Sensitivity analysis on a simulated helpdesk system with respect to input distributions with special reference to the circumference method

Abstract

Simulation analysis makes use of statistical distributions to specify the parameters of input data. It is well known that fitting a distribution to empirical data is more of an art than a science (Banks J., 1998, p. 74) because of the difficulty of constructing a 'good' histogram. The most difficult step is choosing an appropriate interval width. Too small a width will produce a ragged histogram, whereas too large a width will produce one that is overaggregated and block-like. De Beer and Swanepoel (1999) have developed 'Simple and effective number-of-bins circumference selectors' for creating histograms for the purpose of fitting distributions. When using simulation software such as Arena, one can generally fit distributions to input data using a built-in function in the software. If input distributions could be compared regarding their effect on the outcomes of a simulation model, one could assess whether input distributions generated by Arena could be accepted unconditionally or whether one should pay special attention to the input distributions used in the simulation model. In this study a simulation model of a computer helpdesk system is constructed to test the effect of input distributions. Distributions fitted with the 'circumference technique' are compared with those from the simulation package, Arena, and those calculated by the statistical package 'Statistica', and then compared with empirical distributions. In the helpdesk system, calls from employees experiencing problems with any computer hardware or software are logged, redirected when necessary, 'attended to, resolved and then closed. Queue statistics of the simulation model using input distributions suggested by Arena as opposed to input distributions deduced from the other methods are compared and a conclusion is reached as to how important or unimportant it is for this specific model to select appropriate input distributions.