Constructing Numerical Reference Modes from Sparse Time Series

Constructing the reference modes is a critical step in system dynamics modeling. Estimat-ing rates from sparse time series presents a unique problem. Specifically, as the counts per unit time approach zero, the time series start to look increasingly like discrete stochastic variables, i.e., not continuous, even though one might in some situations reasonably hypothesize an un-derlying continuous variable. Smoothing techniques are commonly used to identify patterns in noisy data, but introduce and remove features that could mislead the modeler. There has been considerable research on optimal smoothing techniques for noisy time series. This paper presents initial work toward different approach that side steps the question of optimal smooth-ing and takes advantage of the emphasis in system dynamics of good models being expected to perform well over a range of conditions. Having rich numerical data to support reference modes can be a critical element of successful system dynamics model testing and confidence building (Homer, 1997). While numerical data typically represent only a small portion of stakeholders ’ knowledge of a problem (Forrester, 1980), numerical data that are available should be included. Social services agencies typically maintai