Geek Gene Hypothesis. We consider our evidence weak
because our study was underpowered, and caution should
be taken in interpreting the lack of significance in the second treatment.
We were initially surprised that regularly looking at
histograms of grades was associated with a higher
score for seeing- bimodality. This led us to add our third
research question, based on the idea that it could be
that the more often you look at your grades, the more it
solidifies your conception of what your grades are like.
This supports our observation that categorizing distributions as bimodal increases belief in innate ability.
System justification theory explains how once you are
forced to take a position on a subject, you are more likely
to believe and defend it.
11 Our approach to priming—
stating that it is a commonly held belief that CS grade distributions are bimodal—may have strengthened our
participants’ beliefs about the bimodality of CS grades.
Because the survey presents us, the researchers, as authority figures, and we imply that grades may be bimodal,
some participants could assume it to be true because of
When we piloted our survey, some participants opined
that they believed that some students were predisposed
because of prior experience, rather than inherent
We did not have a representative sample of CS educators. The educators who participate in CS education communities are generally much more invested in their
teaching than their peers who do not. Furthermore, some
of our participants may be familiar with Ahadi and Lister1,
which could have influenced their responses. But we
would expect the SIGCSE community to be less inclined
to believe in innate ability than their non-SIGCSE peers.
We still had enough participants who agreed with the
hypothesis for us to conduct our analysis. Future work is
needed to replicate our findings with a more representative sample of CS educators.
Supporting literature. Our findings agree with the psychology literature: people’s biases affect their decision-making more when they are judging more ambiguous
10 For example, Heilman et al. found that
resumes of extremely qualified candidates were likely to
be judged worthy of a salary increase regardless of the
gender listed on the resume—but for resumes of ambiguously qualified candidates, resumes with male names were
more likely to be viewed positively than those with female
10 Eyesnck et al. studied the interpretation of written sentences as either threatening or nonthreatening by
people who have anxiety and by a control group.
found that unambiguously threatening or nonthreatening sentences were interpreted similarly between groups,
but participants with anxiety were more likely than controls to label ambiguous sentences as threatening. Visual
information is also subject to this phenomenon: Payne et
al. showed participants a series of photos of people holding either guns or ambiguous objects, and participants
were more likely to identify the ambiguous object as a gun
if it was held by a black person.
correlation between all-succeed and innately predisposed.
Those who felt there was an innate predisposition to do well
in CS also felt that not everyone could succeed in the field.
Regression on all-succeed. After finding a one-way
relationship between grade perceptions and the innateness belief, we wanted to see if there was any evidence of a
feedback loop between the two. Because all-succeed and
innately predisposed correlated so highly, we found they
were interchangeable as measures of belief in innate ability. As logistic regression involves only one dependent variable, we had to pick one of the two to use. We chose to do
this analysis with all-succeed because the question item
had been used in another study.
Recall that our study was set up so that a random half
of the participants categorized distributions and then
were asked about innate ability (Treatment 1), whereas
the other half were asked about innate ability and then
categorized the distributions (Treatment 0). If there is a
feedback loop here, we would expect that seeing-bimodality
would predict all-succeed in Treatment 1, but not in
Guidelines for statistical power in logistic regression suggest that an α level of 0.05 requires 10–20 data points per
independent variable in your model.
16 Because this part of
the analysis requires the statistical power to reject a null
hypothesis, we modeled all-succeed as only a function of
seeing-bimodality, and set α = 0.05.
For Treatment 1, we found that seeing-bimodality was a statistically significant predictor of all-succeed. In Treatment 0,
it was not. This indicates that there is a feedback loop
between categorizing distributions as bimodal and agreement with the idea of innate ability. We hence have observed
evidence for the relationships illustrated in Figures 2 and
4. 4. Discussion
With regard to the feedback loop between seeing-bimodality
and all-succeed, we have some weak evidence that categorizing distributions as bimodal increases belief in the
are told that
CS grades are
Figure 2. Individual-level feedback loop leading individuals to
categorize ambiguous distributions as bimodal.
Figure 3. Social-level feedback loops leading individuals to
categorize ambiguous distributions as bimodal.