The Measurement Window, Not the Basket, Drives the Social Security COLA Gap

Every October the Social Security Administration announces a single number, and every year the same complaint follows: the raise does not cover what actually went up. The standard explanation is that the government measures the wrong basket. The cost-of-living adjustment is tied to the Consumer Price Index for Urban Wage Earners and Clerical Workers, a working-age population, rather than to the research index the Bureau of Labor Statistics maintains for Americans 62 and older. Swap the index, the argument runs, and the gap closes.

That explanation is testable, and in the current inflation regime it does not survive the test. The basket-composition channel is real but small, and its sign is unstable. The mechanisms doing the visible work are three, and none concern which goods are in the index: the shape of the measurement window, the netting out of Medicare premiums, and the fact that Social Security runs two different indexation regimes on the same benefit at different points in its life.

The Surface Claim and What Is Actually Fixed by Statute

The formula is narrow and public. The Social Security Act sets the adjustment as the percentage increase in the CPI-W from the average for the third quarter of the last year in which a COLA became effective to the average for the third quarter of the current year, rounded to the nearest tenth of one percent. For the adjustment payable in January 2026: the third-quarter 2024 average of 308.729 against the third-quarter 2025 average of 317.265, producing 2.8 percent. Three things follow from that construction, and they matter more than the basket debate.

• Only three monthly readings enter the calculation. July, August and September of the measurement year. The nine months before and fifteen months after are outside the formula entirely.
• The rate is fixed for twelve payments. Once announced in October it does not adjust intra-year, whatever prices do. There is no true-up mechanism.
• Rounding is to the nearest tenth. A statutory truncation applied every year to the same compounding base.

The recent series shows the variance this produces without any change in method: 5.9 percent payable in 2022, 8.7 percent in 2023, 3.2 percent in 2024, 2.5 percent in 2025, and 2.8 percent in 2026. Those swings are not evidence that the basket is wrong. They show that a three-month window applied to a volatile series produces a volatile escalator.

Why the Basket Argument Fails in the Current Regime

The Bureau of Labor Statistics does publish a research index for the 62-and-older population, the R-CPI-E, with data running back to December 1982. BLS is precise about its limits: the expenditure weights come from a subset of the Consumer Expenditure Survey representing roughly one-fifth of the urban sample, raising sampling error, while the areas priced, outlets sampled and items selected all represent the total urban population rather than the older subpopulation. BLS states that conclusions drawn from analyses of the series should be treated as tentative. That is the agency's own characterization, not a critic's.

Set the caveats aside and look at the twelve months through July 2026. All-items CPI-U rose 3.4 percent; CPI-W rose 3.4 percent. Core, excluding food and energy, ran 2.5 percent. Medical care services — the category the retiree-basket argument leans on hardest — rose 2.7 percent, below headline. Shelter rose 3.2 percent. Energy rose 14.7 percent.

When the marginal inflation is coming from energy rather than from medical care, an index weighted toward working-age households with longer commutes runs hotter than an index weighted toward households aged 62 and older. In that regime, switching to the elderly index would have produced a smaller adjustment, not a larger one.

This is the part that gets skipped. The retiree-basket case is a claim about relative category inflation, and relative category inflation reverses. It was strong when medical care ran several points above headline; it is weak when medical care services run below headline and energy runs eleven points above it. Arguing for CPI-E as a permanent fix is arguing for a change whose sign one cannot control.

Mechanism One: The Window Sits Fifteen Months Ahead of the Last Payment

Take the 2026 benefit year. The window closed with the September 2025 CPI-W reading of 318.139. The rate was announced in October 2025, first appeared in the January 2026 payment, and pays through December 2026 — fifteen months after the window closed.

By July 2026 the CPI-W stood at 327.104. Against the 317.265 base on which the 2.8 percent adjustment was built, that is a 3.10 percent increase, with five months of the benefit year still to run at the fixed rate. That gap is not evidence of a mismeasured basket. It is the arithmetic of applying a backward-looking escalator to a forward-moving price level.

The symmetry point is usually omitted. The lag cuts both ways: following the third-quarter 2022 window, the 8.7 percent adjustment paid through calendar 2023 while headline inflation decelerated sharply, delivering a real gain. A formula that overshoots in disinflation and undershoots in reacceleration is not biased against retirees. It is a low-pass filter, and the complaint is really about the phase of the cycle.

One further fragility surfaced in this cycle. The October 2025 CPI-W was not published, because a lapse in appropriations prevented BLS from collecting that month's survey data. The measurement window was unaffected — it had closed in September — but the episode establishes that the price series feeding the largest federal benefit escalator is one appropriation away from a hole. That is a structural exposure with no analogue in the basket debate.

Mechanism Two: The Premium Net-Out

For most beneficiaries the number arriving in the account is the gross benefit minus the Medicare Part B premium. The two escalate on entirely different schedules.

Item	2024	2025	2026	2026 change
Part B standard monthly premium	$174.70	$185.00	$202.90	+$17.90 (+9.7%)
Part B annual deductible	$240	$257	$283	+$26 (+10.1%)
Part A inpatient deductible	—	—	$1,736	per benefit period
Social Security COLA	3.2%	2.5%	2.8%	—

A 2.8 percent gross adjustment against a 9.7 percent premium increase is the arithmetic most beneficiaries experience, and no index substitution addresses it. Medical care services inflation of 2.7 percent and Part B premium growth of 9.7 percent are different objects: the premium covers a share of program cost including utilization and enrollment mix, not a consumer price series.

The statutory response is Section 1839(f) of the Social Security Act, the hold-harmless provision. Its text is narrow: the monthly premium shall not be increased to the extent that the increase would reduce the December benefit below the November benefit. It protects the nominal check, not purchasing power. A beneficiary whose entire dollar COLA is absorbed by the premium ends the year with the same nominal payment and a lower real one, and the statute has done what it promises.

The distributional consequence is underdiscussed. Hold harmless does not apply to enrollees adjusted under subsection (i) — the income-related adjustment, where higher-income enrollees pay 35, 50, 65, 80 or 85 percent of total program cost — nor to those whose premiums are not deducted from a benefit check, including new enrollees and those covered by Medicaid. When the held-harmless population is large, unprotected cohorts absorb the shortfall through a steeper standard premium. The protection is real; it is also a cost-shifting mechanism invisible in any COLA statistic.

Mechanism Three: Two Indexation Regimes on One Benefit

This is the structural feature the surface debate misses entirely. Social Security indexes a benefit to wages before it is claimed and to prices after.

Initial benefits run through the primary insurance amount formula, whose bend points are re-indexed each year by the national Average Wage Index. For 2026 the bend points are $1,286 and $7,749, applied at 90, 32 and 15 percent of average indexed monthly earnings in the respective brackets. The first was derived by scaling the 1979 value of $180 by the ratio of the 2024 AWI of $69,846.57 to the 1977 AWI of $9,779.44. The AWI rose 4.84 percent in 2024 and 4.43 percent in 2023. The contribution and benefit base moves on the same index: $168,600 in 2024, $176,100 in 2025, $184,500 in 2026.

Once in payment, a benefit escalates at CPI-W and nothing else. A cohort claiming in 2026 enters on a wage-indexed schedule that grew 4.84 percent on the latest AWI reading, while a cohort that claimed in 2016 has compounded at price growth alone for a decade. That wedge is not a measurement error and no consumer price index corrects it. It is a design choice: each entering cohort shares in productivity growth, each retired cohort is insured against price growth only.

A retiree fifteen years into benefits comparing a check against a neighbor who just claimed is observing this wedge, not a CPI-W deficiency. The two effects are reported as one complaint and have entirely different remedies.

What the Market Is Not Pricing

Viewed as a fiscal rather than a household question, the framing inverts. The COLA is an automatic, backward-looking expenditure escalator on the largest single line of the federal budget, and its key property is not accuracy but the mechanical transmission of last year's inflation into this year's outlays. Three implications follow.

1. An energy-led shock reaches the federal outlay line with a long, predictable lag. The 14.7 percent twelve-month energy increase recorded in July 2026 sits inside the CPI-W series that will define the next window. Energy shocks are conventionally treated as transitory, but through the COLA channel a transitory shock landing in a third quarter becomes a permanent addition to the benefit base.
2. The escalator is ratchet-shaped. Benefits cannot be adjusted downward; the formula reads "the percentage increase, if any." Zero adjustments occurred in the years following the 2009, 2010 and 2015 measurement windows. A deflationary quarter produces zero, not a reduction, so the level effect of an inflation spike is retained permanently even if prices later fall.
3. It compounds against trust fund arithmetic. The 2026 Trustees Report projects Old-Age and Survivors Insurance reserve depletion in the fourth quarter of 2032, with 78 percent of scheduled benefits payable, and combined OASDI depletion in the third quarter of 2034, with 83 percent payable. The OASI date moved one quarter earlier than the prior projection. An argument to switch the escalator to a faster-growing index is therefore an argument to move that date closer — a trade-off rarely stated when the basket case is made.

Where the Mechanism Runs Thin

• It does not explain individual budget stress. A national index is a weighted average across geography and tenure. A renter facing a 3.2 percent shelter increase and an owner with no mortgage experience the same COLA very differently. Housing tenure alone probably dominates index choice for most households.
• It does not establish that CPI-W is the correct index. Nothing here defends CPI-W on the merits. The claim is narrower: in the current category mix, substituting the elderly research index would not close the gap. CPI-W can remain a poor conceptual fit for a retired population and that still be true.
• It does not apply to the zero-COLA years. Where the third-quarter comparison produced no increase, the binding mechanism was the absence of any adjustment, not the lag or the premium net-out. Hold harmless behaves differently in those years.
• It does not identify the driver of the funding gap. The escalator interacts with trust fund arithmetic but does not drive it. The dominant inputs are demographic, and reweighting the escalator is a second-order lever against them.
• It cannot be verified for the current benefit year. The 3.10 percent figure uses a single July reading, not a quarterly average, and the remaining months of 2026 are unknown. The direction is established; the magnitude is not.

What to Watch Next Week

• The next CPI-W print, unrounded. The headline release is CPI-U; the COLA runs on CPI-W, and the two diverge when energy and transportation weights matter. Read the index level, not the CPI-U percentage.
• The energy contribution. Energy at 14.7 percent over twelve months is the single largest swing factor in whether the next third-quarter window prints hot. Watch whether the monthly energy declines seen in July continue.
• Medical care services relative to headline. At 2.7 percent against a 3.4 percent headline, the retiree-basket argument is currently inverted. A reversal in that spread would restore it, and the spread is the cleanest single test of the basket thesis.
• Appropriations status at BLS. The October 2025 non-publication set the precedent. A funding interruption spanning July through September would hit the measurement window directly, in a way interruptions in other months do not.
• Shelter deceleration. At 3.2 percent, shelter is the largest single weight and the slowest-moving component. Its trajectory sets the floor under the next window.

Concrete Framework — Step by Step

A monitoring checklist for tracking this as a policy or fiscal variable.

1. Anchor on the base, not the rate. Record the third-quarter CPI-W average that set the current adjustment — 317.265 for the 2026 benefit year. Every subsequent monthly reading is measured against it, and the running gap is the real-terms erosion to date.
2. Track the three window months separately. Only July, August and September enter the next calculation. A hot June is irrelevant to the formula. Build the running quarterly average as prints arrive rather than reacting to headline releases.
3. Net the premium before comparing anything. Convert both the gross adjustment and the Part B premium change to dollars at the relevant benefit level. Percentage-to-percentage comparisons hide the $17.90 monthly deduction.
4. Separate the two indexation questions. Is the concern the annual adjustment on a benefit in payment, or the level at which it was originally set? The first is a CPI-W question, the second an AWI and bend-point question. They have no overlap.
5. Check the sign of the basket argument before accepting it. Compare the twelve-month change in medical care services against all items. When medical care runs below headline, as at 2.7 versus 3.4 percent in July 2026, the elderly-basket substitution would have reduced the adjustment.
6. Hold the trade-off explicitly. Any proposal to accelerate the escalator has a counterpart in the depletion projection — fourth quarter 2032 for OASI at 78 percent payable, third quarter 2034 for OASDI at 83 percent. A proposal silent on those dates is incomplete.
7. Assign probabilities, not forecasts. The honest statement about the next adjustment is a range conditioned on the energy path and shelter deceleration. Two of the three window months are usually unknown when the commentary starts.

The complaint is legitimate. The diagnosis usually attached to it is not the one the data supports. A formula that measures three months and pays for twelve, sitting upstream of a premium deduction growing at three times its rate, produces most of the observed gap on its own — and does so whichever basket is used.

Disclaimer: this article is analysis of policy mechanics and published statistical series. It is not financial, tax, or benefits advice, and it is not a prediction of any future adjustment.