Related Content • External link to Columbia Business School • More news from Columbia Business School

The experimental design allowed the researchers to isolate the effect of gender differences on women leadership. The experiment consisted of two parts. The study first asked MBA students to complete a set of math problems; both men and women performed about the same. One year later, the researchers brought back the same students, asking them to recall their previous years' performance. The researchers found that when they compared actual with recalled performance, most participants overestimated their performance — a tendency documented in different forms in different studies. The major difference was that men consistently rated their past performance about 30 percent higher than it really was. Women, on the other hand, consistently rated their past performance only about 15 percent higher than it actually was.

Next, the researchers asked participants to estimate their performance on a task if chosen to represent a group, and were then divided into groups to complete the same math problem. The group was split into 33 groups of two or four members. Each group had to choose a representative and would compete with the other groups, with a generous cash prize awarded to the highest-scoring team. It was, then, in the best interest of the group to choose the person who had performed best on the problem sets in the past. This time the researchers also added an incentive: for some (but not all) groups, the representatives got an additional payment of either $20 or $75. In groups where leaders get no additional cash prize, individual and group incentives were aligned: that is, if a group knew a woman was better, its best interest was to pick her or sacrifice its competitive edge and the financial reward. In the groups whose leaders received a payment simply for being chosen to lead, an individual could then be chosen as a rep if they lied about their performance, and the group would lose while the leader would gain.

The results revealed that, on average, both men and women were willing to lie about their performance. When participants had an incentive to lie, they lied more, and the incidence of lying increased as the monetary award for being chosen as leader increased. While women kept pace with men on how frequently they lied, women did not exaggerate their performance to the same degree. As a result, women were selected 1/3 less often than their abilities would otherwise indicate. In other words, while there is no gender differential when it comes to lying, there is a significant gender differential when it comes to "honest" overconfidence: the main difference in women not being selected as leaders appears to be attributable to men's overconfidence in their abilities.

The study suggests an important takeaway for firms: recruiters should consider overconfidence when considering male candidates' claims about past performance. Employers who are not aware of the tendency for men to unconsciously inflate their performance could mistake that overconfidence for true performance, and overlook better female candidates. Furthermore, the researchers find this aspect of gender difference is hard to correct. Columbia Business School Professor Ernesto Reuben explains, "It's not just a matter of telling men not to lie — because they honestly believe their performance is 30 percent better than it really is. Similarly, it's not as if you can simply tell women they should inflate their own sense of overconfidence to be on par with that of men."