Statistical Methods & Applications(2003) 11:395404 SMA @ Springer-Vedag 2003 The treatment of substitution bias in consumer price index: an alternative approach Ignazio Drudi Universityof Bologna,Dipartirnentodi Statistica,Via Belle Arti 41,40125 Bologna,Italy (e-mail: drudi@ stat.unibo.it) Abstract. Substitution bias is a well-known problem in fixed-basket price indices. When a new product substitutes an old one, the most of statistical agencies adopt an ad hoc strategy, using the ratio between prices of the two goods (in a previous period) as a measure of quality change. In the present work we propose an alter- native way to manage substitution that can be easily included in the computation process of the index. Price survey is a pure panel survey, and then substitution may be considered as an attrition problem and faced using the estimator for panels with partial overlap. After a brief description of the problem and of the suggested for- mula, an experimental application is presented. The application is based on about 771 elementary prices collected in Milano in March 1997. Main results are that in each category of consumption the two approaches show significant differences in the micro-indices, at the aggregated level, that is when weights are used to combine micro-indices, the differences agree with the conclusions of Boskin's report. Key words: Price indices, substitution bias, panel survey 1. Introduction Almost all Consumer Price Indices (CPI) available in the world are based on a fixed-basket and hence on the hypothesis of time-invariance of the quality of goods included in the basket. Nevertheless, the concept of quality is rather complex as it is a lot of factors affected by. Consumers' behaviour, supply strategies, technological innovations interact in modifying characteristics of products available for final consumption. As a consequence, national statistical agencies have to face the well-known problem of "substitution bias". There has recently been a considerable interest in examining the foundations and measurement biases of the Consumer Price Index