In this article
After this article, you can
- Distinguish two separate medians from paired answers by the same people.
- Identify when a household-level conclusion needs linked respondent data.
Terms at hand: Inflation · Economic Calendar · Central banks · Consumer Price Index (CPI) · Hawkish vs Dovish
Michigan’s report contains sentiment indices and separate inflation expectations. Check the series, units and release vintage before interpreting the dollar. Our fictional month separates monthly change, preliminary-to-final revision and surprise against a forecast for the correct event. The same median can also conceal very different individual responses.
What the Michigan consumer sentiment index actually is
The University of Michigan’s Surveys of Consumers asks US adults about personal finances, buying conditions and economic expectations. It transitioned from telephone interviews to web surveys in April–July 2024; collection has been web-only since the July 2024 preliminary release. The target is adults aged 18 or older; the address-based sample uses postal addresses in the 48 coterminous states and the District of Columbia, rather than every possible US household population.
Mark the method transition in long comparisons. The April 2024 announcement describes gradual changeover and question-dependent method effects; the July 2026 special report revisits historical comparisons. Do not assume every level change is economic, or that changing method invalidates the entire series.
Surveyed sentiment and reported spending describe different things. Confidence may accompany spending changes, but the survey does not establish how every consumer will behave or prove a subsequent business-cycle turn. Compare the answers with hard data and retain uncertainty.
Preliminary and final readings — twice a month
Preliminary and final are two vintages for the same month. Preliminary uses an earlier part of that month’s collection; final includes more responses. They are not two different months or independent censuses. Check the official dates: the 2026 calendar lists February 6 and 20, which already disproves a rigid “second and last Friday” rule. Convert the actual release from 10:00 Eastern to your zone.
The provider updates its main release page with a summary and key indicators at publication; the detailed data site follows a separate public embargo. A missing month in one data table does not prove the release has not occurred. Keep the actual announcement and both vintages without overwriting the earlier economic calendar entry.
Inflation horizons and units
Year-ahead expectations concern general prices over the next 12 months. The longer-horizon question asks for average percentage change per year over the next 5–10 years, not the cumulative change over ten years. These answers are expectations, not realised CPI inflation or a central-bank forecast. The five sentiment-index questions cover finances, business conditions and durable-purchase conditions; inflation questions are not extra arithmetic components of ICS.
Inflation expectations can influence economic decisions, as then-chair Powell discussed in August 2022. They do not make inflation entirely self-fulfilling or determine a current policy decision. Keep household expectations distinct from realised inflation and record any dollar interpretation as conditional. See the article on the CPI release and the dollar for the different price measure.
Two medians cannot identify the same respondents
In two invented four-person samples, short- and longer-horizon expectations have identical sorted values: 1%, 1%, 5%, 5%. Each axis has median 3%. Only linking both answers from each person reveals whether high values occur together.
Scroll sideways to see the full diagram
To claim that the same households expect high inflation at both horizons, you need answers linked by respondent. Two medians alone are insufficient; here, even the complete separate distributions are insufficient.
Assumptions: Each of the four people has equal weight and supplies both answers. For an even observation count, this school example defines the median as the mean of the middle two values: (1 + 5) / 2. We choose the “above 4%” threshold ourselves. No probability or inflation forecast is calculated. Official material: Michigan — inflation-expectation medians and distribution.
Current conditions versus expectations
Read the index’s actual questions. Current conditions concern personal finances compared with a year earlier and durable-purchase conditions; expectations concern personal finances a year ahead and business conditions over a year and about five years. ICS uses scaled response balances against a 1966 base: 60 index points do not mean 60% of respondents are happy. A change of two points is also not a change of 2%.
| Measure | Prior final | Prelim consensus | Prelim | Final |
|---|---|---|---|---|
| ICS, index points | 60.0 | 62.0 | 58.0 | 59.0 |
| Year-ahead expectation, % | 3.1 | 3.0 | 3.4 | 3.3 |
| 5–10-year expectation, %/year | 3.0 | 3.0 | 3.1 | 3.1 |
Prelim ICS surprise is 58 − 62 = −4 points; monthly change is 58 − 60 = −2, or about −3.33%. Final 59 revises prelim up 1 point but remains 1 below the prior month. If the separately recorded consensus before final is 58.5, final surprise is +0.5; comparing with stale 62 would give −3 and reverse its sign.
Year-ahead expectations move 3.1%→3.4%, a +0.3 percentage-point change; surprise versus 3.0% is +0.4 pp and final 3.3% revises prelim by −0.1 pp. These are expectations, not a 0.3% current monthly price rise. The table is entirely invented.
Read current conditions and expectations separately, using the survey’s actual questions. A weakening expectations component can motivate further investigation, but does not prove that spending will decline. Compare it with retail sales data using matching periods and release vintages.
Same median, different response distributions
Our school example has five equally weighted year-ahead expectation answers in each sample: A = 2%, 2%, 3%, 4%, 4%; B = 0%, 0%, 3%, 8%, 8%. It does not reproduce Michigan’s survey weights, imputation or median interpolation.
| Measure | A | B |
|---|---|---|
| Median | 3% | 3% |
| Arithmetic mean | 3% | 3.8% |
| Minimum–maximum | 2%–4% | 0%–8% |
| Range | 2 pp | 8 pp |
Four of five ordered answers change while the middle stays at 3%. Range is a simple school measure here, not Michigan’s published interquartile spread. Two separate median values also do not link responses by person; the existing paired-answer diagram shows why. This counterexample does not prove actual expectations lost their anchor.
Before describing expectations as anchored, identify whether the evidence contains only a median or also response distributions at the same horizon and vintage. You cannot calculate dispersion from one 3% figure. For related measurement distinctions, continue in fundamental analysis.
Common misunderstandings
In our table, weaker sentiment coexists with higher price expectations. The survey methodology contains no conversion from an ICS point to a specified number of pips. Assess a policy hypothesis separately from the reported answers, using Fed communication and other data. A headline alone does not establish which component caused a dollar response.
For a price observation, predefine the instrument, price source and window, and record simultaneous releases. Retain quiet and contrary cases. Preliminary-to-final revision is a different comparison from final surprise and month-to-month change; do not select the price minute or forecast vintage afterward to fit a thesis. A recording sheet can be completed without opening a position.
Your next step
- Find both official release dates. Record the preliminary and final publication times and convert them into your own zone. Note other releases that day without assuming CPI or employment must dominate every market reaction.
- Record several releases for review. Save sentiment, its components and both inflation-expectation horizons with their vintages, then specify the currency-observation window. Three months can illustrate your recording process; it cannot establish which component most often causes a currency move.
- Compare sentiment with consumption data. Match periods and definitions when reviewing the survey alongside retail sales. A divergence is a finding to investigate, not by itself a proven early signal of a business-cycle turning point.
Check your understanding
One short exercise
About 2 min · you can try again
Each axis has 1%, 1%, 5%, 5%. A pairs are (1,1), (1,1), (5,5), (5,5); B pairs are (1,5), (1,5), (5,1), (5,1). How many people exceed 4% at both horizons?
- A: 2; B: 0.
- Two in each sample.
- Zero in each, because both medians are 3%.
Check the answer and explanation
A: 2; B: 0.
The condition must hold for both answers from one person. Everyone in B has one answer of 1%.
Frequently asked
How do Michigan inflation expectations differ from the CPI inflation reading itself?
Michigan records what respondents expect, while CPI measures observed price changes in its covered consumer basket. Year-ahead expectations refer to the next 12 months; the 5–10-year question asks for average annual change, not a ten-year total. These inflation answers are separate from the five questions making ICS. Keep expectation percentages and sentiment index points separate; an ICS level of 60 is not 60% satisfied respondents.
When and how often is the Michigan consumer sentiment released?
Use the official calendar for two vintages of the same month. The 2026 schedule lists February 6 and 20, rather than a fixed second/last-Friday rule; the provider lists release at 10:00 Eastern. Convert for the actual date and your location. Preliminary reflects earlier collection, final more responses, but final is not automatically less relevant. The main release page and detailed data site also have different availability rules.
Does an unchanged median mean unchanged individual expectations?
Not necessarily. In our invented five-answer samples, A = 2,2,3,4,4% and B = 0,0,3,8,8%. Both medians are 3%, yet arithmetic means are 3%/3.8% and ranges 2/8 pp. Four answers differ. These are equally weighted school distributions, not Michigan’s estimation method or evidence of actual de-anchoring. The paired-answer diagram separately shows that even two complete marginal distributions cannot identify which people are high at both horizons.
Does an upward final revision mean positive surprise and monthly growth?
No. Revision compares final with prelim; surprise compares final with its own pre-release consensus; monthly change uses the prior month. In our separate exercise, prior final 60.0, prelim 58.0, final consensus 59.0 and final 59.2 give +1.2 revision, +0.2 surprise and −0.8 monthly points (about −1.33%). Those signs coexist. They neither measure realised inflation nor settle the next USD move.
Sources & bibliography
-
University of Michigan Surveys of Consumers — data tables (Index of Consumer Sentiment and inflation expectations) · University of Michigan consumer-survey tables, including sentiment components and inflation expectations. www.sca.isr.umich.edu ↗
-
Federal Reserve Monetary Policy - the role of inflation expectations · Federal Reserve material on inflation expectations, policy objectives and interest-rate decisions. www.federalreserve.gov ↗
-
The Conference Board Consumer Confidence Index — methodology · US consumer-confidence survey methodology, useful for comparison with the University of Michigan survey. www.conference-board.org ↗
-
University of Michigan Index calculations — the five sentiment questions · Defines relative response scores and scaling. Inflation-expectation questions are separate; index points are not respondent percentages. data.sca.isr.umich.edu ↗
-
University of Michigan; Richard Curtin Procedure to estimate price expectations — January 1996 methodological memo · Distinguishes central tendency and dispersion, with historical imputation/interpolation procedures. The article’s equal-weight examples do not replicate the survey estimator. data.sca.isr.umich.edu ↗
-
University of Michigan Surveys of Consumers FAQ — samples, vintages and data availability · Describes the postal sampling frame and preliminary/final collection, with separate release-page and detailed-data availability. data.sca.isr.umich.edu ↗
