contracts, and many of these small firms did not have the knowledge to properly cost their products. As the procurers tended to choose the lowest offer, many of the firms that had been too optimistic went bankrupt, which caused problems for both the producer and the buyer. Costing practices needed to be improved, and that was done in the USA by creating calculating cartels which collected data to help member firms to make more accurate cost calculations, and by standardizing cost accounting.⁴⁴ Standardization meant first standardizing the terminology, and later the principles of cost accounting.⁴⁵ It was a process that took several decades and, in the depression, the unfair competition argument resurfaced in the National Industrial Recovery Act passed in 1933. As a part of President Roosevelt’s New Deal, this Act prescribed ‘fair competition’ as meaning that firms were forbidden to set prices so low that competitors went into bankruptcy.⁴⁶

Many bidders with poor information about the real cost no doubt create a risk of underbidding and cost overrun, but Giguet and Morlat’s conclusion that underbidding would also cause the total investment budget to exceed its limits does not receive conclusive support in empirical studies. It seems to be more common⁴⁷ that the investment budget is not used up as there are always projects where implementation will be postponed or never happen. We are overoptimistic and plan to achieve more during the coming year than we actually manage to achieve. It has been reported that some companies are aware of this and therefore accept more projects than the investment budget allows so as not to miss out on lucrative investment opportunities.

One question which has been discussed in the literature⁴⁸ is whether underestimation is unintentional or the result of a deliberate act to make a project look more appealing. If the estimator has developed an optimistic bias then the underestimate is unintentional; if the estimator realizes that the estimate is unrealistically low but still submits it to get the investment approved, then it is a deliberate act. The requestor or project planner may be convinced of the merits of the investment but cannot prove it in the way the company prescribes in its investment processes, or the investment may be in the interest of the planner. The first alternative has already been discussed, so now we turn to investment costs that are deliberately underestimated.

Case studies from both the public⁴⁹ and the private sectors⁵⁰ give examples of project costs that have been deliberately underestimated in order to have them investigated and even approved. An example was given in the second section of Chapter 4. The managing director of Frövifors Bruk AB said the new mill would cost 300 million to gain approval to do a feasibility study, well aware that the board would probably find it wiser to build a larger and more expensive machine to better utilize economies of scale later on. Because of this information asymmetry ‘[y]ou have to go about things gradually so that you don’t get a negative answer’, as the MD of Frövifors put it in Chapter 4.

The use of (or, more correctly, the misuse of) cost-plus contracts, which guarantee the contractor full compensation for costs plus a certain markup in military development projects, has long been criticized⁵¹ for not giving contractors incentives to keep costs down. Tougher competition has encouraged contractors to put in lower offers for cost-plus contracts as the benefit is far larger than the cost of not being able to meet the contract. As public authorities are forced to choose the lowest offer due to the rules for public procurement, and sometimes also lack the competence to assess offers correctly, they risk giving the contract to a contractor who consciously or unconsciously has put in a bid that is too low. Furthermore, the penalties for not being able to meet technical requirements are higher than the penalties for not meeting time and cost requirements. Hence, problems in meeting technical requirements are taken as time and cost overrun.⁵² This causes military authorities to reduce the amount of development work done on a cost-plus basis, and to supplement contracts with incentives to hold costs down.

To curb high and rising costs, President Reagan appointed David Packard to chair a commission on US government military procurement. The Packard Commission⁵³ concluded that the problems with cost growth, delays, and performance shortfalls remained as documented in earlier studies, and recommended changes in the acquisition process and in the contracts used. However, this does not seem to have significantly reduced cost overrun in defence contracts. A study of 269 defence acquisition contracts during the period from 1988 to 1995 did not show improvements, and in fact Air Force contracts worsened.⁵⁴

Developing new military technologies carries substantial technological risks, and Kenneth Arrow⁵⁵ has argued that the primary reason for using cost-plus contracts in weapon development and acquisition has not been to reduce cost growth but to allocate risk between industry and government. Developing new military technology that government deems necessary is associated with such high risk and uncertainty that industry needs to share these risks with government. However, this type of contract has also allowed certain defence industry firms to take advantage of the situation by ignoring the financial risk, as they have a guarantee preventing bankruptcy. An instrument applied by government to reduce the risk firms have to take when developing new technology has been exploited by some firms for profit.

In civilian projects, there might also be an incentive for sellers to underestimate the real cost. However, cost can also increase when cost-plus contracts are not used if the seller is more knowledgeable about the project than the buyer. Construction companies sometimes accept deals that do not look profitable on paper because they foresee that the buyer will probably need to ask for additional work or changes in the agreed plan and therefore can be expected to want to renegotiate the deal later on.⁵⁶ The construction company’s information and knowledge advantage allows them to sign unrealistic plans because they know that profitable additional work will be needed as the other party learns more about their own plans. Therefore, the procurer needs to match the seller in competence and knowledge, otherwise the seller might take advantage of the situation. If the procurer lacks the necessary competence, it can be better to let a consultancy firm evaluate offers, as well as handle procurement and contracts.

One might expect lenders to be aware when there is information asymmetry between procurers and sellers, but that is not the case. Even when a project is affected by very large cost overruns, lenders almost never make a loss. If it is a public project or company, then the state, and ultimately the taxpayers, take the risk. If it is a private company, then lenders look at the company’s ability to pay back the loan. The important thing is the borrowers’ ability to repay their loan, not their ability to avoid cost increases. Lenders do not usually evaluate projects; it is easier and cheaper for them to evaluate companies and management.

Information asymmetry exists not only between those proposing and approving a project, but also within organizations. However, removing this asymmetry is not always the best solution because cooperation is built on trust. An approved cost estimate is an agreement between two parties, and a commitment on behalf of those that have requested funding and, in particular, the project manager responsible for implementing the decision, as they have all signed the funding request and by doing so confirmed their belief in it. If those approving the funding question what is in the request too closely, this can be interpreted as questioning the qualifications and loyalty of those making the request. The higher the number of managers who have scrutinized and approved a request, the more difficult it becomes to stop it. Top management cannot reject projects approved lower down the organization without them losing face.⁵⁷ Nor is it is always in the best interest of the company to show distrust openly, as that can damage the trust needed to cooperate in the future.

A questionnaire⁵⁸ sent to Fortune 500 companies showed that management were aware that some proposers were more optimistic than others, and adjusted their profitability estimates downwards to compensate for this optimism. Managers accept projects even when they know that the true cost is underestimated, as they do not want to disbelieve committed project proposers. Company management is aware that investment requests are often too optimistic and learn to compensate for the optimistic biases of requestors.

It is also a widely held view in industry⁵⁹ that it is better to approve a tight budget and be prepared to provide additional funding if needed, as all budgets tend to be used up. ‘Wasn’t cost overrun larger? I expected twice as much’, as the CEO of ASSI is said to have exclaimed when he got the follow-up review of the Frövifors investment. By approving a budget he thought was too small, he got more out of project management and ensured that funds were put to the most efficient use. Not surprisingly, it is extremely unusual for a project to be implemented below cost and unused funding returned; project management always seem to find a way to spend what has been approved.

This can also be an argument for keeping the estimates low in the planning process, and for being prepared for cost to rise gradually, as low-cost estimates put a pressure on project management to find economic solutions. Whether this is effective is another issue. Requesters who know that those responsible for approving requests will always make cuts in the budget can be expected to sneak in extra funds to counteract such cuts. Budgeting is called a game for good reasons.

Cost Overrun and Path Dependency

All projects need finance to be implemented, which means that there are always at least two processes involved in project planning. One is solving technical and economic challenges, and the other is the process through which finance is secured and approved. These two processes are often interconnected, as finance puts restrictions on the size of the project, and the design might have to be altered to attract support and finance. During the process, the number of alternatives is usually reduced to one, and consensus around a specific design and project is gradually built up through commitments from all the interested parties. These commitments are based not only on rational choices but also on emotions. In fact, emotions always come first. Rational choices are reconstructions, efforts to test and prove visions which are sometimes associated with strong emotions. People engaging in getting projects implemented, or indeed stopped, can become very emotionally involved, and assessments and decisions become a mixture of emotions and reason.

The formal investment decision is seldom between different projects. The alternatives are gradually discarded until only one remains. Many different design alternatives, technical solutions, sizes, and localizations can be studied; however, cost estimates are still undetailed and based on comparisons with earlier investments and rough estimates. Only when the board approve the invitation of tenders can reliable estimates be made, and then the choice is not between different localization options but whether to proceed with the alternative chosen or not. The early decisions and choices that are necessary to push the project idea towards implementation can create a path dependency by imposing restrictions on future decisions.

Furthermore, projects are learning processes. We shall not delve into the many theories of learning, but, for learning to take place, there must be some motivation or goal and some form of negative feedback that makes us aware when there is something in our assumptions or images that is not as we assumed. To learn something new, we need to be made aware of the difference between what we think we know and what we do not know. An important supplier of such feedback is the control systems that exist in organizations. They force estimators to learn more about the project before they are allowed to implement it.

Another important aspect of learning is that it has to be focused. One cannot investigate and solve all problems at the same time. To move the project forward, it is necessary to focus on one or a few problems at a time. This typically gives planning processes a sequential character. Richard Cyert and James March termed it ‘sequential attention to goals. … Organizations resolve conflict among goals, in part, by attending to different goals at different times’⁶⁰.

Trying to solve all problems simultaneously would make the level of uncertainty unmanageable. Instead, one has to focus on solving one or a few problems at a time and assume that other factors will remain constant. The same applies when there are multiple and possibly conflicting interpretations of the information available. One has to proceed in steps to keep perceived uncertainty at a manageable level. Keeping some goals or solutions constant while experimenting with others also means that some solutions will be developed before others, and so project planning becomes an iterative process. All this – early decisions, commitments, learning, and many other factors reducing alternatives – can create a path dependency that will drive cost.

Path dependency connects our next step with where we come from.⁶¹ It explains how past decisions limit the choice of future possible decisions. Such path dependence can create cost growth both before and after the decision to invest. We have so far focused on commitments to early decisions created by an interconnection between definition and funding processes, and the reduction of alternatives in the definition process. Path dependence can also be inherent in the technology, be a consequence of specialization, driven by hopes of re-couping sunk cost, and individual commitments, causes that are not always easy to distinguish from each other.

A concept related to path dependence is escalating commitments,⁶² often expressed as the tendency to ‘throw good money after bad’, particularly when this action cannot be ascribed to impartial rational reasoning. The reasons advanced for escalating commitments are much the same as for path dependence, and vary from expectations of good profit, to social pressure from peers and the role, and a hope to recover investment already made, i.e. sunk cost. The term ‘escalating commitment’ is not an explanation but a symptom. The explanations given vary in the relatively rich literature around this topic. We have already described how the definition and impetus processes are interlinked and that those advocating the project must convince those allocating resources of the merit of the project, and get their commitment to the cause. This process of committing support starts right at the beginning of the project, and is one of several explanations for escalating commitment. A related term is ‘entrapment’, which can be defined as instances of escalating commitment leading to an ineffective course of action.⁶³ Escalating commitment can lead to both effective and ineffective courses of action.

Starting with technology as a cause of path dependence, an often cited example is the QWERTY keyboard layout which, due to its early dominance in the market, has been able to retain its dominance in spite of better layouts. The more people using it the more difficult it becomes to switch to another layout, a fact well known in the IT industry, where the number of users creates an increasing rate of return. The concept is not new. Harry Turner Newcomb,⁶⁴ for instance, showed in 1898 that ‘the law of increasing returns’ in industries with low operating costs which did not need to increase their investments or increase them in proportion to increases in production, such as the railway industry, leads to the firm with the largest network out-competing other firms. But W. Brian Arthur⁶⁵ presented the first formal theory showing how increasing rate of return can lead to ‘lock-in due to historical events’ to a less efficient technology. Characteristic here is that there are several possible future paths, that the outcome is unpredictable, that the choices made limit future possible choices, and that the path dependency can lead to inefficient ends.

Specialization is another factor that places a firm on a certain path, or commits an individual to a certain profession. Gary Becker⁶⁶ distinguishes between general and firm-specific human capital. The former is applicable to firms in general, while the latter is of more value to a specific firm in creating an incentive to remain in the organization. Managers and workers with more specialized, firm-specific skills are of lower value outside the firm, which is why they have fewer options to leave the firm. This could make them more prone to continue investing in failing courses of action.⁶⁷

When it comes to investments, this is true also in a broader sense. Competing firms can be expected to have almost the same information about the market. What makes an investment more profitable for one firm over another is that it has capabilities, i.e. firm-specific resources, better suited to making it profitable. Firms have to specialize and invest in developing capabilities further in areas where they have a competitive advantage, which often implies assets that are so firm-specific that they cannot be traded. Thus, specialization locks the firm into a certain path of investment, in the same way that specialized employees become locked into their firm. This also means that the first investment is the most important. Later investment utilizes earlier investment.

To the employee, firm-specific skills become sunk cost when leaving the firm. Sunk cost is gone and irrelevant, and cannot be recovered. It should not influence future decisions, but can in fact do so due to psychological factors. The gambler makes yet another bet to recover earlier losses, investors make additional investments in the hope of recovering earlier investment, or a bank gives additional loans to recover earlier loans. Hopes of recovering losses can lock an investor into path dependence.

Sunk cost should not matter, but financial obligations can make sunk cost relevant. Investments can be divided into basic and additional investment. The basic investment can be a new plant or production unit. Basic investments can be very expensive and may be partly financed by borrowed capital which has to be repaid. The company therefore becomes locked in to making the basic investment profitable. This forces the company onto a path of making additional investments to be able to pay rents and loans back. The basic investment has several names: in the vehicle and software industry, for instance, it is called a platform. A platform is developed upon which several car models, or many customer software projects, can be based. The principle is the same. A platform is not profitable if it is used only to develop one car model or software for one customer one time only. It has to be made profitable by additional investments.

Sunk cost can also be used intentionally to gain additional funding and investments. An extreme example is the Peruvian regional director referred to by Albert O. Hirschman⁶⁸ who, due to lack of funding, used up all his funds to build bridges, which were useless if, later on, funding was not also appropriated to build roads connecting the bridges. Similarly, there is the tale of a regional director in the south of Sweden who spent all his funding on secondary roads, knowing that this would give him strong arguments for extra funding to build highways catering for long-distance transport. This has been claimed to have been behind the Swedish Road Administration developing a fairer allocation system in the 1960s based on cost–benefit analysis.

The world of ideas that exist in organizations can also create path dependence. Gothenburg, the second largest city in Sweden, decided to turn its railway station into a drive-through station by constructing an underground railway loop with two new stations. In the 1980s, a local politician had proposed an underground line connecting three local transport junctions which was considered too expensive, but it later reappeared as a railway line project.⁶⁹ However, it turned out to be too difficult and costly to build a whole railway station underneath one of the three local traffic junctions, due partly to space but also to the blue clay substrata underneath the junction, so the station was moved eastwards where it could be constructed in rock. This then made it necessary to scrap another junction, whereby the new loop did not serve the purpose the original underground line was supposed to fill. Still, this did not stop the project. Some project ideas are the result of a search process and others an idea imported from outside, but many are just based on ideas that have been around for a long time. Such ideas are found in both public organizations and private firms, and can create their own path dependency.

Individual actors may also be locked into a course of action. An ambitious entrepreneurial managing director of a public utility saw an opportunity to reverse the downward economic development in his region by developing a biomass energy industry system. He was not alone in promoting this venture, but became identified with it as he promoted it widely in the media. However, as project planning proceeded and more became known about the risks involved, he became gradually more hesitant, and was already doubtful about the prospects for the project when he made his presentation to the board to get the go ahead:

His fears came true, and, on top of the economics, serious technical problems emerged. The project was shut down, and the entrepreneurial managing director had to shoulder much of the blame as he had become personally associated with the venture through the media. It was seen as his project.

We might wonder why no one hits the brake before a seemingly impossible business idea crashes. Maybe it is not always easy to stop momentum when an idea has become widely accepted as truth, and, as in this case, many people have attached great hopes to the venture. As the highly engaged entrepreneur, our managing director had built up expectations. When he started to doubt his earlier promises, he did not want to make those that had believed in him disappointed, which is why he had to go through with the project and try to make it succeed.

Key actors can become locked in to the promises they make in order to get their project implemented. Even if they realize that the assumptions they have made do not hold, it is often difficult and painful to re-evaluate these assumptions, especially if they were the ones selling the assumptions in the first place. It is easier to hope that they are right and that everything will solve itself.

We started by explaining how emotions, visibility, and rigid decision processes can lock decision makers and other actors in to early estimates and commitments, and make it difficult to revalue old ones and make objective assessments of new information. An estimate needs approval in order to develop an idea further. In this respect, there is no difference between the private and public sectors but, as has been pointed out, it can be more difficult to explain cost growth in the public sector as the requirement for transparency is higher. Large sunk cost and many actors committed to a project can make it more difficult to terminate the project if the cost estimates increases. However, it is not only the organizational context that creates commitments and rigidity. Key actors can become locked in to commitments that they made earlier in the process. Those that have been working with the project for a long time and have put their name on the figures become committed to achieving the figures promised. Personal commitments are built up and can by themselves become a force that helps overcome doubts and drive plans forward, making it difficult to stop plans once they have gathered enough momentum. All these factors mean that plans can be approved even when they should not have been.