Early procedures for the analysis of multivariate panel data do not rest on well-specified statistical models. Recent approaches based on path analysis suffer from the defects of variable standardization and lack of attention to measurement error. The paper formulates a measurement model for quantitatively scaled multivariate panel data. The model is applied to a data set indexing two constructs measured at three time points. Multiple measurement of each construct in conjunction with the measurement model allows estimation of a true variance-covariance matrix. Analysis of this matrix produces substantially different interpretations of variable influence than similar analyses of the original data. (Author)