Australian Industries Ran a 4.4-to-1 Gap in Unfilled Jobs. Their Pay Rises Landed 0.78 Points Apart.

For four years the Australian labour market ran a natural experiment, and it produced a result that the standard account of the period does not predict.

The premise of that account is simple enough to be almost invisible. When an industry cannot find workers, it pays more. Scarcity raises the price. That assumption sits underneath the whole shortage debate of the 2020s: employer groups reporting that jobs could not be filled, skilled migration expanded to fill them, training places subsidised in the occupations named as short, and a central bank watching for the wage pressure a tight labour market was expected to generate.

The Australian Bureau of Statistics publishes what is needed to test it, in two separate collections. The Labour Account reports, every quarter, the proportion of jobs in each industry that sit vacant. The Wage Price Index reports, every quarter, how fast the hourly price of labour in each industry is rising, holding the mix of jobs constant. Both key their industry on the same ANZSIC division letter, so they join exactly, with no matching or apportionment.

Put the two side by side across the shortage and the relationship is not weak. It is absent.

The horizontal axis spans a factor of 4.4. Mining could not fill 4.57% of its jobs. Education could not fill 1.03%. These are not marginal differences in labour-market tightness; they are different worlds.

The vertical axis spans 0.78 percentage points. The fastest-growing industry pay rate in the country belonged to electricity, gas, water and waste services, at 3.97% a year. At the bottom, three industries finish within two-hundredths of a point of each other: financial and insurance services at 3.186%, public administration at 3.195%, other services at 3.204%. Mining, with the most acute shortage in the economy, came in at 3.55%. Education, with the least, came in at 3.36%.

The rank correlation across the eighteen industries is 0.042, with a 95% bootstrap interval running from −0.42 to 0.50. That interval is wide because eighteen points cannot pin a correlation down precisely, and it comfortably contains zero. On the evidence here the vacancy rate tells you nothing useful about which industries raised pay fastest.

It is not the window#

A single window can flatter a null result, so the same cross-section was recomputed on three others: one starting after the last lockdown quarter, one covering the most recent four years, and one running to the latest available data. The correlation moves between 0.03 and 0.15. The vacancy gap stays between 4.3 and 4.4 to one. The wage spread stays between 0.78 and 1.36 percentage points.

Nor is it a quirk of the 2020s. The cross-sectional rank correlation was computed for every one of the 106 usable quarters back to 1998. Its mean is −0.067. It is positive in 42% of quarters and negative in the rest — which is to say it behaves roughly like noise around zero, and has done for a quarter of a century.

Ranked by the size of the pay rise, the industries do not sort themselves by scarcity.

The comparison that matters#

Australia has been here before, and the earlier case is the sharper test — because in the mining boom, the same statistical machinery recorded a very different answer.

Through the mining boom, industry pay rises stood 0.763 percentage points apart on average, against an economy-wide vacancy rate of 1.48%. Through the shortage era, the vacancy rate averaged 2.62% — a labour market three-quarters tighter — and industry pay rises stood 0.480 percentage points apart. The spread narrowed by 37% as the shortage roughly doubled.

The obvious objection is arithmetic: dispersion tends to shrink when the average shrinks, and wage growth in the 2020s was slower than in the boom. It does not survive the test. Take every pre-2020 quarter whose average pay rise was within a quarter of a percentage point of the shortage era’s 3.31% — there are 19 of them — and the spread across industries in those quarters was 0.617 percentage points. At the same average pay rise, the shortage era’s dispersion was 22% lower. Measured as a share of its own mean, dispersion fell from 18.8% before 2020 to 15.0%.

The same narrowing appears in a completely separate ABS collection. Average Weekly Earnings is an employer survey of what is actually paid, including overtime, bonuses and shifts in who holds the jobs — none of which the Wage Price Index is designed to capture. Its cross-industry spread fell from 2.56 percentage points before 2020 to 1.90 during the shortage. Two collections built on different samples and different definitions record the same compression.

This chart is also where a pre-registered prediction failed, and the failure is worth stating plainly rather than burying. The prediction was that dispersion would hit a record low during the shortage. It did not. The lowest reading in the series is 0.313 percentage points, in the March quarter of 2019 — two years before the shortage started. Dispersion during the shortage averaged 0.480 and sits at 0.398 in the latest quarter, low by historical standards but not a record.

What that failure changes is the shape of the story, not its substance. The compression did not arrive with the shortage. It was already there, and the tightest labour market in the series did not disturb it.

The workers did not move either#

If relative pay does not move, the textbook has a fallback: the adjustment happens in quantities instead. Workers go where the jobs are, without anyone having to outbid anyone. That was pre-registered as an open question, with either answer to be reported.

The answer is that it did not happen.

The rank correlation between an industry’s vacancy rate and its growth in filled jobs is 0.007, with a 95% interval from −0.55 to 0.49. Administrative and support services carried the third-highest vacancy rate in the country and finished the window with fewer filled jobs than it started with. Education carried the lowest vacancy rate in the country and added jobs at 5.54% a year.

One caveat belongs on this chart. The window opens in the September quarter of 2021, during the Delta lockdowns, which shut much of hospitality and the arts. Those two industries show the fastest job growth in the window partly because they are measured from a floor. Starting the window after the last lockdown lifts the correlation to 0.17 — still indistinguishable from nothing, and still leaving the conclusion where it was.

So neither price nor quantity responded to where the shortage was. Something else was setting the number.

A pay rise with a date on it#

If wages in an industry are bid up by competition for scarce workers, increases arrive whenever a firm decides it needs to move. They have no particular reason to cluster in one part of the calendar. If wages are set administratively, they do.

The Fair Work Commission conducts an annual wage review, and its determinations, in the Commission’s own description, “usually come into operation on 1 July of the following financial year”. An increase operative on 1 July lands in the September quarter. So for every industry and every financial year since 1998-99, the analysis measured what share of that year’s total wage growth arrived in the September quarter.

Five divisions were grouped in advance as award-reliant — retail; accommodation and food services; health care and social assistance; administrative and support services; and other services — and four as market-set: mining, financial and insurance services, information media and telecommunications, and professional services.

The two groups separate, and they separate at a specific point. Before 2010-11, the award-reliant industries took 35.6% of their annual wage growth in the September quarter and the market-set industries took 33.3% — close to the 25% a year spread evenly would produce, and close to each other. From 2010-11 the award-reliant share steps up to 53.7% and stays there. The market-set share drifts down to 29.1%. In 2024-25 the gap is 60.8% against 25.7%.

The timing of that break coincides with the arrival of the Fair Work Act’s annual wage review cycle, whose determinations take effect on 1 July, in place of the earlier review arrangements that operated on a different calendar. The data cannot prove the causal link on its own; what it establishes is that from that year onward, a majority of the annual pay rise in award-reliant industries has arrived in the single quarter following the Commission’s operative date, and that no comparable clustering developed in the industries that set pay through individual arrangements.

Two features of this chart should temper it. The 2019-20 and 2020-21 readings are distorted — the 2020 review was deferred and staged in three tranches running into February 2021, which pushes growth out of one September quarter and into the next. Both years are plotted, and both are excluded from the averages quoted above. And the grouping into award-reliant and market-set is a simplification of a spectrum: every division contains employees on awards, on enterprise agreements and on individual arrangements, in varying proportions.

What this is worth#

Those five award-reliant divisions held 6.9 million filled jobs at the March quarter of 2026 — 42.8% of the 16.18 million jobs the Labour Account counts. For a plurality of Australian jobs, the size of the annual pay rise is now substantially a decision with a date attached, and the vacancy rate in the industry is not visible in the outcome.

That has a direct bearing on how labour shortages are diagnosed. The instruments used to identify them — employer surveys, occupation shortage lists, the migration and training programs built on top of them — all rest on employers reporting that they cannot fill jobs. The market’s own test of the same claim is what those employers pay. Across eighteen industries and four years, that test returns nothing: the industries reporting the most acute scarcity did not pay more than the ones reporting the least, and did not attract workers away from them either.

What could explain this instead#

The Wage Price Index is built to be flat. The WPI holds job quality and quantity constant precisely so that it measures the price of labour and not the changing mix of who is employed. If shortage industries responded by hiring more senior staff, paying retention bonuses or lifting overtime, the WPI would not show it. This is the strongest counter-explanation, and it is the reason Average Weekly Earnings was pulled as a second measure: AWE does include bonuses, overtime and composition, and its cross-industry spread narrowed too, from 2.56 to 1.90 percentage points. That does not eliminate the objection — AWE is noisier and less frequent — but the compression is not confined to the measure designed to strip these things out.

Nineteen divisions is a blunt unit. An ANZSIC division averages over occupations that may have very different labour markets. A shortage of anaesthetists and a surplus of receptionists both sit inside health care and social assistance and can cancel out. The finding is that shortage does not predict pay at the level at which shortage is publicly reported and at which migration and training policy is set — the same level the shortage debate itself is conducted at. Whether the relationship reappears at the four-digit occupation level is a different question, and this analysis cannot answer it.

The vacancy rate may not measure scarcity. A high vacancy rate can mean an industry cannot find workers, or that it has high turnover and is perpetually re-advertising the same roles, or that posting a vacancy is cheap. Hospitality and mining may show similar vacancy rates for entirely different reasons. If the vacancy rate is a poor proxy for scarcity, then the absence of a correlation with pay carries little information — though it would then also undercut the vacancy statistics that the shortage case has been argued from.

Relative shortage narrowed even as absolute shortage grew. Measured as a share of its own mean, the spread of vacancy rates across industries was in fact lower during the shortage era (30.9%) than before the pandemic (39.8%), because the average vacancy rate itself rose so much. On that reading, shortages became more uniform and uniform pay rises were the right response. The counter is that reallocation responds to absolute gaps, not proportional ones: an industry leaving 4.57% of its jobs unfilled while another leaves 1.03% is a 3.5-point difference in scarcity, and the ratio between the highest and lowest industry stayed at 4.4 to one throughout.

Wage growth is a slow variable. Enterprise agreements typically run three years, so a shortage beginning in 2021 might not reach pay until agreements are renegotiated. The window here runs 3.75 years, and extending it to the latest data does not change the correlation (0.03). But a longer lag than that cannot be ruled out from this data.