On the relevance of t-ratios in empirical modelling: two special cases

Abstract

It is a common practice to rescale data to assist model estimation. Rescaling in general does not affect the model fit or the statistical inference from the estimation results. This article however, describes two cases where care is called upon when interpreting the t-ratio associated with the constant terms. One case is a linear model with a continuous dependent variable and the other case is a nonlinear model with a discrete dependent variable. It is proved in this article that in these two cases, one can arbitrarily manipulate the t-ratio associated with the constant term by rescaling the data. Implications from these results on empirical modelling are also given.

Acknowledgements

The author is grateful to Wiktor Adamowicz, David Hensher, and Kenneth Train for helpful comments. The author is responsible for any remaining errors.