CES vs. CPS: Payroll Jobs vs. Household Employment

The BLS Employment Situation release presents results from the Current Employment Statistics (CES) survey and the Current Population Survey (CPS). CES is also called the establishment or payroll survey; CPS is the household survey. The two headline figures appear together, but they come from separate samples and estimation systems. BLS describes them as complementary measures of labor-market activity, not two attempts to count exactly the same object. Its [comparison of the household and payroll surveys]({source:blsCpsCesComparison}) lays out their different concepts and methods.

That distinction matters when a headline says payroll employment increased by a certain number while the household survey's employed-person estimate moved by a different amount. The first figure is about jobs on covered nonfarm payrolls. The second is about people in the civilian, noninstitutional population age 16 and older who meet the CPS employment definition. Neither number is a correction or substitute for the other.

CES estimates jobs attached to covered employer payrolls. It reports nonfarm employment by industry and area and also produces hours and earnings measures. It does not estimate the unemployment rate. CPS asks about people's work and job-search activity and can classify them as employed, unemployed, or not in the labor force. That is why the national unemployment rate, labor-force participation rate, and employment-population ratio come from CPS concepts. The [CPS definitions]({source:blsCpsConcepts}) explain those measures.

“Jobs” and “employed people” are different units even before coverage is considered. A worker with two covered nonfarm wage-and-salary jobs creates two CES job records but remains one employed person in CPS. Conversely, one job can be held by different people at different times in the same month; the payroll estimate is tied to whether jobs are reported for the reference pay period, not to a unique headcount over the whole month.

CPS is a monthly sample of households. It asks about the calendar week that usually includes the 12th of the month. A person is classified according to work, temporary absence, availability, and job-search activity during the relevant reference periods. CES is a monthly sample of businesses and government agencies. Employers report the pay period that includes the 12th; pay periods can be weekly, biweekly, semimonthly, monthly, or another length. CES counts a covered job when the employer reports pay for at least part of that pay period.

The dates are designed to be near each other, but the reference windows are not identical. Someone can work during the CPS calendar week but not receive pay during an employer's pay period that includes the 12th, or the reverse. A job that begins or ends near the boundary may therefore appear in one measure's current period and another period's estimate. BLS says the effect of the different windows is not known and cannot be adjusted for directly, so timing is a possible source of divergence rather than a precise explanation for a particular month's gap.

The survey inputs also differ. CPS interviews a sample of households to estimate people and their characteristics. CES gathers payroll records from a sample of establishments and estimates jobs, hours, and earnings. Both are subject to sampling and nonsampling error, but the household estimate is more sensitive to month-to-month sampling variation. The BLS [Employment Situation methods guide]({source:blsEmploymentSituationQuickGuide}) explains how the release combines survey results and flags common interpretation issues.

Suppose 100 different people are all employed in jobs covered by CES during the relevant reference periods. Eighty-five each hold one covered nonfarm wage-and-salary job. The other 15 each hold two covered jobs.

CPS counts the 100 people once each. CES counts 85 + (15 × 2) = 115 jobs. The 15-job difference is not 15 additional workers; it reflects second jobs held by people who are already employed. The example is hypothetical and deliberately holds coverage, reporting, and reference timing constant so the unit difference is visible.

The example also shows why the surveys should not be forced into an identity such as “payroll jobs minus household employment equals multiple jobholders.” Real survey totals use different populations, industries, job classifications, pay-period timing, samples, and estimation procedures. Some multiple jobholders also have jobs outside CES coverage, such as a second job in agriculture or unincorporated self-employment. The simple arithmetic isolates one mechanism; it is not a formula for reconciling published national estimates.

CPS's population base is the civilian noninstitutional population age 16 and older. Under CPS rules, employment can include a person who worked at least one hour as a paid employee, worked at least one hour in their own business, profession, trade, or farm for pay or profit, worked at least 15 hours unpaid in a family business, or had a job but was temporarily absent. A person working in a covered job is generally counted once as employed even when they have another job. BLS details these categories in its [CPS concepts page]({source:blsCpsConcepts}) and [CPS Handbook of Methods]({source:blsCpsHandbook}).

CES instead covers nonfarm wage-and-salary jobs at reporting establishments. It excludes most agricultural work, unincorporated self-employment, unpaid family work, workers in private households, and people absent for the entire reference pay period without pay. It includes logging within its nonfarm scope. Incorporated businesses can report their wage-and-salary employees; the distinction is not simply whether someone owns any business interest. For CES's scope and classification details, see [Current Employment Statistics concepts]({source:blsCesConcepts}).

Consider a separate hypothetical case: an unincorporated consultant works at least one hour for profit during the CPS reference week but has no wage-and-salary payroll job at a covered establishment. CPS can count that person as employed, while the consultant's own work does not add a CES payroll job. Agricultural work outside the CES nonfarm scope can produce a similar difference. These are coverage differences, not missing workers in one data source.

Furlough timing can create another contrast. CPS can classify someone absent from a job for the full reference week as unemployed if they are on temporary layoff, expect recall, and are available for work; this exception removes the requirement to have searched during the prior four weeks, not the availability condition. CES can include a payroll job if the employee receives pay for any part of the employer's reference pay period. A paid leave arrangement or a pay-period boundary can therefore affect the two measures differently. The exact classification depends on the survey's definitions, not a general assumption that “furloughed” always means included or excluded.

CES begins with a sample of establishments and revises estimates as more employer reports arrive. Each year, BLS benchmarks March payroll employment to a more complete count based primarily on unemployment-insurance tax records, then revises surrounding months. The benchmark anchors the employment level; sample estimates are used to measure movements between benchmarks. CES also uses a statistical birth-death procedure because new firms and closures cannot all be observed immediately in the sample.

CPS does not benchmark employment to the same employer payroll universe. Its weights use population controls developed by the Census Bureau from census counts and estimates of births, deaths, and migration. BLS normally introduces annual controls with January data and does not usually revise the full historical CPS employment series to match them. A shift in the estimated population base can affect the level or apparent change in CPS employment even when no single survey respondent's status changed. The [BLS survey comparison]({source:blsCpsCesComparison}) explains benchmark revisions, population controls, sampling error, and business birth-death adjustments.

Seasonal adjustment is another setting to keep consistent. Each series has its own patterns and adjustment process. Compare seasonally adjusted data with the corresponding seasonally adjusted series, or compare not-seasonally-adjusted values on a consistent basis. Do not combine one series' adjusted monthly change with the other series' unadjusted level and treat the result as a meaningful gap.

In the long run, both measures tend to track broad employment trends, but they can diverge for periods of time. Differences in the number of multiple jobholders, industry mix, self-employment, population controls, business births and deaths, response rates, and reference-period timing can all matter. A larger one-month difference is a reason to inspect the definitions, revisions, and uncertainty—not proof that the economy created or lost an unexplained number of people.

Use CES when the question concerns nonfarm payroll jobs, industry employment, average hours, or average earnings among covered workers. CES can provide detailed employer-based trends and is generally more precise for month-to-month changes in covered payroll employment. Keep in mind that its job count can include multiple jobs held by one person and does not include all forms of work.

Use CPS when the question concerns how many people are working, how many are unemployed, labor-force participation, or employment by demographic group. CPS is the source for the national unemployment rate and for ratios that use the civilian noninstitutional population as their base. The related guide to the labor-force participation rate and employment-population ratio explains those denominators; U-3 versus U-6 compares narrower and broader measures of labor underutilization.

BLS publishes an adjusted household-survey research series that removes categories not covered by CES and adds nonfarm wage-and-salary multiple jobholders to make the concepts more comparable. The multiple-job adjustment does not identify whether those workers' secondary jobs are also nonfarm wage-and-salary jobs; a secondary job in agriculture or self-employment can therefore make the adjusted count higher than CES coverage would imply. The series can help with trend research, but it is imperfect and does not replace either official headline series. See the Beveridge curve guide for a different comparison: job openings from employer surveys alongside CPS unemployment.

When a monthly report's CES and CPS changes differ, first state the question and unit you care about. Then check industry and worker coverage, reference dates, seasonal-adjustment status, population controls, benchmark revisions, and sampling error. The two surveys are more informative together because one measures payroll jobs and the other measures people and labor-force status. Their differences become interpretable once the objects being counted are kept separate.