The common factor – inhibition by another name?

The “common factor” thought to underlie all examples of executive function is “about one's ability to actively maintain task goals and goal-related information and use this information to effectively bias lower-level processing” (Miyake & Friedman, Reference Miyake and Friedman2012, p. 11). Tasks that require delay of gratification exemplify the common factor. A child, for example, may be told that if she refrains for 30 seconds from eating a marshmallow that is right in front of her, she will later receive two marshmallows; or, if she does not touch an attractive toy for 30 seconds, she will be able to play with it at the end of that time. The child must keep in mind the fact that a larger reward will be hers if she can wait, and use that information to prevent herself from reaching out and taking the lesser reward. Anti-saccade tasks (Supplementary Material, Appendix 1, Section 1.), Stroop tasks (Supplementary Material, Appendix 1, Section 2.), and stop-signal tasks (Supplementary Material, Appendix 1, Section 3.), like delay of gratification, were first considered to be examples of tasks that tap inhibition; in the new analysis they are considered to tap the common factor (Miyake & Friedman, Reference Miyake and Friedman2012).

Since the correlation between inhibition and the common factor is almost perfect (Miyake & Friedman, Reference Miyake and Friedman2012), one could gloss the common factor as the ability to inhibit inappropriate responses (see e.g., Zacks & Hasher, Reference Zacks, Hasher, Dagenbach and Carr1994) and propose inhibition as the factor that is common to updating and shifting.Footnote ³ One advantage of explicitly labeling inhibition as the factor common to all examples of executive function is that it is somewhat more informative about the possible underlying mechanism than is the “ability to actively maintain task goals and goal-related information and use this information to effectively bias lower-level processing” (Miyake & Friedman, Reference Miyake and Friedman2012, p. 12). The more abstract definition, however, has the advantage of being general enough and flexible enough to be seen as a factor in most tasks. Carrying any task through to completion will require one to maintain task goals and to bias lower-level processing that might otherwise prevent the attainment of those goals.

Updating

To turn now to the specific components, updating is required in tasks, like the letter memory task, or n-back tasks in general, that require the participant to constantly refresh the material in working memory. (See Section 4 in Supplementary Material, Appendix 1, for a challenging version of an n-back task.) Updating might be glossed as “effective gating of information” or “controlled retrieval from long term memory” (Miyake & Friedman, Reference Miyake and Friedman2012); or as working memory (Hoffman, Schmeichel & Baddeley, Reference Hofmann, Schmeichel and Baddeley2012); or as a form of executive attention, “the ability to control attention to maintain information in an active, quickly retrievable state” (Engle, Reference Engle2002, p 20). Inhibition and updating show different neural patterns and different patterns with aging (Turner & Spreng, Reference Turner and Spreng2012).

Shifting

Shifting is construed as “flexibility – ease of transitioning to new task-set representations” (Miyake & Friedman, Reference Miyake and Friedman2012). A task that reflects shifting is a color-shape task in which participants shift from responding on the basis of color or shape. (See Section 5 in Supplementary Material, Appendix 1 for a color-shape task.) Other forms of task switching involve other dimensions, such as judging numbers either on the basis of whether they are odd or even or on the basis of whether they are greater or smaller than 5.

Unclassified tasks

Three other tasks – the Simon task (Simon & Wolf, Reference Simon and Wolf1963), the flanker task, and the ANT (Attention Network Task) – have been used frequently in bilingualism research. In the Simon task, the participant sees a red figure (e.g., a rectangle) and must press a key on the right side of the keyboard, or a blue figure and must press a key on the left side of the keyboard. Sometimes the placement of the color on the screen corresponds to the side of the keyboard where the participant must press a key (congruent trials) and sometimes the placement is on the other side (incongruent trials). See Section 6 in Supplementary Material, Appendix 1 for an on-line demonstration of a Simon task.

In the flanker task, a participant typically sees an array of 5 stimuli, commonly arrows. The participant is asked to respond based on the direction the central arrow is pointing. On some trials that arrow is flanked either by arrows pointing in the same direction (congruent, or no-conflict, trials) or arrows pointing in the opposite direction (incongruent, or conflict, trials). The participant's task is to press a key on the right side of the keyboard if the arrow is pointing right and a key on the left side of the keyboard if the arrow is pointing left. The flanker task can be made more complicated by including a go/no-go component. See Section 7 in Supplementary Material, Appendix 1 for an online version of the flanker task. A related task is the Attention Network Task, or ANT. The ANT involves more processes than the flanker task – alerting and orienting in addition to executive function. See Section 8 in Supplementary Material, Appendix 1 for an online version of the ANT.

Yet other tasks have been used to measure executive function. See Section 9 in Supplementary Material, Appendix 1 for short descriptions. The use of so many different tasks makes comparisons across experiments difficult. All the tasks measure slightly different aspects of executive function as well as different ancillary processes, so the implications of failures to achieve consistent benefits across tasks cannot be readily assessed. Nor, for that matter, can the implications of success in achieving consistent benefits.

İn most studies, experimenters subtract the mean reaction time on congruent (or control) trials from the reaction time on incongruent trials, in order to obtain a measure of the cost of incongruency. Or one subtracts the mean reaction times for trials where no switch is required from trials where a switch is required.Footnote ⁴

Task impurity

Tasks are “impure” (Burgess, Reference Burgess and Rabbitt1997; Friedman, Miyake, Young, DeFries, Corley & Hewitt, Reference Friedman, Miyake, Young, DeFries, Corley and Hewitt2008; Marton, Campanelli, Eichorn, Scheuer & Stepanoff, Reference Marton, Campanelli, Eichorn, Scheuer and Stepanoff2013, among others) or, more neutrally, tasks have multiple components. No task measures just one process. Even tasks that are superficially similar, such as the Simon and the flanker, require different cognitive, perceptual, and motor mechanisms. That is evident when one acts as a participant in various tasks.

Consider the Simon and flanker tasks from the point of view of their similarities and differences. Both tasks require the participant sometimes to use one finger and sometimes a different finger in responding. Both tasks have congruent and incongruent trials. The similarity stops there. The stimuli are different – rectangles vs arrows; one is non-directional and the other is inherently directional. More important, the incongruency in the flanker has a different source from the incongruency in the Simon. In the flanker, incongruency is due to a conflict between the direction of the target arrow in the focus of attention and the direction of the arrows in the periphery of attention. The congruent item is focal and the incongruent surround is peripheral (Guiney & Machado, Reference Guiney and Machado2013). The flanker requires one to ignore the arrows surrounding the target. In the Simon, incongruency is due to a lack of alignment between the spatial position of the stimulus and the spatial position of the key to be pressed. There is a single stimulus and it is always in the participant's focal attention whether it is congruent or incongruent. Another difference is that the Simon requires inhibition of a prepotent response whenever the stimulus is on the other side of the screen from the keyboard response, while the flanker does not (Poarch & Van Hell, Reference Poarch and Van Hell2012).

Although the task differences might seem minor, they have consequences. Average reaction time, independent of congruency condition, correlates well between the Simon and flanker tasks: people who are fast overall on the Simon task are fast overall on the flanker task. But the cost of incongruency does not correlate well between the two tasks. Individuals who show a low cost of incongruency on the Simon do not show a similarly low cost on the flanker (e.g., Humphrey & Valian, Reference Humphrey and Valian2012; Paap & Greenberg, Reference Paap and Greenberg2013; J. Van Hell, personal communication, 30 Dec 2012). Even the verbal and numerical versions of the Stroop do not correlate significantly with each other (Duñabeitia, Hernández, Antón, Macizo, Estévez, Fuentes & Carreiras, Reference Duñabeitia, Hernández, Antón, Macizo, Estévez, Fuentes and Carreiras2014). Since the tasks are conceptually extremely similar, it is likely that cognitive processes outside of executive function are responsible for the differences in responding. As a result, it is difficult to know whether differences, when they are found, are due to the aspects of a task that measure executive function or to aspects that measure other cognitive processes.

Now consider the ANT. It includes alerting and orienting components in addition to a conflict component (Supplementary Material, Appendix 1, Section 8.); the conflict component is similar to that in a flanker task. The alerting and orienting components are not specifically tapping executive functions. Active video game players, who show superior performance on many tests of visual performance and attention, perform better than non-game players on the alerting component. When tested on the ANT, participants in four age groups (7-10, 11-13, 14-17, 18-22) were always faster than non-gamers, but showed a greater cost of incongruency than did non-players (Dye, Green & Bavelier, Reference Dye, Green and Bavelier2009). This effect may be due to the greater spatial attentional span that active video game players have (Dye et al., Reference Dye, Green and Bavelier2009). Because players can take in more peripheral stimuli at one time than non-players can, they have an advantage in active video game play. But they have a disadvantage in disattending to the flanking stimuli; their wider attentional purview results in their seeing the conflicting arrows and being slowed down by the conflict. Whether or not that explanation (of the counter-intuitive handicap which video game players experience on incongruent conditions) is correct is less important than the possibility it suggests: cognitive processes that lead to superior performance in some domains can lead to inferior performance in other domains, even if all the domains in question tap one or another aspect of executive function.

In the flanker task in Appendix 1, the central arrow is sometimes smaller than the surrounding arrows, sometimes the same size, and sometimes larger. I found the condition with the smaller arrow easy to respond to, even in the incongruent trials, because there were small blank spaces on either side of the arrow, drawing my attention to the arrow and making it easier to ignore the flanking arrows. That type of visual effect is not itself an executive function process, but it interacts with an executive function process. We need fine-grained task analyses in order to understand what processes are being recruited and how they interact.

Summary of executive function

The brief review demonstrates how much there is still to understand about what the components of executive function are, how those components interact with other cognitive processes (such as alerting and orienting), and how to measure those components. Good performance on one executive function task does not entail good performance on other executive function tasks. Rather small changes in the format of an experiment may alter the patterns of results, and what seem at first blush like minor differences between tasks may instead be sufficiently major that performance on one does not correlate with performance on another.

I draw three implications from the discussion thus far. One is that it will be difficult to make clear predictions about the benefits of cognitively enriching activities, including bilingualism. A given experience may affect one task one way and another task another way. Further, not all ways of being bilingual, or of being musically trained, need be the same; someone who uses all their languages frequently, even if one of them is newly acquired, may differ from someone who uses their non-dominant languages infrequently, even if one of them was their mother tongue. The second implication is that small variations in task components can change whether participants do well or badly on a task, even if those components are orthogonal to executive function. The third is that we should expect inconsistencies from experiment to experiment not only because of our still fragmentary understanding of executive function and the tasks used to measure it, but also because no experiment can control for, or measure the effects of, every experience that improves or enriches high-level cognitive functioning.

Children

Studies of children with large samples have tended to show weaker effects, or no effects, compared to studies with small samples (but the size of the samples is somewhat diluted by the number of different ages and conditions.) For instance, three large studies comparing mono- and bilingual children have failed to find benefits of bilingualism (see Supplementary Material, Appendix 2, for details). In comparisons between monolingual Spanish-speaking and bilingual Spanish–Basque speaking children, there were no bilingual advantages on the ANT, the verbal Stroop, or the numerical Stroop (Antón, Duñabeitia, Estévez, Hernández, Castillo, Fuentes, Davidson & Carreiras, Reference Antón, Duñabeitia, Estévez, Hernández, Castillo, Fuentes, Davidson and Carreiras2014, total n = 360); Duñabeitia et al., Reference Duñabeitia, Hernández, Antón, Macizo, Estévez, Fuentes and Carreiras2014, total n = 504). Participants did show a cost of incongruency, and younger children showed a stronger cost than older children, so the tasks measured what was intended. But there were no differences in the incongruency effects as a function of language status. Both studies controlled for a range of possibly confounding variables. Further, the size of the incongruency effect in the verbal Stroop was not correlated with the size of the incongruency effect in the non-verbal Stroop. Although both tasks tap similar executive functions, for which the difference between words and numbers is irrelevant, it matters enough to change the pattern of responding.

The third large study compared monolingual English-speaking and bilingual Welsh–English speaking children, teenagers, young adults, and older adults from Wales and England (Gathercole, Thomas, Kennedy, Prys, Young, Viñas Guasch, Roberts, Hughes & Jones, Reference Gathercole, Thomas, Kennedy, Prys, Young, Guasch, Roberts, Hughes and Jones2014, total n = 650). On neither a card-sorting task nor a Simon task were there accuracy or reaction time advantages favoring bilinguals within any age group (except for accuracy among old adults); a few favored monolinguals. This study did not, however, control for possible confounding variables.

Among the experiments showing beneficial effects of bilingualism in children is a study comparing 7-month-olds who were exposed only to Italian with 7-month-olds exposed to both Italian and another language (usually Slovenian – Kovács & Mehler, Reference Kovács and Mehler2009). The infants learned to associate either an auditory or visual pattern with a visual reward. When the contingencies for the reward were switched, the infants exposed to two languages accommodated the switch more rapidly. Since the bilingual infants had accumulated only minimal exposure to their two languages and were not yet speaking, their superior executive function performance could not be due to experience in inhibition of one language or conflict monitoring of the two languages (Bialystok, Reference Bialystok2010). The challenge of listening to two different phonologies and prosodic patterns may be sufficient to improve executive function. One might then expect that 7-month-olds who heard both German and Dutch would not show a benefit, given the close overlap between the two languages.

A limited benefit of bilingualism has been reported for bilingual two-year-olds who are still on the cusp of language learning (Poulin-Dubois et al., Reference Poulin-Dubois, Blaye, Coutya and Bialystok2011). Of 5 tasks (see Supplementary Material, Appendix 2), bilinguals had a higher proportion of correct trials on a shape Stroop task, but not on delay or switch tasks. Since bilingual 7-month-olds show benefits on a switch task, it is not clear why the bilingual 2-year-olds in this study did not. Slightly older children, aged 3 in one group and 4½ in another, showed more consistent benefits of bilingualism (Bialystok, Barac, Blaye & Poulin-Dubois, Reference Bialystok, Barac, Blaye and Poulin–Dubois2010), perhaps suggesting that more language experience is needed before effects are visible. Out of four tasks (see Appendix 2 for descriptions), bilingual children performed better on all but the ANT.

An interesting comparison of 4-year-olds who were Korean–American bilinguals, Korean–American (English-speaking) monolinguals, Korean monolinguals, or non-Korean–American (English-speaking) monolinguals was designed to separate language effects from culture effects (Yang, Yang & Lust, Reference Yang, Yang and Lust2011). The bilingual group was more accurate and faster overall compared to the three monolingual groups on all conditions of a child version of the ANT, suggesting that language status was more important than culture, and also suggesting a general advantage to being bilingual. Visual inspection of the means suggests that the bilingual group showed a smaller incongruency effect with respect to accuracy, as one would predict, but a larger effect with respect to reaction time, contrary to what one might predict.

For slightly older children, ranging in age from 5 to 8, there are reaction time benefits of trilingualism on the incongruency effect of Simon task, compared to monolinguals, and benefits of both bi- and trilingualism on the incongruency effect of the ANT, compared to second language learners (Poarch & Van Hell, Reference Poarch and Van Hell2012; see Appendix 2 for fuller description). Here, the benefits are seen for reaction time but not accuracy.

While at least one study has suggested that, in children, apparent benefits of bilingualism are due to correlated SES differences (Morton & Harper, Reference Morton and Harper2007), another has found that SES and bilingualism independently contribute to more accurate (though not faster) flanker performance in children aged 6 and 7 (Calvo & Bialystok, Reference Calvo and Bialystok2014; see Appendix 2 for details). Only overall performance is reported, not performance on incongruent vs congruent trials. Since bilinguals showed superior performance across the board, they may be more accurate in general rather than have better executive function.

Summary and interpretation of data on children

For children, effects are mixed across and within studies. Some studies show no benefits of bilingualism and others show benefits that are localized to particular tasks, components of tasks, or particular measures (e.g., accuracy or reaction time, but not both). The multiplicity of different tasks makes it difficult to isolate the source of the differences (Paap, Reference Paap2014; Paap & Greenberg, Reference Paap and Greenberg2013).

A concentration on bilingualism, instead of the range of experiences that improve executive function, makes it difficult to know what participant features to control for. When no differences are observed between mono- and bilingual participants, we are in the dark about what properties the monolinguals have that might make them and bilinguals perform similarly. Monolingual and bilingual children in a particular study might get different amounts of exercise, or some other beneficial experience. In addition, the lack of a thorough-going analysis of tasks measuring executive function means that we do not know to what extent task properties that are independent of executive function are overwhelming task properties that more directly reflect executive function. Finally, the overall benefits of bilingualism reported in some experiments may signify a global benefit rather than one limited to executive function.

Young adults

Especially when young adults are tested, effects have been inconsistent, with a number of reports of failure to find differences between mono- and bilinguals (see, e.g., the reviews previously mentioned in the beginning of the bilingualism section; studies cited in Paap, Reference Paap2014). Two studies with large numbers of participants failed to find any benefits of bilingualism. One found no bilingual benefits for the Simon or flanker (Humphrey & Valian, Reference Humphrey and Valian2012); the other found no benefits for the Simon, the flanker, a color-shape switching task, or an anti-saccade task (Paap & Greenberg, Reference Paap and Greenberg2013). In neither study did the size of the incongruency effect correlate between the Simon and the flanker.

Earlier studies have varied, sometimes finding no or limited benefits on the Simon, flanker, and ANT, and sometimes finding strong benefits. In one modified version of the Simon, using arrows instead of rectangles (see Supplementary Material, Appendix 2 for details), young bilingual adults averaging age 20 showed no advantage (Bialystok, Craik & Luk, Reference Bialystok, Craik and Luk2008). If anything, the bilinguals showed a larger incongruency effect (22 ms) than the monolinguals did (8 ms). (In contrast, old bilingual adults did show an advantage over monolinguals, though that was because they showed no incongruency effect at all.) Catalan–Spanish bilinguals have shown an advantage on the flanker, but only in cases where high-monitoring was required (Costa, Hernández, Costa-Faidella & Sebastián-Gallés, Reference Costa, Hernández, Costa-Faidella and Sebastián-Gallés2009).

Stronger results with the ANT are apparent in a study comparing 20-year-olds who were early- or late-learning Spanish-English bilinguals. Both types of bilinguals outperformed monolinguals on the ANT (Pelham & Abrams, Reference Pelham and Abrams2014), suggesting that the benefits of bilingualism, when found, are not restricted to those with a great deal of experience in their second language. Similarly, a comparison of Chinese students (average age 20, range from 18–48) who had learned English either early or late also found benefits for both types of bilinguals on a version of the ANT (Tao, Marzecová, Taft, Asanowicz & Wodniecka, Reference Tao, Marzecová, Taft, Asanowicz and Wodniecka2011).

Another study compared second language learning students who were immersed in a second language learning environment with classroom learners; the classroom learners showed more, rather than less, of a benefit on the Simon, or were similar (Linck, Hoshino & Kroll, Reference Linck, Hoshino and Kroll2008). Similarly, less proficient bilinguals, as measured by reaction times to recognize translation equivalents, showed more, rather than less, of a benefit on the Simon or were similar (Linck et al., Reference Linck, Hoshino and Kroll2008). Thus, learning a second language, or being proficient in two languages, conferred a benefit in executive function compared to bilinguals, but degree of immersion or proficiency was irrelevant.

One study has found that early, but not late, bilingual college students (average age = 21) showed an advantage in the flanker task compared to monolinguals (Luk, De Sa & Bialystok, Reference Luk, De Sa and Bialystok2011). Age of bilingualism was operationalized as the time when individuals began using both languages on a daily basis; that was age 5 for early bilinguals and age 16 for late bilinguals. There was also a weak (r = 0.24) but significant correlation between age of active bilingualism and size of the flanker effect: the later the age at which participants began using both their languages actively, the greater the cost of conflict. On a measure intended to be similar, only bilinguals who rated themselves as 4 or 5 (on a 5-point scale) in speaking and listening in both childhood and adulthood were compared with monolinguals (Humphrey & Valian, Reference Humphrey and Valian2012); there were no advantages for bilinguals so defined on either the Simon or the flanker task.

If frequent use of both one's languages is an important variable, then interpreters should show a greater benefit on a task like the Simon than garden-variety bilinguals. Spanish-speaking monolinguals, Spanish–English-speaking bilinguals, and Spanish–English-speaking interpreters were compared on the Simon, but all three groups performed similarly (Yudes, Macizo & Bajo, Reference Yudes, Macizo and Bajo2011). Thus, in general, it is difficult to find advantages in executive function for being a highly proficient bilingual; the simple challenge of dealing with two languages at any level appears to be cognitively enriching enough in the cases where advantages for bilingualism are found.

Overall, then, benefits of bilingualism are sparse on the Simon, flanker, and ANT tasks. Depending on one's task analysis, the Simon, flanker, and ANT may not be the best tests of bilingual benefits. On one analysis, the Simon primarily involves inhibition (Yudes et al., Reference Yudes, Macizo and Bajo2011), while on others it primarily involves interference (e.g., Bialystok et al., Reference Bialystok, Craik and Luk2008). Absent a fine-grained analysis of tasks, it is not possible to determine whether one or another aspect of executive function is primary. On the face of it, the Simon, flanker, and ANT involve all three executive functions – inhibition, updating, and shifting. On a Stroop task, which can be interpreted as either an inhibition task (e.g., Heidlmayr, Moutier, Hemforth, Tanzmeister & Isel, Reference Heidlmayr, Moutier, Hemforth, Tanzmeister and Isel2014) or an interference task (Bialystok et al., Reference Bialystok, Craik and Luk2008), there are bilingual advantages (Bialystok et al., Reference Bialystok, Craik and Luk2008; Heidlmayr et al., Reference Heidlmayr, Moutier, Hemforth, Tanzmeister and Isel2014). Go/no-go tasks, in which participants must refrain from responding on some trials, might appear to be the best example of an inhibition task. In one experiment, participants had to press the space bar for any digit except the digit 3, in which case they were not to respond. Error scores – responding on no-go trials – were equivalent for mono- and bilinguals in both younger and older age groups (Bialystok et al., Reference Bialystok, Craik and Luk2008).

Switching tasks like card-sorting and the color-shape task, almost by definition, highlight shifting, even if they also involve inhibition and updating. Shifting, or switching, might be seen as similar to what some bilinguals do as a matter of course – shifting from one language to another depending on contextual demands. If the conversational context is uniformly in one language, primarily inhibition of the other language(s) might be required, but in a translation environment, both shifting and inhibition would be required. Thus, one might predict that shifting or switching advantages would be reliably found in individuals who use both their languages regularly.

On a card sorting task, interpreters generally out-performed both mono- and bilinguals, who did not differ between each other (Yudes et al., Reference Yudes, Macizo and Bajo2011). (All three groups were equivalent in the number of categories they succeeded at, but interpreters were more efficient in arriving at those categories, making fewer attempts and fewer errors.) Thus, even though interpreters were equivalent to mono- and bilinguals in the Simon task, they outperformed both mono- and bilinguals in card-sorting. This study is like others in not showing differences between mono- and bilinguals among young adults.

Bilingual advantages have been reported in another shifting task, the color-shape switching task (Prior & MacWhinney, Reference Prior and MacWhinney2010; see Supplementary Material, Appendix 2). When trial sets were blocked – only color or shape was required – monolinguals and bilinguals responded with equivalent reaction times. In the mixed condition, where color and shape were each sometimes required, depending on the trial, (or the reverse), monolinguals were slower than bilinguals (Prior & MacWhinney, Reference Prior and MacWhinney2010). But a switching advantage for bilinguals is not generally found (e.g., Paap & Greenberg, Reference Paap and Greenberg2013), even among bilinguals who frequently use two related Romance languages, Spanish and Catalan (Hernández, Costa, Fuentes, Vivas & Sebastián-Gallés, Reference Hernández, Costa, Fuentes, Vivas and Sebastián-Gallés2010; Hernández, Martin, Barceló & Costa, Reference Hernández, Martin, Barceló and Costa2013). One might expect that managing two similar languages would be more difficult than managing two very different languages, because of the phonological, lexical, and grammatical interference between similar languages. But across three experiments using a version of a color-shape task, there were no advantages in switching among bilinguals (though they were faster overall; Hernández et al., Reference Hernández, Martin, Barceló and Costa2013).

Further, bilinguals’ performance on language switching tasks does not always parallel their performance on cognitive switching tasks (e.g., Calabria, Hernández, Branzi & Costa, Reference Calabria, Hernández, Branzi and Costa2011; Calabria, Branzi, Marne, Hernández & Costa, in press; Weissberger, Wierenga, Bondi & Gollan, Reference Weissberger, Wierenga, Bondi and Gollan2012). That is, bilinguals are better – make fewer mistakes – when switching between languages than when switching on cognitive tasks, a point discussed at greater length in the section on older adults.

Summary and interpretation of data on young adults

Especially among young adults, it is common to find no differences between mono- and bilinguals, but positive results also exist. My interpretation is that young adults have accumulated, and are in the midst of accumulating, many cognitively enriching experiences. The participants in most experiments with young adults are college students. We hope that they are being cognitively challenged. Although we do not know the full range of experiences that will improve executive function, the teenage and young adulthood years are periods of experimentation. In the welter of new experiences that require planning, updating, shifting, and inhibiting, it would not be surprising if positive effects of bilingualism would be washed out. Beneficial challenging experiences may include examples that have not yet been systematically studied. For example, an absorbing interest in American football (sometimes referred to as a coach's game because of the many strategies involved), ballet, acting, or cooking may be cognitively demanding and lead to the development of a range of skills.

My hypothesis is slightly different from claims that young adults are operating at peak processing capacity or are extremely efficient processors (e.g., Kroll & Bialystok, Reference Kroll and Bialystok2013). Yes, young adults have faster reaction times than either children or older adults, and reaction times begin their inexorable slowing around age 20 (Fozard, Vercruyssen, Reynolds, Hancock & Quilter, Reference Fozard, Vercruyssen, Reynolds, Hancock and Quilter1994). But, on my analysis, young adult monolinguals who perform on a par with bilinguals succeed not primarily because they are efficient processors, but because they have other experiences that are on a par with bilingualism in their ability to enrich cognitive functioning. A similar line of reasoning can be extended to children. They spend more time sleeping and may (arguably) also spend more time in unchallenging activities than do young adults, but they spend much of their waking time actively learning.

Older adults and executive function

It is with old and very old adults that the benefits of bilingualism for executive function seem clearest (though that may partly be a function of less research with this population). For example, one study found that middle-aged bilinguals between 30 and 60, as well as old bilinguals between 60 and 90, had a smaller incongruency effect in the Simon task than did age-matched monolinguals (Bialystok, Craik, Klein & Viswanathan, Reference Bialystok, Craik, Klein and Viswanathan2004). In some cases, reaction times differed by a factor of 2 or more. That finding is similar to another report that old bilinguals showed an advantage on a conflict version of the Simon and on a Stroop task (though not on a go/no-go task or a reverse Simon task, Bialystok et al., Reference Bialystok, Craik and Luk2008).

In an examination of performance on a color-shape task, a comparison of middle-aged adults (average age, 32) and old adults (average age, 64) in two experiments found that old bilingual adults outperformed their monolingual peers in both experiments, while middle-aged bilingual adults did not show an advantage (Gold et al., Reference Gold, Kim, Johnson, Kryscio and Smith2013). Recall, however, that older Welsh–English bilinguals did not show an advantage on either card-sorting or the Simon task (Gathercole et al., Reference Gathercole, Thomas, Kennedy, Prys, Young, Guasch, Roberts, Hughes and Jones2014; see also Kousaie & Phillips, Reference Kousaie and Phillips2012, and Kirk, Scott-Brown & Kempe, Reference Kirk, Scott-Brown and Kempe2013). A study of old Spanish-speaking immigrants found that bilingual status, when measured objectively, was associated with better executive function even though it did not reduce the likelihood of dementia (Zahodne, Schofield, Farrell, Stern & Manly, Reference Zahodne, Schofield, Farrell, Stern and Manly2014).

Bilinguals’ language system appears to be partially independent of their executive function system (e.g., Calabria et al., in press; Calabria et al., Reference Calabria, Hernández, Branzi and Costa2011; Magezi, Khateb, Mouthon, Spierer & Annoni, Reference Magezi, Khateb, Mouthon, Spierer and Annoni2012). Several studies have compared bilinguals’ ability to switch between their two languages and their ability to shift between color and shape. Aging affects language switching and color-shape switching differently among highly proficient Spanish-Catalan bilinguals. The cost of language switching does not appear to increase with age but the cost of color-shape switching does (Calabria et al., in press). In addition, for neither young nor old participants was there a correlation between the cost of language switching and the cost of color-shape switching. Both findings suggest that bilinguals’ control over their languages is not mediated solely by executive function processes and that executive function is influenced by many different factors. Even when an increasing cost of switching languages with age is reported, there remains a dissociation with executive function; some elderly participants could not perform the color-shape task at all even though they could switch languages easily (Weissberger et al., Reference Weissberger, Wierenga, Bondi and Gollan2012).

A dissociation between language use and executive function should not be surprising. Neither executive function nor language processing are monoliths. But the dissociation carries an interesting implication, namely that not all of the processes involved in managing multiple languages transfer to executive component processes. Skill in managing two languages does not completely transfer to the domain of executive functions. To some extent, systems are modular, even while they have more general effects.

Summary and interpretation of experimental data on older adults

The data overall, although mixed, suggest a benefit in older adults for being bilingual on tasks of executive function. Old adults, in contrast to young adults and children, might be expected to show more benefit from bilingualism because they tend to spend less of their time in active and challenging cognitive activities compared to young adults and children. In the absence of other stimulation, the benefits of bilingualism might be clearer.

Older adults and dementia

Although a natural hypothesis is that pursuit of any challenging cognitive activity will result in healthier aging, that is not a given. Dementia affects executive function but also cognitive systems like language and memory that are not executive functions. Thus, bilingualism might or might not have benefits for aging. The effects of bilingualism in slowing the onset of cognitive decline and dementia have been explored in a number of studies, starting with Bialystok, Craik, and Freedman (Reference Bialystok, Craik and Freedman2007), who reported that bilinguals experienced dementia more than four years later than monolinguals.

Subsequent research has shown variation from study to study (Freedman, Alladi, Chertkow, Bialystok, Craik, Phillips, Duggirala, Raju & Bak, Reference Freedman, Alladi, Chertkow, Bialystok, Craik, Phillips, Duggirala, Raju and Bak2014). Some studies show benefits after equating participants on a range of variables (e.g., Alladi, Bak, Duggirala, Surampudi, Shailaja, Shukla, Chaudhuri & Kaul, Reference Alladi, Bak, Duggirala, Surampudi, Shailaja, Shukla, Chaudhuri and Kaul2013; Craik et al., Reference Craik, Bialystok and Freedman2010), while others do not (e.g., Zahodne et al., Reference Zahodne, Schofield, Farrell, Stern and Manly2014).

Retrospective vs prospective studies

Studies that find a protective effect of bilingualism tend to be retrospective while studies that do not find a benefit tend to be prospective (Yeung, St. John, Menec & Tyas, in press; Zahodne et al., Reference Zahodne, Schofield, Farrell, Stern and Manly2014). In an ideal world, the results from retrospective and prospective studies coincide. Unfortunately, the data on bilingualism are not ideal. In retrospective studies, individuals are consecutively tabulated according to the date at which they were diagnosed with a mild or severe cognitive impairment (often via the Mini-Mental State Examination, or MMSE) upon presentation at a memory clinic. Retrospective studies exhibit what could be called the complement class problem. Many individuals with cognitive difficulties never appear at a memory clinic and their characteristics are unknown; similarly, individuals without cognitive difficulties seldom appear at a memory clinic. Those individuals comprise the complement class. We only know the size and composition of the class that has visited the clinic, a class that may or may not be representative of the general population. Even if both mono- and bilinguals have comparable scores on the MMSE when they first appear in a clinic, and even if they decline at similar rates (Bialystok et al., Reference Bialystok, Craik and Freedman2007), the complement class could be different and thus change our conclusions about the total class.

In prospective studies, in contrast, investigators study a large sample of a community and track their health over a period of years, generally between 4 and 10 years, noting changes in physical health, mental health, and cognition. Here, the whole class is examined, allowing an estimate of the incidence of a phenomenon. Although not everyone agrees to be a participant in such a study, and not everyone continues to participate, demographic comparisons suggest that those who enter and those who do not are similar. In prospective studies, individuals are usually not demented at the time of first being recruited or, if they are already demented, they are not included in the sample. Prospective studies use objective tests at each point along the way and thereby avoid the possibility of group-based differential sensitivity to symptoms of impairment.

Findings from retrospective studies

Retrospective studies tend to find that bilinguals experience dementia 4–7 years later than monolinguals, although the studies differ in their details (Alladi et al., Reference Alladi, Bak, Duggirala, Surampudi, Shailaja, Shukla, Chaudhuri and Kaul2013; Bialystok et al., Reference Bialystok, Craik and Freedman2007; Chertkow, Whitehead, Phillips, Wolfson, Atheron & Bergman, Reference Chertkow, Whitehead, Phillips, Wolfson, Atherton and Bergman2010; Craik et al., Reference Craik, Bialystok and Freedman2010; see review in Freedman et al., Reference Freedman, Alladi, Chertkow, Bialystok, Craik, Phillips, Duggirala, Raju and Bak2014), with some studies finding benefits when controlling for immigration status (Craik et al., Reference Craik, Bialystok and Freedman2010) and others not (Chertkow et al., Reference Chertkow, Whitehead, Phillips, Wolfson, Atherton and Bergman2010). For example, a study in Toronto, Canada, examined consecutive records from a two-year period of individuals who came to a memory clinic and received a diagnosis of probable Alzheimers disease (Craik, Bialystok & Freedman, Reference Craik, Bialystok and Freedman2010; see Supplementary Material, Appendix 2 for details). The average age at which families reported initial symptoms of cognitive decline was 72.6 for monolinguals and 77.7 for bilinguals. The average age at the first clinic visit was 76.5 for monolinguals and 80.8 for bilinguals. Both differences were significant, and were maintained after controls for covariates were introduced.

A later study comparing mono- and bilinguals in Toronto, Canada, with respect to both mild cognitive impairment and dementia, confirmed earlier findings that bilinguals presented with dementia at a later age than did monolinguals (in this study, about 7 years later), and also demonstrated a later onset of cognitive decline than monolinguals (about 4 years later; Bialystok, Craik, Binns, Ossher & Freedman, Reference Bialystok, Craik, Binns, Ossher and Freedman2014; see Appendix 2 for details), even when controlling for immigration status and other factors.

In contrast, a study in Montreal, Canada, found that nonimmigrant monolinguals showed a later decline than nonimmigrant bilinguals; only nonimmigrants who were at least tetralingual showed a protective effect (Chertkow et al., Reference Chertkow, Whitehead, Phillips, Wolfson, Atherton and Bergman2010; see Appendix 2 for details). For immigrants the findings were different. Immigrants had an overall earlier onset of cognitive impairment than nonimmigrants, (except for tetralingual nonimmigrants). Bilinguals, trilinguals, and tetralinguals were impaired significantly later than monolinguals. Because of the large number of variables in this study, and the differences in findings among nonimmigrants and immigrants, it is difficult to arrive at an overall interpretation. A retrospective study in Hyderabad, India, where immigration is not an issue, compared bi- and monolinguals (Alladi et al., Reference Alladi, Bak, Duggirala, Surampudi, Shailaja, Shukla, Chaudhuri and Kaul2013). Bilingualism was protective, even when covariates like education were controlled for.

One exception to the general finding with retrospective studies compared Spanish–English bilinguals in the United States. It investigated whether degree of bilingualism – measured as the extent to which participants were equal in their two languages on a naming task – was associated with delay of cognitive decline. It was, but only for low-education individuals; for high-education individuals there was no effect (Gollan, Salmon, Montoya & Galasko, Reference Gollan, Salmon, Montoya and Galasko2011).

Summary and interpretation of data on cognitive decline and dementia in retrospective studies

Although there is variation from study to study, the retrospective studies show later decline for bilingual compared to monolingual individuals. In some cases knowing two languages is enough, in others it is necessary to know more than two. In one case, an advantage was found only for individuals with lower education, in others an advantage exists when education is controlled for. In one case, immigration status made a difference, but in others it does not. On the whole, the retrospective studies confirm a protective role for using more than one language.

Findings from prospective studies

Prospective studies, in contrast, tend to find that bilinguals and monolinguals show similar trajectories with respect to cognitive decline and dementia (Crane, Gibbons, Arani, Nguyen, Rhoads, McCurry, Launer, Masaki & White, Reference Crane, Gibbons, Arani, Nguyen, Rhoads, McCurry, Launer, Masaki and White2009; Crane, Gruhl, Erosheva, Gibbons, McCurry, Rhoads, Nguyen, Masaki & White, Reference Crane, Gruhl, Erosheva, Gibbons, McCurry, Rhoads, Nguyen, Arani, Masaki and White2010; Sanders, Hall, Katz & Lipton, Reference Sanders, Hall, Katz and Lipton2012; Yeung et al, in press; Zahodne et al., Reference Zahodne, Schofield, Farrell, Stern and Manly2014). For example, Crane et al (Reference Crane, Gruhl, Erosheva, Gibbons, McCurry, Rhoads, Nguyen, Arani, Masaki and White2010; see Supplementary Material, Appendix 2 for details) studied second generation Japanese–American men in Honolulu. Participants were divided into three groups: the monolingual group said they spoke or read no or very little Japanese; the middle group – 59% of the sample – said they spoke but did not read or write Japanese; the bilingual group said they both spoke and read Japanese. Neither the mono- nor bilingual group differed from the middle group in their rate of cognitive decline.

A community-based prospective study of Hispanics in New York City similarly showed no difference in cognitive decline between Spanish–English bilinguals and Spanish monolinguals ranging in age from 64 to 95 (Zahodne et al., Reference Zahodne, Schofield, Farrell, Stern and Manly2014). All the participants had been born outside the US and were assessed periodically over a period of up to 22 years, averaging 6.5 years. Language status was not associated with dementia once covariates for age at enrollment in the study, education, and sex of participant were included. For the subset of participants for whom objective English assessment scores were available, proficiency in English was also not associated with cognitive decline.

A notable exception to the failure to find a benefit of bilingualism in prospective studies is a study of a random national sample of Israelis who were tested three times over a 12-year period (Kavé, Eyal, Shorek & Cohen-Mansfield, Reference Kavé, Eyal, Shorek and Cohen-Mansfield2008; see Supplementary Material, Appendix 2 for details). The participants were bi-, tri-, or at least tetralingual and had an average age of 83 in the first of three waves of data gathering. (The participants’ advanced age in the first wave may be a factor in the findings.) All participants spoke at least Hebrew and at least one other language. The number of languages individuals spoke was an independent predictor of scores on a cognitive screening test at all three waves, and accounted for a significant increase in R² after age, gender, country of origin, education, and age of immigration to Israel had been entered as independent variables. The best scores were obtained by individuals speaking four or more languages (similar to Chertkow et al.'s Reference Chertkow, Whitehead, Phillips, Wolfson, Atherton and Bergman2010 findings). Educational level was a strong predictor, but even among individuals with more than 12 years of education, there was a dose response: the more languages the better. Thus, education and language status had independent effects.

Summary and interpretation of data on cognitive decline and dementia in prospective studies

Overall, community-based samples that follow individuals over time do not show advantages of bilingualism. Although there may be an issue regarding the extent of bilingualism of participants labeled bilingual (Freedman et al., Reference Freedman, Alladi, Chertkow, Bialystok, Craik, Phillips, Duggirala, Raju and Bak2014, note that Zahodne et al., Reference Zahodne, Schofield, Farrell, Stern and Manly2014, did not have data on the age at which individuals became bilingual, and that Sanders et al., Reference Sanders, Hall, Katz and Lipton2012, may have inadvertently included some bilinguals among their monolinguals), that seems less important than the methodological difference between retro- and prospective studies (Zahodne et al., Reference Zahodne, Schofield, Farrell, Stern and Manly2014).

Summary and interpretation of data on old adults

With respect to measures of executive function, there is reason to think that elderly bilinguals show better performance than monolinguals. That conclusion is tentative, because there are fewer studies of executive function in the elderly compared to children or young adults. With respect to conversion to dementia, the safest conclusion thus far is that “bilingualism alone is insufficient to guarantee the postponement of dementia” (Freedman et al., Reference Freedman, Alladi, Chertkow, Bialystok, Craik, Phillips, Duggirala, Raju and Bak2014). The fact that most retrospective studies do show a benefit, while most prospective studies do not, is an inconsistency that needs to be resolved.

Summary and interpretation of education data

Education has a protective effect on cognitive decline and correlates with executive function.

Summary and interpretation of exercise data

Some individuals are more likely than others to engage in exercise. Level of education, for example, correlates with exercise. And the likelihood of maintaining an exercise program once initiated may vary according to motivation, education, or cognition. Nevertheless, when individuals are randomly assigned to exercise groups, those engaging in mindful, stimulating exercise appear to perform better on tasks measuring executive function. A striking feature of the interventions is that it is not necessary to have life-long experience in order to show executive function benefits (as noted by Li et al., Reference Li, Legault and Litcofsky2014). Another striking feature is that the benefits extend to domains other than those on which the enriching experience focused. Aerobic exercise appears overall to improve executive function in older adults (Guiney & Machado, Reference Guiney and Machado2013) and is correlated with later cognitive decline.

Summary and interpretation of music training data

Music training is associated with superior performance on executive function tasks and with later cognitive decline, though there has not been much research in this area.

Summary and interpretation active video game playing data

Active video game players show somewhat inconsistent benefits on executive function tasks and can even show deficits (e.g., Bailey, West & Anderson, Reference Bailey, West and Anderson2010, show a deficit for active video game players compared to nonplayers in one condition of a modified version of a Stroop task). Active video game experience also appears to have different effects on different tasks. For switching tasks there are benefits, but for tasks requiring lack of attention to peripheral distractors there are costs. There is no evidence concerning active video game experience and onset of cognitive decline.

Summary and interpretation of SES data

SES is probably only a proxy for the relevant challenging activities that lead to better executive function.

Summary and interpretation of miscellaneous factors

Leisure and social activities may provide challenges sufficient to delay cognitive impairment, but may also sometimes be a consequence of better cognition. Having an agreeable and conscientious personality may contribute independently to healthy aging or may be a consequence of other variables.