Women in STEM, Women in Computer Science: We’re Looking At It Incorrectly

The usual presentation of data about U.S. women’s undergraduate degrees in STEM disciplines (science, technology, engineering, and math) is based on the percentage of degrees in each field that are awarded to women. For years this by-discipline view of the data has been used to support claims of improvement in women’s participation in these fields. Unfortunately, this perspective serves to inflate the extent of the improvement, and the actual situation of women in STEM fields is more serious than commonly believed. The by-discipline view can also suggest research questions about gender in STEM that turn out to be somewhat misguided.

By combining National Science Foundation (NSF) and National Center for Education Statistics (NCES) data for 1966-2012, we can construct the typical by-discipline view of women in STEM, with its indication of progress over time.

 

During the time period examined, each disciplinary area saw an increase in the proportion of degrees that were awarded to women. The categories used are NSF catch-alls as follows: engineering; physical sciences; earth, atmospheric and ocean sciences; mathematics; computer science; biological and agricultural sciences. The “All” category refers to the combination of these individual disciplines. This view of the data does not, however, tell a completely accurate story.

During the same time period, 1966-2012, there has been a seismic shift in the makeup of the undergraduate college-going population. First, that population has grown considerably. In 1966 there were 524,008 bachelors degrees awarded nationwide. By 2012 that number had more than tripled to 1,810,647. Second, the gender demographics changed quite dramatically, as shown below. In 1966 the graduating baccalaureate pool was 42.6% women, while the 2012 pool was 57.4% women.

 

With a change of this magnitude we would expect to see a larger number of women in every field, and it is easy to mischaracterize the resulting effects. Consequently, a different form of analysis is necessary to accurately gauge the extent of change in women’s participation in the STEM disciplines. As a first step, breaking with the conventional by-discipline analysis, I examine women’s STEM degrees as a percentage of women’s bachelors degrees overall, again comparing 1966 and 2012.

 

 

This view of the data, supported by the table below, shows that women earned biology degrees in 2012 at about twice the rate they did in 1966. They also earned engineering degrees at a significantly higher rate. There was a slight increase in earth sciences, a decrease in physical sciences, a significant decrease in mathematics, and improvement in CS. In fact, driven by the positive shifts in Biology, Engineering, and CS, in 2012 women earned 11.22% of their degrees in STEM, compared to only 7.47% in 1966.

Once we view women’s degree data this way, we should also consider a comparable view of men’s degree data. This will give us a mechanism for gauging the degree of disparity in terms of the rate at which women are earning degrees in STEM disciplines. Analysis of men’s STEM degrees as a percentage of men’s total degrees gives the following:

 

This view of the data shows that, overall, men are pursing STEM fields at a lower rate today than they did in 1966. Men are earning a lower percentage of their degrees in every STEM discipline except for computer science. Overall, in 1966 men earned 28.68% of their degrees in a STEM field, which dropped to 24.06% by 2012.

Now we can do a true side-by-side comparison of the rate at which these two populations earn degrees in the STEM fields. 

 

While the by-discipline view of STEM degrees is far from rosy, this by-gender view of the data facilitates a more accurate assessment of the situation for women in STEM, and we can build on this to understand the ways in which the by-discipline view can mislead. If there were parity between men and women in STEM disciplines then they would graduate with degrees in those fields at the same rate relative to the size of their respective pools. We see this only in Biology where the graduation rate is almost equal (7% of women’s 2012 degrees were earned in Biology versus 6.77% of men’s 2012 degrees). In all other STEM fields men earned degrees at a higher rate and women are far from parity. Below I have repeated the comparison of the rate at which men and women earn STEM degrees, including a “gender disparity ratio”, the ratio of men’s rate/women’s rate, which serves as one measure of how much faster the rate is at which men are earning degrees in each discipline (except Biology).

 

This shows that men earned degrees at 5.7 times women’s rate in the engineering disciplines, and at 6 times women’s rate in computer science. We can extend this analysis by computing the gender disparity ratio for 1966 and comparing that to the ratio we computed for 2012, based on the rate at which men earned STEM degrees in 1966 as a percentage of men’s degrees, and similarly for women. A comparison of the 1966 and 2012 data, given in the next table, shows that there has been improvement in each of the STEM disciplines except for math and CS. In 1966 men earned math degrees at 1.48 times the rate women did, and that increased to 1.8 times by 2012. In 1966, admittedly when CS was a new academic discipline, men earned CS degrees at 3 times the rate women did, and this increased to 6 times women’s rate by 2012.

 

If we look at the by-discipline and by-gender views side by side, certainly both indicate that there is still much work to be done, even with the relatively more optimistic sense of progress presented in the by-discipline view. If we dig deeper, however, into the ramifications of these two perspectives, we see that the by-discipline view can mislead us about the degree of progress and about questions to consider as we continue to try to change the situation for women in STEM disciplines.

Certainly the by-discipline view on the left gives a sense of progress because in 2012 women earned a higher percentage of degrees in each discipline than they earned in 1966. Unfortunately this view, because it does not take into account the overall demographic change in the baccalaureate population, gives a false picture of the gender disparity. The by-gender view on the right gives a much more accurate picture of how far from parity women are in terms of the rate at which they earned STEM undergraduate degrees.

There are a few ways that we can demonstrate how the by-discipline perspective masks the effect of the changing demographics and the true gender disparity. One way to contrast the two perspectives on the data is to convert the by-discipline information to a gender disparity number. For example, in 2012 women earned 18% of all CS degrees, so we can say that the by-discipline view of the data yields a gender disparity ratio of 4.5 (82% / 18%), whereas the by-gender view gives a gender disparity ratio of 6 (men’s rate of 5.08% divided by women’s rate of 0.84%). A similar computation for all of the disciplines (next table) shows the extent to which the by-discipline perspective masks the full extent of the gender disparity. In other words, the by-gender analysis, taking into account the larger size pool of women students, indicates a more serious gender disparity than we see with the by-discipline data.

 

Focusing on two disciplines gives us another way to demonstrate that the by-discipline perspective can mask the effect of the demographic change in the baccalaureate population. The next graph shows both by-discipline and by-gender views of women’s CS undergraduate degrees for the period 2002-2012. The blue line shows the by-discipline percentage of CS degrees earned by women, while the gold line shows the by-gender percentage of women’s degrees earned in CS.

 

If we look only at CS, this graph does not tell us anything that we do not already know from the previous bar charts and tables. The difference between the two ways of viewing the data becomes clearer if we add to the picture information about women’s math degrees. The by-discipline data shows women earning a significantly higher percentage of math degrees than CS degrees (43% vs 18% in 2012, for example). This might lead one to consider constructing studies to examine whether women approach these fields differently or whether the climate in one is significantly different than in the other. It turns out, however, that examining differences between these fields is completely unwarranted. We can see this by adding to the graph the two data views of math degrees. The graph below shows women’s math and CS degrees as a percentage of degrees in those fields, as well as women’s math and CS degrees as a percentage of all women’s degrees.

 

The rate at which women earned math degrees in 2012 was basically identical to the rate at which women earned CS degrees. The by-discipline view, however, deceives us into thinking women are better represented in math than they are in CS. This is because of the combined effect of the larger number of women overall and the large decline in the rate at which men are earning degrees in math. In fact, in 2012 the raw number of men earning math degrees was lower than it was in 1966, even though more than twice as many men graduated from college in 2012 than in 1966. By contrast, the raw number of women earning math degrees was higher in 2012 than in 1966, and almost 5 times as many women graduated from college than in 1966. But the raw number of women’s math degrees earned in 2012 (8536) is almost identical to the number of CS degrees (8730). In the by-discipline view, the overall shifts in the population graduating from college mask the fact that there is no difference in the rate at which women pursue math and CS.

This analysis supports an argument that the “percentage of women by discipline” approach should be abandoned, or at least used cautiously and in conjunction with a “by gender” analysis. Because the by-discipline approach does not take into account overall demographics at the undergraduate level, it is masking exactly how serious the situation is for women is in some disciplines.

A quick look at the graduate degree situation strengthens the argument for use of a by-gender analysis. The overall populations of Masters degree recipients and PhD recipients have changed substantially over time, calling into question the veracity of by-discipline analyses of those groups. The pool of Masters degree recipients was roughly 38% women in the late 1960s, and increased to 59.9% women in 2012. Similarly, the PhD recipient pool was 9.6% women in the late 1960s, up to 51.4% in 2012 (note that NSF and NCES data both show the same trend, although NSF includes fewer types of doctoral degrees in the summary data). By the same reasoning laid out above, any analysis that reports on percentage of total Masters or PhD degrees awarded to women in a discipline could be flawed and lead to incorrect conclusions about the participation of women in many fields.

A closer look at the PhD situation in CS reveals that, in the period 2002-2012, women consistently earned between 19% and 23% of all computer science PhDs. Yet the rate at which women earned those degrees as a proportion of all women’s PhDs is less than 1/3 the rate at which men earned their CS PhDs as a proportion of all men’s PhDs. The following figure compares women’s and men’s CS PhDs as a percentage of all women’s or men’s PhDs respectively.

 

We can see that women clearly are earning CS PhDs at a lower rate than do men, but the rate has been relatively consistent during the last decade, both as a percentage of women’s degrees and as a percentage of total CS PhDs. A comparison, however, between the rate at which women have been earning undergraduate computer science degrees and the rate at which women have been earning CS PhDs may spell trouble ahead because of the slowing rate at which women are pursuing undergraduate CS degrees.

 

It is critical to ongoing analysis of women in STEM overall, and in computer science in particular, that we use a by-gender analysis in order to derive an accurate picture of the situation. The by-gender analysis will serve as a check against the inflated sense of progress that results from using only a by-discipline analysis, and the by-gender analysis can facilitate a more accurate identification of trends and research directions, helping us avoid spurious research efforts that are not supported by actual data.

 

 

Data Sources:
National Science Foundation, National Center for Science and Engineering Statistics. 2013. Women, Minorities, and Persons with Disabilities in Science and Engineering: 2013. Special Report NSF 13-304. Arlington, VA. Available at http://www.nsf.gov/statistics/wmpd.

National Center for Education Statistics. 2013. Digest of Education Statistics. Advanced release of selected 2013 digest tables. Available at http://nces.ed.gov/programs/digest/2013menu_tables.asp.

 

Thanks:
I have discussed these ideas with various people over the course of many years.  For assistance during development of this post, particular thanks go to Roger Hoerl (Union College) who helped me decide how to most clearly graph and contrast the data, and to Eshragh Motahar (Union College), Jodi Tims (Baldwin Wallace University), Elizabeth Hawthorne (Union County Community College), and Susan Yohn (Hofstra University) who provided valuable feedback.

This material is copyrighted 2014 by Valerie Barr. Permission is granted for re-use ONLY with proper attribution.