Construction projects are complex as they include many activities which influence and interact with each other at different stages. The impact of design phase undiscovered rework on construction phase quality has been hypothesized as influential in project dynamics by many. However few empirical studies have measured this impact. In this paper we develop a simple system dynamics model, estimate it using data from 18 construction projects, and validate the model on a validation set of 15 projects. The model provides good fit for the calibration set and strong predictive power on the validation set. It also allows us to estimate the impact of undiscovered design changes on construction phase quality, which appears to be notable.
In this study, using different versions of a growth management game involving two different complexity factors, we compare performances of heuristic rules with experimental results. We present a method for obtaining a statistical distribution of scores resulting from a given simulated decision heuristic, which can be used to compare against and assess experimental gaming results. The method is based on the idea of generating vast number of scores by stochastically simulating a given decision rule and obtaining the resulting score distribution. We use this method to compare scores from different game versions whose scores are essentially not comparable, and to see how the score distributions change from one game version to another. In simulations, we first use a simple random "decision rule" and then develop a more intelligent hill-climbing heuristic. The results show that when the games involve delay, human subjects do not perform better than the random heuristic a primitive rule composed of a sequence of random decisions. On the other hand, in nonlinear games, subjects outperform the random heuristic and their scores fit better the score distribution of the hill-climbing heuristic. We also demonstrate how the score distribution from random heuristic can be used as a reference performance measure.
The aim of this study is to test statistically the effects of delay, nonlinearity and feedback factors on the complexity of a stock management task. The task requires the player to bring the inventory to a target level and keep it there. Each of the individual complexity factors brings different challenges to the game. Using a slightly modified Latin square experimental design, we test the factors at different strength levels. We use two measures of game complexity: game scores and players subjective difficulty ratings. The results show that, with respect to the simple base game, delay and nonlinearity create worsening in game performances. Also, with increased delay duration, delay order and nonlinearity, subjects' performances deteriorate. However, feedback does not deteriorate the game performance. Furthermore, increased feedback strength even improves scores, due to a technical side effect on the performance measure. All subject groups exhibit learning by repeated trials. Nevertheless, there is also evidence that delay prevents transfer of learning to other game versions. The subjective complexity ratings of the players yield parallel results, the overall correlation of game scores with the subjective difficulty ratings being +0.59.