Beth Andres-Beck

A summary from YodasEvilTwin on Slashdot:

"The internet is dominated by sexist men, which discourages women from getting involved in related fields."  

I add a bunch more caveats, references and empirical data, but that is a good summary of how I interpret the evidence.

Introduction

There is currently a responsibility-dodging contest between industry and academia over who is to blame for the declining enrollment of women in Computer Science and declining employment of women in software development. I hear people in industry bemoan the "empty pipeline", while academics maintain that women aren't entering their programs because of perceptions of the industry.  I have compiled some data that may help resolve the question by highlighting a third factor common to both: access to an Internet-based culture of computing.

Two recent South African* studies found that students who had not had Internet access were more likely to major in Computer Science than those with previous experience [1][2]  They also found that the effect was more pronounced among young women than young men.

I put forward a related inflection point in the classic graph of CS enrollment in the US over time [3]:

Access to personal computers increased both men and women’s enrollment proportionally, whereas rising access to the Internet raised men’s enrollment significantly more than women’s enrollment and women’s enrollment began to decline before men’s peaked.  Access to non-social technology led to equal rises in men and women enrollment, whereas access to the Internet increased the gap in enrollment.  It becomes even more apparent if we consider that at the same time women’s enrollment was rising, so the percentage of all degrees awarded to women awarded in computer science rose less and fell further than the total number of women studying computer science:

It appears to me that the introduction of the Internet compounded the pre-existing gap in American enrollment.

While often the gender problem in Computer Science is lumped together with the gender problem in Engineering, the enrollment pattern for Engineering in the United States looks very different. Thus I am highly sceptical of research that conflates the two fields, and I will be focusing exclusively on Computer Science.

 

(Note: though this chart begins at 1970, explicit discrimination was not outlawed until 1972 and almost certainly contributed to very low initial percentages.)  

New Work

Here I examine the predictive power of Internet access to explain international variations in women's enrollment in computer science.

For these I will compare women’s representation in Computer Science to women’s representation among general college enrollment, in order to capture how women’s enrollment in Computer Science differs from other fields. This avoids confounding factors that may be shared by all college-educated professions such as family responsibilities or sexual harassment in order to focus on those contributing to the specific declines in CS.

Thanks to some data from Eurostat [4], I was able to produce cross-national comparisons between 30 nations.  I use a normalized number: the percentage of all college graduates who are women minus the percentage of CS graduates who are women, divided by the percentage of all college graduates who are women.  Thus, 0 is perfect equality and 1 is perfect inequality.  Here is the data for the 30 countries available:

The data most widely and relevantly available regarding Internet use was the percentage of the population who had never accessed the Internet in 2006. This is the time frame where the women above would have been choosing their field. Again, the data for those 30 countries:

I then graphed access to the Internet versus the variation from expected participation. I found a statistically-significant negative correlation:

Among these nations there are five Scandinavian countries that have had 100% internet access among young people for over a decade and very high internet access among the general population since 2004.  This is fundamentally different than the conditions in all other nations, where near-universal internet access, if it has been achieved at all, occurred less than a year before these data points.

If we exclude those nations that have hit that ceiling, we get an extremely statistically significant correlation:

Among the non-Scandinavian group, 39% of the total variation goes away if we control for Internet access.  We would expect the primary difference of women’s enrollment in the field to be national culturally-specific characteristics [5]; that there is a significant shared mediating factor is notable.

In addition, 23 countries had sufficient longitudinal data to do a temporal analysis.  Of those 23 countries, 5 had a positive correlation between rising Internet access and the parity of women’s enrollment in CS relative to other fields, all around the Mediterranean.  3 nations had no significant correlation.  1 had a weak negative correlation and 14, or 61%, had a strong negative correlation:

I conclude that in the last 10 years among many Northern European nations, rising Internet access is correlated with falling interest in computer science relative to other professions among women.  The group of Mediterranean nations that show a positive correlation should be a fruitful area for future research, but seem outliers from the Northern cohort.

Given that the same dynamic has be observed at an individual level, changes in enrollment over time at a national level and in cross-national analysis, a causal relationship is plausible, though not proven.  Notably, I do not have any longitudinal data prior to the introduction of the Internet for these countries to compare the post-Internet trends to and I have found no studies from other nations comparing individual-level Internet use with interest in Computer Science.

Discussion

I would caution against considering this support for any hypothesis relying on generalizations about the characteristics of women as a group, as the literature does not support gender as a distinguishing factor in career interest, skill or motivation [8].  Instead, the evidence suggests we should be looking to cultural factors amplified by the Internet that affect those identified by others as women, since the one thing women have in common is being perceived as female by members of the community. 

A global technical culture was noted by scholars as early as 1988 [6] and has since migrated online [7].  In many ways it has been one of the strengths of the field, but it may also be amplifying and spreading the alienating and exclusionary aspects of the current occupational culture.

The first hypothesis I propose is that Internet culture supports a belief in a meritocratic environment [9], which has been linked, ironically, to an increase in biased behavior [10] as it provides moral cover for prejudiced beliefs.  Encountering overt, covert or benevolent sexism undermines both women’s performance and interest [11].  Even if such beliefs were prevalent in professional spaces before the Internet, as masculine gender performance is common, aggressive and publicly visible in online forums [12]  women no longer have to be the target of such behavior themselves before college in order to associate it with the industry and choose an alternative career.

The second hypothesis is that the Internet encourages a sense of belonging [13] to the masculinized culture of software development [14], which alienates many women [15] by causing them to feel excluded from a camaraderie-focused profession [16].  Again, while this culture may have existed before the Internet, women with Internet access are likely to encounter such attitudes earlier and more frequently. To the best of my knowledge, whether the Internet has changed the culture of computing itself, either in America or internationally, is an outstanding question.

The one thing this correlation does imply is that programs to expose girls and young women to technology wouldn't be effective atboost the number of women entering software development. In fact, they might be counterproductive if they lead to earlier encounters with the existing occupational culture. Basically, free samples only boost sales of products people find enjoyable and we are therefore unlikely to solve these problems by writing off the experiences of the current generation and placing all our chips on the next.

It is much more controversial to say “we must change our culture” than “little girls should get to play with robots”, especially since many current programmers identify with that culture strongly [17].  Yet, if women aren’t becoming programmers because the occupational culture is unappealing, cultural change is the only solution that would lead to integration. The fears among those of us who do identify with the culture may also be unfounded: the aspects that are alienating are not necessarily central to the culture as a whole.

There is a second option: those programmers who do not wish to participate in the current exclusionary culture may spin off and create their own spaces with which the alienating members of the prior culture no longer identify.  Unfortunately, the recent push I have seen towards such an alternative culture is even more sexist than what came before [18].  On the other hand, new fields such as Biological Computation, Analytical Mathematics and Library and Information Sciences involve software development in different cultural contexts, producing vastly different gender balances [19].  By respecting, celebrating and promoting such fields we may be able to create a more inclusive discipline without significantly altering current norms of Internet behavior.

Of course, such respect for other fields would, itself, be a significant cultural change [20].

 

Citations

Where ever possible I have included a URL to the full text of the article or data set.

Not expecting everyone to read the pages and pages of resources here, I’ve added a brief description to many of them.

1.
L.F. Seymour, M. Hart, P. Haralambous, T. Natha,  and C. Weng,    "Inclination of scholars to major in information systems or computer science",  South African Computer Journal, 2005, pp.97-106. 

http://arnetminer.org/publication/inclination-of-scholars-to-major-in-information-systems-or-computer-science-1054288.html 

2.
Jacobs, C. and Sewry, D.A.  Learner Inclinations to Study Computer Science or Information Systems at Tertiary Level.  South African Computer Journal, 2009.
http://sacj.cs.uct.ac.za/index.php/sacj/article/viewFile/25/10

3.
This data is from the Integrated Post-Secondary Education Data System, available from: http://nces.ed.gov/

4.
All the data used to create these charts is available from: http://epp.eurostat.ec.europa.eu

5. 

Neither qualitative nor quantitative: can best be described as “making stuff up on the basis of some open-ended surveys other people gave”.  This is a paper where a number of researchers go hunting for cultural explanations that can explain women’s declining participation in IT.  Despite huge amounts of work on the topic, they do not identify any changing cultural factors that would explain the decline: instead they essentially say, “well, there are probably reasons women make the choices they do, so we should assume they exist!”  They manage to totally mentioning any factor other than computer access that might set IT professions apart from other professions.  It is also notable for being a 43 page long paper that never once mentions men in a professional context: that they are able to erase 79% of the IT workforce that entirely is actually quite impressive.  However, I think it demonstrates the assumption on the part of many researchers that the differences in enrollment between nations can be explained by cultural factors.
Trauth, E.M., Quesenberry, J.L., and Huang, H. “A Multicultural Analysis of Factors Influencing Career Choice for Women in the Information Technology Workforce,” Journal of Global Information Management (16:4), 2008a, pp. 1-23.
http://www.personal.psu.edu/gms/trauth/Trauth%20et%20al-JGIM.pdf 

6. 

1988 study comparing British, West German and American companies found a shared occupational culture of technical excellence among the geographically disparate companies.
Gerpott, T. J., Domsch, M. and Keller, R. T. (1988), CAREER ORIENTATIONS IN DIFFERENT COUNTRIES AND COMPANIES: AN EMPIRICAL INVESTIGATION OF WEST GERMAN, BRITISH AND US INDUSTRIAL R&D PROFESSIONALS. Journal of Management Studies, 25: 439–462. doi: 10.1111/j. 1467-6486.1988.tb00709.x

7. 

Sociological description of the technical culture on the Internet during the late 90’s.
Turner, Fred. "Cyberspace as the new frontier? Mapping the shifting boundaries of the network society." June 6. 1999. Red Rock Eater News Service.
http://www.stanford.edu/~fturner/Turner%20Cyberspace%20Red%20Rock%20Eater%201999.pdf

Also, a case study from Brazil:

A narrative description of the cultural connection between a few Brazilian programmers and an international, Internet-based culture of software development.
Takhteyev, Y. Jeeks: Developers at the Periphery of the Software World, the Annual Meeting of the American Sociological Association, New York, NY, August 10-17, 2007.
http://takhteyev.org/papers/Takhteyev-2007-Jeeks.pdf

8. 

A quantitative study finding that gender alone is not a determining factor for “career anchors”, or what people want from a job.  “Women” was not a useful category when trying to predict what people were looking for from their employment.  While it is unsurprising that gender isn’t a defining characteristic in this domain, some people seem very wedded to the idea.
Quesenberry, J.L. and Trauth, E.M. (2008) “Revisiting career path assumptions: the case of women in the IT workforce” Proceedings of the International Conference on Information Systems (Paris, France, December).
http://www.eileentrauth.com/uploads/4/6/7/6/4676002/quesenberrytrauth_icis-2008.pdf

9.

Narrative description of the variations in hacker culture over time.  Primarily drawn from ethnographic observations.
Hannemyr, Gisle. "Technology and pleasure: Considering hacking constructive" First Monday [Online], Volume 4 Number 2 (1 February 1999)

http://www.firstmonday.org/htbin/cgiwrap/bin/ojs/index.php/fm/article/view/647/562
 

Also, for a historic instance in the creation of culture, see:

Historical description of the online community that sprung up around the Whole Earth Catalog.
Turner, F. (2005) ‘Where the Counterculture Met the New Economy. The WELL and the Origins of Virtual Community’, Technology and Culture 46: 485–512.
http://www.stanford.edu/~fturner/Turner%20Tech%20&%20Culture%2046%203.pdf
 

And the manifestation of this in mainstream offline publication, Wired magazine:

A journalistic piece describing the advertising and values of Wired magazine in the early 90’s.
White, Keith. The Killer App: Wired Magazine, Voice of the Corporate Revolution The Baffler 1994 -:6, 23-28
http://www.zooid.org/~vid/txt/decon_wired.html 

10. 

This study describes three experiments asking managers to make compensation decisions about fictional employees.  The found that when the managers were told the company’s core values emphasized meritocracy the managers favored male employees over equally qualified and performing female employees, while in a control condition where they were not told the organization was meritocratic women were not discriminated against.
Castilla, Emilio J., and Stephen Benard. “The Paradox of Meritocracy in Organizations.” Administrative Science Quarterly 55 (2010): 543-576.

http://dspace.mit.edu/handle/1721.1/65884 

11.

Quantitative experimental study demonstrating the mechanism of sexist behavior among engineering students and the effect such behaviors have on female students.  They first identified male students who held sexist beliefs and observed that they acted in a dominant and sexually interested way towards an ostensible female classmate.  They then had female students interact with either the sexist male engineering students or an actor trained to act the same way and observed that the women performed worse on a subsequent engineering test than women who had interacted with nonsexist men.  They also found that interacting with sexist men didn’t cause women to perform worse on an English test, a field that isn’t subject to masculine stereotypes.
Logel C, Walton GM, Spencer SJ, Iserman EC, von Hippel W, Bell AE. Interacting with sexist men triggers social identity threat among female engineers.J Pers Soc Psychol. 2009 Jun;96(6):1089-103. Erratum in: J Pers Soc Psychol. 2009 Oct;97(4):578.
http://www2.psy.uq.edu.au/~billvh/LWSIvHB.JPSP.09.pdf

12.

Narrative overview of gender performance in online forums.  Includes citations to observational works covering stereotypical performance and cross-gender participation.  Behind a pay wall.
Marshall, Jonathan. "Online Life and Gender Vagueness and Impersonation."Encyclopedia of Gender and Information Technology. IGI Global, 2006. 932-938. Web. 4 Jun. 2012. doi: 10.4018/978-1-59140-815-4.ch147
http://www.irma-international.org/chapter/online-life-gender-vagueness-impersonation/12852/

Qualitative description and quantitative analysis of gendered interactions in mailing list and chat rooms involving the harassment of women.  Describes a morphology of sexual harassment that can be applied to further cases.
Herring, S. C. (1999). The rhetorical dynamics of gender harassment on- line. The Information Society, 15 (3), 151-167. 

http://ella.slis.indiana.edu/~herring/harassment.pdf

Qualitative analysis of patterns of discussion on LINGUIST mailing list in 1995.  Specifically, this describes the pattern of antagonistic and argumentative engagement common on that list, and offers evidence that that discursive approach discouraged some men and most women from participating in the discussion.
Herring, S. C. (1995). Men’s language on the Internet. Nordlyd, 23, 1-20. (Proceedings of the 2nd Nordic Language and Gender Conference, November 2-4, 1994.) 

http://ella.slis.indiana.edu/~herring/men.1995.pdf

13.

Experimental exploration of community engagement.  They did A/B testing with new features displayed in profiles for movielens.org, as well as reproducing the results in a laboratory setting.  One was based on the idea of forming bonds with individual members with similar taste in movies while the other encouraged members to identify with a subgroup formed around shared tastes.  They found that creating subgroups significantly increased both reading and participation, while introducing people to individual other people slightly increased contributions without increasing reading behavior.  This paper is particularly recommended for anyone attempting to engineer social behavior online: the analysis includes more decomposition of the results than I’ve covered here.
Ren, Y., Harper, F. M., Drenner, S., Terveen, L., Kiesler, S., Riedl, J., et al.
Increasing Attachment to Online Communities: Designing from Theory. MIS Quarterly.
http://www.tc.umn.edu/~chingren/pdf/papers/Ren-IncreasingAttachment-2010.pdf

14.
Currently:

Quantitative results measuring gender stereotypes of IT-related skills. Specifically, this compared younger workers to previous studies and found that while they still stereotyped technical skills as masculine and people skills as feminine, business-related skills had migrated from being considered “masculine” to being considered gender-neutral.  Progress has been made, even if technical skills are still currently subject to stereotyping.
Trauth, Eileen M.; Joshi, K. D.; Kvasny, Lynette; Chong, Jing; Kulturel, Sadan; and Mahar, Jan, "Millennials and Masculinity: A Shifting Tide of Gender Typing of ICT?" (2010). AMCIS 2010 Proceedings. Paper 73. http://aisel.aisnet.org/amcis2010/73
http://faculty.ist.psu.edu/lkvasny/Amcis2010-Millenials.pdf

Historic:

Qualitative review of the history of computing as a field and the role gender played in that history, mostly employing primary documents (including a bunch of quite funny ads).  Unlike the following citation, he attributes the ultimate masculine associations of computer programming to the masculine fields it interacted with, specifically systems analysis, accountants and punchcard machine operators.  He does identify data entry as feminized work, but rather than suggest a fuzzy border between that work and programming he suggests that the proximity to data entry has been seen as a threat to programmers’ identities and led the profession to assert strong boundaries between the two, frequently using gender as the dividing line.
Haigh, Thom. Masculinity in the History of Computing(s). May, 2008. CBI Workshop on History.
http://www.tomandmaria.com/tom/writing/Haigh%20Masculinity%20for%20Workshop.pdf

And also:

Historic narrative, with supporting evidence.  This describes how programming swapped from being considered secretarial “women’s work” to companies recruiting programmers with personality, psychology and intelligence tests that were both biased and administered almost exclusively to men.  In the 1950’s coding and programming turned from low-status jobs into an “arcane black art”, which appealed to certain status-seeking, socially-inept men.  Such men were actively recruited because they fit the new stereotype of “programmer” than hiring managers had, since there were few experienced programmers at that point and coding on a white board was still years off as an interviewing technique.  In short, Ensmenger argues that programming became stereotyped because management only hired stereotypical programmers.

He also presents the beginning of the struggle between manager/capitalist/professional attitudes and craftsmanship/labor/cowboy coder attitudes, which continue to this day.

 Ensmenger, Nathan. The Computer Boys Take Over
http://mitpress.mit.edu/catalog/item/default.asp?ttype=2&tid=12264
a short version of which is available for free here: 

http://www.ischool.utexas.edu/~nathan/files/ensmenger-gender.pdf

15. 

Quantitative experimental measurements of interest in CS classes.  They had one group of students classify items according to whether they associated the items with computer science.  They then constructed two virtual classrooms, one containing items identified as “stereotypical” and the other containing “neutral” items.  Then then investigated how interested non-CS majors were in taking an introductory class held in the offered classroom.  When presented with the stereotypical classroom women were far less likely to enroll in a computer science class, while men were slightly more likely to enroll.  Women’s interest in taking the class was determined by “ambient belonging”, which is a measure of whether or not one feels alienated from the activity on the basis of passive cues.
Sapna Cheryan, Andrew N. Meltzoff, Saenam Kim, Classrooms matter: The design of virtual classrooms influences gender disparities in computer science classes, Computers & Education, Volume 57, Issue 2, September 2011, Pages 1825-1835, ISSN 0360-1315, 

10.1016/j.compedu.2011.02.004.
http://ilabs.uw.edu/sites/default/files/Cheryan_Meltzoff_Kim%20Computers%20and%20Education%20(2011).pdf
Side note: it is interesting how much more diverse actual science fiction & fantasy (48% WorldCon attendence), World of Warcraft subscriptions (40%), video games (33%) and Star Trek fan sites (34-48%) are than computer science itself.  This may explain the presence of women in software development who also identify strongly with the culture, despite others viewing it as highly masculine.

16. 

Experimental study where students read a job overview that either used masculine (he), gender-inclusive (he or she) or gender neutral (one) language.  Men did not respond to the choice of language, but women who read the masculine job description reported higher levels of ostracism.
Jane G. Stout Nilanjana Dasgupta. When He Doesn’t Mean You: Gender- Exclusive Language as Ostracism Pers Soc Psychol Bull June 2011 37: 757-769, doi:10.1177/0146167211406434
http://scholarworks.umass.edu/cgi/viewcontent.cgi?article=1288&context=theses

17. 

Qualitative analysis of a data set from Slashdot.  It finds programmers asserting a specific identity defined against the identity of project managers in exchanges that have “an antagonistic quality”.  In their data set, programmers thought they were more technically knowledgable than PMs and cared more about quality code.  PMs asserted that they knew more about actually producing software and were more action-focused, seeing programmers as “whiners”. 
Case, P. and Pineiro, E. (2009) Stop whining, start doing! Identity conflict in project managed software environments. ephemera, 9
(2). pp. 93-112. ISSN 1473-2866
http://eprints.uwe.ac.uk/11246/2/9-2case-pineiro.pdf

I believe most of Dilbert could also be used as evidence for this conflict.

18.

While this hasn't been explored in academic journals yet, but Business Weekly has a piece on brogramming.  It is non-academic, but describes a phenomenon that has been widely discussed online, where, in contrast to the hacker culture, non-work activities are valued in a fraternity-like culture of alcohol consumption, fashion and masculinity.  I know personally I’ve met some members of this sub-sub-culture.
http://www.businessweek.com/articles/2012-03-01/the-rise-of-the-brogrammer 

19. 

A collection of quantitative numbers on gender balance in various college majors that involved IT.  Basically Computer Science, Electrical Engineering and Information Technology Management are heavily male-dominated while Business Management, Mass Communication Technologies and Library/Information Sciences are gender-balanced or have slightly more women than men.
Wendy Cukier, Denise Shortt, and Irene Devine. 2002. Gender and nformation technology: implications of definitions. SIGCSE Bull. 34, 4 (December 2002), 142-148. DOI=10.1145/820127.820188
http://www.jise.org/Volume13/Pdf/007.pdf 

20.

Qualitative interviews that were analyzed to identify the features unique to the subculture of IT workers.  Notably, they identified a common scorn for non-technical people, which makes it harder to collaborate across disciplinary boundaries.
Indira R. Guzman, Jeffrey M. Stanton, Kathryn R. Stam, Vibha Vijayasri, Isabelle Yamodo, Nasriah Zakaria, and Cavinda Caldera. 2004. A qualitative study of the occupational subculture of information systems employees in organizations. In Proceedings of the 2004 SIGMIS conference on Computer personnel research: Careers, culture, and ethics in a networked environment (SIGMIS CPR '04). ACM, New York, NY, USA, 74-80. DOI=10.1145/982372.982388 http://doi.acm.org/10.1145/982372.982388
Full text unfortunately behind a pay wall.

 

* These studies were mistakenly referred to as South American studies in the first version of this post.