How the benchmark works

Where the data comes from

The benchmark is built from salary submissions by professionals in our community, shared through our salary survey. It is a community sample, not a census of the Australian market, and we think you should know what that means: our sample currently leans towards Sydney and Melbourne, a handful of large corporate sectors make up more than half of it, and women are represented more strongly than in the national full-time workforce. Where an employer has verified data with us, that is shown separately and labelled as employer-provided.

All benchmarks are in Australian dollars. Submissions from people working overseas are excluded from Australian comparisons, and employer-uploaded rows in any other currency are rejected rather than converted.

How we build your comparison group

We compare you with people whose work actually looks like yours. Your comparison group is matched on what you do (your job or discipline), where you do it (your industry, and where relevant the tier of firm), and your seniority. Seniority is never relaxed: a Senior Associate is not benchmarked against partners to make a group bigger.

The quick calculator on the home page is a deliberately rougher tool. It compares you with everyone at your level across a broad sector group, without matching your job or your employer's tier, and it says so on the card. Your full benchmark after the survey uses the matching described here, which is why the two can differ.

When several valid groups exist, we serve the tightest one that meets our privacy floor, not the biggest or the most flattering one. The headline figure is the median (the middle value) of your group, shown with the spread around it.

When your group is small

Specific jobs in specific industries can be rare. When your exact group is too small to serve, we widen it one careful step at a time, for example from your industry to a group of similar industries, and we tell you on screen when a figure comes from a widened view. A wider group is only used when it is meaningfully bigger and still mostly made up of people in your discipline; otherwise we would rather show you an honest building state than a number built from the wrong crowd.

The privacy floor

We never publish a benchmark built from fewer than five people. That floor exists to protect the people who share their pay: below it, individual salaries could be guessed from the group figure. For the same reason we never display exactly how many people are in your group, and we never reveal how concentrated a group is in any single employer. The qualitative strength labels exist instead of those numbers, not as decoration on top of them.

Cleaning the data

Self-reported data contains mistakes, so every group is cleaned before we compute anything. Entries that are impossible (a base salary higher than the stated total) are dropped, and statistical outliers are trimmed. The trim is deliberately asymmetric: we are aggressive about junk at the bottom, like a $5 salary that was meant to be $85K, and conservative at the top, because a partner earning several times the group median is often real. Very small groups use a method designed for small samples rather than the standard one.

Keeping numbers current

Pay moves, so old data should not weigh as much as new data. Recent, higher-confidence submissions count more towards your benchmark, and older submissions are adjusted upwards for wage growth using the ABS Wage Price Index before they enter the calculation. When a group does not yet have enough recent responses, we still serve it, but we label it as a modelled figure and soften the claims we make from it, rather than presenting stale data as today's market.

What the confidence labels mean

Labels like Strong cohort, Moderate cohort and Early read tell you how much weight a figure can bear. They start from the size of your group, and for your main comparison we then stress-test the median itself: we recompute it many times over random resamples of the group (a statistical bootstrap) and measure how much it moves. If the plausible range around the median is wide, the label is downgraded. This check only ever makes a label more cautious, never more confident, so a tight-looking small group still reads as an early read.

The gender comparison

Where we show pay by gender, it compares medians for people doing similar work at a similar level in our community sample. That is a like-for-like comparison, and it is not comparable to the national WGEA gender pay gap, which measures average pay across all roles and employers. Both numbers are real; they measure different things.

What to keep in mind

A benchmark is an instrument reading, not a verdict. Small groups carry wide uncertainty even after every cleaning step above, and a community sample reflects who chose to contribute. We would rather under-claim than over-claim: when you see a softened label, a modelled tag or a building state, that is the system telling you the honest limit of what the data supports. If something looks wrong to you, tell us. We read every note.