Unadjusted

BUREAU OF ECONOMIC SENTIMENT

Office of Consumer Measurement

METHODOLOGY NOTE — CSI RELEASE 2026-04

Prepared by: Office of Consumer Measurement, Division of Behavioral Indices Analyst of Record: R. Caldwell (GS-13, Step 4) Classification: Public Use After Embargo Period (Tues. 10:00 AM EST)

Summary of Index Values

The Consumer Sentiment Index (CSI) for April 2026 stands at 98.7, reflecting a 1.2-point increase from the revised March figure of 97.5. The Index of Current Economic Conditions rose 0.9 points to 101.3. The Index of Consumer Expectations rose 1.4 points to 96.8.

All values are seasonally adjusted unless otherwise noted.

Revision History

March 2026 CSI revised from 98.1 (preliminary) to 97.5 (final). Revision reflects updated response weighting. See Technical Note 2026-03-R for methodology. February 2026 CSI revised from 96.4 (preliminary) to 97.2 (final). January 2026 unchanged at 96.9.

METHODOLOGY REVISION MEMORANDUM 2026-02

Subject: Updated Weighting Procedure for Housing Cost Perception, Effective January 2026 Release

From: Division of Behavioral Indices, Office of Consumer Measurement To: Bureau Director, Division Chiefs, Internal Distribution

Beginning with the January 2026 release, the weighting procedure for survey responses related to housing cost perception has been updated to reflect current market conditions. Previously, respondent assessments of housing affordability were weighted equally across income quintiles. Under the revised procedure, responses are weighted according to homeownership probability within each quintile, as determined by the Bureau’s Residential Attainment Model (RAM-IV).

The practical effect of this change is to increase the weight of respondents in the top two income quintiles, whose housing cost perceptions more closely correlate with actual market transactions, and to decrease the weight of respondents in the lower three quintiles, whose perceptions are influenced by rental market conditions not fully captured by the Case-Shiller Index.¹

This adjustment brings the CSI housing component into closer alignment with the Federal Reserve’s preferred measure of shelter inflation (OER) and reduces month-to-month volatility in the index.

No change to headline methodology. No change to survey instrument.

Impact Assessment: Applying the revised weighting retroactively to CY2025 data raises the average annual CSI by approximately 2.3 points. This is within the normal range of methodological revision effects and does not constitute a break in series.

INTERNAL COMMUNICATION — NOT FOR DISTRIBUTION

From: R. Caldwell To: J. Okafor, Division Chief Date: February 14, 2026 Re: Weighting revision impact on Q1 estimates

Jim —

Per your request, I’ve completed the impact assessment for the housing weight revision. The memo is attached. Short version: applying new weights to CY2025 data adds ~2.3 points to the annual average. The effect is larger in Q3 and Q4 2025, when rental cost perceptions were driving most of the downward pressure on the headline number.

I want to flag one thing that won’t appear in the memo. When I ran the revised weights against raw survey responses from December 2025, I noticed that the unadjusted bottom-quintile sentiment figure was 61.2. The adjusted figure, after the new weighting, is 73.8. The headline CSI for December, with all adjustments applied, was 96.4.

I’m not suggesting the methodology is incorrect. Each adjustment has a technical basis. I’m noting the distance.

— R

FILE: CSI_PARALLEL_v3.xlsx (Personal Device — Not Bureau Property)

Sheet 1: Unadjusted Index — Monthly Values

Sheet 2: Methodology Differences

The unadjusted index uses the Bureau’s pre-2024 weighting methodology, before the following revisions were applied:

1. Housing cost perception reweighting (2024-Q2): Reduced weight of bottom three quintiles. Effect: +2.3 annual avg.
2. Employment quality substitution (2024-Q4): Replaced “satisfaction with current employment” with “employment status” as primary labor input. Respondents working part-time, gig, or contract classified as “employed” without quality adjustment. Effect: +1.8 annual avg.
3. Healthcare cost exclusion (2025-Q1): Reclassified healthcare spending perception as “structural” rather than “cyclical,” removing it from the sentiment calculation entirely. Effect: +3.1 annual avg.
4. Digital consumption credit (2025-Q3): Added imputed value of free digital services (social media, search, streaming ad-supported tiers) to consumer welfare estimate, offsetting negative sentiment from price increases in physical goods. Effect: +4.2 annual avg.
5. Response recency bias correction (2026-Q1): Applied temporal smoothing to survey responses, reducing the weight of “recent negative experiences” relative to “general economic outlook.” Rationale: respondents overweight recent events. Effect: +2.1 annual avg.

Cumulative effect of adjustments 1-5: +13.5 points.

The official CSI and the unadjusted CSI diverged by 3.4 points in January 2025. They now diverge by 31.5 points.

Both are defensible. Both use the same raw survey data. Both are produced by the same analyst.

Sheet 3: Notes (Personal)

I don’t keep this spreadsheet because I’m brave. I keep it because I can’t stop. The way you keep scratching at something under the skin even when you know scratching makes it worse. Every Tuesday before the release I open this file and run the same data through two sets of weights and watch the gap get wider and the gap is the size of reality and I close the file and go to bed and in the morning I badge into the building and sit at my desk and produce the number I am paid to produce.

The official number is not wrong.

Say it again.

The official number is not wrong. Each adjustment has a technical justification. Each justification was reviewed by the methodology committee. The methodology committee reports to the division chief. The division chief reports to the Bureau director. The Bureau director reports to the Secretary of Commerce. No one in this chain has committed fraud. No one has lied. The number is the output of a process, and the process has been refined — continuously, incrementally, always in the direction of better alignment with — with what?

With a version of the economy in which things are fine.

METHODOLOGY NOTE — CSI RELEASE 2026-05

Prepared by: Office of Consumer Measurement, Division of Behavioral Indices Analyst of Record: R. Caldwell (GS-13, Step 4)

Summary of Index Values

The Consumer Sentiment Index for May 2026 stands at 99.4, the highest reading since November 2019.

EXPENSE LOG — R. CALDWELL — PERSONAL

Week of April 28, 2026

Metro fare (monthly pass): $142.00. Was $98.00 in 2022. The Bureau classifies this under “transportation services — public transit (urban)” and applies hedonic adjustment for service quality improvements including real-time arrival tracking and contactless payment. The app tells me my train is four minutes away. This is useful. I am not sure it is worth $44.

Lunch (desk, Mon-Fri): $11.40 average. The container is smaller than it was last year. I measured it once with a ruler from my desk drawer — the footprint shrank from 7.5” x 5” to 7” x 4.5”, a 16% reduction in area. The price increased 3% in the same period. In the CPI, this item has been reclassified from “meals, lunch, prepared, workplace” to “meals, lunch, prepared, premium convenience,” a different product category with a different baseline, so the year-over-year comparison resets. The Bureau’s methodology note describes this as “accounting for product evolution in the prepared foods segment.”

Prescription copay (atorvastatin): $45.00. Was $15.00 under previous formulary. The increase reflects a plan design change, not a drug price increase, so it is excluded from the health component of the CPI. The plan design changed because the insurer reclassified the drug from Tier 1 to Tier 2. The insurer reclassified the drug because the manufacturer increased the wholesale price. The manufacturer’s price increase IS captured in the Producer Price Index, which is a different index, maintained by a different division, published on a different Tuesday. The two Tuesdays do not talk to each other.

Rent: $1,840. Lease renewal in June. Landlord’s letter says $2,050. I have not opened the spreadsheet to calculate what this does to my unadjusted index because I already know.

Grocery bill (weekly average): $87.30. This number is almost useless because the composition changes every week — I buy what is on sale, which means I buy what the store has decided to discount, which means my diet is an artifact of pricing strategy. The Bureau calls this substitution. When the price of beef goes up twenty percent and I buy chicken, the CPI does not register that as inflation. It registers it as consumer choice. I chose chicken. I was not forced into chicken by the price of beef. The methodology assumes agency at the exact point where agency has been removed. I wrote a paper about this in graduate school, eleven years ago. The paper received a B+. The professor’s marginal note, which I still have, reads: “Technically sound but overstates the normative implications of a modeling decision.”

A modeling decision. Eleven years later the modeling decision is lunch in a smaller container and chicken instead of beef and a copay that tripled and a rent increase I cannot absorb and none of it, not one line of it, will appear in the index I produce on Tuesday because each of these experiences has been modeled away, classified as structural or reclassified as a different product or adjusted for quality or smoothed for recency, and the modeling decisions are individually reasonable and collectively they describe a country where I am doing fine.

BUREAU OF ECONOMIC SENTIMENT — CONFERENCE PRESENTATION (SLIDE EXCERPT)

Title Slide: Consumer Confidence in the Post-Pandemic Economy: Evidence of Sustained Recovery Presenter: Division of Behavioral Indices, Office of Consumer Measurement Event: Interagency Economic Indicators Summit, May 2026

Slide 7: CSI Trend 2020-2026

[Bar chart. Blue-to-green gradient. Bars ascending from left to right. Y-axis: Index Value (85-105). X-axis: Quarterly averages, Q1 2020 through Q2 2026. The bars reach 100 for the first time in Q2 2026. A dotted horizontal line at 100 is labeled “Historical Average.” An annotation reads: “Recovery complete.”]

I made this slide. I selected the color gradient. I chose the y-axis range — 85 to 105 — which makes the recovery look steep and dramatic. If the y-axis started at 0, the line would be nearly flat, a slight ripple across the top of the chart, indistinguishable from noise. But no one starts the y-axis at 0 for an index. That would be misleading in the other direction. The axis range is a framing decision. Every chart is a framing decision. I was trained to make these decisions and I make them well.

The slide does not show the unadjusted index. If it did, the bars would descend from left to right, reaching 64.1 in Q2 2026, the lowest reading in the unadjusted series. The two lines would form an X, crossing somewhere in early 2025, diverging in opposite directions. It would be a striking visual. I have made it, on my laptop, late at night. I have stared at the X. The place where the lines cross is the place where the methodology began to matter more than the data, and I can identify it to the quarter, and the identification changes nothing.

SURVEY RESPONSE — RAW DATA (Selected)

Respondent 4,271 — April 2026 Wave

Region: South Atlantic. Income quintile: 2. Age: 44. Employment: Part-time (2 positions). Housing: Renter.

Q1: Would you say that you and your family are better off or worse off financially than you were a year ago? A: Worse off.

Q2: Now looking ahead — do you think that a year from now you and your family will be better off financially, or worse off, or just about the same? A: About the same.

Q3: Now turning to business conditions in the country as a whole — do you think that during the next twelve months we’ll have good times financially, or bad times, or what? A: I don’t know. I see the numbers and they say one thing and I go to the store and it’s another thing. My daughter needs braces and the orthodontist doesn’t take our insurance anymore. But they say the economy is strong. I don’t know what economy they’re looking at.

Q4: Looking ahead, which would you say is more likely — that in the country as a whole we’ll have continuous good times during the next five years or so, or that we will have periods of widespread unemployment or depression, or what? A: [Response coded as N/A — respondent asked clarifying question: “Good times for who?” Interviewer unable to resolve. Response excluded per Protocol 7, Ambiguous Response Handling.]²

Q5: About the big things people buy for their homes — such as furniture, a refrigerator, stove, television, and things like that — generally speaking, do you think now is a good or bad time for people to buy major household items? A: Bad time. But we need a refrigerator because ours broke, so I guess we’re buying one anyway.

Weighting applied: Under pre-2024 methodology, respondent contributes 1.0 weight to composite. Under current methodology, respondent contributes 0.43 weight (adjustments: income quintile reweighting [-0.22], renter status structural classification [-0.18], ambiguous response exclusion Q4 [-0.17]).

More than half of this person’s response has been adjusted away. The Bureau’s methodology note calls this “reducing noise to improve signal fidelity.” The signal, in this case, is the absence of the respondent’s actual experience.

INTERNAL COMMUNICATION — NOT FOR DISTRIBUTION

From: J. Okafor, Division Chief To: R. Caldwell Date: May 12, 2026 Re: Re: Re: CSI Methodology — Concerns

Rachel,

I’ve read your memo carefully. I understand your concerns about the cumulative direction of recent methodological changes. You raise valid technical points and I appreciate your thoroughness.

I want to be direct with you. The Bureau’s methodology is reviewed annually by an independent advisory board. The board includes academic economists from four universities and two former Fed governors. The most recent review, completed in January, concluded that all revisions since 2024 are “consistent with best practices in survey methodology and reflect appropriate adaptation to evolving economic conditions.”

You are a GS-13 analyst. You are good at your job. You produce reliable work. But the methodology is not your decision. It is not mine. It is the output of a process that is larger than either of us.

I’m not asking you to agree. I’m asking you to understand the limits of your role.

The May number looks strong. Let’s have it ready for Tuesday.

— Jim

CSI_PARALLEL_v3.xlsx — Sheet 3: Notes (continued)

Jim’s email is correct. My role has limits. I am an analyst of record. I produce the number. The number is reviewed. The number is released. The number enters the world as fact, and by the time it enters the world I am already working on next month’s number, and the process is continuous, and the process is my job, and my job is my lease and my prescription copay and my Metro pass, and all of these things are denominated in the same economy the number describes, so my ability to question the number is funded by the number.

I went to the cafeteria today. I’ve been avoiding it since December, eating at my desk — the $11.40 containers. But today the elevator opened on the wrong floor and I walked through the atrium and the cafeteria was right there, the way it always is, with its salad bar and its rotating soup and its tables by the window that look out onto Constitution Avenue.

There was a meeting in the conference room adjacent to the cafeteria. Through the glass I could see a PowerPoint. Slide 14 of something. A bar chart. The bars were going up. The color scheme was blue-to-green, a gradient that implies improvement. I could not read the axis labels from where I was standing but I knew what they said the way you know the lyrics to a song you hate — the bars were some version of the number I produce, deployed in some version of the briefing I have seen given forty times, in which an economist whose title is longer than mine explains to people whose titles are longer still that the consumer feels good about the economy, that confidence is holding, that the indicators support continued optimism.

The indicators.

I bought a coffee and went back to my desk. The coffee was $4.75. It was $3.50 when I started here. The Bureau counts this as a different product because the cup size changed from 12 oz to 10 oz, so there is no year-over-year comparison. The coffee is the same coffee. The cup is smaller. The price is higher. The index registers neither fact.

CROSS-REFERENCE NOTE — PERSONAL

Costa Rica. GDP per capita: $13,420. Life expectancy: 80.8 years. United States. GDP per capita: $76,330. Life expectancy: 77.5 years.

I put it in the spreadsheet. It doesn’t belong there. It has no methodological relationship to consumer sentiment. But the spreadsheet is the only place where I keep numbers that mean what they say.

MARGIN ANNOTATION — PRINTED DRAFT OF METHODOLOGY REVISION 2026-03

Found in analyst’s desk drawer during routine office reassignment, August 2026. Annotations in blue ballpoint, analyst’s handwriting confirmed by HR file signature sample.

Printed text: “The digital consumption credit reflects the Bureau’s recognition that consumer welfare in the contemporary economy includes non-monetary benefits derived from digital platforms. Imputed values are calculated using the Bureau’s Digital Welfare Estimation Model (DWEM-II), which assigns dollar-equivalent values to time spent on ad-supported digital services based on revealed preference analysis.”

Margin note (blue ink): Time spent on ad-supported services. Revealed preference. A woman in the bottom quintile scrolls her phone for forty minutes on a bus because the bus takes forty minutes and she has already worked one shift and is going to another and the scrolling is not a preference it is the absence of an alternative and the model counts it as $3.20 of imputed welfare that offsets a price increase in gasoline or bread or electricity and the index moves up by a fraction of a point and I am the one who moves it.

Printed text: “DWEM-II has been validated against consumer time-use surveys conducted by the Bureau of Labor Statistics and shows strong correlation (r = 0.81) between imputed digital welfare values and self-reported life satisfaction among digital platform users.”

Margin note (blue ink): Among users. Who are the non-users? What is their self-reported life satisfaction? The validation study excludes respondents without broadband access (14% of bottom quintile). Excluding them improves the correlation. Improving the correlation validates the model. Validating the model justifies the credit.

Margin note (blue ink, different pen pressure, possibly later): I should not be writing in the margins of official documents. I should not be keeping a parallel index. I should not be lying awake calculating the gap. I should not be doing any of this and I cannot stop and the cannot-stopping is not courage, it is the same compulsion that makes me check the decimal places twice, that makes me re-derive the seasonal factors by hand when the software has already computed them, it is the part of me that was trained to care about the number being right and the part of me that cares about the number being right is the same part that produces the number that is wrong. The training does not distinguish. The caring does not distinguish. I am a precise instrument pointed at the wrong object and the precision is the problem.

METHODOLOGY NOTE — CSI RELEASE 2026-06

Prepared by: Office of Consumer Measurement, Division of Behavioral Indices Analyst of Record: R. Caldwell (GS-13, Step 4)

Summary of Index Values

The Consumer Sentiment Index for June 2026 stands at 100.2, crossing the 100 threshold for the first time since February 2020.

This milestone reflects sustained improvement in consumer confidence across multiple dimensions, including employment conditions, income expectations, and purchasing attitudes. The index has risen 5.8 points over the past six months, the longest sustained increase since Q3 2017.

All values are seasonally adjusted unless otherwise noted.

Technical Note on Survey Response Rates

The overall response rate for the June 2026 wave was 34.2%, compared to 51.8% in June 2020 and 68.4% in June 2010. Response rates have declined across all income quintiles, but the decline is steepest in the bottom two quintiles (Q1: 22.1%, Q2: 28.3%) compared to the top two (Q4: 41.7%, Q5: 48.9%).

Declining response rates in lower income quintiles are consistent with national trends in survey participation and do not, in the judgment of the methodology committee, introduce systematic bias, as non-response adjustments are applied per Bureau standard practice.³

REVISION HISTORY — CSI RELEASE 2026-06 (DRAFT)

Version 1.0 — Submitted 07:42 AM, June 30, 2026 Standard release format. All values verified against source data.

Version 1.1 — Submitted 08:15 AM, June 30, 2026 Corrected rounding error in Table 3, Row 12 (Index of Consumer Expectations, unadjusted).

Version 1.2 — Submitted 08:31 AM, June 30, 2026 Revised footnote 3. Previous version read: “Non-response adjustment assumes that non-respondents within a demographic cell hold, on average, the same sentiment as respondents in that cell. This assumption has not been validated against external data sources since 2018.” Current version reads: “Non-response adjustment is applied per Bureau standard practice.” Change made at the recommendation of the Division Chief following pre-release review.

Version 1.3 (FINAL) — Submitted 09:02 AM, June 30, 2026 Technical Note on Survey Response Rates: replaced “the decline is steepest in the bottom two quintiles” with “response rates vary across demographic groups.” Removed quintile-specific response rates from body text. Moved to Appendix D (available upon request).

CSI_PARALLEL_v3.xlsx — Sheet 1 (continued)

The official index crossed 100. Crossing 100 is a symbolic threshold. It means — in the language of the index — that consumers feel better about the economy than the historical average. It will be in the press release. It will be in the briefing. The Secretary of Commerce will mention it in a speech. The number 100.2 will enter the public record as a fact about how Americans feel.

The unadjusted index is 64.1. This is not a recession number. This is not a downturn number. This is a number without a name. The Bureau’s taxonomy does not include a category for a number this low produced by a methodology this old. It is as if the number exists in a room no one has opened in years, and the room is locked, and I have the key, and the key is on a laptop in my apartment that I open every Tuesday night and close every Tuesday night and never show to anyone.

DAILY RECORD — R. CALDWELL — JULY 1, 2026

Badge in: 8:47 AM. The lobby scanner made its small confirming sound. The sound means: you belong here. I have heard this sound approximately 1,200 times. Elevator to 6. The elevator has a screen that shows the Bureau’s latest releases. Today it displays: CSI REACHES 100.2 — HIGHEST SINCE FEBRUARY 2020. The font is Calibri.

Desk by 8:53. Computer on. Two monitors. Left monitor: Bureau network, secure. Right monitor: email, calendar, shared drives. I have a third screen in my bag — the laptop — but I do not take it out at work. The laptop stays in the bag the way a flask stays in a coat, hidden not because anyone is checking but because the hiding is part of the thing.

9:00 AM. Stand-up meeting. Jim asks each analyst for status. My status is: June release final, no outstanding items, beginning July survey data intake. This is true. It is also true that I have a spreadsheet on a laptop in my bag that says the number we released yesterday is 36 points higher than the number I calculate when I do not adjust away the people who are drowning, but that is not my status, that is something else, something I do not have a word for because the vocabulary of my profession was designed for situations in which the methodology is either correct or incorrect and mine is both.

11:30 AM. A new email from the Office of Management and Budget. Subject line: CSI Milestone — Messaging Coordination. The OMB wants to coordinate with the Bureau on “framing the positive trajectory of consumer confidence indicators” in advance of the midterm briefing cycle. They are asking for three bullet points. Three bullet points about how good the number is.

I write the bullet points. They take four minutes. They are accurate. Every word corresponds to a data point. Every data point was produced by the methodology. The methodology is defensible. The bullet points are a lie in the way that a mirror in a funhouse is a lie — the reflection is real, the glass is real, the distortion is a property of the curvature, and the curvature was chosen.

I write: “Consumer sentiment reached 100.2 in June, the highest reading in over six years, reflecting broad-based improvement in household financial assessments and forward-looking expectations.”

I do not write: The improvement is broad-based in the way that a survey with a 22% response rate in the bottom quintile is broad-based. I do not write: The forward-looking expectations are forward-looking in the way that asking people who answer their phones how they feel about the future is forward-looking.

12:15 PM. Lunch at my desk. The smaller container. I eat without tasting, and inside the not-tasting it is possible to function, and functioning is what I do.

4:30 PM. I finish the data intake protocol for July. I proofread the previous month’s Appendix D — the one with the response rates that were moved from the body text to the appendix by request. Appendix D is six pages long and is available upon request and nobody has requested it. It has been downloaded once. The download log does not identify the user. It was me. I downloaded it from my personal laptop on a Tuesday night to check whether the version they published matched the version I submitted. It did not. The quintile-specific rates were rounded to whole numbers, removing the decimal precision that would let a reader calculate the exact non-response differential. The rounding was consistent with Bureau publication standards. The precision I had included was, technically, excessive.

5:22 PM. Badge out. The scanner makes the same sound leaving as entering. It does not distinguish between arrival and departure. It registers presence. I take the Metro. The train is four minutes away. The app tells me so. I paid $44 more per month for the privilege of knowing this. At home I open the laptop and I do not open the spreadsheet. I cook dinner — rice, an egg, greens from the store where the greens cost more than they did and the Bureau says they don’t because the greens are now classified differently, a different cultivar in a different product category, and I eat the greens that are a different product and they taste exactly the same.

I do not open the spreadsheet.

I open the spreadsheet.

July intake data is starting to come in. Early responses only — the people who answer quickly, who tend to be higher-income, more engaged, more optimistic. The early data skews positive. It always skews positive. The final number, when the slower respondents are counted, will be lower. But the preliminary estimate, released first, will carry the early skew, and the revision, released a month later, will be smaller than the distance between early and late respondents warrants, because the revision formula applies temporal smoothing, because respondents overweight recent events, because the methodology has determined that people’s actual feelings about their actual lives are a source of noise.

The preliminary July CSI will be above 100. I know this the way I know that my rent is going up in June and my copay is going up in August and my lunch container is going to get smaller again and none of these things will appear in the index I produce because the index has been adjusted and I have not been adjusted and the adjustment is the normal state and I am the outlier and the methodology was designed to smooth outliers and I am standing in the gap between the two numbers — the one I produce and the one I keep — and the gap is 36.1 points wide and I can feel it in my body the way you feel weather, the way you feel the particular pressure of a room where everyone has agreed not to mention the thing that is happening, and the thing that is happening is that the number is wrong and the number is not wrong and both of these are true and I go to work in the morning.