In the early days, computer scientists conceptualized the computer as a tool for science and a new way of thinking (Von Neumann, 1988; Denning, 2017). Recently, researchers showed a vital interest in computational thinking after Wing published her seminal article on the importance of computational thinking in 2006. Specifically, in her article, Wing (2006) clearly explained the interconnection of computational thinking with many disciplines. Recently, research efforts emphasized the importance of introducing computational thinking in all scientific fields (Weintrop et al., 2016), literature (Burke and Kafai, 2012), arts, and music (Edwards, 2011). Additionally, among many other researchers, Grover and Pea (2013), Smith and Burrow (2016), and Yadav et al. (2016) strongly argued for the systematic integration of the teaching of computational thinking in school curricula.

Computational thinking is a problem-solving process that educators can use in all content areas across grade levels (Barr and Stephenson, 2011; Román-González et al., 2018). In particular, computational thinking, as a problem-solving process, refers to how to (a) formulate problems with the use of a computer, (b) organize data using abstractions, (c) automate solutions using algorithms, and (d) generalize a solution to other issues.

Currently, there is no universal definition of the construct of computational thinking (Angeli et al., 2016; Kalelioglu et al., 2016). Weinberg (2013) reported that 25% of published articles on computational thinking cite their different definitions. Since a review of all computational thinking definitions in the literature is out of the scope of this study, the authors adopted the well-accepted definition offered by Grover and Pea (2013). According to Grover and Pea (2013), abstraction is the skill of removing characteristics or attributes from an object or entity to reduce them to a set of fundamental elements. Abstraction and generalization have similar meanings as abstracts are generalized through parameterization. Decomposition involves breaking down complex problems into simpler ones. Algorithmic thinking involves devising a step-by-step solution to a problem and differs from coding (i.e., the technical skills required to write code in a programming language).

Additionally, algorithmic notions of sequencing and control flow are also considered essential elements of computational thinking. Finally, debugging is about recognizing when actions do not correspond to instructions, and fixing errors in a sequence of instructions. In this study, the authors examined two of the five aspects of computational thinking: algorithmic thinking focusing on sequencing; and the skill of decomposition.

Robotics engage children in a playful exploration of tasks aligned with developmentally appropriate practices for early childhood education (Bers, 2008). Additionally, play improves children’s ability for creativity and mental flexibility. In robotics, the manipulation of physical objects enables children to exercise their symbolic thinking and combine their ideas to generate motor responses (Fenson and Schell, 1985; Russ, 2004; Singer and Singer, 2005; Critten et al., 2022).

Piaget (1963) was the first to conceptualize that the manipulation of physical objects plays a significant role in the learning of all children. Specifically, according to Piaget (1963) and Bredekamp and Copple (1997), children realize the world that surrounds them by dynamically exploiting the objects that constitute it. In this regard, hands-on experiences enable children between five and 6 years of age to form a connection between a concrete representation (e.g., authentic materials) and a symbolic representation (e.g., abstract material, such as ideas; Clements and McMillen, 1996).

Accordingly, in this context, technological tools, like floor robots, can offer children a hands-on way of acquiring new knowledge (Papert, 1993; Reid-Griffin and Carter, 2004; Chaldi and Mantzanidou, 2021). Furthermore, Resnick (1995) states that the best computational tools are the ones that provide the learner with the affordance not only to comprehend the learning content but to transform it and apply it to new concepts and ideas.

Thus, learning with robots can be an attractive activity for young learners in early childhood education settings (Nouri et al., 2020; Chaldi and Mantzanidou, 2021). Children at 4 years old can develop computational thinking skills when they play with robots to solve a problem (Papadakis et al., 2017). Specifically, researchers reported in the literature that learning with robotic devices allows learners to engage in experimentation, research, programming, and problem-solving about things they encounter in their daily lives (Pugnali et al., 2017). Kim et al. (2015) indicate that the contact and use of educational robots in education are significant and influential in teaching STEAM as robotic devices offer opportunities for young children to experiment with knowledge-building activities (Falloon, 2016; Vidakis et al., 2019; Papadakis, 2020). Especially, Falloon (2016) stated that young learners learn to code through decomposition, problem analysis, and evaluation activities.

Accordingly, in the study, the authors used the floor robot Bee-Bot to engage preschool children in various activities to solve several tasks.

Prior research has shown that four-year-old children can successfully build and program a robot (Bers et al., 2002; Cejka et al., 2006; Wyeth, 2008; Sullivan and Bers, 2013). In addition, researchers have used Bee-Bots in research studies with impressive learning outcomes in developing children’s problem-solving skills (Bers et al., 2014; Elkin et al., 2016). The Bee-Bot is a small floor programmable robot in the form of a bee (Elkin et al., 2016). In addition, it is a screen-free coding toy robot that allows children to interact with it without the constraint of screen-time (Livingstone, 2016).

A Bee-Bot uses a LOGO-like programming language with simple programmable steps. It is a robotic device that is developmentally appropriate for preschool children (Beraza et al., 2010; Highfield, 2014). A Bee-Bot has four directional buttons placed on its back (see Figure 1), and children can program it to move forward or backward, turn left or right by 90 degrees, clear its memory, pause, and execute a sequence of commands (Highfield, 2014).

A Bee-Bot does not need additional hardware to work. Also, the large variety of floor mats used with Bee-Bots accommodates educators with many possibilities for designing lessons on different topics and learning objectives. However, a Bee-Bot lacks the support of the algorithm’s visualization imposing an extraneous cognitive load on learners’ mental resources (Angeli and Valanides, 2020). Therefore, teachers need to scaffold young children’s learning during their interactions with Bee-Bots by using appropriate tools that can act as external reference systems for reducing children’s cognitive load during problem-solving (Sarama and Clements, 2009; Angeli and Valanides, 2020). Accordingly, to remedy Bee-Bots’ shortcomings, the authors of this study designed and implemented scaffolding tools for supporting young children’s problem-solving with Bee-Bots.

Papert (1980) strongly supported the view that the concepts of robotics and programming are intertwined. Learners, through interplaying with robotics, develop an understanding of elementary programming concepts, such as sequence and cause-and-effect relationships (Fessakis et al., 2013; Kazakoff et al., 2013), as well as an understanding of more advanced concepts, such as abstraction, debugging, and iteration (Horn and Jacob, 2006; Wing, 2008; Brennan and Resnick, 2012; Horn et al., 2012). Papert (1980), in his seminal book “Mindstorms: Children, computers and powerful ideas,” documented the manifestation of the programming language as a learning tool for the development of problem-solving and reflective thinking. Through programming, children learn to solve problems systematically and express ideas using various tools (Sullivan and Bers, 2013). In particular, programming as a teaching approach can foster cognitive tasks directly linked to the development of computational thinking (Resnick et al., 2005; Grover and Pea, 2013; Lye and Koh, 2014; Macrides et al., 2022).

According to the Macrides et al. (2022), programming can be introduced to young learners as a stand-alone subject matter or integrated into the regular kindergarten curriculum with other subject areas such as music, movement and dance, art, science, mathematics, and literacy. In addition, recent advances in visual language-based systems using digital, tangible, or hybrid media have enabled the introduction of programming and computational thinking concepts and ideas in early-childhood education settings (Elkin et al., 2016; Bers and González-González, 2019). At the same time, research evidence shows that programming synergistically deepens the learning of STEAM subjects and that its teaching should start as early as possible in school at all educational levels and ages (Sengupta et al., 2013; Wilensky et al., 2014). Relevant review studies also supported that computer programming through robotics can be a promising educational tool for integrating technology and engineering in early childhood STEAM education (Jung and Won, 2018; Çetin and Demircan, 2020).

Wood et al. (1976) defined scaffolding as the support provided to learners by their educators in the correct quantity and time, enabling them to participate in a problem’s solution substantially. The construct of scaffolding is directly associated with the socio-cultural theory of Vygotsky (1978) and, more specifically, with his idea of the zone of proximal development. The zone of proximal development is “the distance between the actual development level as determined by independent problem solving and the level of potential development as determined through problem-solving under adult guidance or in collaboration with more capable peers” (Vygotsky, 1978, p. 86). Research studies report that scaffolding in the form of instructional support can use tools, strategies, computer tutors, and animated pedagogical agents to assist learners in comprehending and conceptualizing meanings emerging during teaching beyond their current mental and cognitive capabilities (Puntambekar and Hubscher, 2005).

Providing scaffolding is essential in educational settings because it can positively affect the teaching and learning process (Hogan and Pressley, 1997; Cole, 2006; Pawan, 2008), especially for young students (Belland, 2014). Originally, instructional scaffolding referred to the support or guidance provided by a human expert to a novice learner to complete a task that would otherwise be too challenging to achieve beyond the learner’s unaided efforts (Belland, 2014; Kim et al., 2018). However, due to the advancements of technology, scaffolding is no longer limited in interactions between a human expert and a learner; instead, such interactions now include the use of technological tools, resources, and environments (Cai et al., 2022). When learners’ limited cognitive resources are sufficiently supported and augmented by scaffolding, they can concentrate on and master the task at hand by completing only those elements within the range of their competence (Tawfik et al., 2018). Importantly, learners become genuinely involved with the mission to be achieved only when provided with appropriate help or a supporting structure to initiate and sustain interest (Belland et al., 2019). Subsequently, while the learners’ competence gradually increases, assistance is withdrawn, and learners continue to function independently. However, removing the aid does not diminish learning or functioning; learners operate at the elevated plane reached via scaffolding (Liu et al., 2020).

Consistent findings in the literature support the claim that gender differences influence student learning and school achievement (Pomerantz et al., 2002; Duckworth and Seligman, 2006; Sousa and Tomlinson, 2011). For example, studies in neuroscience (Bonomo, 2010; Sousa and Tomlinson, 2011) reported that the hippocampus, the brain structure where memory and language function, develops more rapidly and is more prominent in size in girls than boys. This fact causes differentiation in sequencing, vocabulary, and reading and writing skills between the sexes. In addition, the cerebral cortex, responsible for spatial perception, is more significant in boys than girls, supporting that boys learn better with visual-motor activities. Because of these differences, girls tend to be more social and prefer group-centered activities than boys, who like personalized and sensory-motor activities (Gurian and Ballew, 2003).

Research on gender disparity in acquiring computational thinking skills in robotics activities is relatively sparse in educational settings. While research has been conducted on gender differences with elementary and high school students (e.g., Atmatzidou and Demetriadis, 2016; Román-González et al., 2017), limited research has been carried out in early childhood settings (Sullivan and Bers, 2013; Atmatzidou and Demetriadis, 2016).

One hundred and seventy young children aged between five to 6 years old participated in the study. Children were randomly selected from 10 different preschools in a European country. Parents signed consent forms for children’s participation in the study. There were 78 girls and 92 boys. Participants were randomly assigned into three research groups, two experimental groups, and a control group, as shown in Table 1. In the first experimental group, children used the model-based scaffolding technique; in the second experimental group, they used the code-based technique. The authors explain both types of scaffolding in sections 4.3. Research intervention for phase 2: Model-based scaffold and 4.4. Research intervention for phase 2: Code-based scaffold. In the last group, the control group, children worked through play with no scaffolding. All research phases were conducted at the site of each school individually with each child because scaffolding tools are exceptionally efficient in individualized learning environments (Tabak, 2004; Puntambekar and Kolodner, 2005). The same researcher completed all of the tasks with the children.

During the first phase of the intervention, children in all three groups became familiarized with the essential functions of Bee-Bots, followed by learning complex programming commands. The structure of each lesson was designed based on the theory proposed by Hubwieser and Aiglstorfer (2009), which approaches the teaching of programming into two parts: the learning of basic process units and the teaching of combined process units. In this study, the basic process units corresponded to the primary sequences of the Bee-Bot’s commands, taught in 11 subtasks, as shown in Table 2. This approach has many benefits as novice programmers (a) are successfully introduced to basic programming commands (Misirli and Komis, 2014), and (b) learn to program successfully even during their initial attempts (Reges, 2006).

The combined process units consisted of complex and more extensive sequences of commands. For learning these sequences of commands, the authors adopted the principles reported by Armoni and Gal-Ezer (2014). In line with their work, the problem-solving tasks consisted of subtasks with increasing levels of complexity. More specifically, the total number of the commands FORWARD/BACKWARD and the total number of the TURN_RIGHT/TURN_LEFT commands increased gradually and were used in various combinations, as shown in Table 3.

As shown in Figure 2, this scaffolding tool included a model (a representation on a smaller scale than the real one) of the actual floor mat, the robotic device Bee-Bot, and laminated cards depicting each programming command. The child had to first think about the algorithm and construct it using the laminated cards on the floor and the model mat. After that, the children tested their algorithm on the actual mat using the Bee-Bot.

As shown in Figure 3, the code-based scaffolding tool incorporated a cardboard and a set of laminated cards to model each Bee-Bot command. First, children were asked to look at the actual mat and then, without simulating the moves of the Bee-Bot, use the laminated cards to form a visual sequence of commands for the Bee-Bot to execute. Next, the children displayed the visual algorithm on a large cardboard, and finally, they tested it using the actual mat and the Bee-Bot.

The researchers used three different problem-solving tasks in the three research phases. Each task/phase was accompanied by a relevant mat, as shown in Figure 4. Children had 20 min to solve a problem-solving task with the Bee-Bot. The first problem-solving task (in the first phase) involved children in 13 subtasks to discover the functions of the Bee-Bot. The second task (in the second phase, which was the phase where the two types of scaffolds were used), which included five subtasks, taught children how to formulate sequences of commands at increasing levels of complexity. Finally, after the researchers removed the scaffolds (in the second phase), they used a third task (in the third phase) comprised of five subtasks to assess children’s computational thinking.

Research procedures consisted of three research phases conducted individually for each participant. There were 3 days elapsed time between each research phase. The duration of each phase was 20 min.

In Phase 1, children learned the basic commands of the Bee-Bot and, more specifically, as shown in Table 4, how to program the Bee-Bot using short sequences of commands.

During Phase 2, children worked with tasks that required sequences of four to seven commands. Some examples are given in Table 5. In Phase 2, the researchers scaffolded children’s learning using the scaffolding tools in the two experimental groups. The control group worked through free play and exploration with no scaffolding. During the last phase, Phase 3, the researchers removed the scaffolding tools and used the third problem-solving task to assess children’s performance.

The researchers videotaped children’s interactions with Bee-Bots during all research phases of the study. Specifically, the researchers recorded children’s command choices to complete each problem-solving task in various attempts. The research data were analyzed using the process coding method by Saldaña, 2015, which is suitable when the observed interactions include problem-solving. At first, both researchers coded four videos from each group to ensure standard grading procedures. Then researchers coded the videos independently. Interrater reliability was calculated and found to be 0.87 (denoted as the percentage of agreement between the first and the second researcher), which is considered satisfactory for problem-solving tasks. Regarding the effect of gender and scaffolding tools on children’s computational thinking performance, the researchers performed two 2 × 3 analyses of variance, one for phase 2 and another for phase 3. In addition, the authors performed a 2 × 3 repeated measures analysis of variance to determine between-subjects effects for boys and girls across the experimental and control conditions.

A vast majority of studies under the umbrella of computational thinking have used rubrics for their assessment (e.g., Sherman and Martin, 2015; Basu et al., 2018; Basu, 2019). In line with Basu et al. (2018) methodology, the researchers herein collected data from all one hundred and seventy students’ solutions for each problem-solving task and then identified whether students solved the tasks correctly on their first attempt or in further attempts. The researchers developed a rubric to assess students’ computational thinking performance based on this holistic analysis. As shown in Table 6, the authors of this study evaluated student performance in (a) the number of attempts in terms of command sequences students formed to program the Bee-Bot and (b) whether students used decomposition to solve a task.

The descriptive statistics shown in Table 7 indicate higher scores for the male participants than their female counterparts in each group. The descriptive statistics also show large standard deviations indicating that there is a lot of variance in the observed scores around the mean. A large spread indicates that there are probably large differences between individual scores that might be related to important initial problem-solving differences among the preschool students who participated in the study.

A 2 × 3 analysis of variance determined a statistically significant difference between boys and girls on the different forms of scaffolds administered in Phase 2. The results revealed that the type of scaffold [F (2, 164) = 64.60, p < 0.01] was statistically significant for the scores of computational thinking. Post-hoc comparisons showed that both model-based scaffolding and code-based scaffolding outperformed the control group, with no statistically significant differences between them. The findings did not show any other statistically significant main effect or an interaction effect between gender and type of scaffold.

During the third research phase, when the researchers withdrew the scaffolds, boys outperformed girls in all groups. According to the descriptive statistics shown in Table 8, boys in the model-based scaffolding group outperformed boys and girls in all other groups. As in Table 7, the descriptive statistics in Table 8 also show large standard deviations indicating a wide variance in the observed scores around the mean. A wide variance indicates, as before, that there are probably large differences in the individual scores among the preschool students who participated in the study.

The researchers conducted a 2 × 3 analysis of variance to investigate the differences between boys and girls and the different forms of scaffolding strategies. The findings showed that gender had a significant main effect [F (1, 164) = 9.06, p < 0.01] in the computational thinking score, revealing the superiority of boys in this study. The results did not indicate any other statistically significant main effect or an interaction effect between gender and type of scaffold.

The authors also performed a 2 × 3 repeated measures analysis of variance to examine the effects of the two scaffolding techniques on boys’ and girls’ computational thinking across Phase 2 and Phase 3 of the research. The results showed statistically significant between-subject effects for type of scaffold, F (2, 164) = 11.03, p < 0.01, in favor of the model-based scaffolding in Phase 2, and gender, F (1, 164) = 10.63, p < 0.01, in favor of boys who performed at their best in Phase 3.