Background · CPI methodology history
A plain-language timeline of the methodological changes the Bureau of Labor Statistics has made to the CPI. What each change does, when it was made, how it affects the reported inflation rate, the official rationale, and the critique.
The Consumer Price Index that the federal government reports today is not the Consumer Price Index of 1980. Over the past four decades, the Bureau of Labor Statistics has made a series of methodological changes that have, in aggregate, produced a lower reported inflation rate than the pre-1983 methodology would produce.
Each individual change has a stated rationale and a body of academic literature defending it. Each change also has critics who argue the cumulative effect is to understate what families actually experience. This page presents the changes factually: what each change does, when it was made, the arithmetic direction it pushes the headline number, BLS's official justification, and the critic's response. Readers can draw their own conclusions about whether the cumulative methodology is more or less accurate than what it replaced.
Before 1983, BLS calculated the housing component of CPI using a "user cost" measure that included actual home prices, mortgage interest rates, property taxes, insurance, and maintenance costs. In January 1983, BLS switched to Owner's Equivalent Rent (OER): an estimate of what homeowners would have to pay to rent their own homes. The change applied to CPI-U immediately; CPI-W (used for Social Security and union contracts) was transitioned in January 1985.
Arithmetic effect
Reduces reported housing inflation when home prices rise faster than rents — which has been most of the period since 1995.
Weight in CPI
Housing services are roughly 33% of CPI; OER alone is about 24%, making it the single largest line in the index.
Primary source
The pre-1983 method conflated investment costs (mortgage interest, home appreciation) with consumption costs (the actual service of shelter). High inflation and double-digit interest rates of the late 1970s made the old approach unworkable. BLS argues a home is fundamentally a capital good, and CPI should measure only the consumption service it provides.
Families actually pay mortgages, property taxes, and home prices — not theoretical equivalent rents. When home prices and mortgage rates rise faster than rents (essentially always since 1995), OER systematically understates the housing inflation real households face. OER is the single largest contested methodology in CPI.
The Senate Finance Committee appointed a commission led by Stanford economist Michael Boskin in June 1995 to study whether CPI overstated inflation. Its final report, "Toward A More Accurate Measure Of The Cost Of Living," issued December 4, 1996, concluded the CPI overstated annual inflation by approximately 1.1 percentage points, identifying four sources of bias: substitution bias, outlet substitution bias, quality change bias, and new product bias. The Congressional Budget Office estimated this overstatement would add $148 billion to the federal deficit and $691 billion to the national debt by 2006 if not corrected — because Social Security COLAs, federal benefits, and tax brackets were all indexed to the "biased" CPI.
Arithmetic effect
Set the intellectual foundation for the methodology changes BLS implemented from 1998 through 2002, all of which had the cumulative effect of reducing reported inflation.
Bias estimate
1.1 percentage points per year overstatement (Boskin's own figure for 1996); prior to 1996, estimated at 1.3 percentage points per year.
Primary source
Final Report (SSA archive)
Consumers substitute between products when relative prices change; CPI's fixed-basket methodology missed this. New products entered too slowly. Quality improvements in technology and durable goods were systematically miscounted as price increases. The commission was bipartisan and its findings were endorsed by mainstream academic economists.
Every Boskin recommendation had the arithmetic effect of reducing reported inflation. None were corrections in the other direction. The commission's timing — exactly as Social Security trust fund pressures mounted in the mid-1990s — suggested to critics that the methodology corrections served the federal budget at least as much as statistical accuracy.
BLS had been using hedonic methods in housing (since 1988) and apparel (since 1991) to adjust prices for measurable quality changes. In January 1998, BLS expanded the methodology to personal computers, then added televisions (1999), consumer audio equipment, VCRs, camcorders, DVD players, and college textbooks. By 2018, BLS extended hedonic adjustment to smartphones, residential telephone services, internet services, and cable and satellite television.
Arithmetic effect
When a product's measurable quality improves, the price increase is reduced by the dollar value of the quality improvement. For technology specifically, this routinely produces large negative price changes — i.e., reported deflation.
BLS's own estimate
Per BLS, hedonic adjustments outside shelter and apparel have reduced annual CPI growth by approximately 0.005 percent per year. Combined weight of hedonically-adjusted categories outside shelter and apparel: about 1% of CPI.
Primary source
A 2024 smartphone has dramatically more capability than a 2014 smartphone at the same price point. Treating it as the same product would systematically overstate inflation. Hedonic methods isolate the price change for an equivalent-quality good.
Hedonic adjustment is asymmetric in practice. When product quality improves, the price change is adjusted down. When quality declines — smaller airplane seats, worse customer service, shrinkflation in packaged goods, less durable appliances — there is no corresponding upward adjustment. The methodology aggressively treats quality improvements as price reductions but rarely treats quality declines as price increases.
Effective January 1999, BLS introduced a geometric mean (geomean) formula for calculating most elementary-level CPI estimates, replacing the modified Laspeyres arithmetic-mean formula used previously. The change applied to about 61% of the CPI by weight; categories where consumer substitution is limited (most medical care services, hospital services, professional services) retained the older formula. The geomean formula assumes consumers substitute toward relatively cheaper items within a category — buying more Granny Smith apples when Red Delicious prices rise.
Arithmetic effect
By construction, a geometric mean index is mathematically smaller than or equal to its arithmetic mean equivalent (when based on the same weights). BLS estimates this single change has reduced annual CPI growth by approximately 0.3 percentage points per year.
Coverage
Approximately 61% of CPI by weight; remaining 39% (mostly medical services and shelter) still uses arithmetic-mean methodology.
Primary source
The pre-1999 arithmetic mean assumed consumers continued buying a fixed basket regardless of price changes within a category. Real consumers substitute. The geometric mean formula incorporates this substitution and "better approximates a cost-of-living index" per BLS.
This is the largest single methodological change in CPI's modern history, by BLS's own estimate reducing annual reported inflation by roughly 0.3 percentage points. Over 25 years, that compounds to roughly 8% lower cumulative inflation than the pre-1999 methodology would produce. Critics argue substitution is a behavioral adaptation to inflation, not a reduction in the inflation experienced.
BLS began publishing the Chained Consumer Price Index for All Urban Consumers in August 2002, using data from July 2002 forward. The C-CPI-U addresses "upper-level substitution" — consumer shifts between major categories (e.g., from pork to beef when pork prices rise). The geometric mean change in 1999 handled within-category substitution; C-CPI-U handles between-category substitution. C-CPI-U has averaged 0.2 percentage points below the standard CPI-U from 2001 through 2023.
Arithmetic effect
C-CPI-U has averaged approximately 0.2 percentage points below CPI-U per year, with substantial variation. It exceeded CPI-U briefly in January 2023 during pandemic spending pattern disruption.
Status until 2017
C-CPI-U was published as an alternative measure but was not used for federal tax brackets or Social Security COLAs.
Primary source
The chained methodology more accurately reflects real consumer behavior. When apples become more expensive than oranges, consumers switch. A "cost of living" index should reflect this substitution. The C-CPI-U is widely endorsed by mainstream economists as the most accurate cost-of-living measure available.
Substitution toward less-preferred goods is not the same as the cost of living holding constant. If a family's grocery budget stays flat in dollars but they're now eating chicken instead of beef and store-brand instead of name-brand, their real consumption has declined even if the C-CPI-U records no inflation. Substitution measures adaptation to inflation, not absence of it.
The Tax Cuts and Jobs Act of 2017, enacted in December 2017 and effective for tax year 2018, permanently switched the inflation index used for federal income tax bracket adjustments from CPI-U to C-CPI-U. This change affected all annually adjusted tax provisions: tax bracket thresholds, the standard deduction, the Earned Income Tax Credit, and contribution limits for tax-advantaged accounts. The Joint Committee on Taxation estimated this single change would raise federal revenue by $134 billion over the first decade by reducing inflation adjustments to tax brackets.
Arithmetic effect
Tax brackets now adjust upward more slowly than they would under standard CPI-U. Over time, more taxpayers cross into higher brackets as wage inflation outpaces the C-CPI-U adjustment. The Joint Committee on Taxation estimated $134 billion in additional federal revenue over the 2018-2027 window.
Permanence
Unlike most TCJA provisions which expire after 2025, the chained-CPI indexing switch is permanent.
Primary source
C-CPI-U is "a closer approximation of a cost-of-living index than other CPI measures" per the TCJA legislative findings. Using a more accurate measure for tax indexing produces a more economically efficient outcome.
The arithmetic effect is a permanent increase in federal income tax revenue, accomplished by changing the inflation measure rather than the tax rates. Whether C-CPI-U is more "accurate" than CPI-U remains contested in the academic literature. Switching the index in only one direction — toward the version that produces more revenue — suggests something other than pure methodological improvement.
In January 2018, BLS extended hedonic quality adjustment to smartphones, residential telephone services, internet services, and cable and satellite television. The methodology uses "directed substitution" — when a new smartphone model replaces an old one in the BLS sample, the quality difference is calculated through regression modeling and deducted from the observed price change. Adjustments are made twice yearly to coincide with manufacturer release cycles.
Arithmetic effect
When successor products are more capable than predecessors (typical for technology), the hedonic adjustment converts a sticker-price increase into a smaller reported price increase, no change, or sometimes deflation. Combined CPI weight of post-2018 hedonic categories is small but growing.
Trend
BLS has indicated it will continue expanding hedonic methods to additional digital-economy categories as those categories grow in household budget share.
Primary source
The technology in a phone, broadband connection, or cable service changes dramatically year over year. Treating the iPhone 17 as identical to the iPhone 8 would systematically overstate price increases in a critical and growing share of household spending.
Hedonic adjustment continues to be asymmetric. A faster phone with more storage at the same dollar price is treated as deflation. A phone with shorter battery life, more advertising, more mandatory subscriptions, less repairability, or planned obsolescence does not trigger a corresponding price increase. The methodology counts gains aggressively and ignores losses.
Reality Index is not the only attempt to address the gap between official CPI and what households experience. Brief characterizations of the major alternatives:
ShadowStats Alternate CPI (John Williams). Estimates what CPI would report under the 1980 methodology and the 1990 methodology, applied to current data. Currently shows inflation rates several percentage points higher than official CPI. Methodology has been criticized as opaque; ShadowStats does not publish its underlying basket or detailed calculations. Influential among monetary-policy skeptics but generally not accepted in the academic literature.
MIT Billion Prices Project (Alberto Cavallo, Roberto Rigobon). Collects retail price data from online retailers in real time. The methodology is transparent and well-cited academically. Has generally tracked CPI closely over the post-2010 period, with some differences in volatility. Stopped publishing US data in 2016 but the methodology continues through PriceStats.
BLS Research Series (CPI-U-RS). BLS's own research series applying current methodology backward to 1978. Useful for consistent-methodology comparisons across decades. Not a critique of current methodology; it just makes historical comparisons mathematically consistent.
Federal Reserve Bank of Cleveland Median CPI. Reports the median (rather than mean) of category price changes, reducing the influence of volatile components. A useful Fed policy tool rather than a competing inflation measure.
Trimmed Mean PCE (Dallas Fed). Trims the top and bottom of category price changes each month, then averages the middle. Another Fed policy tool, increasingly used as an alternative to core PCE for monetary-policy decisions.
Reality Index distinguishes itself from these alternatives by being explicitly fixed-basket rather than substitution-adjusted, using independent retail dollar data wherever it exists (KFF for health insurance, NCES for tuition, FHFA for home prices, AAA for vehicles, EIA for gasoline, BLS APU for retail food), and publishing every component, every weight, and every methodology choice openly for inspection and critique. The full methodology is at methodology_paper.html.
The methodology choices above are not abstract statistical questions. They have direct, measurable consequences for federal benefits, federal revenue, contract escalation, and household budgets:
Social Security cost-of-living adjustments. SS COLAs are indexed to CPI-W (a variant of the standard CPI-U used for urban wage earners). Every methodology change that reduces reported inflation reduces the annual COLA, compounding over decades. A retiree's Social Security check is meaningfully lower than it would have been under the pre-1983 methodology — by some estimates, 20-30% lower in cumulative terms.
Federal income tax brackets. Since 2018, brackets are indexed to C-CPI-U rather than CPI-U. Inflation pushes wages up; brackets adjust upward more slowly than wages; taxpayers cross into higher brackets without real income gains. The federal government collects an estimated $134 billion in additional revenue over the 2018-2027 decade from this single change.
Treasury Inflation-Protected Securities (TIPS). TIPS principal is indexed to non-seasonally-adjusted CPI-U. Every methodology change that reduces reported inflation reduces the inflation adjustment paid to TIPS holders. Foreign central banks and pension funds holding TIPS are paid less than they would have been under the pre-1983 methodology.
Federal contracts, union COLAs, multiemployer pension plans. Many private contracts and pension plans use CPI-U for inflation adjustment. The methodology changes flow through automatically — typically without the contracting parties' attention.
None of this resolves whether the methodology changes were correct. It just establishes that the methodology choices have measurable consequences for who pays and who receives. Reality Index does not advocate for one set of methodology choices over another. It measures what families actually paid, in unadjusted retail dollars, against fixed specifications, over the longest available record — and reports the gap between that measure and official CPI honestly.
This page is part of the Reality Index methodology series. For the headline inflation rate calculation, see the headline rate. For the full methodology, see methodology paper. For the sensitivity analysis showing how different weighting methodologies affect the result, see sensitivity analysis. For what Reality Index does not measure, see what we don't measure.
Corrections, additional methodology changes worth documenting, and source suggestions welcome at info@realityindex.co.