SAMPLE DESIGNS IN CONSTRUCTION OF CONSUMER PRICE INDICES AND THEIR IMPACT ON SAMPLING ERROR

All member countries of the European Union have a consumer price index in their statistical systems. It is used as inflation rate indicator and for price stability comparison among member countries. As cost of living index methodologically better corresponds to consumer price index than retail price index, the Croatian Bureau of Statics has intention in 2004 to transform methodologically the term into consumer price index.The quality of consumer price index depends on the quality of data, and in particular on representation of both the sample of retail outlets used to monitor prices and the choice of items priced. As regards the sample choice practice in the Republic of Croatia it does not differ too much from the European Union practice. In member countries of the European Union are also mainly used judgematic sampling methods which are close to some kind of cutt-off selection, in which a large part of population (usually the items with the lowest expenditures) is left unobserved. The only EU countries that appear to make serious effortsat full “probability” sampling are the United Kingdom and Sweden. The main problem with “non-probability” sampling is that it does not permit the precise statistical estimation of sampling errors. Knowing the sampling error is important not only for the users but forthe statisticians as well. Optimization of the sample size is only possible in the case of a “probability” sample. Countries which optimize their sample allocation use the following criteria: price indices variance, cost of the price collection, travel costs and expenditure weights. Full “probability” sample has been used only in the United States of America so far. This design can be considered as the ultimate goal for each statistical office. The most difficult problem in selecting full “probability” samples for CPI is the lack of appropriate sampling frames of items.