What counts as a STEM job? The answer doubles depending on who is counting.

Two official numbers, both correct

In June 2025 the Bureau of Labor Statistics published an analysis by Audrey Watson in its Beyond the Numbers series, examining what happens to STEM employment and wage figures when you change the definition. It used a special tabulation of May 2023 data from the Occupational Employment and Wage Statistics programme, known as OEWS.

The finding is worth sitting with. The OEWS programme publishes a STEM dataset built on one definition, and by that definition there were nearly 10.2 million STEM jobs, about 6.7 percent of US employment. Apply the broadest option developed by the interagency Standard Occupational Classification Policy Committee, and the same month contains nearly 20.7 million jobs, about 13.6 percent of employment.

Neither figure is wrong and neither is a revision of the other. They are answers to different questions that happen to share a three-letter label.

Where the extra ten million people come from

The committee sorts STEM occupations into four subdomains by subject area, and the gap between the two headline numbers is almost entirely one of them.

Health is the largest STEM subdomain, marginally larger than the science, engineering, mathematics and IT group that most people picture when they hear the word. Exclude health and social science, as the narrower definition does, and STEM halves.

The two largest subdomains together account for over 97 percent of STEM employment under the broad definition. Social science and architecture are small by comparison, and architecture in particular is understated because several split occupations were assigned elsewhere.

The five kinds of STEM work

The same guidance also sorts occupations by the type of work performed, which cuts across the subject areas: research, development, design or practitioner occupations; technologist and technician occupations; postsecondary teaching; managerial; and sales.

Most STEM jobs, about 13.3 million, are in the research, development, design or practitioner group. Technologist and technician occupations account for about 4.9 million. Managerial, postsecondary teaching and sales occupations together make up roughly 2.4 million.

Combining four subdomains with five occupation types produces a grid, and a definition of STEM is really a choice about which cells of that grid to include. That is why there is no single number: there is a matrix, and different users need different slices of it.

What the largest STEM occupations actually are

The individual occupations are more surprising than the totals. In the health subdomain, the largest is registered nurses at nearly 3.2 million jobs, followed by licensed practical and licensed vocational nurses at 630,250, medical and health services managers at 515,100, and pharmacy technicians at 460,280.

In the science, engineering, mathematics and IT subdomain the largest are software developers at 1.7 million, computer user support specialists at 689,700, computer and information systems managers at 592,600, and computer systems analysts at 498,810. The largest occupations in that subdomain unrelated to computing are industrial engineers at 332,870 and civil engineers at 327,950.

Read that list again and notice how much of it is nursing and computer support. The public image of STEM is a laboratory. The employment data is mostly hospitals and IT departments.

The split-occupation problem

There is a wrinkle in the data worth understanding, because it is the kind of detail that quietly moves numbers people then quote for years.

Some occupations genuinely belong to two subdomains at once. Architectural and Engineering Managers is the clearest case: it contains both architectural managers and engineering managers, and the survey does not record which is which. The same problem applies to architectural and civil drafters, to other drafters, and to a residual category of life, physical and social science technicians.

Because the underlying survey cannot split a single occupation code between two cells, the analysis assigned each split occupation entirely to one subdomain, choosing the larger one, to avoid counting the same jobs twice. That keeps the total honest and makes the parts less so: it overstates some subdomains and understates others.

Architecture takes the worst of this. Its 139,880 jobs exclude architectural managers, one of the higher paying occupations that would otherwise sit in it, along with two lower paying drafter occupations. Its employment total and its average wage are both distorted, in opposite directions. The BLS analysis says so plainly, which is more than most secondary coverage of STEM data ever does.

A similar rule applies to occupation types. Four occupations that straddle the research and technician categories, including Computer Occupations, All Other, were assigned to the higher skilled research, development, design or practitioner group, following the classification principle that a worker who could be coded two ways is coded to the one requiring more skill.

Two domains above the four subdomains

The four subdomains also roll up into two broader domains, which is where several other published definitions of STEM come from.

The science, engineering, mathematics and information technology domain combines the social science subdomain with the life and physical science, engineering, mathematics and IT subdomain. The science and engineering related domain combines architecture and health. Those groupings exist because different agencies and programmes historically drew the line in different places, and the committee structure is designed to let each of them express its own definition using the same underlying codes.

This is the real reason STEM counts differ across sources. They are not sloppy. They are deliberately different selections from a shared grid, and unless a report tells you which cells it used, you cannot compare its number to anyone else's.

The wage picture is not uniform

STEM occupations do pay above average, and by a wide margin. The annual mean wage across all subdomains and occupation types was 108,330 dollars in May 2023, against 65,470 dollars for all occupations and 58,720 dollars for non-STEM occupations. Using the narrower definition, which drops health and social science, the STEM mean rises to 112,770 dollars.

That headline hides real spread. Of the 200 STEM occupations in the broadest definition, 165 had mean wages significantly above the all-occupations average, and 31 had below-average wages.

Technologist and technician work is the one STEM occupation type that pays below the national average, and 28 of the 31 below-average STEM occupations sit in it. The lowest paying include dietetic technicians at 38,440 dollars, emergency medical technicians at 43,100 dollars, and pharmacy technicians at 43,330 dollars.

That is not the whole story either. About 45 percent of individual technologist and technician occupations paid at or above the all-occupations average, and the highest paying include medical dosimetrists at 131,850 dollars, nuclear technicians at 97,520 dollars, and nuclear medicine technologists at 95,080 dollars. The category contains both ends of the range.

The extremes sit inside health. That subdomain holds some of the highest paying occupations in the entire economy, including pediatric surgeons at 449,320 dollars and cardiologists at 423,250 dollars, alongside several of the lowest paid STEM roles. An average across a group that wide describes very few of the people in it.

Where health pays more, and where it pays less

The science, engineering, mathematics and IT subdomain has a higher overall average wage than health, at 112,960 dollars against 104,320 dollars. That gap is not what it looks like. It comes mostly from the mix of occupation types in each group rather than from one field paying better across the board.

Of the four occupation types present in both subdomains, the science and IT group pays more in only two. Its managerial roles average 177,520 dollars against 134,440 dollars in health, a difference of more than 43,000 dollars, and it holds a larger share of its employment in that high paying type. Its technicians also earn more, 65,990 dollars against 56,950 dollars, and it has proportionally fewer of them.

Health pays more in the other two. Its postsecondary teaching roles average 122,760 dollars against 104,990 dollars, and its research, development, design or practitioner occupations average 123,700 dollars against 117,660 dollars. Both of those advantages are diluted, because postsecondary teaching is 3 percent or less of employment in either subdomain.

So the familiar claim that technology pays better than healthcare does not survive contact with the breakdown. What is actually true is narrower: technology has a larger share of its workforce in management and a smaller share in low paid technician roles. Composition, not superiority.

Why the definition affects you and not just the statistic

This looks like a methodology argument. It has practical consequences, because definitions get attached to things people want.

Training programmes, scholarships, economic development incentives, visa categories, employer diversity reporting, and university marketing all use some definition of STEM. When a programme says it serves STEM workers, it has picked a definition, and that pick decides whether a registered nurse, an economist, or an architectural drafter is eligible. The same person can be inside the boundary for one programme and outside it for the next.

It also shapes what people believe about their own trajectory. Headlines about STEM growing at nearly three times the rate of non-STEM employment between 2023 and 2033 are based on a particular definition. If your occupation is outside that definition, the headline is not about you, even if your work is unmistakably technical.

What this analysis does not establish

• It does not name a correct definition. The BLS analysis presents options, and it says explicitly that different definitions suit different purposes.
• It is not a forecast. These are May 2023 employment and wage estimates, not projections about any occupation's future.
• Mean wages describe groups, not individuals. They say nothing about what any specific person earns, or would earn after moving.
• The subdomain figures carry a known distortion. Occupations split across two subdomains were assigned to one to avoid double counting, which overstates some cells and understates others.
• Architecture and social science estimates are the most affected by that assignment, so treat those two subdomain totals with the most caution.
• Healthcare support occupations such as nursing assistants and home health aides are not in any of these STEM definitions.
• Being counted as STEM is not a statement about skill, difficulty, or the value of the work.

How to use this when describing your own work

1. Find out which definition a programme or employer is using before you conclude you do not qualify. Ask whether health and social science occupations are included.
2. Describe the work, not the label. What you built, measured, diagnosed, or maintained survives any definition change; the word STEM does not.
3. If your occupation sits in a split or borderline category, say what you actually do rather than relying on a job title to carry it.
4. When you read a STEM growth or wage statistic, check which subdomains it covers before applying it to your own plans.
5. Keep the specifics of technical work in your own record: the systems, the volumes, the outcomes. Those are what a reviewer can verify.
6. Treat the classification as a filter to get past, not as a verdict on whether your work is technical.

The number is a choice, not a fact

The BLS analysis is careful to present options rather than declare a winner, and that restraint is the useful part. A statistic that ranges from 10.2 million to 20.7 million depending on a boundary decision is not really one statistic. It is a family of them, and the label on the outside does not tell you which one you are holding.

So when the next report says STEM employment is growing, the first question is not how fast. It is which definition, and whether it counts the largest group of technical workers in the country: the ten million people working in health.