A Higher Occupational Wage Is Not Yet a Higher-Value Worker
Occupational wage growth is the subject here: what does it mean when the typical pay attached to an occupation rises? The immediate question is not whether the occupation is “good” or “bad,” but which part of the wage signal changed. This series builds a practical way to distinguish a higher price for the work itself from a different mix of workers, a different stage of experience, or a different place and industry in which the work is performed.
That distinction matters because wage statistics are often asked to carry more meaning than they contain. A higher number can influence a salary negotiation, a career decision, an employer’s hiring plan, or a public argument about technology and inequality. But a wage is not a direct score of human worth or a complete measure of productivity. It is an observation produced by a labor market, and labor markets attach prices to work through several channels at once.
The useful judgment is therefore conditional: when occupational wages rise, identify what was compared, who was observed, and which mechanism could have moved. Only then can we ask whether the occupation’s relative price increased, whether workers became more capable or differently selected, or whether the comparison simply shifted to a more highly paid setting.
A wage increase is not a verdict on workers
The simplest story is also the most seductive. Suppose the average wage of computer systems analysts, nurses, or electricians rises. It is tempting to say that the occupation became more valuable and that the people in it therefore became more valuable too. That sentence quietly joins together at least three different claims.
The first claim concerns the work: employers may be willing to pay more for a unit of output because demand has increased, technology has changed, or the work has become harder to replace. Economists often describe this as a change in an occupation-specific wage premium—the pay attached to the occupation after accounting, as far as the design allows, for who is doing the work and how experience is rewarded.
The second claim concerns the people: the average worker in the occupation may have changed. An expanding occupation can attract entrants from other fields. Those entrants may have lower occupation-specific experience than incumbents even if the occupation is becoming more important. A shrinking occupation may lose lower-paid or less productive workers first, leaving a stronger-looking group behind. The average wage can consequently move differently from the price of the work.
The third claim concerns the setting: the same occupational label covers different industries, metropolitan areas, employers, and career stages. A wage observed in a high-cost city or a high-paying industry is not automatically evidence that the occupation itself commands a higher premium everywhere.
The IZA study that anchors this series makes the problem explicit. It examines Swedish workers from 1996 through 2013 and estimates both changes in occupational wage premia and occupation-specific life-cycle wage profiles. Its central finding is not that every high-wage occupation contains more valuable people. Rather, differential premium growth increased between-occupation wage inequality, while changes in worker composition partly counteracted that increase. The study also finds a positive association between premium growth and employment growth, suggesting an important role for demand-side forces.
That conclusion is more informative than a ranking of occupations because it explains why the visible wage measure can mislead. It keeps the occupation—the subject of the analysis—in view while separating the occupation’s price from the characteristics of the people currently sorted into it.
The wage is a layered signal
A useful starting model is to treat an observed occupational wage as the result of several layers rather than one hidden essence. One layer is the occupation-specific premium. Another is the worker’s skills and the composition of the group currently employed there. A third is the wage return to experience, which can differ across occupations and change over time. A fourth is the setting: location, industry, working conditions, and institutional arrangements can all affect the observed number.
This model does not say that every dataset can estimate every layer. It says that an observed wage should not be interpreted as if those layers were already separated.
The Bureau of Labor Statistics’ Occupational Employment and Wage Statistics program illustrates the visible surface of the problem. Its occupational profiles provide hourly and annual wage estimates for roughly 830 occupations, along with occupational definitions, employment information, and wage data by state, area, and industry. A profile can show a mean wage and a distribution through the 10th, 25th, 50th, 75th, and 90th percentiles. That distribution answers a useful descriptive question: how are wages spread among workers identified with this occupation in the specified comparison frame?
It does not, by itself, answer why the spread exists or why it changes. A percentile tells us where a wage falls in a distribution, not whether the worker’s skills caused the wage, whether the occupation’s premium rose, or whether the observed group differs from an earlier group. A national median and a local mean may both be accurate while answering different questions.
The BLS page also makes the contextual problem concrete. Experience and education affect where an individual may fall in the national distribution. Location matters because wages can be higher in large metropolitan areas, where cost of living and other conditions differ. Industry matters as well. For computer systems analysts, the page shows different annual wage distributions across industries such as colleges and professional schools, insurance carriers, computer systems design, and aerospace manufacturing. The occupation has not changed names, but the labor-market setting has changed.
That is why “this occupation pays more” is incomplete unless the comparison specifies the wage measure, the period, the geography, the industry, and the population being described. The BLS data are valuable precisely because they expose these dimensions. They give a reader a disciplined way to compare wages, not a causal explanation of worker value.
From measurement to explanation
The harder question begins when we compare wages across time. An average wage in an occupation can rise because the occupation’s premium rose. It can rise because the people in the occupation became more experienced. It can rise because workers with lower wages left, or because a higher-paying industry or city now contributes more of the observed employment. These possibilities can coexist.
The IZA paper’s motivating example is a positive demand shock to computer programmers. If employers are suddenly willing to pay more for programming output, the occupation’s premium may increase. That higher premium can draw workers from other occupations, including people who are less productive at programming than the incumbents they join. The new entrants dilute the observed average even as the underlying price of programming rises. A raw mean may therefore understate the demand signal; in a different setting, selective departures may make a declining occupation look stronger than it is.
This is why the study focuses on occupation stayers when estimating year-to-year changes. Comparing workers who remain in the same occupation helps difference out time-invariant worker–occupation characteristics in the paper’s framework. It is not a magic filter: selection on idiosyncratic wage shocks can remain, and the authors discuss how that selection can bias premium-growth variation downward. Still, the design makes the question more precise. Instead of asking only whether the average of everyone currently in the occupation changed, it asks how the occupation’s wage relationship moved after accounting for important compositional differences.
Experience introduces another layer of ambiguity. Wage growth early in a career is not the same thing as a rising occupational premium. Different occupations can reward experience at different rates, and those rates can shift as technologies or work practices change. If a statistical model treats all wage growth among stayers as premium growth, it may mistake normal or occupation-specific skill accumulation for a change in the price of the occupation.
The paper addresses this problem with a life-cycle profile centered on a “flat spot”—a point at which the marginal effect of additional experience on wages is assumed to be zero. The study’s baseline places that point at 30 years of potential experience and tests alternatives. This is an identification choice, not a universal fact about when people stop learning. Its importance is precisely that a reasonable-looking change in centering can alter the estimated premium growth.
The broader lesson is methodological but practical: controls do not automatically clarify a wage comparison. A model must represent the way experience enters the wage, and readers must know what assumption makes the premium component identifiable. The paper’s example of physical and engineering science technicians shows the stakes. Without re-centering, the fitted premium growth for 2005–06 is around 8 percent; with flat spots at 25, 30, or 35 years, it is around 2 percent. The same underlying wage-change observations do not speak for themselves. Their interpretation depends on whether early-career growth has been separated from premium growth.
The first discipline for reading occupational wage growth is therefore simple: never ask only whether an occupation pays more. Ask what was measured, who was included, what experience profile was assumed, and which location or industry was compared. The next step is to examine those visible BLS measures more closely, because a mean, a median, and a percentile are not interchangeable—and none is a complete answer to who actually became more valuable.