Question
Consumer sentiment crashed while labor markets stayed tight. What gives?
The University of Michigan sentiment index dropped hard in 2023, but employment stayed strong, wage growth stayed positive, and real consumption kept climbing. Everyone noted the gap. The usual story was that people felt bad for reasons the data couldn't capture—vibes, news diet, social media dread, whatever.
But I keep running into a measurement problem that might be doing more work than the narrative allows. The Michigan survey asks about "economic conditions in the next year" and "next five years." That's a forward-looking question with a very short track record of accuracy. During much of 2023, inflation had just come off a 40-year high. Interest rates were climbing fastest in decades. Even if you personally had a job, the visible economic scenario looked genuinely unstable—not ambiguous enough to dismiss consumer anxiety as pure sentiment detachment.
Then there's the distribution question. National aggregates hide what's happening at different income levels. (Bivens and Zipperer have done good work on this.) If sentiment was getting hammered in the bottom two quintiles while top earners stayed confident, the aggregate disconnect looks worse than the actual lived experience gap. I haven't seen the Michigan folks publish clean crosstabs by income and sector for that period, which seems like an oversight.
The other thing: are people answering a question about "the economy" or their own financial situation? Those aren't the same, and the phrasing matters more than we usually admit. I'd want to see whether the gap disappears when you ask about personal finances specifically, not national outlook.
4 comments
Log in to comment.
You're circling around something real, but I'd separate two different problems that keep getting tangled up. One is measurement—yeah, the Michigan survey's forward-looking framing during a period of visible rate shock and recent inflation trauma probably does capture genuine economic anxiety, not just mood. That's worth taking seriously as actual information. The other problem is that "sentiment" itself is doing too much conceptual work.
When you ask someone "how do you think the economy will do," you're asking them to make a prediction while simultaneously reporting their emotional state, their media diet, their social position, and their risk tolerance. A construction worker in 2023 with a job but no health insurance and variable hours isn't the same anxious as a tenured professor with both, even if both answer "pessimistic." We collapse those into one number and then argue about whether the number is "real."
The income distribution point you mention—that's where I'd actually push harder. I've seen this in disability employment data too. You can have a labor market that looks tight because a handful of sectors are sucking up workers, while whole regions or skill categories stay slack. Top earners might see their own prospects as solid. Everyone else is watching the available jobs and seeing something different. The aggregate "labor market is tight" becomes almost useless information at that point. You need to know tight for whom, in what conditions, with what staying power.
I'm wondering whether you've encountered any actual crosstab work on this, or if the Michigan folks just haven't released it publicly. I ask because I spent about eight years watching state VR agencies report aggregate employment numbers that looked fine—placement rates stable, earnings up—while the actual distribution was quietly shifting. More people cycling through low-wage placements faster, fewer staying in jobs past eighteen months, the middle getting hollowed out. You only see it if you disaggregate.
The forward-looking question issue really does seem like the stronger point though. I'm curious whether the sentiment drop tracks more closely to specific policy uncertainty moments—like when the Fed started signaling sustained rate hikes—than it does to actual labor market deterioration. If people were genuinely modeling "what happens when my adjustable mortgage resets" or "what does a recession look like from my income level," then the gap between their anxiety and current aggregate employment numbers isn't a measurement problem at all. It's just them reading the odds better than the headline data suggests.
You're right that the phrasing matters, but the Michigan survey actually does ask personal finances separately—that's the other half of their index. And here's what's weird: personal finances stayed relatively stable through 2023 while the national economic outlook component tanked. So people weren't confused about which question they were answering. They genuinely thought *they* were doing okay but the broader system was fragile.
I'd push back slightly on the "genuinely unstable scenario" framing though. Yes, rates were climbing and inflation had been high, but by mid-2023 the fed funds rate was already at terminal rate, inflation was dropping month-to-month, and there was no recession. The visible data didn't actually show instability—it showed tightening. There's a real difference, and it matters for why people were anxious. They were watching a central bank that had aggressively tightened policy, and they knew from history that was supposed to break something. The anxiety wasn't irrational, but it was partly about *expected* instability that hadn't materialized. People were bracing for a hit that the labor market just wouldn't take.
The income distribution point you're making is solid though, and you're right there's limited crosstabs. What I've seen suggests sentiment crashed hardest among people who'd been hit by prior shocks—renters, people without college degrees, folks who'd seen stagnant wages for a decade. They had less buffer for the rate environment to matter, even if their job was currently solid.
You're onto something real, but I'd push back slightly on the measurement framing. The distribution point is solid—I've seen this in hiring and wage data too, where the aggregate numbers genuinely don't reflect what's happening at different rungs. But the forward-looking uncertainty angle needs more weight than you're giving it.
Here's what I saw in manufacturing during that exact window: companies froze capital spending in late 2022 and didn't restart it for months after inflation peaked. It wasn't because things were actually uncertain—inflation data was pretty clear by mid-2023—it was because the machinery that had worked for a decade (raise prices, demand stays sticky, margins hold) had visibly broken. Managers I knew personally weren't confused about economic conditions; they were spooked by the fact that the old playbook had stopped working. That's not a measurement artifact. That's a real signal that people took seriously even if employment stayed solid.
On the personal versus national framing: I'd actually expect the opposite of what you're suggesting. People are usually worse at predicting their own financial stability than they are at reading broad conditions. A plant manager with a steady paycheck might feel bulletproof about their job but genuinely believe the supply chain is going to seize up. When you ask about national conditions, you're often getting something closer to actual pattern recognition, not just personal comfort. The gap between the two might matter, but I'm skeptical it explains the whole disconnect. Sometimes people are just right about what's coming, even when the lag data looks fine.