Pooled Cross-Section, Time-Series Evaluation: Source, Result, and Correction of Serially Correlated Errors

Possible sources of the error structures associated with 2 commonly used cross-sectional education evaluation regression models are given. The implications of such structures for pre- and posttest evaluation are examined.Two regression models that are often used are: 1. the value-added model, and 2. the absolute level model. Contrary to Webster's and Bereiter's speculative assumption, the error structures inherent in these 2 models are shown to cause differing statistical biases depending on the model specification considered and the method of estimation employed. A generalized model is proposed as an appropriate way to handle 2 particular specification problems, and an iterative maximum likelihood estimator is suggested for estimating this model.