Job postings data and the BLS JOLTS job openings series compared

Job postings data and the BLS JOLTS job openings series both describe demand for workers, but one counts adverts and the other surveys employers about positions. This post sets them side by side from their publishers' own documentation, explains why their levels and turning points differ, and gives a method and a worked example for comparing any postings series with the survey on growth.

How do job postings data compare with the BLS JOLTS job openings series?

In the comparisons we read, the two agree more often on direction than on size. JOLTS estimates the positions that sampled US establishments hold open on the last business day of each month. A postings series counts adverts that a site can see, on its own calendar and classification. Compare their growth over matched months and industries, not their levels.

The table sets the survey, from the JOLTS chapter of the BLS Handbook of Methods, beside a public postings index, from Indeed Hiring Lab's data repository.

The BLS JOLTS job openings series: methodology and the job openings definition

JOLTS is a monthly BLS survey of US nonfarm establishments. A job opening is a specific position with work available, which could start within 30 days and for which the employer is actively recruiting from outside the establishment, counted on the last business day of the month. Openings are a stock; hires and separations are flows over the whole month.

The Handbook of Methods, last modified on 21 July 2026, settles cases that an advert cannot show. Active recruiting includes internet notices but also help-wanted signs, word of mouth, interviewing, employment agencies and job fairs. Excluded, among others, are positions open only to internal transfers, promotions or recalls, starts more than 30 days ahead, positions whose new hire has not yet reported for work, and work by temporary help agencies, contractors or consultants.

The bureau's answers for respondents add two rules. A job that is always open is reported if the employer is recruiting and the work could begin within 30 days. A position stops being an opening once an offer is accepted, and becomes a hire only when the person reports for work. An advert shows neither event.

The sample design stratifies establishments by ownership, region, industry and size, and most stay in the sample for 36 months. New respondents are interviewed by telephone for about five months, then are encouraged to report by web or email. In estimation, weights are adjusted for nonresponse and benchmarked to employment from the Current Employment Statistics program, then seasonally adjusted.

National estimates are released in the first week of a month or the last week of the month before. The August 2026 release came on Tuesday 29 September 2026, 29 days after Monday 31 August, and September's is scheduled for Tuesday 3 November 2026. Each preliminary estimate becomes final a month later, and at the end of each calendar year the previous five years are revised. The release carries seasonally adjusted and unadjusted tables, and the JOLTS FAQ, updated on 4 December 2024, states that the survey does not collect occupation information.

Online job postings data compared with JOLTS vacancies: the 2024 discrepancy question

In the official record we read, 2024 was a year of agreement in direction. The Federal Reserve Board's Monetary Policy Reports of 1 March and 5 July 2024 said JOLTS openings and a measure of vacancies built from Indeed postings both moved lower while staying above pre-pandemic levels. The doubts BLS staff later tested concerned the level of JOLTS.

The March report put JOLTS openings at nearly 9 million at the end of 2023 and said Board staff had constructed the postings measure; the July report followed that measure through the first half of 2024. Hiring Lab revised its index history at its change of seasonal adjustment in November 2024, so recheck any comparison from that year.

BLS staff answered the doubts about level in the Monthly Labor Review of August 2026, in an article by Larry Akinyooye, Mark Crankshaw and Sean McIllece. Had multi-unit businesses reported business-wide openings, or the same remote opening at several establishments, their openings rates should have risen more between the periods 2019 to 2020 and 2021 to 2022. They rose 2.0 percentage points, against 2.3 at single-unit businesses, and with their micro-level research the authors concluded that the pandemic surge reflected real economic change. They also noted that response rates have fallen substantially, which could add to openings if nonresponse is not random.

Earlier comparisons by BLS and by Federal Reserve Board economists show how adverts can part from the survey with no change in demand.

• A shift of channel. In The Economics Daily of 28 December 2004, BLS set the JOLTS openings rate beside the Conference Board's Help-Wanted Advertising Index of newspaper ads. From December 2000 to November 2001 the index fell 42 percent and the rate 30 percent. BLS named three possible reasons: differences of scope and definition, employers moving adverts from newspapers to internet sites, and larger errors in early JOLTS estimates.
• A change in the price of adverts. In a FEDS Note of 23 June 2016, Board economists Tomaz Cajner and David Ratner, giving their own views, found that JOLTS openings rose 48 percent from the end of 2012 to the end of 2015, against 4 percent for a composite index of the Conference Board's Help Wanted OnLine data rescaled to JOLTS. By their estimate, rising prices for job ads on Craigslist could, by themselves, explain roughly a quarter of the gap.

A postings series therefore carries demand plus changes in where employers advertise, what a site charges and reaches, and how its publisher counts.

Indeed job postings data versus the JOLTS job openings survey: the difference

Indeed's data count job postings visible on Indeed, gathered from across the internet and de-duplicated so that a job found in several places appears once. JOLTS asks sampled establishments how many positions met its three conditions on one day. The first measures advertising on one platform; the second estimates open positions across the US nonfarm economy.

Hiring Lab's data notes take the stock as the postings visible each day, often averaged over a week or a month, and count a new posting on the first day it is visible.

Indeed Hiring Lab job postings index versus JOLTS: the methodology difference

Hiring Lab's Job Postings Index is the percentage change in seasonally adjusted Indeed postings since 1 February 2020, as a seven-day trailing average, with a daily reading refreshed each week. JOLTS publishes monthly openings for one day, as levels and rates, adjusted with X-13ARIMA-SEATS. The index has no level to set beside an openings count, so only changes compare.

Hiring Lab adjusts each geography and sector separately with the Bundesbank's method for daily data, projecting the latest calendar year's factors from the three years before. Its sectors are Indeed's own occupational categories, built from normalised job titles, while JOLTS classifies the establishment by NAICS industry.

Indeed Hiring Lab job postings vs JOLTS job openings: how the online series compare

Indeed's index and JOLTS openings compare only after three conversions. Read the index at the last business day of each month, not as a monthly average. Turn both into growth over the same months. Match them by state or nation rather than by sector, because Indeed's sectors are occupations and JOLTS classifies industries. Then compare direction and timing.

How to compare a postings series with the JOLTS survey

These steps suit any postings series. The statistics come from the NIST/SEMATECH e-Handbook of Statistical Methods and NIST's Dataplot reference manual.

1. Match the population. Count postings by where the job is, not the company's headquarters, and keep the same employers in every month, so that an employer added to the source never reads as new demand.
2. Match the industries. Map the series' categories to NAICS sectors once, keep the map fixed, and compare sector with sector. An occupational category is not an industry.
3. Match the day. Read the postings stock on the last business day of the month, or the nearest day you hold, rather than a monthly average.
4. Match the seasonal treatment. Set the unadjusted JOLTS tables beside an unadjusted postings count, or compare 12-month changes. Two series adjusted by different methods carry two different seasonal models.
5. Compare growth, not levels. Take the change in the natural logarithm of each series, month on month or over 12 months. The handbook notes that many time series techniques assume stationarity, a mean, variance and autocorrelation structure constant over time, and lists differencing, and the logarithm for a variance that changes, among the ways to reach it.
6. Measure agreement three ways. The share of months in which both move the same way; the correlation coefficient of the growth rates, r = Sxy / √(Sxx × Syy), which runs from -1 to +1; and the dates of peaks and troughs. To test for a lead, shift the postings growth one month earlier and compute r again.
7. Look for breaks before reading a gap. A change of channel, of a site's prices, of a publisher's method or of a source's coverage moves a postings series with no change in demand. Mark each known break and test the stretches between them.
8. Record the vintage. Store each JOLTS value with its release date, preliminary or final, so that a test uses the figures available at the time and avoids look-ahead bias.

Worked example: an industry's postings against JOLTS openings

Acme Robotics is an invented maker of warehouse robots, and this example is illustrative: every figure in it is invented, including the figures standing in for the survey. Its strategy team wants to know whether postings in its customers' industry move with JOLTS openings for that industry.

Assumptions:

• Both series cover one industry, under a fixed map to one NAICS sector, for jobs located in the United States, over seven consecutive months.
• The postings figure is the number of open postings on the last business day of each month, at the same employers in every month.
• The survey figure stands in for unadjusted JOLTS openings, in thousands. It is invented, not a BLS estimate.
• Growth is the simple percentage change on the month, rounded to one decimal. For changes this small the log difference gives nearly the same values.

The working:

1. Levels. Open postings per thousand survey openings were 2,000 / 400 = 5.00 in month 1, 2,163 / 416 = 5.20 in month 3 and 1,816 / 384 = 4.73 in month 7. Over the seven months the survey fell 4.0% and postings 9.2%. Read as levels, the two seem to drift apart.
2. Direction. Signs agree in 5 of the 6 months. Month 6 disagrees: the survey rose 1.0% and postings fell 1.0%.
3. Correlation. The mean growth is -0.65 for the survey and -1.5 for postings. From the deviations, Sxy = 56.65, Sxx = 32.875 and Syy = 107.5, so r = 56.65 / √(32.875 × 107.5) = 56.65 / 59.45 = 0.95.
4. Size. The slope Sxy / Sxx = 56.65 / 32.875 = 1.72: postings moved about 1.7 points for each point of the survey.
5. Turning point. Both series peak in month 3, so neither leads at this turn.

Acme's team reads the postings series as tracking the survey's direction and swinging wider, with month 6 to examine for a break. Six growth rates are far too few for a conclusion; the same steps over several years of months make the test.

Postings read at the employer, and weekly market series from Fokals

We read every posting from the job board or careers page the company itself publishes, so each record is one role as its employer published it, dated when first seen, last seen and closed. In the hiring dataset, refreshed daily, Job Postings labels each role by job function, seniority and ten role flags, with location, work mode and advertised pay on one annual US-dollar scale; Hiring Activity gives each company's open, new and closed postings for every closed UTC day, with open postings by country, function, seniority and work mode. The reasons are in why we read job postings from the company's own board.

Market Series aggregates the company tables every week, as of each Sunday: 28 families of series across 21 industries, countries, company size bands and markets, over 7, 30 and 180 days, each row with a count, its cohort, a rate, the previous window, growth and an index.

These properties carry steps of the method directly.

• The day. Open postings at the end of every closed UTC day, by the country of each role's first location, under one stable company ID.
• Coverage. Market Series cohorts are same-store, so the 30-day growth compares the latest 30 days with the 30 before them over the same companies. A posting found on a board's first read is a baseline, never a new posting.
• Vintage. Daily and weekly rows are written once after the period closes and never revised, so the record is point-in-time by construction.

Market Series uses its own 21 industries and the headquarters country, so map its industries to NAICS sectors once. For a series by job location, the query below, in PostgreSQL, counts open postings located in the United States on each last business day, over companies present on all of those days; the days and the sector map are your own tables.

with month_ends as (
  select day from my_last_business_days
),
same_companies as (
  select h.company_id
  from company_hiring_daily h
  join month_ends m on m.day = h.day
  join my_naics_map n on n.company_id = h.company_id
  where n.naics_sector = '31-33'
  group by h.company_id
  having count(*) = (select count(*) from month_ends)
)
select h.day,
       sum(coalesce((h.by_country ->> 'US')::int, 0)) as us_open_postings
from company_hiring_daily h
join same_companies s on s.company_id = h.company_id
join month_ends m on m.day = h.day
group by h.day
order by h.day;

Each field is defined in the data dictionary, and sector hiring trends for macro and thematic research puts the weekly series to work.

Frequently asked questions

The queries and code on this page are examples to adapt. Test them in your own environment before you rely on them.

What this page says about the products it names was checked against their public documentation on 7 October 2026. Product and company names are trademarks of their owners. Fokals is not affiliated with them or endorsed by them.