Introduction
Accounting for quality differences between goods is a perennial concern when constructing a price index. The goal of a price index is to capture a pure price movement across two periods, but if there are systematic differences in the goods being compared over time that also affect price—so-called differences in quality—then a pure price movement cannot be captured using transaction prices alone. Without prior information, there is no way to disentangle a pure price movement from a movement in price due to changing quality over time. The ideal measure of a pure price movement is a constant-quality price index that holds the determinants of price fixed across periods. This allows for an apples-to-apples comparison using transaction prices, and any difference in prices between periods must reflect a pure price movement.
There are a number of techniques available to combat quality differences creeping into a price index. Probably the simplest and most intuitive approach is the pure matched-model index, wherein prices for pairs of similar goods are compared over time. By focusing on pairs of similar goods, it becomes less likely that differences in quality between different goods can contaminate measurement of a pure price movement. Other more exotic techniques, namely stratification and hedonics, offer the same promise of a constant-quality price index, usually at the expense of additional econometric apparatus.
One interesting feature of these different methods—matched model, stratification, and hedonics—is that they really aren’t that different. Each relies on the same key assumption to produce a constant-quality index; what differs between the methods is simply how they go about implementing this assumption. This point is fundamental in order to understand how these different approaches can deliver a constant-quality index, and how they relate to each other.
In the interest of simplicity, this course focuses on geometric price indices. The concepts for a geometric index are directly applicable to arithmetic indices, although there are some extra details to worry about in the arithmetic case.30 In most applications quality adjustments are done using a geometric index-number formula. Throughout this course, attention is focused on building a constant-quality index when the entire population of transactions for goods and services is known. This is almost never the case in practice, but it makes the exposition easier by ignoring issues associated with sampling and statistical inference. It also emphasizes that building a constant-quality index is fundamentally a population-level problem, and not an issue of sampling. Once a constant-quality index is defined in the population, it is simple to estimate it with a sample of transactions by replacing the population-level quantities with their sample counterparts—see Manski (1988).
See Lee (2016, chap. 1) and Manski (2007, chap. 7) for some of the details.↩︎