Most ammunition is produced long before its ultimate consumption and s
tored in a series of different depots for a considerably long period o
f time. During storage, the quality of the ammunition stockpile deteri
orates proportionally to the condition of the depots. We view differen
t conditions associated with a series of depots as step stress. A rand
om effects logistic regression model is employed to predict the qualit
y of ammunition stockpile in terms of the routing information such as
a series of locations and duration of storage of ammunition lots. The
resultant prediction model can be used to determine the appropriate ti
me for reorder or renovation of ammunition before the quality becomes
substandard. An example is given to illustrate the implementation proc
edure of the prediction model suggested in this article. (C) 1994 John
Wiley & Sons, Inc.