The nature and importance of student learning styles are among the most written about and least agreed upon issues in the educational literature. Broadly speaking, learning styles refer to students’ individual preferences for particular educational environments and techniques for learning new information. Scholarly attention to the potential role of learning styles in education began in earnest in the 1970s, but the concept is rooted in much earlier research on cognitive styles (Cassidy, 2004). A long history of research on cognitive styles demonstrates that people do in fact tend to think in different ways. For example, people who are field-dependent prefer to analyze information as part of a larger context, whereas those who are field-independent prefer a more objective analysis of information independent of the surrounding context (Willingham, 2009). Some people prefer to think mainly in concrete terms, while others prefer abstract concepts (see Kozhevnikov, 2007, for a review of cognitive style models). Claiming that such differences in thinking styles do not exist would be akin to claiming that extraversion does not exist.
Students will often report a preference for one type of thinking or another. Debate about learning styles pertains to a separate but related claim made by many educators – that instruction tailored to match students’ learning preferences leads to more successful learning regardless of the nature of the material to be learned (e.g., Gregorc & Ward, 1977; Dunn, 2000; Zapalska & Dabb, 2002). This idea is very widely endorsed among educators. Nearly four decades ago, Arter and Jenkins (1977) reported that 99% of the teachers they surveyed agreed that “A child’s modality strengths and weaknesses should be a major consideration when devising educational prescriptions,” and 96% believed that their students learned more when teachers matched their teaching approach to students’ modality preferences (p. 290). Recent data suggest that such assumptions have changed little over time. In a survey of primary and secondary school teachers, 94% endorsed the belief that learning is improved when students are taught in a manner consistent with their learning style (Dekker, Lee, Howard-Jones, & Jolles, 2012). Learning style matching is also widely endorsed in higher education and among parents (Pashler, McDaniel, Rohrer, & Bjork, 2009), and college students likewise tend to view their own perceived learning styles as important (Krätzig & Arbuthnott, 2006). Popular websites (e.g., “Overview of learning styles,” n.d.) and even the websites of university learning centers (e.g., “Three learning styles,” n.d.) assert that people learn in different ways and that matching learning styles with teaching methods improves learning.
The literature on learning styles is extensive, complex, and fragmented. There is no uniformly accepted definition of what a learning style is, nor is there a universally accepted model of specific learning styles. One useful definition that helps to illustrate the broad concept of learning styles is that they refer to “the way people absorb, process, and retain information” (De Bello, 1990: 204), although many other more complex definitions have been offered (see Hyman & Rosoff, 1984; Cassidy, 2004, for reviews). What is perhaps more important is the remarkable proliferation of learning-style models that scholars have devised over the past several decades. In an important review, Coffield, Moseley, Hall, and Ecclestone (2004) identified 71 different learning-style models. Whereas the established idea that people differ in their cognitive styles originated in psychological research, much of the literature on learning styles has been produced outside the field of psychology – specifically in such fields as education and business – which led to a poorly integrated field of research with “complexities and convolutions difficult to comprehend and assimilate” (Cassidy, 2004: 419). Kozhevnikov (2007) blames this lack of coherence on researchers’ shift toward using self-report measures to assess conscious learning-style preferences, replacing an emphasis on assessments of abilities. Often there is little attention to whether such perceptions of one’s own learning are accurate or important.
Given the proliferation of learning-style models, it is not possible to provide a detailed description of even a meaningful sample of such models here. Several comprehensive reviews are available for readers wishing to better understand the specifics of various models (Cassidy, 2004; Coffield et al., 2004). Instead, this chapter will focus primarily on the evidence for the broad hypothesis that learning is reliably enhanced when teachers attempt to match their style of teaching – irrespective of content – to students’ self-reported learning styles. Despite the popularity and widespread endorsement of the learning-styles approach to instruction, many researchers have noted that there are few methodologically sound studies that provide support for its efficacy (Coffield et al., 2004; Pashler et al., 2009). Unlike most fields of research, more than half the available scholarly literature on learning styles comes from doctoral dissertations. Although dissertations often include well-conducted research, they are student projects that are held to a lower threshold of quality than is found in published, peer-reviewed research.
The most broadly researched and applied model of learning styles addresses students’ preferences for learning through specific sensory modalities – usually visual, auditory, and kinesthetic (Arter & Jenkins, 1977; Stahl, 1999). In their review of several early studies on student learning preferences, Arter and Jenkins first outlined the type of evidence required to conclude that matching instruction to learning styles actually enhances learning. Such evidence, they explain, must come from experimental rather than correlational research, and must meet several specific methodological criteria. First, the method of categorizing students by learning style must have demonstrated reliability and validity. Second, there must be evidence that students classified as particular types of learners perform differently on corresponding ability measures; for example, students with an auditory learning style must perform significantly better on auditory tests than on visual tests. This criterion helps to demonstrate that learning-style instruments are assessing something other than simple preferences. Third, some students must be taught using methods matched to their preferred style, while others are taught with methods that are contrary to their preferred style. Finally, students in all conditions must be assessed with the same outcome test after instruction. Furthermore, to ensure that the experimental findings have utility, the participants must be representative of classroom populations and the teaching objectives and methods must be consistent with actual educational environments.
Perhaps not surprisingly, Arter and Jenkins (1977) found relatively few studies that met all their criteria. They identified 14 such studies assessing the effects of matching teaching styles to auditory or visual learning styles when teaching reading to schoolchildren. The studies they reviewed were diverse with respect to the age of student participants, the duration and method of instruction, and the type of outcome test used to evaluate learning. Only one of the studies yielded a significant result supporting the matching approach, and Arter and Jenkins identified several methodological limitations that limited the validity of this single study. Tarver and Dawson (1978) conducted a similar review, and likewise found virtually no evidence that matching instruction to learning style improves learning. They concluded that the lack of observed benefit across a variety of instructional and assessment methods supports the conclusion that students’ preferences for specific modalities cannot be matched with teaching methods to produce better outcomes.
Kampwirth and Bates (1980) conducted an even more extensive review of 22 studies of elementary school children in which researchers compared learning outcomes when students with preferences for either visual or auditory learning learned via either a matched method or an unmatched method. In only two of the studies was there an effect of matching consistent with learning-styles predictions. In the other 20 studies, there was either no consistent effect of matching or, in some cases, the observed effect was the opposite of what learning-styles models would predict. That is, students in some studies actually learned more when they were taught in their less-preferred modality. Kampwirth and Bates also pointed out that many researchers first screen participants to identify those with a distinct preference for learning via one modality or the other, and exclude from their studies participants with less distinct preferences. This practice would tend to exaggerate the effects of matching, while simultaneously making the results less applicable to real-life educational settings where all students are included.
In 1987, Kavale and Forness touched off a particularly interesting and contentious professional debate about learning styles. These researchers sought to conduct a more rigorous integration of learning-styles research than had previously been conducted. Using a statistical technique called meta-analysis – a method for combining the results of many existing studies – the researchers analyzed data from more than 3,000 elementary and secondary students collected during 39 studies in which learning styles were assessed, teaching materials and techniques were designed to match those styles, and an outcome test was administered. Only 16 of the studies came from published articles; the rest came mostly from dissertations, with a few coming from books. Kavale and Forness found that 13 of the 39 studies revealed some small positive effect of matching. Interestingly, when the researchers classified the 39 studies based on methodological quality using an established procedure, they found that studies of poor quality showed the largest – but still very small – effects of matching, while studies of high quality showed extremely small effects not statistically different from no effect at all. They also reported that more than one-third of students receiving matched instruction performed more poorly on outcome tests than control students who did not receive customized instruction. Kavale and Forness concluded that there was no support for the hypothesis that matching improves learning.
Kavale and Forness’s (1987) analysis and conclusions were met with derision by one of the staunchest advocates of matching instruction to students’ learning styles. Dunn (1990) was highly critical of Kavale and Forness for including studies in their meta-analysis that Dunn claimed were methodologically flawed. Ironically, Kavale and Forness attempted to account for methodological quality and found that the studies that were most flawed were those that provided the strongest support for matching. Dunn cited 10 studies that she claimed provided support for matching instruction to learning style, but 9 of the 10 studies came from unpublished dissertations – mostly from her own institution. As noted above, most dissertation research has the distinct limitation of lacking rigorous peer review.
A few years later, Dunn and her colleagues (Dunn, Griggs, Olson, Beasley, & Gorman, 1995) published their own meta-analysis of 36 studies conducted on Dunn’s own model of learning styles. The researchers concluded that there was an overall average positive effect whereby students performed better after instruction that was consistent with their learning style than with instruction that was inconsistent. Again, however, the quality of the data included in the analysis is indeterminate. Of the 36 studies, 35 came from unpublished dissertations. Dunn and colleagues also eliminated several studies that they claimed had serious methodological flaws, but provided few details on the nature of either the included or excluded studies. It is particularly difficult to evaluate the validity of the researchers’ conclusions when nearly all the data came from unpublished sources. Furthermore, it is difficult to comprehend why research seemingly demonstrating the validity of such an important educational principle has largely failed to appear in peer-reviewed sources. The evidence supporting other learning-style models likewise suffers from the limitation of consisting nearly entirely of unpublished projects not subject to peer review (Stahl, 1999).
A few years after the publication of Dunn and colleagues’ (1995) analysis, Kavale and colleagues (Kavale, Hirshoren, & Forness, 1998) provided a critique. They concluded that Dunn et al. could not have conducted a comprehensive review of research on the Dunn learning-styles model, as evidenced by the fact that nearly all included studies were dissertations and that many relevant databases were not searched. The critics also pointed out that Dunn and colleagues reported average effects of matching, but did not report variability around those averages as is customary in meta-analyses and is necessary for interpreting the findings. Kavale and colleagues cite some evidence from Dunn et al.’s report suggesting that the omitted information could potentially invalidate the findings. They also criticized Dunn for including studies that were likely to be statistical outliers, as evidenced by effect sizes so large as to be unrealistic given the nature of the research. Kavale and colleagues bluntly concluded that the meta-analysis had “all the hallmarks of a desperate attempt to rescue a failed model of learning style” (p. 79).
Harsh critiques aside, recent research evaluating the benefits of matching instruction to students’ preferred learning modality have continued to yield little positive evidence. For example, Massa and Mayer (2006) conducted two experiments comparing visually-oriented learners with verbally-oriented learners in terms of learning from a multimedia lesson that emphasized either verbal or visual presentations of content. In one experiment involving college students and a replication involving non-college-educated adults, the researchers used self-report measures to categorize participants as either visualizers or verbalizers. Participants were then randomly assigned to study a computerized lesson in electronics under training conditions emphasizing either verbal or visual learning. The researchers observed no interaction between student learning preference and the type of instruction in determining performance on any of four composite learning measures. The researchers then analyzed effects on separate components of the composite tests. Among 51 specific outcome measures, significant matching effects occurred in only two cases, which is less than the number of positive effects that would be expected due to chance. Moreover, nearly half of the effects were in the opposite direction to what would be predicted by the matching hypothesis. Massa and Mayer concluded that they had found no support for the claim that learners with visual and verbal orientations will benefit from different methods of instruction.
Other recent research further calls into question the premises of the matching hypothesis with respect to various learning modalities. For example, Constantinidou and Baker (2002) had adults ranging in age from 19 to 77 complete a learning-styles instrument to assign scores for visual and verbal learning preferences. Participants attempted to learn lists of objects presented under one of three conditions: an auditory condition in which participants heard the words out loud, a visual condition in which participants saw drawings of the objects, and a combined condition in which participants saw a drawing and heard the name of the object at the same time. Constantinidou and Baker found no correlation between preferences for a visual learning style and performance under any of the three methods of instruction. Interestingly, preferences for auditory learning did not correlate with performance on the auditory memory task as would be predicted by the matching hypothesis, but did correlate with performance on the visual memory task – contradicting the matching hypothesis.
In a similar study, Krätzig and Arbuthnott (2006) had university students complete an established inventory to assign scores for visual and auditory leaning preferences, as well as kinesthetic preferences. In addition, participants separately reported which of the three learning styles they believed best fitted them. Finally, they completed measures of visual, auditory, and kinesthetic memory. There were no significant positive correlations between scores on the learning style inventory and objective measures of memory associated with the corresponding modality. That is, a preference for visual learning was not associated with better visual memory, nor were auditory and kinesthetic preferences associated with performance on relevant memory tests. The same pattern of findings emerged when researchers used participants’ self-categorizations of learning style. Importantly, only 29 of the 65 participants were classified the same way by the learning-styles inventory and their own self-classification. The researchers repeated their analysis using only the data for those participants who were categorized the same way using both assessments, and still there was no support for the learning-styles hypothesis. Krätzig and Arbuthnott state that, contrary to the predictions associated with learning-style models, they found no evidence that visual, auditory, or kinesthetic learners learn better in their preferred modality.
In the most recent test of the assumptions associated with the matching hypothesis, Rogowsky, Calhoun, and Tallal (2014) had 121 college-educated adults complete an established leaning-styles survey assessing preference for auditory or visual learning, as well as actual aptitude measures assessing auditory comprehension and visual comprehension. The 61 participants who could be definitively categorized as having a specific learning style were randomly assigned to learn content from a nonfiction book by either reading an ebook or listening to an audiobook. They then answered questions assessing their comprehension of the content immediately after exposure and again two weeks later. Across all participants, preference for an auditory learning style was not associated with better performance on a listening comprehension test than on a reading comprehension test; similarly, a preference for visual learning was not associated with better performance on a reading comprehension test than it was on a listening comprehension test. Moreover, self-reported visual learners scored higher than auditory learners on both listening and reading aptitude tests. With respect to the modes of instruction implemented among participants with the most distinctive learning preferences, there was no significant interaction indicating that participants with a particular learning style learned more – in terms of either immediate performance or longer-term comprehension – from a particular instructional method. Rogowsky and colleagues found that only the general comprehension aptitude measures – and not the learning style measures – were associated with how much participants learned.
As noted at the beginning of this chapter and illustrated by the research described thus far, learning styles associated with specific sensory modalities – visual and auditory and to a lesser extent kinesthetic – have been the most widely studied preferences, and the existing research yields little support for matching instruction to these preferences. Research on other learning-style models leads to very similar conclusions. For example, Lundstrom and Martin (1986) tested the matching hypothesis based on Gregorc’s (Gregorc & Ward, 1977) model in which learners are categorized into one of four learning styles based on their preferences for concreteness versus abstraction and sequentialness versus randomness. Based on Gregorc’s model, the researchers predicted that students with certain styles would perform better studying independently and students with other styles would learn more as part of a group. The researchers had college students complete a learning-styles instrument as well as achievement tests and measures of attitudes toward the styles of instruction. Students then participated either in an instructional method involving individual study or one involving interactions between groups of students. There was no interaction between any of the learning styles and either of the instructional methods in affecting either student achievement or attitude toward instruction.
Bostrom (1990) reviewed four studies investigating learning-style matching for purposes of software training. The studies all tested the efficacy of matching on a learning-styles model proposed by Kolb (1984), whereby learners are categorized as having one of four possible styles depending on their preference for concrete versus abstract experiences, and for active versus reflective learning activities. Of the four studies, only the single published study yielded significant results. Nonetheless, Bostrom interpreted all four studies as having a “consistent pattern of findings” supporting the importance of these learning styles (p. 101). This is a curious conclusion given that even for the single published study, only one out of three tested effects was significant according to accepted standards and all effects were extremely small.
Hayes and Allison (1993) reviewed 17 studies investigating the effects of matching based on several different learning-styles models. They reported that 10 of the 17 studies provided some support for the effectiveness of matching, but that the findings were inconsistent. Often the findings supported matching for one particular learning style within a model and not for other styles; some studies revealed support for matching when learning was measured in one way, but not when it was measured in another. Moreover, positive effects tended to be very small in magnitude – suggesting that matching was associated only with very weak effects or no effects at all.
In one of the most recent tests of the effects of matching instruction to learning styles, researchers investigated yet another model (Cook, Thompson, Thomas, & Thomas, 2009). Cook and colleagues used a learning-style measure to assess medical students’ preference for a sensing learning style emphasizing applied learning such as data collection and experimentation, versus an intuitive learning style emphasizing broader patterns and theories. The researchers predicted that participants with a sensing learning style would prefer and learn more from an instructional approach where applied problems were presented in advance of didactic information, and that intuiting learners would prefer and learn more when didactic information was presented in advance of applied problems. Medical residents each completed web-based learning modules containing both didactic content and medical case problems to be solved; all modules contained identical content, but varied in terms of whether the didactic information or case problem was presented first. Participants completed two modules from each format presented in randomized order; they completed a test of applied knowledge after each module and a cumulative final exam at the end of the academic year. Cook and colleagues found no significant effect where sensing or intuitive learners learned more from instructional methods designed to match their learning preferences. There was also no evidence that learning styles predicted faster learning when a matched instructional method was used. The researchers conducted additional analyses of other learning-style dimensions assessed by the instrument they used, and found no evidence that any of the dimensions were associated with improved learning either independently or in conjunction with particular teaching methods.
Researchers have proposed many explanations for why matching teaching methods to learning styles does not have the benefits that most teachers assume. From a measurement perspective, many tests used to identify students’ learning styles have poor reliability which may indicate that students’ preferences are not necessarily stable across time (Stahl, 1999). Stahl also argues that inventories purportedly assessing learning styles often are actually assessing abilities. For example, some students classified as auditory learners might prefer auditory presentations simply because they have poor reading skills, so emphasizing auditory teaching methods could deprive them of opportunities to improve their reading comprehension skills. Accordingly, Stahl recommends tailoring teaching methods to students’ developmental skill level rather than to their supposed learning styles.
Many researchers have also criticized advocates of learning-style models for failing to consider the nature of the content being taught (Snider, 1992; Pashler et al., 2009). Hyman and Rosoff (1984) cited this failure as a major weakness of the learning-styles concept, stating that effective teaching requires attention to the subject matter being learned. For example, learning to read a map requires a visual representation rather than an auditory explanation only (Willingham, 2009). Snider puzzled over the question of why people agree that one cannot learn to play basketball solely through discussion, but believe that students can learn to read solely through visual or kinesthetic methods. Reading, she insists, requires some skills that are neither kinesthetic nor visual. Cook and colleagues (2009) agree that teaching methods should be tailored to learning objectives rather than to students’ learning styles.
Arter and Jenkins (1979) proposed that learning-style models simply may not describe students’ characteristics accurately, the assessed characteristics may not have a meaningful effect on learning, or the effect may be weak compared with the effect of other factors. Accordingly, a very large number of studies indicate that there are many techniques that improve student learning more powerfully and consistently than matching teaching to learning styles (see Walberg, 1984; Kavale et al., 1998). Kavale and Forness (1987) explained that learning-style assessment itself is problematic because when people are assessed and categorized in this way, there is a great deal of overlap across groups in their actual preferences. In other words, most people’s preferences do not follow a clean categorical model where one style is preferred strongly and consistently above all others. Kavale and Forness state that for students with less differentiated preferences – which is most of them – learning preferences make very little difference in comparison with many other factors that affect learning.
Pashler and colleagues (2009) question the very notion that students differ greatly in the ways in which they learn. They assert that although many factors may affect the type of teaching that is most effective for individual students, the assumption that there are vast differences between students in terms of the way they learn can distract from the application of empirically supported principles that can improve learning across the board. Willingham (2009) agrees stating, “Children are more alike than different in terms of how they think and learn” (p. 113). He goes on to explain that although people differ in memory ability associated with specific senses, most memories are stored based on meaning rather than the specific sensory mode through which the information was absorbed. Therefore, having strong auditory memory or strong visual memory does not help when the objective is to learn meaning, because the meaning is not stored based on sense-specific information such as visual and auditory details.
Other researchers have pointed out additional ways that learning-style models neglect important details about how the brain works. Geake (2008) emphasized that that “focusing on one sensory modality flies in the face of the brain’s natural interconnectivity” (p. 130). In general, neither learning nor teaching practices can be cleanly differentiated according to learning-style assumptions, so it is likely all modalities are important for learning (Kavale and Forness, 1987; Arter & Jenkins, 1979). Accordingly, teaching that integrates multiple modalities simultaneously tends to produce greater learning gains across all students than teaching that is tailored to specific modalities (Massa & Mayer, 2006; Tight, 2010). Finally, researchers have questioned the utility of any learning model that neglects or denies – as learning-style models do – the role of intelligence (Hyman & Rosoff, 1984).
Clearly, there is a disconnect between what most teachers assume about students’ learning styles and the experimental evidence supporting such assumptions – a disconnect that Pashler and colleagues (2009) characterized as “striking and disturbing” (p. 117). Researchers have provided a number of explanations for this disconnect. Pashler and colleagues noted that learning-style models are appealing because they classify people into neat categories and focus on treating students as individuals. The models are also consistent with the idea that everyone can learn very effectively if only the right individualized teaching method is used, and justify blaming the educational system rather than any lack of ability if a child is not succeeding academically. Willingham (2009) suggests that the vast majority of teachers believe in the importance of learning styles because the concept is so widely accepted that it seems that it must be true, and because confirmation bias causes people to interpret ambiguous data as confirming their expectations. Other researchers have elaborated by pointing out that teachers do not usually systematically collect data comparing the effectiveness of different teaching methods so they may tend to remember times when matching seemed to work, interpret what they observe primarily in terms of what they expect to see, or misattribute students’ progress to learning-style matching when many other factors could have been at play (Arter & Jenkins, 1977).
Those who criticize the practice of matching instruction to learning styles assert neither that all students are the same nor that all teaching methods will be equally effective across all students and all situations. However, given the pressure on teachers to assess learning styles and develop lesson plans to match those styles (Rogowsky et al., 2014), as well as the commercial interests helping to drive such an agenda (see Pashler et al., 2009), it is important to determine whether such efforts are in the best interest of students. Arter and Jenkins (1977) adeptly pointed out that if matching does not yield benefits under controlled research conditions, it is even less likely to be beneficial in classrooms. Pashler and colleagues noted that even if a study provided the necessary experimental evidence, it would support only a specific type of classification rather than learning-style models in general. Furthermore, they argue, the benefits of implementation would need to be evaluated in the context of the increased cost – in terms of teacher training and other factors – of assessing styles and customizing teaching methods. Others fear that students may be hindered by being told that they have one learning style rather than the ability to learn in a variety of ways (Henry, 2007). It would certainly appear to be good news that, as researchers have found, students can learn via many modes and can shift to learning strategies other than those they most prefer when the situation or content demand it (Krätzig & Arbuthnott, 2006; Constantinidou & Baker, 2009). Arter’s and Jenkins’ admonition regarding learning-style matching made nearly 40 years ago is no less apropos today: “no matter how strongly a given model of learning may appeal to conventional wisdom, that model’s validity and utility is still an empirical question” (p. 282).