Apologies if this is more than anyone wanted to digest about consumer sentiment.
Here’s what else I can offer about the 2024 switch in the Index of Consumer Sentiment (ICS) survey from phone-based to web-based. Based on data that I received about the differences between phone-based ICS and web-based ICS, my back-of-the-spreadsheet estimate is that based on answers from all partisan identifications overall web-based ICS is about 11% lower than phone-based ICS. This is consistent with literature on survey methods indicating that survey respondents tend to give more upbeat answers when talking to a live person on the phone vs. answering questions on a computer.
For the same reason that Dr. Sahm uses 3-month unemployment rate averages in her Sahm rule recession indicator, I like to use 3-month ICS averages to reduce the possibility of spurious volatility in monthly surveys. When I compare the 3-month ICS average over the last 3 months of the phone-based ICS survey (Jan. to March 2024) with the August-to-October 2024 3-month average when ICS was entirely web-based, there was a -11% drop in ICS. I attribute that drop in reported ICS from winter 2024 to autumn 2024 just before the election as almost entirely due to the change in survey method from phone- to web-based.
But, when I look at 3 sub-groups based on party identification, the changes over that same period are: Democrats -8%, Republicans -19%, independents -14%. Not much was going in the economic data over this period. If anything, the principal indicator of interest during the 2024 election year – annual CPI inflation – improved a bit from 3.3% average Jan.-March to 2.5% August-October.
I can think of no reason why Republican ICS would have fallen by so much during this period – no reason why Republicans should have been more disappointed than Democrats or independents that inflation was not falling fast enough and no reason why Republicans would have been more affected by the change in survey method and given more downbeat answers to the web survey than Democrats or independents.
I do recognize that looking at 2 moments in time – Jan.-March and August-October – is not enough to draw firm conclusions.
I certainly hope that UMichigan Surveys of Consumers will provide more information about the change in survey method and the possible effects.
As you can see, I am something of an ICS nerd – some might say crank.
I sure do wish there were data looking back at the prior three decades before 17, especially for Democrats. The trend for both partisans does seem down.
We are also entering times where there is an ever larger disconnect between the investor economy and the paycheck economy.
You put your finger on the thing the whole sentiment debate is dancing around. The investor economy and the paycheck economy used to rise and fall more or less together. Somewhere along the way they came uncoupled, and now one can post records while the other feels like it's slipping — both true at the same time, which is exactly why people argue past each other about whether things are "good."
The piece you're replying to is measuring the feeling of that gap. What it can't quite reach is the mechanism underneath it: there's no longer anything that keeps the two economies tied to each other. When the paycheck side and the output side drift apart, nothing automatically pulls them back. So the gap just widens until it shows up as a mood nobody can quite explain.
That's why I think the pre-2017 history you're asking for would show a narrower gap, not a wider one — not because people were less partisan, but because the two economies hadn't separated as far yet. The partisanship is loud. The uncoupling is the part that actually moves the paycheck.
There is a modest drift of fewer R's and more D's over the past months (and the reverse was true during some of Biden's time).
This means that the timeseries data for subsamples are not the same sample sets, with all sorts of implications about trying to read the tea leaves.
In particular it is perhaps likely that some unhappy members of the party in control of the White House stop identifying as members of that party and become leaning independents.
This sort of movement might explain the odd artifact that R-leaning Independents have become happier over the past 6 months, and everyone else is flat (D's) or declining. Ie. if some former R's are now I-leaning-R, their opinions could have declined but still increased the average optimism of the I-leaning-Rs.
Second, the survey covers about 1000 people, so the 95% confidence margin of error is about 3%. A big danger with looking at subsamples (R, R-lean, I, D-lean, D) is that the MoE blows up.
I couldn't find the full demographic breakdown, but eyeballing chart 4 in the above linked doc, 2026 looks like 20% R, 30% D, so 50% I. True I usually maxes out at about 10%, so if we split the leaners in half you have :
30% D - MoE > 5.5%
20% D-lean MoE about 7%
10% True I MoE almost 10%
20% R-lean MoE about 7%
20% R MoE about 7%
To be clear, the numbers are guesses upon guesses, but the MoE calculation is correct.
So ... the full average is probably a much better indicator than it seems because it isn't perturbed as much by shifting identification and has a much tighter MoE.
Of course that says nothing about the vibecession effect.
On Nextdoor I just saw a series of attacks on democrats for the price of gasoline, including one on Biden and one on our governor. The Republicans have mastered attacks, distortion and misinformation, as well as much more in politics. The dems seem lost and leaderless.
People can be swayed by politics for a while, but every year Jesus doesn't show up, but the home insurance bill gets bigger, more people will end up swayed by reality.
Excellent analysis, as always. I knew that “In the summer of 2024, the Michigan Survey transitioned from a phone to an internet survey to reduce costs.” But, I did not know that “The transition coincided with a drop in the share of Republican respondents and an increase in the share of Democratic respondents. If the shares had remained at their level in early 2024, the current overall sentiment index would be 7 points higher.” In the technical report on the Index of Consumer Sentiment (ICS) website, "Methodological Transition to Web Surveys”, https://www.sca.isr.umich.edu/files/methodtransitionannouncement2024.pdf
there is a discussion of web respondents being less optimistic than phone respondents. There is definitely a break in the ICS data starting in April 2024 as the survey transitions to 100% web-based by July 2024. Anyone doing modeling work with ICS data that includes pre-2024-4 and post-2024-3 reported ICS numbers should think about making some sort of adjustment.
My own back-of-the-spreadsheet estimate is that May 2026 ICS reported as 44.8 from the web-based survey would be equivalent to 50.2 if the survey had been phone-based as was the case prior to April 2024. So, the all-time ICS low is still June 2022 = 50 at the height of the post-covid inflation surge. Ernie Tedeschi’s adjustment places May 2026 ICS higher at 53.7. See https://www.briefingbook.info/p/the-effect-of-online-interviews-on
It’s odd that the web-based ICS survey ended up with a higher share of Democrats. The conventional wisdom is that Republicans are a male-dominated group less likely to take the time to answer surveys in general, but also more web-oriented than talking-on-phone-oriented. You would think that Republican “bros” would be more likely to answer a web survey. Great work digging out that strange result, Dr. Sahm. I do hope UMichigan Surveys of Consumers engages with your work.
I've seen many methodological observations that R populations also skew older, less internet-savvy, and less inclined to respond honestly to surveys these days, phone or otherwise. It's very hard to assess how this all shakes out.
I, knowing nothing about opinion polling: is there some method, probably more intensive and time-consuming, to differentiate the partisan impact on a person’s response and their underlying opinion?
The Hassett-vs-Krugman fight over whose voters are "really" unhappy almost proves the author's deeper point. When the partisan gap is bigger than the income gap, the politics isn't the signal — it's the noise sitting on top of a signal both sides actually share.
Strip the party labels off and the thing underneath doesn't move. Since 1979 the economy's output per hour is up about 90 percent. Typical pay is up about 33 percent. Those two used to track together and then quietly came apart, and nothing was ever built to keep them tied. That's not a red or blue number. It's a design gap, and it shows up in a paycheck the same way regardless of how someone votes.
Which might be why sentiment "feels" off across the board even when the headline data looks fine. People are driving a system that pulls, and they feel the pull in their daily lives long before any survey catches it. The partisanship changes how they explain the feeling. It doesn't change the feeling.
Very thought-provoking, CS. I take your points, too. I try to focus on slope, directionality, and velocity of change over time in each separate political-affiliation cohort. Like you, I'm also concerned about how the shift to internet surveys affects the representativeness of the sample vis-a-vis the general population. This issue has arisen in other survey research.
the length of your rebuttal/argument is the crux of the problem and has remained so since you know who came down the escalator; it only takes short snippets of an argument - no matter how false- repeated over and over in different forms to stick in the mind of our short attention span public.
The media and non you know who supporters have yet to obtain the magic elixir to break the spell.
The 2 party system is for people too lazy to use an analytical critical thinking process. It's just a pandering contest for the party nomination, then finger pointing in the general election.
What a great deeper dive presentation on consumer sentiment, Dr Sahm!
If I may, a couple of thoughts...
In a Trump 'situation,' meaning the economy and all other ancillary components and metrics measuring our monetary and fiscal health and attitudes, it's not surprising that consumers are exhibiting a historical low in current and outlook sentiment.
IMO, the majority of Americans are honest, hardworking, people attempting to continually improve the quality of their familial situations. So, when they hear about the daily impeachable offenses perpetrated by Trump and the regime, essentially all of it self-enriching and none of it to improve America, with no accountability then it's easy to envision their feeling demoralized, frustrated and without meaningful purpose as their lives and freedoms are torn away
Frankly, if I were asked to provide sentiment on the future outlook of all things economy, I'd have to preface the answer as would an attorney, "it depends."The answer is bifurcated because it depends on whether Trump and the regime can be washed away by a blue tsunami or if we will be perpetually stuck with him. We already know he has no intention of leaving, so the illegal slush fund is his attempt to prepay participants for the next insurrection.
Thankfully, it appears that the fund will not be set up and will be crushed; the ballroom won't be built by taxpayers, neither will that ridiculous arch. Michael Wolff has stated that the midterms will determine whether Trump is finished OR if we are finished. Terrifying!
Anyway, for the intermediate term, perhaps the consumer sentiment should add two components that survey attitudes on two vectors:
1) An extension of the regime after the midterms if a blue tsunami fails.
2) The gestation of an actual presidency can occur as a result of successful midterms to oust the regime.
The sentiment gap between Democrats and Republicans is easy to explain - Republicans are in denial, no matter who is in the White House. After all, reality has a well known liberal bias.
Apologies if this is more than anyone wanted to digest about consumer sentiment.
Here’s what else I can offer about the 2024 switch in the Index of Consumer Sentiment (ICS) survey from phone-based to web-based. Based on data that I received about the differences between phone-based ICS and web-based ICS, my back-of-the-spreadsheet estimate is that based on answers from all partisan identifications overall web-based ICS is about 11% lower than phone-based ICS. This is consistent with literature on survey methods indicating that survey respondents tend to give more upbeat answers when talking to a live person on the phone vs. answering questions on a computer.
For the same reason that Dr. Sahm uses 3-month unemployment rate averages in her Sahm rule recession indicator, I like to use 3-month ICS averages to reduce the possibility of spurious volatility in monthly surveys. When I compare the 3-month ICS average over the last 3 months of the phone-based ICS survey (Jan. to March 2024) with the August-to-October 2024 3-month average when ICS was entirely web-based, there was a -11% drop in ICS. I attribute that drop in reported ICS from winter 2024 to autumn 2024 just before the election as almost entirely due to the change in survey method from phone- to web-based.
But, when I look at 3 sub-groups based on party identification, the changes over that same period are: Democrats -8%, Republicans -19%, independents -14%. Not much was going in the economic data over this period. If anything, the principal indicator of interest during the 2024 election year – annual CPI inflation – improved a bit from 3.3% average Jan.-March to 2.5% August-October.
I can think of no reason why Republican ICS would have fallen by so much during this period – no reason why Republicans should have been more disappointed than Democrats or independents that inflation was not falling fast enough and no reason why Republicans would have been more affected by the change in survey method and given more downbeat answers to the web survey than Democrats or independents.
I do recognize that looking at 2 moments in time – Jan.-March and August-October – is not enough to draw firm conclusions.
I certainly hope that UMichigan Surveys of Consumers will provide more information about the change in survey method and the possible effects.
As you can see, I am something of an ICS nerd – some might say crank.
Consider this : https://www.bigdatapoll.com/blog/trump-disapproval-rating-hits-new-high-in-march-amid-unpopular-war/
Republicans are so unhappy with Trump that they are no longer identifying as Republican.
It seems pretty reasonable that there would be an impact on ICS among the R's that are still R's
Very thoughtful post. I once worked at the Center for Survey Research at Indiana University and very much appreciate this peak into the UM survey.
I sure do wish there were data looking back at the prior three decades before 17, especially for Democrats. The trend for both partisans does seem down.
We are also entering times where there is an ever larger disconnect between the investor economy and the paycheck economy.
You put your finger on the thing the whole sentiment debate is dancing around. The investor economy and the paycheck economy used to rise and fall more or less together. Somewhere along the way they came uncoupled, and now one can post records while the other feels like it's slipping — both true at the same time, which is exactly why people argue past each other about whether things are "good."
The piece you're replying to is measuring the feeling of that gap. What it can't quite reach is the mechanism underneath it: there's no longer anything that keeps the two economies tied to each other. When the paycheck side and the output side drift apart, nothing automatically pulls them back. So the gap just widens until it shows up as a mood nobody can quite explain.
That's why I think the pre-2017 history you're asking for would show a narrower gap, not a wider one — not because people were less partisan, but because the two economies hadn't separated as far yet. The partisanship is loud. The uncoupling is the part that actually moves the paycheck.
A couple of random points.
First, the Michigan survey has documented that people shift their claimed political party. See Graph 4 in https://www.sca.isr.umich.edu/files/partisaneconomy202504.pdf
There is a modest drift of fewer R's and more D's over the past months (and the reverse was true during some of Biden's time).
This means that the timeseries data for subsamples are not the same sample sets, with all sorts of implications about trying to read the tea leaves.
In particular it is perhaps likely that some unhappy members of the party in control of the White House stop identifying as members of that party and become leaning independents.
This sort of movement might explain the odd artifact that R-leaning Independents have become happier over the past 6 months, and everyone else is flat (D's) or declining. Ie. if some former R's are now I-leaning-R, their opinions could have declined but still increased the average optimism of the I-leaning-Rs.
Second, the survey covers about 1000 people, so the 95% confidence margin of error is about 3%. A big danger with looking at subsamples (R, R-lean, I, D-lean, D) is that the MoE blows up.
I couldn't find the full demographic breakdown, but eyeballing chart 4 in the above linked doc, 2026 looks like 20% R, 30% D, so 50% I. True I usually maxes out at about 10%, so if we split the leaners in half you have :
30% D - MoE > 5.5%
20% D-lean MoE about 7%
10% True I MoE almost 10%
20% R-lean MoE about 7%
20% R MoE about 7%
To be clear, the numbers are guesses upon guesses, but the MoE calculation is correct.
So ... the full average is probably a much better indicator than it seems because it isn't perturbed as much by shifting identification and has a much tighter MoE.
Of course that says nothing about the vibecession effect.
Also missing is any wealth demographic data. I'd bet that if respondents were broken up into deciles of wealth, we'd see quite a different picture.
On Nextdoor I just saw a series of attacks on democrats for the price of gasoline, including one on Biden and one on our governor. The Republicans have mastered attacks, distortion and misinformation, as well as much more in politics. The dems seem lost and leaderless.
People can be swayed by politics for a while, but every year Jesus doesn't show up, but the home insurance bill gets bigger, more people will end up swayed by reality.
Excellent analysis, as always. I knew that “In the summer of 2024, the Michigan Survey transitioned from a phone to an internet survey to reduce costs.” But, I did not know that “The transition coincided with a drop in the share of Republican respondents and an increase in the share of Democratic respondents. If the shares had remained at their level in early 2024, the current overall sentiment index would be 7 points higher.” In the technical report on the Index of Consumer Sentiment (ICS) website, "Methodological Transition to Web Surveys”, https://www.sca.isr.umich.edu/files/methodtransitionannouncement2024.pdf
there is a discussion of web respondents being less optimistic than phone respondents. There is definitely a break in the ICS data starting in April 2024 as the survey transitions to 100% web-based by July 2024. Anyone doing modeling work with ICS data that includes pre-2024-4 and post-2024-3 reported ICS numbers should think about making some sort of adjustment.
My own back-of-the-spreadsheet estimate is that May 2026 ICS reported as 44.8 from the web-based survey would be equivalent to 50.2 if the survey had been phone-based as was the case prior to April 2024. So, the all-time ICS low is still June 2022 = 50 at the height of the post-covid inflation surge. Ernie Tedeschi’s adjustment places May 2026 ICS higher at 53.7. See https://www.briefingbook.info/p/the-effect-of-online-interviews-on
It’s odd that the web-based ICS survey ended up with a higher share of Democrats. The conventional wisdom is that Republicans are a male-dominated group less likely to take the time to answer surveys in general, but also more web-oriented than talking-on-phone-oriented. You would think that Republican “bros” would be more likely to answer a web survey. Great work digging out that strange result, Dr. Sahm. I do hope UMichigan Surveys of Consumers engages with your work.
I've seen many methodological observations that R populations also skew older, less internet-savvy, and less inclined to respond honestly to surveys these days, phone or otherwise. It's very hard to assess how this all shakes out.
I, knowing nothing about opinion polling: is there some method, probably more intensive and time-consuming, to differentiate the partisan impact on a person’s response and their underlying opinion?
Need to replace the Michigan Consumer Survey with quantitative questions -
1. What percentage of your budget have you spent on core items (food, shelter, transportation, utilities) over the last year.
2. Ditto - more or less for next year.
3-5 Specific projections for inflation, employment, GDP... or at least directional answers.
The Hassett-vs-Krugman fight over whose voters are "really" unhappy almost proves the author's deeper point. When the partisan gap is bigger than the income gap, the politics isn't the signal — it's the noise sitting on top of a signal both sides actually share.
Strip the party labels off and the thing underneath doesn't move. Since 1979 the economy's output per hour is up about 90 percent. Typical pay is up about 33 percent. Those two used to track together and then quietly came apart, and nothing was ever built to keep them tied. That's not a red or blue number. It's a design gap, and it shows up in a paycheck the same way regardless of how someone votes.
Which might be why sentiment "feels" off across the board even when the headline data looks fine. People are driving a system that pulls, and they feel the pull in their daily lives long before any survey catches it. The partisanship changes how they explain the feeling. It doesn't change the feeling.
Very thought-provoking, CS. I take your points, too. I try to focus on slope, directionality, and velocity of change over time in each separate political-affiliation cohort. Like you, I'm also concerned about how the shift to internet surveys affects the representativeness of the sample vis-a-vis the general population. This issue has arisen in other survey research.
the length of your rebuttal/argument is the crux of the problem and has remained so since you know who came down the escalator; it only takes short snippets of an argument - no matter how false- repeated over and over in different forms to stick in the mind of our short attention span public.
The media and non you know who supporters have yet to obtain the magic elixir to break the spell.
It doesn't look good Bob!
The 2 party system is for people too lazy to use an analytical critical thinking process. It's just a pandering contest for the party nomination, then finger pointing in the general election.
https://braddelong.substack.com/p/re-the-truly-abominable-kevin-hassett?r=dvwlc&utm_medium=ios
What a great deeper dive presentation on consumer sentiment, Dr Sahm!
If I may, a couple of thoughts...
In a Trump 'situation,' meaning the economy and all other ancillary components and metrics measuring our monetary and fiscal health and attitudes, it's not surprising that consumers are exhibiting a historical low in current and outlook sentiment.
IMO, the majority of Americans are honest, hardworking, people attempting to continually improve the quality of their familial situations. So, when they hear about the daily impeachable offenses perpetrated by Trump and the regime, essentially all of it self-enriching and none of it to improve America, with no accountability then it's easy to envision their feeling demoralized, frustrated and without meaningful purpose as their lives and freedoms are torn away
Frankly, if I were asked to provide sentiment on the future outlook of all things economy, I'd have to preface the answer as would an attorney, "it depends."The answer is bifurcated because it depends on whether Trump and the regime can be washed away by a blue tsunami or if we will be perpetually stuck with him. We already know he has no intention of leaving, so the illegal slush fund is his attempt to prepay participants for the next insurrection.
Thankfully, it appears that the fund will not be set up and will be crushed; the ballroom won't be built by taxpayers, neither will that ridiculous arch. Michael Wolff has stated that the midterms will determine whether Trump is finished OR if we are finished. Terrifying!
Anyway, for the intermediate term, perhaps the consumer sentiment should add two components that survey attitudes on two vectors:
1) An extension of the regime after the midterms if a blue tsunami fails.
2) The gestation of an actual presidency can occur as a result of successful midterms to oust the regime.
The sentiment gap between Democrats and Republicans is easy to explain - Republicans are in denial, no matter who is in the White House. After all, reality has a well known liberal bias.