Designing Inclusive Computational Thinking Learning Trajectories for the Youngest Learners

Abstract

Foundational thinking for later use of technology, particularly coding, is necessary for an inclusive and sustainable future. Inclusive practices beginning in early childhood recognize children’s innate development of computational thinking—sequencing, repetition and looping, debugging, decomposing and composing, representation, and causality. This qualitative research describes processes of developing and evaluating hypothesized developmental progressions. Inclusive engagement of children with and without disabilities is described in examples for each level of each developmental progression. Implications for teaching and learning with inclusive practices are described for children with and without disabilities.

1. Introduction

Recently, the field has seen a resurgence in the promotion of computational thinking (CT) skills among young children. Seen as an important way of fostering digital competencies along with critical thinking, creativity, and problem-solving abilities, proponents believe CT can also lay a strong foundation for future work in STEM domains. The STEM Innovation for Inclusion in Early Education (STEMI²E²) center seeks to develop and validate models of high-quality inclusion of young children with disabilities in STEM learning, as well as creating and formatively evaluating learning trajectories [1] in the STEM fields. That is, through a partnership with inclusive early childhood programs and early interventionists working directly with families, the center is iteratively generating hypothetical learning trajectories for topics in science, technology, and engineering; that is, developmental progressions with aligned learning experiences. This study describes the initial results of developmental progressions following the testing of a learning model for CT.

Early in the process of defining the T in STEM for the STEMI²E² project, we decided that CT would be the focus of our development of technology-focused learning trajectories and models for inclusion. Although we use multiple technologies with teachers and with children as appropriate, especially assistive devices, we decided the most fecund focus for learning lay in the generalizable competencies of CT. For our audiences of very young children, especially those with disabilities, these were deemed more valuable than other aspects of technology, such as computer use or identification or vocabulary of technological hardware and software. This aligns with other definitions of technology education as the introduction of underlying ways of thinking used to build or create technology—the basic logic underlying computer science.

1.1. Computational Thinking and the Youngest Learners

CT has been defined in a variety of ways. Seymour Papert [2] first used it to describe not just the procedures and concepts involved in the solution of problems and the design of computers as systems, but also the application of these procedures and concepts in the perception and comprehension of natural phenomena [3]. A later seminal definition from Wing’s work was, “CT is the thought processes involved in formulating problems and their solutions so that the solutions are represented in a form that can be effectively carried out by an information-processing agent” [4]. Similarly, others tie the definition to the use of digital devices, such as a “set of thought processes that allows framing and solving problems using computers, robots, and other computational devices” that involve a wide array of cognitive abilities, including but not limited to elements of pattern recognition, conceptualization, sequencing, planning, and problem-solving [5] (p. 84). Others list the “big ideas” of CT, such as abstraction and simulation, systematic data processing and modeling, facilitating knowledge creation, algorithms, decomposition/modularization, iteration, recursion, parallel processing/thinking, constraints and efficiency, networks, and systematic error detection/debugging [2,6,7,8]. Papert [2], for example, contrasts the common but inefficient practice in math where children tend to erase all their work and restart or ask for an answer if an answer is incorrect, to the programming practice of carefully tracing one’s code to find and fix the “bug”.

The Committee on STEM Education of the National Science and Technology Council [9] offers the following: “Computational thinking, including computer science, is not just about using computing devices effectively; more broadly, it means solving complex problems with data, a skill that can be learned at an early age. It seeks to expand the use of digital platforms for teaching and learning, because they enable anywhere/anytime learning; make possible individualized instruction customized to the way each person learns most effectively; and can offer more active and engaging learning through simulation-based activities or virtual reality experiences” (p. vi) and “CT encompasses a set of processes that defines a problem, breaks it down into components, and develops models to solve the problem, then evaluates the result, iterates changes, and does it again. Although the concept was developed in computer science (CS), it is increasingly seen as a set of broadly valuable thinking skills that helps people solve problems, design systems, and understand human behavior, and that can be learned at a very young age without involving computer coding. In an increasingly technological and complex global economy, CT needs to be an integral element of all education, giving every learner the capacity to evaluate information, break down a problem, and develop a solution through the appropriate use of data and logic (pp. 23–24)”.

The Committee goes on to proclaim that the U.S. should “make CT an integral element of all education” (p. 22). They argue “that digital literacy empowers people with the tools to find information, answer questions, and share ideas, and that they need to understand how to use these tools responsibly and safely” and thus we must “advance CT as a critical skill for today’s world” [9] (p. vi).

Unfortunately, there has been limited research connecting younger children’s learning and development with the skills and competencies expected of children in K–12. This is in part due to the lack of articulation of the overall learning goals for children from birth to age 5, the nature of the competencies at these early ages, how they build and interweave during this stage, and the ways in which these levels of thinking can be supported through interactions with instructional activities, adults, and peers.

Nevertheless, a growing body of research indicates that children as young as 4 years of age can begin to acquire CT competencies, even when those are restricted to the use of digital devices [1,5,6,7,10,11]. For example, programming or coding, instructing a computer to follow a set of commands, is possible for children as young as preschoolers with the proper technology environment and guidance from teachers [1,12,13,14]. Preschool-age children are often the youngest age group to have CT experiences or to be studied for their CT skills. More research is needed on the potential foundations for CT from infancy to age 6. Our youngest burgeoning computer scientists may be showing us their abilities, but the observation of these developmental progressions needs a clearly focused lens that includes children as young as infants and toddlers across defined learning topics. We sought to build and refine such lenses around CT processes by hypothesizing a progression, designing instruction, and testing learning trajectories that include goals, developmental progressions, and instruction.

1.2. Inclusive Instructional Design Centering on the Learning Trajectories Approach

Teaching grounded in students’ thinking and learning is most likely to develop concepts, skills, and creativity and do so for all children by recognizing and building on their strengths. To do so, research suggests ensuring that learning progresses along research-based paths [15,16,17]. Learning trajectories are composed of three components: a goal, a developmental progression, and instructional activities [18]. To attain a certain competence in a given domain or topic (the goal), children learn each successive level of thinking (the developmental progression), aided by instruction (activities and teaching practices) designed to build the mental actions-on-objects that enable thinking at each later level. These three components address the three questions of the powerful pedagogical strategy of formative assessment: where are you trying to go? (the LT goal), where are the students now? (a level in the LT developmental progression), and how can you move students to the next levels? (the LT instruction) [19,20,21]. In this way, learning trajectories can facilitate developmentally appropriate teaching and learning for all children [17,22], including children with disabilities [23].

Component 1: Goals of Computational Thinking. There is considerable consensus on the core concepts and processes of CT, including sequencing, repetition, conditionals, decomposition, and debugging [24,25,26,27]. Some add pattern recognition, mapping similarities and differences among problems, allowing people to make predictions and create solution strategies, as well as abstraction, distinguishing between relevant and irrelevant details. A different perspective is that, although these processes are important, there are fewer CT abilities that apply to different domains but are rooted in the particular concepts and processes of the domain. That is, people do not develop “better abstraction” abilities and then apply them to domains, but rather their ability to abstract emerges from learning concepts and practices within each domain. See the related work by Seymour Papert [28] entitled, “You can’t think about thinking without thinking about thinking about something”. Thus, a continued challenge to building the research base for computational thinking is making reasoned choices about how to study the overlap of topics and disentangle evidence of children’s learning within and across topics—choices that may necessarily differ across developmental periods from early childhood to later childhood and beyond.

Component 2: Developmental Progressions within Computational Thinking. Teams of researchers created “emerging” developmental progressions for five areas of CT (these were based on specific concepts more than the “levels of thinking” in the math developmental progressions mentioned above). For each, the researchers conducted two coordinated, extensive reviews of the research literature on what CT concepts students in grades K–8 should learn and what they did learn from engagement with CT activities yielded learning trajectories for the topics of sequence, repetition, and conditionals, decomposition [25,26,27,29]. However, these progressions have not explored the foundations of children’s capacities for learning computational thinking skills from birth. This gap in the research literature is particularly problematic for inclusion, as a lack of information about where children might start their learning limits possibilities for supporting what they might achieve; that is, exploring the earliest opportunities for supporting computational thinking allows for building on children’s existing strengths, regardless of where they might start.

Component 3: Instructional Activities Scaffolding Computational Thinking Activities. Game design and robotics can serve as an effective means for the iterative exploration of CT, as they are motivating and engaging for children and introduce computer science [30,31]. Visual and tangible programming experiences are often followed by exposure to higher-level programming languages [6]. However, building foundations of CT for the youngest children, aged from birth to age 5 requires high-quality experiences that are goal-oriented in terms of the content and competencies to be learned but less structured in terms of the specific tasks children are being asked to “accomplish”. That is, young children respond with curiosity and motivation to explore interesting tasks, but highly structured and imposed tasks may hinder their natural creativity. Regardless of the content of experiences, the essential aspect of instruction within a learning trajectory is that they are not simply isolated activities. Each learning experience should be designed with the intention that the activity leads somewhere new in children’s thinking. Thus, instructional experiences within a learning trajectory are inexplicably linked to the overall goal or topic and a level within the topic’s developmental progression.

Further, teaching computation thinking, as with all domains, must ensure that all children, including children with disabilities, have access to those experiences. Thus, adaptations to the environments, materials, and instructional strategies must be designed from the beginning to the end of the research and development process for full inclusion [32]. The segregation in STEM teaching and learning between pedagogies “for” children with and without disabilities is problematic for both early childhood education generally and for the education of children with disabilities. Lambert and Tan [33] found that “research on math teaching and learning that included disability was less likely to be theoretical (14% qualitative) and more likely to be empirical (81% quantitative)”. Only 5% of those articles included preschool-age children with disabilities [33]. To develop truly inclusive and sustainable efforts for teaching computational thinking, a research base defining the goals for all children is needed. This will generate more specific work about what interventions work well and for whom across STEM domains, including computational thinking. Thus, rather than focusing early efforts on specific disabilities, an inclusive approach will focus on how the same goals can be approached while incorporating adaptations to the environment, materials, and instruction. Thus, an inclusive instructional design with the learning trajectories approach uses two types of adjustments to support individualization—[1] adaptations that target accessibility for children with specific needs and [2] adjustments to instruction that scaffold learning along a developmental progression [34]. Importantly, the goal for all children remains the same—as defined by the first component of the learning trajectory—with children varying on where they are on a progression—the second component. Adaptations are only needed to remove barriers to instruction that move children along the developmental progression.

2. Using Theory and Research to Create Nascent Learning Trajectories

To create research-based experiences in CT for young children, we followed the Curriculum Research Framework [35], which has been used successfully to create high-quality strengths-based instructional activities [36,37,38]. The framework suggests iterative processes of literature review, hypotheses of developmental progressions, creating and testing instructional activities, leading to continued iterations of both the progressions and associated instructional activities.

2.1. Gathering and Synthesizing Extant Research

To develop the nascent learning trajectories of CT for young children, we examined literature aimed at both researchers and practitioners to inform the three components of learning trajectories. Although little has been written about computational thinking for the youngest learners, and even less about inclusive approaches to teaching CT, a targeted search, often of research conducted with older students, yielded some guidance in both areas. We then synthesized relevant research on particular topics in computational thinking—the goals of CT. The review of literature included empirical research, position statements, and standards focused on technology. We used the research to create “emerging” developmental progressions for five areas of CT. Across each age group, the skills, behaviors, and types of thinking children were expected to or observed to engage in were listed. This produced a beginning explanation of advancement through levels of computational thinking—an initial developmental progression.

These developmental progressions then underwent a Delphi process and were revised. The Delphi process involved a variety of experts and practitioners who provided feedback on the concepts and goals that should focus efforts to support young children’s development and learning across age groups. Using the themes from the expert feedback, the results were used to revise the developmental progression levels and associated age groups. Finally, project staff provided additional comments and suggestions to create an iteration to be evaluated, a process described in Section 3.

2.2. Goals and Developmental Progressions

The goals and developmental progressions for computational thinking describe our observations that began with young children’s innate or intuitive understandings and moved toward more specific and generalizable knowledge the children gained with experience. Specific language around these developmental changes is frequently used in the labeling of the developmental progression levels. For example, “noticer” is often used to describe a foundational level of thinking wherein children’s interaction with a phenomenon uses their five senses. Similarly, the term “recognizer” is often used when children not only experience a phenomenon but seem to have enough implicit knowledge about the phenomenon to anticipate it or respond to it. This is contrasted with the use of “identifier” as a level of explicit thinking in which children seem to understand something about a phenomenon with little external guidance. For the sake of brevity and clarity, each learning trajectory’s literature and developmental progression is briefly outlined in this section, then the evaluated and revised progression is presented in tabular form in the following section.

2.2.1. Sequencing

Sequencing is essential to the teaching and learning of computational thinking, as programming entails the thoughtful sequencing of rules. One definition of algorithms in the CT literature is “precise step-by-step plans or procedures to meet an end goal or to solve a problem” [39].

Thinking about or planning a sequence can be applied to events or actions for the young child. Toddlers may attempt to stack rings of ordered sizes, receiving immediate feedback when a smaller ring has gone on before a larger ring. Children build toward an understanding that a sequence will produce a known result as they experience predictable routines, such as putting socks on and then shoes before going outside. They may also experience events that do not need to be ordered, allowing for efficiency of thinking by ignoring irrelevant information.

Although much of the attention on CT has been focused on resources for high school students, new tools and activities directed to elementary children have also proliferated in recent years with the creation of new coding environments (e.g., Scratch Jr., Kodable, HopScotch), activity sequences (e.g., Blockly used by Code.org), apps (e.g., Daisy the Dinosaur), board games (e.g., Robot Turtle), and tangible computing tools using programmable robots (e.g., KIBO, Bee-Bot, LEGO WeDo). In one study, students from five to seven years of age in Australia learned about building and programming LEGO robots through modeling, exploring, and evaluating plans [40]. Similar results have been found with other robots for children as young as kindergartners, with positive effects on CT, such as choosing and sequencing commands [41]. In another algorithmic thinking study, children had to specify a set of steps to help a character navigate to a destination. They relied on trial and error, teachers’ questions and support, and peer interactions to help analyze the results of their steps and to experiment with changing the order systematically. These children, 3–6 years of age, gained more than a control group on sequencing [42]. Children 5 and 6 years of age who engaged in programming in Logo to direct a digital turtle to create geometric paths [43] also improved in sequencing and problem-solving [12,44,45].

These studies provide numerous examples of young children engaging in sequencing and algorithmic thinking by directly interacting with digital and non-digital programming activities. Many other questions remain, such as how early young children can begin sequencing and at what ages they can begin to add complexity to their sequences. These complexities may include increasing the number of steps in the sequence, representing the steps in more difficult abstractions—from clearly representative pictures to more abstract symbols or code. Further research is needed to understand whether sequencing can be divided into developmental sections of sequencing alone and sequencing with conditional or loops. Answering these questions by gathering more evidence about the progression of young children’s thinking about sequencing will edify the teaching and learning of this important component of CT.

Initial progression steps around sequencing included children recognizing when steps are needed, replicating steps, reorganizing steps, and planning steps. These aligned with the literature described above in which children were asked to think about and plan sequences, including whether an order was needed. The literature also suggested that the developmental progression should include the evaluation of a series of steps, the representation of steps, and the results of steps. However, the evaluation of steps could also be categorized as part of debugging because finding and fixing errors begins with evaluation. Similarly, the representation of steps could also align with literacy, as letters are to reading and numerals are to math. Likewise, the results of steps could be categorized in causation and conditionals. To further complicate matters, foundations in young children’s early experiences of predictability in daily routines were expected to lead to these ways of thinking but are not frequently thought of in terms of computational thinking. Each of these complexities created an opportunity for thought partnership in which steps could remain, be removed, added, or re-organized into another more appropriate progression.

2.2.2. Repetition and Looping

Recognition of repetition is important in many contexts and tasks, as exemplified by mathematics practice standard 8: Look for and express regularity in repeated reasoning [46]. This implies the importance of pattern recognition (considered by some as a separate topic)—recognition of a situation or structure as “seen previously” and generalizing the repetition. In CT, this recognition of repetition develops into quantifiable loops and the recognition and use of loops to gain an efficient cumulative effect.

Children’s understanding of loops in computer programming may have psychological roots in infancy as children learn about their ability to repeat an action and modify those repetitions. For example, an infant may bat at an interesting toy in their play gym and realize it produces an interesting sound, which prompts them to repeat their action to continue producing the interesting sounds (e.g., “circular reactions”, [47]). Babies prefer novel and random patterns after habituating to a constrained pattern [48]. Infants as young as 5 months learn to anticipate spatial regularities and tend to disengage from visual–temporal redundancy [49]. These studies indicate that infants possess cognitive structures that notice and respond to repetition from a very young age. As children develop, they can notice repetition in things they see and hear, with pattern recognition developing into a more complex understanding of mathematical functions [1]. Thus, the goal of using loops as efficient components of computer code has foundations in the infant’s experience of repeating and modifying interesting sounds, body movements, or manipulation of objects. This continues to develop into more complex actions, such as using a representation of instructions to repeat actions.

The topic of repetition as a learning goal should move children toward an understanding that instructions that repeat can be represented more efficiently as a loop [25]. A review of lessons for children in grades 1–5 indicated that teachers targeted skills of repetition by the second grade, asking children to find repetition and use loops as more efficient commands [50]. It is unknown at what age children begin to understand repetition and looping in the context of generating instructions.

Thus, the initial progression steps around which this goal’s instructional activities were created included repetition of actions in infancy, modifying repetition as toddlers, and recognizing repetition in preschool. The progression includes looping, particularly as an end goal, with little guidance in the literature as to when children might be able to use or value looping as an efficient strategy. Additional challenges of whether to include different types of loops, as well as the extent to which levels of repetition and looping need to incorporate representation exist. The process of designing and testing instruction for these proposed levels quickly resolved some of the more complex issues, as described in Section 3.

2.2.3. Debugging

Infants and toddlers may use implicit understanding and trial and error to work toward self-defined outcomes, such as moving around an obstacle to get to a desired object. Toddlers show a brain response to puzzles that are incorrectly assembled, indicating an early ability to detect errors [51]. However, toddlers also exhibit developmental immaturity in both visual processing and executive control, which is posited to explain the phenomena of toddler scale errors, in which toddlers try to interact with miniatures of familiar objects in the same way they would with the full-sized objects [52]. This research suggests that infants and toddlers may have innate abilities to recognize some errors while making and experiencing errors very differently from older children and adults.

For preschoolers, the ability to recognize errors is achievable by preschool and kindergarten children, but not attained by all without intentional teaching in math and computer science [1,53]. Understanding that there is an intended outcome for most tasks and being able to tell whether the actual outcome matches the intended outcome, is critical to successful debugging [8,53]. Once this is attained, young children tend to use simple trial and error for iterative improvement, without systematic data collection or solution strategies [53,54]. Given the opportunity, evidence suggests that preschoolers can analyze and solve simple debugging problems (e.g., robots given instruction via arrows and images) [7]. However, children ages 4–6 were observed to debug logical errors more easily than syntax errors—or errors that were represented by symbols [29].

Evidence also suggests that debugging skills can be improved with intentional teaching. Children as young as 3–6 years of age gained from a CT intervention, with more skills than a control group in sequencing, causality, and debugging [42]. In fact, Sullivan & Heffernan [55] suggest that the immediate feedback of tasks focused on an intended outcome, such as those in robotics, can motivate children to search for a reason and a way to fix the problem. Taking a broad conceptualization of CT that includes processes and practices that include coding and also cut across learning domains and contexts, debugging can be viewed as recognizing errors within directions, e.g., [2].

Thus, debugging was conceptualized as the recognition and repair of errors in early childhood. The progression was expected to move from recognizing errors for infants and toddlers, with the addition of repairing errors in preschool. The recognition and repair of errors were expected to vary with the complexity of the error and the amount of assistance needed. We also hypothesized that young children at the top of the age range we observed—age 5–6 years—might systematically compare an actual outcome to an intended outcome to recognize error, whereas young children might only recognize error through explicit feedback—either from an adult or from the environment producing feedback contingent on the child’s unintentional mistakes.

2.2.4. Representation and Abstraction

As applied to CT, representation is often thought of as representing variable codes or data within programming languages, as well as modeling and simulating data [56,57,58,59]. The foundation of using representation is to recognize and express that one thing can be represented by another—encompassing pictures, language, signs, gestures, and actions [55]. A further use of representation is its capacity to reduce complexity, or to abstract information to a higher level—such as asking for “all the fruit” rather than “the banana, the two apples, the four oranges, the twelve grapes, and the mango”. Part of this thinking is to choose only the relevant information to be represented or abstracted [60] and part of it is to understand the underlying tools that can be used to represent or abstract information [61]. Thus, representation could also be aligned with another commonly used term in computational thinking—abstraction.

Although the description of abstraction is frequently cited as “deciding what details we need to highlight and what details we can ignore” [62] (p. 3718) provides a much richer discussion of abstractions, and layers of abstractions, as mental tools. The focus on attention to detail may be an oversimplification if applied primarily to help children decide what should hold their attention. In contrast, an application of abstraction as a mental tool leads us to encourage children to highlight important information by giving it a representation that can be used with other information in a multitude of situations. Wing [62] also discusses one aspect of de-composition as understanding hierarchical layers of abstraction.

Bagge and Grover [63] define four levels of abstraction, or four levels of program development—problem, algorithm, program, and execution. That is, programmers can generate descriptions of the problem they want to solve, descriptions of the steps they will take to solve it, assign abstract code, or representation to each of these steps, and finally execute the code to test that they have solved their initial problem [63]. These steps can be implemented in many ways and in different, potentially iterative, ways through many styles of thinking, learning, and knowing [64]. Human thought about a plan is complex. Across computing, the representation of a plan is a creative and potentially personal process. Carefully abstracted representation of plans translates the important details into useful tools for communicating and eventually coding.

The earliest foundations for this type of thinking can be observed with infants and toddlers using any object to represent another, such as a block that is pretend food [65]. A primary tool they soon use is representation in the form of language, including sign language and the use of augmentative and adaptive communication (AAC) devices, as well as more complex representations such as pictures and symbols, including letters and numerals. Case studies suggest that children use combinations of representations to reduce the cognitive load of translating across levels of abstraction [66], and as an internal dialogue that facilitates their ability to represent abstractions for external use [57]. For example, to translate their stated plan (represented by drawing a map) to a set of code cards (another representation), children used a combination of language and concrete actions (gesture, moving the objects in the problem being solved) to process and complete the task [66]. However, the use of logic to plan, evaluate, and problem-solve may come later than their use of symbols or syntax to represent their logic [29]. This follows a theme of earlier implicit knowledge as a foundation to explicit knowledge, analogous and possibly intertwined with the earlier development of receptive language followed by expressive language.

Toddlers use pretend play that may later translate to symbols such as computer codes that represent a set of actions so that these sets of actions can be understood, broken down, and re-arranged for efficiency or to meet different goals. Thus, an initial hypothesized developmental progression began with using one object to represent another. As this domain of thinking matures, more complex representations can be used and embedded in increasingly complex structures, combining and re-mixing multiple representations in programming languages. That is, these representations, or abstractions, which begin with pictures of objects, language for objects, and symbols for action are organizers of thoughts and plans. These become important in children’s CT because they allow for the most salient pieces of information to be identified and used, while less important details are ignored. Thus, symbols as instructions may transition children into the use of codes as programming commands. We also wanted to test whether and to what extent young children could use a symbol or code to represent a set of codes—using a hierarchy of abstraction. The application of these symbols and hierarchies into variations of instructions and plans moves us to the next topic—decomposition and modularity.

2.2.5. Decomposition and Modularity

Decomposition can be defined as a process of breaking down problems or tasks into smaller and more manageable pieces (“bite-sized pieces”, [2]). What pieces are broken down and what is done with these pieces involves other CT skills and concepts. Specifically, Brennan and Resnick [67] pair modularization with abstraction by emphasizing parts of commands and sets of steps, while pairing decomposition with the re-use and re-mixing of commands. The re-use and remixing of components more closely align with the definitions of Grover and Pea [6] of modularization, which they pair with decomposition by focusing on structured problem-solving. To synthesize this, we could consider that modularity fits better with abstraction when composing compositions to create a larger vision, but with decomposition when breaking down components. Both emphasize the nature of code as pieces or chunks of information. It is this chunking of information that structures the pattern recognition discussed in repetition and looping.

Modularity is included in the K–12 for the earliest ages, and one report claims that it is not yet backed by research on young children’s abilities outside of coding contexts [54]. A study of CT noted that young children needed specific scaffolds to identify parts and wholes [54], which aligns with children’s development in mathematics [1]. In mathematics, recognizing parts and wholes is an important foundation in the early years, with children able to decompose shapes and numbers before kindergarten and can break down a multistep problem to solve it in parts in the primary grades. However, there is little research that establishes the relationship between such decompositions and computer programming, or even whether modularity in computational thinking and decomposition in math are similar cognitive constructions. Further, children may not be motivated by the efficiency of modularity [8,54], indicating that simpler problem decomposition may be most useful in the early years.

Thus, the hypothesized developmental progression for decomposition focused on identifying component parts, or parts of a whole, and applying this knowledge to modify or recompose in ways that solve problems or complete tasks. It began with recognizing that problems or tasks could be broken down, then addressed whether children could break down problems or tasks—either with help or independently. We also wanted to test the extent to which children grasped that components of a problem or task might be re-arranged, the final step in the initial progression.

2.2.6. Causation and Conditionals

The use of causation and conditionals in CT requires an understanding that one event can cause another, that a state can be evaluated as true or false, and to extend true and false evaluations to causes of different events. A suggested goal for learning conditionals is “Each of two conditional states may have its own action” [25]. Children may be able to make inferences about causality, with older children developing the capacity to represent and understand conditional representations [55]. A suggested LT for conditionals hypothesized that students understood, “A conditional connects a condition to an outcome” before “A condition is something that can be true or false” [68]. Thus, conditionals may help connect the notion of true or false statements to the concept of causality.

Across age ranges, some evidence suggests the development of causality concepts. Infant’s and toddler’s ability to understand causality may exist as young as 8 months old. For example, infants’ eye movement indicates anticipation of a previously caused event to occur again and children at 19 and 24 months of age repeated actions that they had observed to cause a novel experience [69]. Evidence even suggests that toddlers (18–30 months) can track multiple causes [70]. Even very young infants are learning about cause-and-effect relationships in their daily interactions as they explore the world around them, such as when a baby bats at a toy hanging from their play gym and notices that it causes a sound every time they bat at it [71].

Children 3–6 years of age improved their understanding of causality, indicating that causality learning is sensitive to experience [42]. In the context of robotics, children in kindergarten were able to infer why a robot would stop [55]. Children in first and second grade applied causality to a representation of instructions [66]. They increased their understanding of control structures to communicate cause–effect and conditionals through an intervention, with first graders improving significantly above a control group [66]. However, conditionals seem to be quite difficult for children to understand [66]. More evidence is needed to understand how early and through what mechanisms these inherent abilities can be supported to grow into applications of conditionals in CT.

The initial developmental progression hypothesized for this goal included infants and toddlers noticing causality, preschoolers beginning to engage in causal reasoning, and kindergarten connecting true and false conditions to causality. The behaviors we thought we might see when testing this hypothesized progression included children repeating an action that had caused a novel event, children identifying what caused a phenomenon, and older children identifying what might happen given two different conditions.

2.3. Developing the Instructional Activities

Following the process of generating initial developmental progressions across the topics of sequencing, repetition and looping, de-bugging, representation, decomposition, and causation and conditionals, learning experiences were developed or adapted to elicit and scaffold learning across these topics. Several types of activities were developed for different purposes. First, plans for teachers to implement were developed, embedding a plethora of information and resources for adapting to children’s needs, providing STEM vocabulary, and extending learning in other contexts. Whenever possible, these were adapted from the research or wider educational literature and fitted with or modified to address the given level of the learning trajectory. Others were created anew, starting with the mental actions-on-objects that constituted the pattern of thinking specified in a specific level [35]. These longer lessons were authored and reviewed initially by staff with expertise in STEM content, early childhood education, and children with disabilities, then a final review was completed by leadership. Shorter “mini-investigations” were also developed for classrooms and early intervention providers, tested, and rapidly revised based on testing. The design of activities provided children with opportunities to try these skills while allowing researchers to observe the children’s level of thinking. The design goal of activities was to create an experience that “… allows, or obliges, the child to externalize intuitive expectations. When the intuition is translated into a program it becomes more obtrusive and more accessible to reflection” [2] (p. 145).

3. Formative Evaluation of the CT Learning Trajectories for Young Children

To formatively evaluate the hypothesized learning trajectories of CT, learning experiences were piloted in inclusive contexts. Learning trajectories should help teachers identify children’s levels of children’s thinking and plan further experiences and discussions. Thus, if the hypothesized learning trajectories were complete in their description and ordered correctly, the observed levels of thinking should be coherent with the hypothesized levels. Similarly, the instructional activities should, for children at level n, support their advancement to level (n + 1). Evidence and counter-evidence were collected to revise the developmental progressions.

The goals of the testing and revisions included simplifying and clarifying for ease of use, identifying examples of development across the ages, eliminating redundancy within and across the topics, and estimating the order of developmental steps based on observations. We organized refined versions of the developmental progressions following testing of instructional activities. The hypothesized progressions are ordered from top to bottom by those that were observed to come earlier in development to those that come later. Levels of topics are given names to aid memory and use of the levels, especially for teacher professional development.

3.1. Sample

Children with and without disabilities were recruited from inclusive early childhood classrooms in North Carolina and Colorado and home-visiting settings in Pennsylvania called “incubator” sites. The center-based settings included 19 classrooms across 13 centers. The home-based settings included 10 early intervention practitioners working with 1–3 families each. Eligibility criteria for center-based settings included having children with diagnosed disabilities enrolled and receiving services. Home-based practitioners were eligible if they were providing services specific to meeting the needs of children with Individualized Education Plans (IEPs) for school-age children or Individual Family Service Plans (IFSPs) for children under 4 years of age. Additionally, the project sought practitioners with a high level of enthusiasm for the project for this early phase of pilot testing. The specific disabilities of children in the project are not part of the data collection, as efforts toward inclusion focused primarily on what children were able to do. However, the types of accommodations generated focused on removing barriers based on whether children could reach or manipulate materials, engage in the key concepts of tasks, and respond in multiple ways to cover ability differences in gross motor, fine motor, cognition, and communication.

3.2. Procedures

Research staff collected videos of the implementation of instructional activities and generated field notes. Video of children’s natural, spontaneous activities that demonstrated the levels of thinking in the developmental progressions were also collected as evidence. These activities were shared in staff meetings to reflect on the levels of thinking children may or may not be exhibiting while engaging in the activities. Analysis of what was engaging to children, scaffolds that were effective in moving children’s level of thinking, and ideas to improve the activities were also shared. Additional documents included a “Questions and Wonderings” document that structured field notes, ideas for testing, and questions for each topic and level. Lastly, a checklist of the hypothesized developmental progressions was used to note what levels of thinking were observed or not observed and to document evidence and counter-evidence (or non-evidence) of the levels of thinking.

3.3. Revisions of the Developmental Progressions

3.3.1. Sequencing

The testing of sequencing led to six levels of progression. These ranged from the earliest levels in which toddlers took two sequential actions to meet an independently defined goal or by following directions from an adult. Later levels of thinking showed an understanding that a symbolic sequence can be decomposed and re-arranged. The revised progression is shown in

Table 1. Developmental progression of sequencing.

Level Name	Description	Observations
Sequence Completer	Completes two-step tasks, prompted or unprompted.	Pretended to wash hands, then got food for a baby doll.
Step Recognizer	Recognizes that there are steps in a sequence.	When asked what she must do when pretending to go grocery shopping, the child said “You get a shopping cart. You get some food. You go home”.
Simple Sequencer	Creates and completes an ordered sequence of steps using simple representation.	Child sequenced a set of two arrows to give directions through a maze. Child used visuals to complete the steps to make a pretend pie.
Complex Sequencer	Creates and completes an ordered sequence of steps using complex representation.	Child used blocks as codes to represent a sequence of dance moves. Child put arrows in order, creating directions to show another child how to reach a hidden treasure in a maze.
Sequence Planner	Plans steps to take to perform a sequence prior to creating or completing steps. May check steps for errors.	Child said “forward, forward, forward” before laying three forward-facing arrows on the maze. Child narrated directions and pointed the correct way through the maze before laying arrows out to give directions.
Early Decomposer	Understands that there may be more than one sequence of steps that complete a task.	After verbalizing a plan to turn at the end, the child put the turn arrow at the beginning, indicating that she understood that both ways would work. Child initially said a 6-step path was the fastest, but when challenged to make a path with 4 steps they did.

, with examples from our research for each level.

The progression changed from the previous by removing, re-organizing, and adding levels. We removed recognition of the need for steps and re-worded it to describe children implementing or describing a sequence. This rewording reflects the evidence we observed. One way we tested children’s understanding of whether steps were needed was to take actions in the “wrong” order—putting food on the table, then a plate—which also aligned with our plans to test debugging. Thus, recognizing the need for steps was incorporated into de-bugging. We also removed language about replicating steps because children’s ability to complete a small number of steps did not seem contingent based on whether verbal directions were given, the steps were modeled, and the child replicated, or if the child had an idea independent of directions. Regardless of where the idea came from, toddlers could complete purposeful sequences of two to three steps to meet their own needs and goals, which were sometimes aligned with the goals of the adults. We labeled these goal-oriented toddlers Sequence Completers. Planning sequences remained in the progression and were observed in preschool children. Proposed additions included following sequences in verbal directions and expressing knowledge of a sequence. However, as noted, following directions differed little from completing an order of steps that was familiar—for example, 1. get the doctor kit, 2. pretend to give the baby dolls shots. Expressing knowledge of a sequence did seem to occur later than children who could complete a sequence—with children aged between two and three years describing or pointing to sequences. Some children disengaged when asked questions about the order of events, while other children seemed to think questions like “Could you go home, then get your groceries?” were silly. We labeled this group the Step Recognizers.

Moving from completing and expressing sequences, we also tested children’s use of sequences when incorporating representations. Children in the 2–3 year age range were able to engage in short sequences represented in pictures (e.g., a picture of an oven to represent the step of “put the pie in the oven”). We called this level of thinking Simple Sequencer while labeling the next level Complex Sequencer. Children in the 3–5 year age range could engage with longer sequences of 3–5 steps and, primarily in classrooms of children aged 4, children could specifically create and use sequences with abstract representations (e.g., blocks represent dance moves). Another distinction between the two was seen in responses to a grid maze in which children were supposed to help the cat get to a snack. Simple Sequencers often added directions (arrows) directly on top of the map, where they wanted the arrow to lead. Complex Sequencers could use more distal methods—arranging all arrows in a line beside the grid maze.

We observed more accuracy, confidence, and independence in the use of representations as sequences in classrooms with preschool children than in children in younger age groups. This included generating directions that required children to translate their memory of symbolic directions to the execution of commands. This process moves children closer to cognitively accommodating similar processes of coding and testing the execution of programming commands, particularly as they plan steps as Sequence Planners and begin to see the consequences, or lack thereof, of re-arranging steps of a plan as Early Decomposers. The re-arranging of steps is a foundation for later modularity, where children may think of a set of steps as one unit. Then, it is the unit itself that can then be usefully rearranged as a single component in a larger set of directions.

Important lessons about the teaching and learning of sequencing were also uncovered. Implicit knowledge about sequencing can be observed in children’s responses to adult prompting about “What is next?” Some children needed orientation as to where the sequence should start and end, an order that may differ by culture. Children with experience seeing adults point to text as they read left to right may assume a left-to-right orientation, but many did not. When given opportunities to orient sequence cards themselves, a guide, such as Velcroing cards into a board, was helpful both for orientation and accommodation of developing fine motor skills. Similarly, children working toward the level of Sequence Planners needed scaffolding, such as allowing children to begin with the start of an existing plan and work to complete it. This type of scaffolding borrows from strategies that are known to be effective with children with disabilities—graduated guidance—beginning with a high level of support and reducing it when children are ready [72].

3.3.2. Repetition and Looping

Interest in repetition and variation in expected repetition can be seen in infancy, such as a common experience of playing “peek-a-boo” with a baby is sure to bring giggles and coos. This innate ability to attend to repetition and variation may be the earliest foundation of computational thinking. Our revised progression moves from this repetition of actions to the explicit understanding that repeating a set of tasks will have a specific and quantifiable effect, as shown in

Table 2. Developmental progression of repetition.

Level Name	Description	Observations
Action Repeater	Intentionally does something more than once and attends to repetition in the environment.	Child knocked over a block tower, then signed “more” to ask the adult to repeat the action.
Repetition Recognizer	Expects repetition after observing it.	Children pretended to drink tea, then held cups out for adults to fill repeatedly.
Simple Looper	Completes a task repeatedly until a natural end is reached.	Children picked up blocks and handed them to a teacher until all blocks were put away.
Complex Looper	Completes a set of tasks repeatedly until a natural or defined end is reached.	When pretending to be a cashier, the child repeated a set of actions (move items on the conveyor belt, scan to get a beep, place item in bag) for each item until all items were gone.
Cumulative Looper	Identifies that many loops can lead to a cumulative effect.	When filling a number of gift bags, the child counted down as each was completed—“I finished that one. Now there are two more to go…”

.

The initial progression was revised by joining two levels described as noticing repetition and modification of repetition. The new level, Action Repeater, is the foundational level of the progression—a level that typically develops without intentional support from adults. We also moved levels described as anticipating or expecting repetition to an earlier location in the progression. Expecting recognition was observed in the behavior of toddlers and aligns with expectations of implicit rather than explicit knowledge of toddlers. This level was labeled Repetition Recognizer.

Early revisions, generated during the creation of activities, included removing types of loops such as open and closed loops, as well as clarifying the definition of loops as repetitions with a given endpoint. We observed and tested looping in children’s behavior when loops came to a natural end and could not be done anymore, as well as loops that were defined by rules such as “jump three times then stop”. We observed some variation between younger and older children, with children able to repeat one or two tasks with a natural end around age 2. These children, Simple Loopers, repeat actions as if they are single actions but repeat these actions to meet a goal. This may include an implicit understanding that their repetitions move them toward their goals—such as repeated scoops of sand leading to a bucket full of sand. The Complex Looper, in contrast, may group a set of tasks into a unit that repeats. Our evidence suggests that children can repeat a known set of tasks or use the representation of a set of tasks to repeat. However, we do not consider the use of representation to be essential to this level of thinking. In fact, we observed counter-evidence that children would recognize, use, or appreciate the efficiency of using representations of loops rather than creating representations of all repetitions of all tasks. Three expected levels related to representation were removed from the progression. The final level of the progression then became the Cumulative Looper, wherein children were observed to explicitly recognize the usefulness of their repeated actions.

3.3.3. Debugging

Tasks intended to elicit debugging of directions were successful only with the oldest children observed. Thus, the final step in this progression is labeled Early Debugger. However, before being able to fix an error in directions, young children were able to identify errors in someone’s actions and identify errors in directions.

Table 3. Developmental progression of debugging.

Level Name	Description	Observations
Simple Error Recognizer	Identifies a simple error in someone’s actions by recognizing that an intended outcome has not been reached	When an adult made a silly error, putting a circle play-doh face on top of the play-doh eyes and mouth, said “If you put the face on top it’s gonna be yucky.” When an adult put glasses on Mr. Potato Head, then eyes, the child said “No”, and pointed to the eyes to indicate that they saw an error.
Simple Error Fixer	Identifies a simple error in someone’s actions and makes a change—reorganizing two steps into correct order when out of order or fixing an error with adult help.	When the adult made the error of putting the lid on a play-doh bucket, and then putting the play-doh on top of the lid, the child said “Let me show you.” The child put the play-doh inside the bucket, then put on the lid. After initially putting play food directly on the table, the child was asked if they should put down the plate and then the food. The child responded by moving the food from the table to the plate.
Guided Error Recognizer	Recognizes more complex errors in someone’s actions with help. Help may come from a variety of sources—an adult, comparing to a nearby example, or remembering experiences in the context of a familiar task.	Pointed out errors in dressing a stuffed doll (a familiar task) but did not recognize errors of an adult who dressed a flat felt doll. Laid out a set of arrows in a sequence to guide an adult through a maze. Recognized error only after adult followed sequence and did not get to the maze’s end.
Error Identifier	Identifies errors in a representation of a sequence but may not be able to fix complex errors	Pointed to an error in a sequence representation (symbols of directions), saying that it would not take their friend to the treasure.
Early Debugger	Identifies and fixes errors in directions by checking directions, comparing them to an example, or observing unexpected results of executed directions.	Fixed errors in a representation of a sequence while watching how the representation was used to move through a maze. Placed arrows in a sequence to lead a friend through a maze, but noticed she misplaced an arrow when she checked her work, then corrected it.

shows the hypothesized progression following testing, with examples of thinking and behavior observed at various age ranges.

The initial progression was focused on recognition and repair, with variations by the complexity of errors and scaffolding needed. The revision of the progression clarified the types of errors recognized by children. Specifically, we could observe that children between 18 months and 3 years were able to recognize when something was wrong with familiar objects or events. That is, the Simple Error Recognizer responded differently when an event or object was different from what they would typically expect than when it was what they expected. For example, children had experience dressing dolls and identified the adult putting the dolls’ socks on over their shoes as an error—children said “Stop!” as directed or answered Yes/No when asked “Did I make a mistake?”. However, the same action taken with less familiar two-dimensional people was not identified as an error. We also observed that children between 2 and 3 years of age could fix errors adults made. The Simple Error Fixer could show an adult the “right way” to do a familiar task after the adult demonstrated an error. Children as young as 3 began to find errors in more novel and complex tasks when specific feedback was available, including tasks that were represented in symbols. Feedback for the Guided Error Recognizer could include an adult prompting the child’s thinking or other environmental input such as the intended outcome not being reached. It should be noted that the “intended outcome” for young children could be fluid or forgotten—creating an interesting instructional challenge. The next level of thinking—Error Identifier—noticed errors in directions with little to no help but was not able to identify a way to fix the error consistently. As noted, the finding and fixing of errors in representations of action plans (i.e., directions) were observed infrequently and only with preschool-age children.

3.3.4. Representation and Abstraction

Despite the importance of representation in CT, its inclusion as a separate topic was problematic. Although the more complex skill of abstraction could be targeted as a learning goal, representation as used by young children is integral to all learning activities. Thus, separating levels of a representation progression from other aspects of CT would produce unnecessary complications for use.

Instead, representation was built into the other progressions. Sequencing includes levels in which children follow verbal directions, then follow and use symbolic directions later. Representation is also integrated into later levels of Debugging when children fix errors in representations of actions (directions on paper) rather than noticing and fixing errors in people’s actual actions.

The simple representations included pictures that looked similar to the objects they represented. Symbols such as arrows were a moderately difficult representation of movement. The more abstract representations necessitated a codebook, where blocks represented specific actions children would take. As such, representation was conceptualized as a tool of comprehension rather than a concept itself. This conceptualization will, of course, differ from the computational thinking of older children who may become proficient in specific programming languages.

3.3.5. Decomposition and Modularity

The goal of decomposing systems and plans, breaking apart a problem to simplify its analysis, may begin with an initial understanding of parts and wholes of objects or sequences. The testing of investigations intended to elicit decomposition provided more “non-evidence” or counterexamples of decomposition than examples. Toddlers were able to compose objects. Younger preschoolers were able to compose single representations of a direction, such as using a block representing a dance move or an arrow representing a step in a specific direction. Some older preschoolers were even able to re-arrange these representations or suggest different paths that would take the robot to the treasure, indicating an early understanding of modularity. However, none of these observations were examples of decomposition, breaking down a task into smaller pieces. Thus, decomposition was removed as a topic, with the expectation that these skills develop after a stronger understanding of sequencing. That is, a strong recognition of steps needs to occur before children build on this thinking to break down and re-arrange steps. These skills also require executive functioning competencies associated with planning thus developmental maturity is likely part of moving toward goals of decomposition. Thus, decomposing is referenced as the last step in the developmental progression for Sequencing—the Early Decomposer.

3.3.6. Causation and Conditionals

The testing of causation and conditionals suggested that children’s use of this concept between birth and the age of 5 years is not specific enough for CT to address it in this topic alone. Instead, we used the evidence for these investigations to improve the progression of the cross-cutting concept of “Cause and Effect”. Additionally, we view cause and effect as more relevant as it applies to children’s observations. In the context of CT, we think of causation and conditionals as they apply to a structure of directions that children can give or follow.

Specifically, the aspect of causation that we viewed as most relevant to CT was the use of conditionals. For example, children were sometimes encouraged to complete tasks with visual organizers showing First/Then statements that are similar to “If/Then” conditionals—first put on your coat, then we will go outside. While important, this single level of thinking did not give us framework for a developmental progression. Although the oldest preschoolers did at times recognize that a condition connected to directions – such as when a child decided that directions did not need to be changed because the robot found the treasure—thiswas rare. Further, this example is about a child recognizing a state (an observation) rather than using information about a state to control the next set of directions. This also connects to the limitation of our repetition and looping progression, as children did not use directions, such as a numerical parameter, to determine that a condition had occurred and the repetition could end (e.g., add two things to a bag; when 5 bags are packed, stop).

4. Discussion

Young children can benefit from learning computational thinking concepts and skills, both for foundational understandings of knowledge critical in an increasingly technological world and for increased comfort with and capacity for creativity and problem-solving. The three learning trajectories of CT developed and refined in this project—Debugging, Repetition and Looping, and Sequencing—borrow from the extensive work of scholars and educators and were adapted and refined with the results of our observations and interactions with very young children, including children with disabilities. Other recent frameworks for CT include (1) Concepts (control flow, representation, and hardware/software), (2) Practices (algorithmic design, decomposition, abstraction, iteration, pattern recognition, debugging, and generalizing), and (3) Perspectives that cover the social-emotional skills that facilitate successful application of CT [73]. We agree that the topic of representation belongs in concepts, as it is a structure for instructions, rather than the planning process itself. Our work aligns primarily with the practices of this framework.

In an attempt to create a shared language for teachers that is specific enough to be applicable while parsimonious enough to be memorable, perhaps practices [73] could be considered as two domains—(1) algorithmic design and (2) optimization. We consider sequencing to align with algorithmic design because the purpose of the practice is to understand and plan what to do. Decomposition is an extension of algorithmic design and the last step in our Sequencing progression, as understanding that a sequence can be broken down and rearranged is a later development of thinking about sequences. Other researchers [66] describe algorithms as steps in a sequence, with modularity described as decomposing multiple steps. As skills develop from sequences that are acted out to sequences that are represented in symbols that can be broken down and re-ordered, the complexity of children’s efforts begins to more closely represent the skills applied in computer programming—a primary application of algorithmic design. To optimize these designs, understanding the connection between expressed sequential plans and the execution of plans transitions into examining and fixing plans when a lack of expected outcomes or output occurs in the execution of plans.

Optimization, as a second domain of Practices [73], would include what we have described as Repetition and Looping, as well as Debugging, but what Zeng [73] described as Abstraction, Iteration, Pattern Recognition, and Generalizability. Specifically, when applied to optimizing an inefficient set of directions, one must find the pattern or repetition and then iterate or loop it to simplify the original. Likewise, when attempting to correct a problem in directions, one may generalize a solution that has worked in one area of the code to another, or one may use abstraction skills to review the intended plan against the written code to find the important details that may be missing or the extraneous details that could be removed.

Another reason that optimization may be an important domain to explore is a developmental pattern of optimization being quite intuitive in actions and observations, but quite difficult for young children in the context of representation. Debugging may begin with simple error recognition—often observed in children’s joyful identification of adults’ silly mistakes. In simple contexts, such as two-step sequences, young children can fix these errors. With guidance from adults, examples, or reminders of previous experience, children can fix errors in more complex situations—such as errors in processes with more than two steps, where one step should not precede another. Finding and fixing silly mistakes of teachers and family members can lead to finding and fixing complex sequences in a programming language, but our observations suggest that applying representation to this intuition requires experience and development. Other researchers noted a similar pattern [73,74]. Educational robotics supported executive function and problem-solving for young children, but coding activities had mixed effects, supporting problem-solving for first graders but not for children between 4 and 5 years of age [74]. The authors suggest that the physical aspects of educational robotics were more accessible to the younger children, while the representation and manipulation of instruction were more developmentally appropriate for first graders [74].

4.1. Implications to the Teaching of Inclusive Early Computational Thinking

Some CT topics, such as sequencing, are familiar within early childhood environments, whether practitioners think of it as computational thinking or not. First, with a better understanding of the foundational skills involved in CT, practitioners may reframe what they already do with their youngest children. Observation of teachers in incubator sites engendered our inclusion of effective adaptations, many of which have a direct connection to CT. For example, task orientation visuals are often used to remind children of what needs to be done now and what is coming next—“First, wash hands. Then, eat snack”. As applied to CT, “First-Then” visuals also teach about contingencies—with conditional states built into the idea that washing hands precedes a snack, so handwashing leads to a snack. The visuals help children understand the steps they need to take to meet their own goals. While washing hands is not often a goal for young children, snack time might be. Thus visuals can clarify that hand washing is a path, not a detour, to a desired state.

Visual schedules are used to provide predictability, letting children know about and identify the order of the day. Visual schedules also teach children about sequencing, with many opportunities for children to answer the question “What’s next?”. Another use of visuals is autonomy support. Choice boards may include what centers are available for children at different points in the day or what ways children can participate in groups (ex. when it is your turn you can tap on the drum, say your name, or sign your name). These differ from task-orientation visuals that orient children to what is expected (ex. Eat then brush teeth, but it is not a choice to skip tooth brushing). Autonomy-supporting visuals clarify the authentic choices children can make. The importance of helping young children practice making independent choices works well for all children and incorporating visuals can work particularly well for non-speaking children while using representation that structures computational thinking. Thus, rich and inclusive CT is already occurring naturally in the routines and activities within early childhood environments.

Once practitioners recognize their children’s CT, they can use their understanding of the progression to support children’s development in their CT skills to move from one level to the next. This has the potential to move practitioners from implementing isolated activities to engaging children in deeper, extended learning opportunities to foster growth in children’s CT skills. For example, during one classroom visit, a researcher observed a preschool teacher engaging with children at the grocery store set up in the dramatic play center. Playing the store clerk attendant, the child rang up the items his teacher was purchasing as he cranked the lever, rang the item up on the scanner, and placed the grocery item in a bag. As he went through this process, he also pointed to the visuals of the items to indicate the price. He continued to loop through this sequence until he rang up all the grocery items the teacher was purchasing. At a brief glance, this may seem like a typical dramatic play interaction in a preschool classroom, but with a lens for CT, we see rich opportunities for supporting the child’s understanding of looping, sequencing, and representation. For example, the “conveyor belt, scanner, bag” process could be visualized as a sequence and followed by other children who had no previous experience as grocery clerks. That sequence could then be represented as a loop indicating “Do this for all grocery items” to remind the child when to start and end. Thus, with knowledge of a CT progression, a common dramatic play routine opens up an opportunity for CT learning. Examples such as this were previously the minority within the CT literature, with a review of CT articles about children aged 2–8 from 2013 to 2021 indicating that less than 10% of articles used “unplugged” teaching strategies while educational robotics was the tool in more than half of the studies [73]. More recently, unplugged activities that teach CT without screens or robotics are appearing in studies [75,76].

Although most early childhood professionals want to implement STEM, one concern voiced by practitioners wondering about bringing STEM learning opportunities to young children is the developmental appropriateness of STEM for the youngest children. Perhaps for this reason, this study is also novel in the age groups to which CT was applied, as many studies of early childhood CT work with children older than age 5 [66,77,78], with preschool children being studied less often [74,79]. Developmental appropriateness, as well as content appropriateness, are important considerations given that STEM is broad and deep—generating a plethora of possibilities for frustratingly difficult or contextually meaningless experiences that can be labeled “STEM.”

The power of the learning trajectories approach, both the systematic process of developing LTs and the principle of beginning with what children know and can do, begins with development. That is, the center of a learning trajectory is a developmental progression, framed by content goals and realized in instructional activities. Thus, teaching STEM through a learning trajectories approach should always be developmentally appropriate.

Beginning with a goal can reduce the trap of doing fun STEM-adjacent activities with little intentionality. Noticing children’s thinking and considering how their behavior aligns with a level of developmental progression can guide the next steps of instructional decision-making. Using instructional activities that intentionally scaffold children’s thinking from one level to the next gives children experiences that fit within their zone of proximal development rather than being too difficult or too easy. The development and continued testing of these learning trajectories will support educational practices that bring appropriately challenging STEM experiences to young children with a diversity of strengths. When teaching children with disabilities, the STEM goal remains the same, while other goals such as IEP or IFSP can be incorporated with planning, creativity, and intentionality. At times, early interventionists may need to prioritize a functional goal. In that situation, learning trajectories might become the context in which intervention occurs. For example, a child with cortical visual impairment may be working on tracking objects; this goal might fit well in the context of repetition and looping.

4.2. Future Directions

Continued work in this area should further examine these hypothesized progressions through the implementation of instructional activities aimed at eliciting and scaffolding the levels of thinking. Although the STEMIE project made substantial progress in the creation and field testing of inclusive learning trajectories for STEM, the study is limited in scope and size. The scope of these phases of the project included knowledge and content development. The number of participants was, therefore, small and specific to the purpose of studying the products and processes of the work, rather than the quantitative impact of those products on people. Future work is needed to further evaluate and refine the learning trajectories, especially in the STE domains, extend the work to other topics in STE, and produce support materials for teachers, faculty and professional development providers, and families.