Journal of Educational Psychology © 2013 American Psychological Association

2014, Vol. 106. No. 1, 105-120 0022-0663/i4/$l2.00 DOI: 10.1037/a0034099

The Role of Anxiety and Working Memory in

Gender Differences in Mathematics

Colleen M. Ganley Marina Vasilyeva

Boston College and University of Illinois at Urbana-Champaign Boston College

This research examined a potential mechanism underlying gender differences in math performance by
testing a mediation model in which women's higher anxiety taxes their working memory resources,
leading to underperformance on a mathematics test. Participants for the 2 studies were college students
{N = 87, Af = 118) who completed an anxiety measure, 2 working memory tasks (verbal and
visuospatial), and a challenging math test including both geometry and algebra items. Findings showed
a significant gender difference in math performance, anxiety, and visuospatial working memory. Further,
there was a mediating chain from gender to the worry component of anxiety to visuospatial working
memory to math performance. The results suggest that women's heightened worry may have utilized
their visuospatial working memory resources, and the resulting gender differences in working memory
were associated with gender differences on a math test. The present research contributes to our
understanding of affective and cognitive factors underlying gender differences in mathematics. The
findings are discussed in terms of their implications for interventions aimed at reducing anxiety and
improving working memory skills.

Keywords: gender differences, mathematics performance, anxiety, working memory, college students

There is a large body of literature investigating gender differ- sources needed for the task, potentially leading to poorer test
ences in math performance, and multiple explanations for these performance. Although there is much research supporting the
differences have been proposed. Some studies have emphasized possibility of this model, this is the first empirical investigation of
the role of affective processes, such as anxiety, whereas others the full mediation model that incorporates botb cognitive and
have focused on cognitive differences that may be associated with affective factors.
a male advantage in mathematics. The present investigation seeks
to examine the joint role of affective and cognitive factors under-
lying gender differences in math performance. Specifically, we test Gender Differences in Math Performance
a mediational model, whereby females have higher anxiety about
Research shows that gender differences in math performance are
a math test, which subsequently utilizes working memory re-
generally small or nonexistent early in development but become
larger and more consistent later on (Hyde, Fennema, & Lamon,
1990; though see Robinson & Lubienski, 2011). In particular, a
This article was published Online First August 19, 2013. recent meta-analysis has indicated that differences favoring males
Colleen M. Ganley, Department of Counseling, Developmental, and appear in high school students and persist into college (Lindberg,
Educational Psychology, Boston College, and Departments of Curriculum Hyde, Petersen, & Linn, 2010). High-school males outscore fe-
and Instruction and Educational Psychology, University of Illinois at males on national achievement tests including the quantitative
Urbana-Champaign: Marina Vasilyeva, Department of Counseling, Devel- portion of the Scholastic Assessment Test (SAT) and the mathe-
opmental, and Educational Psychology, Boston College. matics Advanced Placement (AP) exams, both of which have
This research was supported by Boston College through a dissertation
implications for students' academic futures (College Board, 2009,
fellowship from the Lynch School of Education and an individual research
grant from the Graduate Student Association, both to Colleen M. Ganley,
2010). These differences emerge even though female students take
and by the Institute of Education Sciences, U.S. Department of Education, advanced mathematics courses at a similar rate to male students in
through Grant R305B10O017 to the University of Illinois at Urbana- high school (College Board, 2010). As females progress through
Champaign. The opinions expressed are those of the authors and do not college and graduate school, they tend to drop out of the mathe-
represent views of the Institute of Education Sciences or the U.S. Depart- matics pipeline at higher rates. Although in the last decade the
ment of Education. We thank Martha Makowski, Paul Mathewson, and majority (57%) of college students have been female, they made
Katie-Beth Price for assistance with data collection as well as Beth Casey,
up 44% of college math majors and only 31% of recipients of
Lillie Albert, Elida Laski, Michelle Perry, Alana Dulaney, Sarah Lubien-
ski, and Joe Robinson for helpful feedback on this project. doctoral degrees in mathematics (National Science Foundation,
Correspondence concerning this article should be addressed to Colleen
National Center for Science and Engineering Statistics, 2011).
M. Ganley, who is now at the Department of Psychology and the Leaming Even larger gender gaps are seen later in math-related career
Systems Institute, Florida State University, 1107 West Call Street, Talla- choices in Science, Technology, Engineering, and Mathematics
hassee, FL 32306. E-mail: ganley@psy.fsu.edu (STEM) careers (Snyder & Dillow, 2011).
105
106 GANLBY AND VASILYEVA

In attempting to understand the age-related increase in the members of their gender, they may become familiar with the
observed gender gap, it is important to consider the types of stereotype that women are not as good as men at mathematics,
mathematical tasks that tend to reveal gender differences. Recent which may lead them to develop higher levels of anxiety about
studies have shown that a male advantage is only apparent in their own mathematical abilities.
certain areas of math, whereas females maintain an advantage in Evidence of a negative relation between anxiety and math
other areas. Females perform better than males on computation performance combined with evidence of heightened math anxiety
tasks, as well as other tasks that rely heavily on recall of proce- in women raise the possibility that gender differences in anxiety
dures and information (Gibbs, 2010; Hyde, Fennema, & Lamon, play a role in gender differences in math achievement. To address
1990; Vasilyeva, Casey, Dearing, & Ganley, 2009). However, this issue, two past studies have tested anxiety as a potential
males tend to outperform females in more complex math that mediator of the relation between gender and math performance
involves problem solving, particularly in areas that rely on spatial with mixed results. In considering the discrepancy in these find-
thinking (Becker, 1990; Gibbs, 2010; Hyde, Fennema, & Lamon, ings, it is important to note differences in the studies' methodol-
1990; Lindberg et al., 2010; Vasilyeva et al, 2009). ogies that could be relevant. In particular, Osbome (2001), who
In the context of these findings, Gibbs (2010) suggested that the found evidence for mediation, used a measure of state anxiety
fact that gender differences favoring boys appear to emerge with during math testing, whereas Casey, Nuttall, and Pezaris (1997),
development is due, in part, to the nature of skills required on math who did not find such evidence, examined a measure of trait math
assessments at different ages. Specifically, as students progress anxiety. It has been argued that the relation with test performance
through their schooling, a more pronounced male advantage is is stronger for state, compared to trait, anxiety because the former
observed in association with increased task complexity. Gibbs's directly captures the affective state during testing (e.g., O'Neil &
research suggests that we may underestimate existing gender dif- Fukumura, 1992). In the present studies, we chose to use a measure
ferences by aggregating performance across assessments that re- of state anxiety, as it reflects feelings associated with a particular
quire a wide range of skills. Many curriculum-based standardized testing situation.
assessments do not include items at a higher level of complexity,
which could explain why we do not see gender differences on
Relation Between Anxiety and Working Memory
these types of assessments (Hyde, Lindberg, Linn, Ellis, & Wil-
liams, 2008). Thus, despite the fact that some researchers empha- Given the evidence of the link between anxiety and math per-
size that gender differences in math are small—and perhaps dis- formance, a further question concems the psychological mecha-
appearing—it appears that if more complex items are included in nism underlying this relation. Past research suggests that anxiety
math assessments, the male advantage may be more robust across may impact math performance through its relation with working
ages. memory. Working memory is the part of the memory system that
involves temporarily storing information while simultaneously
manipulating it. According to traditional models, there are two
Explaining Gender Differences distinct types of working memory, verbal and visuospatial, which
in Math Performance are coordinated and controlled by the central executive component
of working memory (Baddeley & Hitch, 1974). Additionally, more
recent models incorporate the concept of the episodic buffer,
Potential Role of Anxiety
where information is integrated across multiple dimensions (i.e.,
Researchers consistently find that math anxiety is negatively verbal and visual; Baddeley, 2000, 2012). Together, the various
related to math performance. This relation has been found both components of working memory provide a mental space for prob-
with measures of trait anxiety, which reflect a general tendency to lem solving.
feel anxious about mathematics (Ganley & Vasilyeva, 2011; Ma, There is extensive research showing that working memory is
1999; Miller & Bichsel, 2004), and measures of state anxiety, related both to anxiety and math performance. First, several studies
which capture anxious feelings experienced in the context of a have documented a relation between anxiety and working memory
math testing situation (Beilock, Rydell, & McConnell, 2007; Bro- functioning (e.g.. Beilock, 2008; Eysenck & Calvo, 1992). In
dish & Devine, 2009; Cadinu, Maass, Rosabianca, & Kiesner, particular, in the context of the processing efficiency theory, it has
2005; Osbome, 2001). Although most of these studies are corre- been theorized that individuals with high anxiety would perform
lational in nature, there is also some research that has experimen- less efficiently on tasks requiring working memory resources
tally tested the effect of interventions designed to alleviate anxiety because their worrisome thoughts interfere with working memory,
on later performance. These studies have generally found that making them unable to fully utilize their working memory capacity
behavioral interventions that reduce anxiety also improve perfor- for task performance (Eysenck & Calvo, 1992; Eysenck, Derak-
mance (in math and other academic areas), suggesting a causal shan, Santos, & Calvo, 2007). The processing efficiency theory
relation between anxiety and performance (Hembree, 1988; posits that anxiety could affect both verbal and visuospatial com-
Ramirez & Beilock, 2011; Wood, 2006). ponents through its effect on the central executive, which is uti-
This relation is particularly important in the context of gender, lized by both types of working memory. Existing research, how-
as women are consistently found to be more anxious about math ever, shows inconsistent results in regard to whether anxiety
tests than men (e.g., Hembree, 1988, 1990; Hyde, Fennema, Ryan, impacts the verbal or visuospatial component, or both. Some
Frost, & Hopp, 1990; Miller & Bichsel, 2004). It has been sug- researchers have found that anxiety exclusively has its impact on
gested that heightened anxiety may reflect the process of gender verbal working memory—presumably because intmsive worry-
socialization (Eccles & Jacobs, 1986). As women are socialized as related thoughts may require processing of verbal information
 ANXIETY, WORKING MEMORY, GENDER, AND MATH 107

(Ikeda, Iwanaga, & Seiwa, 1996; Lee, 1999; Owens, Stevenson,

Norgate, & Hadwin, 2008; Rapee, 1993)—whereas other research- Gender Math Performance
ers have found that anxiety leads to lower performance on visu-
ospatial working memory tasks but not on verbal tasks (Crowe,
Matthews, & Walkenhorst, 2007; Shackman et al., 2006). Due to
the inconsistencies in past research, in the current studies we
examine both types of working memory.
In addition to establishing the link between anxiety and working
memory, researchers have also examined the link between working
memory and academic performance. It has been shown that work-
ing memory plays a critical role in problem solving across aca-
demic domains. This can be expected as most cognitive tasks
require holding multiple pieces of information in one's mind while Figure 1. Proposed mediation model for the relation between gender and
also processing information (Raghubar, Bames, & Hecht, 2010). math test performance.
With respect to mathematics, some studies find evidence that
verbal working memory (Adams & Hitch, 1997; Gathercole, Pick-
ering, Knight, & Stegmann, 2004; Owens et al., 2008), visuospa-
tial working memory (Holmes, Adams, & Hamilton, 2008; Kyttälä two studies testing this model in two different samples to better
& Lehto, 2008), or both (Berg, 2008; Holmes & Adams, 2006; assess the generalizability of findings. In both studies, we chose
Jarvis & Gathercole, 2003; Miller & Bichsel, 2004) are related to college students as participants, because gender differences in
performance. Given the integrative functions of the central exec- math have been found to be particularly robust at this age, and we
utive and episodic buffer, it is not surprising that both visuospatial presented these students with challenging math problems (Lind-
and verbal working memory have been implicated in solving math berg et al, 2010). We first examined whether the math test
problems. The extent to which visuospatial versus verbal process- revealed gender differences in each of the samples. Then we tested
ing is involved in problem solving may vary—it may be dependent the model in which anxiety and working memory create a medi-
on the particular features of the mathematical task. ating chain between gender and math performance.
At present, there has been only one study that examined working In considering potential mediators of the gender difference in
memory as a mediator of the relation between anxiety and math mathematics, we examined both types of working memory (verbal
performance (Owens et al., 2008). These researchers found that and visuospatial), so that the component(s) of working memory
working memory indeed partially explained the relation between that are impacted by anxiety and related to math performance
anxiety and math performance in fifth-grade children. However, could be better understood in the context of gender differences.
gender was not examined as part of this model, and therefore it is The type of anxiety examined in the present studies was the worry
not clear if the same explanatory mechanism can be applied to component of state anxiety. As noted earlier, state anxiety reflects
understanding the nature of gender differences in math perfor- the individual's experience during a testing situation and thus may
mance. be a stronger predictor of performance than trait anxiety (O'Neil &
Fukumura, 1992). The worry component refers to anxious
Anxiety and Working Memory as Mediators of thoughts, such as concems related to performing poorly, as op-
Gender Differences in Math Performance posed to the emotionality component, which refers to the physio-
logical arousal response (Deffenbacher, 1977, 1978; Liebert &
The integration of the bodies of research reviewed above sug- Morris, 1967). We focus on the worry component of anxiety
gests the possibility that anxiety and working memory together because some research suggests that it is more strongly related to
play a role in gender differences in math performance. First, there working memory and test performance than is emotionality (Def-
is substantial research demonstrating gender differences in anxiety, fenbacher, 1980; Hembree, 1988; Kim & Rocklin, 1994; Zeidner
in its many manifestations (Hembree, 1988, 1990; Miller & Bich- & Nevo, 1992).
sel, 2004). Second, there are studies showing that anxiety is related
to working memory (Owens et al., 2008; Rapee, 1993; Shackman
et al., 2006). Finally, there is evidence that anxiety and working Study 1
memory are both related to math performance (Hembree, 1988;
The first study was conducted as part of a larger investigation,
Holmes & Adams, 2006; Jarvis & Gathercole, 2003; Miller &
which also examined effects of stereotype threat on the math
Bichsel, 2004). Combining these findings suggests that female
performance of female students. Thus, half of the students in the
students may be more anxious about a math testing situation, that
sample were tested under stereotype threat conditions, whereas the
their heightened anxiety may interfere with working memory, and
other half were not. We found that performance did not vary as a
this interference may be related to poor performance in math, as
function of the stereotype threat manipulation—this variable did
shown in Figure 1. To our knowledge, the present studies are the
not produce either main effect or interactions. Therefore, in the
first to empirically test this entire model.
present study we included students from both conditions. Because
the main focus of our study was on the relation between working
The Present Research
memory, anxiety, and math performance, and stereotype threat
This research expands on previous work that has examined condition did not interact with any of these variables, we did not
individual pieces of the model depicted in Figure 1. We conduct include it in the analyses reported here. It should be noted that
108 GANLEY AND VASILYEVA

when we tested our models using stereotype threat condition as a Verbal working memory. To assess verbal working memory,
covariate, we found the same pattem of results. we used a word recall task (listening recall) from the Automated
Working Memory Assessment (AWMA; Alloway, Gathercole, &
Method Pickering, 2004). In each trial, the students heard a series of
sentences and judged whether they were tme or false. After they
Participants. Participants were 87 undergraduate students heard 1-6 sentences in a row that they judged tme or false, they
from a private university in the Northeast region of the United were asked to recall the last word of each sentence in the order in
States. The university has been categorized as "most selective" which the sentences were presented. For example, in a trial with
(U.S. News & Worid Report, 2012). Students were recruited from two sentences, the participants heard "Magazines have pages" and
introductory level psychology courses by advertising the study in said true or false, then heard "Apples play football" and said tme
their classes. There were no particular selection restrictions; ev- or false, and then were asked to say the last word of each of the
eryone who signed up for the study was tested. Students received sentences in order ("pages, football"). The task was divided into
extra credit in their course for participation. The sample included six blocks: The first block consisted of trials that included only one
63 female students and 24 male students. This gender distribution sentence, the second had trials with two sentences, and so forth, up
was approximately proportional to that of the students in the to six sentences. Each block had a maximum of six trials with a
courses from which participants were recmited. On average, par- particular number of sentences. Due to the ease of trials within the
ticipants were 19 years, 11 months of age. Sixty-six percent of first three blocks (i.e., with 1, 2, or 3 sentences), students com-
participants took calculus in high school (62% of female students pleted only one trial from each of these blocks. For the remaining
and 75% of male students), and, on average, students had taken hlocks (4, 5, or 6 sentences), the participants completed trials
between one and two math courses in college (M = 1.15; female within the block until they either (a) got 4 trials correct, in which
students M = 0.97, male students M = 1.63).' case they moved on to the next block without completing the last
Materials. Participants completed four measures during the two trials in the current block, or (b) got 3 trials incorrect, in which
testing session: a measure of the worry component of state anxiety, case the participants were finished with the task. Thus, the verbal
two working memory assessments (verbal and visuospatial), and a working memory task measured the ability to remember words
mathematics test. from several sentences while making tme/false judgments about
Worry. Worry was measured with a four-item self-report the meaning of each sentence. The outcome measure was the recall
scale, adapted from Morris, Davis, and Hutchings (1981). Partic- score: the number of times a participant accurately recalled the
ipants rated their feelings on a scale ranging from 1 {strongly words in the correct order. The test-retest reliability for this
disagree) to 4 {strongly agree) for statements about their worry measure is .81 (Alloway, Gathercole, & Pickering, 2006).
(e.g., "I am worried that I may not do well on this test"), two of
Visuospatial working memory. To assess visuospatial working
which were reverse coded (e.g., "I feel very confident about my
memory, we used the spatial recall task, also from the AWMA
performance on the test I'm about to take"). A worry score was
(Alloway et al., 2004). The task followed a parallel stmcture to the
calculated by summing the responses on the four items after
word recall task but with visual, instead of verbal, stimuli. On each
appropriate reverse coding. The intemal consistency for the worry
trial, two shapes were presented next to each other (see Figure 2
scale as measured by Cronbach's alpha was .78.
for sample). The shape on the right was either identical to or a
Working memory. Two dual-task working memory measures
mirror image (opposite) of the shape on the left. Importantly, the
were administered to the participants on a computer: a verbal
shape on the right was rotated relative to the shape on the left by
working memory measure (a word recall task) and a visuospatial
0, 120, or 240 degrees; a red dot marked the top of the shape to
working memory measure (a spatial recall task).
indicate the degree of rotation. The task was divided into seven
blocks: The first block consisted of trials that included only one
pair of shapes; the second had trials with two pairs of shapes, and
TrMI Trial 2 so forth, up to seven pairs of shapes. Within a trial, participants had
to determine if the shape on the right was the same or opposite of
the shape on the left. After making the same-opposite judgments
a ^ for all pairs of shapes within a trial, participants had to point to one
of three dots (at 0, 120, and 240 degrees) that matched the dot
location for the shape on the right in the order in which the pairs
of shapes were presented. Thus, the visuospatial working memory
task measured the participants' ability to remember the location of
a dot across trials while making same/opposite judgments about
RectfZ
shapes. The administration procedure for the trial blocks was
parallel to that in the verbal working memory task (i.e., students
completed one trial for 1, 2, and 3 pairs of shapes and up to 4 trials
for 4, 5, 6, and 7 pairs). Participants' recall scores, which were
based on their correct recall of the dot locations, were used as the
Figure 2. Sample visuospatial working memory task tdal. From Auto-
mated Working Memory Assessment by T. P. AUoway, S. E. Gathercole,
and S. Pickering, 2004. Reprinted with permission. Copyright by Pearson ' Analyses were also run with the number of college courses that
Education. students had taken as a covariate, which did not change the results.
 ANXIETY, WORKING MEMORY, GENDER, AND MATH 109
Table 1
Study 1: Descriptive Statistics: Means, Standard Deviations, and Effect Sizes for
Gender Differences

Female students Male students

Measure M SD M SD Cohen's d
Worry scale 10.10 2.09 7.63 1.88 -1.25"
Verbal working memory 22.63 4.16 23.21 3.99 0.14
Visuospatial working memory 25.83 6.39 32.79 5.41 1.18**
Math score (%) 40.67 18.99 56.71 21.62 0.79**
Note. The scale for worry is from 4 to 16; the scale for verbal working memory is from 0 to 36; the scale for
visuospatial working memory is from 0 to 42; the scale for math score is percent accuracy, ranging from 0 to
100.
•> < .01.

outcome measure for this task. The test-retest reliability for this Results
measure is .82 (AUoway et al., 2006).
Math performance. Participants solved 12 math problems Preliminary and descriptive analyses. Prior to conducting
(see the Appendix) presented in a multiple-choice format (six the main analysis of interest, we ran multiple sets of preliminary
algebra items, six geometry/measurement items). These test items analyses. In one set we examined whether the order of task
were selected from several widely-used standardized tests: Na- administration affected performance. We found that the order of
tional Assessment of Educational Progress (NAEP) for 12th grade, administering the two types of working memory tasks and the
the Massachusetts Comprehensive Assessment System (MCAS) order of administering the two types of math problems did not
for 10th grade, and the SAT. The items were selected by a panel affect the outcome variable, students' math scores (p = .47 and
of experis in math and math education based on the coverage of .94, respectively), and therefore the data were collapsed across the
mathematical content. Another criterion for item selection was different orders. Then we examined whether performance varied
difficulty level. Because gender differences are often found on across the two types of math items (algebra and geometry/mea-
more difficult problems (Gibbs, 2010; Lindberg et al., 2010), we surement). The algebra items (M = 42%) were slightly more
chose items that showed 30%-40% accuracy rates in national difficult than the geometry/measurement items {M = 47%), but
samples of high school seniors. In a separate study, we obtained this difference was not statistically significant (p = .11). There-
evidence for the criterion-related validity of the math test by fore, in the main analysis we relied on the overall math test score
showing a high correlation between accuracy scores obtained on across content areas. Finally, as indicated earlier, a preliminary
this test and the scores on the math portion of the state- analysis showed that the stereotype threat manipulation did not
administered standardized test, MCAS (r = .63, p < .001). The produce either a main effect or any interactions, and thus the data
internal consistency for the mathematics test as measured by from both conditions were combined in furiher analysis.
Cronbach's alpha was .61. Pariicipants completed the six algebra It should be noted that when conducting analyses of variance
problems in one block and the six geometry/measurement prob- (ANOVAs), unequal sample size can be problematic (as was the
lems in another, and they were given 6 min for each block. Math case for the gender distribution in the present study). However, it
scores were calculated as the percent of items answered correctly is only an issue if there is heterogeneity of variance (Keppel &
(out of the total number of items attempted).^ Wickens, 2004). Thus, we conducted Levene's test of homogene-
Procedure. Participants were tested individually; each student ity of variance, which was nonsignificant across all outcomes
received a booklet that included the worry scale and the math test. (.28 < ps < .93). This indicated that the unequal sample size does
Prior to conducting the study, the purpose was explained to the not present a serious problem in our subsequent analyses because
students as follows, "The goal of this study is to understand the we found no evidence for heterogeneity of variance.
relationship between what students think while doing math, how Descriptive statistics by gender are presented in Table 1. As
they do in math, and their memory skills." At the beginning of shown in the table, significant gender differences were found in
testing, students were told, "The main thing we are going to do is worry, visuospatial working memory, and math performance
a math test. The math test is made up of 12 word problems that are (.79 < Irfsl < 1.25) but not in verbal working memory. Table 2
multiple-choice. Also you will be answering some questions about presents correlations between the variables. These correlations
your thoughts and feelings and doing some memory tasks." Prior show that worry, visuospatial working memory, and math perfor-
to completing test measures, participants were given a practice mance were all significantly correlated with one another. Verbal
math problem, which included a stereotype threat manipulation working memory was related to visuospatial working memory and
(the content of the problem either activated the gender stereotype worry but not to math performance. This nonsignificant relation, in
or did not). Immediately after the practice math problem, they combination with the lack of a gender difference in verbal working
completed the worry scale. The memory tasks were administered memory, informed the decision to exclude verbal working memory
next, with the order of the verbal and visuospatial measures coun-
terbalanced. Finally, students completed the math test, with the
order of the algebra and geometry/measurement blocks counter- ^ The same pattern of results was obtained when analyses were run with
balanced. the percent correct out of all items as well as when the scores were
corrected for guessing (Frary, 1988).
lio GANLEY AND VASILYEVA

Table 2
Study I: Correlations Among Measures
Worry Verbal Visuospatial
Measure Gender scale working memory working memory

Worry scale -.48"

Verbal working memory .06 -.25*
Visuospatial working memory .46" -.41" .33*
Math score .35" -.47" .19 .51"
"p < .05. < .01.

as a potential mediator in the model. Thus, only visuospatial {b2). Mediation is likely evident if the relation between gender
working memory was used in the subsequent mediation analysis. and math test score (c) disappears or significantly decreases
Analytic strategy for mediation analysis. After we estab- after the addition of worry and visuospatial working memory
lished gender difference in students' math test scores, the next step into the model (c').
was to test whether worry and working memory served as a The indirect effects were tested by calculating the product of the
potential mechanism through which gender was linked to math test path coefficients (henceforth referred to as a point estimate) be-
performance. In statistical terms, this pattern is captured in a model tween the predictor and mediator(s) and between the mediator(s)
in which worry and working memory mediate the relation between and the outcome. The three specific indirect effects were (1) the
gender and math performance (Hayes, Preacher, & Myers, 2011). indirect effect of gender on math test score through worry only
To test for mediation, we ran a series of regression models.'' The (a, X b¡), (2) the indirect effect of gender on math test score
first two regression analyses tested whether the predictor, gender through visuospatial working memory only (ÛJ X èj), and (3) the
(females = 0, males = 1), was related to the outcome measure, indirect effect of gender on math test score through worry and then
math test score (Equation 1), and the first potential mediator, visuospatial working memory (a, X «33 X ¿2). The test of the total
worry (Equation 2). Next, we ran a regression model that included indirect effect is the sum of the three specific indirect effects (o, X
gender and worry as predictors of visuospatial working memory ¿1 -t- 02 X ¿>2 -I- a, X ^3 X ¿72)- This is equivalent to testing the
(Equation 3). The last regression analysis tested whether math test difference between the total effect of gender on math performance
performance was predicted from gender, worry, and visuospatial (c) and the direct effect of gender on math test performance (c')-
working memory (Equation 4). Note that the total effect, c, captures the relation between gender
and math test performance without taking any other variables into
Math Test Score = ßo: + c(Gender) -He, (1) account, whereas the direct effect, c', captures the remaining
relation between gender and math test score after taking into
Worry = ai(Gender) -H (2)
account the contributions of worry and working memory to math
Visuospatial Working Memory = ßo3 + a test performance. The main interest of the present study is the third
indirect effect: the path from gender, to worry, to visuospatial
+ «3 (3) working memory, to math test score.
To estimate confidence intervals for the point estimate of the
Math Test Score = ßo4 + c / (Gender) -f è, (Worry)
mediated effect, we used bias-corrected bootstrapping (Dearing &
-H i)2(Visuospatial Working Memory) -I- e^ (4) Hamilton, 2006; Hayes et al., 2011 ; Preacher & Hayes, 2008). This
method was recommended by Taylor, MacKinnon, and Tein
Figure 1 depicts the model and the coefficients that represent (2008), who conducted a Monte Carlo study comparing the Type
each effect from these four analyses. Critical to the present anal- I error, power, and coverage of six different methods for estimat-
ysis is the combined path from gender to worry (a,), worry to ing this type of mediation model (i.e., a three-path model). Boot-
working memory (03), and working memory to math test score strapping has an advantage over other tests of the indirect effect
(e.g., the Sobel test) in that it does not assume that the sampling
distribution of the indirect effect is normal (Hayes et al., 2011).
c=16.O5{5.11)*' The procedure, instead, involves resampling to estimate the char-
Gender Math Test Score
acteristics of the sampling distribution and uses this information to
create confidence intervals around the point estimate. Specifically,
the bootstrapping strategy used in the present study involved
selecting 50,000 bootstrap samples with replacement from the

' Another way to test the paths that we were interested in is to use
structural equation modeling with maximum likelihood estimation. It
should be noted that this approach has some advantages, but ordinary least
squares (OLS) regression is better here because it is more difficult to
establish whether one has satisfied the strict assumptions necessary for
using maximum likelihood estimation when the sample size is small (Ding,
Figure 3. Study 1 mediation analysis results. *p < .05. " p < .01. Velicer, & Harlow, 1995).
 ANXIETY, WORKING MEMORY, GENDER, AND MATH 111

Table 3
Study 1: Magnitude of Indirect Effects

95% bias-corrected confidence

interval
Effect type Formula Point estimate Lower limit Upper limit
Total indirect effect a, X i , -H a^ + a¡ X 14.74 9.40 21.87
Indirect effect through worry only a, x b¡ 6.86 2.34 13.22
Indirect effect through visuospatial
working memory only 02 x ¿^ 5.81 1.82 12.45
Indirect effect through worry, then
visuospatial working memory a¡ X a^ X b2 2.07 0.51 5.26
Note. If the confidence interval does not overlap with 0, it indicates a significant indirect effect.

current sample and then calculating the point estimates for the Study 2
indirect effects within each of these samples (Preacher & Hayes,
2008)."* Then, the distribution of the point estimates obtained in Study 2 addressed some of the potential weaknesses of Study 1.
these samples was used to create 95% confidence intervals based First, we recmited a larger sample of students (118 vs. 87) that had
on the sampling distribution of these estimates (Hayes et al., 2011). a more balanced distribution of males and females (56 male
The mediation analysis was mn with the PROCESS SPSS macro students, 62 female students). Second, we recmited students from
created by Hayes et al. (2011), which uses bootstrapping and a university that varied in a number of ways from that in Study 1 ;
reports bias-corrected confidence intervals. Using this method, the It was a less selective public university with a more diverse student
total indirect effect as well as the three specific indirect effects body and was located in a different region of the United States.
were examined. An indirect effect was considered significant if its Therefore, we were able to determine whether the results obtained
95% confidence interval did not overlap with zero (Preacher & in Study 1 would generalize to another group of college students.
Hayes, 2004). Third, we did not use a stereotype threat manipulation in this
Estimation of the mediation model. Results of the mediation study, and thus any potential infiuence of stereotype threat was
analysis are presented in Figure 3. In the first step of the analysis, eliminated.
we ran the regression models described in Equations 1 and 2. The
results showed that gender was significantly related to both the
outcome measure—math test score (Equation 1; c = 16.05, 5,. = Method
5.11, p = .002)—and the potential first mediator—worry (Equa-
tion 2; a, = -2.47, .s„, = 0.47, p < .0001). The next regression Participants. Participants were 118 undergraduate students
analysis (Equation 3) included gender and worry as predictors of from a public university in the Midwest region of the United
visuospatial working memory performance. The findings revealed States. The university has been categorized as one of the "more
that both gender (Oj = 5.13, i^j = l.51,p= .002) and worry (a, = selective" national universities (U.S. News & World Report,
—0.74, s„3 = 0.33, p = .03) were significant predictors of visu- 2012); its acceptance rate is about 2.5 times higher than the
ospatial working memory. The final regression analysis predicted university where participants were recmited for Study 1. Partici-
math performance from gender, worry, and visuospatial working pants were recmited from introductory level psychology courses
memory performance. The results showed that gender was no by advertising the study in their classes. There were no particular
longer a significant predictor of math test performance (c' = 1.31, selection restrictions; everyone who signed up for the study was
V = 5.09,p = .80), but both worry (fc, = -2.78, 5^; = I.ll,p = tested. Students received either subject pool credit, extra credit in
.014) and visuospatial working memory (¿2 = 1.13, s^2 — 0-31- their course, or $5 for participation. The sample included 62
p = .0005) were significant predictors. female students and 56 male students. Participants were on average
As a final step in the analysis, we estimated confidence intervals 20 years, 7 months of age. Sixty-two percent of participants took
for the different mediation effects. Specifically, we used bias- a calculus course in high school (57% of female students and 68%
corrected bootstrapping to calculate confidence intervals—both of male students), and, on average, students had taken between one
for the entire model and for each of the three specific indirect and two math courses during college (M = 1.79; female students
effects (through worry only, visuospatial working memory only, M = 1.52, male students M = 2.09).
and both worry and visuospatial working memory). As seen in
Table 3, the total indirect effect was significant. Critical to our
research question, the indirect effect of gender on math test score
" It is important to note that the bootstrapping algorithm will give
through both worry and visuospatial working memory (a^ X a^X somewhat different values for the confidence intervals around the point
¿2) was significant with a point estimate of 2.07 (95% CI [0.51, estimate each time the bootstrapping is done because it is using different
5.26]). Thus, there was evidence of a mediating chain from gender samples within the larger sample to compute the point estimates. Despite
to math test performance through worry and then visuospatial this issue, this method still offers multiple advantages over other existing
methods.
working memory.^ The indirect effect through worry and then
' We compared the current model that includes both mediators with
visuospatial working memory (2.07% on the math test) is 12.92% models that include each mediator separately and found that the model with
of the total effect (16.05% on the math test). both mediators explained a larger proportion of the total effect.
112 GANLEY AND VASILYEVA

Table 4
Study 2: Descriptive Statistics: Means, Standard Deviations, and Effect Sizes for
Gender Differences

Female students Male students

Measure M SD M SD Cohen's d

Worry scale 9.04 1.98 7.39 2.2! -0.79"

Verbal working memory 22.42 7.38 22.20 4.38 -0.04
Visuospatial working memory 26.53 6.40 31.66 6.59 0.79"
Math score (%) 36.29 20.62 56.76 22.69 0.94"
Note. The scale for worry is from 4 to 16; the scale for verbal working memory is from 0 to 36; the scale for
visuospatial working memory is from 0 to 42; the scale for math score is percent accuracy, ranging from 0 to
100.

Materials. Participants completed the same four measures Estimation of the mediation model. We conducted a medi-
used in Study 1: a worry scale (a = .84), two working memory ation analysis parallel to that in Study 1—testing the indirect path
tasks (verbal and visuospatial), and a mathematics test (a = .70). from gender to worry to visuospatial working memory to math test
Procedure. The procedure was identical to that in Study 1 performance (see Figure 4). In the first step of the analysis, we ran
except that this study did not include a stereotype threat manipu- the regression models described in Equations 1 and 2 above. The
lation. All other measures were administered in the same way. results showed that gender was significantly related to both the
outcome measure—math test score (Equation 1 : c = 20.47, s^ =
Results 4.04, p < .0001)—and the potential first mediator—worry (Equa-
tion 2: a, = - 1.65, s^¡ = 0.39, p = .0001). The next regression
Preliminary and descriptive analyses. Preliminary analyses analysis (Equation 3) included gender and worry as predictors of
showed that the order of administering the two types of working visuospatial working memory performance. The findings revealed
memory tasks and the order of administering the two types of math that both gender (oj = 3.64, .5^2 = 1.33,/? = .007) and worry (03 =
problems had no effect on students' math test scores (p = .94 and
-0.90, .s^3 = 0.27, p = .001) were significant predictors of
.53, respectively). Thus, the data were collapsed across different
visuospatial working memory performance. The final regression
orders. We also found that the algebra items (M = 44%) were
analysis predicted math test scores from gender, worry, and visu-
slightly more difficult than the geometry/measurement items (M =
ospatial working memory performance. Results showed that al-
48%), but this difference was not statistically significant (p = .08),
though the coefficient for gender decreased from 20.47 to 13.67, it
and thus, parallel to Study 1, we used the overall math test score
was still a significant predictor of math test performance (c' =
as the dependent variable in main analyses.
13.67, s^. = 4.36,/? = .002), as was visuospatial working memory
Descriptive statistics by gender are presented in Table 4. As
(¿2 = 0.86, ÎM = 0.31, p = .007), but worry (¿, = -1.46, i^,, =
shown in the table, significant gender differences were found in
0.88, p = .10) was not a significant predictor.
worry, visuospatial working memory, and math performance
(0.79 < Wsl < 0.94), but not in verbal working memory. Corre- As a final step in the analysis, we estimated bias-corrected
lations among the variables are presented in Table 5. These cor- bootstrap confidence intervals for the mediation effects. The re-
relations show that worry, visuospatial working memory, and math sults are presented in Table 6, and as can be seen, the total indirect
performance were all significantly related to one another. Verbal effect was significant. The key finding is that the indirect effect of
working memory was related to visuospatial working memory, but gender on math test score through both worry and visuospatial
it was not correlated with either worry or math performance and working memory (ay y. a-¡ y~ b-^ was significant with a point
was therefore excluded from mediation analysis. In sum, descrip- estimate of 1.28 (95% CI [0.35, 3.33]). Thus, there was again
tive analyses—both in terms of gender differences on the variables evidence of a mediating chain from gender to math test perfor-
of interest and correlations between these variables—showed sim- mance through worry and then visuospatial working memory. The
ilar pattems to those in Study 1. indirect effect through worry and then visuospatial working mem-

Table 5
Study 2: Correlations Among Measures
Worry Verbal Visuospatial
Measure Gender scale working memory working memory

Worry scale -.37"

Verbal working memory -.02 .01
Visuospatial working memory .37" -.39" .19*
Math score .43" -.34" -.06 .41"
*p<.05. "p<.0\.
 ANXIETY, WORKING MEMORY, GENDER, AND MATH 113

c = 20.47 (4.04)*' math achievement (Hyde et al., 2008; Lindberg et al., 2010). Some
Gender Math Test Score
evidence suggests that these differences have been decreasing over
time. In particular, when looking at grades or standardized test
performance in school students, researchers often find comparable
levels of performance for males and females, and sometimes a
female advantage, especially when measured in terms of grades
(e.g., Pomerantz, Altermatt, & Saxon, 2002). The findings of a
male advantage appear to be more persistent at the college level,
where the evidence consistently shows a small but significant
gender difference (Hyde, Fennema, & Lamon, 1990; Lindberg
Gender Math Test Score
et al., 2010).
Further, it has been argued that math problems that are more
Figure 4. Study 2 mediation analysis results. " p < .01. complex tend to show larger differences favoring males (Gibbs,
2010). Thus, in the present studies we utilized a test instrument
ory (1.28% on the math test) is 6.24% of the total effect (20.47% that included items proven to be challenging in earlier large-scale
on the math test). assessments, to maximize our chances of finding a gender differ-
ence. Our results showed that the mean scores for both genders
Discussion were quite low, indicating that this was indeed a difficult assess-
ment for our participants (on average, female students = 38%
The main goal of this work was to examine the cognitive and correct, male students = 57% correct). More importantly, there
affective factors related to gender differences in math perfor- was a robust gender difference (Study 1 d = 0.74; Study 2 d =
mance. This investigation grew from research on gender social- 0.94). The magnitude of this difference was substantially larger
ization, which suggests that as women are socialized, they are than in research using less difficult assessments (Hyde, Fennema,
exposed to gender stereotypes, and this may lead them to be & Lamon, 1990; Lindberg et al, 2010). Note that the ability to
anxious about their ability to do well in mathematics (Eccles & solve challenging math problems may be particularly predictive of
Jacobs, 1986). Past research suggested that heightened anxiety an individual's desire to persist in seeking a career in a math-
may have negative consequences for cognitive processing, which
intensive field. The present findings, consistent with Gibbs's
in tum, could lead to poor performance on math tests. Both lower
(2010) arguments, highlight the importance of considering the
levels of performance and the anxiety itself may deter female
nature of the math task, and particularly, the level of difficulty, in
students from choosing educational paths leading to STEM ca-
discussing the existence and magnitude of gender differences.
reers. To develop educational programs aimed at reducing the
gender gap in math-intensive fields, it is critical to better under-
stand whether the affective and cognitive factors suggested in the Mediators of the Relation Between Gender
literature as potentially underlying gender differences in math
and Math Performance
achievement form a mediating chain from gender to math perfor-
mance. The present research provided the first empirical investi- The key novel finding of the present work is the mediating
gation of which we are aware testing the entire mediational rela- pathway from gender to math performance created by anxiety and
tion from gender differences in math anxiety to working memory working memory. These results suggest that one possible reason
to math performance. gender differences in math performance might exist is because of
increased worry in female students, which taxes their visuospatial
working memory, which, in tum, leads them to perform more
Gender Differences in Math Performance
poorly on difficult math assessments. We should he clear: The
Recently, there has been a discussion in the literature concerning finding of a mediated relation between variables does not provide
the magnitude and even the very existence of gender differences in evidence of a causal link between them. Yet, it provides a critical

Table 6
Study 2: Magnitude of Indirect Effects

95% bias-corrected
confidence interval
Lower Upper
Effect type Formula Point estimate limit limit
Total indirect effect a, X fc, + a^ a, X 6.80 3.27 11.82
Indirect effect through worry only "l X 2.40 -0.13 5.71
Indirect effect through visuospatial
working memory only X 3.13 0.75 7.25
Indirect effect through worry, then
visuospatial working memory a. X a, X è. 1.28 0.35 3.33
Note. If the confidence interval does not overlap with 0, it indicates a significant indirect effect.
114 GANLEY AND VASILYEVA

piece of evidence for the possibility of such a link, which can be ality) are critical dimensions to consider in examining the relation
further addressed in experimental work. between anxiety and math performance.
The present results generally fit with prior work that has shown Working memory and its relation to worry and math
evidence for the individual pathways involved in the model. Here, performance. Prior research has allowed us to identify working
we combined pathways associated with cognitive and affective memory as a potential mediator of the relation between anxiety
predictors and found that together they explain a significant pari of and math performance. There are multiple findings pointing to the
the relation between gender and math performance. In fact, the link between anxiety and working memory (e.g., Ikeda et al., 1996;
model that includes both mediators is stronger (i.e., explains a Lee, 1999; Owens et al., 2008; Shackman et al, 2006), and
larger poriion of the total effect) than models that include each of between working memory and academic performance (e.g., Adams
the mediators separately. Furiher, our studies allowed us to take a & Hitch, 1997; Berg, 2008; Holmes et al., 2008; Owens et al.,
more in-depth look into specific affective and cognitive factors 2008). The findings, however, have been somewhat inconsistent in
that are associated with gender differences in math performance. implicating verbal versus visuospatial working memory, or both,
in math performance. The present research contributed to the
The role of worry. In these studies, the worry component of literature by directly examining both types of working memory in
anxiety, as measured before a math test, was found to be involved in relation to gender, anxiety, and math performance. Consistent with
the mediational path from gender to math performance. The present previous research, we found that visuospatial and verbal compo-
analyses confirmed the bivariate relations documented in prior re- nents of working memory were moderately correlated with each
search, namely, relations between gender and anxiety (Hembree, other (Gilhooly, Wyn, Phillips, Logie, & Delia Sala, 2002; St
1988, 1990; Miller & Bichsel, 2004), anxiety and working memory Clair-Thompson & Gathercole, 2006). We found mixed results in
(Ikeda et al., 1996; Lee, 1999; Owens et al., 2008; Shackman et al., terms of the relations between the two types of working memory
2006), and anxiety and math performance (Ganley & Vasilyeva, and worry. In both studies, visuospatial working memory was
2011; Miller & Bichsel, 2004; Osbome, 2001). Two previous studies related to worry; however, verbal working memory was correlated
examined math anxiety in relation to gender differences in math with worry in Study 1 but not in Study 2. The results of Study 1
performance with mixed results. Osbome's (2001) findings supported fit with the processing efficiency theory (Eysenck & Calvo, 1992),
the mediational role of anxiety, whereas Casey et al.'s (1997) findings which suggests that anxiety affects the central executive compo-
did not. To understand potential sources of inconsistency between the nent of working memory that is responsible for processing infor-
findings, we take a closer look at the specific dimensions of anxiety mation stored in verbal and visuospatial memory. However, the
measured in these studies. results of Study 2 suggest that perhaps there is something more
The first potentially relevant distinction is between trait and state specific about visuospatial working memory that worry interferes
anxiety. The study by Casey et al. (1997) utilized a measure of trait with. Future research should furiher examine these relations.
math anxiety (i.e., a general tendency to feel anxious about math The pattern of relations between working memory and the other
tests), whereas the present studies as well as Osbome's (2001) study two variables examined in the present studies—gender and math
explicitly tested state anxiety experienced during the current testing performance—was different for visuospatial and verbal measures. In
situation. Of course, the two types of anxiety are not completely particular, visuospatial working memory was strongly related to both
independent: Individuals with higher trait anxiety are more hkely to gender and math performance, whereas verbal working memory was
have higher state anxiety (Hembree, 1990; Miller & Bichsel, 2004). not significantly related to either gender or math test performance.
At the same time, these two types of anxiety are viewed as separate Thus, verbal working memory could not act as a mediator, given a
constructs. It has been suggested that due to its proximity to the testing nonsignificant relation with both the predictor and outcome.
situation, state anxiety may show a stronger relation to performance The lack of involvement in the mediation model for verbal
than trait anxiety (e.g., O'Neil & Fukumura, 1992). This may, in part, working memory is somewhat surprising given other studies that
explain why studies looking at trait anxiety arrive at different findings have implicated this component of working memory as a mediator
than the studies testing state anxiety. In the current studies, state of the relation between anxiety and math performance (Owens et
anxiety in some students may have been particularly heightened by al., 2008). Based on current conceptualizations of working mem-
their knowledge that they would have a rather limited amount of time ory, which emphasize the integrative role of the episodic buffer,
to take the math test. one could expect a mixture of visuospatial and verbal working
memory effects on math performance. Yet, it is imporiant to note
Another potentially relevant distinction is between the two that the present studies, in contrast to the study done by Owens et
components of anxiety—worry (anxious thoughts) and emotional- al. (2008), focused specifically on the role of working memory in
ity (affective arousal; e.g., feeling tense or nervous/jittery). These the context of gender differences. There have been numerous
components of anxiety have been found to differentially relate to studies revealing a male advantage in visuospatial abilities (e.g.,
performance (Deffenbacher, 1980; Hembree, 1988; Kim & Rock- Voyer, Voyer, & Bryden, 1995), thus the inclusion of gender in the
lin, 1994; Zeidner & Nevo, 1992). Worry, in particular, appears to model may explain the strength of visuospatial working memory
be strongly linked to working memory, as monitoring anxious as a predictor of the observed differences in math performance.
thoughts utilizes a substantial amount of working memory re-
sources. In the study by Casey et al. (1997), the anxiety scale
combined worry and emotionality items. It is possible that the Limitations and Future Directions
relation between worry and math performance was diluted by One limitation of the present studies periains to the measures used.
combining it with emotionality. Thus, the type of anxiety (trait vs. With regard to the math assessment, the test consisted of only 12
state) and the component of anxiety measured (worry vs. emotion- items (6 in each content area) due to constraints on how much time we
 ANXIETY, WORKING MEMORY, GENDER, AND MATH 115
could request from undergraduate students for test taking. We did not sources of individual difference within groups. After all, there are
find differences between algebra and geometry/measurement items, substantial individual differences within each gender that need to
and it is possible that the small number of problems within each be better understood. Furthermore it may be useful to examine the
content area made it difficult to find such differences. A more exten- strength of the predictive relations for males compared to females.
sive test with a more comprehensive representation of each domain For example, the relation between worry and math performance
would be useful in comparing gender effects and the mediational and the extent to which this relation is mediated by working
model across mathematical content. memory may be different for the two gender groups.
With regard to the measure of anxiety, we used a worry scale that
was based on self-report. Although this is a common approach to Educational Implications
assessing affective states, it may have introduced a confound—it is
possible that male students were less likely to report worry as it would The findings of the present studies not only lead to important
not be seen as acceptable. However, there is evidence suggesting that suggestions for future research but also have potential implications for
the gender difference in worry is unlikely to be due to males' general educational practice. One of the key findings in this respect is the
tendency to downplay their worry. In particular, in other subjects (e.g., evidence of large gender differences on a difficult mathematics test
English) male students report lower levels of confidence than do among college students. This finding adds to the recent evidence
female students (Correll, 2001). This may refiect cultural expectations (Gibbs, 2010), suggesting that, despite improvements in female stu-
about the strengths and weaknesses of certain groups in certain dents' performance on standardized test and classroom-based assess-
academic areas. Using multiple measures of anxiety and not just ments, the issue of gender differences in mathematics has not been
self-report may allow one to get a more complete picture of a person's completely resolved. Thus, it is important for policy makers and
emotional state during problem solving. educators to continue their efforts to improve the math achievement of
With respect to the overall method, as we mentioned earlier, the female students and incorporate research evidence conceming factors
mediation analysis is correlational in nattire, and thus causal relations related to gender differences in mathematics achievement.
cannot be established definitively. Most likely, some relations re- In this context, the other key finding of the present studies—
ported here are bidirectional in nature (e.g., the relation between conceming the role of anxiety and working memory as mediators
worry and math performance). Conceptually, mediation offers a po- of gender difference in math performance—appears to be partic-
tential mechanism underlying the relation between the predictor and ularly relevant. The results raise the possibility that efforts to
outcome (Dearing & Hamilton, 2006). Establishing this type of rela- ameliorate the problem of gender differences in math-related fields
tion is critical to the investigation of gender differences in math may not be adequate unless they target specific factors, such as
performance. Yet, the strongest test of the relations uncovered in the worry about math, in girls and women. In fact, recent findings
present studies would be provided by an experimental investigation of suggest that heightened math anxiety in female teachers at the
whether changes in anxiety and working memory could lead to beginning of the school year is associated with lower math per-
changes in math outcomes. formance over a school year in their female students (Beilock,
We cannot make causal claims based on the present studies; bow- Gunderson, Ramirez, & Levine, 2010). Thus, addressing the prob-
ever, there have been a few studies exploring the causal relation lem of math anxiety in females may have far-reaching implica-
between test anxiety and performance, although not focusing specif- tions. Other recent findings indicate that female students' perfor-
ically on mathematics (Hembree, 1988; Wood, 2006). In a meta- mance can be improved through interventions targeting specific
analysis of intervention studies, Hembree (1988) found that anxiety cognitive skills, including visuospatial reasoning (Terlecki, New-
during testing can be reduced through behavioral and cognitive- combe, & Little, 2008; Wright, Thompson, Ganis, Newcombe, &
behavioral interventions and that this diminished anxiety is followed Kosslyn, 2008). The converging evidence from the present studies
by an improvement in test performance. Further, a number of studies and other recent research highlights the importance of considering
bave provided experimental evidence tbat cognitive interventions can both cognitive and affective processes in targeting gender differ-
lead to improvements in working memory (Klingberg, Forssberg, & ences in educational settings.
Westerberg, 2002; Thorell, Lindquist, Nutley, Gunilla, & Klingberg,
2009) and that there are subsequent gains in matb performance References
(Holmes, Gathercole, & Dunning, 2009). Although these studies
Adams, J., & Hitch, G. (1997). Working memory and children's addition.
provide important evidence of malleability of the factors related to Journal of Experimental Child Psychology, 67, 21-38. doi:10.1006/jecp
gender differences in mathematics, none of them have yet looked at .1997.2397
the joint role of anxiety and working memory. In future work, it Alloway, T. P., Gathercole, S. E., & Pickering. S. (2004). Automated
would be useful to design intervention studies aimed at examining Working Memory Assessment. Test battery available from authors at
whether the mediational chain uncovered in the present research can http://www.pearsoncIinical.co.uk/Psychology/ChildCognitionNeuropsy
be reproduced through experimental manipulations. That is, would the chologyandLanguage/ChildMemory/AutomatedWorkingMemory
interventions designed to alleviate math anxiety increase the working Assessment(AWMA)/AutomatedWorkingMemoryAssessment(AWMA)
memory of female students during testing, and would this increase .aspx
lead to an improvement in math performance? Researchers' ability to Alloway, T. P., Gathercole, S. E., & Pickering, S. J. (2006). Verbal and
visuo-spatial short-term and working memory in children: Are they
answer such questions could have important consequences for wom-
separable? Child Development, 77, 1698-1716. doi:I0.111 l/j.1467-
en's achievement in math-intensive fields. 8624.2006.00968.x
Finally, it would be important to explore the mechanisms ad- Baddeley, A.D. (2000). The episodic buffer: A new component of working
dressed in the present studies not only with the goal of explaining memory? Trends in Cognitive Sciences, 4, 417-423. doi:10.1016/
between-group gender differences but also to identify potential S1364-6613(00)01538-2
 GANLEY AND VASILYEVA
116
Baddeley, A. (2012). Working memory; Theories, models, and controver- Ding, L., Velicer, W. F., & Harlow, L. L. (1995). Effects of estimation
sies. Annual Review of Psychology, 63, 1-29. doi; 10.1146/annurev- methods, number of indicators per factor, and improper solutions on
psych-120710-100422 structural equation modeling fit indices. Structural Equation Modeling:
Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. H. Bower A Multidisciplinary Journal, 2, 119-143. doi; 10.1080/
(Ed.), The psychology of leaming and motivation: Advances in research 10705519509540000
and theory (Vol. 8, pp. 47-89). New York, NY; Academic Press. Eccles, J. S., & Jacobs, J. E. (1986). Social forces shape math attitudes and
Becker, B. (1990). Item characteristics and gender differences on the performance. Signs, 11, 367-380. doi; 10.1086/494229
SAT-M for mathematically able youths. American Educational Re- Eysenck, M. W., & Calvo, M. G. (1992). Anxiety and performance; The
search Joumal, 27, 65-87. doi;10.3102/00028312027001065 processing efficiency theory. Cognition & Emotion, 6, 409-434. doi;
Beilock, S. L. (2008). Math performance in stressful situations. Current 10.1080/02699939208409696
Directions in Psychological Science, 17, 339-343. doi;10.1111/j.l467- Eysenck, M. W., Derakshan, N., Santos, R., & Calvo, M. G. (2007).
8721.2008.00602.x Anxiety and cognitive performance; Attentional control theory. Emo-
tion, 7, 336-353. doi: 10.1037/1528-3542.7.2.336
Beilock, S. L., Gunderson, E. A., Ramirez, G., & Levine, S. C. (2010).
Frary, R. B. (1988). Formula scoring of multiple-choice tests (correction
Female teachers' math anxiety affects girls' math achievement. Proceed-
for guessing). Educational Measurement: Issues and Practice, 7, 33-38.
ings of the National Academy of Sciences, USA, 107, 1860-1863.
doi;10.1111/j.l745-3992.1988.tb00434.x
doi;10.1073/pnas.0910967107
Ganley, C. M., & Vasilyeva, M. (2011). Sex differences in the relation
Beilock, S. L., Rydell, R. J., & McConnell, A. R. (2007). Stereotype threat
between math performance, spatial skills, and attitudes. Joumal of
and working memory; Mechanisms, alleviation, and spillover. Joumal
Applied Developmental Psychology, 32, 235-242. doi:10.1016/j.appdev
of Experimental Psychology: General, 136, 256-276. doi;10.1037/0096-
.2011.04.001
3445.136.2.256
Gathercole, S. E., Pickering, S. J., Knight, C , & Stegmann, Z. (2004).
Berg, D. H. (2008). Working memory and arithmetic calculation in chil-
Working memory skills and educational attainment: Evidence from
dren; The contributory roles of processing speed, short-term memory,
national curriculum assessments at 7 and 14 years of age. Applied
and reading. Joumal of Experimental Child Psychology, 99, 288-308.
Cognitive Psychology, 18, 1-16. doi;10.1002/acp.934
doi;10.1016/j.jecp.2007.12.002
Gibbs, B. G. (2010). Reversing fortunes or content change? Gender gaps in
Brodish, A. B., & Devine, P. G. (2009). The role of performance-avoidance math-related skill throughout childhood. Social Science Research, 39,
goals and worry in mediating the relationship between stereotype threat 540-569. doi:10.1016/j.ssresearch.2010.02.005
and performance. Joumal of Experimental Social Psychology, 45, 180- Gilhooly, K. J., Wyn, V., Phillips, L. H., Logie, R. H., & Delia Sala, S.
185. doi;10.1016/j.jesp.2008.08.005 (2002). Visuo-spatial and verbal working memory in the five-disc Tower
Cadinu, M., Maass, A., Rosabianca, A., & Kiesner, J. (2005). Why do of London task: An individual differences approach. Thinking & Rea-
women underperform under stereotype threat? Evidences for the role of soning, 8, 165-178. doi: 10.1080/13546780244000006
negative thinking. Psychological Science, 16, 572-578. doi;10.1111/j Hayes, A. F., Preacher, K. J., & Myers, T. A. (2011). Mediation and the
.0956-7976.2005.01577.X estimation of indirect effects in political communication research. In
Casey, M. B., Nuttall, R. L., & Pezaris, E. (1997). Mediators of gender E. P. Buey & R. Lance Holbert (Eds.), Sourcebook for political com-
differences in mathematics college entrance test scores: A comparison of munication research: Methods, measures, and analytical techniques (pp.
spatial skills with internalized beliefs and anxieties. Developmental 434-465). New York, NY: Routledge.
Psychology, 33, 669-680. doi;10.1037/0012-1649.33.4.669 Hembree, R. (1988). Correlates, causes, effects and treatment of test
College Board. (2009). SAT percentile ranks for males, females, and total anxiety. Review of Educational Research, 58, 4,1-11. doi: 10.3102/
group. Retrieved from http;//professionals.collegeboard.com/data- 00346543058001047
reports-research/sat/data-tables Hembree, R. (1990). The nature, effects, and relief of mathematics anxiety.
College Board. (2010). AP summary reports: National report. Retrieved Joumal for Research in Mathematics Education, 21, 33-46. doi;
from http;//www.collegeboard.com/student/testing/ap/exgrd_sum/2010 10.2307/749455
.html Holmes, J., & Adams, J. W. (2006). Working memory and children's
Correll, S. J. (2001). Gender and the career choice process; The role of mathematical skills; Implications for mathematical development and
biased self-assessments. American Joumal of Sociology, 106, 1691- mathematical curricula. Educational Psychology, 26, 339-366. doi;
1730. doi;10.1086/321299 10.1080/01443410500341056
Crowe, S. F., Matthews, C , & Walkenhorst, E. (2007). Relationship Holmes, J., Adams, J. W., & Hamilton, C. J. (2008). The relationship
between worry, anxiety and thought suppression and the components of between visuospatial sketchpad capacity and children's mathematical
working memory in a non-clinical sample. Australian Psychologist, 42, skills. European Journal of Cognitive Psychology, 20, 272-289. doi:
170-177. doi;10.1080/00050060601089462 10.1080/09541440701612702
Dearing, E., & Hamilton, L. C. (2006). Contemporary approaches and Holmes, J., Gathercole, S. E., & Dunning, D. L. (2009). Adaptive training
classic advice for analyzing mediating and moderating variables. Mono- leads to sustained enhancement of poor working memory in children.
graphs of the Society for Research in Child Development, 71, 88-104. Developmental Science, 12, F9-F15. doi;10.1111/j.1467-7687.2009
doi; 10.1111/j. 1540-5834.2006.00406.X .00848.x
Deffenbacher, J. L. (1977). Relationship of worry and emotionality to Hyde, J. S., Fennema, E., & Lamon, S. J. (1990). Gender differences in
performance on the Miller Analogies Test. Joumal of Educational mathematics performance; A meta-analysis. Psychological Bulletin,
Psychology, 69, 191-195. doi; 10.1037/0022-0663.69.2.191 107, 139-155. doi;10.1037/0033-2909.107.2.139
Deffenbacher, J. L. (1978). Worry, emotionality, and task-generated inter- Hyde, J. S., Fennema, E., Ryan, M., Frost, L. A., & Hopp, C. (1990).
ference in test anxiety; An empirical test of attentional theory. Joumal Gender comparisons of mathematics attitudes and affect; A meta-
of Educational Psychology, 70, 248-254. doi; 10.1037/0022-0663.70.2 analysis. Psychology of Women Quarterly, 14, 299-324. doi: 10.1111/j
.248 .1471-6402.I990.tb00022.x
Deffenbacher, J. L. (1980). Worry and emotionality in test anxiety. In I. Hyde, J. S., Lindberg, S. M., Linn, M. C , Ellis, A. B., & Williams, C. C.
Sarason (Ed.), Test anxiety: Theory, research, and application (pp. (2008, July 25). Gender similarities characterize math performance.
111-128). Hillsdale, NJ; Erlbaum. Science, 321, 494-495. doi:10.1126/science.U60364
 ANXIETY, WORKING MEMORY, GENDER, AND MATH 117
Dceda, M., Iwanaga, M., & Seiwa, H. (1996). Test anxiety and working diator modeh. Behavior Research Methods, 40, 879-891. doi: 10.3758/
memory system. Perceptual and Motor Skills, 82, 1223-1231. doi: BRM.40.3.879
10.2466/pms. 1996.82.3c. 1223 Raghubar, K. P., Bames, M. A., & Hecht, S. A. (2010). Working memory
Jarvis, H. L., & Gathercole, S. E. (2003). Verbal and non-verbal working and mathematics: A review of developmental, individual difference, and
memory and achievements on national curriculum tests at 11 and 14 cognitive approaches. Leaming and Individual Differences, 20, 110-
years of age. Educational and Child Psychology, 20, 123-140. 122. doi:10.1016/j.lindif.2009.10.005
Keppel, G., & Wickens, T. D. (2004). Design and analysis: A researcher's Ramirez, G., & Beilock, S. L. (2011, January 14). Writing about testing
handbook (4th ed.). Upper Saddle River, NJ: Prentice Hall. worries boosts exam performance in the classroom. Science, 331, 211-
Kim, S. H., & Rocklin, T. (1994). The temporal patterns of worry and 213. doi: 10.1126/science.l 199427
emotionality and their differential effects on test performance. Anxiety, Rapee, R. M. (1993). The utilisation of working memory by worry.
Stress & Coping: An Intemational Joumal, 7, 117-130. doi: 10.1080/ Behaviour Research and Therapy, 31, 617-620. doi: 10.1016/0005-
10615809408249339 7967(93)90114-A
Klingberg, T., Forssberg, H., & Westerberg, H. (2002). Training of work-
Robinson, J. P., & Lubienski, S. T. (2011). The development of gender
ing memory in children with ADHD. Joumal of Clinical and Experi-
achievement gaps in mathematics and reading during elementary and
mental Neuropsychology, 24, 781-791. doi:10.1076/jcen.24.6.781.8395
middle school: Examining direct cognitive assessments and teacher
Kyttälä, M., & Lehto, J. E. (2008). Some factors underlying mathematical
ratings. American Educational Research Joumal, 48, 268-302. doi:
performance: The role of visuospatial working memory and non-verbal
10.3102/0002831210372249
intelligence. European Journal of Psychology of Education, 23, 77-94.
doi:10.1007/BF03173141 Shackman, A. J., Sarinopoulos, I., Maxwell, J. S., Pizzagalli, D., Lavric,
Lee, J. H. (1999). Test anxiety and working memory. Joumal of Experi- A., & Davidson, R. J. (2006). Anxiety selectively disrupts visuospatial
mental Education, 67, 218-240. doi:10.1080/00220979909598354 working memory. Emotion, 6, 40-61. doi:10.1037/1528-3542.6.1.40
Liebert, R. M., & Morris, L. W. (1967). Cognitive and emotional compo- Snyder, T. D., & Dillow, S. A. (2011). Digest of education statistics 2010
nents in test anxiety: A distinction and some initial data. Psychological (NCES 2011-015). Washington, DC: National Center for Education
Reports, 20, 975-978. doi:I0.2466/pr0.1967.20.3.975 Statistics, Institute of Education Sciences, U.S. Department of Educa-
Lindberg, S. M., Hyde, J. S., Petersen, J. L., & Linn, M. C. (2010). New tion.
trends in gender and mathematics performance: A meta-analysis. Psy- St Clair-Thompson, H. L., & Gathercole, S. E. (2006). Executive functions
chological Bulletin, 136, 1123-1135. doi:l0.1037/a0021276 and achievements in school: Shifting, updating, inhibition, and working
Ma, X. (1999). A meta-analysis of the relationship between anxiety toward memory. The Quarterly Journal of Experimental Psychology, 59, 745-
mathematics and achievement in mathematics. Joumal for Research in 759. doi: 10.1080/17470210500162854
Mathematics Education, 30, 520-540. doi: 10.2307/749772 Taylor, A. B., MacKinnon, D. P., & Tein, J. (2008). Tests of the three-path
Miller, H., & Bichsel, J. (2004). Anxiety, working memory, gender, and mediated effect. Organizational Research Methods, 11, 241-269. doi:
math performance. Personality and Individual Differences, 37, 591- 10.1177/1094428107300344
606. doi:10.1016/j.paid.2003.09.029 Teriecki, M. S., Newcombe, N. S., & Little, M. (2008). Durable and
Morris, L. W., Davis, M. A., & Hutchings, C. H. (1981). Cognitive and generalized effects of spatial experience on mental rotation: Gender
emotional components of anxiety: Literature review and a revised differences in growth pattems. Applied Cognitive Psychology, 22, 996-
worry-emotionality scale. Joumal of Educational Psychology, 73, 541- 1013. doi:10.1002/acp.l420
555. doi: 10.1037/0022-0663.73.4.541 Thorell, L. B., Lindquist, S., Nutley, S. B., Gunilla, B., & Klingberg, T.
National Science Foundation, National Center for Science and Engineering (2009). Training and transfer effects of executive functions in preschool
Statistics. (2011). Women, minorities, and persons with disabilities in children. Developmental Science, 12, 106-113. doi:10.111 l/j.1467-
science and engineering (NSF 11-309). Retrieved from http://www.nsf 7687.2008.00745.x
.gov/statistics/wmpd
U.S. News & World Report. (2012). College search. Retrieved from
O'Neil, H. F., & Fukumura, T. (1992). Relationship of worry and emo- http://colleges.usnews.rankingsandreviews.com/best-colleges
tionality to test performance in the Juku environment. Anxiety, Stress &
Vasilyeva, M., Casey, B. M., Dearing, E., & Ganley, C. M. (2009).
Coping: An International Journal, 5, 241-251. doi: 10.1080/
Measurement skills in low-income elementary school students: Explor-
10615809208249525
ing the nature of gender differences. Cognition and Instruction, 27,
Osbome, J. W. (2001). Testing stereotype threat: Does anxiety explain race 401-428. doi:10.1080/07370000903221809
and sex differences in achievement? Contemporary Educational Psy-
Voyer, D., Voyer, S., & Bryden, M. P. (1995). Magnitude of sex differ-
chology, 26, 291-310. doi:10.1006/ceps.2000.1052
ences in spatial abilities: A meta-analysis and consideration of critical
Owens, M., Stevenson, J., Norgate, R., & Hadwin, J. A. (2008). Processing variables. Psychological Bulletin, 117, 250-270. doi:I0.1037/0033-
efficiency theory in children: Working memory as a mediator between 2909.117.2.250
trait anxiety and academic performance. Araiety, Stress & Coping: An
Wood, I. (2006). Effect of anxiety reduction on children's school perfor-
Intemational Joumal. 21, 417-430. doi:10.I080/10615800701847823
mance and social adjustment. Developmental Psychology, 42. 345-349.
Pomerantz, E. M., Altermatt, E. R., & Saxon, J. L. (2002). Making the doi: 10.1037/0012-1649.42.2.345
grade but feeling distressed: Gender differences in academic perfor- Wright, R., Thompson, W. L., Ganis, G., Newcombe, N. S., & Kosslyn,
mance. Joumal of Educational Psychology. 94. 396-404. doi:10.1037/ S. M. (2008). Training generalized spatial skills. Psychonomic Bulletin
0022-0663.94.2.396 & Review. 15, 763-771. doi:10.3758/PBR.15.4.763
Preacher, K. J., & Hayes, A. F. (2004). SPSS and SAS procedures for Zeidner, M., & Nevo, B. (1992). Test anxiety in examinees in a college
estimating indirect effects in simple mediation models. Behavior Re- admission testing situation: Incidence, dimensionality, and cognitive
search Methods, Instruments. & Computers. 36, l\l-l'i\. doi:10.3758/ correlates. In H. M. van der Ploeg, R. Schwarzer, & C. D. Spielberger
BF03206553 (Series Eds.) and K. A. Hagtvet & T. B. Johnsen (Vol. Eds.), Advances
Preacher, K. J., & Hayes, A. F. (2008). Asymptotic and resampling in test anxiety research (Vol. 7, pp. 288-303). Amsterdam/Lisse, the
strategies for assessing and comparing indirect effects in multiple me- Netheriands: Swets & Zeitlinger.

{Appendix follows)
118 GANLEY AND VASILYEVA

Appendix
Mathematics Test Items

Test item Female percent correct" Male percent correct"

(1) Yvonne has studied the change in cost of tickets over time for her favorite sports team.
She has created a model to predict the cost of a ticket in the future. Let C represent the
cost of a ticket in dollars and y represent the number of years in the future. Her model
is as follows: 63%
C = 2.50^ + 13 52%
Based on this model, how much will the cost of a ticket increase in two years?
(A) $5* (B) $8 (C) $13 (D) $18 (E) $26

(2) The length of a rectangle is 3 more than its width. If L represents the length, what is
an expression for the width? 71%
(A) 3 - Í. (B) L - 3 (C) L X 3 (D) Z. -H 3 (E) ¿ - 3* 42% 57%

(3) Which of the following is the graph of 12jr - 51 a 3? 25% 38%

A) •4 1 1 1 h
-2 -1 0 t 2 3 4

— I — I — I — I — I — « I I I
-4 -3 -2 -I 0 I 2 3 4

• I I i
- 4 - 3 - 2 - 1 0 1 2 3 4

*-é 1 11—I—I—I—h
I I I h
-2 -1 0 1 2 2 3 4

H 1—»—( 1
-4-3-2-10 1 2 3 4

(4) For what value of x is 8'^ = 16"? 16%

(A) 3 (B) 4 (C) 8 (D) 9* (E) 12 23%

(5) If 3x + 2)' = 11 and 2jr -I- 3}' = 17, what is the average (arithmetic mean) of x and 57%
(A) 2.5 (B) 2.8* (C) 5.6 (D) 5.8 (E) 14 55%

(6) The table below shows a linear relationship between x and y.

X y
-7 a

-3 10

-1 6

0 4

5 -6

38%
What is the value of a?
(A) - 1 8 (B) - 1 4 (C) 14 (D) 18* (E) 36

{Appendix continues)
 ANXIETY, WORKING MEMORY, GENDER, AND MATH 119
Appendix (continued)

Test item Female percent correct" Male percent correct'

67%

If triangles ADE and ABC shown in the figure above are similar, what is the value of x"!
(A) 4* (B) 5 (C) 6 (D) 8 (E) 10

(8) A circle with diameter 10 centimeters is to be cut from a square of paper 10

centimeters on a side. Of the following, which is closest to the amount of paper left
over after the circle is cut out? 32% 29%
(A) 9 square centimeters (B) 21 square centimeters* (C) 24 square centimeters
(D) 69 square centimeters (E) 84 square centimeters 33% 45%

42% 74%

Semicircles are constructed on the sides of an equilateral triangle, as shown in the figure
above. Of the following, which best approximates the sum of the lengths of the three
darkened arcs? 39%
(A) 4.4258 (B) 4.7124* (C) 6.0000 (D) 6.7124 (E) 9.4258

(10) In the ;i:>'-plane, a line parallel to the .t-axis intersects the y-axis at the point (0, 4).
This line also intersects a circle in two points. The circle has a radius of 5 and its center
is at the origin. What are the coordinates of the two points of intersection?
(A) (2, 1) and ( 2 , - 1 ) (B) (3, 4) and (3, - 4 ) (C) (3, 4) and ( - 3 , 4)* (D) (5, 4) and ( - 5 ,
4) (E) (5, 0) and ( - 5 , 0)

(11) In the figure shown below, triangle TUV is formed by joining the midpoints of the
sides of equilateral triangle QRS. Triangle WYZ is formed by joining the midpoints of
the sides of triangle TUV.
Q

If the area of triangle QRS is 64 square inches, what is the area of triangle WTZ?
(A) 1 square inch (B) 4 square inches* (C) 8 square inches (D) 16 square inches
(E) 64 square inches

(Appendix continues)
120 GANLEY AND VASILYEVA

Appendix (continued)

Test item Female percent correct" Male percent correct"

41% 30%

How many of the unit cubes above would it take to make the object below? 47%

(A) 15 (B) 16 (C) 30* (D) 32 (E) 45

Note. An asterisk indicates the correct answer.
" Study 1 means are presented first with Study 2 means below.

Received January 31, 2012

Revision received July 1, 2013
Accepted July 5, 2013
Copyright of Journal of Educational Psychology is the property of American Psychological
Association and its content may not be copied or emailed to multiple sites or posted to a
listserv without the copyright holder's express written permission. However, users may print,
download, or email articles for individual use.