Bias in the CPI: "Roughly Right or Precisely Wrong"


Brian Motley

FRBSF Economic Letter 1997-16 | May 23, 1997

Many economists argue that our most closely watched indicator of inflation, the consumer price index (CPI), is biased and overstates inflation. In December 1996, a group of economists appointed by the Senate Finance Committee reported on a study of the CPI and estimated that the index overstates annual inflation by about 1.1 percentage points (Boskin, et al. 1996); so, instead of the official 2.8% rate of inflation in 1996, it might have been 1.7%.

Many economists argue that our most closely watched indicator of inflation, the consumer price index (CPI), is biased and overstates inflation. In December 1996, a group of economists appointed by the Senate Finance Committee reported on a study of the CPI and estimated that the index overstates annual inflation by about 1.1 percentage points (Boskin, et al. 1996); so, instead of the official 2.8% rate of inflation in 1996, it might have been 1.7%.

While economists’ debates over statistical constructs don’t often make the headlines, in this case they did, because adjustments to the CPI could have important implications for policymakers and the public as well. For example, if the index does overstate inflation, then Fed policymakers are doing better at keeping inflation low than the official numbers suggest. Furthermore, the CPI is the basis for adjusting tax brackets, benefits in certain entitlement programs, and some labor contracts, to account for the effects of inflation. If it is overstated, this would mean that the government has been raising entitlements by more than required to keep up with inflation and that taxpayers have been paying lower income taxes than Congress intended. The Congressional Budget Office estimates that correcting this bias would lower annual outlays in the first year by $4 billion and increase revenues by $2 billion, a decline in the overall deficit of about $6 billion. These deficit reductions would compound over time and also would imply reduced interest payments on the federal debt. The CBO estimates that by the year 2007, the annual deficit would be reduced by $140 billion if the bias were removed.

It is not a simple matter to correct the bias, however. To shed light on the difficulties, this Letter describes the CPI, the sources of bias, and their possible resolutions.

Constructing the CPI

The sources of bias are related to the way the index is constructed. The Bureau of Labor Statistics (BLS) constructs the CPI monthly. The process begins with a Consumer Expenditure Survey, which provides information on spending habits and is used to construct a market basket of goods and services. Currently the CPI is based on a survey from the early 1980s; next year, a survey from the early 1990s will be incorporated. This survey is supplemented by a Point-of-Purchase Survey, which identifies the outlets where households make their purchases. Each month, BLS enumerators visit outlets in cities countrywide to price the goods and services in the market basket. Great effort is expended in trying to price exactly the same items each month or to find comparable items when this is not possible. The price index weights the price of each item according to its importance in the market basket. So it measures how much the cost of the basket has risen since the previous month.

Substitution bias

One element of the bias is known as “substitution bias.” As time passes some prices rise more than others. Since 1983, for example, shelter costs have risen more than 75% while apparel prices have increased only 30%. Households tend to buy less of the items with larger price increases and more of those with smaller price increases. The CPI, however, assumes that the market basket that households buy does not change. This means that items whose prices have risen most receive too much weight in the index (because households substitute away from them), while those whose prices have risen least are given too little weight (because households shift their spending toward them). In other words, the economies that households obtain by substituting cheaper items for more expensive ones are not captured by the index, which, as a result, shows more inflation than households actually suffer.

If the measured increase in prices over a period were based on the market basket households bought at the end of the period rather than the beginning, inflation would tend to be understated rather than overstated. The prices that had risen the least would get more weight than they deserved (since households would buy more of these items at the end of the period) and, conversely, those that had risen the most would get less weight than appropriate.

This suggests a way of dealing with substitution bias. If price changes over a period are measured using the market baskets bought at both the beginning and the end and these two measures are then averaged, we should get a truer measure of inflation. Indeed, providing certain technical conditions are met, this procedure delivers a very good measure of the true rise in the cost of living, which can then be used to estimate the size of the substitution bias in the official CPI.

Unfortunately, this procedure is not practical in real time, because we never have up-to-date information on what households are buying this month. But the Consumer Expenditure Survey now is conducted quarterly and could be used to estimate the bias on an ongoing–though always somewhat delayed–basis. So it would be possible to make a good estimate of the degree of substitution bias and to correct for it.

In addition to the effects of substitutions among the broad groups of goods and services distinguished in the CPI–between, say, apples and oranges–the effects of substitutions within categories–between, say, pippins and galas within the overall category of apples–also are ignored. In many categories, this is almost certainly a bad assumption. Changes in prices between different types of apples surely would cause widespread substitution, and ignoring it will cause inflation in apple prices to be overstated. But it may be a good assumption for medications: households surely would not substitute aspirin for a laxative just because its price dropped.

Because present data collection methods do not provide timely information on purchases of individual items, this bias cannot be measured precisely. However, BLS researchers have tried to estimate its order of magnitude. The BLS recently published an alternative CPI measure that assumes that when relative prices change, households alter their purchases so that the share of their spending devoted to each individual item remains the same. Thus, if the prices of different types of apples change, households spend the same shares of total apple spending on each type. Comparing this experimental index with the official CPI provides one estimate of substitution bias at the individual commodity level. The BLS plans to decide later this year whether to adopt this alternative assumption.

Research on substitution bias suggests that it amounts to slightly less than 1/2 percentage point a year. About half of this bias represents the effect of substitutions at the level of broad commodity groups and the remainder is due to substitutions between individual items within these groups.

Outlet bias

A second bias occurs through failure to take proper account of changes in the types of outlets favored by households in their purchasing. Households can reduce the impact of rising prices by switching toward cheaper outlets, though often these provide less customer service. The share of discount outlets has risen over the years, indicating that for some customers the lower prices are not offset by the associated reduction in service quality.

The CPI ignores price changes that occur when customers switch between outlets. This introduces an upward bias into the index, because some of the savings associated with the move toward cheaper outlets represent true price reductions that are not offset by the lower level of service. One estimate suggests that this bias raises the CPI by 0.1 percentage point a year (Lebow, Roberts and Stockton, 1994). This estimate is rough because it is difficult to disentangle differences in prices from differences in service quality.

Quality change and new goods

The most difficult parts of price level measurement are how to handle changes in the quality of goods and services and how to deal with completely new items. Even if the quantities of goods and services that households purchased remained the same over time, a bias would remain if their quality changed. Clearly an auto purchased in 1997 is quite different from one purchased in 1957 and delivers more services in terms of comfort, safety, and longevity. Similarly, the cost of medical care has risen substantially in recent years, but medical advances have greatly improved its effectiveness. Hence, the true cost of care has risen less than the observed price because the quality of services purchased has risen.

A special kind of quality bias occurs when completely new products are introduced. Inevitably, new products are not introduced into the CPI until they are in widespread use. The microwave oven was not added until 1978 and the VCR until 1987, many years after they first became available. The ubiquitous cellular phone will not enter the index until 1998. The bias arises because prices frequently decline very rapidly in the few years after a new item enters the market. If the CPI market basket contains mostly mature items and few newly emerging ones, these early price declines will be missed, thus contributing to upward bias in the overall index.

BLS statisticians, of course, are well aware of the need to account for quality change. The quality adjustments made in constructing the index have a substantial effect on measured inflation, and new adjustments often are introduced. For example, one recent change altered the way in which hospital services are priced. Instead of pricing the cost of a hospital room or an hour of a physician’s time, the BLS is moving toward pricing the cost of treating individual medical conditions. So if improved surgical techniques reduce the length of a hospital stay, this will register as a price reduction even if the daily room rate remains the same. It is estimated that this change, which represents a better method of measuring quality change in medical care, will shave up to 0.1 percentage point off measured annual inflation.

But are there other quality changes that are not accounted for? The Boskin report argues that the BLS misses many quality improvements; and it estimates that in recent years this has caused CPI inflation to be biased upward by 0.6 percentage point. This estimate is based on a category-by-category evaluation of each of the components of the index. However, although many of these evaluations are based on empirical data, others are derived from judgments that not everyone would agree with. For example, the report guesstimates that the greater variety of foods and beverages now available represents a quality improvement that introduces an annual bias of almost 0.1 percentage point into the CPI.

BLS officials have argued that the quality bias is smaller than the Boskin Commission estimates. They also argue that making bias adjustments on the basis of judgments that–although plausible–cannot be fully defended with statistical evidence would open the BLS to endless criticism and controversy and might undermine the credibility of the CPI and other BLS data, too (Abraham 1997).


The BLS and other researchers are working on a number of fronts to identify and deal with biases in the CPI. Many of the estimates cited in the Boskin report came originally from this research. Already a variety of changes are in train that will have the net effect of cutting almost half a percentage point off measured inflation by the end of the decade.

Nonetheless, there is significant uncertainty about the size of the remaining bias. But although we may be uncertain about its exact size, we can be pretty sure that the bias is not zero, but positive. Assuming that there is no bias until we are able to measure it precisely is a bad strategy because, in fact, we shall never be able to measure it precisely. In these circumstances, assuming that the bias is, say, between 1/2 % and 1 1/2%, is almost certainly preferable to assuming that it is zero. As Chairman Greenspan argued recently–quoting Maynard Keynes–it is better to be roughly right than to be precisely wrong.

Brian Motley
Research Officer


Abraham, Katharine G. 1997. Testimony. House Budget Committee (March 12).

Boskin, Michael J., et al. 1996. Toward a more Accurate Measure of the Cost of Living. Final Report to Senate Finance Committee from Advisory Commission to Study the Consumer Price Index (December 4).

Lebow, David E., John M. Roberts, and David J. Stockton. 1994. Monetary Policy and “The Price Level.” Mimeo. Federal Reserve Board.

Opinions expressed in FRBSF Economic Letter do not necessarily reflect the views of the management of the Federal Reserve Bank of San Francisco or of the Board of Governors of the Federal Reserve System. This publication is edited by Anita Todd and Karen Barnes. Permission to reprint portions of articles or whole articles must be obtained in writing. Please send editorial comments and requests for reprint permission to