Educ Psychol Rev https://doi.org/10.1007/s10648-018-9444-8 R E V I E W A RT I C L E

Drawing Boundary Conditions for Learning by Drawing Logan Fiorella 1

& Qian Zhang

1

# Springer Science+Business Media, LLC, part of Springer Nature 2018

Abstract Learning by drawing can be an effective strategy for supporting science text comprehension. However, drawing can also be cognitively demanding and time consuming, and students may not create quality drawings without sufficient guidance. Furthermore, evidence for drawing is often based on comparisons to weak control conditions, such as students who only read the text without provided illustrations. In this review, we synthesize past research to help draw boundary conditions for learning by drawing, focusing on the role of comparison conditions and drawing guidance. First, we analyze how drawing compares to each of four control conditions: reading only, text-focused strategies (e.g., summarizing), other model-focused strategies (e.g., imagining), or viewing instructor-provided illustrations. Next, we distinguish among four levels of drawing guidance: minimal guidance, drawing training, partially provided illustrations, and comparison to instructor-provided illustrations. Our findings indicate that when compared to only reading the text or using text-focused strategies, creating drawings is consistently more effective at fostering comprehension and transfer, regardless of the level of drawing guidance provided. However, when compared to other model-focused strategies or to viewing instructor-provided illustrations, effects of creating drawings are mixed and may depend on the level of drawing guidance provided, among other factors. We discuss the theoretical and practical considerations of our findings and suggest several directions for broadening research on drawing. Keywords Learning strategies . Generative learning . Learning by drawing . Science Visualizations are central to teaching, learning, and communicating in science (Ainsworth et al. 2011; Cook 2006). Illustrations, graphs, diagrams, animations, and models are often used to support verbal explanations of complex scientific systems and processes (Rau 2017). A vast body of past research has focused on how to design instructor-provided visuals to support learning (Ainsworth 2006; Mayer 2014; Rau 2017). In general, lessons containing provided words and corresponding images result in better learning than lessons containing words alone * Logan Fiorella [email protected]

1

Department of Educational Psychology, University of Georgia, Athens, GA, USA

Educ Psychol Rev

(Mayer 2014). However, some students may not actively integrate the correspondences between these two representations (Bodemer et al. 2004; Hegarty et al. 1991)—a crucial process for fostering understanding of multimedia lessons (Ainsworth 2006; Mayer 2014: Schnotz 2014). To address this issue, in recent years, there has been growing interest in exploring the benefits of learner-generated visuals, in which students actively construct their own visual representations during learning, thereby forcing students to integrate the text with a nonverbal representation. This work most often examines the effects of asking students to create drawings while reading scientific texts—what is referred to as learning by drawing (Fiorella and Mayer 2015, 2016a). For example, students might draw the structure of a neuron while reading a text describing the central nervous system (Van Meter 2001). Learning outcomes of students who created drawings can then be compared to those of students who only read the text or who were asked to use other learning strategies. Although there is growing empirical support for learning by drawing (Fiorella and Mayer 2015; Leutner and Schmeck 2014; Van Meter and Firetto 2013; Van Meter and Garner 2005), its implementation can vary considerably across studies, and not all studies report positive effects (Van Meter and Garner 2005). Most notably, past studies differ in the control conditions to which drawing is compared and the level of drawing guidance provided to students. These inconsistencies have made it difficult to specify the boundary conditions of learning by drawing. For example, how does creating drawings compare to other types of learning strategies? When is creating one’s own drawings more (or less) effective than viewing instructor-provided illustrations? To what extent does the effectiveness of drawing depend on the level of drawing guidance provided? In this review, we address these and other related issues by synthesizing the existing empirical research on learning by drawing. Specifically, we focus on two open questions regarding the boundary conditions of learning by drawing. One major focus is to analyze how drawing compares to stronger control conditions than only reading the text, including the use of other types of learning strategies (e.g., summarizing, imagining) and situations in which students read the text and view instructor-provided illustrations. A second focus is on distinguishing among different levels of drawing guidance used in past research to evaluate the extent to which learning by drawing depends on high levels of guidance and to provide a framework for future research. Finally, we discuss the theoretical and practical implications of our findings, and we recommend ways to broaden research on drawing to further explore underlying processes, individual differences, and the use of drawing in different learning contexts.

What Is Learning by Drawing? To understand the boundaries of learning by drawing, it is important to first distinguish drawing from related text comprehension strategies (Fiorella and Mayer 2015; McNamara 2012). According to Van Meter and Garner (2005), drawing is a learning strategy in which students create representative illustrations to achieve a learning goal. Critically, drawings depict the physical characteristics of a system rather than solely abstract spatial relationships, such as depicted by tables, outlines, or concept maps. McCrudden and Rapp (2017) make a similar distinction between semantic visual displays, which use symbols to represent comparisons, sequences, and hierarchies, and pictorial visual displays, which use images to represent physical structures and processes. Much like pictorial displays, learner-generated drawings

Educ Psychol Rev

primarily serve to depict a system’s physical characteristics, though both pictures and drawings can vary in their precise level of physical and conceptual fidelity (Butcher 2006) and often contain the integration of words (Bobek and Tversky 2016; Mayer 2014). Drawing can be distinguished among other learning strategies based on the cognitive and behavioral activity it fosters in learners (Fiorella and Mayer 2015). Cognitive activity refers to the types of mental representations students construct during learning. According to the construction-integration model of text comprehension (Kintsch 1988), readers form two main levels of mental representations during reading: a text base and a situation model. A text base is formed by selecting and organizing propositions from the text into a propositional network, whereas a situation model is formed by actively integrating the text with one’s existing knowledge. The text base allows a reader to recall main ideas from the text, but the situation model is required for making inferences and applying one’s knowledge to new situations. Accordingly, some learning strategies such as paraphrasing or summarizing primarily focus students’ cognitive processing toward the construction of a text base (text-focused strategies), whereas other learning strategies such as drawing, self-explaining, or imagining support the deeper cognitive processes necessary for constructing a situation model (model-focused strategies; see Leopold and Leutner 2012). Model-focused strategies are closely related to what are referred to as generative (Fiorella and Mayer 2015, 2016a) or constructive (Chi and Wylie 2014) learning strategies—activities that encourage students to organize and integrate the learning material with their existing knowledge. Learning strategies can also differ in their behavioral activity, or the overt behaviors learners produce during learning. This is important because differences in behavioral activity can potentially influence cognitive activity, which is ultimately responsible for learning. For example, some strategies involve no behavioral output (e.g., rereading or imagining), whereas others primarily involve generating words (e.g., summarizing or self-explaining), creating depictive or abstract spatial representations (e.g., drawing or mapping), or engaging in physical movements (e.g., gesturing or manipulating objects). Therefore, strategies can differ in whether they involve creating an external representation and whether that representation explicitly depicts visuospatial relationships. As summarized in Table 1, drawing is best classified as a model-focused strategy characterized by creating representative illustrations of what is described in the text. Based on this classification, drawing should better support meaningful learning outcomes such as comprehension and transfer than text-focused strategies (Leopold and Leutner 2012). However, less is known about the relative effectiveness of other model-focused strategies, such as imagining and self-explaining (e.g., Leutner et al. 2009; Lin et al. 2017; Scheiter et al. 2017). These model-focused strategies target similar cognitive activity but utilize very different behavioral activities to achieve this goal. Drawing involves generating an external representation (unlike imagining) that makes the structural relations of the learning material explicit (unlike selfexplaining). Thus, understanding how model-focused strategies differentially impact learning may depend on considering how differences in behavioral activity support or hinder productive cognitive activity. In this review, we explore these issues by examining contemporary theoretical accounts of drawing and by systematically analyzing past research comparing drawing to text-focused and other model-focused strategies. As discussed below, constructing drawings may offer unique cognitive and metacognitive benefits compared to other learning strategies and to instructor-provided (rather than learnergenerated) illustrations.

Educ Psychol Rev Table 1 Primary cognitive and behavioral activity of learning strategies used in drawing research Strategy

(Re)reading Summarizing Self-explaining Imagining Drawing

Primary cognitive activity

Primary behavioral activity

Text-focused

None (internal)

✓ ✓

Model-focused

✓ ✓ ✓ ✓

✓

External

Words

✓ ✓

✓ ✓

✓

Images

✓ ✓

How Does Drawing Foster Learning? The rationale for drawing is rooted in theories of learning from multiple representations, including dual-coding theory (Paivio 1990), generative learning theory (Wittrock 1990), and theories of multimedia learning (Ainsworth 2006; Mayer 2014; Schnotz 2014). Broadly, these theories assume that integrating verbal and nonverbal (visuospatial) information supports the construction of a coherent mental representation (e.g., a situation model), thereby fostering meaningful learning outcomes. Although theories of multimedia generally focus on supporting integration of words and visuals that are both provided to the learner (Mayer 2014), in learning by drawing, often only the verbal representation is provided (e.g., a scientific text), and learners must use their existing knowledge to translate the text into their own visuospatial representation. According to Van Meter and Garner’s (2005) original generative theory of drawing construction—derived from generative learning theory (Mayer et al. 1995; Wittrock 1990)— translating provided text into a drawing supports comprehension by fostering the cognitive processes of selecting, organizing, and integrating. First, students must select critical elements from the text for further processing in working memory. Then they mentally organize the verbal elements into a coherent verbal representation, or propositional network, based on descriptive conceptual relationships. This propositional network is used to support the construction of a visuospatial representation, or situation model, by integrating it with students’ existing knowledge from long-term memory. Integration occurs either by connecting propositions from the text with stored pictorial representations or by using stored verbal representations to convert the text into a new nonverbal representation. Finally, learners convert their constructed situation model into a perceptual image by creating a representative drawing on paper. Thus, one unique cognitive benefit of drawing is that it involves forced integration of verbal and nonverbal representations (Cox 1999; Van Meter and Garner 2005). In contrast, providing students with illustrations does not guarantee that students will spontaneously integrate the words and images (e.g., Bodemer et al. 2004; Hegarty et al. 1991). Similarly, text-focused strategies such as summarizing do not require that students translate the text into a coherent nonverbal mental representation that incorporates their existing knowledge (Leopold and Leutner 2012). Another important component of drawing is that the cognitive processes of selecting, organizing, and integrating occur recursively and are guided by the metacognitive processes of self-monitoring and self-regulation. According to Van Meter and Firetto’s (2013) updated cognitive model of drawing construction, students monitor and regulate the drawing construction process by iteratively consulting the instructional materials, updating their mental

Educ Psychol Rev

representations, and making necessary modifications to their drawings. This involves setting goals or standards of performance, applying specific strategies to achieve those goals, and evaluating whether one’s performance has met the standards (Nelson and Narens 1990; Winne and Hadwin 1998). For example, when students are prompted to draw, they must determine what to draw, including which elements, how much detail, and how many drawings to include. They may then apply other strategies to help them construct a descriptive representation of the lesson (such as by summarizing) and a depictive representation (such as by imagining) to support the construction of their drawing. Throughout the process, learners continually monitor their performance to evaluate whether their drawing accurately and sufficiently depicts the structural relations described in the text. Van Meter and Firetto propose that drawing uniquely encourages self-monitoring because the act of constructing an integrated external visuospatial representation (i.e., converting text into an image) confronts students with information that explicitly reveals their level of understanding. In other words, it should be clear to students when they are unable to create quality drawings, thereby providing important cues that enhance metacognitive accuracy and, potentially, subsequent self-regulation (Schleinschok et al. 2017; Thiede et al. 2010). A few important implications follow from past theoretical accounts of learning by drawing (Fiorella and Mayer 2015; Van Meter and Garner 2005; Van Meter and Firetto 2013). First, students must have the requisite cognitive strategies and prior domain knowledge available to convert the text into an accurate image. Further, the process of constructing a drawing must not exceed students’ limited working memory resources. This suggests that students with low prior knowledge or cognitive ability may need explicit guidance to support quality drawing construction. The specific type or level of guidance provided will influence how students regulate and execute the drawing strategy during learning. Finally, drawing is expected to provide unique cognitive and metacognitive benefits compared to other strategies or methods that do not require forced integration of multiple representations or that do not provide explicit self-monitoring cues.

Previously Drawn Boundaries of Learning by Drawing In their early review, Van Meter and Garner (2005) presented evidence supporting three core hypotheses concerning the boundary conditions of learning by drawing. First, drawing will only be effective if students are able to create quality drawings that accurately depict the structural relations and processes described in the text. Second, to create quality drawings, learners with relatively low prior knowledge will need strong levels of external support. For example, learners might be asked to compare their created drawings to an instructor-provided illustration (Van Meter 2001). Finally, Van Meter and Garner posited that drawing is most effective for fostering meaningful learning outcomes, such as comprehension and transfer, rather than rote outcomes, such as recall of individual idea units, consistent with the classification of drawing as a model-focused, or generative learning strategy (Fiorella and Mayer 2015; Leopold and Leutner 2012). More recent reviews generally support the hypotheses proposed by Van Meter and Garner (Fiorella and Mayer 2015, 2016a; Leutner and Schmeck 2014). For example, several studies have demonstrated a strong relationship between drawing quality and learning—or what is referred to as the prognostic drawing effect (Schwamborn et al. 2010). Furthermore, almost all past studies have focused on students with low prior knowledge and many studies include

Educ Psychol Rev

some form of external drawing support, such as providing students with partially completed illustrations (e.g., Schmeck et al. 2014; Schwamborn et al. 2010, 2011). However, as discussed below, few studies beyond the early work of Van Meter and colleagues have systematically compared different levels of drawing support. Finally, whereas classic drawing research did not always include measures of meaningful learning (e.g., Lesgold et al. 1975), contemporary research often includes measures of comprehension and transfer (e.g., Leopold and Leutner 2012; Schmeck et al. 2014). This research generally demonstrates that asking students to create drawings of scientific texts supports meaningful learning outcomes—or what is referred to as the generative drawing effect (Schwamborn et al. 2010)—but does not necessarily promote rote outcomes such as recognition test performance (e.g., Van Meter et al. 2006). Despite these advances, open questions remain regarding the specific boundary conditions of learning by drawing. First, the evidence for drawing is often based on comparisons to students who only read (or reread) the text or use a text-focused strategy. Previous reviews have not systematically examined how drawing compares to stronger control conditions, making it difficult to determine whether drawing uniquely supports learning beyond other model-focused strategies or beyond situations in which learners are provided with illustrations from the instructor. Second, although past research suggests that students need guidance to support the construction of quality drawings, researchers have employed a wide range of guidance methods that vary in their level of support and proposed functions. Previous reviews have not systematically distinguished among these different levels of drawing guidance, making it difficult to derive precise predictions regarding when, how, and for whom different types of support are likely to be effective compared to other strategies. In the following sections, we address these two issues by synthesizing the past empirical research on learning by drawing.

The Present Review In the present review, we analyzed peer-reviewed articles testing the effects of creating drawings while learning from a scientific text on meaningful learning outcomes. Articles were first selected from four published reviews of the learning by drawing literature (Fiorella and Mayer 2015; Leutner and Schmeck 2014; Van Meter and Firetto 2013; Van Meter and Garner 2005). We then selected from recently published articles that cited these past reviews or any of the empirical studies reported within them. Studies were excluded from our primary review if they involved a domain outside of science (e.g., math problem solving; Van Essen and Hamaker 1990), if they involved learning materials other than text (e.g., animations; Ploetzner and Fillisch 2017), or if they did not include a measure of meaningful learning outcomes (e.g., verbatim recall; Lesgold et al. 1975). This was done to focus on studies directly relevant to the role of comparison conditions and drawing guidance on meaningful learning from scientific texts. In total, 14 studies (a total of 17 unique experiments) met our criteria for inclusion in the primary review. Although these studies are the focus of our review, at the end of this article, we also consider how drawing has been studied within other learning contexts and how future work can broaden the study of learning by drawing. In examining the core evidence, we analyzed how the drawing condition(s) from each experiment performed on measures of meaningful learning compared to each of four control conditions: read only, text-focused strategy, model-focused strategy, or provided illustrations. Learning outcomes consisted of comprehension, indicated by the ability to explain or make

Educ Psychol Rev

inferences from key concepts presented in the text; or transfer, indicated by the ability to apply knowledge of key concepts from the text to new situations. Next, we classified each study based on the level(s) of drawing guidance employed, distinguishing among four levels of guidance, which we describe later. Finally, we coded whether the study controlled for total learning time, a limitation of some past studies on learning by drawing, and an especially relevant factor when comparing drawing to studying instructor-provided illustrations. Table 2 summarizes the key characteristics and findings of each study across comparison conditions, level of drawing guidance, and learning outcomes. In total, our review yielded 49 comparisons between drawing conditions and non-drawing control conditions on measures of comprehension or transfer. We calculated Cohen’s d effect sizes for each comparison, with values of .2, .5, and .8 corresponding to small, medium, and large effects, respectively (Cohen 1992). The mean effect size across all studies was .41 for comprehension (based on 20 comparisons; median = .49) and .37 for transfer (based on 29 comparisons; median = .32). This is comparable to an earlier review by Fiorella and Mayer (2015), which reported an overall median effect size of .40 across 28 comparisons. As we discuss below, however, the effect of creating drawings on learning may depend on the control conditions to which drawing is compared and the level of drawing guidance provided.

Learning by Drawing—Compared to What? According to Chi’s ICAP framework (Chi and Wylie 2014), the critical test of a model-focused (or constructive) learning activity is when it is compared to other model-focused strategies. Such a comparison can help clarify theoretical and practical issues regarding unique benefits of a specific strategy. Unfortunately, these important comparisons are rare in the learning strategies literature (e.g., Fiorella and Mayer 2015; Fonseca and Chi 2011). Indeed, our review indicated that drawing is most often compared to only reading the text or using a text-focused strategy. However, some recent studies have compared drawing to other learning strategies, including other model-focused strategies (e.g., Leutner et al. 2009; Scheiter et al. 2017; Schmidgall et al. 2018), as well as situations in which students are not instructed to use a strategy but are provided with both the text and instructor illustrations (e.g., Schmidgall et al. 2018; Van Meter et al. 2006). Table 3 presents the mean effect sizes for comprehension and transfer when drawing is compared to (a) reading only or using a text-focused strategy, (b) using a model-focused strategy, or (c) studying an instructor-provided illustration.

Drawing Compared to Reading Only or Text-Focused Strategies As shown in Tables 2 and 3, the existing evidence consistently indicates that students who are asked to create drawings perform better on comprehension and transfer tests than students only asked to read (or reread) the text or asked to use text-focused strategies, with a mean effect size of .46 for comprehension (based on 12 comparisons) and .70 for transfer (based on 15 comparisons). The stronger effect size for transfer performance is consistent with the prediction that drawing is most effective for deeper learning (Van Meter and Garner 2005). A classic study by Alesandrini (1981) found that creating drawings of electrochemistry concepts led to better comprehension test performance than only reading or paraphrasing the text. A more recent study by Leopold and Leutner (2012) found that students who created drawings to

High school

High school

Leopold and Leutner 2012 (Exp. 2)

Leopold et al. 2013

a

High school High school

High School

Leopold and Leutner 2012 (Exp. 1)

Schmeck et al. 2014 (Exp. 1) Schmeck et al. 2014 (Exp. 2)

High school High school

Schwamborn et al. 2010 Schwamborn et al. 2011

a

High school

Elementary

Leutner et al. 2009

Van Meter et al. 2006

Elementary

Van Meter 2001

a

Elementary

College

Gobert and Clement 1999

Hall et al. 1997

College

Alesandrini 1981

a

Sample

Experiment

Influenza Influenza

Water molecules

Water molecules

Water molecules

Washing clothes Washing clothes

Water molecules

Bird wings

Nervous system

Plate tectonics

Air pump

Electrochemistry

Topic

Read only Model-focused Read only Read only Provided illustration Read only Provided illustration Read only Text-focused Read only Text-focused Read only Text-focused Read only Text-focused Text-focused Provided illustration Text-focused Provided illustration Read only Read only

Provided illustration

Read only Text-focused Read only Provided illustration Read only Text-focused Provided illustration

Comparison

Table 2 Evidence for learning by drawing across comparison conditions, guidance level, and learning outcomes

Comprehension Comprehension

Transfer

Level 1 Level 3 Level 3

Comprehension

Transfer

Level 1 Level 1

Comprehension

Transfer

Level 1 Level 1

Comprehension

Transfer

Levels 3 and 4 Level 1

Transfer Transfer

Comprehension

Transfer

Comprehension

Transfer

Transfer

Comprehension

Learning outcome

Level 3 Level 3

Level 1 Level 4 Level 4+ Level 1 Level 4 Level 4+ Level 1

Level 1

Level 1

Level 1

Guidance

.56 .26 .71 .26 1.02 .51 .72 .76 1.08 .23 .64 .70 .23 − .57 .91 .17 − .31 .44 .00 .15 .81 .32 1.17 .15 .55 .69 1.08 .89 − .25 .72 − .74 .85 .52

d

Educ Psychol Rev

Biomechanics of swimming

Greenhouse effect Biomechanics of swimming

Cardiovascular system

Topic Provided illustration Read only Provided illustration Read only Model-focused Model-focused Read only Text-focused Provided illustration Read only Text-focused Provided illustration Model-focused Provided illustration (dynamic) Model-focused Provided illustration (dynamic)

Comparison

Transfer (delayed) Comprehension Transfer Comprehension Transfer

Level 1 Mean Median

Transfer (immediate)

Transfer (delayed)

Level 1

Level 1

Transfer Transfer (immediate)

Comprehension

Comprehension

Learning outcome

Level 2 Level 1

Level 1

Levels 3 and 4

Guidance

.49 − .05 − .10 .62 .40 .21 −.03 .57 .03 .40 1.83 .05 − .11 − .27 − .25 − .17 .41 .37 .49 .32

d

a

Learning time was not controlled across experimental conditions

Drawing guidance: level 1 = minimal guidance; level 2 = drawing training; level 3 = provided partially completed illustrations; level 4 = prompted comparison to provided illustrations (level 4+ = specific prompting questions for comparison to provided illustrations)

College

High school College

Scheiter et al. 2017 a Schmidgall et al. 2018 (Exp. 1)

Schmidgall et al. 2018 (Exp. 2)

College

Lin et al. 2017

a

Sample

Experiment

Table 2 (continued)

Educ Psychol Rev

Educ Psychol Rev Table 3 Mean effect sizes of learning by drawing by comparison condition and learning outcomes Comparison condition

k

Learning outcomes

d

Read only or text-focused

12 15 2 3 6 11

Comprehension Transfer Comprehension Transfer Comprehension Transfer

.46 .70 − .09 − .05 .45 .04

Model-focused Provided illustrations

represent the structure of water molecules outperformed students who wrote verbal summaries on both comprehension and transfer tests. These findings are consistent with past reviews of learning by drawing and with predictions from theories of text comprehension and drawing construction. Text-focused strategies may help students select the key propositions from the lesson to create a text base, but modelfocused strategies like drawing support the deeper generative processes (organizing and integrating) necessary for constructing a situation model. Furthermore, past research suggests that when unprompted to use a specific strategy (i.e., read only conditions), students rely heavily on text-focused strategies (e.g., Fiorella and Mayer 2017). In short, asking students to create drawings while learning from a scientific text is likely to result in better meaningful learning outcomes than asking students to only read the text or use text-focused strategies.

Drawing Compared to Other Model-Focused Strategies Drawing is a model-focused learning strategy because it aims to help students create an integrated mental representation, or situation model, of the to-be-learned material during learning. An important theoretical and practical issue concerns how drawing compares to other model-focused strategies (Chi and Wylie 2014). In other words, to what extent does creating drawings uniquely impact cognitive processing and learning? As presented in Tables 2 and 3, the available research evidence comparing drawing to model-focused strategies such as imagining and self-explaining is mixed. First, a study by Leutner et al. (2009) found that creating drawings to depict the content of a text on the dipole character of water led to significantly worse comprehension performance than mentally imagining the material (d = − .57). This negative effect of drawing was mediated by higher levels of self-reported cognitive load, suggesting that externalizing a mental image on paper can create extraneous cognitive demands. In contrast, a study by Lin et al. (2017) reported no significant differences between students who created drawings and students who formed mental images during learning (d = .40, favoring the drawing group). Similarly, a recent study by Schmidgall et al. (2018) found no benefits of generating drawings compared to imagining on immediate (d = −.11) or delayed (d = −.25) transfer tests. Finally, in a recent study by Scheiter et al. (2017), high school students learned from a scientific text about the greenhouse effect by either creating drawings or generating written self-explanations. The results indicated no overall difference between the two strategies (d = .21, favoring the drawing group), although creating higher quality drawings was more strongly associated with learning outcomes than generating higher quality self-explanations. The studies described above suggest that drawing is not necessarily more effective than other model-focused strategies such as imagining or self-explaining, though further research is

Educ Psychol Rev

needed. On the one hand, model-focused strategies all aim to support similar cognitive activity—the construction of a situation model—and therefore may not be expected to differentially impact learning. This view emphasizes the similarities among model-focused strategies and the idea that there are multiple routes to support similar processes of knowledge construction (e.g., selecting, organizing, and integrating). Lin et al. (2017) concluded that drawing and imagining B…may engage learners in a similar manner by facilitating dual processing and building referential connections between words and pictures^ (p. 11). On the other hand, there are important differences in the behavioral activity among modelfocused strategies, which can impose different cognitive demands on learners and make different aspects of the learning material explicit to students. In comparing drawing and imagining, constructing a drawing by hand can be more time consuming and result in higher extraneous cognitive load than generating a mental image (Leutner et al. 2009). Yet drawing may allow students to externalize (i.e., offload) their mental representation processes onto the paper, whereas imagining requires students to maintain and update their mental image in working memory (e.g., Cox 1999; Hegarty and Steinhoff 1997; Schmidgall et al. 2018). Similarly, other researches on imagining suggest it may be most effective for students with high prior knowledge (Cooper et al. 2001), whereas some have suggested that drawing is more appropriate for students with lower prior knowledge (Fiorella and Mayer 2015; Lin et al. 2017). In comparing drawing and self-explaining, drawing involves generating an external nonverbal representation, whereas self-explaining involves generating an external verbal representation (e.g., written or spoken words). Self-explaining does not require learners to generate an overt nonverbal representation, but it similarly serves to support the construction of a nonverbal (visuospatial) mental representation, or mental model, in learners (e.g., Chi 2000). Furthermore, self-explaining may be less cognitively demanding and allow for greater elaborations on the material than creating drawings by hand. It is important to note that the one study by Schleinschok et al. (2017) comparing self-explaining and drawing asked students to create written (rather than spoken) explanations, which can lead to more organized but less elaborated explanations (Lachner et al. 2017). For its part, drawing can offer unique benefits beyond self-explaining because it requires forced integration between verbal and nonverbal representations (Van Meter and Firetto 2013), and it makes the spatial relations of the learning material explicit (Bobek and Tversky 2016). Given their different strengths, it also seems likely that self-explanation and drawing can serve as mutually supportive strategies, such that drawing may support better quality self-explanations (Cox 1999), although this prediction has not yet been directly tested in the literature. Taken together, strong conclusions cannot be drawn regarding the relative effectiveness of drawing, imagining, and self-explaining. Further research is needed to disentangle the cognitive costs and benefits associated with the behavioral differences across these and other modelfocused strategies (e.g., creating concept maps), as well as the potential interactions among the use of multiple model-focused strategies (e.g., self-explaining and drawing). It is also possible that the effectiveness of drawing compared to other model-focused strategies depends in part on the level of drawing guidance students receive, an issue we will return to later.

Drawing Compared to Instructor-Provided Illustrations Another important consideration is whether creating drawings from provided text is more effective than reading the text and viewing instructor-provided illustrations. Research on learning from multimedia indicates that adding instructor-provided visuals to text generally

Educ Psychol Rev

results in better understanding than when students learn from text alone (Mayer 2014). However, studies also suggest that provided visuals are often processed passively (e.g., Hannus and Hyönä 1999), and students may not spontaneously integrate provided words and visuals into a coherent representation (e.g., Hegarty et al. 1991). Drawing may support more active cognitive processing because students are required to convert the text into their own representative illustration (Fiorella and Mayer 2015; Van Meter and Garner 2005). However, drawing may also be too cognitively demanding and time consuming (Leutner et al. 2009; Van Meter 2001), and students may not be able to create quality drawings on their own (Van Meter 2001). As shown in Tables 2 and 3, research comparing provided illustrations to generated drawings has produced mixed results, with an overall mean effect size of .45 for comprehension (based on 6 comparisons) and .04 for transfer (based on 11 comparisons). Studies have found that asking students to create drawings is more effective (Van Meter 2001; Van Meter et al. 2006; Schmeck et al. 2014), less effective (Leopold et al. 2013; Schwamborn et al. 2011; Van Meter 2001; Van Meter et al. 2006), and not significantly different (Hall et al. 1997; Schwamborn et al. 2011; Schmeck et al. 2014; Schmidgall et al. 2018) than providing students with illustrations. Alesandrini (1981) concluded based on early work that B…science learning is better facilitated by showing relevant pictures rather than asking learners to draw their own pictures^ (p. 366). In contrast, Hall et al. (1997) posited that B…an accurate representation generated by a student from a text is as good as or better than a picture provided by the experimenter^ (p. 679). Research by Van Meter and colleagues (Van Meter 2001; Van Meter et al. 2006) has consistently found positive effects of constructing drawings compared to studying provided illustrations, particularly when students were provided with high levels of drawing guidance. However, it is important to note that these studies did not control for differences in the amount of time students spent studying—students who constructed drawings spent considerably more time with the learning material than students who studied provided illustrations. Other studies by Schmeck and colleagues (Schmeck et al. 2014; Schwamborn et al. 2011) have found mixed effects of drawing compared to studying provided illustrations when learning time was controlled. The recent study by Schmidgall and colleagues found no significant differences between drawing and studying provided illustrations on both immediate and delayed transfer tests. In their experiment 1, drawing was compared to provided static visuals, whereas in experiment 2, drawing was compared to observing videos of the instructor dynamically create

Table 4 Levels and functions of drawing guidance used in past research Level Guidance 1 2

3

4

Minimal

Timing Description

Proposed Function(s)

Before Basic instructions to create a Induce strategy selection drawing Drawing Before Explicit training and practice in Provide strategic knowledge; reduce training what and how to draw extraneous cognitive load; increase drawing quality Constrain drawing process; increase drawing Partially During Partially completed quality; reduce extraneous cognitive load completed illustrations of the learning material Feedback; foster self-reflection; constrain Comparison After Prompts to compare drawing construction and increase drawing to provided learner-generated with quality (during revision) instructor-provided illustrations

Educ Psychol Rev

the videos (e.g., Fiorella and Mayer 2016b). In both experiments, students who created drawings spent significantly more time on task; thus, there was again evidence that viewing instructor-provided illustrations (static or dynamic) may be more efficient. Overall, the findings comparing learner-constructed and instructor-provided visuals are mixed, though they also may be moderated somewhat by the level of drawing guidance provided, an issue we turn to next.

Levels of Drawing Guidance According to Van Meter and colleagues (Van Meter and Firetto 2013; Van Meter and Garner 2005), drawing guidance primarily serves to enhance drawing quality by constraining the drawing process or providing students opportunities to check the accuracy of their drawings. Schwamborn et al. (2011) similarly characterized the benefits of guidance in terms of reduced extraneous cognitive load—that is, cognitive processing irrelevant to constructing a situation model. However, past research has not clearly distinguished between different levels of drawing guidance, which can vary considerably from minimal instructions to explicit support and feedback. This creates challenges when interpreting findings across studies, including how drawing compares to other learning contexts, as discussed above. As summarized in Table 4, our review of the literature identified four distinct levels of guidance: minimal guidance, drawing training, providing partially completed illustrations, and comparing one’s drawing to a provided illustration. Although similar guidance methods have been identified in other instructional contexts (e.g., pre-training, scaffolded or faded instruction, feedback), past research on drawing has not produced a clear framework for distinguishing between levels of drawing guidance and few studies have systematically compared different levels of guidance. After briefly describing each of the levels of guidance below, we discuss how this new framework can be used in the context of learning by drawing to interpret existing findings and guide future work.

Level 1: Minimal Guidance Minimal guidance methods provide students with basic instructions to draw, without providing explicit support or practice in how to create quality drawings. At this level, students may receive initial instructions and an example but do not receive additional assistance during or after drawing construction. For example, the classic study by Alesandrini (1981) provided students with simple instructions to either focus on specific details (analytic strategy) or more general concepts (holistic strategy) when creating drawings to represent concepts in electrochemistry—although these different types of instructions did not differentially impact the effectiveness of drawing. Other studies have employed minimal guidance but without differentiating and comparing different types of instructions. For example, Hall et al. (1997) instructed students to read a text-based lesson about how a tire pump works and draw a corresponding picture. Students received specific instructions on the elements that should be included in their drawing (e.g., the piston, inlet valve, outlet valve) but did not receive any additional support to foster drawing quality. The study by Leopold and Leutner (2012) implemented level 1 guidance by providing students with a simple example drawing before students created drawings on their own while learning about the structure of water molecules. In short, the primary function of minimal guidance is to induce students to Bcreate a drawing^ during learning, but students are ultimately on their own throughout the drawing process.

Educ Psychol Rev

Level 2: Drawing Training Drawing training methods provide students with explicit training and practice on how to draw before students are expected to create drawings on their own during learning. Surprisingly, only one study applied level 2 guidance to test the effects of creating drawings on science text comprehension. The study by Scheiter et al. (2017) provided high school students with direct instruction, instructor modeling, and practice in how to create drawings from example text passages. Then students created drawings on their own while learning from a chemistry text. Other students received similar pre-training in how to generate written self-explanations of the material. Although the effects of drawing training have not often been tested, the primary function of pre-training methods tested in other contexts is to provide students with background knowledge prior to learning (e.g., Mayer et al. 2002), thereby helping students manage the complexity of the material during learning and enhancing understanding. Therefore, drawing training may serve to build and automatize strategic knowledge, reducing the cognitive demands associated with creating drawings during learning. This should result in better quality drawings and meaningful learning outcomes, such as when compared to minimal (level 1) forms of drawing guidance.

Level 3: Partially Completed Illustrations Level 1 and 2 methods provide instructions and support prior to learning but do not support the drawing process during learning. Level 3 methods constrain drawing construction during learning by presenting learners with partially drawn illustrations, such as background templates or images of essential elements from the learning materials. Therefore, students are not expected to generate visuals on their own but to create drawings that integrate provided visuals. For example, in the study by Schwamborn et al. (2010), ninth graders created drawings from a scientific text explaining the chemical processes involved in washing laundry with soap and water. To support drawing construction, students were provided with drawings of all the relevant elements (e.g., water molecules, hydrogen bond) and a partially drawn background. Students used the provided elements to construct their own drawings on the background. Another study by Schmeck et al. (2014) followed a similar procedure, providing students with partially completed illustrations of the key elements for understanding the biological processes of influenza. Since students are provided with the visuals they need to construct their drawings, they do not have to rely as heavily on their prior knowledge to connect the text to stored images or to convert the text into a new visual representation. Therefore, level 3 methods primarily serve to constrain the drawing construction process, thereby reducing extraneous cognitive load and enhancing drawing accuracy.

Level 4: Comparison with Provided Illustrations Level 4 methods prompt students to compare their own self-generated drawings to an instructor-provided illustration. Therefore, whereas level 3 methods provide students with partially completed illustrations to use during the drawing process, level 4 methods provide students with a fully completed drawing for them to use as feedback after they create their drawings. For example, in the studies by Van Meter and colleagues (Van Meter 2001; Van Meter et al. 2006), students created their own drawings representing texts about the nervous system or the structure and function of a bird’s wings. Then they were shown an instructorcreated illustration and asked to compare it to their own. Some students in Van Meter’s studies were additionally provided with explicit prompting questions designed to guide them during

Educ Psychol Rev Table 5 Mean effect sizes of learning by drawing by comparison condition and guidance level Comparison condition

Guidance level

k

Learning outcomes

d

Read only or text-focused

Level 1

9 12 2 2 1 1 2 2 1 2 7 1 1 1 1 2 2

Comprehension Transfer Comprehension Transfer Comprehension Transfer Comprehension Transfer Transfer Comprehension Transfer Comprehension Transfer Comprehension Transfer Comprehension Transfer

.46 .75 .69 .54 − .05 .44 − .09 − .18 .21 .24 − .09 .49 − .31 − .10 .00 .92 .67

Level 3 Levels 3 and 4 Model-focused

Level 1

Provided illustrations

Level 2 Level 1 Level 3 Levels 3 and 4 Level 4/4+

the comparison process. Students then had the opportunity to revise their own drawing based on the comparisons they made with the instructor-provided illustrations. Overall, level 4 methods involve both prompted comparison between learner-generated drawings to provided illustrations and an opportunity to revise one’s drawings accordingly. Therefore, level 4 guidance primarily functions as feedback, encouraging students to reflect on their own drawing performance and correct inaccuracies or misconceptions, thereby constraining drawing construction and enhancing drawing quality during revision.

Implications of Drawing Guidance Framework The levels of guidance described above highlight the wide range of methods used in past research on learning by drawing. As summarized in Table 4, these methods vary based on the time at which they are implemented in the drawing process (before, during, or after), the amount of support they provide students, and their primary instructional functions. Given that relatively few studies have systematically compared the effects of different levels of drawing guidance, the proposed framework primarily aims to guide future research exploring the benefits and boundaries of drawing, though it can also be used to provide some additional insight into the existing evidence regarding how drawing compares to other learning strategies. Table 5 presents the average effect sizes of learning by drawing by comparison condition and level of guidance. First, when drawing is compared to only reading the text or using textfocused strategies, level of drawing guidance does not appear to play a significant role. That is, even studies utilizing minimal guidance (level 1) generally report positive effects of drawing (e.g., Alesandrini 1981; Gobert and Clement 1999; Hall et al. 1997). This is consistent with the idea that simply prompting students to construct drawings is generally more effective than the types of strategies (i.e., read only or text-focused strategies) that students generally employ spontaneously when learning from text (Fiorella and Mayer 2017). Next, the four studies comparing drawing to other model-focused strategies implemented either level 1 guidance (Leutner et al. 2009; Lin et al. 2017; Schmidgall et al. 2018) or level 2

Educ Psychol Rev

guidance (Schleinschok et al. 2017). It is possible that the null or small effects from these studies were in part because students needed higher levels of drawing guidance during (level 3) or after (level 4) the drawing construction process. For instance, the study by Leutner et al. (2009) found that drawing resulted in higher extraneous cognitive load than imagining and led to worse learning outcomes. Stronger drawing support may constrain the drawing process and help students better manage the cognitive demands of converting one’s mental image into an external drawing. At present, the available research evidence is too limited to draw strong conclusions regarding the relative effectiveness of model-focused strategies and the role of strategy guidance; this is an important direction for future research. Finally, Table 5 indicates that when drawing is compared to instructor-provided illustrations, level of guidance may moderate the effectiveness of drawing. Drawing was often less effective than studying provided illustrations when students were only provided with minimal (level 1) drawing guidance, but it was at least as effective (and sometimes more effective) when students were provided with the highest level of drawing guidance (level 4). However, when interpreting this research, one should also consider the additional time and mental effort demands associated with constructing drawings (Leutner et al. 2009; Van Meter 2001; Schmidgall et al. 2018), and the additional instructional resources required to implement high levels of drawing guidance. Taken together, providing illustrations may in some cases provide a more efficient route to a similar learning outcome, especially when the illustrations are welldesigned (Mayer 2014). At the same time, when drawing construction is strongly supported, it may provide unique benefits beyond viewing provided illustrations, such as by forcing students to integrate the text with a nonverbal representation and by encouraging students to actively self-monitor and revise inaccuracies. Prompting students to compare their drawings to provided illustrations (level 4) may help optimize the benefits of both learner-generated drawing and learning from provided visuals. Another possibility is that simply observing the instructor dynamically create drawings by hand more efficiently provides the benefits of learner-constructed and instructor-provided visuals (Fiorella and Mayer 2016b). In short, further research is needed to clarify the unique and interacting effects of different levels of guidance, particularly when drawing is compared to stronger control conditions.

Broadening Research on Learning by Drawing In our review of the literature above, we focused on studies testing the effects of creating drawings on science text comprehension. Although this has been the focus of much past research, recent work has explored other important aspects of learning by drawing, including its metacognitive effects, its association with student individual differences, its effectiveness with other types of learning materials, and its use as a retrieval activity. This related work provides further insight into when and how drawing might uniquely support learning, particularly considering our findings regarding comparison conditions and guidance. In the sections below, we discuss the implications of this work for understanding boundary conditions and for broadening future research on learning by drawing.

Role of Metacognition in Learning by Drawing One potential unique benefit of creating drawings is that it helps students better monitor and regulate their learning (Van Meter and Firetto 2013). This metacognitive advantage may especially apply when drawing is compared to viewing instructor-provided visuals. For

Educ Psychol Rev

example, in the early study by Van Meter (2001), participants in drawing conditions engaged in significantly more self-monitoring statements than participants who read texts with provided illustrations. This difference only appeared while participants were inspecting provided illustrations (not while reading or drawing), suggesting that comparing one’s drawings to provided illustrations may offer unique metacognitive benefits compared to only viewing provided illustrations. This is important because monitoring accuracy is generally expected to influence subsequent self-regulation behavior (Koriat 1997; Pilegard and Fiorella 2016; Thiede et al. 2010). However, one recent study by Schleinschok et al. (2017) indicated that while drawing was related to better metacognitive monitoring (i.e., more accurate judgments of learning), it did not translate into better subsequent regulation, as indicated by students’ restudy decisions. Further work is needed to clarify the metacognitive effects of creating drawings compared to other learning activities. For example, self-explaining can also help students monitor their understanding and revise their mental models during learning (Chi 2000). Research should also explore how different forms of guidance may be used to support self-regulatory processes during learning by drawing.

Role of Spatial Ability in Learning by Drawing There is also some evidence that individual differences in spatial ability may moderate the effects of learning by drawing. For example, one recent experiment by Bobek and Tversky (2016) indicated that creating drawings after learning a lesson about the mechanics of a bicycle pump only benefited students with lower spatial ability (indicated by performance on a comprehension test), when compared to students who wrote verbal explanations. In contrast, drawing may not be as effective or necessary for students with higher levels of spatial ability. Such students may instead benefit more from learning by imagining, a model-focused strategy that relies on mentally forming images and does not require creating external representations by hand (Leutner et al. 2009). However, more work is needed because few past studies have explicitly focused on the role of spatial ability or prior knowledge in learning by drawing. Most studies that incorporate spatial ability measures often do so to verify homogeneity across experimental conditions (e.g., Leopold and Leutner 2012; Lesgold et al. 1975; Schmeck et al. 2014) rather than to test for interactions. A recent study by Fiorella and Mayer (2017) found complementary evidence that drawing may help compensate for lower spatial ability. In this study, students who spontaneously used more spatial note-taking strategies (which included maps or drawings) while studying a scientific text about the human respiratory system performed better on a subsequent comprehension test than students who used more text-focused strategies (e.g., creating lists). Importantly, spontaneous use of spatial strategies predicted comprehension performance even after accounting for the role of general spatial ability, suggesting that spatial ability and model-focused strategy use (including drawing) provide somewhat independent paths for enhancing learning from scientific texts. One implication is that rather than attempting to train general spatial ability, which may not transfer to educational tasks (Stieff and Uttal 2015), students with low spatial ability might benefit from learning to construct quality drawings, which can be achieved with appropriate forms of guidance. Future research should systematically explore how learning by drawing interacts with individual differences and at different levels of drawing guidance.

Educ Psychol Rev

Learning by Drawing Across Lesson Formats and Topics Although most drawing research has focused on scientific texts, some studies have investigated the effects of creating drawing on learning from other lesson formats and topics, including math word problems, dynamic visuals (e.g., animations and simulations), video lessons, and physical models (Van Essen and Hamaker 1990; Mason et al. 2013; Zhang and Linn 2011; Zhang and Linn 2013; Ploetzner and Fillisch 2017; Bobek and Tversky 2016). Recent studies exploring the effects of drawing while learning from animations have produced mixed results. Mason et al. (2013) found benefits of generating drawings when compared to copied drawing or no drawing when learning from an animation on Newton’s laws of motion. In contrast, Ploetzner and Fillisch (2017) showed no benefits of drawing compared to prompted reflection when students learned from an animated lesson on how a single-cylinder engine works. Ploezner and Fillisch argued that while past studies showed benefits of drawing when learning from relatively simple animations (Mason et al. 2013; Zhang and Linn 2011), creating static drawings may not be appropriate for representing dynamic spatiotemporal changes in complex systems. Another possibility is that drawing may not be as effective when it involves learning from provided visualizations because students are not translating across representations (i.e., from verbal to nonverbal; Cox 1999), since a visuospatial representation is already provided. In such cases, it may be more appropriate to self-explain the provided visuals (e.g., Ainsworth and Loizou 2003; Butcher 2006). Most of past research on learner-generated drawings has involved materials in STEM disciplines. This is likely because STEM relies heavily on visualizations for researching, communicating, and teaching complex concepts, and therefore, these fields may be more conducive to drawing (Ainsworth et al. 2011). As discussed, drawing can help make underlying nonlinear structural relations more explicit and salient than text-focused strategies, such as summarizing (Leopold and Leutner 2012). Although drawing may be uniquely suited for STEM disciplines, future research should investigate its efficacy across other disciplines, such as the humanities and social sciences. It is possible that in such disciplines, strategies such as self-explaining or creating more abstract spatial representations (e.g., concept maps or graphic organizers) may be more appropriate.

Learning by Drawing as a Retrieval Activity Finally, drawing has mostly been studied as an encoding strategy that students employ during learning. However, learners can also benefit from creating drawings after learning as a retrieval activity following learning (Bobek and Tversky 2016). For example, a recent study by Bobek and Tversky (2016) found that creating drawings after interacting with a physical model or watching a video lesson was more effective than writing verbal explanations. In line with research on retrieval-based learning (Roediger and Karpicke 2006), practice in retrieving information from memory supports long-term learning because it consolidates the information and makes it more accessible over time. At present, few studies have used drawing as a retrieval activity or examined the long-term effects of creating drawings (e.g., Mason et al. 2013; Wu and Rau 2017; Schmidgall et al. 2018). Further research exploring the timing of drawing and its effects on long-term learning is important for clarifying when drawing is most effective and whether it provides unique benefits. For example, given the effortful and constructive nature of drawing (either as an encoding or retrieval strategy), its learning benefits compared to studying provided illustrations might be most pronounced on delayed measures

Educ Psychol Rev

of learning, though this idea of drawing as a desirable difficulty was not supported in the recent study by Schmidgall et al. (2018).

Recommendations for Practice The evidence reported above provides implications for implementing drawing into authentic educational contexts. First, when compared to the types of learning strategies students typically use (i.e., read only or text-based strategies), drawing can be an effective learning activity. These benefits have primarily been found for learning from scientific text materials that describe complex spatial relationships. Other strategies might be more appropriate for learning from different types of materials. For example, self-explaining can be an effective strategy when learners are provided with an instructional visual (Ainsworth and Loizou 2003). Second, to achieve the most benefit from learning by drawing, students need to be able to generate quality drawings that accurately depict the different structural elements and relationships that comprise the system. Therefore, students may need explicit instructional support, such as by training students how to translate texts into drawings, providing partially completed illustrations as drawing scaffolds, or prompting students to compare their drawings to a complete instructor-provided illustration and then to revise their drawings accordingly. Third, when an appropriate illustration is not available, students are likely to benefit from constructing their own drawing, especially when provided with appropriate guidance. However, when an appropriate illustration can be made available to students, asking students to instead create their own drawings may be inefficient, given the extra time demands required for drawing. In such cases, it is important for the provided visual to be well-designed and for students to be equipped with strategies that encourage students to actively integrate verbal and nonverbal information and construct meaning from the visuals, such as by self-explaining. Fourth, when strong guidance is provided, drawing may offer unique benefits that cannot be easily achieved through viewing provided illustrations. The act of constructing drawings may help students better monitor their own learning and identify specific gaps or misconceptions that are unique to them. To resolve such knowledge gaps, students may then benefit from comparing and revising their drawing based on an instructor-provided illustration. Similarly, to support long-term retention, constructing drawings may serve as a useful retrieval activity, which could then be followed up with a provided illustration to use as feedback. Finally, drawing is only one learning strategy among many others, each which can serve complementary instructional goals. Therefore, in addition to comparing strategies against each other, it is important to consider how strategies can complement each other. For example, drawing may support self-explaining, and imagining may support drawing, and vice versa. Though research is needed to directly address this issue, considering the different behavioral and cognitive activities associated with different learning strategies may provide educators with a useful heuristic for integrating these strategies into the classroom.

Limitations Given that learning by drawing is an emerging research area, there are not enough published studies to conduct a full-scale meta-analysis examining the role of potential boundary

Educ Psychol Rev

conditions such as comparison conditions and drawing guidance. The existing research also primarily consists of short-term controlled experiments that isolated the effects of drawing compared to other types of learning activities. As a result, our review focuses on studies that involved learning from scientific text and included measures of meaningful learning outcomes (e.g., comprehension and transfer). Therefore, conclusions in this review should be constrained within this context; the boundary conditions of drawing may be different for different learning materials or learning goals. In addition, there are other factors that may moderate the effectiveness of drawing beyond the scope of this review and the available research evidence. For example, learning by drawing may be influenced by specific features of the text, such as its coherence and complexity; the way in which drawings are constructed, such as by hand or on the computer; and other individual differences across students, such as drawing ability or drawing self-efficacy. Finally, although drawing is best characterized by constructing representative illustrations, the drawings students construct can occur at different levels of abstraction, such as using symbols or arrows, and they often incorporate text, such as using labels. Thus, in practice, drawing often extends beyond solely pictorial representations and can incorporate other types of representations. It is important for future research to address each of these issues to further specify the benefits and boundaries of learning by drawing.

Conclusions In conclusion, understanding the boundary conditions that constrain instructional methods and learning strategies is critical for advancing both learning theory and instructional practice. This review contributes to a shift toward understanding not only whether a learning strategy is effective, but under what conditions it is effective (i.e., compared to what other strategies and with which level of guidance). Given the inconsistencies of past research regarding comparison conditions and levels of guidance, we proposed several areas for future research to systematically analyze its unique benefits and boundaries and to broaden its scope across learning contexts. Our findings indicate that drawing is a promising strategy for fostering meaningful learning under certain conditions, and students and teachers should add it to their strategy toolbox. This review aimed to draw some important boundary conditions for determining when drawing serves as the appropriate tool and when students and teachers might be better served by selecting other tools. Acknowledgements We thank Deborah Barany for her helpful comments on an earlier draft of this article. Funding Information This research was supported by a grant from the National Science Foundation (1561728) awarded to Logan Fiorella.

References Ainsworth, S. (2006). DeFT: a conceptual framework for considering learning with multiple representations. Ainsworth, S., Prain, V., & Tytler, R. (2011). Drawing to learn in science. Science, 333(6046), 1096–1097. Ainsworth, S., & Loizou, A. T. (2003). The effects of self-explaining when learning with text or diagrams. Cognitive Science, 27(4), 669–681. Alesandrini, K. L. (1981). Pictorial-verbal and analytic-holistic learning strategies in science learning. Journal of Educational Psychology, 73(3), 358–368.

Educ Psychol Rev Bobek, E., & Tversky, B. (2016). Creating visual explanations improves learning. Cognitive Research: Principles and Implications., 1, 27. Bodemer, D., Ploetzner, R., Feurlein, I., & Spada, H. (2004). The active integration of information during learning with dynamic and interactive visualizations. Learning and Instruction, 14(3), 325–341. Butcher, K. R. (2006). Learning from text with diagrams: promoting mental model development and inference generation. Journal of Educational Psychology, 98(1), 182–197. Chi, M. T. H. (2000). Self-explaining expository texts: the dual processes of generating inferences and repairing mental models. In R. Glaser (Ed.), Advances in instructional psychology (pp. 161–238). Mahwah, NJ: Lawrence Erlbaum Associates. Chi, M. T., & Wylie, R. (2014). The ICAP framework: linking cognitive engagement to active learning outcomes. Educational Psychologist, 49(4), 219–243. Cohen, J. (1992). A power primer. Psychological Bulletin, 112(1), 155–159. Cook, M. P. (2006). Visual representations in science education: the influence of prior knowledge and cognitive load theory on instructional design principles. Science Education, 90(6), 1073–1091. Cooper, G., Tindall-Ford, S., Chandler, P., & Sweller, J. (2001). Learning by imagining. Journal of Experimental Psychology: Applied, 7(1), 68–82. Cox, R. (1999). Representation construction, externalized cognition and individual differences. Learning and Instruction, 9(4), 343–363. Fiorella, L., & Mayer, R. E. (2015). Learning as a generative activity: eight learning strategies that promote understanding. New York: Cambridge University Press. Fiorella, L., & Mayer, R. E. (2016a). Eight ways to promote generative learning. Educational Psychology Review, 28(4), 717–741. Fiorella, L., & Mayer, R. E. (2016b). Effects of observing the instructor draw diagrams on learning from multimedia messages. Journal of Educational Psychology, 108(4), 528–546. Fiorella, L., & Mayer, R. E. (2017). Spontaneous spatial strategy use in learning from scientific text. Contemporary Educational Psychology, 49, 66–79. Fonseca, B. A., & Chi, M. T. H. (2011). Instruction based on self-explanation. In R. E. Mayer & P. A. Alexander (Eds.), Handbook of research on learning and instruction. New York, NY: Rougledge. Gobert, J. D., & Clement, J. J. (1999). Effects of student-generated diagrams versus student-generated summaries on conceptual understanding of causal and dynamic knowledge in plate tectonics. Journal of Research in Science Teaching, 36(1), 39–53. Hall, V. C., Bailey, J., & Tillman, C. (1997). Can student-generated illustrations be worth ten thousand words? Journal of Educational Psychology, 89(4), 677–681. Hannus, M., & Hyönä, J. (1999). Utilization of illustrations during learning of science textbook passages among low-and high-ability children. Contemporary Educational Psychology, 24(2), 95–123. Hegarty, M., Carpenter, P. A., & Just, M. A. (1991). Diagrams in the comprehension of scientific texts. In Hr. Barr, M. L., Kamil, P. B. Mosenthal, & P. D. Pearson (Eds.), Handbook of reading research, Vol. 2 (pp. 641668). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Hegarty, M., & Steinhoff, K. (1997). Individual differences in use of diagrams as external memory in mechanical reasoning. Learning and Individual Differences, 9(1), 19–42. Kintsch, W. (1988). The role of knowledge in discourse comprehension: a construction-integration model. Psychological Review, 95(2), 163–182. Koriat, A. (1997). Monitoring one’s own knowledge during study: a cue-utilization approach to judgments of learning. Journal of Experimental Psychology: General, 126(4), 349–370. Lachner, A., Ly, K-T., & Nuckles, M. (2017). Providing written or oral explanations? Differential effects of the modality of explaining on students’ conceptual learning and transfer. Journal of Experimental Education. Leopold, C., & Leutner, D. (2012). Science text comprehension: drawing, main idea selection, and summarizing as learning strategies. Learning and Instruction, 22(1), 16–26. Leopold, C., Sumfleth, E., & Leutner, D. (2013). Learning with summaries: Effects of representation mode and type of learning activity on comprehension and transfer. Learning and Instruction, 27, 40–49. Leutner, D., & Schmeck, A. (2014). The generative drawing principle in multimedia learning. In R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning (2nd ed.) (pp. 433–448). New York: Cambridge University Press. Leutner, D., Leopold, C., & Sumfleth, E. (2009). Cognitive load and science text comprehension: effects of drawing and mentally imagining text content. Computers in Human Behavior, 25(2), 284–289. Lesgold, A. M., Levin, J. R., Shimron, J., & Guttman, J. (1975). Pictures and young children’s learning from oral prose. Journal of Educational Psychology, 67(5), 636–642. Lin, L., Lee, C. H., Kalyuga, S., Wang, Y., Guan, S., & Wu, H. (2017). The effect of learner-generated drawing and imagination in comprehending a science text. Journal of Experimental Education, 85(1), 142–154.

Educ Psychol Rev Mason, L., Lowe, R., & Tornatora, M. C. (2013). Self-generated drawings for supporting comprehension of a complex animation. Contemporary Educational Psychology, 38(3), 211–224. Mayer, R. E. (Ed.). (2014). The Cambridge handbook of multimedia learning (2nd ed.). New York: Cambridge University Press. Mayer, R. E., Steinhoff, K., Bower, G., & Mars, R. (1995). A generative theory of textbook design: using annotated illustrations to foster meaningful learning of science text. Educational Technology Research and Development, 43(1), 31–41. Mayer, R. E., Mathias, A., & Wetzell, K. (2002). Fostering understanding of multimedia messages through pretraining: evidence for a two-stage theory of mental model construction. Journal of Experimental Psychology: Applied, 8(3), 147–154. McCrudden, M. T., & Rapp, D. N. (2017). How visual displays affect cognitive processing. Educational Psychology Review, 29(3), 623–639. McNamara, D. (Ed.). (2012). Reading comprehension strategies: theories, interventions, and technologies. New York: Lawrence Erlbaum Associates. Nelson, T. O., & Narens, L. (1990). Metamemory: a theoretical framework and some new findings. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 26, pp. 125–173). New York, NY: Academic Press. Paivio, A. (1990). Mental representations: a dual coding approach. New York: Oxford University Press. Pilegard, C., & Fiorella, L. (2016). Helping students help themselves: generative learning strategies improve middle-school students’ self-regulation in a cognitive tutor. Computers in Human Behavior, 65, 121–126. Ploetzner, R., & Fillisch, B. (2017). Not the silver bullet: learner-generated drawings make it difficult to understand broader spatiotemporal structures in complex animations. Learning and Instruction, 47, 13–24. Rau, M. A. (2017). Conditions for the effectiveness of multiple visual representations in enhancing STEM learning. Educational Psychology Review, 29(4), 717–761. Roediger III, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249-255. Scheiter, K., Schleinschok, K., & Ainsworth, S. (2017). Why sketching may aid learning from science texts: contrasting sketching with written explanations. Topics in Cognitive Science, 9(4), 866–882. Schleinschok, K., Eitel, A., & Scheiter, K. (2017). Do drawing tasks improve monitoring and control during learning from text? Learning and Instruction. Schmeck, A., Mayer, R. E., Opfermann, M., Pfeiffer, V., & Leutner, D. (2014). Drawing pictures during learning from scientific text: testing the generative drawing effect and the prognostic drawing effect. Contemporary Educational Psychology, 39(4), 275–286. Schnotz, W. (2014). Integrated model of text and picture comprehension. In R. E. Mayer (Ed.), Cambridge handbook of multimedia learning, 2nd ed. (pp. 72-103). New York, NY: Cambridge University Press. Schwamborn, A., Mayer, R. E., Thillmann, H., Leopold, C., & Leutner, D. (2010). Drawing as a generative activity and drawing as a prognostic activity. Journal of Educational Psychology, 102(4), 872–879. Schwamborn, A., Thillmann, H., Opfermann, M., & Leutner, D. (2011). Cognitive load and instructionally supported learning with provided and learner-generated visualizations. Computers in Human Behavior, 27(1), 89–93. Schmidgall, S. P., Eitel, A., & Scheiter, K. (2018). Why do learners who draw perform well? Investigating the role of visualization, generation and externalization in learner-generated drawing. Learning and Instruction. Stieff, M., & Uttal, D. (2015). How much can spatial training improve STEM achievement? Educational Psychology Review, 27, 605–615. Thiede, K. W., Griffin, T. D., Wiley, J., & Anderson, M. C. M. (2010). Poor metacomprehension accuracy as a result of inappropriate cue use. Discourse Processes, 47(4), 331–362. Van Essen, G., & Hamaker, C. (1990). Using student-generated drawings to solve arithmetic word problems. Journal of Educational Research, 83(6), 301–312. Van Meter, P. (2001). Drawing construction as a strategy for learning from text. Journal of Educational Psychology, 93(1), 129–140. Van Meter, P., Aleksic, M., Schwartz, A., & Garner, J. (2006). Learner-generated drawing as a strategy for learning from content area text. Contemporary Educational Psychology, 31(2), 142–166. Van Meter, P., & Firetto, C. M. (2013). Cognitive model of drawing construction. In G. Schraw, M. T. McCrudden, & D. Robinson (Eds.), Learning through visual displays (pp. 247–280). Charlotte, NC: Information Age Publishing, Inc. Van Meter, P., & Garner, J. (2005). The promise and practice of learner-generated drawing: literature review and synthesis. Educational Psychology Review, 17(4), 285–325. Winne, P. H., & Hadwin, A. F. (1998). Studying as self-regulated learning. In D. J. Hacker, J. Dunlosky, & A. Graesser (Eds.), Metacognition in educational theory and practice (pp. 277–304). Wittrock, M. C. (1990). Generative processes of comprehension. Educational Psychologist, 24(4), 345–376.

Educ Psychol Rev Wu, S. P. W., & Rau, M. A. (2017). Effectiveness and efficiency of adding drawing prompts to an interactive educational technology when learning with visual representations. Learning and Instruction. Zhang, Z. H., & Linn, M. C. (2011). Can generating representations enhance learning with dynamic visualizations? Journal of Research in Science Teaching, 48(10), 1177–1198. Zhang, Z. H., & Linn, M. C. (2013). Learning from chemical visualizations: comparing generation and selection. International Journal of Science Education, 35(13), 2174–2197.