OMEGA Int. J. of Mgmt Sci., Vol. 17, No. I, pp. 21-25, 1989 0305-0483/89 $3.00 + 0.00 Printed in Great Britain. All rights resetwed Copyright ~ 1989 Pergamon Press plc The Current State of Network Simulation in Project Management Theory and Practice C RAGSDALE University of Georgia, USA (Received June 1988) Sehonberger Ilnterfaees 11($), 66-70 (1981)1 has demonstrated the advantages of Monte-Curio simulation of activity networks over the mote traditional PERT/CPM techniques. He concluded however, that simalatiom were too expensive and required too extensive a hoekgroud in probabili~' for most pm~ct managers. This paper highUghel the problems with PERT/CPM and reviews the approaches researchers and practitioners have taken to network simulation since Schonberger's article. It then offers suggestions for future work and reevaluates Schonberger's conclusions in light of current technology. INTRODUCTION FOR MANY YEARS, the project evaluation and review technique (PERT) has been a standard tool used for planning and control in project management. Having defined the intricate re- lationships among the activities in a project and estimated their completion times, the critical path (or longest path) of the project can be determined. Various statements can then be made concerning the probability of completing the project by a given time. As Schonberger [20] has shown, such practices lead to overly opti- mistic estimates of project completion times and results in projects 'always' being late. THE PROBLEMS WITH PERT IN PROJECT MANAGEMENT A number of problems exist with the tradi- tional PERT approach presented in many project management and quantitative methods textbooks. First, PERT focuses on each path in the network as ifit were independent of all others [20]. While the completion time for each path may vary, no consideration is given to the fact that noncritical paths merging into the critical path might be delayed. Without vigilant management this could easily lengthen the project, but the effects of such path interactions are ignored by PERT. 21 A second problem is that of the determining the critical path [24]. Given that actual activity times may vary, numerous parallel paths may have the potential of becoming critical. The standard textbook approach is to ignore this fact, find the one deterministic critical path and use it to calculate the probability of completing the project by a given date. Assuming (con- veniently) that activity times are independent and normally distributed we have: PIT,, < t] = P[Z ~ (t - E(Top)) S~I where: t = a desired completion time; E(T~,) = expected time of the project along the critical path; S,p = standard deviation of the critical path time; Z = a standard normal random variable. Now suppose the critical path in the network has an expected length of 30 days with a stan- dard deviation of 2 days. Then the probability of completing this path within 35 days is: P[Z < (35-30)/21 = P[Z < 2.5] = 0.9938 We might then inform management that there is a 99.38% chance of completing the project within 35 days. But suppose another 'non- critical' path in the network has an expected length of 25 days with a standard deviation of