Performance Sampling and Weibull Distributions
研究了威斯康星州主管人员流动数据，发现简单同质马尔可夫模型预测不准，探讨了用异质非平稳分布或组织过程建模的利弊，指出威布尔混合模型可能误导对个体异质性的推断。
March 1981, volume 26 March and March (1977) reported that data on Wisconsin superintendent mobility deviate in several ways from what would be predicted by a simple homogeneous, stationary Markov model. Such results can be modeled usefully by fitting any of several standard heterogeneous, nonstationary distributions, or by specifying an organizational process and deriving its distributional implications. The choice of tactics is partly a matter of taste and neither eliminates the risk of specification error. For example, some inferences about individual-level heterogeneity drawn from fitting a gamma mixture of Weibulls to duration data (Schmittlein and Morrison, 1981) can be misleading if the data are, in fact, generated by a performance sampling process (March and March, 1978).