Is it worthwhile to invest in diversity training? Or is it not pulling its weight?

A new report in the ‌Proceedings of the National Academy of Sciences has an answer: a little from column A, a lot more from column B.

The report from the University of Pennsylvania’s Wharton School summarizes research involving more than 3,000 employees at a global professional services firm. The study examined whether a brief online diversity training program changed attitudes and behaviors toward women in the workplace.

The top-line finding? Participants’ attitudes before the training were a significant factor in the training’s effectiveness.

“Among groups whose average untreated attitudes—whereas still supportive of women—were relatively less supportive of women than other groups, our diversity training successfully produced attitude change but not behavior change,” the report stated. “On the other hand, our diversity training successfully generated some behavior change among groups whose average untreated attitudes were already strongly supportive of women before training.”

The researchers also gathered some data that reflected attitudes toward racial minorities and found parallel results.

In comments to Fortune, lead researcher Edward Chang, a Wharton doctoral candidate, said the fact that diversity training had the least effect on the workers it was intended for is a problem. “Men and white people hold the most power, and we didn’t see much behavior change among those groups,” Chang said. “It’s not enough to get to people whose attitudes weren’t already supportive.”

Chang added that “a lot of companies are doing these things because they think they work, but we don’t have any evidence to support that.”

(Anna Vereshchak/iStock/Getty Images Plus)