Discrepancies between ADP and the BLS are vast. Let’s start with the BLS.
Nominal Percentage Changes Since 1979
- Nominal Total: 419.8 percent
- Nominal Male: 372.7 percent
- Nominal Female: 528.7 percent
BLS nominal weekly earnings come from the BLS Household Survey. People tell the BLS what they make in a pay period, and the BLS adjusts that to a weekly value.
The BLS Explains. Also see Weekly and Hourly Earnings from the Current Population Survey
The earnings data are collected from one-quarter of the Current Population Survey CPS monthly sample and are limited to wage and salary workers. All self-employed workers are excluded. Most series pertain to workers who usually work full time on their sole or primary job.
Usual weekly earnings represent earnings before taxes and other deductions and include any overtime pay, commissions, or tips usually received (at the main job in the case of multiple jobholders). Prior to 1994, respondents were asked how much they usually earned per week. Since January 1994, respondents have been asked to identify the easiest way for them to report earnings (hourly, weekly, biweekly, twice monthly, monthly, annually, or other) and how much they usually earn in the reported time period. Earnings reported on a basis other than weekly are converted to a weekly equivalent. The term “usual” is determined by each respondent’s own understanding of the term. If the respondent asks for a definition of “usual,” interviewers are instructed to define the term as more than half the weeks worked during the past 4 or 5 months.
The BLS Current Population Survey (CPS) samples approximately 60,000 eligible households monthly to determine labor force statistics. These households are replaced over time through a rotation system to ensure a broad and current sample.
For wages, the BLS only uses 1/4 of that sample, ~15,000.
BLS data is quarterly and as good as the responses are accurate and properly weighted.
Response rates during and after Covid have proven to be problematic.
BLS Real Weekly Earnings
Real means inflation adjusted.
Real Percentage Changes Since 1979
- Real Total: 12.2 percent
- Real Male: 1.7 percent
- Real Female: 35.9 percent
One can cherry pick starting dates to make any case they want. But the key point is nearly the entire increase in real wages is due to increases in real wages for females.
BLS Real Weekly Earnings Detail
Real Percentage Changes Since 2021
- Real Total: 0.8 percent
- Real Male: 1.0 percent
- Real Female: 0.0 percent
Those are actual BLS reported numbers, not Mish calculations. The BLS uses the CPI price index 1982-1984=100 as its deflator.
The Fed prefers to use Personal Consumption Expenditures (PCE) as its measure of inflation.
PCE and CPI Price Indexes
The BLS uses 1982-1984=100 for its CPI inflation calculation.
The BEA uses 2017=100 for its PCE inflation calculation.
To compare the two, I normalized to 2017.
PCE and CPI Price Indexes Detail
Inflation Calculation Details Since 2021
- CPI: +23.1 percent
- PCE: +20.0 percent
ADP data starts in November of 2020 but the first full quarter is 2021 Q1, and BLS wages are quarterly. So my charts start 2021.
ADP Data
ADP Data Collection Procedures are more robust than the BLS. For discussion, please see the section ADP Technical Notes.
Pay Insights is based on the payroll data of private employers.
We track the same cohort of workers over 12-month intervals to compute each individual’s year-over-year change in pay. This matching process results in more than 14.8 million pay change observations each month. From these observations, we calculate the median year-over-year pay change for job-stayers and job-changers.
Each month, we use the payroll data of more than 13.8 million U.S. job-stayers to estimate national median annual pay change.
Salaried workers make up about 39 percent of our job-stayers sample; hourly wage workers comprise about 61 percent.
Pay Insights tracks gross pay, which includes overtime, tips, bonuses, and commissions in addition to salary and hourly wages.
If a worker holds multiple jobs during a single month, we measure the job with the highest rate of pay for that month.
BLS vs ADP Wage Sample Size
- BLS: 15,000
- ADP: 14.8 million
The BLS has a much bigger sample size by far and far better procedures. But what exactly is the BLS measuring? The BLS only counts full-time jobs.
BLS Nominal Weekly Earnings X52 vs ADP Annual Wages
BLS vs ADP Nominal Earnings Since 2021
- ADP Male: +8.6 percent
- ADP Female: +10.5 percent
- BLS Male: +23.0 percent
- BLS Female: +21.8 percent
To compare BLS weekly estimates to ADP annual moving averages, I multiplied BLS weekly wages by 52.
The difference is stunning. If this is solely a part-time issue, why is ADP dramatically higher in 2021 Q1?
Something is seriously wrong somewhere, possible multiple places, but the big increases in BLS nominal wages vs ADP explains why I calculated ADP real earnings as negative.
Real Median Annual Wages Are Down Over 9 Percent Since May 2021
Yesterday, I commented Real Median Annual Wages Are Down Over 9 Percent Since May 2021
Want to know why voters are angry? I’ll show you in pictures.
This post explains in more detail how I arrived at negative real numbers for ADP.
ADP nominal wages are up in the single digits since 2021, but the CPI is up 23.1 percent. If you want better numbers, use PCE inflation. It’s “only” up 20.0 percent since 2021.
If you prefer to believe the BLS, hooray! Real earnings are up a grand total of 0.8 percent since 2021.
Presumptions
These numbers presume you believe PCE and CPI measures of inflation are accurate.
I don’t. Do you?
Neither the CPI nor the PCE includes property taxes, or homeowners’ insurance in their measures.
Both include content insurance if your house burns down or is ripped up by a hurricane, but neither includes the house itself.
And when eating out, tips are not included. Heck, that’s an expected 20 percent of the bill.
Is Homeowners Insurance Understated in the CPI?
I discussed homeowners insurance on August 11, 2025 in Is Homeowners Insurance Understated in the CPI? Shop Around!
Our Insurance went up by $2,000. Then another $2,000. Here’s our story.
Also consider Where Do You Spend Money on Food? How Screwed Up Are the BLS Weights?
Does the BLS match your budget?
Don’t worry, Trump says “There is virtually no inflation.”
Nonetheless, he has changed his message: Trump Adopts Chicago Cubs’ Perpetual Message, “Wait Till Next Year”
“One Big Beautiful Bill” did not resonate. Trump opts for “Wait Till 2026”.
Addendum
A reader accurately points out the BLS is a sample, while ADP is a population.
This is correct. And as I pointed out, the BLS is full-time workers, defined as 35 hours per week, not to be confused with the Obamacare definition of 30 hours.
Presumably the BLS is a valid sample. But we know that it isn’t. Evidence is massive revisions to jobs data. The BLS underestimated jobs by millions coming out of Covid and is overestimating them now.
Those revisions are with establishment data that is easier to model and sample. The household samples are a genuine mess due to Covid response rates and now deportations.
We do not have good data. And when BLS does make revisions, it’s in one big bang. Historical errors are left intact.
To answer the lead question: Very, but unquantifiable, and historically guaranteed inaccurate.
Addendum
A reader accurately points out the BLS is a sample, while ADP is a population.
This is accurate.
And as I pointed out, the BLS is full-time workers, defined as 35 hours per week, not to be confused with the Obamacare definition of 30 hours.
Presumably the BLS is a valid sample. But we know that it isn’t. Evidence is massive revisions to jobs data. The BLS underestimated jobs by millions coming out of Covid and is overestimating them now.
Those revisions are with establishment data that is easier to model and sample. The household samples are a genuine mess due to Covid response rates and now deportations.
We do not have good data. And when BLS does make revisions, it’s in one big bang. Historical errors are left intact.
To answer the lead question: Very, but unquantifiable, and historically guaranteed inaccurate.
How hard would it be to pay ADP to produce a big and valid sample?
still don’t understand all that hoopla about nominal wages and=or prices!
Mish, you have best info on real economy. it is called new cars sales, and new house units sales.
housing is most important purchase for human, and car sales are unique because of USA: size of country and cultural thing.
in Europe it is not. public transportation is way better, and number people per sq km is way bigger.
so new cars sales stuck in 30*40 years range , per capita! per 1000 adults
========
and new house units is even worse. new housing is 40*50 years low. per capita!!
because housing is expensive, and 99% bought by on loan from banks , NOBODY BUYS BIGGER house than one can afford, so overall each person/family buys thinking rationally!
so no matter salaries are up, or down, eventually consumer’s buying (aka purchasing power) is ultimate arbiter.
alx
Median number of months on sale for newly Completed homes is 2.7 months. It’s well below the historically average of 5 months.
“According to Mark Mitchell, a Rasmussen Reports survey that will soon be released will show that only 48 percent of U.S. adults under the age of 30 currently have a full-time job.”
https://x.com/honestpollster/status/1972715817587319122
Your BLS Real Weekly Earnings chart provides the evidence showing that inflation is being grossly undercounted by the BLS. If the index was baselined to “100” in 1982-1984, then if the chart were truly showing “real” earnings (aka inflation adjusted), it would have had a constant value of 100 from 1982 through today.
No, inflation-adjusted weekly earnings should not show a constant value, because of productivity changes. Especially in manufacturing and computer-related roles. Even grocery store checkout clerks move more merch now than in 1982.
That explanation has the look and feel of someone who drank the kool-aid. What does productivity have to do with the ratio of income to living expenses? Nothing! And what is increased living expenses relative to income? Inflation!
Excellent article. I think there’s a typo in the section on BLS vs ADP Wage Sample Size (pasted below, italics added by me). In the sentence in italics I think you meant ADP has larger sample size and better methodology? Or else BLS has “smaller” sample size?
ADP samples 9% of the population in a non-systematic way (whoever their customers are). BLS attempts to proportionately sample the whole population but has a much smaller sample size. I suspect neither the BLS nor ADP samples is accurate.
I also note that if employment growth is healthy, it’s a mystery why the ADP and PAYX stock shares are suffering…
BLS vs ADP Wage Sample Size
The BLS has a much bigger sample size by far and far better procedures. But what exactly is the BLS measuring?
ADP vs BLS bs. JP will have a total control of the narrative.
In that case, don’t you mean Jay Powell BS?
Gold vs SPX: in Feb 2025 Gold tested 2024 high @4,802 and moved higher in Apr. It stayed in a trading range for 3 months. In Aug Gold closed above Apr 2025 high, before rising to 4k.
Trump will benefit from Frosty and MPO realized gains. That’s why they hate Trump.
SPX, QQQ and IWM made all time highs. They are between 1% to 15% above 2024 high. Gold is 40% higher.
mags etf seems double top..
scary !
In Dec 2024 the mag7 etf high was 59. Last month: 66, up 10%. QQQ, SPX, DJI, IWM and mag7 market breadths are low. King dollar will deflate them and gold.
Read your charts: In 2025 SPX dropped 1,300 pts from Feb high. SPX is only 600 pts above Feb high: More selling than buying in 2025. The selling might cont in the next few months. If SPX drops 1,000 pts 1Y SPX will have a large selling tail, well below 2024 high. Gold 5,000 bs. Dec 2024 close was :5,881.
The numbers are goop…
ADP says 32,000 jobs were lost last month.
Revelio says 60.000 were added.
Carlyle says 15,000 are being added each month.
Pick your number to fit your opinion and you can run with it…
In a sample of one ~ I have highly qualified individuals dropping in to visit my farm looking for employment or opportunities. I simply skim the cream and move forward with providing limited capital for them to realize their incredible motivation.
Build a better society one talented individual at a time…
>>>
Goldman Hikes Gold Forecast To $4,900 As Ken Griffin Worries About “Substantial De-Dollarization”
=====
hahah!! and it is only for 2026!
Interesting article: I usually remain skeptical of Goldman calling for outsized gains in any commodity after their call that crude oil would never fall below $200/barrel while Bush II was filling the SPR at record highs ~ nearing the end of his and Cheneys war in Iraq. That was laughable.
With the US adding a trillion in debt every hundred and fifty daze, I have to admit having a measure of over~exuberance in the gold stocks at the moment. It is impossible for me to close the gap mathematically on how this administrations reliance on tariffs will solve the debt crisis.
Trade is moving away from the US in a substantive way. Follow activity on the shipping lanes between Brazil and Asia and inter-asia. Traffic has increased dramatically on those routes and slowed to the US.
My brokers are barking at me to stop selling the mag seven and Q’s as AI is going to save the day!
I think it is better to keep harvesting the crop of profits.
We’ve been stuck in quick sand for decades
All these statistical series (incl GDP) have had revisions to baselines, methodology, and definitions; are based on questionable data; have exhibited many flaws empirically; are paid far to much attention; have become political footballs.
They are not fit for purpose and only serve to eclipse real analysis and insight.
Gentleman, we have to do something to protect our phony baloney jobs…we must do something immediately…. Phony, and baloney are what are at the heart of all things BLS/CPI and whatever other measurement you want to use. The real inflation number, without all of the behind the scenes monkey business to flatten the number like utility adjustments and the like would give us easily an annual inflation number in normal years of between 7-10 percent. When it spikes, well you can easily double those numbers most likely. Anyone buying groceries over the past 25 years could tell you that. You literally would need to a take a new job every two years to get what used to be the necessary 20% bump in order to keep up with all of this. Purchasing power is only there for many if you use credit cards. Per FRED, all consumer and credit card debt in nominal dollars was $224 billion on this date in 2020. It has come down from the high, but today it stands at $1.047 trillion dollars. We are looking at on average over those 20 years a 19% annual increase in revolving debt- 475% for the entire period. You want to see where inflation starts, no way CPI/PCE are remotely close to the real inflation number. This is “income/debt” above what someone is getting paid, and often necessary to purchase goods and services for most these days.
What should happen is that every company that issues a W-2 should have an AI agent provided by the BLS that does a random of good sample size of wages & reports back to the BLS each month with actual wages without any identifiable information. This agent could also be tailored to work with financial companies where the income is harder to determine, and Congress should pass a law that requires companies paying employees in cash to report as well.
And you could program the AI agent to report back on the number of employees as well. That would mostly solve these problems. So, we just killed 2 birds with one stone, & we can lay off most of 2,000 BLS employees who do the surveys.
The bad news is this offers an idea of how bad AI is going to be for workers.
How hard would it be to pay ADP to produce a big and valid sample?
Is ADP W-2 jobs only? Are they able to capture Wall Street wages that are largely dividends, capital gains from the sale of stocks & bonds?
We need al all-inclusive methodology that mostly automates the data collection.
How do we attempt to accurately capture cash wages?
Capital gains are not wages.
Cash wages are a small issue.
But what about “non-wages”. The unemployed don’t have wages. More importantly, those not in the labor force, a big number, don’t have wages either.
Got it. The BLS, I guess, only looks at wages which means it’s not looking at all the ways that wealthy people generate income not through wages.
If that’s correct, then that in of itself is idiotic unless the BLS tracks these other sources of income through other means?
Everyone knows that the top 30-40% or so of income earners are the ones powering the economy forward, and this is where stocks, bonds & real estate holdings are concentrated.
This is a good example of why statistical analysis is not as easy as most people think.
The BLS uses a standard statistical methodology. It samples people from the entire working population. Yes, the sample is ‘small’ – 15,000 households- because giving and answering surveys is not easy to do. But 15,000 can yield a statistically significant sample of the entire workforce – especially if the 15,000 households can be linked to demographic averages of the entire workforce after their responses are recorded.
The ADP is a population, not a sample. The ADP calculates statistics from everyone in its particular dataset (which is fairly large). But if the people in that dataset aren’t like the rest of the US working population (ADP has data for <9% of the US workers), those statistics are biased if they are being used to represent the entire US population. I don’t get to choose to have my data available to ADP; my CFO does when she uses that company for payroll purposes. ADP tracks the same people over time, but those people probably won’t be employed or their firm providing info to ADP over the long run. So ADP numbers are statistical averages for a certain small subset of the population only.
So I am not surprised at all the BLS and ADP numbers yield different results. But this is why most economists trust the BLS data more (even if not perfect surveys) because the statistical methodology is more trusted.
Seems like BLS is politicized, especially during Biden’s term, when the divergence from ADP grew.
Mish,
Commentary seems quite muted over the past few days. Do you think people are becoming afraid to participate?
As for government published numbers? Without credibility, they really do not carry that much interest for me.
Anyhow I want to blabber on that gold (above the $4,000 mark) has passed the 11:00 time hurdle and that is a big deal for gold traders. We shall see how it goes through the close but with debt rising $1 trillion every 150 days, I really do not have that much confidence in the dollar as a great place to store my buying power.
Beware the December Futures options expiry date… That is a big delivery month and there could/should be some fireworks as the sellers of paper contracts scramble for inventory.
Lotsa noise out there but it is far better being invested in the quality mining stocks at this time IMO.
AEM is on sale this morning so I picked up another lump as I took off most of what remained of my AMD on the extremely bloated valuation (taxes be damned). The AI trade has gotten well ahead of itself IMO.
>>>
It’s about time women make more …
That said, gold has broken through the $4,000 glass ceiling with relative ease this morning and is now being tested at that level.
Surprisingly the big miners are getting sold off a bit so this portends a solid “Bitch Slapping” of gold as the day moves forward.
Enjoy!
Make more what? Cars? Buns? Coffee?
And make more for what ?
More in wages for their work, whatever they ally their talent and ambition to…
Real earning (2nd chart): Since Nixon real male wages were oppressed for 50 years. Real female wages were up from $250/week to $340/week. After 50 years women still lag behind. In 2020 a shingle woman with kids got $10K and other gov goodies. In Jasa single women sell their body to feed their hungry kids. A country which imports essential goods from hostile places, allow unruly minorities to destabilize them, finance impaired elderly for years, finance shingle women, pay lucrative pensions, medicaid, long vacations…bribing citizens, while piling up debt to prevent centrifugal forces from breaking it apart ==> will eventually collapse. Their leader have to cut “essential” programs sharply, instead of deflecting it with foreign sleaze. Most people will vote for him, but hate him thereafter.
Screw Sam Altman ass. If AMD shares drop OpenAI will not be able to buy AMD 10% shares for $0.01. NVDA will not pay Sam Altman $100 million to finance the competition. Thus, OpenAI will not be able to meet deployment keystones.
They keep telling me how oppressed women are. Yet gains in real wages have gone exclusively to women? Hmmm. At a local University, demographics are 65% female, 35% male. Just buy AI stonks, don’t worry about inflation. The future is unlimited! Lol.