Stochastic regression modelling of cost and duration overrun of construction projects implemented in Kuwait

Construction cost and duration of construction projects are affected by various variables at varying degrees during the construction stage which is although considered to be only a part of the life-cycle of a construction projects' cost and duration, it is increasingly important for several reasons. The construction stage period demands the focus of attention of all the key participants in the construction process because it is during this period that the greatest part of the resources for a project is irreversibly committed. Furthermore, construction cost and duration are the basis for evaluating the success of a project and the efficiency of the project organisation. Researchers such as (Nkado, 1994, Kaming, 1997, EI-Mashaleh & Chasey, 1999, and Jovanovic, 1999) argued that the identification and prioritisation of cost and duration influencing factors are useful basis for modelling and predicting construction cost and duration variations. However, there is no consensus in the literature on the identification of those variables that affect stipulated, planned or achieved construction cost and duration of construction projects because researchers have largely viewed the subject from diverse perspectives. The present research study investigates the significant factors influencing construction cost and duration overrun during the construction stage of projects implemented in Kuwait by conducting a questionnaire survey which was designed incorporating 80 potential causes of construction cost and duration overruns grouped into nine major categories. A questionnaire manuscript was structured and mail distributed to a randomly selected set of 432 construction professionals in the Kuwaiti construction market, 203 completed questionnaires were received back achieving a 47% response rate for the conducted questionnaire survey. Participants were asked to rate, according to their experience and perception, the level of sensitivity of each construction cost and duration overrun variable included in the questionnaire. The rated cost and duration overrun variables were ranked according to their measured severity indices; and the influential variables to construction cost and duration overruns were identified and their relationships were analysed. The conducted statistical analysis of the data obtained from the questionnaire responses revealed that 27 contractor-related overrun factors, two client-related overrun factors, and two design-related overrun factors were viewed by the respondents to be highly important, relevant, and significant to cause construction cost and duration variations. Regression analysis techniques were applied to develop regression models that can predict the percentage cost and duration overrun of construction projects based on the identified significant cost and duration overrun variables revealed from the data analysis of the conducted questionnaire survey. The developed cost and duration overrun regression models were tested and validated against a data sample of 38 completed construction project cases that were not used during the modelling stage, it was found that the actual and model predicted values of the percentage cost and duration overrun have achieved a high correlation coefficient value of R = 0.95 indicating good fit by the developed regression models. Statistical analysis of the error measurement associated with the percentage cost and duration overrun predictions indicate that there is an average error of 0.87% in the %C.O.R. predictions, and 1.02% in the %D.O.R. predictions. The calculated coefficients of variation values for the developed %C.O.R. and %D.O.R. regression models were found to be approximately 16%, and 18% respectively. Stochastic modelling based on the application of Monte Carlo simulation techniques was conducted to predict the best-fitted-probability-distributions for all possible estimated outputs of the developed cost and duration overruns regression models. The conducted fitting process has revealed that the' Triangular' distribution was found to be the best-fit for the estimated %D.O.R. values, whereas the' LogNormal' distribution was found to be the best-fit for the estimated %C.O.R. values. The best-fittedprobability- distribution for the actual observed %D.O.R. values obtained from the testing data sample were found to be a 'Triangular' distribution indicating a perfect match with the estimated outputs of the %D.O.R. On the other hand, the' Weibull' and very closely the' LogNormal' distributions were found to be the best-fitted for the actual observed %C.O.R. values indicating a perfect match with the estimated %C.O.R. values obtained from the developed regression model as a result of the conducted Monte Carlo simulation.