A new study suggests online education may not be the panacea to income-based education gaps that proponents have claimed.

Published in the journal Science, the study found a strong correlation between socioeconomic status and the likelihood of enrollment in and completion of Massive Open Online Courses. The study examined U.S. residents enrolled in 68 online courses offered by Harvard and MIT through edX, the non-profit virtual education platform the schools co-founded in 2012.

Using census data and user-provided addresses, researchers John A. Hansen and Justin Reich found that those who lived in more affluent neighborhoods were more likely to register and complete MOOCs. For each $20,000 rise in a neighborhood’s median income, the odds of participation in a MOOC increased by 27 percent, according to the study. {shortcode-9ce07c0ec107712fc0f45e6532e70344d083ad1b}

While the average MOOC user lives in a neighborhood with a median income $11,998 above the national average, the difference is $23,181 among 13 to 17-year-olds, an age range of critical importance in determining post-secondary focus, the researchers say.

EdX publishes online courses created by dozens of partner schools. HarvardX, also founded in 2012, is Harvard’s MOOC-production arm.

Three years since its founding, HarvardX has drawn some praise from administrators but is still not financially sustainable, and it has faced criticism from some professors for the large—and largely unpaid—time investment required to make courses.

Hansen and Reich’s study, which was conducted using enrollment data provided by Harvard and MIT, raises concerns that MOOCs “can exacerbate rather than reduce disparities in educational outcomes related to socioeconomic status.”

According to edX CEO Anant Agarwal, the study draws attention to a problem of which edX is already aware. Two-thirds of enrollees in edX MOOCs are “highly educated,” he wrote in a statement.

“We know that we have a long way to go to realize the promise of offering quality education to more people around the world without access or means,” Agarwal wrote.

According to Government professor Dustin Tingley, who oversees research on HarvardX pedagogy, part of the disparity in MOOC enrollment could be a result of the advanced nature of Harvard and MIT courses.

“We weren’t going to be offering courses in remedial education and auto mechanics,” Tingley said.

While Robert A. Lue, HarvardX’s faculty director, praised the study for “concretizing” the problem, Lue defended HarvardX’s contributions to access.

“To claim that we have not made a contribution to global access, in my opinion, is inaccurate,” Lue said.“To think that any one institution can solve the access problem, frankly, is naive.”

Technology alone is not enough to improve access, Lue said, citing HarvardX’s efforts with the Allston Ed Portal as a model for community-conducted support for online education.

Despite the concerns outlined by the study, Hansen said online education could still improve education disparities.

“Despite my finding, I think this is still something net beneficial to society, even if my findings aren’t the ones we wish were in the data,” Hansen said.

The researchers drew parallels between MOOCs and earlier attempts to democratize education through technology like radio and television courses.

“A new broadcast technology comes along, and at some point it seems that someone will say, ‘This poses great promise for reducing educational inequities in the country,’” said Hansen, a doctoral candidate at the Graduate School of Education.

The extent to which the impact of these broadcast technologies can be measured was unclear—until now, the researchers say.

“With MOOCs, it is nice because you actually have more data you can use to address it empirically,” Hansen said.