Transferability and sustainability of task-switching training in socioeconomically disadvantaged children: a randomized experimental study

ABSTRACT

Most empirical studies on executive function (EF) and socioeconomically disadvantaged children are largely restricted to understanding and confirming the link between them. The current study extended previous research by examining the near-transfer of task-switching training to a structurally similar new switching task, and far-transfer to a structurally dissimilar EF task (i.e. inhibition and working memory) and academic performance through a 4-week task-switching training programme using a randomised experimental design. Fifty low SES primary school students (M_age = 8.6 years, SD = 0.7) in Hong Kong participated in pretest, posttest, and a one-year follow-up on task-switching performance, EF, and academic performance in three core subjects. Results showed that compared to an inactive control group, the app training group showed improved performance in similar (untrained task-switching) and dissimilar (inhibition) EF tasks even at one-year follow-up. No training effect was found at posttest and in academic performance. Potential implications for future research in task-switching training are discussed.

KEYWORDS:

• Executive function
• poverty
• switching cost
• task-switching
• working memory

1. Introduction

Numerous studies (e.g. Chung et al., 2016; Lawson et al., 2018) have investigated the links between family socioeconomic status (SES) and early cognitive development in children. Socioeconomic adversity was negatively related to developmental outcomes in children such as language acquisition, nonverbal reasoning, general intelligence, and executive function (EF) (de Rosa Piccolo et al., 2016). A wide range of conceptual frameworks of child development have examined the mediating role of early cognitively-enriching environments in the relationship between poverty and children’s early outcomes. Socioeconomically disadvantaged children are less likely to be exposed to early childhood cognitive stimulation or experience high-quality early education and social interaction than their socioeconomically advantaged counterparts (Haft & Hoeft, 2017; Lugo-Gil & Tamis-LeMonda, 2008). This finding suggests that children with low SES are likely to have reduced social and cognitive functioning before their crucial school years, as well as later in life. Among the cognitive disparities related to SES, disparities in EF appear to be larger than those in other cognitive abilities. EF refers to a set of cognitive abilities to perform tasks that guide goal-directed behaviours (Ackerman & Friedman-Krauss, 2017; Zelazo et al., 2013). Many researchers have posited cognitive flexibility, inhibition, and working memory as the main components of EF (Diamond, 2012).

Previous findings suggest that poverty and poverty-related stressors are associated with lower EF ability and compromised self-regulation in young children (e.g. Blair et al., 2011). A study of Chinese-American children between six and nine years of age (Chen et al., 2015), showed that SES was significantly associated with effortful control, which overlapped with EF (Bridgett et al., 2013). SES was also significantly associated with behavioural regulation (heads-toes-knees-shoulders) in children aged six to nine in Taiwan (Wanless et al., 2013), which involves inhibitory control, cognitive flexibility, and working memory (Eisenberg et al., 2004). Poon and Ho (2014) also revealed a significant association between family income and EF, specifically interference control, among 117 Hong Kong Chinese adolescent boys comprising both delinquents and non-delinquents. Structural and functional brain imaging suggests a link between SES and the thickness and surface area of the prefrontal cortex of children (Lawson et al., 2013), a direct and accurate measurement confirming SES disparities in EFs.

1.1. EF and academic skills

EFs have been found to be significantly related to an array of academic skills in children (Fuhs et al., 2015; Sektnan et al., 2010; Wanless et al., 2011). For instance, the domain of working memory is widely recognised as being critical to reading and mathematical abilities (Birgisdóttir et al., 2015; Peng et al., 2016; Peng et al., 2018). Similar results were found for inhibition associated with vocabulary and reading, ranging from medium to highly significant (Clark et al., 2013; Clark et al., 2014; Miller et al., 2013; Nesbitt et al., 2013; Viterbori et al., 2015; Wanless et al., 2011; Welsh et al., 2010). Meta-analysis suggests that cognitive flexibility is associated with performance in both mathematics and reading (Yeniad et al., 2013). Given these findings, which show a strong link between EF and academic capability, it is natural to assume that EF improvement might have a beneficial impact on the academic performance of children.

1.2. EF plasticity

Many studies suggested that EF can be enhanced with appropriate training (Buitenweg et al., 2012; Diamond, 2012; Diamond et al., 2007; Karbach & Schubert, 2013; Lillard & Else-Quest, 2006; Melby-Lervåg & Hulme, 2016; O’Connor et al., 2000). Knowledge concerning the malleability of EF comes from a growing body of literature which shows that EF can be trained through extensive practice requiring the prefrontal cortical circuits (cf. Hebb, 1949; Hsu et al., 2014; Wass et al., 2012; Zhang et al., 2019). According to research into neural systems, EF develops rapidly in early childhood (Blair, 2016) and shows continuous development well into young adulthood (Diamond, 2013).

Moreover, research suggested that EF skills develop significantly when children begin to participate in formal learning environments, as the brain becomes more adaptive to change and functionally responsive to the environment, suggesting that this might be a period of high malleability. For example, in a study into the malleability of EF in childhood, Zhang et al. (2019) showed that a first-grade schooling group outperformed a pre-school group in working memory and inhibitory control at baseline. After receiving a 15-minute computerised training four times per week for five weeks, the children in the pre-school group not only showed improvements in the trained tasks but were also able to achieve a performance comparable to the participants in the school group both at posttest and follow-up. This provides evidence for the malleability of EF, with even a short-term intervention facilitating the acquisition of important EF skills during childhood.

1.3. Cognitive training and EF

Computerised training programmes have become widespread in the training of EF. These typically employ game tasks that specifically require one or more EF skills (e.g. Alloway et al., 2013; Bennett et al., 2013). One training regime commonly used to test executive control functioning is the task-switching paradigm (e.g. Anguera et al., 2013; Karbach & Kray, 2009; Kiesel et al., 2010; Kray et al., 2012; Kray & Fehér, 2017; Minear & Shah, 2008; Pereg et al., 2013; White & Shah, 2006; Zinke et al., 2012). The task-switching training paradigm is a process-based computerised training regime which specifically trains the ability to switch rapidly between two or more cognitive demands. The core EF in this switching ability are inhibition, cognitive flexibility, and working memory (e.g. Koch et al., 2010) and enables adaptation to a rapidly changing environment (Miyake et al., 2000). It allows for separate measures of executive control processes within the same experimental paradigm (Kiesel et al., 2010). Although there are a few variants, most studies have the participants switch between two tasks (A and B) within a mixed-task block and perform only one of the tasks within a single-task block. Two types of costs are thus determined: the switching cost resulting from reconfiguration of the task set and overcoming the interference effect between the previous and current task execution (e.g. Kiesel et al., 2010; Monsell, 2003); and the mixing cost resulting from resolving conflict or stimulus ambiguity during mixed-task trials or differences in arousal level or working memory load between single-task and mixed-task blocks (e.g. Rubin & Meiran, 2005).

Both the switching cost (e.g. Berryhill & Hughes, 2009; Karbach & Kray, 2009; Kray et al., 2012; Kray & Fehér, 2017; White & Shah, 2006; Zinke et al., 2012) and mixing cost (e.g. Minear & Shah, 2008; Soveri et al., 2013; Strobach et al., 2012) can, in general, be reduced through practicing task-switching. However, there are studies of training-induced transfer effects which have yielded mixing findings; the near-transfer effects do reduce switching and/or mixing costs for untrained switching tasks (e.g. Anguera et al., 2013; Karbach & Kray, 2009; Kray et al., 2012; Kray & Fehér, 2017; Minear & Shah, 2008; Pereg et al., 2013; White & Shah, 2006; Zinke et al., 2012) but far-transfer effects were negligible. For instance, von Bastian and Oberauer (2013) reported no evidence for far-transfer effects on reasoning, inhibition, and working memory after cue-based task-switching training in adults. Zhao et al. (2018) discovered no far-transfer effects on untrained EF tasks and general IQ after extensive task-switching training in adults, while short-lived transfer effects on working memory were found for young children. On the contrary, Karbach and Kray (2009) studied children aged from seven to nine years, asking them to perform two simple decision tasks (A and B) and switch between them at a specific signal. In task A, children were asked to indicate whether the object presented on the computer was a fruit or a vegetable and in task B to indicate whether the object was small or large. Although the training duration was short with four training sessions once a week over four weeks, the results showed that those who received training performed significantly better, not only in similar (untrained) switching tasks, but also in dissimilar EF domains, such as inhibition, verbal and visual-spatial working memory, and reasoning. A similar task-switching training paradigm to Karbach and Kray (2009), Kray et al. (2012) was used and this confirmed a training-induced transfer effect on children with attention deficit hyperactivity disorder (ADHD), with children in an intervention group not only showed improvement in task-switching performance, but also in inhibitory control and verbal working memory.

1.4. Task-switching training and academic achievement

Several studies examined the role of task-switching abilities in the context of academic achievement, with most reporting positive correlations (e.g. Arán Filippetti & Richaud, 2017; Cantin et al., 2016; Clark et al., 2010; Gerst et al., 2017). Critically, two meta-analyses have confirmed this association. Firstly, Yeniad et al. (2013) reported that switching was positively correlated achievement in mathematics (r = .26) and reading (r = .21). Secondly, Jacob and Parkinson (2015) reported similar results, but with a slightly higher estimation of effect sizes, i.e. switching ability was positively correlated with achievement in mathematics (r = .34), and reading (r = .32). A recent study examined the association between task switching ability and academic achievement among 10th grade Chinese adolescents (N = 221) suggested that task-switching ability was positively related to achievement in sciences and mathematics but not humanities (Li et al., 2020).

Considering that the aforementioned findings between task-switching and academic abilities are consistent, it is assumed that task-switching training has the potential to improve academic performance. While there is plenty of evidence of a transfer to structurally similar tasks and other EF domains, as well as the child’s reasoning ability (e.g. Karbach & Kray, 2009; Kray et al., 2012; Liu et al., 2015), the transfer of skills to academic abilities has not been established as research has mostly focused on computerised working memory training.

Lastly, some studies have examined the role of individual differences in baseline performance in EF training success. Past evidence showed that individuals with low EF benefitted most from the intervention, resulting in compensation effects (e.g. Bherer et al., 2008; Cepeda et al., 2001; Karbach et al., 2015; Karbach et al., 2017; Zinke et al., 2014). Hence, research into whether EF could be enhanced or even whether the EF training effect could be transferred to other areas for low SES children would be extremely meaningful and provide valuable information for early intervention.

1.5. Research gaps and study objectives

Although consistent research findings on the relationships between EF and socioeconomically disadvantaged children are available, a number of research gaps still need to be addressed. For instance, most of these empirical studies are largely limited to understanding and confirming the association between these two factors. Research on whether EF could be enhanced through a task-switching paradigm or whether the training effects of EF could be transferred to other areas in a sample of Chinese children with low SES, would provide valuable insight into the understanding and development of early interventions. In sum, the main objectives of the present study were: (1) to examine group differences in the transfer of task-switching training to a similar new switching task; (2) to examine the far-transfer of task-switching training to other EF tasks including inhibition, working memory, and academic performance in socioeconomically disadvantaged children; and (3) to explore the sustainability of the training effect (immediate vs. one year after the intervention).

2. Methods

2.1. Participants

Fifty primary school students (M_age = 8.6 years, SD = 0.7) participated in the present study. This sample size was obtained for a power of .80, a medium effect size of .25, and an alpha level of .05 using G*Power (Erdfelder et al., 1996). They were recruited from primary schools in Hong Kong. All of them had household incomes below half of the median household income adjusted according to household size. They had not been diagnosed with psychological disorders such as attention deficit and hyperactivity disorder, reading disability, and autism. Their first spoken language was Cantonese, they were of normal intelligence (≥ 80), with no suspected brain damage, neurological, sensory, or psychiatric problems. The present study was approved by the Human Research Ethics Committee at the first author’s institution. Prior to the commencement of data collection, written consents were obtained from all selected students and their parents.

2.2. Overview of the procedures

The current study adopted a cluster randomised controlled design, consisting of a screening stage, an assessment stage and an intervention stage. During the screening stage, participants were tested with Raven Standard Progressive Matrices (Raven et al., 2000) as a proxy of intelligence to rule out those with low intellectual functioning. Demographic data including educational level and family income were collected through questionnaires from their parents. The assessment stage consisted of pre-test, post-test and one year follow-up. A pre-test measuring three behavioural tasks on EF (untrained task-switching paradigm, Stroop test, and Backward Digit Span task) was administered individually in a pseudo-randomised order two weeks prior to intervention stage. Participants’ performance on three core subjects (Chinese, English, and Mathematics) was also collected through teachers’ reports as one of the pre-test measures. The pretest was followed by four intervention sessions. During the intervention stage, participants were randomly assigned to two groups with 24 participants in the app training group and 26 participants in the inactive control group. The app training group participated in four 1-hour app training sessions on task-switching once a week over four weeks. Participants in the inactive control group did not receive any training during the intervention period. After the intervention stage, a posttest and a follow-up test were conducted a week and approximately a year after the completion of the four intervention sessions.

2.3. Training situations

The training regime was based on Karbach and Kray’s (2009) study, which showed a reduction in switching and mixing costs after four training sessions in children. A 7.9-inch touch-screen Apple mini iPad was used for item presentation and response collection. During the training, participants were asked to perform two simple decision tasks and to switch between them due to a specific signal. The task-switching paradigm included both single-task and mixed-task blocks. The single-task block contained only one of the two tasks while in the mixed-task block, trials with both tasks were included with the task switched on every other trial. Each training session began with two mixed-task practice blocks followed by 24 experimental mixed-task blocks with 17 trials each, with the task switched in every other trial. Trials began with a fixation cross for 1400 ms followed by the appearance of a target word. The inter-trial interval was 25ms and feedback on their accuracy was given after each trial. The procedure of the untrained task-switching paradigm at the assessment stage was similar to that of the task-switching test in the pretest, posttest, and follow-up but with some modifications concerning the objects presented and the task instructions.

2.4. Transfer situations

Transfer of training was assessed through a pretest, posttest, and follow-up design. This was defined as performance improvement in two posttests (immediate vs. one year after intervention) and then compared to the baseline performance at pretest. This pretest baseline measure was conducted two weeks prior to the intervention. The contents of these tests were identical to the pretest session and lasted 60–70 min each. All participants in the app training and inactive control groups completed the assessments individually in a classroom setting at schools. Participants’ performance on three core subjects (Chinese, English, and Mathematics) was also collected through teachers’ reports as one of the pre-test measures. All training and assessments were administered by well-trained research assistants.

2.5. Measures

2.5.1. Untrained task-switching task

In the pretest, posttest, and follow-up tests, participants were shown one of 16 vegetables and 16 fruits of different sizes (i.e. large vs. small) in different trials (see Figure 1). They had to categorise the object shown as either vegetable or fruit in the object categorisation task, and to categorise the object shown as either large or small in the size categorisation task as quickly and accurately as possible. Overall accuracy rate and reaction time (for correct trials only) were measured for the task-switching paradigm. Mixing cost was calculated as the mean reaction time for the performance between single-task and mixed-task blocks. Switching cost was calculated as the difference in the mean reaction time between the switch and non-switch trials within mixed-task blocks. The first trials of both single-task and mixed-task blocks were excluded from analysis.

Figure 1. Example trials in the task-switching paradigm in the pretest, posttest, and follow-up tests.

2.5.2. The task-switching paradigm

During the intervention stage, participants were asked to perform two simple decision tasks and to switch between them due to a specific signal. In task A, they are asked to indicate the object presented was either a car or a plane and then in the task B, they are asked to indicate the number of object (i.e. one or two car(s)/plane(s)). Example trials are shown in Figure 2.

Figure 2. Example trials in the task-switching test in app training.

2.5.3. General intellectual ability

The general intellectual ability of the participants was examined using the Raven’s Progressive Matrices (Raven et al., 2000). This test contained 60 items in total and each item consisted of a visual pattern with a missing piece. Participants had to identify the correct piece to fill in the missing part and make the pattern intact. Participants who scored below 80 were excluded from the current study. This criteria was used to exclude those with poor executive function because of low intellectual abilities.

2.5.4. Stroop color and word test (Stroop, 1935)

Inhibition plays a key role in task-switching as it is primarily triggered by task-irrelevant and task-relevant information, in which the participants are instructed to inhibit or control impulsive responses (Koch et al., 2010). Inhibition was measured under the word condition, colour condition, and colour-word condition in three separate 1-minute rounds (Stroop, 1935). In the word condition (W), participants were instructed to read the Chinese words and name the Chinese names of colours printed in black ink. In the colour condition (C), participants were required to name the colours of colour patches of red, blue, and green inks. Finally, in the colour-word condition (CW), the name of a colour (e.g. BLUE) was presented using a congruent colour (e.g. blue) or an incongruent colour (e.g. yellow) on paper in Chinese. Participants had to name the colour, instead of reading the word on the paper as quickly and accurately as possible. To obtain the interference score as a measure of inhibition, the colour and word score was subtracted from the actual number of items correctly named in the incongruous condition using the formula: CW–[(C × W)/(C + W)] (Golden, 1987). The word score (W) represents the number of items (words) completed. The colour score (C) represents the number of items (name of colour) completed. This scoring system referenced the standardised version of similar test for other populations (Golden, 1987). The Chinese version of the non-computerised Stroop Color and Word Test with normative data for Chinese populations was validated by the Hong Kong Psychological Society (2020).

2.5.5. Backward digit span

Working memory was measured using the Backward Digit Span subtest of the Wechsler Intelligence Scale for Children, Third Edition [WISC-III] (Wechsler, 1981). Participants were orally presented with 18 sequences of single-digit numbers with increasing length from two to nine (two sequences per length), and they had to repeat the numbers in backward order. One point was given for each completely recalled sequence, and the test was terminated after unsuccessful recalls of two consecutive sequences.

2.5.6. Academic subjects

The academic achievement of the students was measured by teachers’ ratings in three key learning areas (i.e. Chinese, English, and Mathematics). Teachers chose the most suitable range of performance level (bottom 20%, 21%−40%, 41%-60%, 61%-80%, and top 20% in the class) for each participant and the values were transformed to a scale from 1 (bottom 20%) to 5 (top 20%) for data analysis. This scale has been adopted in previous research that measured academic achievement in students in Hong Kong (Chen, 2005, 2008).

2.6. Data analysis

An independent sample t-test and a chi-square test were conducted to explore whether the demographic characteristics differed across the two groups of socioeconomically disadvantaged children. A 2 (app training vs. inactive control) x 3 (pretest vs. posttest vs. follow-up) repeated measures ANOVA was also carried out, with time as a within-subjects factor and group as a between-subjects factor. The dependent variables were overall accuracy rate, reaction time, mixing cost, switching cost, inhibition, working memory, Chinese, English, and Mathematics.

3. Results

Table 1 summarises the demographic background of the two groups of participants. Results from an independent sample t-test indicated that there were no significant group differences in age (p = .55), intellectual ability (p = .85), paternal education level (p = .98), maternal educational level (p = .06), and family income (p = .90). A chi-square test was also conducted to examine the group differences in gender. No significant difference was found in two groups, χ² (1, 50) = 2.83, p = .09. In the app training group, there were 10 males and 14 females. In the inactive control group, there were 17 males and 9 females.

Table 1. Demographic characteristics in app and inactive control groups

	App (n = 24)		Inactive Control (n = 26)
	M	SD	M	SD	t
Age	8.56	.41	8.67	.86	.62
IQ	110.04	15.46	109.19	16.31	.19
Father education level	3.27	.65	3.28	.57	.02
Mother education level	3.00	.60	3.44	.62	1.95
Family income	4.05	1.32	4.00	1.38	.12

To prevent baseline differences, participants were matched based on their pretest performance in accuracy rate, reaction time, mixing cost, and switching cost. There were no baseline differences between the two groups in accuracy rate (F(1, 48) = .87, p = .36; η² = .02), reaction time (F(1, 48) = .62, p = .43; η² = .01), mixing cost (F(1, 48) = 3.15, p = .08; η² = .06), and switching cost (F(1, 48) = .65, p = 42; η² = .01). Table 2 shows the descriptive statistics of the two groups of participants in pretest, posttest, and follow-up tests for all outcome measures.

Table 2. Descriptive statistics of the app training group and inactive control group of participants in pretest, posttest, and follow-up for all outcome.

Measures		App (n = 24)			Inactive Control (n = 26)
		Pretest	Posttest	Follow-up	Pretest	Posttest	Follow-up
Accuracy rate (%)	M	92.61	84.10	94.19	91.72	90.88	92.78
	SD	2.78	10.05	3.41	3.80	5.86	4.82
Reaction time (all trials)	M	1519.77	1727.97	1267.99	1563.89	1498.96	1258.54
	SD	156.65	307.64	244.35	229.09	217.16	209.19
Mixing cost (ms)	M	182.14	53.82	41.04	132.40	113.37	121.30
	SD	101.92	143.44	107.57	96.19	112.22	81.77
Switching cost (ms)	M	147.85	52.29	57.71	128.21	65.50	63.98
	SD	70.47	100.59	61.22	98.21	97.11	73.26
Inhibition	M	9.17	9.96	11.43	9.20	9.79	10.13
	SD	1.99	1.64	1.93	2.06	1.67	2.11
Working memory	M	9.23	9.67	10.04	9.55	10.48	9.80
	SD	2.09	2.24	2.12	2.63	2.95	3.63
Chinese	M	3.04	2.92	2.83	3.17	3.29	3.29
	SD	.91	1.02	1.13	1.52	1.33	1.23
English	M	2.50	2.50	2.58	2.83	3.22	3.17
	SD	.98	1.32	1.38	1.44	1.57	1.47
Mathematics	M	2.95	2.95	2.90	3.13	3.38	3.38
	SD	1.17	1.05	1.23	1.45	1.41	1.41

3.1. Training data

3.1.1. Overall accuracy rate

The main effect of time (F(2, 47) = 11.72, p < .001; η² = .33), and the interaction effect of time x group on accuracy rate (F(2, 47) = 6.14, p < .01; η² = .21) were significant. The main effect of group was not significant, F(1, 48) = 1.95, p = .17, η² = .04. An analysis of simple effects showed that the main effect of time was significant for the app training group (F(2, 22) = 11.81, p < .001; η² = .52), but not for the inactive control group (F(2, 24) = .90, p = .42; η² = .07). Post-hoc comparisons revealed that the overall accuracy rates of the students in the app training group did not differ between the pretest and follow-up test (p = .08), however they had lower overall accuracy rate in the posttest compared to the rates in the pretest and follow-up test (p < .001) (see Figure 3).

Figure 3. Accuracy rate in task-switching task between the app training and inactive control groups in pretest, posttest, and follow-up. Error bars represent standard errors.

3.1.2. Reaction time

There were significant main effect of time (F(2, 47) = 51.45, p < .001; η² = .69) and its interaction with group (F(2, 47) = 7.77, p < .01; η² = .25) on reaction time. No significant main effect of group was found, F(1, 48) = 1.58, p = .21, η² = .03. Simple effects analysis showed that the main effect of time was significant for both app training group (F(2, 22) = 30.75, p < .001; η² = .74) and inactive control group (F(2, 24) = 20.38, p < .001; η² = .63). Post-hoc comparisons showed that students in the app training group had a significantly shorter reaction time in the follow-up test than in the pretest (p < .001) and posttest (p < .001). However, their reaction time in the posttest was longer than that in the pretest (p < .01). Students in the inactive control group had significantly shorter reaction time in the follow-up test than in the pretest and posttest (p < .001), but there was no significant difference between the pretest and posttest (p = .15) (see Figure 4).

Figure 4. Reaction time in task-switching task between the app training and inactive control groups in pretest, posttest, and follow-up. Error bars represent standard errors.

3.2. Near-transfer effects

3.2.1. Mixing cost

There were significant main effect of time (F(2, 47) = 7.98, p < .01; η² = .25) and interaction effect of time x group (F(2, 47) = 5.52, p < .01; η² = .19) on mixing cost. The main effect of group was not significant, F(1, 48) = 2.88, p = .10, η² = .06. Simple effects analysis showed that the main effect of time was significant for the app training group (F(2, 22) = 14.08, p < .001; η² = .56), but not for the inactive control group (F(2, 24) = .22, p > .05; η² = .02). Specifically, students in the app training group showed that their mixing cost was lower in posttest (p < .01) and follow-up test (p < .001) compared to pretest. The difference between posttest and follow-up test for students in the app training group was not significant (p = .67) (see Figure 5).

Figure 5. Mixing cost in task-switching task between the app training and inactive control groups in pretest, posttest, and follow-up. Error bars represent standard errors.

3.2.2. Switching cost

The main effect of time on switching cost was significant, F(2, 47) = 14.64, p < .001; η² = .38. However, time did not interact with group (F(2, 47) = .50, p > .05; η² = .02). There was no significant main effect of group on switching cost, F(1, 48) = .00, p = .99, η² = .00. Post-hoc comparisons revealed that the switching cost was generally lower in posttest and follow-up test compared to pretest (p < .001). The difference between posttest and follow-up test was not significant (p = .91) (see Figure 6).

Figure 6. Switching cost in task-switching task between the app training and inactive control groups in pretest, posttest, and follow-up. Error bars represent standard errors.

3.3. Far-transfer effects

3.3.1. Inhibition

There were significant main effect of time (F(2, 44) = 27.15, p < .001; η² = .55), and interaction effect of time x group (F(2, 44) = 5.00, p < .05; η² = .19) on inhibition. No significant main effect of group was obtained, F(1, 45) = 1.19, p = .28, η² = .03. An analysis of simple effects revealed that the main effect of time was significant for both app training (F(2, 21) = 19.24, p < .001; η² = .65) and inactive control group (F(2, 22) = 7.49, p < .01; η² = .41). Specifically, students in the app training group had higher inhibition in the follow-up test than in the pretest (p < .001) and posttest (p < .001). Their posttest score was also higher than pretest score in inhibition (p < .05). None of the differences across the three different time points were found for students in the inactive control group (p = .15) (see Figure 7).

Figure 7. Differences in inhibition between the app training and inactive control groups in pretest, posttest, and follow-up. Error bars represent standard errors.

3.3.2. Working memory

The main effects of group (F(1, 47) = .21, p = .65, η² = .01) and time (F(2, 46) = 2.37, p = .11; η² = .09), as well as the interaction effect of time x group (F(2, 46) = .90, p = .42; η² = .04) on working memory were not significant (see Figure 8).

Figure 8. Differences in working memory between the app training and inactive control groups in pretest, posttest, and follow-up. Error bars represent standard errors.

3.3.3. Academic performance in Chinese, English, and Mathematics

The main effects of time on Chinese (F(2, 45) = .09, p = .92; η² = .00), English (F(2, 44) = 1.25, p = .30; η² = .05), and Mathematics (F(2, 43) = .75, p = .48; η² = .03) subjects were not significant. The interaction effects of time x group on these three academic subjects also did not reach statistical significance (Chinese = F(2, 45) = .91, p = 41; η² = .04, English = F(2, 44) = .87, p = .43; η² = .04, and Mathematics = F(2, 43) = .86, p = .43; η² = .04). There were no significant main effects of group (Chinese: F(1, 46) = 1.00, p = .32, η² = .02; English: F(1, 45) = 2.17, p = .15, η² = .05; Mathematics: F(1, 44) = .95, p = .34, η² = .02).

4. Discussion

The overall aim of this study was to evaluate the transfer effect of app task-switching training on EF and academic achievement of socioeconomically disadvantaged children. The current study not only examined the transfer effect of the app training programme but also explored the sustainability of the training effect (immediate vs. one year after the intervention).

4.1. Findings from the app task-switching training group

The present findings confirmed the effectiveness of app task-switching training for cognitive enhancement of disadvantaged children. There were two major findings. First, the current study found evidence for a substantial transfer effect of task-switching training to a structurally similar new switching task. The mixing cost of the app task-switching group significantly improved at both posttest and one-year follow-up. This finding is in line with earlier studies examining other age groups with switch task training (Karbach & Kray, 2009; Kray et al., 2012; Soveri et al., 2013; Zhao et al., 2018; Zinke et al., 2012). Mixing cost represents the difference in mean reaction time between single task and mixed task blocks. The reduction in mixing cost signifies that students from the training group spent less time in reconfiguring the demands of the task and also in overcoming interference effects in mixed-task blocks than students from the inactive control group. There may be two reasons for the reduction in mixing cost. Previous studies explained that extensive cognitive training may make a skill more automatic, and cause changes in brain activation that are suggestive of this automaticity (Rogers & Monsell, 1995). It is also possible that learning across training sessions would equip students in task-set anticipation, which results in cognitive load reduction in the face of task-set conflicts and stimulus ambiguity (e.g. Jausovec & Jausovec, 2012; Olesen et al., 2004).

On the other hand, switching cost represents the difference in non-switch and switch trials within the mixed task blocks (Karbach & Kray, 2009). The fact that switching cost did not reduce at a follow-up stage is probably due to high demands made on cognitive control induced by task uncertainty (i.e. the absence of task cues). Unlike mixing cost that involves sustained effects and is thought to reflect a more global ability to maintain multiple task sets, switching cost is attributed more to transient control and depends on carry-over effects (Philipp et al., 2008). This is because these affect both relevant and irrelevant task sets during the process of overcoming interference from a previously activated task which is no longer valid. Plus, this also involves the reconfiguration of a new task set (Philipp et al., 2008).

Although the overall accuracy rate of the app task-switching group dropped at the posttest, it returned to the pretest level at one-year follow-up. The drop-in accuracy rate at posttest might be due to the increased cognitive load associated with the intensive training, which possibly affected students’ performance over time (e.g. Endres et al., 2015). Another possible explanation might be a speed-accuracy trade-off as participants responded more quickly at posttest after repeated training, but with more errors (e.g. Samavatyan & Leth-Steensen, 2009).

The second result is that the task-switching training had a beneficial effect on students’ performance on a structurally dissimilar EF task ‒ Inhibition. This training effect was found even at one-year follow-up and was not attributed to age as no training effect was seen in the other group. To understand how far-transfer effect took place, some studies suggested that task-switching training facilitates attentional control, a prefrontal cortex (PFC)-mediated ability that fosters knowledge and the transfer of skills to novel contexts (Sabah et al., 2018). Other studies suggested that content-related features and the intervals between training sessions (also known as the spaced training) might be crucial factors in producing the amount of far-transfer effect. The current study examined a four-week training paradigm focused on the ability to alternate between two cognitive tasks (transportation task vs. number task). Similarly, the Stroop test contained two dimensional stimuli (i.e. name vs. ink colour) and involved incongruent trials in which the task-irrelevant feature has to be inhibited (when the word does not match the ink colour). Although there are a few important differences (i.e. switching demand and nature of interference), both task-switching and the Stroop test required rapid and flexible processing of visual input, reinforced activation of relevant task demand and the inhibition of irrelevant information (Koch et al., 2010). The current study provides additional support that training-related improvement may not be specific to structurally similar task mechanics, it may generalize to other stimulus-response modalities that share a similar cognitive structure (Kattner et al., 2019; Miyake et al., 2000; Zhao et al., 2018).

In fact, inhibition is closely linked to classroom learning. Good inhibition skills make it possible for students to sit still, pay attention, and follow rules (Lyons & Zelazo, 2011; Marcovitch et al., 2008; Zimmerman, 2008). Inhibitory control has been reported to be significantly weaker in poor children in several studies (Evans & Kim, 2012; Hackman et al., 2010; Noble et al., 2015). Future research is needed to determine the factors that enable this transfer. Since transfer distance is an important factor for evaluating training programmes, it seems that this type of task-switching training is suitable for promoting not only one, but several executive control abilities.

Consistent with previous research, there were no beneficial effects on academic performance after training. Transfer of training to academic abilities depends on the training regime and the characteristics of the study sample (Titz & Karbach, 2014). Some studies (e.g. Traverso et al., 2019; Weissheimer et al., 2019) reported the far-transfer effects of EF training to academic abilities implemented in much longer training protocols (e.g. six sessions for two hours). It is also possible that younger children are less likely to develop their own strategy without being trained in strategy development. In a meta-analysis by Powers et al. (2013) on the efficacy of intensive app training, they pointed out that although there were significant overall effects for all age groups, the transfer effects for the youngest group was comparatively less. In other words, young children may require more support on strategy development and application after training (Holmes et al., 2009; Klingberg et al., 2005). Moreover, it is possible that the training effect on strategy development needs time to take effect or may vary depending on the level of the strategy developed by the participant (Gibson et al., 2011). Previous research suggests that such changes may occur later, thereby requiring a long-term follow-up (Holmes et al., 2009).

4.2. Limitations

The present study has some limitations that must be addressed to inform future research. Firstly, the academic performance of the students was measured by teachers’ rating on three key learning areas (i.e. Chinese, English, and Mathematics) using an interval scale. Although easier to administer, these interval scales might be less suitable for detecting subtle effects or changes, especially in detecting a training-related improvement in academic performance. Future studies should use standardised achievement tests to examine the training-gained improvement among groups. Secondly, it should be noted that this present study replicated the EF tasks based on Karbach and Kray’s (2009) study. Researchers may wish to consider opting for the gold standard of training studies – training that refers to the most validated and most adaptive task-switching training task available for Chinese children. Thirdly, researchers should remain cognisant of the active control vs. inactive control groups for the game-training studies. While an active control group is superior to an inactive control group for examining the effectiveness of the intervention (Green et al., 2014; Karbach & Kray, 2009; Shawn Green et al., 2019), the active control group is suggested to be used when it shares the same expectation of improvement as the experimental group so that researchers could attribute differential improvements to the efficacy of the intervention (Boot et al., 2013). Mahncke et al. (2006) compared between an active control group and an inactive control group using a brain plasticity-based training programme. They found no difference in memory function, suggesting that both types of control groups may produce similar controlling effect. Nevertheless, researchers should be fully aware of the design limitations, the expectation effects, and the placebo effects in future intervention studies. It is hoped that with better designs, future research would provide more compelling evidence for the effectiveness of interventions. Finally, to address the potential role of training effects that might be relevant for the amount of transfer produced, replicating the current findings with a larger sample of participants is also recommended in future research.

5. Conclusions

Childhood poverty not only has negative outcomes on child development but also leads to an economic burden on society through reduced productivity and output and the cost of crime – and it increases health expenditures. To advocate for a long-term, effective poverty prevention initiative, the current study examined the potential of a simple app task-switching training intervention in influencing the cognition and academic outcomes of disadvantaged children in Hong Kong. The results of the current study may have potential theoretical and practical implications for future research in task-switching training. On the theoretical side, since the transfer distance is an important aspect for evaluating the effectiveness of the task-switching training programmes, future studies should examine to what extent training-based EF skills transfer to real-world applications such as improved academic performance. On the practical side, it is recommended that the task-switching training programme has a well-refined design which uses a larger sample should be tested. If this training programme is proven to be effective, it can be implemented in educational settings.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research project [project number: 2015.A5.015.15D] is funded by the Public Policy Research Funding Scheme from the Policy Innovation and Coordination Office of the Hong Kong Special Administrative Region Government.