The Consumer Price Index: Data Quality

Questions have recently arisen surrounding the reliability and usefulness of certain federally produced data, including the Consumer Price Index (CPI). The President's recent removal of the head of the Bureau of Labor Statistics (BLS, which produces the CPI) as well as the President's FY2026 budget proposal, which would further reduce BLS staffing levels, have brought renewed interest to the topic. The CPI is one of the leading measures of U.S. consumer inflation and is tracked closely by both private and public actors. New CPI data may affect consumption, financial market, and policy decisions. Given the widespread use of the CPI, it is important to the U.S. economy that it be timely, accurate, and trusted.

This Insight covers what the CPI is and how it is measured, recent changes to CPI data collection and methodology, and longer-term trends in response rates.

CPI Methodology

BLS is a statistical agency housed within the Department of Labor. It publishes CPI data on a monthly basis, including series for all urban consumers as well as wage and clerical workers. (A chained index is also produced with a lag.) Data breakdowns are available by expenditure category and geographic detail.

BLS creates the CPI using several surveys:

• The decennial Census provides population data used to create the geographic sample of the CPI.
• The Consumer Expenditure Surveys provide expenditure data used to create the "market basket" of goods and services for which prices are collected and weight those expenditure categories based on average spending patterns.
• Surveys on prices provide the data used to calculate how prices have changed from one month to the next. BLS uses two different surveys, one in which roughly 100,000 prices per month are collected for commodities and services and another in which roughly 8,000 rental housing unit quotes are collected for all housing.

BLS personnel collect most price data via personal visit or phone, but some commodities and services prices are obtained on websites and apps or from secondary sources.

2025 Collection Reductions

BLS announced this year that, in order to "align survey workload with resource levels," it would suspend data collection entirely in Lincoln, NE, and Provo, UT, beginning in April and Buffalo, NY, beginning in June. For the other 72 areas in which price data is collected, on average, roughly 15% of the sample was suspended. Altogether these collection reductions have resulted in fewer prices and rents being used to calculate inflation.

BLS estimates the impact of the collection reductions to be minimal. A simulated backward-looking series that removed Lincoln, Provo, and Buffalo produced results that differed on average from the actual CPI by less than 1/100^th of a percentage point for 12-month inflation. However, BLS also notes that it has not measured the impact of collection reductions on subnational or individual price indexes, which may exhibit more volatility as a result of reductions.

According to BLS, these reductions are meant to be temporary.

Other Methodology Changes in 2025

As of August 2025, BLS has announced three methodology changes to the CPI in 2025:

• As of April 2025, the source data for leased cars and trucks was changed from survey data to transaction data purchased from a vendor.
• As of July 2025, the source data for wireless telephone services was changed from survey data to secondary source data, and the calculation was changed to a non-traditional indexing method.
• As of October 2025, long-term care insurance will be removed from the insurance index owing to changes in this market.

These changes are meant to improve the accuracy and precision of the CPI.

Survey Response Rates

Surveys are generally voluntary and have been exhibiting a downward trend in the number of responses elicited for many years. This is true of the price surveys used to calculate the CPI as well as most other surveys conducted by statistical agencies. As shown in Figure 1, response rates for the price surveys for commodities and services as well as housing have both trended downward over the past two decades. This trend accelerated during the pandemic.

Survey response rates are important for precision, because the more people from a sample respond, the more likely the survey is to be statistically representative of the sample. Higher response rates are typically associated with data of higher "quality." Response rates can and do differ between surveys for various reasons, including differing personnel or time constraints and data collection methods.

Figure 1. Survey Response Rates, 2003-2024 Responding Sample Units Divided by Eligible Sample Units

Source: Bureau of Labor Statistics, "Response Rates for the Consumer Price Indexes." Notes: Housing response rate data in this graph refer specifically to the shelter component of housing. Housing response rates shown are for units designated "data reported," meaning usable information was obtained. Categories also exist for "found vacant," when a unit is located but is unoccupied, and "other," when a unit is eligible but no data are available. Vacant units may be used in the ultimate calculation of CPI.

BLS has taken certain steps to account for declining response rates. For the CPI, some data collection methods have changed from in person to phone, website, or web-scraping tools. Additionally, BLS has increased the use of secondary data sources, as illustrated by two of the methodology changes to the CPI discussed in the above section. More generally, BLS has, at times, increased the sampling time or size for certain surveys.

While CPI response rates are up from pandemic lows (perhaps as a result of some of BLS's efforts to improve data quality), they do remain depressed and are likely to continue to be a concern of BLS, the statistical community, and the greater economic and policymaking community moving forward. Nonetheless, BLS does commit to applying guidelines and standards to ensure data protection, privacy, and quality. There is additionally some evidence to suggest that declining response rates may not have had significant impacts on data quality given that the magnitude of revisions has not significantly changed in recent years as compared to pre-pandemic years.