A plain-English guide to ADP vs nonfarm payrolls.

Two payroll prints land each month, three days apart. ADP National Employment, on Wednesday morning, samples roughly 25 million private-sector employee records and reports private payroll growth. The Bureau of Labor Statistics Employment Situation, on Friday morning, samples 651,000 worksites and reports total nonfarm payroll growth, the unemployment rate, hours, and hourly earnings. They measure adjacent things and they often disagree. This is the explainer on what each one is, why they diverge, and the rule of thumb that lets a desk translate an ADP surprise into an NFP expectation.

What each report actually measures

• ADP National Employment Report. Co-produced by Automatic Data Processing and the Stanford Digital Economy Lab, from a panel of roughly 460,000 client firms representing about 25 million workers, or one in six US private-sector employees. The report covers private payrolls only and uses a same-store methodology, so newly onboarded ADP clients do not contribute to the headline. Released the Wednesday before the BLS release.
• BLS Nonfarm Payrolls. A survey of 651,000 worksites (the Current Employment Statistics or CES survey), benchmarked once a year against the Quarterly Census of Employment and Wages, which covers essentially the universe of US payrolls. The report covers nonfarm payrolls (private plus government, minus agriculture) and includes the household-survey-derived unemployment rate, average hourly earnings, and average weekly hours. Released the first Friday of the month.

Two structural differences fall out of those definitions. First, NFP includes government, ADP does not; the federal, state, and local payrolls average roughly 30,000 jobs per month in recent samples and can swing by ±50,000 around census hiring or fiscal cycles. Second, ADP's same-store methodology systematically misses new-business formation; in periods when small-business formation accelerates, ADP will print below NFP.

How they line up across a year

The translation rule

There is a working rule of thumb that desks use to translate an ADP surprise into an NFP expectation. It is approximate, but the correlation in the post-2020 sample is strong enough to be useful when nothing else is moving the market.

Expected NFP ≈ ADP + 30k (government) + 20k (same-store gap)

Today's ADP at 122k would imply an NFP expectation of around 170k headline, plus or minus the structural noise. The standard error on that translation is roughly 60k, which is also roughly the standard error on the NFP first release versus the eventual benchmarked total. That is the reason a fundamental fund will fade an immediate dollar reaction to an ADP-versus-NFP gap and wait for the revision.

Why they diverge in any given month

Three drivers of divergence, in rough order of contribution:

1. Survey timing. ADP samples its panel during the pay-period that includes the 12th of the month. So does the BLS. The samples are different and the same-week sampling catches different sets of firms. In months when pay periods span calendar-month boundaries, this matters.
2. Same-store vs population. ADP rebuilds its panel as clients churn; the same-store methodology means new ADP clients don't show up until they have a year of continuous data. In start-up-heavy periods, ADP under-counts.
3. Revisions. BLS publishes the previous two months of NFP as revised data every release. ADP publishes a single-revision series for the prior month. Watching the revisions matters more than the headline.

What it means for FX

The ADP-to-NFP path is informative for FX in two senses. First, a clean ADP surprise that the translation rule says implies a similar NFP move usually fades by Friday because the market has already priced the implied NFP move. Second, an ADP that diverges meaningfully from the translation rule (a beat of 50k when the rule says it should be a miss) generates an over-reaction on Wednesday that the Friday print typically corrects. The trade is to fade the Wednesday move, not to chase it.

The signal is cleaner when read through the economic-surprise index: a one-standard-deviation ADP surprise contributes roughly 6 points to the index; one-standard-deviation NFP surprise contributes roughly 12. The weights tell you NFP has twice the FX gravity of ADP.

What neither tells you

1. Wages. ADP publishes a separate Pay Insights report with wage data; NFP includes Average Hourly Earnings. Both are dirtier than the headline employment figure.
2. Hours worked. NFP includes average weekly hours, which moves the aggregate income picture more than the payroll count does. A weak headline with strong hours is the same income story as a strong headline with weak hours.
3. The composition. Both reports publish industry breakdowns. Reading only the headline is leaving half the signal on the table. A 150k print driven by healthcare and government is a different story than a 150k print driven by construction and manufacturing.
4. The household survey. The unemployment rate comes from a separate survey of 60,000 households, not the establishment survey. The two can disagree on direction in any given month. That is not an error; they are measuring different things.

Related reading

• Today's analysis applies this directly: Two more upside prints. The thesis is in trouble.
• The framework reading the labour data into a one-number summary: A plain-English guide to economic surprise indices
• The Beveridge-curve angle on the same data: A plain-English guide to the Beveridge curve
• How wages connect to the inflation picture: A plain-English guide to PCE vs CPI