One of the more remarkable and discouraging facts about the U.S. economy is that even while women have increasingly entered the workforce, they still make less than comparable men on a variety of metrics. Even though women now outnumber men in college, there is still a difference in earnings. Part of this gap is because men are more likely to study in high-paying fields such as math, science, and various technology fields. But why does this gap exist?

A new study argues that gender biases early in a child’s education may be a root cause.

Economists and sociologists already know a lot about gender bias in the workforce, of course. A variety of earlier economic studies showed how gender biases, even unconscious ones, can result in unequal work outcomes. Consider a paper published in 2000 by Harvard University economist Claudia Goldin and Princeton University economist Cecilia Rouse. They looked at how gender biases might affect auditions for seats in a symphony orchestra. Goldin and Rouse found that making the audition blind by having the musician play behind a screen resulted in the hiring of many more female musicians. The orchestra staff appeared to have an unconscious bias against female applicants.

The new paper, by economists Victor Lavy and Edith Sand of the University of Warwick and Tel Aviv University, respectively, looks at how this kind of bias from teachers might affect the future educational path of students. Specifically, they look at the implicit gender biases of primary school teachers. Lavy and Sand have access to a data set from Tel Aviv, Israel that tracks the progress of students from primary, or elementary school, all the way through high school. This way they can see how events from primary school affect a student’s educational trajectory.

The authors measure gender bias by looking at the difference between two tests that students take on math and language skills. One test is external and graded by an outside evaluator. This person has no information about the student’s identity, including his or her gender. The other test is internal and graded by the student’s teacher who knows quite a bit about the student. Lavy and Sand take the difference between the two test scores, which cover very similar areas, as a measure of gender bias.

Lavy and Sand find that these gender biases do exist and they have long-term effects. Male students that had more biased teachers do better on standardized tests later in their schools years. And the opposite happens for female students: they will do less well. And the effects aren’t just limited to test scores. Boys with biased primary school teachers are more likely to take math classes in high school and girls are less likely. Considering that these courses serve as a base for further course in math and science, this could explain future gaps. The authors show a strong correlation between test scores and future earnings.

And the effect is larger for certain kinds of students. In particular, girls that come from a family with a large difference in education levels are more affected by the early gender bias. In other words, if a girl’s father is more educated than her mother, she’ll be more affected by the gender bias of an early teacher.

The direct applicability of this study in terms of its exact findings to other countries or even other cities in Israel is debatable. But given the other research on the extent of implicit gender biases in the United States, the broader point of the paper could very likely hold up. And this would mean that actions early in a student’s life can affect not only their adult experiences but the overall economy if gender biases distort the allocation of potential scientists and engineers. In other words, the economy might miss out on a great scientist.

Research shows again and again that the quality of education and other factors early in a child’s life can affect their outcomes later in life. Lavy and Sand’s findings are another indication that early life circumstances need rapt attention.