The establishment survey counts jobs, not people, so a second job is counted twice and the self-employed are excluded. Its sample is far larger, which is why nfp is the headline number even though the unemployment rate comes from the other survey.
Two adjustments deserve attention. The birth-death model estimates jobs at firms too new or too recently closed to survey, and it systematically overstates at turning points. Annual benchmark revisions to state unemployment insurance records then correct the accumulated error, occasionally by hundreds of thousands of jobs.
Example: payrolls average 180,000 a month for a year. The annual benchmark revision cuts the level by 600,000, implying the true average was closer to 130,000 and the labour market was weaker than anyone traded on.
Related: household-survey, nfp, average-hourly-earnings, data-revision, jolts