Over-control bias: what people have in common can hide how their group is treated differently
Bring up the gender pay gap, and someone will often have an explanation ready: "Women just choose lower-paid jobs."
At first glance, the data seems to back this up. Many jobs are still done mostly by women or mostly by men. And some of the jobs done mostly by women, such as aged care, have long been low-paid.
But why do women end up in these jobs? And why are they paid less?
The answers depend on how you look at the data.
What is over-control bias?
Let's say you want to know whether women are paid less than men.
You could compare the pay of all women with all men. But that comparison mixes up gender with other differences between people. A woman who’s a surgeon and a man who’s a cleaner earn very different amounts, which tells you more about their education than their gender.
So, instead, analysts often compare the pay of women and men who do the same job. This is called "controlling for" the variable (in this case, the job). It’s a fair comparison if you want to know whether women are paid less for the same work. But it can’t tell you whether gender also affects which jobs people end up in, or what those jobs pay.
Take aged care as an example. If a man and a woman both work in aged care and earn the same, it looks like there’s no gender pay gap.
But what if the aged care field is paid less because most of its workers are women?
Then the man and the woman are both underpaid, and comparing people in the same job hides part of how gender affects pay.
This is called over-control bias.
It happens when the variable you control for (the job) is itself affected by the groups you’re comparing (women and men). Gender affects who ends up in a job, and what that job pays.
Why women and men end up in different jobs
From a young age, we pick up ideas about what women and men should do. Traditionally, women have been expected to care for others, and men to provide. These expectations are called gender roles. They come from patriarchy, a social system that gives men (as a group) more power and treats what’s seen as masculine as more valuable. It shows up in laws, workplaces, and everyday life, so it can work even when no one intends it to.
In a 2019 study of more than 7,000 New Zealand children aged 7 to 13, girls were ten times more likely than boys to want to be a teacher, and boys were nine times more likely to want to work in a trade.
Gender roles influence the home too. In New Zealand’s time-use surveys, women did most of the unpaid work. And when you’re the one picking up the kids or looking after a parent, you need a job that fits around it.
Expectations also seem to limit men. Since 1976, women have moved into many jobs men used to dominate, but men have barely moved the other way.
Most GPs are now women. Most registered nurses still are.
Proportion of women among GPs and registered nurses, 1976 and 2023
So when a woman or a man chooses a job, they’re choosing from options that patriarchy has already narrowed.
Why jobs done mostly by women pay less
"Women just choose lower-paid jobs" assumes pay comes first: some jobs pay less, and women end up in them.
But what if it works the other way round?
Research from the US and Europe has found that when more women move into a job, its pay tends to fall, even when the job still needs the same skills.
Why would a job’s pay fall as more women do it?
It comes down to what we as a society value. Work done by women, and work seen as feminine, tends to be valued less. Care is a clear example. At home, raising children and looking after family is usually treated as free, as if someone will always be there to do it. When the same work is paid, as aged care or childcare, it’s often seen as an extension of what women "naturally" do. So it’s often not recognised as skilled work, or paid like it.
What the comparison means for policy
Expectations affect which jobs women end up in, and values affect what those jobs pay. Comparing women and men in the same job hides these social forces.
Read without context, the pay gap seems to come down to the jobs women choose. That invites no action.
Read with context, the same data reflects choices made from narrowed options, in jobs that may be paid less because women do them. That invites questions about how work done mostly by women is valued.
Those questions went to court in 2013, when aged care worker Kristine Bartlett took her employer to the Employment Court. Her union argued her pay was low because most caregivers were women.
Her employer’s defence used a same-job comparison. The few men working as caregivers were paid the same as the women, so where was the discrimination? The union said those men were underpaid too, because the whole job was paid less for being done mostly by women.
The court sided with the union, ruling that caregivers’ pay could be compared with jobs in other fields. In 2017, the government agreed to a $2 billion settlement that raised pay for about 55,000 care and support workers.
But in 2025, the government changed the law. Claims like Kristine Bartlett’s must now compare work with men at the same employer first. Only if there are no men there who do comparable work can the claim look to similar employers, then the same industry or sector. The 33 claims already under way, which the Public Service Association says covered at least 150,000 workers, were stopped.
The law now starts with the kind of comparison Kristine Bartlett’s employer argued for in court: men at the same employer. It’s over-control bias, written into policy.
What to do about it
See the gap. Start with the overall difference between groups, such as the pay of women and men.
Find where the gap comes from. Compare people in the same job, and see how the gap changes.
Ask why. Why do women and men end up in different jobs? Why do those jobs pay differently?
Look outside the data for the answers. Pay data shows who’s in which job and what each job pays. It can’t show what directed people into those jobs, or how their pay was set. Research, history, policy, and the people your data is about can help answer those why questions.
Report the data and the context. Say what the data shows, what it can’t show, and what other evidence suggests.
Kristine Bartlett’s case came down to a comparison choice. Her employer compared her with the men in her job. Her union compared her job with others. The second comparison opened the way to a pay rise for 55,000 care and support workers.
You make choices like this when analysing data, often without noticing. Every time you choose who to compare, you’re choosing what your data can show.
I write about data and power, so if you'd like to follow along, the best place to do that is my newsletter. This article is part of the Things We Tell Ourselves series. Each article looks at an everyday saying, what the data shows, and what it means if you tell stories with data.
Kat Greenbrook is a data storytelling consultant, author, and workshop facilitator based in Wellington, New Zealand. She is the founder of Rogue Penguin and the author of The Data Storyteller's Handbook.
