Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
Augmenting a first-order dynamic regression model by adding particular redundant regressors gives a least-squares estimator of the lagged-dependent variable coefficient that is independent of nuisance parameters under a null hypothesis. This estimator and its t ratio have finite sample null distributions that differ considerably from Student's, but they can be determined exactly by simulation. The invariance properties indicate that for cointegration tests one redundant regressor (the constant or the linear trend) suffices for obtaining similarity. These results generalize characteristics of the Dickey-Fuller tests for unit roots. In a simulation study, the authors examine the invariance and power of various tests in simple illustrative models. Copyright 1992 by Blackwell Publishing Ltd