The Truth Is At Level Seven: Five Lessons from Eric Pachman
The Data For The People founder on the importance of looking behind the numbers
When the jobs number crosses the wire each month, that headline is what the Bureau of Labor Statistics calls level zero, the top of a hierarchy that descends, industry by sub-industry, to the sewing machine fabricators down at level seven. Eric Pachman lives at level seven. He runs Data For The People, a soon-to-be nonprofit that turns largely unrecognized public data into interactive tools anybody can access. When he joined us recently, we opened with two level zero questions. Where is employment heading? Where is inflation heading? We spent the next hour drilling toward the bottom of the hierarchy.
What came back was a better way to read the numbers everyone relies on.
Lesson 1: The Headline Number Hides the Story
The monthly employment report is really two surveys stapled together. Non-farm payrolls come from a survey of businesses while the unemployment rate comes from a survey of households. The two statistics sometimes tell completely different stories, and response rates to both are, in Eric’s words, “falling off a cliff.” He walked us through a recent BLS press release. Overall unemployment unchanged, every subgroup unchanged, no story anywhere, and then, almost in passing, a note that the labor force participation rate fell four tenths of a point to 61.5 percent. Eric read it the way a storyteller reads a buried lede: “Your entire story is that there’s no story,” while the number that actually moved gets a single line and a shrug.
The Fed knows all of this, Eric said, and it’s stuck. “They can’t all of a sudden come out and say, ‘For the last 30 years, we’ve been focusing on the unemployment rate, and guess what? That no longer works. The world has changed.’ They’ve just kind of painted themselves into a corner to say the unemployment rate is the one ring that rules them all.” So Eric built a dashboard that opens up everything else in the household survey. Participation for men 45 and over sits at 54.7 percent, the lowest level on record. Women, he added, are the ones holding the labor force up, and he suggested the men among us go thank one.
I can tell you, investor person that cares about America, you will not see the recession coming with the unemployment rate.
Lesson 2: Not All Job Growth Pays the Bills
There have been stretches where healthcare contributed several hundred percent of all US payroll growth while everything else contracted. That much made the Wall Street Journal. What Eric wanted to know was where inside healthcare, so he drilled into the Quarterly Census of Employment and Wages, built from state unemployment insurance reports covering 98 percent of W-2 employees. It arrives six to nine months late, which is why investors ignore it, and it’s about the only BLS data that isn’t a model. The answer sat several levels down, in social assistance, then individual and family services, then services for the elderly and disabled and home healthcare aides, jobs that meet a real need as the boomers age. His wage ledger, built by Jonathan Pickens, a rising high school senior and one of Eric’s data fellows, makes the arithmetic vivid. Sorted by pay, pediatric surgeons top the list at a median of $559,000, all 1,190 of them. Sorted by employment, home health and personal care aides come out first, 4.3 million of them at a median of $17.21 an hour, about $35,800 a year.
Why are they not telling us that the quality of the job growth is coming all from home healthcare and personal care aides that make $35,000 a year?
Wages like those leave no cushion, which is where Eric’s single income stress test comes in. It maps the median wage in each city against the poverty line for a family of four on one income, and lets you type in a surprise expense like a healthcare bill. “If you put $15,000 in there, the whole country turns red,” and that’s pre-tax, at the median. Americans working two jobs are at an all-time high. The engine also runs on a single funding stream, Medicaid, and by Eric’s count roughly 32 states would have had negative job growth over the past year without the Medicaid care economy. A rising jobs count can describe a thriving labor force or a straining one, and only the quality numbers can tell you which.
Lesson 3: Oil Is a Fruit Salad
Matt pitched petroleum charts as useful even for people who don’t invest in oil, and Eric pushed the point further. “Everyone needs to become an oil investor to be an equity investor.” Before the data work, Eric was a chemical engineer modeling refineries at ExxonMobil, and he knows the territory cold. “I tried to forget it. I can’t.” The analogy he landed on for his 11-year-old is that oil is a fruit salad. Some of the fruit in a barrel of crude becomes propane, some becomes gasoline, some diesel or jet fuel or asphalt. American refineries spent billions on equipment to crack heavy Middle East crude. The light, sweet stuff we frack in the Permian is a different mix, which explains some strange headlines, like record US production alongside record exports and a Strategic Petroleum Reserve drawn down near its lowest level in 45 years.
Send the wrong oil to the wrong refinery and the yields come out broken. The market’s tell is the crack spread, and when the mismatch gets acute, simple refineries suited to light crude can earn four to five times their historical margin. The wartime closing and reopening of the Strait of Hormuz gave Eric a live demonstration.
When we closed the strait, we created COVID for oil markets.
Disrupted flows take months or years to untangle, with ships rerouting around the Cape. Diesel matters most, because trucks and railroads run on it and their fuel surcharges pass through to groceries and lawn care, the second-order effect a headline inflation print will never show you in advance. On a single day of the conflict, diesel inventories dropped three million barrels and prices jumped 13 percent. His charting advice outlasts any one crisis, though. Government inventory reports compare today to a five-year average, a window easily skewed by whichever unusual years fall inside it, so he charts inventories against 40-plus years of history instead. “Don’t take my word for it. Look at the charts.”
Lesson 4: There's a Model Behind Most Big Numbers
Eric recently watched a strange thing happen in the data: working age Americans started disappearing. As the BLS and the Census Bureau updated their models, millions of people were reassigned upward in age, many past 75. Nobody actually got older any faster. The model changed. Digging into census microdata, Eric found the country had lost about two and a half million prime age workers to the revision, people who were supposed to fund Social Security and Medicare. America, he wrote afterward, is far older than we thought. His summary borrowed the statisticians’ proverb and gave it a twist.
All models are wrong. Some are useful. Well, they just made their model a little more useful. It’s still wrong.
Inflation numbers work the same way. When Matt moved the conversation to CPI, Eric went straight for the heaviest item in it, owners’ equivalent rent, which makes up about 26 percent of the index. The government estimates this figure by calling homeowners and asking what they would, hypothetically, rent their house for right now. Eric has never gotten that call. He assumes people just look at Zillow. Turns out a quarter of America’s most-watched inflation number is a mere hypothetical. Why did CPI go up so much in 2022? Shelter. Why did it then look muted? Also shelter. His tools tell a richer story underneath, with 42 to 45 percent of 180 underlying items more than a standard deviation above their historical average, moving in unrelated places like gardening, beef, coffee, and hair care. Core inflation dates to the 1973 oil shock, when the Fed chair of the day decided food and energy were noise, and nobody has revisited it since, even though food is nowhere near as volatile as the healthcare costs core happily keeps.
Lesson 5: AI Expands Analysis, but Judgment Still Matters
Nearly every tool in this conversation has the same origin story. Eric is largely a one-man shop, and the household survey dashboard, with its thirty intersecting dimensions, was, in his words, “impossible to do without AI. Not with my skill set. I’m not smart enough to do it.” What used to take him two weeks now takes two hours, which is why he can publish an original data study every single day. Census microdata, the source of the aging discovery, is so impenetrable that his advice is not to go anywhere near it unless you’re using AI. It’s much easier to be a coding expert these days, he said, and the same logic extends to investors, who no longer have an excuse to ignore rich, unglamorous databases like the QCEW.
None of it runs unsupervised, though. Eric describes a tremendous amount of oversight, testing the code and checking the claims because the models sometimes push a conclusion further than the data allows and sometimes not far enough.
Sometimes it will make one little statement that you’re like, “You’re gonna torpedo the entire credibility of the entire piece based on that statement.”
His overall verdict held both halves at once, that where AI is good it is universe-changing good, and where it’s not, it can destroy all the work that it’s good at. So he keeps humans as the source of creativity and storytelling and lets the machine do 80 to 90 percent of the crank-turning. For investors adopting these same tools, that’s the enduring division of labor. The analysis scales. The judgment doesn’t nearly as well.
The Bottom Line: Our Own Sets of Data
Every answer Eric gave us came from the same motion, taking a number everyone accepts and drilling into how it was built, who collects it, what model sits underneath, and what got left out. The headline numbers will keep insisting there’s no story, and the story keeps sitting down at level seven, in databases anyone can open and almost nobody reads. That work is now his full-time mission, a nonprofit with no paywall, ever. He’s careful about the seam between fact and opinion, putting plenty of bias into his interpretations and none into how the data is worked, and the methodology is free for anyone who wants to rebuild the tools and argue. “I just have to diagnose all this, and you figure out what to do with it.” For all the red maps, the project is a hopeful one. His vision is a shared basis of reality, and “data can do that,” he said, “if we choose to use it for good.” We are allowed polar opposite viewpoints, “but we can’t have our own sets of data.”
Watch the full episode here: