La dialéctica del deseo

Quantitative analytical techniques, which analyse cause-effect relationships, are generally grouped as Mathematical models (An et al., 2007). Models are reductions of reality replicating an intricate system using variables within those systems (Ahiagu Dugbai, et al., 2014). Mathematical models have however been scantly used in the literature to infer causality in

relation to project cost overruns. The techniques which have being used to analyse cost overrun in a limited number of older and more contemporary studies include: Linear modelling techniques such as regression modelling, Networking and data mining techniques such as Artificial Neural Networks, Heuristics based models such as Case Based Reasoning /Reference Class Forecasting, Stochastic techniques such as Monte-Carlo simulations; and Logic based methods such as Binary Logistic modelling and Fuzzy Logic (Trost and Oberlender 2003; Attala and Hegazy 2003; Love et al., 2014; Ahiagu Dugbai et al., 2014; Lee and Kim 2015 and El-Kholy 2015):

3.6.2.1 Regression Models

Approaches to parametric cost estimation based on statistics using linear regression analysis have been developed since the 1970s (An et al., 2007). As such an empirical estimation of cost overrun based on cost drivers requires statistical functions (Hendrickson 1998). Cost overrun analysis using regression analysis generates the best fit parameter values for the cost overrun function based on which empirical cost inference can be made. Older studies such as Trost and Oberlender (2003) as well as Attala and Hegazy (2003) have used linear modelling techniques, based on regression analysis, to analyse cause-effect relationships in explaining recorded cost overruns in projects, and further tested the validity of these models with respect to their use in decision making for future projects, at specified levels of confidence. A more recent study by El-Kholy (2015) generated a regression based model, while comparing its predictive capacity to a Case Based Reasoning (CBR) model for similar data sets derived from 30 projects. The outcome of the study showed that the regression modelling had higher levels of accuracy to predict potential cost overrun in projects.

3.6.2.2 Case Based Reasoning Models

Case-based reasoning (CBR) models have high similarity to expert systems, being described as the foundations of artificial intelligence (An et al., 2007). CBR derives its logic from a rule-based reasoning dependent on experience or memory with adaptations made for deviations from the rule (Chen and Burrell, 2001). Kim (2004) stated that the CBR approach is similar to expert judgments, that rely on the use of experience to solve problems, following the logical steps: outline of the key attributes describing the current problem; identification of similarity of attributes in previous past problems; and prediction within the interface of similarity with subjective correlative adjustment. A CBR system is thus heavily dependent on structure and content of the input case base, which have to

be adequately indexed, organised and adapted for new cost estimating queries El-Kholy (2015).

Although the use of CBR systems have been reported as having lower predictive capacities, they are very useful in decision making and feasibility studies, as such models learn from both past project success and failures in solving new problems, by adopting/adapting solutions that were used to solve old problems (Kolodner, 1992). El- Kholy (2015) applied a CBR model to predict the likely cost overrun given the degree of similarity of the project characteristics to 20 past projects, using a similarity value of attribute, dichotomously assigned one or zero. El-Kholy (2015) used this method to analyse cost overrun factors, whose presence as part of a future project, is indicative of a potential to result in a similar degree of cost overrun, useful in reference class forecasting and decision making to minimise cost overruns for future projects.

3.6.2.3 Artificial Neural Networks

Artificial neural networks are expert non-parametric statistical estimators. The neural network model, rooted in artificial intelligence, has a "brain-like" structure, which can be used to make intuitive decisions similar to the human brain and find complex patterns within the data that often elude conventional analytical techniques (Turouchy et al., 2001). The potential for using artificial neural network relies on the development of genetic algorithms from project data sets, through a process of network training designed to learn and identify trends and relationships in the project input data (Al-Tabtabai et al., 1999). More data input thus optimizes the accuracy of the network model, as outliers and erroneous data are identified and selectively filtered out (Al-Tabtabai et al., 1999; Turouchy et al., 2001). The ability of the neural network to perform sensitivity analysis of a new input parameter to the computer database is an added advantage as it establishes the relative significance of each input parameter within the model, and how it affects the outcome variable (Turouchy et al., 2001).

Ahiagu Dugbai et al. (2014) used data mining techniques based on artificial neural networks, to analyse the complexity of non-linear interactions amongst quantitative project variables such as compensation events, project duration, as well as qualitative information on tendering method, location, project type, fluctuation measure and project’s delivery partner. All analysed projects were completed at cost of between

£1,000 and £14 million, with a maximum duration of 22 months. Sensitivity analysis was further carried out to identify significant cost drivers in a selective elimination process, necessary to optimise the predictive capacity of the model. The least significant input factor identified was the location of the project, while the choice of the project’s delivery partner, was inferred by Ahiagu Dugbai (2014), as majorly influencing the ultimate extent to which the projects ran over budget.

3.6.2.4 Probabilistic Based Methods: Monte Carlo Simulations and Analytic Methods Monte Carlo simulations have also been carried out to analyse the probabilistic effect of specific cost drivers on cost overrun, based on the data distribution of the analysed projects. The core focus of probabilistic techniques is on the quantification of the variability of construction cost overrun based on inherent risks and uncertainties associated with a project (Zwaing, 2014). This is necessary to gain sufficient knowledge about the distribution of the cost parameter of interest. Typically, Lee and Kim (2015) analysed change orders issued during the construction period, due to changes in scope made by the client and changes made due to design errors, which lead to significant cost overruns. This study investigated 9028 change orders from 237 projects that were completed between 2005 and 2011 in South Korea. It was shown that an average of 4.57 of total project cost, was induced as cost overruns solely due to change orders. Love et al. (2014) developed a probabilistic Log-Logistic distribution of cost overruns for 49 road projects (new roads including upgrades and elevated highways) in relation to rework occasioned by errors and omissions in contract documentation, leading to cost overruns. This was used to develop best fit statistical distributions so that probabilities of occurrence of cost overruns was determined, and used to provide a basis to assess the adequacy of the construction cost contingency bench marked, as not less than 13.55%.

It is thus self-evident, that the findings from this class of technical studies on cost overruns which rely on modelling to understand cause-effect relationship, have a robust and methodologically valid underpinning, rooted in the specifics of project data, with a direct applicability useful in monitoring and reassessing future projects, prior to initiation, as a basis of subverting cost overruns. The use of modelling techniques, thus provide logical and repeatable relationships between independent cost overrun drivers and cost overrun, for practical purposes of risk analysis and management by clients.