Attendees of the Google I/O developers conference. (Justin Sullivan/Getty Images)

Women, as a now-former Google engineer recently reminded the world, are underrepresented in several mathematical fields, including physics, computer science and engineering.

The causes of that underrepresentation are hotly debated, with barriers to entry and gender differences in test scores widely discussed as underlying reasons.

But new data analysis underscores the importance of another factor in the gap: confidence.

Young women, including women who are exceptionally talented in math, tend to underestimate their ability.

That lack of confidence likely pushes some of the field’s best minds to pursue other avenues. After all, even if someone has elite skills, if she doesn’t believe herself to be a talented mathematician, she’s unlikely to pursue a STEM education and subsequent career.

(As an aside, some women may also find their confidence shaken by highly publicized memos insisting they are prone to being inferior mathematicians, but that’s a topic for another day.)

Recently, I came across some data that illustrates this point in a particularly powerful and familiar context: the SAT.

When high school students take the SAT, they fill out a questionnaire that asks them demographic and other high school and college-related questions. An older version of this questionnaire asked students about their beliefs in regards to their intellectual ability. Specifically, students were asked if they believed they were in the highest 10 percent in mathematical ability.

What makes this setting particularly interesting, is that the data contains not just students’ beliefs about their own ability, but also their SAT scores, which provide a rough measure of actual ability. Using data for a sample of more than 4 million SAT takers in the late 1990s and early 2000s, the figure below shows the fraction of males and females who believe they are in the top 10 percent in mathematical ability for each 10-point bucket of actual SAT math scores.

Not surprisingly, students are more likely to report being in the top 10 percent if they score higher on the math SAT section (the curves in the figure are upward sloping). However, this figure shows that at all levels of SAT scores, men are more confident in their math ability than women. For example, the figure demonstrates that 67 percent of men who received a 700 on the SAT math section report a belief that they are in the top 10 percent in math ability while only 56 percent of women who also received a 700 on the SAT math section report the same belief. These results show that women are less confident in their own mathematical ability than men with the same SAT math score.

One might argue that this finding has less to do with gender differences in beliefs about math ability, and more to do with gender differences in confidence across all subjects. Fortunately, the data can speak to whether men are more confident than women in general or if this is something specific to mathematics.

The figure below illustrates data for the same approximately 4 million SAT takers on their response to the question of whether they believe they are in the top 10 percent in writing ability by their SAT verbal score. Unlike the math ability graph, here we see that women are slightly more confident in their writing skills than men who achieved the same SAT verbal score. Thus, it is not the case that men are more confident than women in all subjects. Rather, math ability seems to be a particular subject for which men and women show a confidence imbalance.

This data provides compelling evidence in a familiar domain that men and women differ in their confidence levels about mathematical ability — even when comparing individuals with the same test scores. These differences in beliefs are likely an important factor in creating gender imbalance in college majors and occupations. Efforts intended to improve the accuracy of beliefs about mathematical ability are likely to be an important strategy in combating gender imbalance in STEM fields.

Devin Pope is a professor of behavioral science and does research in behavioral economics at the Booth School of Business at the University of Chicago. Follow him on twitter @Devin_G_Pope