Executive Functions: What are they Good for? A Perspective from Intervention Research

Papers on the topic of executive functions (EFs) have appeared steadily over the past three decades and can be found across many literatures ranging from cognitive neuroscience and clinical psychology to the developmental and education sciences (Müller & Kerns, 2015). The study of EF across these various fields reflects the fact that almost every adult and child neurological, psychiatric, learning, and behavioral condition is associated with disturbances in EF. This association with many disorders, including dyslexia and other learning disabilities (LDs), presents a scientific challenge: namely, EFs lack specificity in their predictive validity for the identification and assessment of LDs. On the other hand, if EFs are implicated in many aspects of social–emotional, cognitive, and academic performance and are so vulnerable to perturbations in neurobehavioral development, perhaps it is informative to ask whether EFs are important for understanding learning and response to intervention in children with LDs. Because the relations between EF and academic achievement are robust in correlational studies, and because there are many interventions that are effective for improving academic achievement in students with LDs, examining the relation of EF and academic achievement in the context of interventions seems ideal. In this paper, I discuss the utility of EFs in the context of interventions, using work conducted by our research group and with colleagues as examples.

I do not adopt a particular theoretical framework for EFs in terms of whether EF is a “thing” or a “process,” how many there are (I use EF and EFs interchangeably), or how they should be measured. For a discussion of these topics, see Cirino (2023) and Müller and Kerns (2015) for a developmental perspective. The approach I take is that EF is a characteristic of both persons and interventions. That is, I assume that individuals have relatively less or more capacity in certain situations (consciously or not, Evans & Stanovich, 2013) to attend to and process information relevant to the task at hand, control interference from external (environmental) or internal (one's own thoughts) sources, and to update memory with task-relevant information. In terms of intervention, I assume that instructional materials, procedures, techniques, and the measures used to assess learning vary in the EF load they impose on the learner. One theory of instruction that takes such interactions of the learner and the learning task into account is cognitive load theory (Sweller, 2010). Instruction is hypothesized to be most effective if it: (1) takes the learner's capacities (i.e., prior knowledge and working memory, hereafter, WM) into account when designing instruction; (2) enhances germane load, which results in learning; and (3) reduces extraneous load, which interferes with learning. I will refer to this theory as one lens through which intervention findings related to EF in children with LDs are interpreted.

Two classes of studies are relevant for discussing EF in the context of interventions. One looks at whether EFs moderate the effects of interventions for children with LDs; that is, whether person characteristics such as EFs interact with treatment variables (e.g., the materials and procedures that are part of that intervention). These studies ask for whom and under what conditions the intervention has beneficial effects. Although customizing interventions for children with LDs is an important goal of Special Education research and practice, moderation analyses have only recently become more commonplace (Fuchs & Fuchs, 2019). The second class of studies is those that test the nature of the relation of EFs and academic skill learning, asking whether EF interventions have any value for academic skill learning and conversely, whether academic skill interventions have any value for increasing EF. Practically, such studies speak to the increasing commercialization and popularization of EF interventions for children with LDs and other neurodevelopmental disorders. Scientifically, such studies speak to theories about the direction and nature of the relationship between EF and academic skills.

EFS AS MODERATORS OF INTERVENTION EFFECTS IN VOCABULARY LEARNING AND MATHEMATICS

Two studies are presented in this section. The first takes a micro-genetic approach, asking whether a common technique in vocabulary interventions—the use of non-examples, is similarly beneficial to learners across a range of comprehension and WM levels (Barnes, Davis, Kulesz, & Francis, 2021). The second study is a systematic review of mathematics interventions for children with math learning disabilities (MLD) that asks whether a child's level of EFs moderates treatment response (Barnes, Martinez-Lincoln, Miller, & Agrawal, 2023). The studies with human subjects that are discussed in various sections of this paper were approved by Institutional Review Boards at the University of Texas Health Science Center at Houston and Vanderbilt University. Parental informed consent and child assent were obtained in accordance with IRB standards at each institution.

THE INTERACTION OF VOCABULARY LEARNING TECHNIQUES AND READER CHARACTERISTICS

Non-examples are a frequently used technique in vocabulary learning interventions (Gentner & Namy, 2004; Swanson, Wanzek, Vaughn, Roberts, & Fall, 2015; Williams & Vaughn, 2020). The technique draws on a principle called contextual interference (Battig, 1979) in which an example context (e.g., for the word, conclave, the sentence The football team held a conclave in their locker room to discuss the game) is presented during learning along with a non-example context (e.g., The football team held a conclave on national television to discuss the game). This example/non-example pairing takes advantage of an additional learning technique called structural mapping or alignment (Gentner, Levine, Dhillon, & Poltermann, 2009) in which the non-example differs from the example in only one critical feature; that is, the “meeting” definition of conclave is true for both sentences, but the fact that a conclave is “secret” is violated in the non-example.

Contextual interference and structural alignment set up information processing conditions that are thought to be generally advantageous for long-term retention and are an example of what Bjork (e.g., Bjork & Bjork, 2020) calls desirable difficulties in learning: the deployment of significant cognitive resources during learning leads to better long-term retention than easier learning situations that result in superior short-term performance at the expense of long-term retention. In terms of the cognitive load theory, the use of a structurally-aligned non-example requires considerable comparative processing that requires WM resources (Bjork & Kroll, 2015). For some learners, this type of processing may be germane for learning, but for other learners (i.e., those with less knowledge and WM resources), the use of non-examples may pose an extraneous load on processing that hinders learning.

In Barnes et al. (2021), we compared the learning of new words (e.g., conclave) in two conditions—contextual reinforcement learning where both sentences presented during learning provided correct examples of the word (e.g., The football team held a conclave in their locker room to discuss the game paired with The robbers held a conclave deep in the woods to plot their next robbery), and contextual interference learning as in the earlier example using conclave where one sentence is a correct example, and the other a structurally aligned non-example. In this single-session instructional experiment, we tested the immediate recall of the meaning of a new word right after having heard its definition along with two example sentences or an example and a non-example sentence. Once all words had been exposed in this way, students had to choose the correct word when provided with definitions (see study for a third delayed test). Whether learning was better in the contextual reinforcement or contextual interference condition and whether these effects were moderated by student characteristics including reading comprehension level and WM were tested.

Recall of word meanings immediately after learning each word was better in the contextual reinforcement condition, suggesting better immediate recall in the lower load condition; this effect was not moderated by student characteristics. In contrast, memory for word meanings after all words had been exposed, was better in the contextual interference condition, but only for students who started higher in comprehension. For these students, the findings are consistent with the hypothesis that initial difficulties or challenges in learning are associated with better longer-term retention. The WM level did not interact with any of these effects although it did predict performance on the delayed multiple-choice definitions task.

There are three take-aways from this study. First, instructional techniques such as the common use of non-examples during vocabulary learning, which draw heavily on EFs such as WM, may be more advantageous for students starting higher in language/reading comprehension than for students with lower skills—the very students these interventions are meant to help. Indeed, neither of the learning conditions was particularly beneficial for students with lower reading comprehension. Learning techniques that result in germane load for some students, then, might produce extraneous load during learning for their lower performing peers. Second, the tasks used to assess learning may vary in the degree to which they draw on EFs—here, WM predicted performance on the delayed multiple-choice definition task, but not on the immediate definitions task. Finally, generally effective learning techniques such as contextual interference, structural alignment, and interleaved practice, that have largely been developed and tested with college students and typically developing school-age students (Dunlosky, Rawson, Marsh, Nathan, & Willingham, 2013), may not be helpful for all learners, or perhaps not at all points in the learning process. Micro-genetic studies that experimentally test the conditions under which these generally high-yield techniques for learning are most beneficial for children who have long histories of learning difficulties and associated difficulties in EF provides one direction for future research. Some have suggested that the rich corpus of research on cognitive learning principles and techniques (e.g., Dunlosky et al., 2013) be applied in a more methodical way to intervention design for individuals with learning disabilities (Barnes, Clemens, & Miller, 2022; Jordan, Barbieri, Dyson, & Devlin, 2020). In the present context, the application of these techniques that serve to link EFs required during encoding to long-term retention may also need to consider the EF capacity and prior knowledge of the learner, consistent with cognitive load theory (Sweller, 2010).

BIDIRECTIONAL EFFECTS OF EF AND MATHEMATICS INTERVENTIONS

The moderator studies reviewed in the preceding section suggest that EFs, particularly WM, affect learning of new vocabulary and mathematics. Moderator effects are correlational analyses within experimental designs, and although important for advancing knowledge about what works for whom under what conditions, they do not provide evidence that EFs and academic skills learning are causally related. However, moderator effects can lead one to ask such causal questions as a next step; that is, if EF could be improved would academic learning also increase? Several meta-analyses have now provided a resoundingly consistent “No”: training EFs such as WM does not result in transfer effects to reading or mathematics (e.g., Jacob & Parkinson, 2015; Melby-Lervåg, Redick, & Hulme, 2016; Sala & Gobet, 2020; Shipstead, Redick, & Engle, 2012). But, the story does not end here.

First, these meta-analyses ask whether training WM results in far transfer to mathematics; however, we already know what improves math learning—math interventions. A different question is whether integrating EF with or within math interventions provides synergy for math learning over and above the effects of a generally effective mathematics intervention (e.g., Barnes et al., 2016). Second, looking for far transfer effects from EF training to academics assumes that the developmental relationship of EF and academics is unidirectional. Because the development of any skill—EF or academic—is always based on one's experience with physical, social, and emotional environments (Doebel, 2020), it is just as valid to propose that engaging in academic learning leads to changes in EF or that EF and academic skills learning are bidirectional and mutually beneficial (e.g., Peng & Kievit, 2020). Third, most of the participants in the studies in these meta-analyses are typically developing children and most often arithmetic is the far transfer task. Given that relations of EFs and mathematics are relatively large for math word problem solving (Peng et al., 2016; Spiegel et al., 2021), and for children with math learning disabilities (Peng, Wang, & Namkung, 2018), it is of interest to ask what the effects of EF training might be for children with learning disabilities, and in academic domains that are most strongly related to EF. The study discussed below addresses all these issues.

Fuchs and her colleagues (Fuchs et al., 2022) conducted a word problem solving intervention with second graders with or at risk for MLD (in the Peng et al., 2016, meta-analysis the effect size for WM and math word problem solving for second graders was r = .45). There were three conditions in this RCT: (1) word problem intervention where each 30-min lesson included 6 min of arithmetic and word problem games to develop fluency; (2) the same word problem intervention with 6 min of arithmetic and word problems games designed to develop WM; and (3) a WM intervention (CogMed) lasting 25 min followed by 5 min of practice with arithmetic and word problems and simple corrective feedback. WM and calculation and word problem solving (WPS) outcomes were assessed at pretest, posttest, and WPS was also assessed at delayed posttest. Each research question is presented, followed by its findings and implications.

Given that previous WM training studies have not tested WPS and do not focus on children with MLD, might general WM training result in far transfer to WPS for these children? Yes, there were significant effects of WM training on WPS (and on calculations), but these effects were small compared to those for the WPS interventions. Cognitive training on its own does not provide a solution to improving mathematics in children with MLD. The findings also suggest that to obtain far transfer effects of EF training, there may need to be explicit links provided between cognitive training and academic tasks; however, this is uncertain because WM training without the subsequent 5 min of math practice was not tested.

Given that effect sizes for the relation of WM and WPS do not tell us about the direction of those effects, might WPS intervention result in far transfer to WM? Yes, the WPS intervention without any WM training resulted in effects on WM (also see DeFlorio et al., 2019), although these effects were smaller than they were for the intensive WM training condition. The findings from the first two questions are consistent with bidirectional models in which development of EF and academic skills influence each other (Peng & Kievit, 2020) and with situated developmental models of EFs (Doebel, 2020).

Does providing WM training embedded in the WPS intervention result in synergistic effects on mathematics? No, not in this study. One might argue that the 6-min WM training component of the WPS + embedded WM training condition was not intensive enough to show effects. The few other studies that have employed embedded or math-specific WM training at higher intensities are not informative here; either they lacked a necessary comparison group (i.e., there was no math intervention comparison group without WM training in Kroesbergen, van't Noordende, & Kolkman, 2014) or they failed to show improvements in WM in addition to mathematics (e.g., Muñez et al., 2022).

Given that some comparisons showed effects of interventions on WPS (and calculations), what is the mechanism by which improvement occurs? Longitudinal mediation analyses showed that improvements in WPS in the WPS intervention condition, the WPS + embedded WM condition, and the general WM training condition were mediated by WM growth from pretest to immediate posttest. In contrast, improvements in calculations in these conditions were not mediated by WM. Such findings bring us closer to understanding the mechanisms by which intervention effects may be occurring; in this example, positive effects on math word problem solving seems to accrue from both instruction and practice with WPS as well as from WM regardless of whether WM improves through direct training or through experience in solving word problems. In contrast, improvement in calculations accrued only from instruction or practice with calculations.

What are the implications of this study for EFs in the context of interventions? First, in order to test hypotheses about the nature of the relationship of EF to academic skills, it is necessary to conduct experimental studies, such as this one, that allow for causal inferences about the direction and mechanisms of effects. What is clear from this study is that the relationship of some EFs to some aspects of mathematics learning is bidirectional, but also that EF interventions cannot take the place of academic interventions if improvement in academic skills is the primary desired outcome. The study also provides some clues about the conditions under which the testing of potential causal relationships between EF and academic skills might be most productive. These include: (1) focusing on populations such as children with LDs and domains of academic functioning such as math word problems for which there is evidence of larger associations with EFs based on meta-analyses; (2) using analytic strategies such as longitudinal mediation to test hypotheses regarding potential mechanisms of effect with sample sizes that provide sufficient power to do so; (3) using reliable measures of both EF and academic outcomes, perhaps using a latent variable approach as was employed in this study; and (4) routinely adding measures of EF to intervention studies, not only at pretest for moderator analyses, but also at posttest, as a strategy for determining which of our academic interventions lead to growth in EF.

SUMMARY AND FUTURE DIRECTIONS

In this paper, I posed the question—What are EFs good for? To answer it, I presented studies in which EFs are viewed through the lens of academic interventions for individuals with LDs. Two studies of moderation effects were used to support the idea that considering EFs to be characteristics of both persons and interventions can inform us not only about for whom an intervention is more or less effective, but can also provide valuable information about the cognitive (i.e., EF) load of our interventions. When an intervention is less effective for children with lower levels of EF, altering the EF characteristics of the intervention that may be producing extraneous load during learning would seem to be in order. Similarly, we might ask whether, for individuals with higher levels of incoming EF, increasing the cognitive load of an intervention might be most effective for them, in line with the idea of desirable difficulties in learning (Bjork & Bjork, 2020). I suggested that newer experimental designs may be one way to speed up the acquisition of knowledge important for engineering interventions to fit the needs of a broader range of learners. Such an enterprise is consistent with the spirit of Chronbach's (1957) call for the “joint application of experimental and correlational methods” (p. 679) such that “The greatest social benefit will come from applied psychology if we can find for each individual the treatment to which he can most easily adapt” (p. 679).

The studies of moderation effects provide one type of answer to the question, What are EFs good for? Experimental studies that train EF and measure the effects on academic skills and those that train academic skills and measure the effects on EF are needed to answer this question from a causal perspective. The third study demonstrated bidirectionality in the relation of EFs and academic achievement and further showed that one of the mechanisms of change in math word problem solving was growth in working memory. More studies of this type are needed in order to answer causal questions about the relation of EF and academic achievement. Academic intervention studies without a distinct EF training component also have the potential to contribute to knowledge about causal effects if EFs were not only measured as potential moderators of intervention effects, but also as potential outcomes.

I close with a few thoughts about the emerging experimental research on EF and academic achievement. Recently published intervention studies attempt to conjointly address EFs and academic achievement (e.g., Fuchs et al., 2022; Muñez et al., 2022) and other studies are underway (e.g., Our Mathematical World project, Hornburg, Purpura and colleagues; our Attention Integrated with Mathematics-Kindergarten or Aim-K project). In surveying this work, it is noteworthy that different approaches to EF intervention are being adopted across studies. In some studies, an embedded approach is taken where the EF and academic interventions are integrated. In other studies, a complementary approach is used where the EF and academic interventions are separate, even if delivered in the same session. In some studies, materials and procedures used in the EF intervention are domain-general, and in others they are domain-specific (i.e., specific to the reading or math content of the academic intervention). Will EF interventions that are integrated/embedded into academic interventions provide synergistic effects for academic outcomes that are superior to complementary approaches? Will EF interventions that are domain-specific (in either embedded or complementary approaches) be superior to those that are domain-general? These are empirical questions, the answers to which await completion of ongoing and future studies that allow for inferences about mechanisms of effects. Such studies may help us to design more effective interventions for individuals with LDs and are likely to contribute more broadly to theories about the relationship of EF and academic achievement.

CONFLICT OF INTEREST STATEMENT

The author has no conflicts of interest to declare.