What about the women?

“Education is one of the most important means of empowering women with the knowledge, skills and self-confidence necessary to participate fully in the development process.”

(from International Conference on Population and Development)

Today, in celebration of International Women’s Day, we are acknowledging our female students around the world. Whether they know it or not, they are actively participating in the ongoing movement for gender parity.

Coursera’s commitment to gender parity is woven into its DNA, with a leadership team that includes a female co-founder, a female president, and many other women in positions of substantial influence. But female leadership within Coursera isn’t enough; we recognize there are still barriers toward ensuring that every woman–around the world–has access to Coursera’s courses.

Women around the world

We conducted a demographic survey of over 250,000 Coursera students (thank you to all who participated!) and estimated the fraction of Coursera’s students in each country who are female:1

Fraction of female students by country.

Figure 1. Fraction of female students by country. This chart shows the estimated proportion of female students from each country in the Coursera user base. The dotted vertical line indicates the estimated proportion of female students overall.

A few noteworthy numbers:

  • Romania leads the pack in gender parity with a nearly exact 50/50 split between men and women.
  • On the opposite end of the spectrum, only 26% of students from India are female.
  • In the United States, women account for roughly 44% of the total population.
  • Overall, female students comprise 40% of the Coursera user base.

You can also view a full interactive map of all countries.

Changing the status quo

Changing the status quo requires understanding its roots. So we looked at which characteristics of a country correlated with its Coursera male-to-female ratio. We found a strong correlation with the country’s Gender Equity Index (GEI), compiled by the organization Social Watch. The GEI is composed of three subscores:

  • Economic activity
  • Empowerment
  • Education

We found that equity in education for a given country is the most significant indicator of Coursera female proportion.2 In other words, the access women have to education in their home countries overall correlates with whether they sign up for Coursera classes. Female economic activity in a country is also significantly associated with the Coursera female proportion, although this association is weaker.

Because most enrollees in Coursera classes are college graduates, we compared the percentage of college graduates in a country who are female with the percentage of Coursera students from that country who are female. There is a significant correlation ($r$ = 0.33, $p$ = 10-4), but in 95% of countries, the percentage of Coursera students who are female is smaller than the percentage of college graduates who are female. The mean difference between the two percentages (across countries) is 18%.

One explanation for this is that many Coursera classes are in engineering, which has lower rates of female enrollment. When we separately compare the fractions of enrollees in Coursera science, engineering and humanities classes who are female to the fractions of science, engineering, and humanities college graduates who are female, the discrepancies are smaller: 4% for engineering and 15% for humanities. In science, Coursera actually comes out ahead by 5%:

Comparison of proportion of women among Coursera enrollments and college students by course category.

Figure 2. Comparison of proportion of women among Coursera enrollments and college students by course category. To compute college graduation rates, we averaged over countries, weighting by the number of Coursera students from a country.

These differences might be explained by a variety of factors, among them other demographic variables (e.g., level of education). More broadly, a globally-relevant contributing factor to gender differences in participation is the global gender gap in Internet access, especially for sophisticated broadband applications.

Over time, advances in technology like mobile devices will eventually reduce gender inequity in access to the Internet. At the same time, however, there remain disparities in rates of female participation in different subject areas independent of access concerns, as illustrated below:

Median proportion of women among classes for each topic.

Figure 3. Median proportion of women among classes for each topic. The dotted line indicates the overall proportion of women on the Coursera platform.

The magnitude of gender differences in course participation also varies by other demographic factors, such as age:

Fraction of Coursera students who are female as a function of student age.

Figure 4. Fraction of Coursera students who are female as a function of student age.

Could the dip around age 40 in the United States be due to child-rearing? In India, where women bear children at younger ages, the dip shifts earlier. This is an intriguing hypothesis, but there are many other possible explanations: one might expect Coursera enrollments to be affected by when students in a country graduate from college or start their careers, both of which may vary by gender.

The existence of gender-based differences in Coursera course participation is consistent with differences observed in traditional learning institutions, but nonetheless poses a challenge to gender equality in the online setting. To have any hope of using online education to achieve gender equality worldwide, we must not be afraid to tackle the challenges head on. Only with measurement and analysis can we hope for improvement.

Gender and performance

We also looked at the grades women earned in classes and noticed a striking correlation: the gender gap in course performance shifts in favor of females in classes with a larger proportion of women. This is illustrated in the scatterplot below:

Relationship between gender composition and gender differences in course performance.

Figure 5. Relationship between gender composition and gender differences in course performance. Each point represents a single course session. The $x$-axis is the proportion of students in the class who are female. The $y$-axis shows the difference between the percentile of the median grade achieved by female course completers and the percentile of the median grade achieved by male course completers; positive values indicate courses where females outperform males. In general, the positive correlation ($r$ = 0.38, $p$ < 2 $\cdot$ 10-15) shows that the gender gap shifts in favor of women in classes with a larger proportion of women.

While the correlation in this graph is clear, inferring causality is more complicated. It is possible, for example, that when women see more women in a class (through interaction on class forums) it makes them more comfortable and motivates them to try harder. It is also possible that women who enroll in courses with high female proportions start out being more comfortable with the material, or intending to take the course more seriously, and the gender dynamics in the class have no effect on their performance.

The way forward

We have work to do to achieve gender parity in online education, but there is a bright side: the proportion of female students in Coursera classes has been increasing significantly over time.

Trends in female proportion among users and enrollments over time.

Figure 6. Trends in female proportion among users and enrollments over time. The proportion of Coursera users who are female (blue line) and the proportion of enrollments from females (green line) have been increasing over time.

We hope to see this trend continue. And even though the gender parity on the Coursera platform is not yet perfect, the access to education Coursera provides is already having a tangible impact on the lives of women – impact that has a clear downstream effect on their families and their communities. From heroes like Balesh Jindal, a physician in New Delhi whose Coursera experience drove her to help prevent sexual violence towards girls in her community, to Sharon Watkins, whose efforts to build local learning communities helped inspire Coursera’s Learning Hubs program, female Courserians themselves are living proof of how online education empowers women.

Let’s all celebrate!

We are proud to stand up for gender equality around the world and are determined to continue to find ways to increase access to education for women everywhere. So today, on International Women’s Day, we thank all of our students, male and female, for being part of our global classroom. Let’s all advance education and give women everywhere the power to learn anything they choose.


Notes

1 In order to eliminate biases due to differential survey response based on gender, we combined two independent techniques for gender estimation. Specifically, for each country, we first estimated the number of male ($n_M$) and female ($n_F$) Coursera students using a name-based gender classifier. We assessed the recall and precision of this classifier for males and females separately, using the subset of students who answered the demographic survey:

$$ \begin{align} recall_F &= P(\text{predicted female} \mid \text{female}) \end{align} $$ $$ \begin{align} precision_F &= P(\text{female} \mid \text{predicted female}) \end{align} $$ $$ \begin{align} recall_M &= P(\text{predicted male} \mid \text{male}) \end{align} $$ $$ \begin{align} precision_M &= P(\text{male} \mid \text{predicted male}). \end{align} $$

Using these accuracy metrics, we then computed corrected estimates of the number of male and female Coursera students per country as

This approach allows us to derive bias-corrected estimates of per-country proportions of female students on Coursera:

Here, the modeling assumption is that responding to the demographic survey is conditionally independent of a student’s name given his or her gender within each country stratum (so that the estimates of recall and precision are unbiased).

2 We ran a multiple regression on the three subscores of GEI, and found that the coefficients for both education and economic activity were significant but the former was more so ($p$ < 10-5, $p$ = 0.002, respectively). Each subscore of GEI is measured on a 0-1 scale: a 0.1 point increase in a country’s education subscore corresponded to a 4% increase in Coursera students who were female, while a 0.1 point increase in a country’s economic activity subscore corresponded to a 1.6% increase in Coursera students who were female. Histograms of the subscores are shown below:

Distribution of GEI subscores.

Comments