Participants were 680 preteens (M = 11.52; SD = 0.51), all attending the first year of lower secondary education in France (referred to as 6e, the equivalent of Grade 6 in North American system), and were enrolled by their teachers from 13 French middle schools. Among these middle schools, 3 were private and 10 public, 8 urban and 5 rural, and none was located in educational priority networks. Schools were recruited through a call for expressions of interest coordinated by a regional education inspector and disseminated to middle school principals. Interested schools volunteered by contacting the research team directly. All parents gave written informed consent for their children to take part in the experiment. This study was conducted in compliance with the Declaration of Helsinki and was approved by the local ethics committee (CER Grenoble Alpes, approval no. 2020-09-01-4). Thirty-four participants with a history of psychiatric disorders or diagnoses indicating special educational needs were excluded from the analyses, resulting in a final sample of 646 preteens (336 boys and 310 girls). At the start of each test session, we ensured that children requiring visual or auditory aids were using their corrective devices.

The tasks were written with the jsPsych library (de Leeuw, 2015), on top of the Experiment Factory,⁵ a container-based platform for running online experiments (Sochat, 2018). All code related to the CAM battery are available online at: https://gitlab.com/mpinelli/CAM.

The graphical design of the tasks is suitable for middle school students (11 years on average), and features a cartoon character named Martin who visits a zoo (Figure 1B). For each task, the instructions are displayed on the screen. The tasks include training trials followed by testing trials and automatically begin one after the other in a random order. During the training trials, feedback is given by Martin, indicating whether the response is correct or not.

Computerized attention measure (CAM): eight tasks of the CAM battery were used in this first study: Stroop, Flanker, Go/No-Go, selective attention, working memory, visual sustained attention, auditory sustained attention, and visuo-manual coordination (Figure 1A). The variables and measures of each task were selected to reflect the cognitive processes of interest, according to the literature on cognitive processes. For each task, we paired a gamified instructional approach with the cognitive task (Figure 1B). The parameters of the tasks described can be found in the Git repository. They were defined within the framework of the project but can be directly modified in the source files of each task.

Flanker task: participants were instructed to observe fish appearing on the screen. They had to respond as quickly as possible by pressing the directional arrows that correspond to the direction of the central fish flanked by two other fish on both sides (swimming in the same or in the inverse direction of the central fish). On each trial, a row of five fish appeared at the center of the screen for 1,500 ms, preceded by a 1,000 ms interstimulus interval (ISI). According to the condition (congruency of flankers), Flanking fish swam either in the same direction or the opposite direction. The task began with 4 practice trials, followed by 50 test trials. ER and RT scores were recorded.

Stroop task: in this task, participants had to count tickets to help the zoo’s head. The task entailed swiftly counting the items presented within canonical patterns, and participants were instructed to press the corresponding numbers on the keypad (ranging from 1 to 4) to indicate the count. Importantly, the identity of the items within the patterns (digits from 1 to 4) was inconsequential; the focus was solely on accurately quantifying the number of items present. Participants had to press the number pad as quickly as possible to indicate the correct amount of items (4 possible responses between 1 and 4). Each pattern was presented for 2,000 ms with an ISI of 1,000 ms. This task included trials for which the amount of items corresponded to the written digits and trials for which the amount of items did not correspond to the written numbers. Fifty trials were preceded by 4 training trials. ER and RT scores were recorded.

Go/No-Go task: participants were instructed to give a “candy” response (Go trial) when an animal appeared and to withhold the response when a caretaker was shown (No-Go trial). They had to press (or not) the space bar as quickly and accurately as possible. Items consisted of images representing either a caregiver or an animal. Items remained on screen until 500 ms. The ISI was between 1,250 and 1,750 ms. The task comprised 10 training trials and 80 test trials (No-Go frequency 25%). ER and RT scores were recorded.

Sustained attention tasks: the tasks were adapted into two versions: a visual and an auditory one. Participants either saw or heard animals and were instructed to count them. Depending on the version, they had to press the spacebar whenever an animal was seen or heard. At the end of the task, they reported the total number of animals they had counted. The ISI varied randomly between 1,500 and 4,000 ms. Two practice sequences were followed by 30 to 40 randomly presented test trials. The number of animals counted, ER and RT scores were recorded.

Selective attention task: the aim of this task was to find a target (a brown monkey) among distractors (other animals). They had to press the spacebar if the visual target was present among the set of distractors (filler animals). The target consisted of a round brown monkey and the distractors were brown square monkeys or round blue monkeys. Participants had to respond as quickly as possible by pressing the “P” key if the target was present, and the “A” key if the target was absent. There were 3 sets of 30 trials with either 8, 16, or 32 monkeys. Among the 90 test trials, there were “present” trials and “absent” trials, and 6 training trials preceded the test trials. The stimuli were displayed until the participant’s answer and with an ISI of 1,000 ms between trials. ER and RT scores were recorded.

Visuo-manual coordination task: the task required participants to play with an animal by sending a ball. Participants had to click on the animal that appeared at the bottom of the screen and move the mouse to the ball that appeared at the top of the screen (top left, top center or top right). The animal was displayed for 5,000 ms with a 1,000 ms delay between trials. The test involved 6 training and 30 trials. Movement Time (MT) and ER scores were recorded.

Working memory task: the gamified goal of the working memory (WM) task was to feed animals and recall food items. Participants had to memorize 5 ingredients and indicate, after each ingredient, the position of 3 animals (distractors) that appeared on the upper or in the lower part of the screen. For this distracting task, they had to press the “up” or “down” keys. At the end of each sequence, children had to recall the memorized ingredients via a recognition task in which the 5 target ingredients were mixed with 5 new food items. The task involved 2 training sequences with a low cognitive cost and 14 test sequences presented randomly (7 sequences with a low cognitive load and 7 with a high cognitive load).

Attention network test: the ANT, (Fan et al., 2002) was used to test the validity of the CAM tests. The ANT is a task that assesses three attentional networks (alerting, orienting, and executive). The task and instructions were adapted for use by preteens in the computer rooms of middle schools. Participants had to determine the direction of a central arrow (left or right) with cue arrows appearing above, below fixation, or not (depending on the condition). RT and ER scores were measured according to cue type (no cue, center cue, double cue, spatial cue) and flanker type (neutral, congruent, incongruent). The ANT involved 24 training trials with feedback and three experimental blocks without feedback for a total of 288 test trials. The three networks were computed as in Fan et al. (2002).

The study comprised two sessions (Figure 1C). In the first session, participants performed the 8 CAM tests, dedicated to evaluating reliability and feasibility. Before this session, teachers received coaching on the testing protocol and were instructed to address any technical problems. Teachers received a total of 2 h of training. The first hour focused on presenting the CAM battery and the overall study protocol. The second hour involved testing the full set of tasks in each school’s computer room under the supervision of a psychology researcher. This session aimed to identify any technical issues, train the teachers in task administration, and verify that data were being properly recorded. The experiment took place in classrooms equipped with individual computers for each student, complete with a keyboard/keypad, mouse and headphones. Session 1 lasted about 1 h, during which the 8 web-based tasks were administered randomly. Although there was no formal break during the session, a short pause occurred between each task while the next one was loading. The second session took place in the classrooms of the middle schools that had completed session 1. During this session, participants carried out the ANT, which took about 1 h and was scheduled approximately 1 month after session 1. While the whole experiment was supervised by a psychology researcher, a volunteer teacher managed the organization of the two group sessions in each class.

The completion rate corresponding to the percentage of participants who completed all the 8 tasks in session 1 varied across the 13 middle schools. The completion rate ranged from 29.17% to 100% with a total mean completion rate of 67.80% (see Table 1). The reasons for not completing the full battery were either technical issues that occurred during the testing session (poor network quality preventing tasks from loading and significantly extending the session time; see the middle schools concerned in Table 1), or insufficient time allocated by the teacher in the computer room. In session 2, a total of 100 preteens completed the session, which included the ANT.

Interference scores were computed using RT differences for the Flanker and Stroop tasks (Paap et al., 2020). For the Go/No-Go and selective attention tasks, condition differences were evaluated using ER scores (Meule, 2017; Woods et al., 2013), while the cognitive cost effect in the WM task was examined by considering the percentage of correctly recalled items (Barrouillet et al., 2007). RTs lower than 200 ms and higher than 3 SD from the mean were excluded, along with RTs from incorrect responses. Except for MT and accuracy-based tasks (i.e., selective attention, visuo-manual coordination, and WM tasks), participants with an ER exceeding 40% were excluded from all tasks. Furthermore, participants who consistently pressed the same button throughout all trials of a given task condition were removed. We present below the performance analysis of preteens for each condition of each task for the CAM battery. Due to data loss caused by technical problems, particularly affecting the selective attention and auditory sustained attention tasks, and the exclusion of outliers, some individual data were not recorded, leading to variations in sample size across tasks. The proportion of data loss was 49.07% for the auditory sustained attention task, 38.85% for the selective attention task, and ranged from 21.36% to 31.42% for the other tasks. Hence, we chose to present the performance results for each task individually. Performance data for each task are presented in Table 2.

Inhibitory control was assessed via the Flanker task, the Stroop task and Go/No-Go task, by calculating ER and RT scores. As shown in Figure 2, the distribution of ER scores presented differences between preteens and between the Flanker, Stroop, and Go/No-Go tasks. The mean ER score was 10.10% for the Stroop task, 16.54% for the Go/No-Go task and 9.15% for the Flanker task. Moreover, Flanker task and Stroop task elicited a significant difference on mean RT score between incongruent and congruent trials, F_{(1, 442)} = 154.76, p < 0.001, η²_p = 0.25 (mean congruent = 595.85 ms, SD = 96.12; mean incongruent = 627.11 ms, SD = 111.89; delta = 31.27 ms) and F_{(1, 478)} = 654.76, p < 0.001, η²_p = 0.58 (mean congruent = 890.79 ms, SD = 123.81; mean incongruent = 998.76 ms, SD = 116.15; delta = 107.97 ms), for Flanker and Stroop, respectively. Results also showed a significant difference in ER score between Go (M = 5.06, SD = 10.47) and No-Go trials (M = 28.01, SD = 17.16) [F_{(1, 507)} = 735.14, p < 0.001, η²_p = 0.59] (Figure 2).

Sustained attention was evaluated with the visual and auditory sustained attention tasks, for which RT and ER were computed (Figure 3). The ER for the auditory and visual tasks corresponded to the proportion of trials in which preteens missed a target (i.e., failed to press the “space” key when required). The mean ER was 5.26% and 8.27% for the visual and auditory version, respectively. However, only 40.12% of preteens produced the correct number of heard stimuli and 47.42% for visual stimuli. We nonetheless observed a significant correlation between RT and ER for the visual sustained attention task (r = 0.43, p < 0.001) and for the auditory sustained attention task (r = 0.17, p < 0.01). The more preteens missed targets, and the longer they took to respond (Figure 3).

Results on the selective attention task showed a significant interaction between the set size and the presence of the target [F(2, 798) = 69.55, p < 0.001, η p 2  = 0.15]. The interaction was decomposed into simple effects, with results reported using Bonferroni correction. Results showed a significant difference between present and absent condition with a set size of 32 items, t = 15.52, p < 0.001 (delta presence/absence = 12.66%), 16 items, t = 8.02, p < 0.001 (delta presence/absence = 4.83%) and 8 items, t = 5.47, p < 0.001 (delta presence/absence = 3.40%). In other words, the presence of a target had a greater impact on ER when the set size was larger (Figure 4).

WM performance was evaluated through the mean rate of correctly recognized items in the complex span task. Moreover, the ER score evaluated performance on the distracting task. These two measures were computed as a function of the cognitive load imposed by the distracting task (low vs. high). The ER score of the distracting task was added as a covariate to control any impact on the relationship between the cognitive load and the percentage of correctly recognized items. Results showed that memory performance did not differ across the two cognitive load conditions with 61.15% and 60.85% of correctly recognized items in the low and high cognitive load condition respectively [F_{(1, 505)} = 0.17, p = 0.67] (Figure 5). The ER score in the distracting task (24.06%) did not affect the cognitive load effect on correctly recognized items.

For the visuo-manual coordination task, ER score reflected the situations in which the target was missed. The MT corresponded to the time to move from the initial position to the target, according to the three target position conditions of the visuo-manual coordination task. Data distribution for the visuo-manual coordination task showed large individual differences in performance on both ER score and MT (Figure 6). The mean ER score was 19.49%.

Reliability was assessed for both response speed (mean RTs) and accuracy (correct and incorrect responses). The split-half method (R package “splithalf,” Parsons, 2021) consisting of randomly splitting 10,000 times trials into halves was applied. A correlation was computed for each split, and the reliability score was calculated as the mean of the 10,000 correlations and by applying the Spearman–Brown formula. Results of the split half for the RT and ER scores are presented in Table 3. With regard to RT scores, the results showed satisfactory reliability coefficients, ranging from 0.90 to 0.97. For ER scores, results showed acceptable coefficients (ranging from 0.61 to 0.90), but a low coefficient for the Go/No-Go task (0.48). Moreover, RTs difference reliability was low for Flanker and Stroop, respectively 0.29 and 0.03.

We first evaluated validity with a subsample consisting of preteens from middle schools who completed all CAM battery tests in session 1 and then participated in session 2 to complete the ANT (n = 100). Pearson correlation (significance threshold: α = 0.05) results are presented in Figure 7. Next, an exploratory factor analysis was carried out to study the latent structure of the CAM battery, with the ANT measures included in the analysis. A total of 25 variables were selected for convergent validity and exploratory factor analysis. These variables were drawn from the tasks based on measures used in the literature and their relevance for evaluating the cognitive processes under investigation. The selection aimed to balance speed-based and accuracy-based measures, while also limiting the number of variables included in the factor analysis in accordance with the sample size constraints (see Figure 7 and Table 4).

Effect sizes were interpreted according to Gignac and Szodorai (2016), with r = 0.10 considered small, r = 0.20 typical, and r = 0.30 relatively large. The results indicated that all RT scores from the CAM battery were significantly correlated with the ANT RT score, ranging from typical to relatively large correlations (0.22 < r < 0.61). In addition, a typical, significant correlation was observed between the ANT RT score and working memory performance in the high-load condition (r = −0.20). Regarding sustained attention tasks, the RT score for visual sustained attention showed a typical significant correlation with the ER score from the ANT orienting component (r = 0.21). However, neither the visual nor the auditory sustained attention tasks showed significant correlations with the ER score from the ANT alerting component or the alerting delta score. A typical significant negative correlation was also observed between the RT score from the selective attention task and the ANT orienting delta score (r = −0.20). Additionally, ER scores from the No-Go and Stroop tasks, as well as both RT and ER scores from the Flanker task, were significantly correlated from typical to relatively large with the ER score from the ANT executive control component (0.21 < r < 0.37). Finally, the ANT executive delta score was typical significantly correlated with the RT score from the Flanker task (r = 0.20).

Between-task correlations revealed significant associations between ER and RT scores across the Stroop, Flanker, and Go/No-Go tasks, ranging from typical to relatively large (0.24 < r < 0.51). A typical correlation was also found between the Stroop delta RT score and the ER score from the incongruent condition of the Flanker task (r = 0.21). Additionally, typical correlations were observed between ER scores from the selective attention task and ER scores from both the incongruent (r = 0.23) and congruent (r = 0.25) conditions of the Stroop task. RT score from the selective attention task was also correlated with RT scores from the auditory and visual sustained attention tasks, with relatively large correlations (r = 0.33 and r = 0.30, respectively). A typical negative correlation was observed between the low cognitive load condition of the WM task and visual sustained attention (r = −0.25). The ER score from the visuo-manual coordination task showed typical correlations with the ER score from the No-Go condition (r = 0.24) and the RT score from the Stroop task (r = −0.24), as well as with the delta RT score from the Stroop task (r = 0.26). The MT score from the visuo-manual coordination task correlated with the RT scores from visual sustained attention, the Go condition, Stroop, and Flanker tasks (−0.24 < r < 0.57). Finally, sustained attention tasks showed correlations with the six tasks of the CAM battery, ranging from typical to relatively large (−0.25 < r < 0.50).

The exploratory factor analysis was conducted on ANT and CAM tasks to determine the latent structure (Table 4). The number of factors was selected according to a parallel analysis. The analysis showed a 5-factor solution (using maximum likelihood as the extraction method and a varimax rotation) explaining 45.01% of the variance (only factor loadings greater than 0.35 are reported). The adequacy of the exploratory factor analysis showed that Bartlett’s test was significant (p < 0.001) with an acceptable level of sampling adequacy (Kaiser-Meyer-Olkin = 0.60). Most RT-based variables loaded onto Factor 1. The ANT loaded onto Factor 2, the WM task loaded onto Factor 3, Stroop task loaded onto Factor 4 and Flanker task loaded onto Factor 5. The Go/No-Go task loaded onto the same factor as the Stroop task (Factor 4). Selective attention did not load onto any factor.

Two hundred two preteens (M = 11.06; SD = 0.37, 99 girls and 103 boys) from French middle schools completed the CAM battery in this second study. No middle school was located in educational priority networks. Moreover, all middle schools were public, with 7 located in urban areas and 2 in rural areas. The recruitment procedure was similar to that of study 1. Regarding parental occupation, 16 (mother = 11, father = 5) were retired or unemployed, 126 (mother = 66, father = 60) were employees/workers, 100 were intermediate professions (mother = 53, father = 47), 51 (mother = 20, father = 31) were executive professionals, 25 (mother = 7, father = 18) were craftsmen, merchants, company managers, and 4 (mother = 2, father = 2) were farmers and 82 did not provide information. Written informed consent was given by parents. The data presented here correspond to a sample of students who completed all tasks of the CAM battery and did not receive any intervention. As in study 1, visual and auditory corrections were verified. Participants had no history of psychiatric disorders or diagnoses indicating special educational needs.

The material and measures of the present study were the same as those described in study 1, with adjustments made to the testing procedure. Since we introduced a new test evaluating divided attention online, we were required to discard the visual sustained attention task due to time constraints. Moreover, the WM task was adapted to address the lack of a cognitive load effect identified in the first study. Participants were required to press the “M” key (for “up”) or the “Q” key (for “down”) to indicate the spatial position of the stimulus, which increased the difficulty of the distracting task and enhanced the cognitive load effect. In addition, the number of sequences was reduced to 8 (compared to 14 in the first study) to shorten the task duration. Finally, auditory instructions were added for each task of the CAM battery allowing preteens to read and/or listen to the instructions at their convenience. All task materials and procedures are available in the Git repository.

Divided attention task: participants had to follow two goals when performing this task: to find a visual target (an animal) among distractors and to count how many animals sounds they heard (Figure 1A). There were 30 trials with a single set of 32 monkeys. Among the 30 test trials there were 15 “present” trials and 15 “absent” trials. Six training trials preceded the 30 test trials. The stimuli were displayed until 5,000 ms with an ISI of 1,000 ms. Animal count, ER and RT scores were recorded.

The general procedure for administering the CAM battery was similar to that of the first study, except that groups of 15 participants were preferred and provided with headphones to listen to the auditory instructions. Moreover, students completed the 8 CAM tests in the computer rooms during a single session.

All performance results were significant and consistent with those from study 1 (see Table 2). It should be noted that technical issues with the Flanker task led to variability in the number of trials across congruence conditions (descriptive data are presented as indicative). However, results remained comparable in terms of performance, reliability, and validity (the inclusion or exclusion of the Flanker task did not alter the factor analysis pattern of the other tasks). Results for the WM task showed no significant effect of the cognitive load on memory performance [F_{(1, 154)} = 0.03, p = 0.86], with 66.03% of items correctly recognized in the low cognitive load condition and 65.74% in the high cognitive load condition. Error rates in the distracting task did not influence the effect of cognitive load on item recognition.

Results on the new divided attention task (Figure 8) showed a significant difference between the two conditions [F_{(1, 154)} = 30.57, p < 0.001, η²_p = 0.16]. Compared to the condition in which the target was absent (13.72%), participants made more errors when the target was present (21.25%). Participants heard an average of 17.80 sounds.

Reliability results are presented in Table 3 and were consistent with those of Study 1 for both RT and ER scores, as well as for interference scores. Coefficients for RT scores ranged from 0.88 to 0.96, with similar values across tasks. In contrast, ER scores showed greater variability, with coefficients ranging from 0.39 to 0.88. The divided attention task showed a split-half reliability of 0.88 for RT score and 0.83 for ER score.

The variables from the CAM battery were similar to those in study 1, except that we included the presence condition of the divided attention target (RT and ER scores), as well as the recall variable (number of items recalled). Correlation analyses were conducted to examine the relationships between the divided attention task and other tasks of the CAM battery (see also the results in the Git repository). The ER from the divided attention task showed significant correlations with the ER from the Flanker task (r = 0.16), the RT score from the Stroop task (r = 0.16), and the ER from the selective attention task (r = 0.18). The RT from the divided attention task correlated significantly with all RT scores from other CAM battery tasks (0.17 < r < 0.45), except for visuo-manual coordination.

The same procedure as in study 1 was used to determine the factorial structure of the tests used in the present version of the CAM battery (parallel analysis with a factor loading cutoff of 0.35). Twenty variables were included in the exploratory factor analysis. Table 5 presents a 7-factor solution (using Maximum Likelihood as the extraction method and Varimax rotation), accounting for 47.36% of the variance. Bartlett’s test of sphericity was significant (p < 0.001), and the Kaiser-Meyer-Olkin value was 0.57. Except for the auditory sustained attention task, all speed-based variables loaded onto Factor 1. Visuo-manual coordination task loaded onto Factor 2, WM onto Factor 3, the Flanker task onto Factors 4 and 5, and the Stroop task onto Factor 6. Divided attention and selective attention tasks both loaded onto Factor 7. The No-Go condition of the Go/No-Go task loaded on the same factor as the Flanker task (Factor 4).