SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Background
  4. Method
  5. Results
  6. Discussion
  7. Acknowledgement
  8. References

Cognitive interference theories (e.g. attentional control theory, processing efficiency theory) suggest that high levels of trait anxiety predict adverse effects on the performance of cognitive tasks, particularly those that make high demands on cognitive resources. We tested an interaction hypothesis to determine whether a combination of high anxiety and low working memory capacity (WMC) would predict variance in demanding cognitive test scores. Ninety six adolescents (12- to 14-years-old) participated in the study, which measured self-report levels of trait anxiety, working memory, and cognitive test performance. As hypothesized, we found that the anxiety-WMC interaction explained a significant amount of variance in cognitive test performance (ΔR2 .07, < .01). Trait anxiety was unrelated to cognitive test performance for those adolescents with average WMC scores (β = .13, > .10). In contrast, trait anxiety was negatively related to test performance in adolescents with low WMC (β = −.35, < .05) and positively related to test performance in those with high WMC (β = .49, < .01). The results of this study suggest that WMC moderates the relationship between anxiety and cognitive test performance and may be a determinant factor in explaining some discrepancies found in the literature. Further research is needed to fully understand the mechanisms involved.


Background

  1. Top of page
  2. Abstract
  3. Background
  4. Method
  5. Results
  6. Discussion
  7. Acknowledgement
  8. References

Evidence from a range of sources shows that cognitive test performance can be reduced by trait anxiety in adults and children (Ackerman & Heggestad, 1997; Eysenck, Derakshan, Santos, & Calvo, 2007; Hembree, 1988; Ma, 1999; Owens, Stevenson, Norgate, & Hadwin, 2008). The consequences of elevated or clinical levels of anxiety on cognitive test performance can be far-reaching leading to poor educational outcomes that include early school-leaving and failure to enter college or university (Andrews & Wilding, 2004; Breslau, Lane, Sampson, & Kessler, 2008; Lee et al., 2009; Van Ameringen, Mancini, & Farvolden, 2003). Understanding more about the mechanisms that underpin associations between emotion and cognitive test performance is therefore particularly important to ensure a positive developmental outcome for children and adolescents who experience high levels of anxiety. One important factor to emerge from research in explaining this phenomenon is working memory.

Working memory refers to the cognitive ability of maintaining task-relevant information in complex cognition (Miyake & Shah, 1999). It is the ability to both hold in memory and manipulate information to produce an output. The components of working memory are instrumental in cognitive activity from an early age, and they develop throughout childhood (Marcovitch, Boseovski, Knapp, & Kane, 2010; Nevo & Breznitz, 2011; Tam, Jarrold, Baddeley, & Sabatos-DeVito, 2010). It has also been well established that working memory skills strongly predict test performance in school-aged children (Gathercole, Pickering, Knight, & Stegmann, 2004) and very low working memory capacity (defined as at or below the 10th percentile by age) is associated with poor educational outcome (Alloway, Gathercole, Kirkwood, & Elliott, 2009).

The interplay between anxiety and working memory has been described in theoretical accounts most notably by Eysenck and colleagues (Eysenck & Calvo, 1992; Eysenck & Derakshan, 2011; Eysenck et al., 2007). Attentional control theory (ACT) proposes that anxiety disrupts working memory processes leading to lowered cognitive performance in terms of reduced task efficiency and effectiveness, particularly on complex tasks. In support of this argument, recent evidence suggests that writing about test-specific worry prior to an exam may alleviate the burden on working memory thus boosting exam performance, particularly for high-anxious adolescents (Ramirez & Beilock, 2011). It has also been suggested that trait anxiety is associated with poor recruitment of pre-frontal cortex resources, providing a potential biological correlate and mechanism for the lack of attentional control in high-anxious individuals (Bishop, 2009), although see Eysenck and Derakshan (2011) for an alternative view on this issue. ACT also suggests that high-anxious individuals are motivated to improve their test performance due to the anxiety surrounding perceived failure and negative evaluation. This argument may account for findings where high-anxious individuals perform at equivalent levels to their low-anxious counterparts in terms of performance per se, but are less efficient in achieving parity (e.g. Hadwin, Brogan, & Stevenson, 2005). That is, increased motivation and effort drives high-anxious individuals to try to improve cognitive test performance.

Trait anxiety can disrupt working memory processes in adults (Eysenck & Calvo, 1992; Eysenck et al., 2007) and although the literature is still relatively sparse, several studies have now shown that working memory processes in children are also adversely affected by anxiety (Ng & Lee, 2010; Owens, Stevenson, Hadwin, & Norgate, 2012; Owens et al., 2008; Visu-Petra, Cheie, Benga, & Packiam Alloway, 2011). Nevertheless, the precise mechanisms and conditions under which negative effects of anxiety occur are still uncertain. In particular, it remains unclear how to account for the null findings in studies that test for a ‘main effect’ of anxiety on performance. For example, in a meta-analysis on the effect of anxiety on test performance, Seipp (1991) found a negative-weighted correlation between anxiety and test performance, yet the effect was heterogeneous, reflecting the fact that the study revealed null and even positive associations as well as the expected negative finding. Seipp points out that the variation suggests the presence of moderating factors in the data. However, few of those tested (gender, country, trait vs. state) proved to be significant and none revealed positive findings. A reliance on simple effects of anxiety is therefore problematic if there are in fact moderating factors involved in the anxiety-cognitive performance relationship. Currently, there is a dearth of studies examining the interplay between cognitive and emotional factors in the cognitive test performance of children and adolescents, although interest is increasing (Johnson & Gronlund, 2009; Valiente, Lemery-Chalfant, & Swanson, 2010).

Consistent with these theoretical and empirical views, we suggest that cognitive test performance will be lowest in adolescents if WMC is low and trait anxiety is high. Conversely, if WMC is high and anxiety is high, cognitive test performance should improve as a consequence of the additional motivation and drive to perform well; where this motivation to succeed can only be acted upon given sufficient cognitive resources. Given that prior research shows that anxiety may be particularly important in cognitive tests that require substantial working memory resources such as maths (Ashcraft & Krause, 2007; Ma, 1999) and IQ (Ackerman & Heggestad, 1997; Hembree, 1988), we hypothesized that an interaction between negative effect and WMC would be most associated with tests measuring these domains.

Method

  1. Top of page
  2. Abstract
  3. Background
  4. Method
  5. Results
  6. Discussion
  7. Acknowledgement
  8. References

Participants

Ninety six adolescents from three schools in the UK participated in the study (mean age = 13.4; SD = 0.66; range = 12–14; 52 males, 44 females).

Sampling procedure

Informed written parental consent was obtained in the first instance for each participant via each school. Subsequently, adolescents gave informed written assent before participating in the study. Ethical approval was given by the University ethics committee (id PG/03/96) and Research Governance Office. Participants were tested in school classrooms or libraries during the school day. Trait anxiety measurement and cognitive testing were made in small groups of between 8 and 10 participants. Participants were seen individually to administer the working memory test battery.

Measures

Anxiety

We used the Spielberger trait anxiety form (Spielberger, Edwards, Lushene, Montouri, & Platzek, 1973) to measure anxiety. This measure consists of 20 items and uses a 3-point (1 = almost never, 2 = sometimes, or 3 = often) Likert-type scale, where higher scores indicate higher levels of anxiety. A single score can be obtained by summing the scores on all items (scores range from 20 to 60). Example items include, ‘Unimportant thoughts run through my mind and bother me' and ‘My hands get sweaty’. Good internal consistency has been demonstrated for this measure, with Cronbach's alpha of .91 reported by others (Muris, Merckelbach, Ollendick, King, & Bogie, 2002) and .88 in the current sample.

Working memory capacity

To assess working memory capacity, we compiled a battery of tests using the automated working memory assessment (AWMA; Alloway, 2007) and the Cambridge automated neuropsychological test battery (Cambridge Neuropsychological Test Automated Battery (CANTAB), 2004). The CANTAB uses non-verbal tasks to measure a range of executive functions and has been validated for use with children (Luciana, 2003). We used the forwards and backwards versions of the spatial span test (score ranges = 2–9) on the CANTAB to tap into spatial WMC. The AWMA is comprised of 12 working memory tests and has shown good test–retest reliability (2 weeks) in a sample of 10- and 11-year-old children (Gathercole et al., 2004). The forwards (score range = 0–54) and backwards (score range = 0–36) digit recall tests were used from this assessment to measure verbal WMC.

Cognitive tests

We used two cognitively demanding tests recognized to involve working memory as dependent variables including the maths computation test of the wide range achievement test (WRAT 4; Wilkinson & Robertson, 2006) and the Raven's standard progressive matrices (SPM; Raven, Raven, & Court, 1998). The WRAT 4 is a well-established measure of educational attainment validated for a wide range of ages (5–94). In the maths test, participants are asked to solve as many maths problems of increasing difficulty as they can in 15 min. Scores range from 0 to 55. The SPM is designed to measure eductive reasoning and consists of 60 problems that involve analysing a series of spatial designs with a part missing. The participant must select the correct missing part from a number of options. The scores range from 0 to 60. Participants were given 20 min to complete as many matrices as possible.

Data analysis

Statistical analyses were carried out in Stata 11.1 (StataCorp, 2009) and the Mplus programme (Muthén & Muthén, 1998–2011). To estimate WMC, we calculated factor scores from the four working memory indicators using confirmatory factor analysis (CFA) with maximum likelihood estimation. Significant factor loadings and model fit indices (χ2 > .05, root mean square error of approximation (RMSEA) < .06 and the comparative fit index (CFI) > 0.95; Hu & Bentler, 1999) were used as criteria to initially assess the WMC model. To use factor scores in subsequent analyses, the measure of factor determinacy should be inspected where larger values indicate better measurement of the factor by the observed indicators. Factor determinacy values greater than .80 indicate that the factor scores can be very reliably used in subsequent analyses (Gorsuch, 1983). The resultant standardized factor scores have a mean of zero and a standard deviation of 1.

The two cognitive tests were used to form a standardized composite dependent variable by summing the z scores. To test the anxiety-WMC interaction on cognitive test performance, we used a hierarchical linear regression model entering the continuous predictors of anxiety and WMC in step 1, followed by the product of these as an interaction term in step 2.

The Gaussian distribution of standardized residuals was tested for each step of the hierarchical regression using post-hoc Kolmogorov–Smirnoff tests. To confirm that any associations detected were not caused by undue influence of individual scores, two further post-hoc diagnostic tests were performed; Cook's D, a measure of an individual observation's effect on overall fit and DFBETA, a measure of individual influence on a given beta. The interaction effects were subsequently adjusted for age, gender, and also time of testing to control for any fatigue effects that may have been present due to testing at different times of the school day.

Results

  1. Top of page
  2. Abstract
  3. Background
  4. Method
  5. Results
  6. Discussion
  7. Acknowledgement
  8. References

Means, standard deviations, range, and correlations among study variables are shown in Table 1.

Table 1. Means, standard deviations, range, and correlations between study variables
 Mean (SD); range1234567
Note
  1. fdigit, forward digit span (AWMA); bdigit, backward digit span (AWMA); fspat, forward spatial span (CANTAB); bspat, backward spatial span (CANTAB); anxiety, Spielberger trait anxiety; Maths, WRAT Math computation; Raven's, Raven's standard progressive matrices.

  2. *< .05; **p < .01; ***p < .001.

1. Age13.44 (0.66); 12–14       
2. Fdigit31.59 (5.94); 23–540.09      
3. Bdigit15.06 (5.21); 6–360.020.62***     
4. Fspat6.46 (1.41); 3–90.080.25*0.35***    
5. Bspat5.95 (1.38); 3–90.190.180.28**0.42***   
6. Anxiety33.5 (7.5); 19–520.090.090.170.050.05  
7. Maths37.99 (5.48); 24–520.100.31**0.45***0.22*0.30**0.05 
8. Raven's44.66 (6.35); 27–590.060.29**0.37**0.23*0.23*0.080.51***

Working memory

The working memory CFA provided a good fit to the data (χ2 = 0.03, df = 1, > .10, CFI = 1.00, RMSEA = .00) and all four loadings were significant (ps < .01). Importantly, the factor determinacy score of the working memory capacity scores was also high (0.96).

Hierarchical regression

Post-hoc tests showed that the standardized residuals followed a Gaussian distribution after step 1 (χ2 = 1.65, > .10) and step 2 (χ2 = 0.93, > .10). In addition, no Cook's D scores were >1 nor any DFBETAs >±1.

In step 1, there was a significant main effect of WMC on cognitive test performance (B = .87, = 5.23, < .001). Anxiety alone did not affect the test performance (B = −.00, = −0.10, > .10). The addition of the WMC × Anxiety interaction (Table 2) term in step 2 accounted for a significant amount of variance (ΔR2 = 0.06, ΔF2 = (1,92) = 8.36, < .01). In total, the model accounted for 30% of the variance. As a final step, we ran a regression model adjusting for age, gender, and also time of testing to control for any fatigue effects that may be present due to testing at different times of the school day. The interaction effect in this model remained significant after these adjustments (B = .06, = 2.89, < .01).To fully understand and simplify the interaction (Figure 1), we modelled the anxiety-cognitive performance relationship at different levels of WMC (holding WMC constant at the 33rd median and 66th percentiles). We probed the significant interaction by carrying out a simple slopes analysis using a tertile split on the WMC variable to derive Low (n = 32), Median (n = 32), and High (n = 32) WMC groups, which were based on the percentile markers outlined above. We found a significant positive relationship between anxiety and test performance in the High WMC group (B = .12 [β = .49], = 2.89, < .01), no effect in the Median WMC group (B = .02 [β = .13], = 0.70, > .10), and a significant negative relationship in the Low WMC group (B = −.07 (β = −.35), = −2.05, < .05).

image

Figure 1. An illustration of the anxiety x working memory capacity interaction effect on cognitive test performance. High WMC refers to the effect of trait anxiety on performance when WMC was held constant at the 66th percentile, whereas low WMC is held at the 33rd percentile. Median WMC shows the effect of anxiety on performance when WMC is held at the 50th percentile.

Download figure to PowerPoint

Table 2. Hierarchical regression results predicting cognitive test performance
Predictors R 2 Δ R2 F B t p
Note
  1. WMC, working memory capacity; R2, total explained variance; Δ R2, change in explained variance by step; change in F-ratio by step; B, unstandardized regression coefficient; associated t-statistic.

  2. *p < .01; **p < .001.

Step 1 .23 **
WMC  .87 5.23 <.001
Anxiety −.00−.10 
Step 2 .30 ** .07 8.36 *  
WMC  .66 3.75 <.001
Anxiety .00.11 
WMC × Anxiety  .06 2.89 <.01

Discussion

  1. Top of page
  2. Abstract
  3. Background
  4. Method
  5. Results
  6. Discussion
  7. Acknowledgement
  8. References

We tested the hypothesis that trait anxiety and WMC would interact to explain variance in working memory demanding cognitive test performance (i.e. maths and Raven's SPM) in a sample of adolescents. We found that for individuals with Low WMC, increases in trait anxiety were related to decreases in cognitive test performance. For those with High WMC, however, the pattern of results was reversed. An increase in trait anxiety was linearly associated with higher test scores. These effects were not better accounted for by gender, age, or time of testing. There was no relationship between anxiety and test performance for the overall sample and a restricted subsample of those with average WMC.

This Cognition × Emotion interaction finding is consistent with previous cognitive interference theories in anxiety (e.g. Eysenck et al., 2007) and helps to explain under what circumstances elevated anxiety negatively impacts cognitive test performance. The finding also accords with research addressing the relationship between maths anxiety, working memory, and maths test performance in particular (Ashcraft & Krause, 2007). Moreover, the moderating effect of WMC may be a crucial factor in explaining some of the discrepancies found in the literature concerning the relationship between anxiety and performance. For example, the meta-analysis conducted by Seipp (1991) showed that as well as negative effects of anxiety, there were also positive and null effects.

A novel finding in the results was that the students with good WMC and higher levels of anxiety showed better test performance than other individuals (see Figure 1). This is consistent with reports in the literature that moderate levels of anxiety can facilitate performance on tests (Fernández-Castillo & Gutierrez-Rojas, 2009) and are linked with better performance over time (DiLalla, Marcus, & Wright-Phillips, 2004). Eysenck et al. (2007) suggested that, one of the theoretical limitations of cognitive frameworks of anxiety and performance (e.g., ACT; Eysenck et al., 2007) is that they typically fail to fully account for situations where individuals with high levels of anxiety perform relatively well on cognitive tasks. ACT does, however, suggest that those individuals high in anxiety will be motivated to do well on tests to avoid negative evaluation (Eysenck & Derakshan, 2011). Our results extend this proposition to suggest that this advantage is only possible if individuals have the cognitive resources to offset or cope with the negative effects of anxiety. Other studies have suggested that the relationship between emotional factors and academic achievement is moderated by specific processes such as effortful control (e.g. Valiente et al., 2010) which should be tested in future work with adolescents.

Therefore, the results of the present study suggest that given a strong working memory capacity, higher levels of anxiety may be associated with increased rumination on negative evaluation that, in turn, may facilitate test performance. Future research should test this possibility along with the hypothesis derived from ACT that anxiety increases motivation to avoid negative evaluation/consequences resulting in individuals striving to perform well on tests. In an update to ACT, Eysenck and Derakshan (2011) consider further the possible role of motivation in the anxious individual. It is suggested that when a task is undemanding, or the goals are unclear, high-anxious individuals may lack motivation, whereas when a task is demanding and there are clear goals they make more use of goal-directed attentional system, through effortful control and attentional control mechanisms. If these processes are indeed in operation in young people with high levels of anxiety, our research suggests that they could be moderated by WMC. Our data also suggest a possible expansion of theory to encompass situations where anxiety can actually facilitate performance.

According to ACT, high-anxious individuals may be motivated to perform well on tasks but this can result in poor task efficiency. That is, on reaching parity with low-anxious participants in terms of performance, per se, high-anxious participants may exert extra effort or need more time to complete tasks. Only performance and not efficiency was measured in the current study and so this hypothesis cannot be directly answered here. It may have been that given more time to complete tests, participants with Low WMC and high levels of anxiety would have improved their scores. Nevertheless, it might be argued that in the context of tests requiring large amounts of cognitive resources, task performance within given time frames is a reality for many young people. The current findings may also have implications for interventions that target anxiety in children. The results suggest that reduction of anxiety should be considered in conjunction with individual differences in working memory (Roughan & Hadwin, 2011). Given that the relationship between anxiety and cognitive performance was only a negative one in the Low WMC group, young people with poor working memory skills are likely to benefit the most from any intervention that aims to reduce symptoms of anxiety.

In conclusion, the present results show that anxiety interacts with WMC to predict cognitive test performance. Anxiety was found to have a differential association with performance depending on available working memory resources. Furthermore, given that working memory and anxiety (or correlates of this such as temperament and behavioural inhibition) develop relatively early in childhood, prospective studies beginning as early as 5- or 6-years-old are feasible and desirable. Future research should additionally test for the mechanisms underlying the findings in this study which were not addressed. We suggest that decreased performance with low WMC in anxiety represents an extension of cognitive interference theories in that low capacity essentially acts much like a ‘second task’ does in dual task paradigms consuming cognitive resources. In the case of low WMC, the resource pool is already significantly depleted before factoring in negative effects of anxiety. The increased performance in high WMC individuals with anxiety is likely to be explained by an increased motivation to do well on the task which is driven by anxiety. Whereas for those with low capacity such anxiety becomes deleterious, in high working memory capacity individuals have the resources to act successfully on those motivation drives. To address this proposition more clearly, motivation should be explicitly measured in future research.

Acknowledgement

  1. Top of page
  2. Abstract
  3. Background
  4. Method
  5. Results
  6. Discussion
  7. Acknowledgement
  8. References

The authors would like to thank all the parents, teachers, staff, and children at Purbrook Park School, Bitterne Park School, and Twynham School, for their participation in the project.

This research was funded by an ESRC CASE studentship (award number PTA-033-2004-0052) and Hampshire Educational Psychology Service and Action Medical Research (award number SP4598).

References

  1. Top of page
  2. Abstract
  3. Background
  4. Method
  5. Results
  6. Discussion
  7. Acknowledgement
  8. References