Wednesday, October 29, 2014

Stereotype Threat and the Math Gender Gap

Negative stereotypes about women’s and girls’ abilities in STEM persist despite considerable gains in the last few decades. Stereotype threat is related to the anxiety women face in a situation where they have the potential to confirm a negative stereotype about women as a group. Two common stereotypes are prevalent in society and form the foundation for the threat condition: (1) girls are not as good as boys in STEM; and (2) scientific work is better suited to men than to women. As early as elementary school, kids are aware of these stereotypes and can express stereotypical beliefs about which STEM courses are suitable for girls and boys (Farenga & Joyce 1999; Ambady et al. 2001).
Girls and young women are aware of, and negatively affected by, the stereotypical image of a scientist as a man. Although largely unspoken, negative stereotypes about women and girls in STEM are very much alive (Buck et al. 2008). A large body of experimental research has found that negative stereotypes affect women’s and girls’ performance and aspirations in STEM. Even girls who strongly identify with math - who think that they are good at math and being good in math is important to them - are susceptible to the effects (Nguyen & Ryan 2008).
Stereotype threat may help explain this discrepancy: girls get higher grades in STEM classes but lower scores on high-stakes tests like the math SAT and AP calculus exam. Stereotype threat may also help explain why fewer girls express interest in careers in mathematically demanding fields. Girls may be trying to reduce the likelihood that they will be judged through the lens of negative stereotypes by avoiding these fields all together.

Stereotype threat was first discovered in studies of African American students taking an intelligence test. When the experimenter was white, the students performed significantly worse than white students. The scores improved, however, when the experimenter was African American (Katz, Roberts, & Robinson 1965). This ground-breaking study was the first to show that task performance could be reduced by invoking racial stereotypes. African American students under stereotype threat conditions had higher blood pressure levels than whites; no difference in blood pressure occurred in the non-threat situation (Blascovich et al. 2001). So the affects of the increased anxiety were both physical and measurable.
When testing under the threat condition, stereotyped individuals: (1) made more of an effort; (2) attempted more items; (3) reread items more often; (4) worked slower; and (5) worked with less accuracy (Steele & Aronson 1995). None of these are desirable when taking a timed test, especially one with penalties for wrong answers.
Example 1
College students with similar math abilities are divided randomly into two groups. Both groups contain men and women. The groups are separated into two rooms and all the participants are given the same math test. The groups don’t interact until the study is complete. Before the test starts, Group 1 is told that men perform better than women do on the test; this is the stereotype-threat group. Group 2 is told specifically that “this test was designed to be gender neutral;” this is the no-stereotype-threat group.

The results are shown in the graph. The researchers found that women did significantly worse than men did in the stereotype-threat group. Women scored 5 on average and men scored 25 on average. But in the no-stereotype-threat group, women and men performed equally well, within uncertainties. The researchers concluded that because women’s performance improved when there was no threat, it must have been something about the testing situation rather than women’s ability that accounted for the difference in their performance in the threat group compared to the no-threat group. This type of result has been shown again and again in other experiments. Negative stereotypes about girls’ and women’s abilities in math and science adversely affect their performance in these fields. These findings also point to some good news:  stereotype threat is largely situational, and girls’ performance improves when the threat is removed (Spencer et al. 1999).
Example 2
In this study, students are selected based on their excellent math SAT scores and divided randomly into two groups. Both groups are given a math test of GRE level problems. For Group 1, the test is given under normal “GRE-type” conditions. Group 2 is told specifically that “this test was designed to be gender neutral.” In Group 1, the men and women got the same average score, but in Group 2, the women did significantly better than the men. Why? On the SAT test, women got scores equal to the men, but with stereotype threat in place. On the GRE test, there was no stereotype threat, so women were able to perform at their “real” level (Good et al. 2003).
The figure shows the “gender gap” on the math SAT as a function of time. Part of the reason that the gap has narrowed is that the test designers have taken more care to eliminate word problems that would be easier to answer is the student had taken an auto mechanics or machine shop class then if the student had taken a cooking or sewing class. This gender separation into auto mechanics/shop versus cooking/sewing lasted well into the 1970s in many US school districts. Despite these efforts, however, a persistent gap of 30+ points remains. Researchers find that stereotype threat may account for ~ 20 of these points. 20 points out of 800 may seem small, but reducing stereotype threat could eliminate two-thirds of the remaining gender gap on the math SAT.
If the gender gap were innate, and boys were really better at math than girls, then the results from Iceland in the figure would be impossible. The top plot shows the math (green) and reading (grey) gender gaps for different countries. (Note: the reading gender gap is serious, but it is beyond the scope of this post.) Note that for Iceland, the math gap is positive, meaning that Icelandic girls score higher on standardized math tests than Icelandic boys. The bottom plot shows a World Economic Forum's gender gap index for women’s emancipation for the same countries. The higher the level of women’s emancipation, the smaller the math gender gap. The index is calculated from a variety of features, including the:
-fertility rate;
-female unemployment rate;
-women who have an account in a formal financial institution;
-ability of women to rise to positions of enterprise leadership;
-firms with female top managers, share of women on boards of listed companies; and
-firms with female participation in ownership.
Thank you, Iceland!
Here’s some good news to take away from these studies. For women and girls that have taken a standardized math test, your innate math ability may be higher than your score suggests!
Much of this information is summarized in the 2010 report entitled, Why So Few? Women in Science, Technology, Engineering, and Mathematics, by the American Association of University Women (AAUW). For lots more studies of stereotype threat, check here.

Ambady et al. (2001) Psychological Science, 12(5), 385–90.
Blascovich et al. (2001) Psychological Science, 12(3), 225–29
Buck et al. (2008) Science Education, 92(4), 688–707
Farenga & Joyce (1999) Science Education, 83(1), 55–76
Good et al. (2003) Applied Developmental Psychology, 24, 645–62
Nguyen & Ryan (2008) Journal of Applied Psychology, 93(6), 1314–34
Spencer et al. (1999), Journal of Experimental Social Psychology, 35(1), 13
Steele & Aronson (1995) Jour of Personality & Social Psych, 69(5), 797–811