Abundant research shows that logical reasoning is one key component underlying skills in science (e.g., van der Graaf et al., 2015; Vilia et al., 2017; Berkowitz and Stern, 2018). The reason for this dependence upon logical reasoning is rather straightforward as acquiring skills in science involves learning about abstract scientific facts, theories, and applying complex scientific methods (Klahr et al., 2011; Kuhn, 2011). Thus, in order to acquire the complex and abstract content of science, the child must be able to perform logical and abstract thinking (Roth et al., 2015).

Abstract scientific phenomena and concepts (electricity, magnetism, molecular structure, cell structure) are often described and explained by the use of graphs, diagrams, or physical models (Hegarty, 2014; Newcombe, 2016). The interpretation and comprehension of these forms of visual-spatial representations should theoretically place demands on the individual’s spatial processing abilities (Hegarty, 2014; Stieff and Uttal, 2015; Newcombe, 2016; Verdine et al., 2017). Consistent with the assumption, numerous studies on adults show that measures of spatial ability such as mental rotation and spatial visualization are predictive of concurrent and future accomplishment in science (Hegarty and Sims, 1994 spatial visualization; Paper Folding Test; speeded rotation, spatial orientation; Kell et al., 2013 spatial visualization; Kozhevnikov et al., 2007 spatial visualization; Shea et al., 2001; Wai et al., 2009 spatial visualization; Webb et al., 2007 mental rotation; Yoon and Mann, 2017 mental rotation). Furthermore, a few intervention studies provide evidence that training spatial ability can improve science learning in university students (Sorby, 2009; Miller and Halpern, 2013). However, the mechanisms for how spatial ability supports skills and learning in science are still not well understood. Nevertheless, it has been suggested that spatial processing has several different functions during science problem-solving such as creating spatial-schematic images of abstract concepts and performing spatial transformations of these mental images (Kozhevnikov et al., 2002; Miller and Halpern, 2013; Hegarty, 2014). Important spatial transformations entail the ability to mentally rotate images, integrate or relate different components of visual-spatial information, and to decompose images into parts for subsequent individual analysis (Kozhevnikov et al., 2002; Miller and Halpern, 2013; Hegarty, 2014; Verdine et al., 2017).

Similar to mathematics, a number of researchers suggest that working memory constitutes a key mechanism for scientific reasoning (Hegarty and Sims, 1994; Isaak and Just, 1995; Kozhevnikov et al., 2007; Hegarty, 2014). Working memory refers to a multi-purpose mental workspace responsible for coordinating and executing the performance of simultaneous processes such as temporarily storing information, shifting from one strategy or operation to another, and inhibiting activation of irrelevant information (Engle et al., 1992; Shah and Miyake, 1996; Baddeley, 1997). Science is an intricate academic domain involving a complex interplay of reading/language processes, comprehension processes (i.e., logical reasoning), and visual-spatial processes (e.g., transformations; Hegarty, 2014; Bergey et al., 2015; Roth et al., 2015; Newcombe, 2016). As such, it should require a flexible and efficient mental workspace that can supervise, coordinate, and execute the different process involved in science/physics problem-solving (Kozhevnikov et al., 2007). Evidence in support of this assumption is mounting. For example, Gathercole and colleagues have, in a number of studies, observed relationships between science achievement and verbal working memory in children aged 14–15 years (Jarvis and Gathercole, 2003; Gathercole et al., 2004; see also Danili and Reid, 2004) and children aged 11–12 years (St. Clair-Thompson and Gathercole, 2006).

To date, few researchers have simultaneously investigated to what extent different cognitive abilities support children’s learning of science. However, a few exceptions to this state of affairs do exist. For example, Rhodes et al. (2014) found in a sample of 56 12- to 13-year-olds that proficiency in biology was supported by visual working memory (16.0%) and planning ability (9.6%), but not inhibition control, or attention shifting. In a later study performed on 63 12- to 13-year-olds, Rhodes et al. (2016) observed that proficiency in chemistry was supported by visual working memory (10.9%) and vocabulary (20.2%), but not inhibition control, attention shifting, or planning ability.

In a recent large-scale study on 5,838 16-year-olds, Donati et al. (2019) investigated the unique contribution of working memory, inhibition control, processing speed, vocabulary, non-verbal logical reasoning, and socio-economic status (SES) to attainment in science. Structural equation modeling showed that working memory (10.3%), non-verbal logical reasoning (0.01%), vocabulary (4.7%), and SES (1.1%) accounted for unique variance in science at age 16 while controlling for previous attainment in science at age 11.

Moreover, Mayer et al. (2014) explored the association between scientific reasoning (understanding the nature of science, understanding theories, designing experiments, and interpreting data) and spatial ability (mental rotation), inhibitory control, problem-solving skills, reading and logical reasoning in 155 10-year-olds (fourth grade). Multiple regression analysis showed that spatial ability accounted for 2.9% unique variance, while problem-solving skills and reading comprehension accounted for 6.7% variance each.

Recently, Hodgkiss et al. (2018) examined to what extent four different categories of spatial abilities (intrinsic-static; intrinsic-dynamic; extrinsic-static; and extrinsic-dynamic; Uttal et al., 2013; Newcombe and Shipley, 2015) and vocabulary contribute to children’s (7–11 years old; N = 123) achievements in specific science domains (i.e., physics, biology, chemistry). Multiple regression analyses revealed that all three science domains (biology, chemistry, physics) were supported by vocabulary and spatial abilities, but by somewhat different constellations of spatial abilities. Individual differences in biology scores were accounted for by mental folding (6%), an intrinsic-dynamic skill, and spatial scaling (2%), an extrinsic-static skill. Mental folding also accounted for 4% of the variation in physics scores. Spatial scaling (2%) and embedded figures (3%), an intrinsic-static skill, accounted for variance in chemistry.

In addition to the studies performed on 7- to 16-year-olds, three studies performed on young children (4–6 years old) are relevant for the present study.

For instance, van der Graaf et al. (2016) examined if variation in 100 kindergartners’ (4–5 years old) scientific reasoning (evidence evaluation; experimentation) was supported by verbal working memory, visuospatial working memory, inhibition control, spatial visualization, vocabulary, and grammar. They observed that evidence evaluation was supported by verbal working memory, inhibition control, vocabulary, and grammar but not visuospatial working memory or spatial visualization, while experimentation was only supported by inhibition control.

In a subsequent study on a sample of 100 5- to 6-year-olds, van der Graaf et al. (2018) found that verbal working memory (6.2%) and inhibition (4.8%) provided independent contribution to growth in evidence evaluation but language skills (vocabulary; grammar) did not. Vocabulary (6.8%) and grammar (6.8%), on the other hand, were the only cognitive abilities that accounted for variation in growth in experimentation.

Zhang et al. (2017) tested a sample of 584 6-year-old Chinese children on skills in life sciences (biology), earth and physical sciences (physics; chemistry), language (vocabulary), spatial processing (spatial perception, spatial visualization, and mental rotation), and verbal working memory. Multiple regression analyses revealed that verbal working memory (12.8%), spatial processing (mental rotation; 3.5%), and language (1.3%) contributed to achievement in life sciences, while only language (0.5%) contributed to achievement in earth and physical sciences (physics; chemistry).

In sum, the overall empirical picture regarding the cognitive mechanisms supporting children’s skills in the natural sciences is far from clear. The complicated empirical picture is due to a number of reasons. First, researchers have focused on different domains of science (e.g., biology, chemistry, physics, and generic science) or different aspects of science (e.g., factual knowledge, conceptual reasoning, evidence evaluation, and experimentation). Second, researchers have included somewhat different cognitive abilities in their test batteries, or used different measures to tap the same ability. Third, there is a large variability in age among the samples used in the different studies, ranging from 4–5 years to 16 years. Thus, it is difficult to draw any firm conclusions based on existing research. Hence, more research is required to pinpoint the cognitive mechanisms supporting children’s science learning. However, working memory and language abilities appear to be key components, but their contribution varies among studies from very small (0.5%) to quite large (12.8%). The findings from Mayer et al. (2014), Hodgkiss et al. (2018), and Zhang et al. (2017) corroborate the spatial-science link previously found in adults, by emphasizing the role of spatial abilities in children’s science learning.

The aim of the present study was to simultaneously investigate to what extent logical reasoning, spatial processing ability, and verbal working memory, tapped in third grade, are independent cognitive abilities underlying children’s future physics skills in sixth grade. Based on prior research and theoretical reasoning, it was hypothesized that all three cognitive abilities should independently account for variation in sixth graders’ understanding of physics.

Unlike prior research that has mainly focused on young adults, this study focused on 12- to 13-year-olds’ physics skills measured by a broad curriculum-based test. In addition to measures of logical reasoning, spatial ability, and verbal working memory, measures of basic arithmetic and reading comprehension were included in the study.

Mathematics and physics are two STEM (science, technology, engineering, and mathematics) disciplines, and as such, they may share certain underlying cognitive processes. Consistent with this, an abundance of research demonstrates a link between mathematics and science proficiency (e.g., Ma and Ma, 2005; Maerten-Rivera et al., 2010; Barnard-Brak et al., 2017). As a matter of fact, when performing science experiments, mathematical tools are used in order to collect (i.e., measure), organize, and analyze data (Batista and Matthews, 2002).

A number of studies also display a link between reading and science proficiency (O’Reilly and McNamara, 2007; Maerten-Rivera et al., 2010; Mayer et al., 2014; Barnard-Brak et al., 2017). This link is theoretically reasonable as almost all classroom teaching is provided verbally or via text, reading and language comprehension should play a role in children’s science learning. Moreover, the graphs and diagrams used to describe and explain abstract scientific concepts usually also include textual information that the child has to decode and comprehend (Cromley et al., 2013; Bergey et al., 2015). In view of existing research and theorizing, it was hypothesized that arithmetic ability and reading comprehension should account for unique variation in sixth graders’ understanding of physics.

A final sample of 81 Swedish children (37 girls) took part in the study. A letter of consent was distributed in the classroom that the children brought home to the parents. All children with written informed consent from the parents were included in the study. In grade 3, the mean age was 9.62 years (SD = 0.30, min = 9.03, max = 10.33). In grade 6, the mean age was 12.88 years (SD = 0.25, min = 12.34, max = 13.32). All 81 children had Swedish as their native language, no hearing loss, and normal or corrected-to-normal visual acuity. The study did not include children with neuropsychological diagnoses (e.g., ADHD).

This study was approved by the regional ethics committee in Linköping, Sweden (protocol number 33–09).

In grade 3, 292 children performed the reading test, Raven’s progressive matrices, the arithmetic test, and the mental rotation test in group sessions of 3–5 children. The working memory test was administered during an individual session. All children were tested in familiar rooms at their respective schools. The test instructions were presented orally, and all children received the tests in the same order. During the spring semester in grade 6, 81 children performed the physics test in a classroom session. The test was administered by the classroom teachers as it was part of the national curriculum assessment provided and monitored by the Swedish National Agency for Education. The large attrition is due to that tests of science (biology, chemistry, physics) and social sciences (history, religion, geography, social studies) are optional for the schools to perform, while tests of mathematics, English (first foreign language), and Swedish (native language) are mandatory. A total of 139 (48%) of the 292 children performed tests of either biology (38), chemistry (20), or physics (81).

The regression model, F(5, 80) = 4.68, p = 0.001, R² = 0.238, predicted 24% of the variance in physics. Verbal working memory, spatial processing (mental rotation), and Raven’s progressive matrices emerged as significant predictors. They accounted for 4.5, 4.6, and 5.7% unique variance, respectively, as indicated by their squared partial correlations depicted in the right part of Table 2. The reading and mathematics tasks did not account for any unique variance (p’s > 0.05). More detailed information concerning the results of the multiple regression analysis is presented in Table 2.

The present findings should be interpreted with some care, as the sample size was rather small compared to most prior studies reviewed in the introduction (median sample size = 112). Thus, future studies should replicate the present study with a larger sample. A larger sample should also make it possible to include larger number of cognitive variables. For example, it would be theoretically interesting to include both verbal and visual-spatial working memory tasks, and measures of spatial visualization, spatial perception as well as mental rotation. Then it would be possible to examine the unique relative involvement of different working memory resources and different spatial processing abilities in children’s and adults’ physics skills. Still, this study presents new and theoretically important findings by showing that physics is a multivariate discipline that draws upon numerous cognitive resources. Logical reasoning, verbal working memory, and spatial processing appear to have equally important roles in 12- to 13-year-old children’s physics skills.

The present study suggests that in order to facilitate children’s learning of physics, the regular science teaching should be supplemented with training of general cognitive abilities. Consistent with prior intervention studies focusing on the STEM complex, mental rotation is a spatial processing ability that should be targeted in such intervention (Uttal et al., 2013; Stieff and Uttal, 2015). Although, the effect of working memory training on science achievement has not yet been studied, the present finding suggests that this training might be a way to enhance children’s physics skills. The fact that working memory training has demonstrated positive effects on mathematics learning, another STEM domain, supports this assumption (Holmes et al., 2009; Loosli et al., 2012; Kuhn and Holling, 2014). Logical reasoning is also an important ability to exercise in order to enhance physics skills. This ability has traditionally been seen as none-malleable, but recent studies suggest that this might not be the case (Buschkuehl and Jaeggi, 2010; Au et al., 2015). Given the present findings and prior interventions studies, the most effective approach to enhance children’s science learning, above and beyond regular science teaching, might be to implement a combination training program that targets all three abilities: logical reasoning, working memory, and spatial processing.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.