A systematic review of computational thinking in science classrooms

ABSTRACT

Computational thinking (CT) has been described as an essential skill that should be learned by everyone and can, therefore, be included in their skill set. Computational thinking uses essential principles in computer science for solving problems, understanding complex systems, and human behaviour. This way of thinking has significant consequences for teaching and learning science subjects at elementary and high school levels. In this review, we analyse and discuss the results from 23 studies and highlight the methodology, different strategies, and assessment practices used to promote the integration of computational thinking within science classrooms. We also give an overview of how computational thinking is being taught in science classrooms and describe tools available for teaching computational thinking in science instruction. Findings showed the value of using modelling-based pedagogy in incorporating key computational thinking skills within science instruction and suggests that educators should deploy effective technology tools to enhance the deductive and inductive teaching of science concepts using computational thinking framework.

KEYWORDS:

• Computational thinking
• science classrooms
• schools
• stem education
• systematic literature review

Introduction

As technology becomes a core feature of our daily lives and even more pervasive in the classroom, it is becoming evident that students must be equipped with skills that would enable them to think critically and proffer solutions to multifaceted problems using emerging technologies. Yet while schools may subscribe to the importance of teaching students to think critically and solve problems, there seems to be a lack of agreement on how, when, and what instructional tools to use in teaching these essential skills. Studies have however indicated that one approach to teaching these skills is to teach CT (Cansu & Cansu, 2019; Hickmott et al., 2018) since CT is acclaimed as an approach to teaching the essential 21st-century skill set that every student requires to thrive in the changing world (Sneider et al., 2014; Yadav et al., 2016).

Computational thinking was first described as a generic term referring to a set of computational ideas that people use to represent their work through the design of computer hardware systems, software, and computations (Papert, 1980), and later referenced as a process of procedural and probabilistic thinking in defining the relationship between a problem, its solution and data structuring (Papert, 1996). However, CT became widely accepted following Jeannette Wing’s description of CT to include the use of basic computing principles to solve problems, build structures and understand human behaviour (Wing, 2006, 2008) and the thought process in formulating problems and representing their solutions in a form that can be effectively carried out by an information processing agent (Wing, J.M, 2011). Although CT was intended to be a basic skill that complements reading, writing, and arithmetic for everyone (Wing, 2006), there is no consensus on a formal definition of CT (Barr & Stephenson, 2011; Cansu & Cansu, 2019; Voogt et al., 2015). However, due to the prominence being given to CT, there is a growing global interest in defining and implementing computational thinking in classrooms. Hence, computational thinking has been characterised in a variety of ways ranging from a collection of computer science principles and thought processes that help formulate problems and their solutions to understand the natural and artificial world around us (Aho, 2012; Mannila et al., 2014; Royal Society (The), 2012) and its relationship to the application of the high degree of abstraction and algorithmic approach in solving problem (García-Peñalvo et al., 2016). Moreover, studies have also suggested that CT can be described as a problem-solving procedure involving problem design, logical structure and evaluation of data, representation of data through abstractions, automation of solutions using a sequence of organised steps through algorithmic reasoning, finding, evaluating and enacting feasible solutions to create the most effective method, taking a broad perspective and applying solutions to a wide range of problems (Barr & Stephenson, 2011; International Society for Technology in Education and Computer Science Educators Association; ISTE & CSTA, 2011). Later, Brennan and Resnick (2012) outlined a computational thinking paradigm covering three key dimensions: computational principles (sequences, loops, events, parallelism, conditionalities, operators and data); computational activities (experimentation and iteration, testing and debugging, reuse and remixing, abstracting and modularising); and computational perspectives (expressing, linking and questioning). The United Kingdom’s Computing at School organisation argued that the central and peripheral facets of computational thinking include six separate concepts: logic, algorithms, decomposition, patterns, abstraction, and evaluation; and five approaches to classroom work: tinkering, designing, debugging, persevering, and collaborating (Barefoot, 2014). Likewise, Sullivan and Heffernan (2016) argued that computational thinking includes activities that support the practice of effective computer programming, including problem-solving (e.g., algorithmic development, heuristic development, organisation, planning, search), abstraction (creating new representation of a problem), and design (creating models and simulations). Yadav et al. (2018) described CT as the ways of thinking, or mental habits that computer scientists use and that such mental habits might be useful for science and/or maths inquiry. When reviewing the definitions in the literature, it can be implied that computational thinking is a viable problem-solving approach which requires five basic concepts, including:

• Decomposition – This involves breaking down a complex task into smaller, and more manageable components;

• Recognition of pattern – This involves identifying and defining trends or patterns within a problem;

• Abstraction, which involves identification of particular similarities and differences between comparable problems to work towards a solution;

• Algorithm design, which involves the development of step by step guidelines for solving a problem and can be used again to answer similar problems; and

• Automation, which involves the use of technological tools to mechanise problem solutions (ISTE & CSTA, 2011; Kalelioglu et al., 2016; Yadav et al., 2018).

To have a more nuanced understanding of computational thinking as it applies to the teaching and learning of science, Weintrop et al. (2016) developed a comprehensive and contextualised definition of CT as a set of highly interrelated and dependent practices classified into four groups, namely data practices, modelling and simulation practices, computational problem-solving practices, and systems thinking practices. These groups are defined as follows:

• Data practices involve the collection, creation, manipulation, analysis, and visualisation of data;

• Modelling and simulation practices include the use of computational models to understand a concept, use of computational models to find and test solutions, assessing, designing, and constructing computational models;

• Computational problem-solving practices entail preparing problems for computational solutions, programming, choosing effective computational tools, assessing different approaches/solutions to a problem, developing modular computational solutions, creating computational abstractions, troubleshooting, and debugging; and

• Systems thinking refers to investigating a complex system as a whole, understanding the relationships within a system, thinking in levels, communicating information about a system, defining systems, and managing complexity (Weintrop et al., 2016).

While the use of computational methods, hypotheses, knowledge, and algorithms has been at the forefront of science and engineering since the mid-20^th century, the increase in the importance of computation has revolutionised how research activities are carried out. Consequently, the science education community has recently turned its attention to the concept of computation and mathematics as valuable methods for characterising physical variables and their relations. According to the National Research Council (NRC) of the United States, computation, and mathematics “are used for a range of tasks such as constructing simulations; statistically analysing data; and recognising, expressing, and applying quantitative relationships. Mathematical and computational approaches enable the prediction of the behaviour of physical systems along with the testing of such predictions. Further to this, the Next Generation Science Standards (NGSS) emphasise doing authentic investigations in the classroom through a list of eight essential practices, one of which includes the use of quality mathematical and computational thinking practices in teaching and learning of science (Next Generation Science Standard Lead States. (NGSS), 2013). For instance, the NGSS advocate the use of mathematical and computational thinking for teaching middle school science to draw on students prior experiences in K-5, and progress to include the use of digital tools and/or mathematical representations to analyse large data sets, identify patterns, and create algorithms to design or solve scientific and engineering problems (Next Generation Science Standard Lead States. (NGSS), 2013).

It is advanced that many of the empirical and abstract approaches involved in the NGSS outlined scientific practices such as asking questions, defining problems, creating and using models (technology), planning and conducting research, gathering data, analysis, and interpretation of data, making predictions and communicating results, using mathematical and computational thinking, as well as providing clarifications and answers to questions are principal features of computational thinking (Basu et al., 2013). In light of this, studies show that the thoughtful use of computational tools and related set of skills can deepen the learning of science content from a pedagogical perspective, as science provides a relevant context and set of questions within which computational thinking can be applied (Wilensky et al., 2014; Yadav et al., 2018). This suggests that ‘engaging students in scientific investigation requires not only skill but also knowledge unique to each practice’ (Next Generation Science Standard Lead States. (NGSS), 2013, p. 15). While programming is often used to teach CT, teaching CT does not require students to automatically learn new codes or create programs (Lye & Koh, 2014). In light of this, researchers have proposed agent-based modelling and visual programming language learning environment called CTSiM (Computational Thinking in Simulation and Modelling) for K-12 science students to enable them to achieve the synergistic learning of scientific knowledge and CT practices using learning by modelling approach (Basu et al., 2013)

Theoretical background

Existing reviews on computational thinking

Only a few review studies have been done on the use of CT in teaching and learning science in K-12 classrooms. However, in recent years, numerous systematic reviews and meta-analysis have been actively carried out on studies related to the computational thinking curriculum in K-12 education, without a specific focus on school science. This includes a major analysis by Lye and Koh (2014). They focused on teaching and learning computational thinking through programming and found that programming languages, such as Scratch and Logo, were popularly used as an intervention to promote the development of CT concepts for kindergarten and middle school students while studying language and mathematics material. The review also notes that most studies focused on the use of a constructionism-based problem-solving learning environment with information processing, authentic problems, reflection, and scaffolding activities as other intervention approaches that could be used to foster the development of students’ computational thinking skills.

They also report that teaching computational thinking through programming exposes students to the use of coding exercises and computer science concepts in solving problems and learning subject content. Also, review studies by Cutumisu et al. (2019) and Tang et al. (2020) which looked at computational thinking assessment coded sub-sets of empirical research that examined the methods used to measure the different dimensions of computational thinking at all levels of education. Tang et al. (2020) found that most CT activities were related to programming and/or CS subjects because of the current reliance on programming concepts as major constructs of CT that require the use of computers in solving problems. It is used more commonly in elementary and middle schools than in higher grades, and more in formal educational settings than informal settings. In another pertinent study, Kalelioglu et al. (2016) provided a framework about the notion, scope, and elements of CT. Studies were coded according to theoretical foundations and final analysis showed the use of game-based learning, constructionism, the National Research Council’s (NRC) framework, Positive Technological Development (PTD), STEM and Vygotsky’s Zone of Proximal Development as theories that could be used as roadmaps for teaching, studying, and applying CT principles in many disciplines. The most recent review report on the incorporation of computational thinking into K-12 STEM education by Lee et al. (2020) offers an overview of the literature on what CT looks like from the instructional perspective of STEM education, CT incorporation, complexities of incorporating CT, and new ways of evaluating CT practices to closely connect CT with science practices. However, none of these reviews provides a detailed analysis of how computational thinking has been used in the teaching and learning of science, which is the focus of this study. Therefore, in adopting an exploratory approach, this study reviews empirical evidence of the use of computational thinking in teaching and learning science in elementary and secondary education.

Research questions

Despite the educational benefits of embedding computational thinking in the classroom as a pedagogical strategy, many science educators do not seem to understand how to adopt CT practices to support students’ sensemaking discussion and active engagement in scientific practices (Waterman et al., 2020; Wilkerson & Fenwick, 2017). This is evidenced, for example, by the fact that many science educators still teach science as a body of content to be learned and engage learning in rote procedures to confirm theory only or mostly (Y. Li & Schoenfeld, 2019; Wilkerson & Fenwick, 2017). This tends to limit the conceptual process in which a learner is supposed to actively engage with the use of mathematical and computational thinking in explaining the natural or designed world, wonder about it, and then develop, test, and refine ideas (Next Generation Science Standard Lead States. (NGSS), 2013). With the recent developments in recognising the importance of CT in science education, there is a renewed interest in how CT can support conceptual understanding in science instruction and scientific practices. The purpose of this study is, therefore, to systematically review and synthesise published empirical studies focusing on the incorporation of CT in science classrooms. It is envisaged that this review will inform on how educators can use CT to enhance the teaching and learning of science. To keep this review to a viable scale, we have decided to restrict our study to papers dealing with CT in science classrooms at the elementary and secondary school levels. The decision to focus on school science, beginning at the elementary school and extending to the secondary school, was motivated by the claim that interest in science education is often assumed to be sparked and sustained at an early age (Sanford & Sokol, 2017). Hence, the following research questions are addressed in this study:

1. What research methodologies, subject areas, and education level dominate computational thinking studies in the science classroom context?

2. How has computational thinking been incorporated into the teaching and learning of school science?

3. What are the pedagogical approaches used in incorporating CT in science classrooms?

4. What tools have been used to integrate CT in science classrooms?

5. How are CT skills assessed in science classrooms?

Methodology

Search strategy

For this systematic review, a search strategy was developed to identify relevant literature on computational thinking in science education. The search strategy was tailored using keywords that included ‘computational thinking’, ‘science classrooms’ AND ‘science education’ in searching the Ebscohost, Eric, and Scopus databases. The initial search without data parameters resulted in 563 articles. However, a total of 229 documents consisting of conference papers, conference reviews, dissertations, book chapters, editorials, magazines, reports, lecture notes, and errata were excluded from the initial search. Hence, 334 documents containing ‘peer-reviewed’ academic journal articles and review articles documented in English and published between 2010 to 31 May 2020, emerged from these criteria.

Selection Criteria

The selection criteria were based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (Moher et al., 2009). The PRISMA statement consists of 27 items evidence-based checklist and a four-phase flow diagram that can be used for critical appraisal of published systematic reviews. PRISMA is not intended to be a quality assessment tool, but its purpose is to ensure consistency and accountability when documenting the systematic analysis of literature. The search mainly focused on mapping existing literature and empirical studies on computational thinking in the field of STEM education. The search was then narrowed down to science teaching in K-12 classrooms. Papers reviewed in this study were selected based on the following inclusion criteria:

• Studies related to the use of CT in teaching and learning science subjects that includes physics, chemistry, and life sciences, in K-12 schools

• Studies that describe CT tools used in science education for K-12 context

• Studies describing the evaluation of CT methods and strategies in science classroom scenarios

• Studies published in referenced or peer-reviewed articles and documented in the English language only

• Published between January 2010 and May 2020

Exclusion criteria included studies:

• That was in book chapter format, conferences, and grey literature (opinion pieces, technical reports, blogs, presentations, etc.);

• Not focused on K-12 education and science subjects;

• Not published in reputable (i.e. peer-reviewed) sources

• Not published in English; and

• Studies that mention the term ‘computational thinking’ but are actually about use of technology in science teaching or other topics (and the term appears only in the references section).

The selected computational thinking in science classroom articles reviewed in this study are provided in Table 1.

Table 1. List of the 23 selected articles reviewed in this study.

No.	Authors and year	Title	Context	Subject learning area
1.	Tucker-Raymond et al. (2019)-	“I Broke Your Game!”: Critique among Middle Schoolers Designing Computer Games about Climate Change	Middle school	Climate science
2.	Peel and Friedrichsen (2018)	Algorithms, Abstractions, and Iterations: Teaching Computational Thinking Using Protein Synthesis Translation.	High school	Biology
3.	Kaya et al. (2019)	Animatronic Lions, and Tigers, and Bears, Oh My! How computational thinking and 3D printing can help students create an animatronic zoo.	Elementary school	Elementary science
4.	Hutchins et al. (2020).	C2STEM: A System for Synergistic Learning of Physics and Computational Thinking	High school	Physics
5.	Aksit and Wiebe (2020)	Exploring Force and Motion Concepts in Middle Grades Using Computational Modelling: a Classroom Intervention Study	Middle school	Physics
6.	Luo et al. (2020).	Exploring the evolution of two girls’ conceptions and practices in computational thinking in science	Elementary school	Life science
7.	Basu et al. (2016).	Identifying middle school students’ challenges in computational thinking-based science learning	Middle school	Physics and Biology
8.	Waterman et al. (2020)	Integrating Computational Thinking into Elementary Science Curriculum: An Examination of Activities That Support Students’ Computational Thinking in the Service of Disciplinary Learning	Elementary school	Biology
9.	Sengupta et al. (2013).	Integrating computational thinking with K-12 science education using agent-based computation: A theoretical framework	Middle school	Physics and Biology
10.	Irgens et al. (2020).	Modelling and Measuring High School Students’ Computational Thinking Practices in Science	High school	Biology
11.	Dickes et al. (2020)	Sociomathematical Norms for Integrating Coding and Modelling with Elementary Science: A Dialogical Approach	Elementary school	Elementary science (kinematics and Ecology)
12.	Garneli and Chorianopoulos (2019)	The effects of video game making within science content on student computational thinking skills and performance	Middle school	Physics
13.	Basawapatna (2016)	Alexander Meets Michotte: A Simulation Tool Based on Pattern Programming and Phenomenology.	Middle school	Life science
14.	Yin et al. (2020)	Improving and Assessing Computational Thinking in Maker Activities: the Integration with Physics and Engineering Learning.	High school	Physics and Engineering
15.	Lui et al. (2020)	Communicating computational concepts and practices within high school students’ portfolios of making electronic textiles.	High school	Life science
16.	Boulden et al. (2018).	Computational Thinking Integration into Middle Grades Science Classrooms: Strategies for Meeting the Challenges.	Middle school	Life sciences
17.	Yadav et al. (2018)	Computational Thinking for All: Pedagogical Approaches to Embedding 21st Century Problem Solving in K-12 Classrooms.	K- 12 classrooms	STEAM
18.	Clark and Sengupta (2020).	Reconceptualising games for integrating computational thinking and science as practice: collaborative agent-based disciplinarily integrated games	Middle school	STEM disciplines
19.	Sullivan and Heffernan (2016)	Robotic construction kits as computational manipulatives for learning in the STEM disciplines	P − 12	Physics and Biology
20.	Lee et al. (2020).	Computational Thinking from a Disciplinary Perspective: Integrating Computational Thinking in K-12 Science, Technology, Engineering, and Mathematics Education	K − 12 Classroom	STEM
21.	Weintrop et al. (2016)	Defining Computational Thinking for Mathematics and Science Classrooms	High school	Science and Mathematics
22.	Martin and Jacobsen (2018)	Coding and Computational Thinking in Maths and Science.	Middle and High school	Science and Maths
23.	Lee et al. (2014)	Integrating computational thinking across the K-8 curriculum	Middle school	Life science

Quality assessment

To maintain the quality of the review, we iteratively reduced the initial sample of studies by removing duplicate records found from the three databases. The filtration and removal of duplicate records yielded a reduced sample of 71 articles from the study. Furthermore, the titles of articles were then screened, resulting in the immediate rejection of 119 papers that were out of scope and not relevant to the study. A simultaneous title and abstract review were later conducted to check deeply for the analysis and purification of articles to ensure the quality and relevance of academic literature included in the review process. A careful evaluation of each research paper was carried out at this later stage, after which 121 papers were excluded. After applying these steps for eligibility purposes, the final systematic review included 23 articles in the data extraction phase, which were explicitly linked to the learning of computational thinking in science classrooms. The process is summarised using the PRISMA framework in Figure 1.

Figure 1. Study selection chart (Adapted from Moher et al.’s (2009) PRISMA Framework.

Data extraction/analysis

The Evidence for Policy and Practice Information and Co-ordinating Centre (EPPI-Centre) in London suggested several guidelines containing detailed sets of questions for coding at keywording and data-extraction for mapping and synthesis of primary research; as well as guidelines for the reporting of primary research (which then assists with coding for systematic reviews). In this review, the selected studies were inserted into an excel spreadsheet that was constructed based on the parameters defined for inclusion and exclusion. The required information was later extracted following the research questions raised, using the Guidelines for the Extraction of Information and Quality Evaluation of Primary Studies in Educational Research (Evidence for Policy and Practice Information and Co-ordinating Centre (EPPI Center), 2003). The guideline also includes how the research method used to connect basic CT elements with relevant evidence provided by each study is judged. Although the selected papers included some quantitative, mixed-method, and review research methodology, the majority of the studies examined offered qualitative explanations for their results.

The different types of research methodology employed in the reviewed papers provide an opportunity to identify teaching strategies that are peculiar to the implementation of computational thinking in science classrooms. These strategies assess the efficacy of some of the approaches used in integrating CT in science classrooms based on quantitative studies and provide rich explanations of effective approaches that are insightful for both theory and practice. This form of analysis later formed the basis for drawing conclusions and recommendations from all the studies described, based on evidence. The texts from the selected studies were classified and coded according to content review procedures (Fraenkel et al., 2015). The coding framework was developed based on instruments that have been used in five separate studies. This method was used in conjunction with the four research questions to extract detailed information from each selected article systematically. All the data was collected and analysed with the use of an Excel spreadsheet. A thorough description of the trends emerging from the reviewed studies to support the four research questions was given, along with examples for each.

Results

From the systematic analysis of educational research conducted between 2010 and 2020 on the incorporation of computational thinking as explicitly linked to the teaching and learning of science from elementary to high school in this paper, only 19 empirical studies and four review papers were found, and these provided the basis for answering the study questions.

Research Question 1: What methodologies, subject areas, and education level dominate computational thinking studies in the science classroom context?

Research methodologies

Based on the research classification by Creswell and Creswell (2017) and Snyder (2019), an analysis of the type of research method used in the reviewed studies is shown in Figure 2. The most common research method used was found to be a qualitative research design in 11 studies. This approach emphasised the use of unstructured and non-numerical information to gather an in-depth understanding of how CT is incorporated in science classrooms, and this forms 47.8% of the entire reviewed studies. The next most popular approach was mixed-method research design with six studies, which involves the combination of quantitative and qualitative research within the same project to facilitate a full understanding of a research problem (Bryman, 2008). This was followed by a systematic review which involved the use of explicit methods to identify existing studies that addressed formulated research questions. This approach had a total number of four studies. Lastly, three studies adopted a quantitative research design which focuses on explaining and interpreting the CT integration in the science classroom using numerical data.

Figure 2. Article distribution based on the methodology.

Data collection method

The data collection methods included the following:

1. Observation, which involved the systematic processing of information through unobtrusive visual means without questioning or communicating with participants (Nieuwenhuis, 2014). This included the use of field notes, video, and audio recordings;

2. Interviews, which is defined as a two-way conversation between the interviewer and individual or group of participants in discussing their perceptions of the world in which they live, and to convey how they perceive a subject of mutual interest (Cohen et al., 2018; Nieuwenhuis, 2014);

3. Survey, involving the use of open and/or closed-ended questions to assess information from respondents (Maree & Pietersen, 2014);

4. Portfolios, which is a tool used to collect student works like artefacts, drawings, projects, task representation, assignment, reflection statements, activity sheets, and other academic work products that provide evidence of student learning or programme improvement;

5. Tests, these involve achievement, aptitude, attitude, personality, performance, intelligence, and social adjustment assessment tools designed to measure specific knowledge, skills, behaviour, or cognitive activity that participants take away from their training or classroom instruction experience (Cohen et al., 2018); and

6. Literature review, involving the systematic collection and synthesis of documents and texts from previous research findings to demonstrate evidence at a meta-level and to identify areas where more research is needed. This is a critical component for the creation of theoretical frameworks and conceptual models (Snyder, 2019).

The results indicated that seven of the reviewed studies used tests as their instrument, five used information from literature reviews, 11 used interviews, nine employed observation, ten used portfolios while survey was used in three studies. These figures are illustrated in Table 2.

Table 2. Distribution of Data Collection Methods.

Method	Frequency
Tests	7
Literature review	5
Interviews	11
Observation	9
Student Portfolios	10
Survey	3

Subject area

The 23 papers included in our analysis spanned across a range of science subjects with nine in Biology/Life Sciences, one in Climate Science, four in Elementary Science, seven in Physics, and four in STEM subjects which generally focused on science without a specific subject/content area as shown in Figure 3.

Figure 3. Article distribution based on subject areas.

Education level

In terms of school type, four studies took place in elementary school, ten in middle school, six in high school, and three in what was termed K − 12 classroom spanning from Kindergarten to Grade 12 as indicated in Figure 4.

Research Question 2: How has computational thinking been incorporated into the teaching and learning of school science?

Figure 4. Article distribution based on the school level.

The incorporation of CT into science education is most prevalent in recent years due to its increasing importance as an essential problem-solving approach, and correspondingly attempts have been made by researchers to elucidate what CT integration in science classrooms entails (Waterman et al., 2020; Weintrop et al., 2016). One approach that has been employed to achieve this objective is the proposed use of abstraction and decomposition as important CT concepts required for solving problem, thus stimulating processes that permeate the pedagogy of science (Lu & Fletcher, 2009; Yadav et al., 2014). In addressing this research question, the researchers started by exploring the different CT elements measured in the reviewed studies using identified CT constructs based on Brennan and Resnick (2012), ISTE & CSTA (2011), Weintrop et al. (2016) and Yadav et al., (2014). The common constructs that emerged were further thematically classified according to Brennan and Resnick (2012) classification of CT structures which tended to constitute advanced dimensions of CT as shown in Table 3 and included:

1. Computational concepts include fundamental ideas and/or mental representations that are used to write programs which help us understand the natural world.

2. Computational practices include the use of various design methods, construction procedures, and techniques to create interactive content that fosters the teaching and learning of science.

3. Computational perspectives include changes in the dispositions and understandings of participants about themselves, views on their participation in CT, their relationships with others, and the technological world around them.

Table 3. Summary of constructs used in incorporating CT in science teaching.

Computational Thinking Constructs	Frequency
Computational Concepts Branching Conditionals Data Events Loop Operators Parallelism Random values Repetition Sequences Variable	1 7 3 3 7 3 6 1 1 3 5
Computational Practices Abstractions Algorithm thinking Automation, Computational problem-solving practices Coordination & synchronisation Data practices Debugging Decomposition Defining approximations Experimentation Inventing measures Iteration/iterative design Refinement Revising Logical thinking Modelling and simulation practices Synchronisation Pattern generalization/recognition Place-measure-point command System thinking practices Troubleshooting User interactivity	10 7 1 2 1 6 8 6 1 1 1 2 4 1 1 1 1 1 5 1 2 1 1
Computational Perspectives Collaborating Connecting Dealing with practical situations Expressing	1 1 3 1

From the analysis of the findings, we found that activities that depict conditionals, loops, parallelism, and variables were the top four concepts used to implement CT in science classrooms; while CT practices such as abstraction, algorithm thinking, data practices, debugging, decomposition, iterative design, and pattern recognition were commonly used in the reviewed studies. Nonetheless, the findings revealed that the integration of CT dispositions and perspectives in science classrooms. The study went further to explain how the various CT constructs were used to enhance the teaching and learning of science in K-12 classrooms. For example, Aksit and Wiebe (2020) conducted a study that explored the integration of CT concepts and practices in the context of a simulation-based model building through a block-based programming environment in a middle school science classroom. The students engaged in hands-on activities that encouraged them to construct, modify and experiment with a model of force and motion by simulating a car’s motion on a frictionless road with a given force and mass of the car. The students were subsequently provided with the sprite (i.e., the car) and the stage (i.e., the road), and they were asked to develop an algorithm and construct the script for the model. The researchers claimed that students learned about common CT practices such as abstraction and algorithms as well as fundamental concepts in computer programming such as variables, conditionals, and loops using the Scratch environment to develop their practical and conceptual understanding of force and motion.

Sengupta et al. (2013) also describe a study on the integration of CT in K-12 science education in which they explained how engaging students in the basic practical details of computational representational practices such as abstractions, generalisation of patterns, repeated loops, troubleshooting or debugging and iterative refining of models become central to the development of students CT skills, as well as the way students learn, think and practice science. Overall, across the reviewed studies, the findings indicate that central to the development of students CT, students need to be exposed to the ideas of abstraction, modelling, and automation, while offering more learning outcomes for specific subject areas.

Research Question 3: What are the pedagogical strategies used in incorporating CT in science classrooms?

Integrating computational thinking in any discipline requires the use of a wide range of teaching strategies that can enhance students’ acquisition of critical skills needed for the 21^st century. The analysis of reviewed studies reported in this section reveals the use of strategies such as inquiry-based learning, design-based learning, problem-based learning, project-based learning, modelling-based learning and game-based learning as pedagogical approaches used in incorporating CT in science classrooms. Table 4 provides a summary of the pedagogical approaches reported in the examined studies.

Table 4. Categories of strategies used in integrating CT in science classrooms.

Approaches	Definition	Frequency
Design-based learning	It is a form of inquiry/project learning which involves the integration of design thinking/activities, process, and projects/games into the classroom to foster students’ creativity, problem-solving skills, and engagement in real-world, cross-curricular tasks (Scheltenaar et al., 2015).	5
Game-based learning	Game-based learning is defined as the implementation of game elements such as core mechanics, challenges, and goals into real-life settings to enhance learning (M.C. Li & Tsai, 2013).	4
Inquiry-based learning	Inquiry-based learning is a form of active learning that involves a process of investigation, problem formulation and solving, as well as the construction of knowledge for better understanding (Pedaste et al., 2015).	4
Modelling based learning	Modelling based learning refers to learning from models, learning with models, and learning by modelling (Spector, 2009). It involves the use of instructions to support the development and evolution of learners’ mental models as representations of physical phenomena and integrated knowledge that includes components of a dynamic system and their interactions (Louca & Zacharia, 2012).	7
Project-based learning	Project-based learning (PBL) is a model that organises learning around the creation of presentations, projects and/or products involving complex tasks, based on challenging questions or problems, that involve students in design, problem-solving, decision making, or investigative activities. PBL allows students to work relatively autonomously over extended periods and culminates in realistic products or presentations (Kokotsaki et al., 2016).	4
Problem-based learning	Problem-based learning is an approach in which students work in groups to learn about a subject, concept, and/or principles using open-ended or complex real-world problems (McGuinness, 2011).	2

The results from the reviewed studies showed that there was reasonable evidence of student-centred pedagogies used in integrating CT in science classroom contexts. For example, Boulden et al. (2018) described a research project designed to infuse CT concepts and practices directly into middle school science classrooms through a game-based learning environment. The game required students to immerse themselves in a 3-D world using a block-based programming language interface where they played the role of a computer scientist. The computer scientist was charged with applying higher cognitive processes like abstraction and algorithmic thinking to solve programming challenges that enabled the protagonist to save an underwater research station that has been taken over by a nemesis in a life sciences class. Another study conducted by Kaya et al. (2019) used the 5E instructional model as an inquiry-based learning approach encompassing the phases of engaging, exploring, explaining, elaborating, and evaluating to teach CT with robotics to elementary students. The results reveal that the 5E instructional model encouraged students’ personalised learning as they developed a better understanding and recognition of the connection between CT concepts, practices, and personal life at the end of the intervention. A further study used ViMAP (a visual programming language) as a model-based learning approach to support students’ creation of agent-based models through scaffolded activities that enhanced their conceptual understanding of motion as a process of continuous change (Dickes et al., 2020). The findings from the 23 reviewed studies also suggested that design-based learning, problem-based learning, and project-based learning were used as a guided inquiry process to incorporate CT in some classroom context (Basawapatna, 2016; Tucker-Raymond et al., 2019).

Moreover, it was found that some of the reviewed studies utilised multiple pedagogical strategies to maximise the incorporation of CT in their science classroom settings. For instance, Basu et al. (2017) used a design-based and modelling-based learning approach to scaffold students’ ability to construct, enact, and envision their computational models for a given science phenomenon in a CTSim learning environment. The findings suggested that the sequencing of activities allowed students to tackle modelling and reasoning with a single agent in the first lesson and then build more complex computational models with multiple agents in the second lesson. Furthermore, students’ pre- to post-test gains were found to be statistically significant due to the combined effect of multiple pedagogical strategies employed in the learning environment. Similarly, a study conducted by Hutchins et al. (2020) utilised a modelling-based and problem-based learning approach to synergise the learning of high school physics and computational thinking in a collaborative, computational STEM (C2STEM) learning environment. The authors found that students who worked in the C2STEM learning environment used a step-by-step model-building and problem-solving approach to develop a better understanding of concepts and practices of physics and CT than students who learned through a traditional curriculum.

Research Question 4: What kind of tools have been used in integrating CT into science classrooms?

Many tools have been developed and/or used for teaching CT in various classroom contexts. From the collection of studies examined in this literature, 12 different software tools were found to be used in the introduction of CT concepts and activities in science classrooms. These tools are classified according to four generic categories, namely block-based modelling, agent-based modelling, data processing software, and MakerSpace, as shown in Table 4.

• Block-based modelling tools are programming tools that use a graphical interface of interlocking blocks to integrate the learning of CT concepts and coding into subjects. These include Scratch, Blockly, Snap, etc.

• Agent-based modelling tools are techniques used to invoke actions and interactions of agents (entity, notion or software abstraction similar to the well-known programming specifications such as objects, methods, procedures, and functions) in a shared environment to determine the design, performance, behaviour, and properties of a system as a whole. These are used to incorporate CT concepts into subjects and include CTSim, ViMAP, Netlogo, Star Logo, StarLogo Nova, a simulation creation tool kit and C2STEM.

• Data processing tools are scientific instruments for gathering, creating, manipulating, arranging, evaluating, displaying, visualising, and interpreting data sets. Examples found in the reviewed studies include the application of iSENSE and spreadsheet software.

MakerSpace is a resource developed for informal environments within a school, library, or separate public/private facility dedicated to designing, studying, exploring, and sharing projects that do not require high technologies. Examples of MakerSpace projects/activities include coding, robotics/robot construction, electronics/Arduino, 3D printing, e-textiles, sewing, learning circuits and electricity with paper circuits, etc.

Table 5 provides a summary of tools used in integrating CT.

Table 5. Summary of tools used in integrating CT.

Teaching tools	Frequency
Block-based Blockly Scratch Snap	1 6 1
Agent-based C2STEM CTSim Simulation creation tool kit Starlogo ViMAP	1 2 1 1 1
Data Processing iSENSE Spreadsheet	1 2
Makerspace	2

In the review, six studies utilised the Scratch programming framework as a block-based modelling tool to incorporate CT in science lessons. These include Aksit and Wiebe (2020), Boulden et al. (2018), Garneli and Chorianopoulos (2019), Martin and Jacobsen (2018), Peel and Friedrichsen (2018), and Tucker-Raymond et al. (2019). These studies have primarily centred on using Scratch programming to construct games, create simulation-based models, and learning experiences that help students understand science concepts while they interact with other aspects of CT concepts and practices. For example, Tucker-Raymond et al. (2019) reported a study on integrating computer science in science classrooms where students used Scratch as a graphical drag and drop programming language to build games about climate systems and climate change. The findings indicate that using the Scratch programming environment supports students’ learning of several dimensions of computational thinking such as modelling, troubleshooting, designing user interfaces, and offering sophistication in game design communities with solutions. This is further reinforced by results from another intervention study in a middle school science classroom, which indicated that involving students in block-based programming such as Scratch, offered affordances in building models for simulation-based activities and generative dimensions. This was done to develop basic CT practices such as abstraction and algorithms, as well as fundamental concepts in computing such as conditionals, variables, and loops while improving students conceptual understanding on force and motion (Aksit & Wiebe, 2020).

From the analysis, some studies use more than one type of tool to teach CT concepts. For instance, Kaya et al. (2019) focused on an eight-week post-school intervention provided to elementary students in which they were taught CT concepts and activities such as algorithms, abstraction, pattern recognition, debugging, decomposition and iterative design. The intervention also included the participation and collaboration of students in hands-on artistic engineering activities using robotics with SNAP software (a block-based programming language similar to Scratch) and the development of an animatronics zoo project with MakerSpace tools. They found that participation in the project offered opportunities for the students to learn basic CT skills while teaching them animal habitats and engineering. This also stimulated the creative imagination of students as they learned about the various animal environments and diets. The use of MakerSpace, spreadsheets, and agent-based models such as CTSim (computational thought in simulation and modelling) were also identified as another category of tools recognised from the examined studies. Sengupta et al. (2013) discussed a project in which middle school students were introduced to programming constructs during a science lesson using computational thinking in simulation and modelling (an agent-based computation model). They claimed that the CTSim learning environment is an effective tool in the physics and ecology domain for learning and modelling aggregate level and emerging phenomena.

Research Question 5: How are CT skills assessed in science classrooms?

Research emphasises the crucial role of assessment in effectively incorporating/introducing CT into K-12 classrooms (Grover & Pea, 2013). Given the importance of CT as a required competency that supports the effective teaching and learning of 21^st century skills and science education (García-Peñalvo et al., 2016; Wing, 2006; Yin et al., 2020), the integration of computational thinking in classroom activities, such as assessment, is believed to provide opportunities for students to grow and learn. In light of this, the study also sought to understand ways that CT has been assessed in the science classroom. However, research into the assessment of CT in science classrooms as noted in the examined studies revealed that there was a partial overlap of assessments used for teaching and research which leads to the conflation of assessments used for teaching CT and assessment used for research (data collection purposes). The assessment techniques used for teaching CT in the examined studies include:

• Achievement tests designed to determine if students have met specific learning goals in terms of acquired knowledge and skills. Achievement tests may be taken before, during, or after a learning experience. This includes the use of diagnostic assessment, standardised tests, multiple-choice questions, short answer questions, formative assessment technique like pop quizzes, classroom critiques and anecdotal notes which are used to guide instructions.

• Surveys/questionnaires

• Interviews which includes group discussion, informal conversations,

• Observation as a formative assessment technique that provides the opportunity to assess and document evidence of students learning. It provides insights and understanding of activity or situation being evaluated

• Portfolio assessment as a collection of projects and related tasks used to document what a student has learned (knowledge and skills). This includes the collection of drawings, projects, artefacts, reflective statements, written and other related tasks. These portfolios were mostly used for the summative assessments of students at the end of the intervention programme. Table 6 sets out the assessment techniques.

Table 6. Distribution of assessment techniques.

Assessment	Frequency
Achievement test	11
Surveys/questionnaires	4
Interviews	12
Observations	7
Portfolio assessment	13

The data illustrated in Table 6 show that portfolios (n = 13), interviews (n = 12), and achievement test (n = 11) are the most commonly used techniques for assessing the incorporation and development of CT skills in the examined studies (e.g., Basu et al., 2016; Luo et al., 2020; Tucker-Raymond et al., 2019). For instance, Lui et al. (2020) used portfolios to assess how high school students communicated their computational thinking experiences. They found that the portfolios contained media, videos, texts, and images relating to students’ reports on computational concepts like events, sequences, loops, conditionals, data, operators, polarity, connection types, and current flow. The portfolios also contained practices such as debugging and troubleshooting, iterating and devising, as well as testing and revision. Other techniques used for assessing the development of CT skills as found in the studies include observations (n = 7) and surveys (n = 4. For example, Peel and Friedrichsen (2018) and Yin et al. (2020) developed a self-report survey to measure high school students’ CT dispositions and experiences with using CT skills. It was also noted that some of the studies utilised multiple assessment approaches to evaluate students’ understanding of computational thinking concepts and practices in science classrooms. For instance, Aksit and Wiebe (2020) used students’ responses to reflective statement questions, interviews, and multiple-choice assessments in the form of pre/post achievement tests to assess the CT skills of learners and conceptual understanding of force and motion concepts. From the pre/post achievement test, they found that engaging students in computational modelling intervention substantially improved their conceptual understanding of force and motion concepts. Although no substantial learning gains with a major impact on students’ CT abilities were identified in the study due to participation in the computational modelling intervention, the researchers argue that students can achieve higher levels of CT ability if computationally rich activities are specifically incorporated into science instructions. In addition to the pre/post-test results used in the study, students’ responses to a written reflective statement as well as their responses to the interview questions on what they learned during the computational modelling activities, offered useful insight into students’ improved conceptual understanding of force and motion concepts. The interview responses also showed that some students shared the intention to spend more time using the Scratch programming framework to learn computational concepts.

Discussion

The first objective of this study was to identify the different constructs being used to integrate computational thinking into science classrooms. The various constructs identified were explored based on existing operational frameworks (Brennan & Resnick, 2012; ISTE & CSTA, 2011; Weintrop et al., 2016; Yadav et al., 2014) and later collated into three components: concepts, practices, and perspectives, using Brennan and Resnick (2012) framework. Concepts such as conditionals, loop, and parallelism, as well as practices such as abstraction, algorithmic thinking, debugging, data practices, decomposition, and pattern recognition were commonly identified as major constructs used in incorporating CT into the teaching and learning of science as identified by research question 1. These constructs are foundational to the development and application of students’ CT skills in addressing scientific problems. The large number of studies that utilised abstraction as a CT component in the reviewed studies show that abstraction is a key practice that is closely connected to the effective teaching of science for conceptual understanding.

Another main research question that this study answered was to investigate the various pedagogical approaches that are used to integrate CT into school science. Emphasis on these teaching techniques was necessary to understand how the learning of preferred CT concepts and activities in science classrooms was improved by educators. In this review, modelling-based learning formed a large percentage of the pedagogical strategies used in the studies (30.4%), followed by design-based learning (21.7%), problem-based learning (8.7%) as well as game-based learning, inquiry-based learning and project-based learning (17.4%) respectively. The use of modelling-based learning as a pedagogical strategy is possibly due to its acknowledgement and inclusion in the United States NGSS as an important practice for enhancing students’ understanding of scientific processes (Next Generation Science Standard Lead States. (NGSS), 2013). Nevertheless, these results show how different pedagogical strategies have been used to incorporate CT into science classrooms, supporting the findings of Hsu et al. (2018) who identified project-based learning, problem-based learning, game-based learning as the most widely used teaching methods in their review literature on how to learn and teach computational thinking. The use of project-based learning in computational thinking is possibly due to the popularisation of computer technology, which gave schools more opportunities to design and establish CT courses (Hsu et al., 2018). The researchers claim that various learning strategies were adopted in CT activities in the period 2014–2017, indicating that CT ability gained more recognition as educators and researchers set out to implement various learning strategies to help students enhance their learning output through CT activities. This is well supported by studies highlighting the cultivation of the problem-solving capacity of students through problem-based learning (PBL), in which students practise computational thinking through completing team projects (Liu et al., 2017). Moreover, the use of model-based learning strategies also emphasises the development of student problem-solving skills through problem-based and project-based learning (Liu et al., 2017).

With regards to the types of tools used to teach computational thinking, eight out of the 23 studies employed block-based programming tools such Scratch, SNAP and Blockly apps. Seven studies used agent-based modelling tools including Netlogo, ViMAP, CTSim, C2STEM, simulation creation tool kit, and StarLogo. Two of the studies employed data processing tools such as spreadsheet and iSENSE; while two other studies used MakerSpace components like e-textiles, LilYPad, Arduino microcontrollers, LEDs, sensors and switches for integrating CT in science classrooms. Analysis of the studies reviewed reveal that the kind of tools accessible for teaching CT offers a foundation on which the pedagogical strategies for implementing CT can be applied in science classrooms. For example, Scratch is a programming language (open source software) that enables users to think creatively, reason actively, and work collaboratively by building their own interactive stories, games, and animations (Lockwood & Mooney, 2017). This implies that the use of the Scratch programming environment appears to promote self-paced learning through students’ engagement in project-based, design-based, modelling-based, and game-based learning activities.

The strategies and tools used in assessing of computational thinking in this study shows that 13 of the reviewed studies employed portfolio assessment, 12 used interviews, 11 used achievement tests, seven used observations, and four used surveys/questionnaires as assessment instruments for measuring students’ CT skills and progress within the context of the science classroom. These findings are consistent with Tang et al. (2020), who in their systematic review of empirical studies on assessing computational thinking reported portfolio assessment as the most used assessment tool in the reviewed studies. The need to use portfolio assessment in evaluating CT is emphasised as it captures the holistic view of students’ understanding of computational concepts, engagement, and learning with computational practices, as well as computational communication about their projects (Lui et al., 2019). The use of a portfolio assessment for evaluating CT also serves as a formative assessment tool that can potentially provide feedback to students (Tang et al., 2020). However, the limited use of surveys/questionnaires as an assessment tool in the studies reflects a limitation if this assessment tool is used alone. This is because the use of surveys/questionnaires alone cannot provide in-depth information and evidence of students’ thinking processes on computational thinking (Tang et al., 2020). In light of this, it was found that some of the studies examined provided evidence of the use of multiple assessment methods such as portfolios and interviews, surveys and achievement tests, as well as portfolios and achievement tests to examine students’ higher-level thought processes about how they improve their CT skills during science instruction (Boulden et al., 2018; Hutchins et al., 2020; Yin et al., 2020).

Implications

Our inclusion criteria listed 23 studies that presented conceptualisations about how CT was implemented, taught, and assessed in science classrooms, indicating that concerns about the lack of empirical evidence in this field are recognised. Nonetheless, the findings of this systematic analysis have many consequences for the teaching and learning of science. Overall, it was found that different pedagogical approaches and tools were used for embedding CT constructs such as modelling and problem-solving practices in science classrooms. Secondly, the results show the importance of modelling-based approaches in actively engaging students in CT activities that are inquiry-oriented. The current move towards incorporating CT concepts into science classrooms goes beyond computer science learning and towards more comprehensive advantages of improving students’ comprehension of scientific practices and procedures, system designs, and human behaviours (Barr & Stephenson, 2011; Next Generation Science Standard Lead States. (NGSS), 2013; Wing, 2006). Through taking the approach of describing various dimensions of CT, this systematic analysis has helped to define common features of CT activities that tend to be more closely linked to increased learning of science among students.

Limitations of the study

One of the drawbacks of this study is that the search was primarily limited to three educational repositories. The collection of papers included was limited to peer-reviewed journal articles that focused on teaching CT in science classrooms and were published between 2010 and May 2020. We recognise that the collection of included articles and reviews might not include papers of significance published in other types of publications, such as conference papers, editorials, books, or grey literature. However, most of the reported studies were performed as brief intervention programs at the classroom level, focusing on activities, games, and/or a novel approach designed to teach the concept of CT inside science instruction. Also, papers addressing CT in relation to university education and professional development for educators were excluded, since we aimed to find important contributions of CT to the field of science teaching in K-12 classrooms.

Conclusion and recommendations

The purpose of this systematic review was to analyse empirical evidence on the use of computational thinking in science education. The most commonly measured CT practices were abstractions, algorithm thinking, and debugging. The findings showed that CT can be taught using single or multiple learning approaches in science classrooms. The integration of CT into science classrooms is actively promoted through model-based, project-based, and inquiry-based initiatives designed to involve students and encourage learning-on-demand as students take responsibility for their learning. The literature consistently demonstrates that creative instructional approaches, such as the use of model-based programs, have advantages in promoting the conceptual understanding of science content and the development of related CT abilities among students. Block-based programming applications like Scratch were more frequently used as teaching tools for integrating CT in science classrooms than agent-based programming tools. Finally, the majority of the studies employed a portfolio assessment to evaluate CT. Despite the nuanced and diverse nature of CT as a collection of mental skills used to solve problems and system design, the findings indicate that a significant effort has been made to develop a key set of CT practices and principles across science classrooms. Nevertheless, there is still a need for significant investment in more research on the creation of instructional resources, assessment constructs, and how to combine learning strategies that can be used to support the synergistic learning of CT and science instruction as common practices. It is suggested that the successful integration of CT in science classrooms needs to address three interlocking frameworks for change: the teacher, the school, and policymakers.

Acknowledgments

The authors would like to appreciate Ms Anna M de Wet, language editor, for technical and editorial preparation of the article.

Disclosure statement

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