A net 20 percent plan to hire. A net 5 percent just cut. Same survey, same month

Two numbers sit on facing pages of the same report.

On page 16 of the Small Business Economic Trends report for July 2026, a seasonally adjusted net 20 percent of small business owners say they plan to add staff over the next three months. NFIB calls that the highest hiring plans reading since October 2022, and it is 9 points above the historical average. Turn back one page. On page 15, the same owners, answering the same questionnaire in the same month, report a seasonally adjusted net negative 5 percent change in the number of people actually working for them over the last three months. That is the weakest actual reading of 2026.

Neither number reached the wire. The one that did was 99.8.

What is the NFIB Small Business Optimism Index?

Answer: It is a monthly composite of ten questions put to members of the National Federation of Independent Business, seasonally adjusted and indexed to 1986 equals 100. The NFIB Research Center has run the survey quarterly since 1973 and monthly since 1986, and it draws the sample from NFIB membership files.

The July 2026 reading was 99.8, up 2.4 points from June, which NFIB describes as the highest level since August 2025 and above the 52-year average of 98.0 that the report uses as its benchmark. Eight of the ten components increased and two decreased. Hiring plans contributed most.

Two details about that benchmark are worth holding onto. The monthly series charted in the report begins in January 1986, and the report itself says quarterly collection started in 1973, so the 52-year average spans a longer stretch than the monthly line the reader is looking at. And the front page of the report lists the ten seasonally adjusted component changes, which sum to 29 points, while the index itself moved 2.4. The headline is a rescaled composite, not a total.

The ten components are printed on that front page with their July levels: plans to increase employment at a net 20 percent, plans to make capital outlays at 25 percent, plans to increase inventories at a net 1 percent, expect economy to improve at a net 15 percent, expect real sales higher at a net 7 percent, current inventory too low at a net negative 2 percent, current job openings at 36 percent, expected credit conditions at a net negative 4 percent, good time to expand at 12 percent, and earnings trends at a net negative 16 percent.

What does the index actually ask about?

Answer: Seven of the ten questions ask what owners expect, plan or think about a period that has not happened yet. Two ask about the present. One asks about the last calendar quarter. The survey questions are printed in full on pages 26 and 27 of the report, so this is checkable rather than a matter of interpretation.

Go through them against that list. Do you think the next three months will be a good time to expand. Do you think general business conditions six months from now will be better. What do you expect to happen to sales volume during the next three months. In the next three months do you expect to increase or decrease the number of people working for you. Do you expect to find financing easier or harder during the next three months. Looking ahead three to six months, do you expect to increase inventories. Looking ahead, do you expect to make capital expenditures in the next three to six months. That is seven.

The remaining three are grounded. Do you have any job openings you are not able to fill right now. At the present time, do you feel your inventories are too large, about right, or too low. Were your net earnings during the last calendar quarter higher, lower or about the same.

Now apply that split to July. Of the 29 points of total component change the report prints, the seven forward-looking questions supplied 23, and the three questions about the present or the recent past supplied 6. The move was made almost entirely of intentions.

NFIB's own labelling on page 8 sorts the components into hard and soft, and puts job creation plans, inventory plans and capital expenditure plans in the hard bucket alongside job openings and earnings. That is a defensible convention, because a plan to spend is a firmer object than an opinion about the economy. It is still worth knowing that three of the five hard components are statements about the future.

Why did the index rise in a month when payrolls fell?

Answer: They measure different things. The optimism index counts the direction of owners' answers to ten survey questions, most of them about the coming quarter. Payroll employment counts jobs on employer payrolls. In July 2026 the first went up and the second went down.

The Bureau of Labor Statistics reported total nonfarm payroll employment down 23,000 for July 2026, in release USDL-26-1291 on 7 August 2026, four days before the NFIB report was published. The unemployment rate was 4.1 percent. The release notes that the July change followed an average monthly gain of 34,000 over the prior 12 months, with declines in local government education and retail trade and a continued rise in health care.

None of that makes either number wrong. A survey of owners' intentions and a count of payroll jobs can move in opposite directions in the same month without either being in error, and the NFIB report does not claim otherwise. What the pairing rules out is the reading that gets attached to the index most often, which treats a rise in the composite as evidence that small firms are adding staff. In July 2026 the survey that produced the rise also produced the weakest reading of the year for staff actually added.

How wide is the distance between what owners plan and what they report doing?

Answer: In July 2026 it was 25 percentage points, the widest since September 2022. That is Money & World's own computation from the two employment tables in the NFIB report, subtracting the net percent reporting an actual employment change over the last three months from the net percent planning a change over the next three.

Set the two series side by side for every month from January 2021 to July 2026, which is 67 months. The hiring plans reading is positive in all 67. The actual employment reading is negative in 53 of them, positive in 11 and zero in 3. The mean spread across the period is 20.0 points.

So the distance is not news. It is the normal state of the two questions.

What changed in July was the size. The spread had not touched 25 points since September 2022, when it reached 27, and it had not exceeded 22 in any month of 2023, 2024 or 2025. Through the first half of 2026 it ran between 9 and 15 points. Then it jumped to 25.

Before that becomes a finding, the reasons it might mean less than it looks deserve a hearing, and there are three. The two questions cover different windows, one looking forward a quarter and one looking back a quarter, so the difference between them is not a forecast error and cannot be scored as one. The actual employment series is chronically negative in this panel, averaging negative 2.1 across the period even through stretches when total payrolls grew, which points to something structural in how NFIB members answer a question about their own headcount rather than to a four-year collapse in small business employment. And one month is one month.

The scale of that second point is worth stating in figures. Total nonfarm payroll employment stood at 142,863,000 in January 2021 and 158,858,000 in July 2026, a gain of roughly 16 million jobs, across the same 67 months in which NFIB members reported net employment declines in 53. Whatever the actual employment series is measuring, it is not the direction of American employment.

What does a net percent actually mean?

Answer: A net percent is the share of respondents giving the positive answer minus the share giving the negative answer. A net 20 percent planning to hire means the share planning to add staff exceeded the share planning to cut by 20 percentage points. It says nothing about the number of jobs or the size of the firms.

This is the most durable misreading of the index, and it survives because the number wears a percent sign. Consider what a net 20 percent is compatible with. It could be 25 percent of owners planning to add against 5 percent planning to cut, with 70 percent standing still. It could be 45 percent against 25 percent. It could be a firm adding one part-time worker and a firm cutting forty, counted identically, because a diffusion measure weights every respondent the same regardless of payroll.

Compare that with what the Job Openings and Labor Turnover Survey publishes. For June 2026, in release USDL-26-1289 on 4 August 2026, BLS put job openings at 7.4 million and the openings rate at 4.4 percent, hires at 5.3 million and 3.4 percent, and quits at 3.2 million and 2.0 percent. Those are levels and ratios drawn from establishment payroll records. They can be added up, and they carry a denominator.

Do the hard labour data agree with the survey?

Answer: On direction, roughly. On level, they are not comparable statistics. NFIB reports that 36 percent of owners had at least one job opening they could not fill. JOLTS reports that establishments with 1 to 9 employees had 1,421,000 openings in June 2026, a rate of 5.7 percent. Those are not versions of each other.

The NFIB figure is a share of firms with any unfilled opening. The JOLTS figure is openings as a proportion of employment plus openings. A shop with one vacancy out of four staff and a shop with one vacancy out of forty count the same in the first and very differently in the second. Both are published monthly, both concern small employers, and putting them in the same sentence as though one confirms the other is a category error.

Where the two do speak to the same thing, the smallest establishments look reasonably healthy in the federal data. JOLTS table 7 shows establishments with 1 to 9 employees making 724,000 hires in June 2026 against 713,000 total separations, a net gain of about 11,000, with an openings rate of 5.7 percent that runs above the 4.6 percent for total private. Their quits rate, at 1.5 percent, is the lowest of any size class other than the very largest employers.

That table also carries a warning about precision. The openings rate for establishments with 1 to 9 employees ran 4.1 percent in March 2026, 6.3 percent in April, 6.1 percent in May and 5.7 percent in June. A series that moves 2.2 points in a month and gives most of it back over the next three is not delivering the small-firm labour market in high definition either.

What about sales and prices?

Answer: The NFIB sales series has been negative almost continuously since mid-2022, which does not mean small business revenue has fallen for four years. It means more members report lower dollar sales than higher, quarter on quarter, in a survey that measures direction and not amount.

A seasonally adjusted net negative 4 percent of owners reported higher nominal sales in the past three months, unchanged from June. Counting from the report's own table, that series has been negative in 49 of the 50 months from June 2022 to July 2026. The single exception is February 2026, at positive 1. Over that same stretch total nonfarm payroll employment rose by 6.5 million.

The Census Bureau, meanwhile, put advance retail and food services sales for July 2026 at 763.6 billion dollars in release CB26-131 on 14 August 2026, down 0.6 percent on the month and up 5.0 percent on the year. The NFIB question is quarter on quarter and the Census figure is a dollar total, so this is not a contradiction to be resolved. It is a reminder that a diffusion reading of negative 4 is a statement about how many owners moved which way, and about nothing else.

Prices are where the survey is most informative, because there the direction is the point. A net 31 percent of owners reported raising average selling prices in July, down 7 points from June after four consecutive monthly increases, and a net 28 percent plan to raise prices in the next three months, down 4. The report gives the historical average for actual price increases as a net 14 percent, so the July reading sits at more than double it. The 12-month change in the consumer price index in the same month was 3.4 percent.

How reliable is the survey?

Answer: The July 2026 report rests on 887 usable responses from a sample of 10,000 owner-members, a response rate of 8.9 percent. That is the largest monthly count of 2026 and the smallest July count in the six years the report tabulates. The report publishes no standard errors and no confidence intervals.

The response counts are printed on page 25 of the report, and the trend in them is steep. July responses ran 1,440 in 2021, 1,351 in 2022, 1,313 in 2023, 1,309 in 2024, 953 in 2025 and 887 in 2026. That is a decline of about 38 percent in five years.

July is also one of the four heavy months. NFIB fields a larger survey in January, April, July and October, and the intervening months are much thinner: the June 2026 reading of 97.4, the one that July's 2.4 point rise is measured against, came from 405 responses. February 2026 rested on 428, March on 432. A 2.4 point move between a 405-response month and an 887-response month is being asked to carry a great deal.

Set that against how a federal statistical agency handles the same problem. The Census Bureau's advance retail release draws a stratified random probability sample of roughly 4,800 firms, weights and benchmarks it to a universe of over three million retail and food services firms, and publishes standard errors, coefficients of variation and a 90 percent confidence interval on every headline change. Where that interval contains zero, the release marks the figure and states in plain language that there is insufficient evidence the change differs from zero. The June to July 2026 change of positive 0.2 percent carries exactly that mark.

The NFIB report carries no equivalent. Searching the full text of all 29 pages returns no instance of margin of error, standard error, confidence interval, sampling error or weighting. The only statistical statement in the document is the response-rate sentence.

Who is being surveyed?

Answer: NFIB members. The report states on its inside cover that the sample is drawn from the membership files of the National Federation of Independent Business, which is a trade association rather than a statistical agency, and which lobbies. That is disclosed and lawful, and it is a fact about the survey frame rather than about the survey's conduct.

NFIB filed two lobbying reports with the Senate Office of Public Records for 2026, disclosing expenses of 1,280,000 dollars for the first quarter and 1,670,000 dollars for the second, which is 2,950,000 dollars across the first half of the year. The organisation is both registrant and client on those filings.

What follows from this is narrow, and being precise about it matters, because the loose version of the point is unfair. There is no evidence in the document that the survey is conducted improperly, and none is claimed here. What the membership frame does mean is that respondents are self-selected twice over, first by choosing to join a business association and then by choosing to return a questionnaire, and that a reader treating the result as a representative sample of American small business is treating it as something the report itself never says it is.

The commentary section is signed work and reads as such. Chief Economist William C. Dunkelberg used July's edition to attribute capital spending strength to investment in chips and the buildings to house them, to note that equity market gains are supporting spending by wealthier consumers, and to argue that a resolution of the war with Iran would reduce oil prices. Those are his arguments, presented as commentary, and they sit in a section clearly separated from the tables.

What to look at in the next release

The August 2026 edition is survey 537. Three things in it will be worth more than the headline.

The pair, first. Hiring plans on page 16 and actual employment change on page 15, read together. If plans hold near 20 while the actual series stays around negative 5, the spread is a level shift rather than a turn. If the actual series climbs toward zero, something has moved.

The response count on page 25, second. August is an off-quarter month, so a count in the 400s should be expected, and the size of any month-on-month change in the index should be read against it.

The component table on the front page, third. The ten changes and their sum tell you whether a move came from questions about the next quarter or from questions about the last one. In July 2026, 23 of 29 points came from the former. That is a fact about the survey's construction, and it is available before anyone has to form a view about what small business owners are feeling.