Cognitive load theory (CLT; Sweller, Reference Sweller, Mestre and Ross2011) is a contemporary educational psychology theory that applies findings from cognitive science to generate effective instructional designs for learning. The theory stipulates a working memory (WM) whose capacity and duration limitations can be circumvented by knowledge in schematic form in long-term memory. Baddeley's (Reference Baddeley1986) WM model has underpinned the human cognitive architecture used by CLT to generate novel instructional designs. Recently, Schweppe and Rummer (Reference Schweppe and Rummer2014) have argued CLT might benefit from considering other WM models as a means of generating novel instructional designs. For example, Kane and Engle's (Reference Kane and Engle2002) WM model holds that WM capacity is a function of the ability to control attention, so that information is held in an active, easily retrievable state. Kane and Engle's WM model raises questions as to what instructional activities or designs might best enhance attention control; or, to put it another way, reduce mind wandering as a failure of attentional control. Mind wandering, defined by McVay and Kane (Reference McVay and Kane2010) as ‘off-task thoughts during an ongoing task or activity’ (p. 188), is one of the most ubiquitous of all mental activities (Killingsworth & Gilbert, Reference Killingsworth and Gilbert2010), and its impact in educational settings is just beginning to be understood (Smallwood, Fishman, & Schooler, Reference Smallwood, Fishman and Schooler2007; Wammes, Boucher, Seli, Cheyne, & Smilek, Reference Wammes, Boucher, Seli, Cheyne and Smilek2016). Instructional designs to support attentiveness and/or reduce mind wandering thus stand in contrast to the traditional focus of CLT instructional designs on managing WM capacity limitations through a focus on intrinsic cognitive load (a function of the complexity of the learning task given the learner's knowledge base) and/or extraneous cognitive load (resulting from cognitive processes unrelated to learning).The current iteration of cognitive load theory incorporates insights from evolutionary theorising, holding that humans have evolved to attend to, process, and guide behavioural responses through biologically primary knowledge (Geary, Reference Geary2008). Examples of biologically primary knowledge include facial recognition, physical gestures, and learning a native language. In contrast, a separate category of biologically secondary knowledge requires extensive, effortful engagement in learning, often through structured activity within educational institutions; examples include learning to read, write, and solve mathematical problems. Recent research informed by both embodied cognition perspectives and cognitive load theory has focused on both observed gestures (e.g., Wong et al., Reference Wong, Marcus, Ayres, Smith, Cooper, Paas and Sweller2009) and self-generated gestures (e.g., Ginns, Hu, Byrne, & Bobis, Reference Ginns, Hu, Byrne and Bobis2016) as forms of biologically primary knowledge that might support the construction of biologically secondary knowledge (see Paas & Sweller, Reference Paas and Sweller2012). We propose that a specific drama education activity involving movement, mirroring, may likewise incorporate biologically primary behaviours and cognitive processes that can support learning by enhancing attention. In this practice, ‘two actors face each other and mirror each other's movements’ (Cohen, Reference Cohen2013, p. 203). One participant may begin as the leader and one as the follower, while maintaining eye contact throughout the process (Boal, Reference Boal2002; Cohen, Reference Cohen2013). Cohen (Reference Cohen2013) argues mirroring forces ‘total situational concentration’ in ‘driving out all contextual awareness’ (p. 203). According to these arguments, then, mirroring is an embodied drama activity that might fruitfully be incorporated into lessons requiring sustained attention.

Such effects on attention and learning can be predicted by several biologically primary attentional pathways. The first is ‘a bias to orient to the human face and to automatically process key pieces of facial information’ (Geary, Reference Geary2008, p. 180; see also von Grünau & Anston, 1995). Attention capture by gaze can be attributed to the social significance and survival value conveyed by eye contact (Baron-Cohen, Reference Baron-Cohen, Moore and Dunham1995). The second potential attentional pathway mirroring may activate is through simple social presence. Ybarra et al. (Reference Ybarra, Burnstein, Winkielman, Keller, Manis, Chan and Rodriguez2008) randomly assigned participants to conditions involving 10 minutes of social interaction, involving discussion of a social issue, 10 minutes of intellectual activities, or 10 minutes watching a video clip with social interactions. Participants in the social interaction condition outperformed the control participants on cognitive assessments of speed of processing and working memory, and performed at levels equivalent to the intellectual activities condition. Ybarra et al. explained these results by proposing that ‘social interaction can “exercise” general cognitive processes (working memory, speed of processing, inhibition) in the service of social cognition (e.g., empathy, mentalizing, symbolic interaction)’ (p. 255). Third, social interactions may often result in non-conscious mimicry of a range of non-verbal behaviours, including postures, mannerisms, and expressions (Chartrand & Bargh, Reference Chartrand and Bargh1999). Vicaria and Dickens (in press) describe interpersonal coordination behaviours as a ‘social glue’ with evolutionary advantages, and their meta-analysis of the interpersonal coordination literature established reliable effects of such behaviours across a range of intrapersonal and interpersonal outcomes; this body of research did not, however, include any studies with an educational focus.

Anecdotal arguments made by drama educators thus suggest mirroring enhances attention through mimicking in a social context. The present study hypothesised that mirroring will temporarily enhance attention, in turn enhancing learning. Thus, a precursor mirroring activity may be an effective ‘warm-up’ activity prior to a mathematical lesson where sustained attention is required. We propose the following hypotheses:

1. 1. Participants who begin a short mathematics lesson with a mirroring exercise will develop more effective problem-solving schemas than participants who perform the same movements without mirroring, as measured by subsequent test performance.

2. 2. Participants who begin a short mathematics lesson with a mirroring exercise will develop more efficient problem-solving schemas than participants who perform the same movements without mirroring, as measured by subsequent time to solution on test questions.

Results

Descriptive statistics (means and standard deviations) for all variates under analysis are given in Table 1.

TABLE 1 Descriptive Statistics for Variates Under Analysis

The Anderson-Darling test of normality was used to evaluate the assumption of normally distributed scores in each condition, and followed Keselman, Othman, and Wilcox's (2013) recommendation of a type 1 error rate for this test of .20. Using this standard, data were found to be non-normally distributed in one or both conditions for several of the variables under analysis. Levene's test, which is frequently used to test the assumption of homogeneity of variances across experimental conditions, is also dependent on the assumption of normality. The recently developed non-parametric Levene test (Nordstokke, Zumbo, Cairns, & Saklofske, Reference Nordstokke, Zumbo, Cairns and Saklofske2011), based on a rank transformation of all scores followed by Levene's test on these scores, provides a suitable alternative under violations of the assumption of normality.

Since for some variates data were non-normally distributed but the non-parametric Levene test was not statistically significant, the Mann-Whitney U test using an exact p value (Mehta & Patel, Reference Mehta and Patel2012) was used to analyse these data. Additional to these analyses, standardised mean difference effect size, Cohen's d, with its 95% confidence interval is also reported (for Mann-Whitney tests, effect sizes and confidence intervals are based on methods developed by Newcombe, Reference Newcombe2006a, Reference Newcombe2006b). The present study adopted Hattie's (2009) benchmarks for effect sizes in educational studies suggesting small d = .20, medium d = .40, and large d = .60 and above. Analyses controlled the Type 1 error rate at 0.05.

Were the Groups Equivalent in Prior Knowledge?

A preliminary analysis investigated whether the mirroring and non-mirroring conditions were equivalent in general mathematical knowledge by analysing participants’ Brief Mathematics Assessment-3 scores. The Anderson-Darling test of normality verified that the distribution of scores in each condition was not normally distributed with the mirroring condition (p < .001) and non-mirroring condition (p = .101). A non-parametric Levene test verified the equality of variances across the experimental conditions (F = 3.88, p = .061). A Mann-Whitney test determined that there was no significant difference between the mirroring group (mean rank = 14.62) and the non-mirroring group (mean rank = 12.38) on the Brief Math Assessment scores; U = 70.00, p = .444, d = .31 [95% CI -.47, 1.05]. Thus, random assignment of participants to conditions was effective in distributing mathematical ability across conditions.

Does a Mirroring Intervention Improve Efficiency of Mathematics and Mathematical Learning Outcomes?

The major variables under analysis were number of correct solutions to problems (hypothesis 1 — effectiveness of learning) and time to correct solution across the two tests (hypothesis 2 — efficiency of learning). For Elementary test questions, a ceiling effect was obtained on number of questions correctly answered (mirroring condition mean score = 92.3% correct; non-mirroring condition mean score = 90%), precluding analysis of this variate. For time to solution of Elementary test questions, the Anderson-Darling test of normality showed that both the mirroring (p = .249) and non-mirroring data (p = .211) were normally distributed. Subsequently, a non-parametric Levene test verified the equality of variances in the samples, which were not statistically significant (F = .212, p = .649). An independent groups t test established that participants in the mirroring condition solved the Elementary test questions more quickly (M = 182.38 seconds, SD = 38.96) than the non-mirroring condition (M = 212.46, SD 30.72), t(24) = -2.19, p = .039, d = -.86, 95% CI [-1.66, -0.05]. Thus, hypothesis 1 of the study was not supported in the number of questions correctly answered. However, participants in the mirroring group were more efficient in answering these questions by the lower amount of time taken to answer questions, compared to the non-mirroring group, supporting hypothesis 2.

As with the Elementary test questions, a ceiling effect was obtained on number of Brain Builder questions correctly answered (mirroring condition mean score = 96.9% correct; non-mirroring condition mean score = 96.9%), precluding analysis of this variate. For time to solution of Brain Builder test questions, the Anderson-Darling test of normality showed that the mirroring condition was not normally distributed (p = .117), whereas the non-mirroring group was normally distributed (p = .976). As these results varied across conditions, a conservative approach is to infer that the assumption of normality was not met. Subsequently, the non-parametric Levene test showed the equality of variance in the samples was not statistically significant (F = .358, p = .555). The Mann-Whitney non-parametric test found the mirroring condition solved the more complex ‘Brain Builder’ test questions more quickly (mean rank = 9.88) than the non-mirroring condition (mean rank = 17.12), U = 37.50, p = .015, d = -1.08 [95% CI -1.83, -.19]. Thus, hypothesis 1 of the study was also not supported in the number of questions correctly answered in this harder test. However, participants in the mirroring group were also considerably more efficient in answering the harder questions, supporting hypothesis 2.