Bias in the compilation of consumer price indices when different models of an item co-exist

Abstract

This paper draws attention to a potential bias in the measurement of inflation arising from the coexistence in the market of new and old models of a product. This is a well-recognised problem which has been analysed in the literature in terms of difficulties in estimating quality differences. In this paper it is shown how the bias extends beyond the issue of the technicalities of such quality adjustments, but arises from the very nature of pricing behaviour in this market context. We show why there might be quite different pricing policies for a new and existing model at the time of launch and subsequent to it. Since a price collector can only either continue to monitor the prices of the old model, or switch (with a quality adjustment) to the new model, a bias will arise. The nature of the bias can be considered as biased sampling, there being a tendency for statistical offices to keep collecting prices on old models rather than new ones which necessitate a quality adjustment. On subsequent rebasing there may well be a reversal of the bias towards new models. A formal model of the nature and extent of the bias is developed and some simulated results are provided. The paper concludes with some practical suggestions.