We develop estimation methods for continuous-time Markov and semi-Markov non-homogeneous manpower systems using the notion of calendar time divided into 'time windows' by change points. The model parameters may only change at these change points but remain constant between them. Our estimation methods employ a competing risks approach and allow for left truncated and right censored data. Maximum likelihood estimators are given for the hazard and survivor functions describing length of stay in any grade of the manpower system. The models are fitted to data from the Northern Ireland nursing service. (C) 1998 John Wiley & Sons, Ltd.