Test Anxiety and Metacognitive Performance in the Classroom

Abstract

Test anxiety is a context-specific academic anxiety which can result in poorer academic and metacognitive performance. We assessed how the quantity and relative weight of assessments contribute to the effects of test anxiety on performance and metacognitive accuracy in a smaller seminar-style class on human memory (study 1) and a larger lecture-style class on cognitive psychology (study 2). Students took six low-stakes quizzes each worth 10% of their final grade in study 1 and two high-stakes exams each worth 40% of their final grade in study 2. All students provided their state anxiety and predicted their scores before and after each assessment. Students in both classes also provided their trait (overall) anxiety after the final assessment. In both studies, students' higher post-state anxiety appeared to be associated with worse assessment performance; however, pre-and post-state anxiety decreased across the quarter in study 1 but remained constant in study 2. Additionally, we found that metacognitive accuracy moderated the effect of post-state anxiety on performance in study 1. Students with higher trait anxiety in study 1 were underconfident in their scoring predictions, while in study 2 students with higher trait anxiety performed worse on their assessments. Thus, students' metacognitive accuracy appears to be influenced by trait anxiety when taking low-stakes quizzes, while performance is related to trait anxiety when taking high-stakes exams.

Full text

INTERVENTION STUDY
Test Anxiety and Metacognitive
Performance in the Classroom
Katie M. Silaj
1
&Shawn T. Schwartz
1,2
&Alexander L. M. Siegel
1,3
&
Alan D. Castel
1
Accepted: 26 January 2021/
#Springer Science+Business Media, LLC, part of Springer Nature 2021
Abstract
Test anxiety is a context-specific academic anxiety which can result in poorer academic
and metacognitive performance. We assessed how the quantity and relative weight of
assessments contribute to the effects of test anxiety on performance and metacognitive
accuracy in a smaller seminar-style class on human memory (study 1) and a larger
lecture-style class on cognitive psychology (study 2). Students took six low-stakes
quizzes each worth 10% of their final grade in study 1 and two high-stakes exams each
worth 40% of their final grade in study 2. All students provided their state anxiety and
predicted their scores before and after each assessment. Students in both classes also
provided their trait (overall) anxiety after the final assessment. In both studies, students’
higher post-state anxiety appeared to be associated with worse assessment performance;
however, pre- and post-state anxiety decreased across the quarter in study 1 but remained
constant in study 2. Additionally, we found that metacognitive accuracy moderated the
effect of post-state anxiety on performance in study 1. Students with higher trait anxiety
in study 1 were underconfident in their scoring predictions, while in study 2 students with
higher trait anxiety performed worse on their assessments. Thus, students’metacognitive
accuracy appears to be influenced by trait anxiety when taking low-stakes quizzes, while
performance is related to trait anxiety when taking high-stakes exams.
Keywords Test anxiety .Metacognition .Trait anxiety .Low-stakes testing .Classroom study
Educational Psychology Review
https://doi.org/10.1007/s10648-021-09598-6
*Katie M. Silaj
kmsilaj@ucla.edu
1
Department of Psychology, University of California, Los Angeles, 502 Portola Plaza, Los Angeles,
CA 90095, USA
2
Present address: Department of Ecology and Evolutionary Biology, University of California, Los
Angeles, Los Angeles, CA, USA
3
Present address: Leonard Davis School of Gerontology, University of Southern California, Los
Angeles, CA, USA

Test anxiety is a context-specific academic anxiety in which an individual reacts to an
evaluative event like a quiz, midterm test, or final exam in an anxious, fearful, or nervous
manner (Cassady, 2010). It has been estimated to impact 25–40% of students (Cassady, 2010)
and can result in a combination of negative outcomes inducing poor test performance (Cassady
& Johnson, 2002). Test anxiety can be viewed as both a state anxiety that is triggered by the
testing event, but also as a trait anxiety in which the effects of test anxiety are more severe for
people with high levels of baseline anxiety (hereafter “trait anxiety”; Goetz et al., 2013;Harris
et al., 2019; Spielberger, 1972). According to Zohar (1998), the impact of test anxiety depends
on both the individual’s trait anxiety and the situational factors surrounding the test, such as the
individual’s level of confidence in their content knowledge or the level of stakes associated
with the test.
Interventions for test anxiety can increase students’performance while having no effect on
trait anxiety (Harris et al., 2019). Thus, it is important to consider individual differences in trait
anxiety when investigating the impact of test anxiety on performance.
Furthermore, external factors may also play a role in the effects of test anxiety, such as the
stakes (Wood et al., 2016) and format (Hembree, 1988) of the assessment. For example, higher
test anxiety is associated with poor performance on high-stakes standardized tests (Cassady &
Johnson, 2002; Putwain, 2008), and the multiple-choice testing format is associated with
higher test anxiety when compared to a matching format (Hembree, 1988)oranopen-ended
format (Birenbaum & Feldman, 1998).
Lastly, metacognition, or the ability to assess and control our cognitive processes (Everson
et al., 1994), is related to test anxiety such that students with high test anxiety often show a
lower level of metacognitive skillfulness compared to students with low test anxiety. Some
work has suggested that metacognition may be a mediator between test anxiety and perfor-
mance (Veenman et al., 2000) and between test anxiety and study strategies (Spada et al.,
2006). Hence, the focus of the current study was to examine how self-reported state anxiety
affects student performance and metacognitive accuracy, and how those measures vary with
levels of self-reported trait anxiety along with the number and weight of assessments students
take in a given course.
Test Anxiety and Academic Performance
Test anxiety is associated with poorer academic performance in elementary, high school, and
college students (Cassady & Johnson, 2002;Putwain&Best,2011; Williams, 1991) and is
considered to be one of the most disruptive factors in exam performance (Cizek & Burg, 2006;
von der Embse & Hasson, 2012). High test anxiety is related to lower test scores in general
and, specifically, to poor performance on college entrance exams and standardized tests, which
provides evidence that the stakes of assessments may induce higher levels of test anxiety (von
der Embse et al., 2018). In one study, von der Embse and Hasson (2012) examined the effects
of test anxiety on student performance on a high-stakes state graduation exam in an urban and
a suburban high school. They found that test anxiety accounted for 4–15% of the variance in
test scores when controlling for school. This result illustrates the wide-ranging effects of test
anxiety and raises the concern that the effects of test anxiety and poor performance on high-
stakes assessments may be compounded for students in school districts that fail to meet
standards for funding. Additionally, poor performance on high-stakes exams could lead to
other negative consequences for students such as low course grades (Segool et al., 2013), less
Educational Psychology Review

attentional control during exams (Fernández-Castillo & Caurcel, 2015), and increased anxiety
and depression (Leadbeater et al., 2012; von der Embse et al., 2018). Thus, it is important to
better understand which factors associated with real classroom environments are related to the
relationship between test anxiety and performance.
Interventions for Test Anxiety
Despite decades of research on test anxiety, there remains a lack of consensus amongst experts
in the field regarding the most effective interventions to combat its detrimental effects on
student performance. One intervention that an instructor might implement to encourage the
development of better study habits and a more positive outlook on testing events is to frame
assessments as “retrieval practice.”There is a considerable amount of research that supports
the idea that testing oneself is a much more powerful learning event than simply restudying
material (Roediger & Karpicke, 2006), with retrieval practice being described as a desirable
difficulty (Bjork, 1994). The positive implications of testing oneself are often not apparent
until after the learner has taken the assessment and may seem like a challenge to the learner
initially. This knowledge has led researchers in applied educational settings to implement
testing as an intervention to promote student retention of learned content (Karpicke, 2017).
The results of these classroom experiments have led to recommendations that teachers
should use testing as a tool to further student learning (Dunlosky et al., 2013;Dunnetal.,
2013). Perhaps regarding test anxiety, more frequent testing events provided by the instructor
can help the learner to engage in retrieval practice, which may negate the feelings of
anxiousness in response to tests over time. A recent study investigated the effects of retrieval
practice on test anxiety in middle and high school students (Argarwal et al., 2014) and reported
that 72% of students credited retrieval practice with making them feel less nervous for tests and
exams. As such, this may be a promising avenue for reducing test anxiety in a classroom
setting, although it has yet to have been explored within a real-world college setting.
An intervention that may be more easily implemented in an undergraduate education
context is varying the frequency of testing events within a given course. In a recent study,
participants studied five word lists, and then after a distractor task, they either took an interim
test or restudied the words (Yang et al., 2020). Performance on a cumulative test at the end of
the study was significantly higher for participants who took interim tests after each list
compared with the group that restudied the words. Additionally, participants took a Test
Anxiety Inventory (TAI; Chinese version; Yue, 1996) which revealed that the low-stakes
assessments can improve performance regardless of levels of test anxiety. Although this was
demonstrated in a controlled, experimental context, we are currently interested in examining
the effects of format and relative weight of assessments on performance and metacognitive
accuracy in a classroom setting.
Test Anxiety and Metacognitive Accuracy
Test anxiety and metacognition are intimately linked (Spada et al., 2006). Thus, the process of
monitoring oneself throughout the testing cycle may be an important factor in determining
performance outcomes. Heightened anxiety brought upon by metacognitive awareness of a
lack of preparation or ability causes the learner to become more metacognitively aware of their
Educational Psychology Review

failure to succeed on the test, thus inducing more anxiety (Covington, 1985;Naveh-Benjamin,
1991). This can lead to additional cognitive interference because these resources that could
have been devoted to retrieving test-relevant knowledge are essentially divided due to
heightened levels of anxiety (Cassady & Johnson, 2002; Kurosawa & Harackiewicz, 1995;
Veenman et al., 2000). Furthermore, individual differences exist such that students with high
test anxiety may display lower levels of metacognitive abilities when compared to students
with low levels of test anxiety (Veenman et al., 2000).
Veenman et al. (2000) assessed this relationship between metacognition and performance
using the metacognitive word knowledge task. In this task, participants were asked if they
knew the definitions of 31 health science words, and then were asked to provide definitions.
The authors assessed metacognition by computing hits (e.g., the number of words participants
said they knew and correctly identified their meaning on the vocabulary task) and false alarms
(e.g., the number of words participants said they knew but did not correctly identify their
meaning on the vocabulary task) rates. Participants with low test anxiety exhibited a superior
metacognitive skillfulness during math performance relative to participants with high test
anxiety (Veenman et al., 2000). In another study, test anxiety exerted a negative influence
on students’performance on a metacognitive word knowledge task, independent of overall
reading ability (Everson et al., 1994).
Metacognitive accuracy can interact with math anxiety such that higher math anxiety
predicts lower test performance except in students that are high in metacognitive skillfulness
(Legg & Locker, 2009). Additionally, when students are more metacognitively aware, they
report higher confidence in their ability to answer test problems correctly. Some work has
shown that with feedback, students can improve their metacognitive accuracy from one exam
to another (Callender et al., 2016). Thus, more retrieval practice opportunities within an
academic term may lead to increased metacognition, which may especially benefit students
that are high in test anxiety.
Overview of the Present Study
In the current study, we examined the effects of self-reported state anxiety on performance and
metacognitive accuracy in a true classroom setting, and how these measures vary with
students’levels of self-reported trait anxiety and (i) the number and (ii) relative weight of
assessments on a student’s overall grade within a given course. In a smaller, seminar-style
class in human memory, students took six quizzes that were each worth 10% of their final
grade (study 1), while in a larger, lecture-style class in cognitive psychology, students took two
exams, each worth 40% of their grade (study 2). Thus, study 1 provided a context in which we
could examine the impact of state and trait anxiety on student performance and metacognition
in a smaller, low-stakes setting where students had more opportunities for retrieval practice,
while study 2 provided a context in which we could examine the impact of state and trait
anxiety on student performance and metacognition in a larger, high-stakes environment with
less retrieval practice opportunities.
Students in both classes completed a questionnaire both before and after each assessment
within the span of the academic quarter (11 weeks: 10 instructional weeks + 1 final exam
week). The questionnaires prompted students to predict and postdict their scores on the
assessment and rate their level of state anxiety pre- and post-assessment. Students also gave
a rating of trait anxiety either after the final assessment of the academic quarter (study 1) or
Educational Psychology Review

after each assessment (study 2). We used single-report measures as opposed to a formal
questionnaire to avoid introducing additional anxiety into the testing environment and as a
practical means to address time constraints imposed by the classroom setting. We were
interested in measuring self-reported state anxiety and its relationship with performance and
metacognitive accuracy while controlling for individual differences in self-reported trait
anxiety.
We expected that student scores on assessments and their metacognitive accuracy in
predicting their scores would significantly increase across the quarter for both classes due to
experience with the content and expectations of the course. We also expected state anxiety to
significantly decrease from pre- to post-assessment for students in both classes due to a sense
of relief that the test is over, in addition to across the quarter for students who engaged in more
frequent, relatively lower-stakes assessments as they will gain retrieval practice (Argarwal
et al., 2014). We expected students who report high ratings of trait anxiety to also report
significantly higher ratings of state anxiety and have significantly lower scores on the
assessments compared to students who report low ratings of trait anxiety (Zohar, 1998).
Lastly, we expected metacognition to significantly interact with test anxiety, such that students
with high test anxiety would be less accurate in their assessment score predictions and
postdictions compared to students with low test anxiety (Veenman et al., 2000).
Study 1
Method
Participants
Participants were 49 University of California, Los Angeles (UCLA), undergraduate students
enrolled in Psychology 124C: Human Memory during Winter Quarter 2018. One outlier was
removed due to providing ratings outside of the requested range, so there were 48 students
included in the final sample.
Materials and Procedure
There were six, short-answer format, low-stakes quizzes given during weeks 3–9(noquiz
week 6). The quizzes covered materials from the lecture and assigned readings from the
previous week. Possible quiz scores ranged from 0 to 10 and the lowest quiz score was
dropped at the end of the quarter. Quizzes accounted for 50% of the final grade (each quiz was
worth 10%). There was also a written assignment, research proposal, and in-class presentation
that in combination contributed the remaining 50% of the final grade (not included in the data
analyzed here). Quiz topics included (in this order) short-term and working memory, encoding
and studying, forgetting and retrieval/testing, autobiographical and eyewitness memory, self-
regulated learning, metamemory, memory and aging, and expertise and applications. An
example of the type of question a student may have answered on a quiz would be of the
following form and level of conceptual understanding: “Define and distinguish between
proactive and retroactive interference and give unique examples of each. How can you show
a release of proactive interference, and what does this suggest about the nature of
forgetting?”
Educational Psychology Review

Before each quiz, students were asked “How nervous/anxious do you feel at the
current moment?”and would respond on a Likert scale from 1 (not at all anxious)to
10 (very much anxious). We used this measure to assess student levels of state
anxiety in an effort to minimize any effects of anxiety a full questionnaire might
induce. Single-scale measures have been found to correlate well with more extensive
scales (Davey et al., 2007; Núñez-Peña et al., 2013)andaremorepracticalandless
invasive for our purposes when considering inherent student anxiety and time con-
straints in a more naturalistic, classroom setting.
Students were then asked to predict their scores out of 10 points. They had 45 min
to answer the three short answer questions. After the quiz, students reported their state
anxiety on the same scale a second time. Lastly, they were asked to postdict their
scores out of 10 points. The students were told that their responses to the pre- and
post-quiz surveys were voluntary and would not influence their grades. This procedure
was followed for all six quizzes. At the end of quiz 6 (week 9), students were asked
“In general, how anxious of a person do you think you are?”and responded on a
Likert scale ranging from 1 (not at all anxious in general)to10(very much anxious
in general). This measure was requested following the final post-quiz state anxiety
rating of the academic quarter and was used to assess individual levels of trait
anxiety.
Additionally, an assigned course “reader”graded all assignments for the course and was
blind to the hypotheses and predictions of the current study as well as unaware of participants’
self-reported ratings when grading the quizzes. All materials and procedures were performed
ethically and approved by the UCLA Institutional Review Board.
Results and Discussion
For the following analyses, we first conducted a repeated-measures analysis of variance
(ANOVA) on each dependent variable while treating quiz number throughout the quarter as
a within-subjects independent measure (for ANOVAs where Mauchly’stestofsphericitywas
violated, Greenhouse-Geisser sphericity corrections were used). Then, we conducted both
multiple regression analyses and interrupted regressions (Simonsohn, 2018)totestifquiz
number significantly predicted participants’overall quiz performance, estimation accuracy,
and anxiety throughout the quarter. We did not assume monotonicity of each relationship, and
therefore tested the plausibility of both the presence of linear and non-linear relationships
between quiz number and each dependent variable (Bowman et al., 1998). Interrupted
regressions were subsequently used to reveal the potential existence of U-shaped (or
inverted U-shaped) relationships and were conducted using the “Robin Hood”algorithm to
identify a break point along measures of the dependent variable (two-lines test, Simonsohn,
2018). Upon examining dependent variables throughout the academic quarter, a U-shaped
relationship might have occurred if, for example, a dependent measure was higher or lower at
both the beginning and end of the academic quarter relative to the middle. Such a pattern may
result from external factors potentially elevating levels of anxiety or performance due to the
nature of uncertainty when becoming acquainted with a new course at the beginning of a term
or knowing that stakes are high for students with borderline grades at the end of the term.
Additionally, the interrupted regression model using the “Robin Hood”algorithm generally
yields higher statistical power for detecting U-shaped relationships than other statistical
alternatives, and also is less likely to result in higher rates of false-positive and false-
Educational Psychology Review

negative results that can occur when assuming and forcing a standard quadratic functional
form on the data (Simonsohn, 2018). In accordance with Simonsohn (2018), we deemed a U-
shaped relationship to exist, and therefore rejected the null hypothesis of an absence of a U-
shaped relationship, if both slopes of the two lines were significant and of opposite sign. For
each dependent variable, if both the results from the multiple regression model and the
interrupted regression model were significant, we opted for the most parsimonious explanation
by choosing the multiple regression (linear) model to explain the apparent relationship
between the data. Additionally, we reported adjusted R2values as percent of variance
explained for all linear regressions.
Quiz Performance Throughout the Quarter
To examine overall quiz performance with respect to quiz number, we conducted a 1 (Exam:
quiz) × 6 (Quiz number: 1, 2, …,6) repeated-measures ANOVA on overall quiz scores, which
revealed a significant main effect of quiz number, F(2.78, 108.50) = 10.03, p<.001,η2=.20
(Table 1). Post hoc paired-samples t-tests with a Bonferroni correction indicated that scores
were higher on quizzes 3 (M=8.58, SD = 1.31), 4 (M=9.35, SD = .62), 5 (M=8.85, SD =
1.03), and 6 (M=8.98, SD = 1.07) relative to quiz 1 (M=7.88, SD = 1.91), higher on quiz 4
(M=9.35, SD = .62) relative to quiz 2 (M=8.45, SD = .95), and higher on quiz 4 (M=9.35,
SD = .62) relative to quiz 3 (M=8.58, SD = 1.31), adjusted ps < .03. No other comparisons
were significant, adjusted ps > .17.
Next, we conducted regressions using both pre-quiz and post-quiz state anxiety as predic-
tors. The multiple regression model took the following form: quiz score = β0+β1(quiz
number) + β2(pre-quiz state anxiety) + β3(post-quiz state anxiety) + β4(trait anxiety). Four
predictors significantly explained 14.8% of the variance, R2=.16, F(4, 269) = 12.84, p<.0001,
and we found that quiz number (β1=.20,p< .0001) and post-quiz state anxiety (β3=−.19, p<
.0001) significantly predicted quiz score, but pre-quiz state anxiety (β2=.05,p= .24) and trait
anxiety (β4=.04,p= .27) did not (Table 2). The interrupted regression model took the same
form as the multiple regression model and revealed no significant U-shaped relationship β1–4=
.38, p<.0001;β5–6=−.12, p= .24). Overall, these findings indicate a significant positive linear
relationship between quiz number and average quiz score, such that when controlling for both
pre-quiz and post-quiz state anxiety along with trait anxiety, the average quiz score for each
student increased with each subsequent quiz throughout the quarter (see Fig. 1).
Table 1 Repeated-measures analysis of variance coefficients in study 1
Outcome Fdf p η2
Quiz performance 10.03 (2.78, 108.50) < .001*** .20
Predictions 4.71 (3.70, 136.95) .002** .11
Postdictions 7.62 (3.17, 104.74) < .001*** .19
Prediction calibration 2.01 (5, 195) .08 .05
Postdiction calibration 1.93 (3.40, 132.55) .12 .05
Overall calibration 2.66 (5, 195) .02* .06
Calibration difference score 4.12 (3.62, 170.26) .004** .08
Pre-quiz state anxiety 4.66 (2.21, 10.26) < .001*** .11
Post-quiz state anxiety 3.16 (5, 170) .009** .09
State anxiety difference sc ore 3.81 (5, 235) .002** .08
*p<.05, **p<.01,***p<.001
Educational Psychology Review

Table 2 Multiple regression analyses coefficients using pre-quiz and post-quiz state anxiety as a predictor in study 1
Outcome % variance explained
Explained
R2Fdf β0β1β2β3β4
Quiz score 14.8% .16**** 12.84 (4, 269) 8.32**** .20**** .05 −.19**** .04
Prediction 5.7% .07*** 5.13 (4, 267) 8.33**** .05 −.11* −.02 −.09*
Postdiction 18.5% .20**** 16.39 (4, 267) 9.54**** −.02 .09 −.29**** −.11**
Prediction calibration 8.0% .09**** 6.90 (4, 269) −.22 .17* .16* −.16* .15**
Postdiction calibration 11.2% .12**** 9.57 (4, 269) −1.20** .19*** −.03 .13* .15***
Overall calibration 9.1% .10**** 7.81 (4, 269) −.71 .18** .06 −.02 .15***
Calibration difference score 7.1% .08**** 6.25 (4, 269) 1.16*** −.13** .13** −.12** .04
*p<.05, **p<.01,***p< .001, ****p< .0001
Model: Outcome = β0+β1(quiz number) + β2(pre-quiz state anxiety) + β3(post-quiz state anxiety) + β4(trait anxiety)
Educational Psychology Review

Estimation Accuracy Throughout the Quarter
To examine quiz score metacognitive prediction (prior to taking each quiz) with respect to quiz
number, we conducted a 1 (Exam: quiz)×6(Quiznumber:1, 2, …,6) repeated-measures
ANOVA on quiz score prediction, which revealed a significant main effect of quiz number,
F(3.70, 136.95) = 4.71, p=.002, η2=.11(Table1). Post hoc paired-samples t-tests with a
Bonferroni correction indicated that metacognitive score prediction was higher on quiz 4 (M=
7.87, SD = 1.25) than on quiz 1 (M=6.96, SD = 1.94), and lower on quiz 5 (M=7.07, SD =
1.53) than on quiz 4 (M=7.87, SD = 1.25), adjusted ps < .006. No other comparisons were
significant, adjusted ps > .055.
Next, we conducted separate regressions examining quiz score prediction and postdiction,
each using both pre-quiz state anxiety and post-quiz state anxiety as predictors. For quiz score
prediction, the multiple regression model took the following form: quiz score prediction = β0+
β1(quiz number) + β2(pre-quiz state anxiety) + β3(post-quiz state anxiety) + β4(trait
anxiety). The four predictors significantly explained 5.7% of the variance, R2=.07, F(4, 267)
=5.13,p=.0005, and we found that pre-quiz state anxiety significantly predicted quiz score
prediction (β2=−.11, p= .04) as did trait anxiety (β4=−.09, p=.04),butquiznumber(β1=
.05, p= .36) and post-quiz state anxiety (β3=−.02, p= .76) did not. The interrupted regression
model took the same form as the multiple regression model and revealed no significant U-
shaped relationship (β1–3=.31,p=.05;β4–6=−.10, p= .29) (Table 2). Thus, while there
appeared to be a main effect of quiz number, when controlling for participants’state anxiety
and trait anxiety, quiz number was no longer a significant predictor of quiz score prediction
(see Fig. 2a).
Secondly, to examine quiz score metacognitive postdiction (directly following taking each
quiz) with respect to quiz number, we conducted a 1 (Exam: quiz) × 6 (Quiz number: 1, 2,…,
6) repeated-measures ANOVA on quiz score postdiction, which revealed a significant main
effect of quiz number, F(3.17, 104.74) = 7.62, p<.001, η2=.19(Table1). Post hoc paired-
Fig. 1 Mean quiz score performance in study 1 across the academic quarter. All error bars represent ± 1 standard
error
Educational Psychology Review

samples t-tests with a Bonferroni correction indicated that metacognitive score postdiction was
higher on quizzes 2 (M=8.71, SD = 1.02), 3 (M=8.25, SD = 1.10), and 4 (M=8.32, SD =
1.79) relative to quiz 1 (M=7.37, SD = 1.63), and lower on quiz 5 (M=7.65, SD = 1.70)
compared to quiz 2 (M=8.71, SD = 1.02), adjusted ps < .008. No other comparisons were
significant, adjusted ps > .10.
For quiz score postdiction, the multiple regression model took the following form: quiz
score postdiction = β0+β1(quiz number) + β2(pre-quiz state anxiety) + β3(post-quiz state
anxiety) + β4(trait anxiety). The four predictors significantly explained 18.5% of the variance,
R2=.20, F(4, 267) = 16.39, p<.0001, and we found that post-quiz state anxiety significantly
predicted quiz score postdiction (β3=−.29, p< .0001) as did trait anxiety (β4=−.11, p=
.005), but quiz number β1=−.02, p= .63) and pre-quiz state anxiety (β2=.09,p=.08)did
not. The interrupted regression model took the same form as the multiple regression model and
revealed a significant inverted U-shaped relationship (β1–2=1.34,p<.0001;β3–6=−.23, p<
.0001) (Table 2). Thus, while there appeared to be a main effect of quiz number, controlling for
participants’state anxiety and trait anxiety revealed that quiz number was no longer a
significant linear predictor of quiz score postdiction; however, quiz score postdiction was
seemingly lower at both the beginning and end of the academic quarter relative to the middle
as revealed by the significant inverted U-shaped relationship (see Fig. 2b).
We then examined another form of estimation accuracy using students’pre- and postdiction
calibration (actual score –predicted/postdicted score; such that a positive value would indicate
underconfidence and a negative value would indicate overconfidence). To examine quiz score
prediction estimation accuracy (for metacognitive predictions provided prior to taking each
quiz) with respect to quiz number, we conducted a 1 (Exam: quiz) × 6 (Quiz number: 1, 2, …,
6) repeated-measures ANOVA on quiz score prediction calibrations, which did not reveal a
significant main effect of quiz number, F(5, 195) = 2.01, p=.08,η2=.05(Table1).
For quiz score prediction calibration, the multiple regression model took the following
form: quiz score prediction calibration = β0+β1(quiz number) + β2(pre-quiz state anxiety) +
β3(post-quiz state anxiety) + β4(trait anxiety). The four predictors significantly explained
8.0% of the variance, R2=.09, F(4, 269) = 6.90, p<.0001, and we found that quiz number
significantly predicted quiz score prediction calibration (β1=.17,p= .01), as did pre-quiz state
anxiety (β2=.16,p= .02), post-quiz state anxiety (β3=−.16, p= .01), and trait anxiety (β4=
Fig. 2 Mean quiz score predictions (a) and postdictions (b) of performance in study 1 across the academic
quarter. All error bars represent ± 1 standard error
Educational Psychology Review

.15, p= .003). The interrupted regression model took the same form as the multiple regression
model and revealed no significant U-shaped relationship (β1–5=.20,p=.03;β6=−.01, p=
.97) (Table 2). Overall, these findings indicate a significant positive linear relationship for quiz
score prediction calibration through time, as when controlling for both state and trait anxiety,
the average quiz score prediction calibration for each student increased with each subsequent
quiz throughout the quarter (see Fig. 3a).
Secondly, to examine quiz score postdiction estimation accuracy (for metacognitive
postdictions provided directly following taking each quiz) with respect to quiz number, we
conducted a 1 (Exam: quiz)×6(Quiznumber:1, 2, …,6) repeated-measures ANOVA on quiz
score postdiction calibration, which did not reveal a significant main effect of quiz number,
F(3.40, 132.55) = 1.93, p=.12,η2=.05(Table1).
For quiz score postdiction calibration, the multiple regression model took the following
form: quiz score postdiction calibration = β0+β1(quiz number) + β2(pre-quiz state anxiety)
+β3(post-quiz state anxiety) + β4(trait anxiety). The four predictors significantly explained
Fig. 3 Mean score calibration (actual −pre-/postdicted score) before (pre-quiz; panel a) and after (post-quiz;
panel b) each quiz in study 1 across the academic quarter. Mean calibration difference scores (absolute value of
difference between pre-quiz and post-quiz score calibration; panel c) in study 1 across the academic quarter. All
error bars represent ± 1 standard error
Educational Psychology Review

11.2% of the variance, R2=.12, F(4, 269) = 9.57, p<.0001, and we found that quiz number
significantly predicted quiz score postdiction calibration (β1=.19,p= .0005), as did post-quiz
state anxiety (β3=.13,p= .01) and trait anxiety (β4=.15,p= .0005), but pre-quiz state
anxiety (β2=−.03, p= .53) did not. The interrupted regression model took the same form as
the multiple regression model and revealed no significant U-shaped relationship (β1–3=−.25,
p=.15;β4–6=.31,p=.001)(Table2). Similarly, these findings indicate a significant positive
linear relationship for quiz score postdiction calibration through time, as when controlling for
both state and trait anxiety, the average quiz score postdiction calibration for each student
increased with each subsequent quiz throughout the quarter (see Fig. 3b).
Additionally, to test whether or not students’overall calibration (the average of the
prediction and postdiction score calibration) was generally underconfident or overconfident,
we conducted a one-sample t-test with the alternative hypothesis that students’average overall
calibration would be greater than 0 (i.e., underconfident), and found this to be the case, t(47) =
6.71, p< .001, Cohen’sd=.97. We then conducted a 1 (Exam: quiz) × 6 (Quiz number: 1, 2,
…,6) repeated-measures ANOVA on students’overall calibration, which revealed a signifi-
cant main effect of quiz number, F(5, 195) = 2.66, p=.02,η2=.06(Table1). Post hoc paired-
samples t-tests with a Bonferroni correction indicated no significant differences for overall
quiz score calibration between any other quiz, adjusted ps > .25. Next, we used a multiple
regression model that took the following form: quiz score overall calibration = β0+β1(quiz
number) + β2(pre-quiz state anxiety) + β3(post-quiz state anxiety) + β4(trait anxiety). The
four predictors significantly explained 9.1% of the variance, R2=.10,F(4, 269) = 7.81, p<
.0001, and we found that quiz number significantly predicted quiz score overall calibration (β1
=.18,p= .001) as did trait anxiety (β4=.15,p= .0004), but pre-quiz state anxiety (β2=.06,p
= .28) and post-quiz state anxiety (β3=−.02, p=.76)didnot(Table2). The interrupted
regression model took the same form as the multiple regression model and revealed no
significant U-shaped relationship (β1–5=.20,p=.007;β6=−.08, p= .80). These findings
indicate a significant positive linear relationship for quiz score overall calibration through time,
as when controlling for both state and trait anxiety, the overall quiz score calibration for each
student increased with each subsequent quiz throughout the quarter (i.e., students became more
underconfident over time); this finding is consistent with prior work showing that students may
become more underconfident as their performance improves with practice (Koriat et al., 2002).
Finally, to examine how the difference in calibration from pre-quiz to post-quiz changed
throughout time, we computed calibration difference scores (the absolute value of the differ-
ence between pre-quiz and post-quiz calibration) each week, for each student in the course. We
then conducted a 1 (Exam: quiz) × 6 (Quiz number: 1, 2, …,6) repeated-measures ANOVA on
students’calibration difference scores, which revealed a significant main effect of quiz
number, F(3.62, 170.26) = 4.12, p=.004, η2=.08(Table1). Post hoc paired-samples t-tests
with a Bonferroni correction indicated significantly lower calibration differences scores for
weeks 3 (M=1.05, SD = 1.37), 4 (M=1.01, SD = 1.57), 5 (M=1.01, SD = 1.22), and 6 (M
=.91, SD = 1.32) compared to week 2 (M=1.95, SD = 2.02), adjusted ps < .02. No other
comparisons were significant, adjusted ps > .74. Next, we used a multiple regression model
that took the following form: calibration difference score = β0+β1(quiz number) + β2(pre-
quiz state anxiety) + β3(post-quiz state anxiety) + β4(trait anxiety). The four predictors
significantly explained 7.1% of the variance, R2=.08, F(4, 269) = 6.25, p< .0001, and we
found that quiz number significantly predicted quiz calibration difference scores (β1=−.13, p
= .004), as did pre-quiz state anxiety (β2=.13,p= .003) and post-quiz state anxiety (β3=
−.12, p= .002), but trait anxiety (β4=.04,p= .17) did not (Table 2). The interrupted
Educational Psychology Review

regression model took the same form as the multiple regression model and revealed no
significant U-shaped relationship (β1–5=−.15, p=.01;β6=−.24, p= .38). Thus, we revealed
that metacognitive calibration scores appear to be more discrepant from pre- to post-
quiz earlier on in the quarter, becoming more stable from quiz 3 onward (see Fig. 3c).
Anxiety Throughout the Quarter
To examine pre-quiz state anxiety with respect to quiz number, we conducted a 1 (Exam: quiz)
× 6 (Quiz number: 1, 2, …,6) repeated-measures ANOVA on pre-quiz state anxiety, which
revealed a significant main effect of quiz number, F(2.21, 10.26) = 4.66, p<.001,η2=.11
(Table 1). Post hoc paired-samples t-tests with a Bonferroni correction indicated significantly
lower pre-quiz state anxiety for quizzes 3 (M=5.73, SD = 2.33)and4(M=5.49, SD = 2.20)
relative to quiz 1 (M=6.93, SD = 2.06), adjusted ps < .006. No other comparisons were
significant, adjusted ps > .31. Next, we conducted separate regressions examining pre-quiz and
post-quiz state anxiety, each using both quiz number and trait anxiety as predictors. For pre-
quiz state anxiety, the multiple regression model took the following form: pre-quiz state
anxiety = β0+β1 (quiz number) + β2(trait anxiety). Two predictors significantly explained
17.0% of the variance, R2=.18, F(2, 277) = 29.60, p< .0001, and we found that trait anxiety
significantly predicted pre-quiz state anxiety (β2=.37,p< .0001), but quiz number (β1=−.13,
p= .06) did not (Table 3). The interrupted regression model took the same form as the multiple
regression model and revealed no significant U-shaped relationship (β1–3=−.56, p=.009;β4–
6=.18,p= .20). Thus, although there was a significant main effect of quiz number, controlling
for trait anxiety revealed there to be no significant linear relationship between quiz number and
pre-quiz state anxiety (see Fig. 4a).
Secondly, to examine post-quiz state anxiety with respect to quiz number, we conducted a 1
(Exam: quiz)×6(Quiznumber:1, 2, …,6) repeated-measures ANOVA on post-quiz state
anxiety, which revealed a significant main effect of quiz number, F(5, 170) = 3.16, p=.009,
η2=.09(Table1). Post hoc paired-samples t-tests with a Bonferroni correction indicated
significantly lower post-quiz state anxiety for quizzes 2 (M=4.76, SD = 2.44) and 4 (M=
4.74, SD = 2.28) relative to quiz 1 (M=5.99, SD = 2.21), adjusted ps<.019.Noother
comparisons were significant, adjusted ps > .056. For post-quiz state anxiety, the multiple
regression model took the following form: post-quiz state anxiety = β0+β1(quiz number) +
β2(trait anxiety). Two predictors significantly explained 8.8% of the variance, R2=.09, F(2,
271) = 14.14, p< .0001, and we found that trait anxiety significantly predicted pre-quiz state
anxiety (β2=.27,p< .0001), but quiz number (β1=−.12, p= .12) did not (Table 3). The
interrupted regression model took the same form as the multiple regression model and revealed
no significant U-shaped relationship (β1–3=−.65, p=.004;β4–6=.09,p= .50). Therefore,
Table 3 Multiple regression analyses coefficients for state anxiety in study 1
Outcome % variance explained R2Fdf Intercept β1β2
Pre-quiz state anxiety 17.0% .18**** 29.60 (2, 277) 4.37**** −.13 .37****
Post-quiz state anxiety 8.8% .09**** 14.14 (2, 271) 3.80**** −.12 .27****
Overall state anxiety 15.8% .16**** 27.09 (2, 277) 4.12**** −.14* .32****
State anxiety difference score 1.0% .02 2.31 (2, 271) 1.78*** −.09 .05
*p<.05, **p<.01,***p< .001, ****p<.0001
Model: Outcome = β0+β1(quiz number) + β2(trait anxiety)
Educational Psychology Review

even though there was a main effect of quiz number, controlling for trait anxiety revealed no
significant relationship between quiz number and post-quiz state anxiety (see Fig. 4b).
Thirdly, to examine overall state anxiety with respect to quiz number controlling for trait
anxiety, we used a multiple regression model of the form: overall state anxiety = β0+β1(quiz
number) + β2(trait anxiety). Two predictors significantly explained 15.8% of the variance, R2
=.16, F(2, 277) = 27.09, p< .0001, and we found that quiz number significantly predicted
overall state anxiety (β1=−.14, p= .04) as did trait anxiety (β2=.32,p< .0001) (Table 3).
The interrupted regression model took the same form as the multiple regression model and
revealed no significant U-shaped relationship (β1–5=.91,p<.0001;β6–10 =0,p= .99). Thus,
we revealed that overall state anxiety appears to decrease throughout the quarter even when
controlling for trait anxiety.
Lastly, to examine how the difference in state anxiety changed from pre-quiz to post-quiz
throughout time, we computed state anxiety difference scores (the absolute value of the
Fig. 4 Mean state anxiety measures before (pre-quiz; panel a) and after (post-quiz; panel b) each quiz in study 1
across the academic quarter. State anxiety difference score (absolute value of difference between pre-quiz and
post-quiz anxiety;panelc). All error bars represent ± 1 standard error
Educational Psychology Review

difference between pre-quiz and post-quiz state anxiety) each week, for each student in the
course. We then conducted a 1 (Exam: quiz) × 6 (Quiz number: 1, 2, …,6) repeated-measures
ANOVA on students’state anxiety difference scores, which revealed a significant main effect
of quiz number, F(5, 235) = 3.81, p=.002, η2=.08(Table1). Post hoc paired-samples t-tests
with a Bonferroni correction indicated significantly lower state anxiety differences scores for
weeks 4 (M=1.53,SD = 1.49) and 6 (M=1.44, SD = 1.30) compared to week 2 (M=2.55,
SD = 2.28), adjusted ps < .006. No other comparisons were significant, adjusted ps > .057.
Next, we used a multiple regression model that took the following form: anxiety difference
score = β0+β1(quiz number) + β2(trait anxiety). The two predictors did not significantly
predict state anxiety difference scores, R2=.02,F(2, 271) = 2.31, p=.10; β1=−.09, p=.12;
β2=.05,p= .16 (Table 3). The interrupted regression model took the same form as the
multiple regression model and revealed no significant U-shaped relationship (β1–2=.47,p=
.18; β3–6=−.15, p= .047). Thus, we revealed that state anxiety difference scores tend to vary
with quiz number as indicated by the significant main effect (see Fig. 4c).
Anxiety from Pre-quiz to Post-quiz
To test our hypothesis that state anxiety would decrease from pre- to post-quiz, we conducted a
paired samples t-test comparing pre-quiz anxiety and post-quiz anxiety averaged across all six
quizzes. The t-test was significant, confirming that pre-quiz anxiety (M=5.97,SD = 1.82) was
significantly higher than post-quiz anxiety (M=4.92,SD =1.78);t(47) = 5.39, p<.001,
Cohen’sd=.78.
Anxiety and Quiz Performance
We first computed Pearson’s correlations between overall quiz performance and measures of
trait anxiety, overall state anxiety, pre-quiz anxiety, and post-quiz anxiety. The correlations
between quiz performance and trait anxiety, r(46) =−.008, p=.96, quiz performance and
overall state anxiety, r(46) =−.28, p=.05, and quiz performance and pre-quiz anxiety, r(46) =
−.10, p=.49, were not significant. The correlation between quiz performance and post-quiz
anxiety was significant, r(46) =−.42, p=.003, thus indicating that students’quiz performance
varied inversely with their anxiety after taking each quiz.
Due to the possibility that Pearson’s correlations between overall quiz performance and
measures of state anxiety could be inflated due to the overlap of their variances, we computed
partial Pearson’s correlations that held trait anxiety constant. The partial correlations between
quiz performance and overall state anxiety, r(45) = −.32, p= .03, and post-quiz anxiety, r(45)
=−.46, p= .001, were significant, while the partial correlation between quiz performance and
pre-quiz anxiety, r(45) = −.11, p= .45, was not significant. These analyses suggest, rather, that
when holding trait anxiety constant, students’quiz performance varied inversely with both
their overall state anxiety and post-quiz anxiety.
Anxiety and Estimation Accuracy
We first computed Pearson’s correlations between overall score calibrations and measures of trait
anxiety, overall state anxiety, pre-quiz anxiety, and post-quiz anxiety. The correlation between
overall score calibrations and trait anxiety was significant, r(46) =.36, p=.01, indicating that
metacognitive accuracy varied with levels of trait anxiety; however, the correlations between
Educational Psychology Review

overall score calibrations and overall state anxiety, r(46) =.16, p=.28, pre-quiz anxiety, r(46) =
.19, p=.19, and post-quiz anxiety, r(46) =.10, p=.51, were not significant.
Due to the possibility that Pearson’s correlations between overall score calibrations and
measures of state anxiety could be inflated due to the overlap of their variances, we computed
partial Pearson’s correlations that held trait anxiety constant. None of the partial correlations
between overall score calibrations and overall state anxiety, r(45) = −.02, p= .87, pre-quiz
anxiety, r(45) = .01, p= .93, and post-quiz anxiety, r(45) = −.06, p= .71, were significant.
Thus, when holding trait anxiety constant, we no longer find significant relationships between
any measure of anxiety and estimation accuracy.
Lastly, we examined the relationship between state anxiety and metacognition and how
they might interact to influence overall quiz performance. Table 3illustrates the different
models that were tested. We found there to be a significant moderation effect such that the
effect of post-quiz anxiety on quiz performance seems to significantly depend on participants’
metacognitive prediction calibration, R2=.39,F(4, 269) = 43.40, p<.0001;βint =.03,p=.01.
We also found there to be a significant moderation effect such that the effect of post-quiz
anxiety on quiz performance seems to significantly depend on participants’metacognitive
postdiction calibration, R2= .32, F(4, 269) = 31.71, p< .0001; βint =−.05, p= .01.
Consequently, there was also a significant moderation effect of post-quiz anxiety on quiz
performance such that it significantly depends on the metacognitive calibration difference
score, R2=.18,F(4, 269) = 14.75, p< .0001; βint =−.05, p= .02. No other interactions were
significant, all ps > .06 (see Table 3).
Study 1 Results Summary
In study 1, quiz performance increased with each subsequent quiz and post-quiz state anxiety
was a predictor of quiz performance, while pre-quiz state anxiety and trait anxiety were not
related to performance. Metacognitive pre- and postdictions did not vary with quiz when
controlling for state and trait anxiety. Metacognitive pre- and postdiction calibration increased
across the quarter when controlling for state and trait anxiety, and post-quiz anxiety was also a
predictor of postdiction calibration. Thus, while students’performance increased throughout
the quarter, they became increasingly more underconfident with each subsequent quiz indi-
cating that they were underestimating their performance. Overall state anxiety decreased with
quiz number, but this varied with trait anxiety, such that students with higher trait anxiety had
higher overall state anxiety. Finally, the effect of post-state anxiety on performance was
moderated by pre- and postdiction calibrations and pre-postdiction difference scores (Table 4).
Study 2
Method
Participants
Participants consisted of 299 UCLA undergraduate students enrolled in Psychology 120A:
Cognitive Psychology during Spring Quarter 2018 (92 students were excluded for missing
data and 4 students were excluded as outliers: 203 students included in the final sample.
Excluded outliers ±2.5 SDs from Z-score for overall score calibration).
Educational Psychology Review

Materials and Procedure
There were two, non-cumulative exams administered during weeks 5 and 10. Exams
were 60 multiple-choice questions based on the previous four weeks’worth of
information. Scores ranged from 0 to 60, and exams accounted for 80% of the final
grade (each exam worth 40%). An example of the type of multiple-choice question a
student may have answered on an exam would be of the following form and level of
conceptual understanding: “Which of the following tasks is least appropriate as a
means of testing implicit memory?”There was also a written assignment, research
proposal, and in-class presentation that factored into the final grade. The procedure for
each exam was the same as the procedure for each quiz administered in study 1 and
the same single-report scales were used (pre-quiz anxiety measure, prediction, test,
post-anxiety measure, postdiction). The only difference was that trait anxiety was
measured after each of the two exams. Students were informed that their participation
in the surveys would not affect their test grades.
Similar to study 1, assigned course teaching assistants graded all assignments for
the course and were blind to the hypotheses and predictions of the current study, as
well as students’metacognitive ratings and reported anxiety measures. Additionally,
all materials and procedures were performed ethically and approved by the UCLA
Institutional Review Board.
Table 4 How metacognition moderates effects of state anxiety on performance in study 1
Model % variance explained R2Fdf Intercept β1β2β3βint
A 33.3% .34**** 35.80 (4, 275) 8.69**** .14*** .26* −.18*** .02
B 29.1% .30**** 29.66 (4, 275) 8.32**** .15*** .23* −.11** .02
C 37.7% .39**** 42.36 (4, 269) 8.63**** .13*** .35** −.18**** .01
D 36.4% .37**** 40.98 (4, 275) 8.60**** .15*** .16 −.17**** .03
E 33.3% .34**** 35.75 (4, 275) 8.27**** .15**** .24** −.11*** .02
F 38.3% .39**** 43.40 (4, 269) 8.57**** .14*** .14 −.18**** .03*
G 21.1% .22**** 19.64 (4, 275) 8.69**** .17**** .29** −.15**** −.01
H 16.2% .17**** 14.43 (4, 275) 8.19**** .18**** .22* −.06 −.004
I 31.0% .32**** 31.71 (4, 269) 8.86**** .12** .60**** −.19**** −.05*
J 12.1% .13**** 10.59 (4, 275) 8.44**** .19**** .12 −.08 −.03
K 8.48% .10**** 7.46 (4, 275) 8.10**** .20**** .04 −.02 −.02
L 16.8% .18**** 14.75 (4, 269) 8.49**** .19**** .14 −.09* −.05*
*p<.05, **p<.01,***p< .001, ****p<.0001
Significant interactions are bolded. Model A: Quiz score = β0+β1(quiz number) + β2(overall calibration) + β3
(overall state anxiety) + βint (β0*β3). Model B: Quiz score = β0+β1(quiz number) + β2(overall calibration) +
β3(pre-quiz state anxiety) + βint (β2*β3). Model C: Quiz score = β0+β1(quiz number) + β2(overall
calibration) + β3(post-quiz state anxiety) + βint (β2*β3). Model D: Quiz score = β0+β1(quiz number) + β2
(pre-quiz calibration) + β3(overall state anxiety) + βint (β2*β3). Model E: Quiz score = β0+β1(quiz number) +
β2(pre-quiz calibration) + β3(pre-quiz state anxiety) + βint β2*β3). Model F: Quiz score = β0+β1(quiz
number) + β2(pre-quiz calibration) + β3(post-quiz state anxiety) + βint (β2*β3). Model G: Quiz score = β0+β1
(quiz number) + β2(post-quiz calibration) + β3(overall state anxiety) + βint (β2*β3). Model H: Quiz score = β0
+β1(quiz number) + β2(post-quiz calibration) + β3(pre-quiz state anxiety) + βint (β2*β3).ModelI:Quizscore
=β0+β1(quiz number) + β2(post-quiz calibration) + β3(post-quiz state anxiety) + βint (β2*β3). Model J: Quiz
score = β0+β1(quiz number) + β2(calibration difference score) + β3(overall state anxiety) + βint (β2*β3).
Model K: Quiz score = β0+β1(quiz number) + β2(calibration difference score) + β3(pre-quiz state anxiety) +
βint β2*β3). Model L: Quiz score = β0+β1(quiz number) + β2(calibration difference score) + β3(post-quiz
state anxiety) + βint (β2*β3)
Educational Psychology Review

Results and Discussion
Exam Performance
In order to determine if students performed differently between exam 1 and exam 2, we
conducted a paired samples t-test between students’scores for each exam. Scores improved
from exam 1 (M=49.67, SD = 5.91) to exam 2 (M=51.90, SD = 4.92), t(206) = 6.67, p<
.001, Cohen’sd=.46.
Estimation Accuracy
We conducted a 2 (Exam: midterm, final)×2(Calibration:pre-exam, post-exam)repeated-
measures ANOVA on overall score calibration to examine how well students metacognitively
estimated their scores before and after each exam (see Fig. 5). There was a significant main
effect of exam such that overall score calibration was significantly more positive (i.e., more
underconfident) for exam 2 (M=2.13, SD = 5.61) compared to exam 1 (M=.33, SD = 5.97),
regardless of whether these metacognitive estimations of accuracy were provided before or
after each exam, F(1, 202) = 13.56, p<.001,η2= .04. These analyses also revealed a
Fig. 5 Mean overall metacognitive score calibration (actual score −predicted/postdicted score) measures for
both exams in study 2. More positive scores indicate higher underconfidence. All error bars represent ± 1
standard error
Educational Psychology Review

significant main effect of calibration such that postdiction score calibration (M=1.83, SD =
4.59) was significantly more underconfident than was prediction score calibration (M=.63,
SD = 5.32), F(1, 202) = 22.42, p<.001, η2= .02. Finally, there was a significant interaction
between exam and calibration, F(1, 202) = 9.60, p=.002, η2= .005. Follow-up post hoc
independent samples t-tests with a Bonferroni correction revealed there to be significant
differences between prediction calibration for exam 1 (M=.18, SD = 7.64) and exam 2 (M
=−2.20, SD = 7.58), between pre- (M=.18, SD = 7.64) and postdiction (M=−1.69, SD =
7.32) calibration for exam 1, and between the prediction calibration for exam 1 (M=.18, SD =
7.64) and the postdiction calibration for exam 2 (M=−3.00, SD = 6.84) (adjusted ps<.001).
Anxiety Across Exams
In order to determine whether or not measures of anxiety remained relatively constant from
exam 1 to exam 2, we conducted multiple paired samples t-tests between trait anxiety, pre-
exam anxiety, and post-exam anxiety for exam 1 and exam 2. We found no significant
differences for any measure of anxiety between exam 1 and exam 2, indicating that students
were no more anxious for the final exam compared to the midterm exam: trait anxiety at exam
1(M=6.22, SD = 2.36) vs. exam 2 (M=6.37, SD = 2.34), t(206) = 1.66, p=.10,Cohen’sd
=.12; pre-exam anxiety for exam 1 (M=6.28, SD = 2.18) vs. exam 2 (M=6.50, SD = 2.39),
t(206) = 1.44, p= .15, Cohen’sd=.10; post-exam anxiety for exam 1 (M=5.85, SD = 2.39)
vs. exam 2 (M=5.83, SD = 2.44), t(206) = .12, p=.91,Cohen’sd=.01.
Anxiety from Pre- to Post-exam
To test our hypothesis that anxiety would decrease from pre- to post-exam, we conducted a
paired samples t-test comparing pre-exam anxiety and post-exam anxiety averaged across both
exams. The t-test was significant confirming that pre-exam anxiety (M= 6.35, SD = 2.00) was
significantly higher than post-exam anxiety (M=5.80,SD = 2.18); t(202) = 4.50, p<.001,
Cohen’sd=.32.
Anxiety and Exam Performance
Similar to study 1, we first computed Pearson’s correlations between overall exam perfor-
mance and measures of trait anxiety, overall state anxiety, pre-exam anxiety, and post-exam
anxiety. The correlations between exam performance and trait anxiety, r(201) =−.20, p=
.004, overall state anxiety, r(201) =−.32, p<.001, pre-exam anxiety, r(201) =−.25, p<.001,
and post-exam anxiety, r(201) =−.33, p< .001, were all significant, thus indicating that
students’exam performance varied inversely with all measures of their anxiety before and after
taking each exam.
Then, due to the possibility that Pearson’s correlations between overall exam
performance and measures of state anxiety could be inflated due to the overlap of
their variances, we computed partial Pearson’s correlations holding trait anxiety
constant. All partial correlations between overall exam performance and overall state
anxiety, r(200) = −.26, p< .001, pre-exam anxiety, r(200) = −.17, p= .01, and post-
exam anxiety, r(200) = −.27, p< .0001, were significant, thus indicating that
students’exam performance varied inversely with all measures of their
state anxiety before and after taking each exam when holding trait anxiety constant.
Educational Psychology Review

Anxiety and Estimation Accuracy
We also computed Pearson’s correlations between overall score calibrations and measures of
trait anxiety, overall state anxiety, pre-exam anxiety, and post-exam anxiety. None of the
correlations between overall score calibrations and trait anxiety, r(201) = .02, p= .73, overall
state anxiety, r(201) = .07, p= .32, pre-exam anxiety, r(201) = .09, p= .19, and post- exam
anxiety, r(201) = .04, p= .59, were significant.
Then, due to the possibility that Pearson’s correlations between overall score
calibrations and measures of state anxiety could be impacted by trait anxiety, we
computed partial Pearson’s correlations holding trait anxiety constant, and found that
none of the partial correlations between overall score calibrations and overall state
anxiety, r(200) = .07, p= .33, pre-exam anxiety, r(200) = .09, p= .18, and post-
exam anxiety, r(200) = .03, p= .67, were significant. Thus, when holding trait
anxiety constant, we still do not find significant relationships between any measure
of anxiety and estimation accuracy.
Study 2 Results Summary
In study 2, higher state anxiety (pre- and post-quiz and overall) was associated with lower
exam performance when controlling for trait anxiety. Neither state nor trait anxiety impacted
the estimation accuracy of performance for participants. Students also became more
underconfident from exam 1 to exam 2 while their scores increased, indicating that students
were metacognitively unaware of their performance gains.
General Discussion
The goal of the current research was to explore the effects of test anxiety on undergraduate
students’overall performance and metacognitive accuracy and how the frequency and stakes
of the assessment influence those effects. In both studies, we found that state anxiety decreased
from before to after assessments, and overall score calibration increased throughout the
academic quarter (see Table 5for a summary of the convergence and divergence of results
from study 1 to study 2). Additionally, higher post-exam state anxiety was associated with
worse assessment performance. Although students were less anxious once they completed the
quiz or exam, they did not recognize that their performance increased. Students were more
anxious after the exam when they knew they had performed poorly (or were not anxious
knowing that they had done well).
In study 1, we found there to be a main effect of quiz number on pre- and post-quiz state
anxiety; however, regression analyses controlling for trait anxiety revealed there to be no
significant linear relationship between quiz number and pre/post-quiz state anxiety. Overall,
students appeared to be more anxious at the beginning of the quarter relative to the middle of
the quarter, with state anxiety generally stabilizing after acclimation to the course. Accord-
ingly, since pre- and post-state anxiety did not decrease with quiz in study 1, this further
suggests that the impact of frequent, low-stakes quizzes on state anxiety may vary with
individual differences in trait anxiety. Similarly, pre-exam state anxiety was constant across
the two exams (middle and end of the term) in study 2.
Educational Psychology Review

In study 1, trait anxiety varied positively with overall score calibration, but was not
associated with quiz performance; however, in study 2, higher trait anxiety was associated
with poorer exam performance, but not with estimation accuracy. Specifically, we found that
students’metacognitive predictions increased throughout the term in both studies, in addition
to overall performance increasing throughout the term for both classes. Additionally, in study
1, overall score calibrations increased with each subsequent quiz when controlling for both
state and trait anxiety, which illustrates that students became more underconfident with
practice even though their performance increased across the quarter. Similarly, students
became more underconfident from exam 1 to exam 2 in study 2, despite their increase in
performance.
Metacognitive postdictions increased throughout the term in study 1 but remained un-
changed in study 2. Though, this increase in pre-/postdictions in study 1 can be accounted for
by holding trait anxiety constant in analyses. When considering individual differences in trait
anxiety, pre-postdictions stayed constant across the quarter suggesting that metacognitive
predictions are related to trait anxiety. Additionally, prediction calibration increased from
exam 1 to exam 2 in study 2, and both prediction calibration and postdiction calibration
increased throughout the term in study 1; trait anxiety was a predictor of this increase, but state
anxiety was not related to this increase. This is consistent with our findings that trait anxiety
seems to be related to metacognition when taking frequent, low-stakes quizzes. Though,
postdiction calibration did not significantly change throughout the term in study 2 while
overall score calibration did increase throughout the term for both courses. As such, this
observed change in metacognitive performance throughout the term in both studies reveals a
potential “underconfidence with practice effect”(Koriat et al., 2002), such that students
became more underconfident throughout the quarter upon completing more assessments.
Students in study 1 with higher trait anxiety exhibited higher metacognitive
underconfidence when taking the more frequent, lower-stakes quizzes, though this trait anxiety
was not related to the students’overall performance. Yet, when controlling for trait anxiety,
there does not appear to be a significant relationship between state anxiety and metacognitive
accuracy. Conversely, trait anxiety was not related to students’metacognitive accuracy in
study 2, despite its relationship with overall performance. These results reveal a critical
implication for determining testing practices that are more inclusive to individuals with higher
levels of baseline anxiety. The current findings are compelling in that they reveal the
importance of considering students’levels of test anxiety to motivate the implementation of
Table 5 Comparison of the main findings across studies
Results summary Study 1 Study 2
Assessment performance1++
Overall calibration1++
Pre-assessment state anxiety1−No change
Post-assessment state anxiety1−No change
Changes in state anxiety2−−
Overall state anxiety and performance3−−
“+”indicates an increase across time points or a positive correlation while “−”indicates a decrease across time
points or a negative correlation
1Throughout the quarter
2From pre-assessment to post-assessment
3Partial Pearson’s correlation controlling for trait anxiety
Educational Psychology Review

more frequent, lower-stakes quizzes in courses. A primary goal for educators and those
designing more modular and adaptable course curricula at all levels of education should be
to consider the various intrinsic (Deci et al., 1999) and extrinsic (Ryan & Weinstein, 2009)
factors that may perpetually enhance (Putwain et al., 2012; Vansteenkiste et al., 2004)or
decline (Ratelle et al., 2007) the students’educational outcomes in the classroom.
To further examine how metacognitive accuracy influences the relationship between state
anxiety and performance, we ran a series of regression models, which revealed that
metacognitive accuracy was a moderator of the effects of post-quiz state anxiety on perfor-
mance in study 1. Specifically, pre- and postdiction calibrations and pre-postdiction calibration
difference scores moderated the relationship between post-quiz state anxiety on performance
(see Table 4).
Thus, it seems that although we found that higher post-quiz state anxiety is related to lower
performance outcomes, this varies with prediction accuracy, postdiction accuracy, and differ-
ences between pre- and postdiction measures. Individual differences in metacognitive abilities
seem to be related to the effects of post-state anxiety on performance when taking more
frequent, lower-stakes quizzes.
We should also note that although many of our regression models were significant in study
1, R2values were relatively small, explaining between 3 and 17% of the variance amongst the
data. The implications of these results indicate an inherent amount of unexplainable variability
within our dataset, most likely due to the practical considerations that our sample was
constrained to a real, classroom setting comprised of enrolled students in the course, as
opposed to a controlled, randomly sampled in-lab group of recruited participants.
More broadly, integrating optimized tools and techniques to develop more equitable and
customized approaches for different students to achieve learning outcomes, regardless of their
background and preparation, is a critical consideration that must be accounted for in future
course planning. For instance, students from lower socioeconomic backgrounds may already
be prone to higher levels of stress and anxiety (Lantz et al., 2005), which may further instigate
higher levels of test anxiety in these individuals. Additionally, there are intergenerational
effects of certain academic anxieties (Maloney et al., 2015) such that students of parents with
math anxiety experience lower math achievement when their parents have been highly
involved in their math development. These considerations are essential when designing
assessments that provide optimal and equitable conditions for learning achievement. In fact,
it is vital to consider students’diverse backgrounds at all stages of the test cycle as learners
with high test anxiety have been shown to have deficits in encoding, organization, and storage
of learned information (Mueller, 1980). These deficits during the test preparation phase often
lead to poor test performance for students with high test anxiety such that anxiety prompts
interference, distractibility, or inefficient cue utilization strategies (Cassady, 2004; Schwarzer
& Jerusalem, 1992). During the test reflection phase, learners typically form attributional
biases, which can lead to negative future attitudes and behavior patterns towards testing. If a
student attributes their failure on the assessment to a lack of ability, they are more likely to
avoid proper test preparation activities for future tests (Elliot & McGregor, 1999).
Some limitations of the current study include only investigating student anxiety both
immediately before and after the assessment without considering other factors that could
potentially explain why these effects were observed. More specifically, we did not consider
how the differences in low-stakes quizzes may prompt students to adjust their study practices
throughout the term (study 1) compared to a different class with a completely different
assessment structure (study 2). Therefore, future research should examine how students’
Educational Psychology Review

individual levels of preparation for various types of assessments mediate their levels of post-
quiz anxiety and performance. Additionally, future classroom-based studies examining test
anxiety should seek to collect a variety of demographic factors, along with reports and
measures of students’efforts and strategies to prepare for quizzes and exams, to subsequently
compare differences in test anxiety both within and across groups of students at all stages of
the testing cycle. These findings, when considered with students’individual circumstances,
may provide a framework for designing more equitable assessments that take into account
factors potentially interacting with or provoking students’test anxiety. It is also worth noting
that although single-report measures were most practical for this study, future research should
consider these relationships between characteristics of assessments (i.e., format, frequency,
and stakes) and state and trait anxiety while using validated scales to measure individual
differences in different types of test anxiety.
In conclusion, the findings of the current study suggest that less-frequent, higher-stakes
assessments may trigger students’baseline levels of anxiety in that students may inherently
feel more pressure to perform better on these exams compared to testing environments with
more-frequent, lower-stakes quizzes, as each individual assessment would not necessarily have
as substantial of an impact on a student’s overall course grade comparatively. This finding
complements prior work showing that students are overwhelmed with stress, anxiety, and
worry due to testing in high-stakes contexts (e.g., Segool et al., 2013; Triplett & Barksdale,
2005).
Though, students’feelings regarding testing seem to be inconsistent across studies exam-
ining the relationship between test anxiety and the relative weight of an assessment (Mulvenon
et al., 2001;Mulvenonetal.,2005; Putwain, 2008). Our findings, therefore, contribute to these
prior equivocal results in the literature that prompted us to further investigate how the
relationship between test anxiety and the stakes of an assessment vary with one another, albeit
including an examination of how test anxiety interacts with students’metacognitive accuracy.
Acknowledgements We thank Gerardo Ramirez, Barbara Knowlton, Robert Bjork, Elizabeth Bjork, and
Courtney Clark for helpful comments regarding this work. We also thank Mary Hargis, Dillon Murphy, Julia
Schorn, Mary Whatley, and the rest of the members of the Memory and Lifespan Cognition Laboratory at UCLA
for their guidance and support.
Code Availability No custom code or software were written for this project. Code to run analyses in R
are provided in the OSF link below.
Author Contribution A. D. Castel and A. L. M. Siegel conceived of the study. A. L. M. Siegel collected the
data, and S. T. Schwartz analyzed the data with A. L. M. Siegel and K. M. Silaj. K. M. Silaj and S. T. Schwartz
wrote the manuscript. A. L. M. Siegel and A. D. Castel contributed to the editing of the manuscript.
Funding This research was supported in part by the National Institutes of Health (National Institute on Aging;
Award Number R01 AG044335 to Dr. Alan Castel).
Data Availability Data and survey materials are available via OSF (https://osf.io/pv89n/).
Declarations
Conflict of Interest The authors declare no competing interests.
Educational Psychology Review

References
Argarwal, P. K., D’Antonio, L., Roediger, H. L. I., McDermott, K. B., & McDaniel, M. A. (2014). Classroom-
based programs of retrieval practice reduce middle school and high school students’test anxiety. Journal of
Applied Research in Memory and Cognition, 3(3), 131–139.
Birenbaum, M., & Feldman, R. A. (1998). Relationships between learning patterns and attitudes towards two
assessment formats. Educational Research, 40(1), 90–98.
Bjork, R. A. (1994). Memory and metamemory considerations in the training of human beings. In J. Metcalfe &
A. P. Shimamura (Eds.), Metacognition: knowing about knowing (pp. 185–205). MIT Press.
Bowman, A. W., Jones, M. C., & Gijbels, I. (1998). Testing monotonicity of regression. Journal of
Computational and Graphical Statistics, 7(4), 489–500.
Callender, A. A., Franco-Watkins, A. M., & Roberts, A. S. (2016). Improving metacognition in the classroom
through instruction, training, and feedback. Metacognition and Learning, 11(2), 215–235.
Cassady, J. C. (2004). The impact of test anxiety on text comprehension and recall in the absence of external
evaluative pressure. Applied Cognitive Psychology, 18(3), 311–325.
Cassady, J. C. (2010). Test anxiety: contemporary theories and implications for learning. In J. C. Cassady (Ed.),
Anxiety in schools: the causes, consequences, and solutions for academic anxieties (pp. 7–26). Peter Lang.
Cassady, J. C., & Johnson, R. E. (2002). Cognitive test anxiety and academic performance. Contemporary
educational anxiety, 27,270–295.
Cizek, G. J., & Burg, S. S. (2006). Addressing test anxiety in a high-stakes environment: strategies for
classrooms and schools.CorwinPress.
Covington, M. V. (1985). Test anxiety: causes and effects over time. In H. M. van der Ploeg, R.
Schwarzer, & C. D. Spielberger (Eds.), Advances in test anxiety research (Vol. 4, pp. 55–68).
Swets & Zeitlinger.
Davey, H. M., Barratt, A. L., Butow, P. N., & Deeks, J. J. (2007). A one-item question with a Likert or Visual
Analog Scale adequately measured current anxiety. Journal of Clinical Epidemiology, 60(4), 356–360.
Deci, E. L., Koestner, R., & Ryan, R. M. (1999). A meta-analytic review of experiments examining the effects of
extrinsic rewards on intrinsic motivation. Psychological Bulletin, 125(6), 627–668.
Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Improving students’
learning with effective learning techniques: promising directions from cognitive and educational psycholo-
gy. Psychological Science in the Public Interest, 14(1), 4–58.
Dunn, D. S., Saville, B. K., Baker, S. C., & Marek, P. (2013). Evidence-based teaching: tools and techniques that
promote learning in the psychology classroom. Australian. Journal of Psychology, 65(1), 5–13.
Elliot, A. J., & McGregor, H. A. (1999). Test anxiety and the hierarchical model of approach and avoidance
achievement motivation. Journal of Personality and Social Psychology, 76(4), 628–644.
Everson, H. T., Smodlaka, I., & Tobias, S. (1994). Exploring the relationship of test anxiety and metacognition
on reading test performance: a cognitive analysis. Anxiety, stress, and coping, 7(1), 85–96.
Fernández-Castillo, A., & Caurcel, M. J. (2015). State test-anxiety, selective attention and concentration in
university students. International Journal of Psychology, 50(4), 265–271.
Goetz, T., Bieg, M., Lüdtke, O., Pekrun, R., & Hall, N. C. (2013). Do girls really experience more anxiety in
mathematics? Psychological Science, 24(10), 2079–2087.
Harris, R. B., Grunspan, D. Z., Pelch, M. A., Fernandes, G., Ramirez, G., & Freeman, S. (2019). Can test anxiety
interventions alleviate a gender gap in an undergraduate STEM course? CBE—Life Sciences Education, 18,
1–9.
Hembree, R. (1988). Correlates, causes, effect, and treatment of test anxiety. Review of Educational Research,
58(1), 47–77.
Karpicke, J. D. (2017). Retrieval-based learning: a decade of progress. In J. T. Wixted (Ed.), Cognitive
psychology of memory, of learning and memory: a comprehensive reference (Vol. 2, pp. 487–514).
Oxford: Academic Press.
Koriat, A., Sheffer, L., & Ma’ayan, H. (2002). Comparing objective and subjective learning curves: judgments of
learning exhibit increased underconfidence with practice. Journal of Experimental Psychology: General,
131(2), 147–162.
Kurosawa, K., & Harackiewicz, J. M. (1995). Test anxiety, self-awareness, and cognitive interference: a process
analysis. Journal of Personality, 63(4), 931–951.
Lantz, P. M., House, J. S., Mero, R. P., & Williams, D. R. (2005). Stress, life events, and socioeconomic
disparities in health: results from the Americans’Changing Lives Study. Journal of Health and Social
Behavior, 46(3), 274–288.
Educational Psychology Review

Leadbeater, B., Thompson, K., & Gruppuso, V. (2012). Co-occurring trajectories of symptoms of anxiety,
depression, and oppositional defiance from adolescence to young adulthood. Journal of Clinical Child &
Adolescent Psychology, 41(6), 719–730.
Legg, A. M., & Locker, L. (2009). Math performance and its relationship to math anxiety and metacognition.
North American Journal of Psychology, 11,471–486.
Maloney, E. A., Ramirez, G., Gunderson, E. A., Levine, S. C., & Beilock, S. L. (2015). Intergenerational effects
of parents’math anxiety on children’s math achievement and anxiety. Psychological Science, 26(9), 1480–
1488.
Mueller, J. H. (1980). Test anxiety and the encoding and retrieval of information. In I. G. Sarason (Ed.), Test
anxiety: theory, research, and applications (pp. 63–86). Erlbaum.
Mulvenon, S. W., Connors, J. V., & Lenares, D. (2001). Impact of accountability and school testing on students:
is there evidence of anxiety?.
Mulvenon, S. W., Stegman, C. E., & Ritter, G. (2005). Test anxiety: a multifaceted study on the perceptions of
teachers, principals, counselors, students, and parents. International Journal of Testing, 5(1), 37–61.
Naveh-Benjamin, M. (1991). A comparison of training programs intended for different types of test anxious
students: further support for an information-processing model. Journal of Educational Psychology, 83(1),
134–139.
Núñez-Peña, M. I., Suárez-Pellicioni, M., Guilera, G., & Mercadé-Carranza, C. (2013). A Spanish version of the
short Mathematics Anxiety Rating Scale (sMARS). Learning and Individual Differences, 24, 204–210.
Putwain, D. W. (2008). Deconstructing test anxiety. Emotional and Behavioural Difficulties, 13(2), 141–155.
Putwain, D. W., & Best, N. (2011). Fear appeals in the primary classroom: effects on test anxiety and test grade.
Learning and Individual Differences, 21(5), 580–584.
Putwain, D. W., Kearsley, R., & Symes, W. (2012). Do creativity self-beliefs predict literacy achievement and
motivation? Learning and Individual Differences, 22(3), 370–374.
Ratelle, C. F., Guay, F., Vallerand, R. J., Larose, S., & Senécal, C. (2007). Autonomous, controlled, and
amotivated types of academic motivation: a person-oriented analysis. Journal of Educational Psychology,
99(4), 734–746.
Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: taking memory tests improves long-term
retention. Psychological Science, 17(3), 249–255.
Ryan, R. M., & Weinstein, N. (2009). Undermining quality teaching and learning: a self- determination theory
perspective on high-stakes testing. Theory and Research in Education, 7(2), 224–233.
Schwarzer, R., & Jerusalem, M. (1992). Advances in anxiety theory: a cognitive process approach. In K. A.
Hagtvet & T. B. Johnsen (Eds.), Advances in test anxiety research (Vol. 7, pp. 2–31). Swets & Zeitlinger.
Segool, N. K., Carlson, J. S., Goforth, A. N., von der Embse, N., & Barterian, J. A. (2013). Heightened test
anxiety among young children: elementary school students’anxious responses to high-stakes testing.
Psychology in the Schools, 50(5), 489–499.
Simonsohn, U. (2018). Two lines: a valid alternative to the invalid testing of U-shaped relationships with
quadratic regressions. Advances in Methods and Practices in Psychological Science, 1(4), 538–555.
Spada, M. M., Nikcevic, A. V., Moneta, G. B., & Ireson, J. (2006). Metacognition as a mediator of the effect of
test anxiety on a surface approach to studying. Educational Psychology, 26(5), 615–624.
Spielberger, C. (1972). Anxiety as an emotional state. In C. D. Spielberger (Ed.), Anxiety: current trends in
theory and research (Vol. 1, pp. 23–49). Academic Press.
Triplett, C. F., & Barksdale, M. A. (2005). Third through sixth graders’perceptions of high- stakes testing.
Journal of Literacy Research, 37(2), 237–260.
Vansteenkiste, M., Simons, J., Lens, W., Sheldon, K. M., & Deci, E. L. (2004). Motivating learning, perfor-
mance, and persistence: the synergistic effects of intrinsic goal contents and autonomy-supportive contexts.
Journal of Personality and Social Psychology, 87(2), 246–260.
Veenman, M. V. J., Kerseboom, L., & Imthorn, C. (2000). Test anxiety and metacognitive skillfulness:
availability versus production deficiencies. Anxiety, Stress, and Coping, 13(4), 391–412.
von der Embse, N., & Hasson, R. (2012). Test anxiety and high-stakes test performance between school settings:
implications for educators. Preventing school failure: alternative education for children and youth, 56(3),
180–187.
von der Embse, N., Jester, D., Roy, D., & Post, J. (2018). Test anxiety effects, predictors, and correlates: a 30-
year meta-analytic review. Journal of Affective Disorders, 227,483–493.
Williams, J. E. (1991). Modeling test anxiety, self-concept and high school students’academic achievement.
Journal of Research & Development in Education, 25(1), 51–57.
Wood, S. G., Hart, S., Little, S., & Phillips, S. A. (2016). Test anxiety and a high-stakes standardized reading
comprehension test: a behavioral genetics perspective. Merrill-Palmer Quarterly, 62(3), 233–251.
Educational Psychology Review

Yang, C., Sun, B., Potts, R., Yu, R., Luo, L., & Shanks, D. R. (2020). Do working memory capacity and test
anxiety modulate the beneficial effects of testing on new learning? Journal of Experimental Psychology:
Applied Advance online publication.
Yue, X. (1996). Test anxiety and self-efficacy: Levels and relationship among secondary school students in Hong
Kong. Psychologia: An International Journal of Psychology in the Orient, 39(3), 193–202.
Zohar, D. (1998). An additive model of test anxiety: role of exam-specific expectations. Journal of Educational
Psychology, 90(2), 330–340.
Publisher’sNote Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.
Educational Psychology Review