Evidence that ageing yields improvements as well as declines across attention and executive functions

Abstract

Many but not all cognitive abilities decline during ageing. Some even improve due to lifelong experience. The critical capacities of attention and executive functions have been widely posited to decline. However, these capacities are composed of multiple components, so multifaceted ageing outcomes might be expected. Indeed, prior findings suggest that whereas certain attention/executive functions clearly decline, others do not, with hints that some might even improve. We tested ageing effects on the alerting, orienting and executive (inhibitory) networks posited by Posner and Petersen’s influential theory of attention, in a cross-sectional study of a large sample (N = 702) of participants aged 58–98. Linear and nonlinear analyses revealed that whereas the efficiency of the alerting network decreased with age, orienting and executive inhibitory efficiency increased, at least until the mid-to-late 70s. Sensitivity analyses indicated that the patterns were robust. The results suggest variability in age-related changes across attention/executive functions, with some declining while others improve.

Full text

Articles
https://doi.org/10.1038/s41562-021-01169-7
1Center of Linguistics, School of Arts and Humanities, University of Lisbon, Lisbon, Portugal. 2Department of Linguistics, University of Potsdam, Potsdam,
Germany. 3School of Psychology, Georgia Institute of Technology, Atlanta, GA, USA. 4Office of Population Research, Princeton University, Princeton,
NJ, USA. 5Center for Population and Health, Georgetown University, Washington, DC, USA. 6Department of Neuroscience, Georgetown University,
Washington, DC, USA. ✉e-mail: jlverissimo@edu.ulisboa.pt; michael@georgetown.edu
Life expectancy has risen substantially, increasing the impor-
tance of understanding age-related changes in cognition. Not
surprisingly, ageing negatively affects many aspects of cogni-
tion, as demonstrated by a large body of work indicating that older
adults show worse performance than younger adults in a variety of
cognitive tasks. For example, older adults consistently display lower
accuracy and/or slower performance in tasks measuring episodic
memory1–3, word recognition and retrieval4,5, word learning6,7,
implicit learning of complex skills and sequences8–10, and various
visuo-spatial abilities11,12. Though some of these age differences may
be due to age-related decreases in overall processing speed or other
general factors13–15, reductions in many cognitive functions are still
detectable even when controlling for such variables16–19 and thus
seem to represent specific neurocognitive declines1,20. (For conve-
nience, we use terms such as ‘decline’ and ‘improvement’ to refer
to age-related differences in both longitudinal and cross-sectional
studies, the latter of which constitute the vast majority of work on
this topic; however, these terms should be treated with caution,
since alternative explanations for age effects cannot be completely
ruled out in cross-sectional studies21,22; see Discussion.)
The trajectories of age-related declines vary considerably,
depending on the particular cognitive function in question. Some
functions decrease robustly across the whole adult lifespan, while
others show smaller age-related declines or show decreases that
begin or become more pronounced later in life3,22–28. Such differen-
tiation can even be detected within particular cognitive domains1,29.
This variability raises the possibility that not all aspects of cognition
decline. Indeed, evidence suggests that some functions may be fully
preserved over the course of ageing, such as automatic processes of
memory retrieval30–32, aspects of lexical and grammatical processing
in language comprehension33–37, and skills that were well practised
throughout life38,39.
Furthermore, evidence suggests that some areas of cognition
actually show improvements with increasing age. This has long
been posited for ‘crystallized’ (knowledge-based) aspects of cog-
nition40–42, such as vocabulary, lexical–semantic knowledge, verbal
comprehension and general information, which tend to exhibit
age-related increases until quite late in life; these have tradition-
ally been contrasted with ‘fluid’ (information processing) aspects
of cognition40–42, such as spatial visualization and processing speed,
which often show age-related declines24,26,43–49. However, older adults
have been found to outperform younger adults in other domains
as well. These include ‘wisdom’50,51, theory of mind52, emotional
regulation21,53–56, aspects of decision-making abilities57–60, motiva-
tion related to one’s job61,62 and certain dimensions of personality
such as agreeableness and conscientiousness63. Given that some of
these domains seem to involve fluid (perhaps in addition to crystal-
lized) aspects of cognition64,65, this raises the possibility of multifac-
eted ageing outcomes in fluid processes, including not only decli
nes24,26,43–49 and age invariance (for example, automatic processes of
memory retrieval30–32) but also improvements.
Going beyond which abilities show improvements, research has
also begun to reveal the shapes of these trajectories—that is, when
improvements may occur over the course of ageing. At least for
crystallized cognition, and perhaps for aspects of fluid cognition,
age-related improvements are often nonlinear, with increases tend-
ing to continue up to a certain point, often in one’s 70s, after which
age invariance or declines may set in1,22,26,52,61,66,67. Finally, the why of
cognitive improvements is also beginning to be understood, though
primarily for crystallized abilities thus far. In particular, age-related
improvements seem to be largely explained by the lifetime accu-
mulation of experience or practice, resulting in (neurobiologically
based) increases in knowledge or perhaps in skill efficiency, which
may outweigh any neurobiological declines1,38,48,66,68–76.
We investigated the effect of age on attention and executive
functions—the critical set of processes that allow us to focus on
selective aspects of information in a goal-directed manner, while
ignoring irrelevant information77–79. This set of functions is crucial
Evidence that ageing yields improvements as well
as declines across attention and executive functions
João Veríssimo 1,2 ✉ , Paul Verhaeghen3, Noreen Goldman4, Maxine Weinstein5 and
Michael T. Ullman 6 ✉
Many but not all cognitive abilities decline during ageing. Some even improve due to lifelong experience. The critical capacities
of attention and executive functions have been widely posited to decline. However, these capacities are composed of multiple
components, so multifaceted ageing outcomes might be expected. Indeed, prior findings suggest that whereas certain atten-
tion/executive functions clearly decline, others do not, with hints that some might even improve. We tested ageing effects on
the alerting, orienting and executive (inhibitory) networks posited by Posner and Petersen’s influential theory of attention, in
a cross-sectional study of a large sample (N = 702) of participants aged 58–98. Linear and nonlinear analyses revealed that
whereas the efficiency of the alerting network decreased with age, orienting and executive inhibitory efficiency increased, at
least until the mid-to-late 70s. Sensitivity analyses indicated that the patterns were robust. The results suggest variability in
age-related changes across attention/executive functions, with some declining while others improve.
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles Nature HumaN BeHaviour
for everyday life80,81 and supports numerous higher-level cognitive
capacities82–86.
Theories of the neurocognition of ageing have widely pos-
ited age-related declines in attention and executive functions87–96.
Moreover, these theories have generally assumed that such declines
affect attention/executive functions quite broadly87–93,95,96. This view
is consistent with broad declines in the neurobiological substrates of
these functions97–102. However, reviews and meta-analyses of behav-
ioural findings have suggested that age-related weakening of atten-
tion/executive functions is not universal. Rather, there seems to be
large variability in age effects across tasks and functions, ranging
from clear declines (for example, in working memory and dual-task
paradigms) to age invariance (for example, in aspects of selective
attention and inhibitory control)26,29,79,103–113. This variability is not
surprising, given that different aspects of the broad capacities of
attention and executive functions seem to constitute at least some-
what independent neurocognitive processes78,114–117. Moreover, even
where age-related declines have been observed, including in vari-
ous studies of inhibitory control, at least some of these effects can
be attributed to general age-related slowing rather than to specific
deficits of attention/executive functions118,119. There thus seems to
be a mismatch between the theories as well as the neurobiological
evidence on the one hand and at least some behavioural ageing pat-
terns on the other.
Even more provocative than an absence of declines would be the
existence of age-related improvements in specific aspects of atten-
tion or executive function. This is not as implausible as it might
seem, given that age-related increases have been observed for a
range of functions, including some that seem to involve fluid abili-
ties. As far as we know, no reviews or meta-analyses have uncovered
reliable age-related improvements in aspects of attention or execu-
tive function. It is possible, however, that reviews and meta-analyses
have missed such improvements for various reasons. These reasons
include not only experimental and analytical factors in the included
studies (for example, categorical designs with younger and older
adults preclude the examination of nonlinear ageing patterns; see
below) but also the aggregation of studies with tasks probing dif-
ferent attention/executive functions that show different trajectories,
such as declines and improvements that could cancel each other out.
Examining the effect of age on clearly distinct theoretically moti-
vated functions may therefore be useful.
Here we focus on the highly influential theory of attention pro-
posed by Posner and Petersen, which posits that this broad capacity
in fact comprises three ‘networks’—namely, alerting, orienting and
executive control117,120. The attentional network of alerting is charac-
terized as a state of enhanced vigilance and preparedness to respond
to incoming information, and it features both a phasic aspect (con-
cerned with rapid changes of attention) and a tonic aspect (refer-
ring to more stable vigilance or arousal)121. The orienting network
involves the selection of information from sensory inputs, which
is achieved by shifting processing resources to a given location.
Finally, the executive network consists of a set of top-down pro-
cesses involved in inhibitory function—that is, in detecting conflict
and inhibiting distracting or conflicting information.
The three attentional networks are at least partially independent.
Such independence is supported by substantial behavioural and
neurobiological evidence that associates each network with distinct
processing mechanisms117,120,122–125, neuroanatomical substrates126–129,
neuromodulators130–133 and genetic polymorphisms126,134. Given that
these attention/executive functions show neurocognitive differen-
tiation, we suggest that they may also show distinct susceptibilities
to ageing.
The most common paradigm used to examine the three atten-
tional networks is the Attention Network Test (ANT)122,135. The
ANT is a simple task that nonetheless simultaneously measures
the efficiency of all three networks by combining a cued reaction
time task136 with a flanker task137. Importantly, the ANT shows
moderate-to-high reliability122, including in older adults138–140.
Moreover, the task shows high construct and criterion valid-
ity126,127,141, again including in older adults138. In the ANT, the tar-
get stimulus in every trial typically consists of a central arrow that
points either left or right, with two flanker arrows on each side
(Supplementary Fig. 1). These flanker arrows all point either in the
same direction (congruent) or in the opposite direction (incongru-
ent) as the central arrow. Participants are simply instructed to per-
form a flanker task—that is, to decide the direction of the central
arrow (left or right) by pushing one of two buttons. Each target
stimulus (the central arrow with its flankers) appears either above
or below the centre of the screen. Three general types of warning
cues are presented before the target: (1) no cue; (2) alerting cues
such as central cues, which immediately precede the target, thus
serving as a temporal signal for incoming information; and (c) ori-
enting (spatial) cues, which are displayed above or below the centre
of the screen, consistent with the location of the upcoming target,
and thus convey spatial information relevant for the task. By com-
paring response times (RTs) to the target stimuli between the vari-
ous conditions (the different types of cues and the different types of
flankers), it is possible to estimate the efficiency of each attentional
network (accuracy is generally near ceiling in these tasks and so is
usually not analysed). Specifically, the efficiency of the alerting net-
work is measured as the benefit (RT speed-up) of alerting cues rela-
tive to trials with no cues, the efficiency of the orienting network is
measured as the benefit of orienting cues relative to alerting cues
and the efficiency of the executive network is measured as the cost
(RT slowdown) produced by incongruent flankers relative to con-
gruent flankers—that is, the greater the interference caused by the
flankers pointing in the opposite direction, the lower the efficiency
of the executive network.
We are aware of 13 studies that have previously examined effects
of ageing on the ANT; all were cross-sectional140,142–153. As expected,
and not of primary interest here, these studies found general slow-
downs with ageing in the task—that is, age-related increases in
overall RTs across all conditions. Importantly, the studies also sug-
gest that the three network efficiencies (defined by RT differences
between conditions; see above) show somewhat different ageing tra-
jectories. As can be seen in Table 1, decreases in efficiency with age-
ing have generally been observed for the alerting network, whether
or not age-related declines in processing speed were controlled for
(achieved in most studies by transforming RTs into proportions or
z-scores relative to a participant’s mean RT, or by treating partici-
pant mean RTs as a covariate). In contrast, no age-related changes in
the efficiency of the orienting network have usually been reported,
though sometimes efficiency increases have been found, especially
when processing speed was not controlled for. The ageing trajectory
of executive network efficiency has been less clear, with decreases
or no changes reported when processing speed was not controlled
for, but an apparent shift away from declines when processing speed
was controlled for, with several studies even finding improvements.
Thus, despite the appearance of broadly different ageing tra-
jectories for the three attentional networks, previous studies have
demonstrated a fair bit of variability in this regard, even when pro-
cessing speed was (or was not) controlled for. Such inconsistencies,
which can obscure the true pattern of trajectories, may be due to
various experimental and analytical factors, including the follow-
ing. First, previous studies have had relatively small sample sizes:
between 13 and 77 older adult participants (in addition to similar
sample sizes of younger adults), except for two studies that focused
on older adults, which tested 145 (ref. 149) and 184 (ref. 140) par-
ticipants. Second, almost all studies employed categorical designs
comparing younger and older adult groups rather than a continu-
ous age design across the age range of interest (but see refs. 140,149).
Categorical (factorial) designs lead to information loss because the
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles
Nature HumaN BeHaviour
variability in ages within each group cannot contribute to poten-
tial age effects109,154. Indeed, in such designs, any changes that occur
within an age group cannot be detected; this can be especially
problematic in ageing studies, since many cognitive changes seem
to occur within middle to older adulthood1,3,24,26,29. Furthermore,
categorical designs with two groups cannot examine nonlinear age
effects, even though such nonlinearities are common in cognitive
ageing (see above). Third, age effects (or the lack thereof) in previ-
ous ANT ageing studies may have been distorted by an absence of
precise control of potentially confounding factors, such as the basic
demographic variables of education and sex. Such variables can be
correlated with age109 and may play a role in the efficiency of the
attentional networks155–157. Even when mean levels of education or
sex ratios are matched groupwise (as in many previous studies), dif-
ferent distributions of these variables between the groups and dif-
ferent correlations with age can inflate both false positive and false
negative errors154.
The present study was designed to examine the effect of age on
the efficiencies of the three attentional networks while address-
ing these issues. The ANT was given to a large sample of adults
ranging in age from 58 to 98 years, in a cross-sectional continu-
ous age design (N = 702 in our main analyses; see Extended Data
Fig. 1 for demographic information). We focused on adults from
late middle age to older adulthood to capture changes during the
period when many cognitive changes occur. Our sample included
participants at very old ages (higher than in most previous ANT
studies), allowing us to examine ageing effects across a wide age
range within older adulthood. The analyses were performed on RTs
using mixed-effects regression. Our primary analyses examined the
linear effect of age on the efficiency of each attentional network.
We also examined nonlinear age effects on the efficiency of each
network. We statistically controlled for both sex and education. We
additionally controlled for processing speed to avoid confounding
effects from general age-related processing slowdowns as well as
speed decreases due to motor or perceptual declines in ageing158 or
to slowed responses from lower alertness122. Such processing speed
differences were primarily addressed by log-transforming the RTs,
yielding proportional effects while also reducing skewness and
minimizing the influence of outliers. Finally, we performed a wide
range of sensitivity analyses, including analyses further controlling
for processing speed as well as not controlling for it at all.
On the basis of previous findings, we predicted age-related
decreases in the efficiency of the alerting network, but no age-related
differences or perhaps even increases in the efficiency of the orient-
ing network. The expected ageing trajectory of executive efficiency
was less clear, though we speculated that no differences or even
increases might be possible. Indeed, a combination of age-related
declines, invariance and improvements across and/or within the
networks would not be inconsistent with the broader literature,
given that all have been observed across as well as within cognitive
domains, including in fluid aspects of cognition. (The predictions
were not altered after data analysis or while writing or revising this
paper.)
Results
Age and attentional network effects. Linear mixed-effects model-
ling revealed that age had a significant main effect on RTs (across
all cue and flanker types), such that greater age was associated
with overall slower responses on the task (regression coefficient
(β), 0.0083; 95% confidence interval, (0.0068, 0.0098); t = 10.70;
P < 0.001). (For the accuracy analyses, see Extended Data Fig. 2.)
Each additional year of age was associated with a mean cost of
6.3 ms (in back-transformed RTs; because the analyses were con-
ducted on log RTs, the cost per year in milliseconds was obtained by
back-transforming the effect of age, over the entire age range). All
three attentional network effects (the average effects from the whole
sample of participants across the full age range) were significant:
the alerting effect was 3.3 ms in back-transformed RTs (β = 0.0048
(0.0007, 0.0088), t = 2.31, P = 0.021), the orienting effect was 11 ms
(β = 0.0160 (0.0120, 0.0201), t = 7.79, P < 0.001) and the executive
effect was 47 ms (β = 0.0680 (0.0636, 0.0723), t = 30.40, P < 0.001).
See Extended Data Fig. 3 for the complete model results, including
Table 1 | Summary of results of prior studies examining effects of ageing on the ANT
Network Age-related changes in eciency Summary of results
Alerting Decrease: older adults benefit less from
alerting cues Nine studies reported efficiency decreases, generally regardless of whether processing speed
was controlled for142,145–148,150–153 or not145–148,151,153.
Age invariance: no effect of age on the
benefit of alerting cues Four studies reported no effects of age on alerting efficiency: three while controlling for
processing speed140,143,149 and two while not controlling for it149,152.
Increase: older adults benefit more from
alerting cues One study (with a small sample size) found age-related efficiency increases144, both with and
without controlling for processing speed.
Orienting Decrease: older adults benefit less from
spatial cues One study found age-related decreases in orienting efficiency, with processing speed
controlled for142.
Age invariance: no effect of age on the
benefit of spatial cues Eleven studies reported no effects of age on orienting efficiency, particularly when controlling
for processing speed140,144–153, though sometimes also when not controlling for it144,146,149,153.
Increase: older adults benefit more from
spatial cues Six studies reported efficiency increases, in one case when controlling for processing speed143
but mainly when not controlling for it145,147,148,151,152 (these five studies all reported age invariance
when controlling for processing speed).
Executive Decrease: older adults show more
interference from incongruent flankers Eight studies found age-related decreases in executive efficiency, sometimes when controlling
for processing speed140,1 49,153 but mainly when not controlling for it145–147,149,151–153.
Age invariance: no effect of age on
interference from incongruent flankers Seven studies reported no effects of age on executive efficiency, mainly when controlling for
processing speed143,145–148,151 but occasionally when not controlling for it144,148.
Increase: older adults show less interference
from incongruent flankers Four studies found age-related increases in executive efficiency, in all cases when controlling
for processing speed142,144,150,152.
All statements of efficiency decreases and increases refer to significant effects, as reported in the cited papers. Results from analyses that tested for but did not find significant age effects on network
efficiency are classified as showing age invariance. Several studies reported results only with processing speed controlled for140,142,143,150. This must be taken into account in interpreting the pattern of findings
described in the table, since in these cases the absence of reported effects without controlling for processing speed does not imply that a given pattern was more common when processing speed was
controlled for.
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles Nature HumaN BeHaviour
degrees of freedom, which are the same for all effects in the model.
All reported P values here and elsewhere are two-tailed, with signif-
icance determined as P < 0.05. Extended Data Fig. 4 shows the mean
untransformed RTs for each condition (each cue and flanker type)
in milliseconds, as well as the mean attentional network effects in
milliseconds, presented in five-year age brackets.
Linear effects of age on each attentional network. Age interacted
with all three attentional network effects—that is, with the efficien-
cies of the alerting, orienting and executive networks. Figure 1 dis-
plays these interactions as the effects of age on each of the three
network effects, where each network effect is expressed as the con-
trast between the relevant cue or flanker types (see Extended Data
Fig. 5 for an equivalent figure displaying the effects of age over scat-
ter plots of the attentional effects for each of the 702 participants).
These interactions can also be seen in Extended Data Fig. 6, which
displays age effects separately for each of the cue and flanker con-
ditions; although RTs in all conditions increased with age, these
increases differed among the cue and flanker types, leading to the
observed interactions between age and the three attentional net-
work effects.
First, the alerting effect (that is, the response speed-up in tri-
als with a central cue relative to trials with no cue) decreased with
age (β = −0.0008 (−0.0013, −0.0003), t = −3.05, P = 0.002)—that
is, older participants showed less of a benefit from alerting cues
than younger participants, suggesting that ageing was associ-
ated with decreasing efficiency of the alerting network (Fig. 1a).
Second, the orienting effect (that is, the response speed-up in tri-
als with a spatial cue relative to trials with a central cue) increased
with age (β = 0.0007 (0.0002, 0.0011), t = 2.61, P = 0.009)—that is,
older participants benefited more from spatial cues than younger
participants, suggesting increasing efficiency of the orienting
network with ageing (Fig. 1b). Third, the executive effect (that is, the
response slowdown in trials with incongruent flankers relative to
trials with congruent flankers) also interacted with age (β = −0.0012
(−0.0017, −0.0007), t = −4.42, P < 0.001). Specifically, the executive
effect decreased with age, such that older participants showed less
interference from incongruent flankers relative to congruent ones,
suggesting that ageing was associated with increasing efficiency of
the executive inhibitory network (Fig. 1c).
Linear effects of age: sensitivity analyses. The results were robust,
in that the same pattern of significant age effects for the three network
efficiencies was also obtained in a range of sensitivity analyses, each
of which involved one type of change to the main analysis presented
above. See Supplementary Table 1 for the results from all sensitivity
analyses. First, the same pattern was found when participant mean
log RT was included as a covariate140,142 in addition to sex, education
and trial position, to further control for age-related declines of pro-
cessing speed. Note that in this analysis, processing speed declines
are accounted for not only by employing log-transformed RTs as
the dependent measure, but also by controlling for speed differences
between participants, thereby further reducing the likelihood that
age-related differences in processing speed from general, motor or
perceptual slowdowns158, or from slowdowns from other factors
such as reduced alertness122, could explain the observed patterns.
Second, and conversely, we ran the original statistical model on raw
(unlogged) RTs rather than log-transformed RTs as the dependent
measure, and we again found the same pattern of results. This shows
that the age effects on the three attentional networks were also
detectable as linear effects on the millisecond scale. Thus, the pattern
was found even when processing speed was not controlled for at all.
Third, because correlations between predictors (see ‘Participants’ in
Methods) can substantially change regression estimates and even
Decreasing efficiency
with ageing
–0.05
0
0.05
0.10
100
Age (years)
Alerting effect (log-RT difference)
Alerting network
Increasing efficiency
with ageing
–0.05
0
0.05
0.10
100
Age (years)
Orienting effect (log-RT difference)
Orienting network
Increasing efficiency
with ageing
–0.05
0
0.05
0.10
60 70 80 90 60 70 80 90 60 70 80 90
100
Age (years)
Executive effect (log-RT difference)
Executive network
ab c
Fig. 1 | Linear effects of age on the efficiencies of the three attentional networks. a, Effect on the efficiency of the alerting network. b, Effect on the
efficiency of the orienting network. c, Effect on the efficiency of the executive network. The shaded bands represent pointwise 95% confidence intervals.
The regression lines and bands are shown from the minimum to the maximum age in our sample (ages 58–98). For each network, we followed up on
the interaction between age and efficiency with analyses examining efficiency at different ages. For alerting (a), at the minimum age of 58 years, the
benefit of a central cue was significant (8 ms benefit; β = 0.0121 (0.0059, 0.0182), t = 3.82, P < 0.001), but by age 74 this effect was no longer positive
(β = −0.0001 (−0.0053, 0.0050), t = −0.05, P = 0.957). By the maximum age of 98 years, the central cue had a significant detrimental effect relative
to no cue (16 ms cost; β = −0.0184 (−0.0339, −0.0030), t = −2.34, P = 0.019). As there were only a few participants above age 90 (and thus the sparse
data in this age range may not generalize to other samples), we additionally estimated the alerting effect at age 90; there was also a significant cost at
this age (β = −0.0123 (−0.0240, −0.0006), t = −2.06, P = 0.039). For orienting (b), although there was already a significant benefit from spatial cues at
age 58 (6 ms; β = 0.0098 (0.0036, 0.0160), t = 3.10, P = 0.002), this benefit was much larger by age 98 (32 ms; β = 0.0359 (0.0204, 0.0513), t = 4.55,
P < 0.001)—almost four times larger as indicated by the β values, which reflect differences in log-transformed RTs, and more than five times larger in
back-transformed RTs (values in milliseconds). A similar orienting benefit was found at age 90 (β = 0.0306 (0.0189, 0.0423), t = 5.13, P < 0.001) as at age
98. For the executive network (c), although the interference cost of incongruent flankers was present throughout the age range, it was much larger at age
58 (51 ms; β = 0.0794 (0.0727, 0.0861), t = 23.14, P < 0.001) than at age 98 (28 ms; β = 0.0316 (0.0149, 0.0483), t = 3.71, P < 0.001)—more than twice
as large as revealed by the regression coefficients and almost twice as large in back-transformed RTs. A similar executive effect was obtained at age 90
(β = 0.0411 (0.0285, 0.0538), t = 6.37, P < 0.001) as at age 98.
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles
Nature HumaN BeHaviour
lead to sign reversals159,160, we examined whether the observed age
effects were also obtained when age was not adjusted for covariates,
by refitting the original model without any covariates (that is, with-
out sex, education and trial position). This model also yielded the
same pattern of results. Fourth, age-related visual declines161 could
result in slowed or inaccurate perception of the stimuli, which might
explain some of the observed patterns. For example, visual declines
resulting in difficulties identifying the arrows could lead to reduc-
tions in congruent/incongruent RT differences, providing an alter-
native account for the observed age effect on the executive network.
However, a model in which we also covaried out a measure of visual
acuity (see ‘Participants’ in Methods) again produced the same pat-
tern of significance, even though decreases in visual acuity nega-
tively impacted RTs in the task (β = −0.0454 (−0.0906, −0.0001),
t = −1.98, P = 0.048). Note also that the stimuli (cross, asterisk and
arrows) were enlarged as compared with the original task162 to mini-
mize problems due to age-related declines in visual acuity, while the
distance between the central point and the arrow positions above or
below it was still maintained from the original task, to minimize dif-
ficulties from declines in peripheral vision or from any changes in
participants’ spatial attentional gradient163 or ‘useful field of view’164.
Fifth, it might be argued that older participants showed less of an
interference effect (an apparent increase in executive efficiency)
because, being slower overall than younger participants, they were
more likely to bump up against the trial timeout of 1,700 ms (par-
ticularly on the slower incongruent trials), thus yielding less of a
difference between the congruent and incongruent conditions.
However, as shown in Extended Data Fig. 7, very few of the RTs
of even the older participants approached 1,700 ms for incongru-
ent let alone congruent trials, arguing against such a timeout expla-
nation. Furthermore, a sensitivity analysis restricted to responses
below 1,600 ms as well as below 1,500 ms showed the same pattern
as the main analysis, with age effects on executive function of the
same magnitude. Sixth, although excluding double cues allowed
for the analysis of all three attentional effects in the same statisti-
cal model140,143,149,150 (see ‘Data exclusions and analysis’ in Methods),
it resulted in information loss, specifically in the computation of
the alerting and executive effects, both of which can be analysed
with both central and double cues122,145,146. An alternative analysis of
alerting and executive effects that included both the double and
central cues again yielded the same pattern as the main analysis.
Seventh, the same pattern of results remained when analysing a
larger sample (734 participants) that additionally contained 32
‘low-accuracy’ participants, by employing a more lenient inclu-
sion criterion of at least 50% task accuracy, rather than 75% (see
‘Participants’ in Methods). Eighth, and conversely, when we used
a more stringent criterion of including only participants with at
least 90% accuracy (643 participants), again the same pattern was
obtained. Finally, the same results were also found when extremely
old participants (that is, four participants aged 90–98) were excluded
from the original model (since they may have constituted outliers).
Nonlinear effects of age on each attentional network. Given non-
linearities in other cognitive ageing effects1,22,26,66,67, we also tested for
nonlinearities in the effect of age on each of the three networks. First,
we included a quadratic term for age (as an orthogonal polynomial),
which interacted significantly with the executive effect (β = 1.3482
(0.3886, 2.3078), t = 2.75, P = 0.006). This interaction indicated that
whereas the executive effect (response slowdown of incongruent rel-
ative to congruent flankers) was present throughout the age range,
it first diminished (that is, efficiency increased) but then plateaued
(with a maximum efficiency at age 78), after which there seemed
to be a reversal, with decreasing efficiency from this point onwards
(Fig. 2a). In contrast, the quadratic term for age did not interact
significantly with either the alerting (β = 0.4394 (−0.4552, 1.3337),
t = 0.96, P = 0.336) or orienting effects (β = −0.3221 (−1.2148,
0.5704), t = −0.71, P = 0.480). Note that the non-significance of
these interactions should not be taken as evidence for the absence
of nonlinearities in the effect of age on these two networks: even in
a large sample, nonlinearities can remain undetected due to insuf-
ficient statistical power or because the particular nonlinearity that
is tested (in this case, a quadratic effect) may not capture the shape
of the underlying effect.
Indeed, despite their wide use, quadratic polynomials impose a
particular functional form on the estimated effects. The observed
quadratic effect thus might not indicate a true reversal for older par-
ticipants but rather some other nonlinearity, such as a stable execu-
tive effect at older ages (consistent with the wide confidence interval
Increasing efficiency
with ageing until age 78
0
0.05
0.10
0.15
60 70 80 90 100
Age (years)
Executive effect (log-RT difference)
Quadratic model
Increasing efficiency
with ageing until age 76
0
0.05
0.10
0.15
60 70 80 90 100
Age (years)
Executive effect (log-RT difference)
Breakpoint model
ab
Fig. 2 | The nonlinear effect of age on the efficiency of the executive network. a, The model with a quadratic term for age. b, The breakpoint model with
the optimal breakpoint—that is, at age 76. The shaded bands represent pointwise 95% confidence intervals. The regression lines and confidence interval
bands are shown from the minimum to the maximum age in our sample (ages 58–98). See Extended Data Fig. 9 for an equivalent figure showing the
executive effect for each of the 702 participants.
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles Nature HumaN BeHaviour
on the right side of Fig. 2a). We therefore also examined the non-
linearity with linear regression-with-breakpoints. This technique
estimates two linear slopes joined at a ‘breakpoint’165,166. Such break-
point models allow various nonlinear shapes to be detected, includ-
ing the reduction, elimination or reversal of an effect at a particular
point. We employed a breakpoint discovery procedure to estimate
the best location (age) of the breakpoint by fitting a set of regression
models, each with a breakpoint at a different age, and selected the
model with the best fit165,167,168.
The discovery procedure (Extended Data Fig. 8) revealed that the
optimal breakpoint model contained a breakpoint at age 76. The esti-
mates of this model showed that the two linear slopes (that is, before
versus after age 76) differed significantly from each other for the
executive network (β = 0.0039 (0.0014, 0.0063), t = 3.08, P = 0.002).
Thus, consistent with the quadratic model, a nonlinear age effect was
found for the executive network. In this model (Fig. 2b), the slope of
the age effect from age 58 to 76 was significantly different from zero
and indicated decreasing interference costs during this age range—
that is, increasing executive efficiency (β = −0.0022 (−0.0030,
−0.0014), t = −5.20, P < 0.001). Follow-up analyses revealed that the
interference produced by incongruent flankers relative to congruent
flankers was much larger at the minimum age of 58 years (interfer-
ence cost of incongruent flankers, 54 ms; β = 0.0851 (0.0775, 0.0927),
t = 21.95, P < 0.001) than at 76 years (33 ms; β = 0.0452 (0.0351,
0.0554), t = 8.73, P < 0.001)—almost twice as large as revealed both
by the regression coefficients and in back-transformed RTs. In con-
trast, the slope from ages 76 to 98 showed a numerical reversal of
the effect (greater interference of the incongruent flankers with
increasing age), though this was marginally significant (β = 0.0017
(−0.0002, 0.0035), t = 1.72, P = 0.086). Unlike the executive network,
the difference between the two linear slopes was not statistically sig-
nificant for either the alerting (β = 0.0009 (−0.0014, 0.0032), t = 0.76,
P = 0.445) or orienting networks (β = −0.0014 (−0.0037, 0.0009),
t = −1.22, P = 0.221). Extended Data Fig. 9 displays the nonlinear
effect of age on the efficiency of the executive network, from both
the quadratic and breakpoint models, over scatter plots of the execu-
tive effects for each of the 702 participants.
Nonlinear effects of age: sensitivity analyses. The observed non-
linear breakpoint pattern was robust, in that the same results were
obtained in the nine sensitivity analyses reported above for the linear
analyses (see Supplementary Table 2 for the model results), namely:
(1) with the mean log RT of each participant included as a covari-
ate to further control for age-related processing slowdowns even
beyond employing log-transformed RTs as the dependent measure;
(2) with raw (unlogged) RTs as the dependent measure and thus not
controlling for processing speed at all; (3) without any covariates;
(4) when controlling for visual acuity; (5) with responses restricted
to those below 1,600 ms as well as below 1,500 ms, to test for poten-
tial timeout effects; (6) when including both double and central
cues in the analysis; (7) with a more lenient participant inclusion
criterion, yielding a larger sample of participants; (8) with a more
stringent participant inclusion criterion, yielding a smaller sample
of participants; and (9) when excluding extremely old participants.
All of these models yielded a significant difference between the age
effect on executive efficiency from age 58 to 76 and the age effect
from 76 to 98, but no such significant differences for the alerting
and orienting networks.
Figure from McDonough et al.’s (2019) review107
Normal ageing
Attention network efficiency
Orienting
Executive
Alerting
Age effects in the present study
Attention network efficiency
Normal ageing
Orienting
Executive
Alerting
ab
Fig. 3 | Comparison between a recent qualitative review of age effects on the three attentional networks and the findings from the present study. a, A
graphical summary presented by McDonough et al., which examined effects of age on the three networks in the ANT and related tasks in prior studies.
Image adapted with permission from ref. 107. b, The findings obtained in the present study. The age effects in b are fitted values from the regression models
presented in the Results: linear age effects for alerting and orienting, and a quadratic age effect for the executive network. To make the trajectories of
the three networks more comparable, the executive effect was transformed from a cost metric to an efficiency metric (by multiplying the effect by −1);
the three trajectories were made approximately equal at the origin (by subtracting the predicted effect at the minimum age from all predicted values,
for each network); and the three trajectories were standardized (by dividing the predicted values by the standard deviation of by-participant effects, for
each network). The graphical summary from the recent review is both similar to and critically different from our findings. On the one hand, the shapes of
the three ageing trajectories in the summary show striking parallels to the pattern of our results, with strong declines for alerting but not for orienting or
executive function. On the other hand, the summary emphasizes age-related stability rather than improvements for the orienting and executive networks,
because significant age effects were not found in a number of prior studies of orienting107 (which may have been partly due to low statistical power from
small sample sizes) and because variability was found across studies for the executive network107 (which may have been partly due to the use of linear
rather than nonlinear analyses). Thus, while the recent review (as well as other reviews and theories79,87–93,95,96,105,106,108) indeed seems to have captured
certain aspects of the effect of ageing on attention/executive functions, the results of the present study suggest updating this view to also include
age-related improvements.
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles
Nature HumaN BeHaviour
Discussion
This study examined effects of ageing on the alerting, orienting and
executive attentional networks posited by Posner and Petersen’s the-
ory of attention117,120. The efficiency of the networks was tested with
the ANT—a behavioural task with high reliability and validity—in
a large sample of middle-aged to older adults. Linear mixed-effects
regression indicated that whereas the efficiency of the alerting net-
work decreased with increasing age, the efficiency of both the ori-
enting and executive networks increased. In other words, whereas
older participants benefited less from the presence of a central cue
relative to no cue (alerting effect), they benefited more from a cue
that directed their attention to a particular location (orienting effect)
and showed less interference from incongruent flankers (executive
effect). Nonlinear mixed-effects modelling revealed that the effect
of age on the executive network was in fact nonlinear: the increase
in executive efficiency held until the mid-to-late 70s, but after that,
the increase was eliminated and possibly reversed, with an appar-
ent decrease in executive efficiency during older adulthood. (Note
that since the ANT yields only behavioural measures, these findings
should not be interpreted as direct reflections of age-related changes
in the brain networks underlying the attentional processes.)
The linear and nonlinear age effects were robust and did not
seem to be explained by a variety of alternative accounts. We have
seen that these effects held across a wide range of sensitivity analy-
ses. Moreover, further examination revealed that the age effects
were probably not due to speed–accuracy trade-offs: unlike in the
RT analyses, the accuracy analyses (Extended Data Fig. 2) revealed
no significant age-related changes in efficiency for the alerting or
orienting networks, and executive function actually showed signifi-
cant efficiency increases, arguing against speed–accuracy trade-offs
in all three networks. In fact, this accuracy result for the executive
network underscores the robustness of the increases in executive
efficiency, which were thus found in both speed and accuracy. The
increases in orienting and executive efficiency also do not seem
to be explained by sample selectivity at older ages—for example,
survivor effects or the self-selection of fitter adults in old age169–171.
First, whereas the efficiency of the orienting and executive networks
indeed increased with age, alerting efficiency decreased in the same
sample. Working memory abilities were also found to decrease
with age in the same sample of participants, as reported in a recent
study109 (also see Extended Data Fig. 1). Finally, the nonlinear effect
of age on executive function indicates that improvements occurred
during earlier rather than later older adulthood, whereas selection
biases should be stronger at later ages. The nonlinear effect of age-
ing on executive function also argues against the possibility that
the increases in executive efficiency might actually be due to con-
comitant decreases in alertness122, since alerting efficiency showed a
linear decrease (with no evidence for a nonlinearity), which would
predict a similar linear increase in executive efficiency. More gener-
ally, it is not clear how a variety of other alternative accounts (for
example, age-related slowing or age-related changes in participants’
visual abilities, spatial attentional gradient or useful field of view)
could easily explain the different age-related trajectories of the three
networks (and working memory), as well as the nonlinear effects
obtained for executive function.
The fact that age-related efficiency decreases were not observed
across all three networks is consistent with prior data, which sug-
gest that ageing leads to declines in some but not other aspects of
the wide domain of attention/executive functions. First, our results
are broadly consistent with the pattern of clear declines observed
for alerting efficiency but not for orienting or executive efficiency
in previous ANT ageing studies (Table 1); see Fig. 3 for a com-
parison between a recent review of these studies and the find-
ings presented here. Moreover, meta-analyses have suggested that
attention/executive functions demonstrate different trajectories
for different functions, including an absence of reliable declines
in executive inhibitory tasks110–113. We emphasize that we are not
suggesting that no executive functions show age-related declines,
and indeed, meta-analyses and reviews reveal that other aspects
of executive function decline robustly, including working mem-
ory79,103,104,106,111–113. In fact, as mentioned above, a recent study of
the same sample of participants examined here found that working
memory showed strong declines109, underscoring that age-related
decreases can be found in some but not other aspects of attention/
executive function even in the same group of individuals. Thus, our
findings, together with other data, argue against theories positing
general age-related declines in attention and executive functions,
including proposals such as that put forth by Hasher, Zacks and
colleagues that ageing leads to broad deficits in inhibition87–93,95,96.
Negative impact Positive impact
Neurobiological
mechanisms
Declines Efficiency
Alerting
Orienting
Executive
Attentional
networks
Maintenance
Compensation
Reserve
From experience
Fig. 4 | A neurocognitive account of age effects on the three attentional
networks. We propose that the observed age-related efficiency changes
in the three networks are explained by a combination of neurobiologically
based mechanisms that have previously been implicated in other cognitive
ageing trajectories1,38: declines, maintenance, compensation and reserve
(“cumulative improvement…of neural resources”38), particularly from
experience. First, the neurobiological substrates of all three networks show
age-related declines, which probably differ in their trajectories99–102,214–218.
Countering these declines, some age-related maintenance and
compensation may occur across the networks1,38,88,193. This seems best
studied for compensation, especially for the executive network88,107, as
well as more specifically for dopaminergic processes, which underlie
executive function219,220. Crucially, the declines should also be countered
by reserve-related gains in neural efficiency, in particular from lifelong
experience and practice with the networks43,48,66,69–72,75 . Depending on the
degree of these gains and the extent of the neural declines (as well as the
scope of any maintenance and compensation), this may not only mitigate
performance decreases38 but also lead to observable improvements1,74,195.
We hypothesize that learning and thus experience-based gains are much
more likely to occur in the orienting and executive than alerting networks.
The orienting and executive networks involve skills of selective attention
(that is, selectively attending to particular types of information, such as
a spatial location or a particular object in space)79, and as skills they can
presumably improve with experience221. By contrast, alerting involves
obtaining and maintaining an alert state and thus depends more on
basic processes of vigilance and preparedness117,120,131. Evidence suggests
that the inhibition of conflicting information, and more generally the
efficiency of executive function, can benefit from practice and training,
further supporting the hypothesis that age-related experience can lead
to improvements in executive inhibitory function221–223. Additionally,
a substantial if controversial literature suggests that individuals with
more experience in certain aspects of executive function (including
inhibitory control), such as bilinguals or musicians (versus monolinguals
or non-musicians), may show broader improvements in such functions—
especially in older adulthood, including in tasks such as the ANT162,224–231.
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles Nature HumaN BeHaviour
Rather, the data support a view in which some but not other aspects
of attention and executive functions decline94,110,163. Importantly,
variability in age effects across attention/executive functions is
also consistent with independent evidence indicating the neuro-
cognitive separability of such functions, not only among the three
attentional networks117,122,123,127,134 but even to some extent among
executive functions, including working memory and inhibitory
control78,114–116,172.
Our results suggest that aspects of attention/executive func-
tion are not just resistant to age-related declines but can even show
improvements, which were observed for both orienting and execu-
tive inhibitory function. These improvements are not surprising.
Age-related improvements have been found not only in various other
domains (including some involving fluid intelligence)21,40,42,43,52,53,60,66
but also crucially in orienting and executive efficiency in a number
of prior ANT studies (Table 1). Indeed, these orienting and execu-
tive improvements were observed in a wide range of Western popu-
lations143–145,147,148,150–152, suggesting that the present findings are not
unique to people living in Taiwan, the population examined here
(‘Participants’). Furthermore, a number of previous studies using
flanker tasks (which are incorporated in one form into the ANT)
have found age-related improvements in the efficiency of the execu-
tive network163,173–177. Interestingly, a recent study examining age
effects on multiple inhibition tasks reported that whereas tasks that
tap into the inhibition of prepotent responses (such as stop-signal
and Simon tasks) show performance declines with age, those that
involve the inhibition of distractors (such as arrow flanker, let-
ter flanker and number Stroop tasks) show improvements163. This
generalizes our finding of age-related improvements in the execu-
tive network beyond the arrow flanker task (increasing the validity
of the finding) and provides a possible framework for identify-
ing which inhibitory functions improve and which decline. This
fine-grained distinction, in which different ageing trajectories may
be found for different inhibitory processes, is in fact consistent with
empirical data and theoretical positions suggesting neurocognitive
distinctions among different types of inhibition, including between
the inhibition of distractors and prepotent responses175,178–181. More
generally, such a finer-grained distinction agrees with the view
espoused in this paper that different attention/executive (sub)
processes may show different ageing trajectories, potentially even
across different types of inhibition110,111,113,175,182, and further coun-
ters unitary views of executive or inhibitory declines87–93,95,96. Note
that, as with the ANT, not all studies using flanker183–187 or num-
ber Stroop188 tasks have reported age-related improvements (some
have reported declines and others age invariance); we suggest that
variability across ageing studies using these tasks may be due to
some of the same experimental and analytical factors discussed in
the Introduction.
Overall, our findings suggesting improvements in the orient-
ing and executive networks have clear precedents in the litera-
ture, despite the emphasis on age-related declines of attention and
executive functions in prior theoretical frameworks. Given these
precedents, in addition to our relatively large sample size, the exam-
ination of age as a continuous variable (moreover within the period
of likely changes, from middle age to older adulthood), the use of
both linear and nonlinear analyses, the robustness of the effects and
evidence against alternative accounts, our results do not seem likely
to be false positives.
How important are the observed age-related improvements?
The millisecond differences between middle-aged and older adults
in orienting and executive efficiency were relatively small. Such
small age effects are not surprising, given that the attentional effects
over the full age range were also quite small—in fact, smaller than
have generally been reported in older adults in previous ANT
studies140,143,145,146. Indeed, this was not unexpected, since stud-
ies with larger sample sizes often yield smaller effect sizes that are
probably closer to the true effects189–192. Nonetheless, the efficiency
of the orienting network increased about four to five times over the
age range in our sample (that is, from middle age to older adult-
hood), while the efficiency of the executive network increased about
twofold from middle age to the mid-to-late 70s. In fact, this execu-
tive efficiency increase was large enough that even the subsequent
apparent decline never led to worse predicted performance at very
old ages than at the minimum age in our sample (Fig. 2). The rel-
evance of the age-related improvements in executive efficiency is
further underscored by the finding that executive improvements
were observed for accuracy as well as RTs. Finally, given that ori-
enting and perhaps especially executive function underlie many
other—particularly higher-level—cognitive capacities, age-related
improvements in these attention/executive functions probably have
downstream effects on the wide range of cognitive abilities that
use them, including spatial navigation, long-term memory encod-
ing and retrieval, decision-making, reasoning, mathematical abili-
ties and language processing82–86. The importance of the observed
effects may therefore be substantial, perhaps especially for executive
inhibitory function.
The multifaceted pattern of ageing observed across the networks
(and working memory in the same sample) is probably explained
by a combination of neurobiologically based mechanisms that
have been invoked in the broader literature on ageing and cogni-
tion1,20,38,48,66,69–75,88,193–195: neural declines that are countered by main-
tenance (repair), compensation and experience-related gains, all of
which may vary across the networks (Fig. 4). We suggest that ori-
enting and executive function are especially susceptible to learning
and thus to experience-based gains in neural efficiency (Fig. 4). An
account invoking neural declines countered by experience-related
gains can also explain the nonlinear effect of age found for the
executive network: up to one’s mid-to-late 70s, the impact of the
declines may be outweighed by the benefits accrued from continual
experience with the network; however, at older ages, the cumulative
effect of the declines (perhaps together with an asymptotic learning
effect) may yield detectable performance decreases. This account
may also explain similar patterns of improvement till around one’s
70s in other (crystallized and perhaps fluid) abilities3,26,196. Note
that because different domains that yield improvements in ageing
can show somewhat different trajectories, including the apparent
absence of late declines24,26,46,52,53, the lack of an observable late-age
decline in orienting is not particularly unexpected. Future studies
may reveal whether orienting in fact shows late or shallow declines
that were not detectable here.
This study’s limitations and findings suggest new lines of
research. Studies should test the generalizability of the results to
(1) other tasks, including additional tasks that involve the inhibi-
tion of distractors163 as well as tasks probing other types of inhibi-
tion163,175,178,179,181; and (2) other populations, including samples with
higher mean levels of education, more individuals at the oldest ages
and younger adults—while still controlling for sex, education, pro-
cessing speed, visual abilities and perhaps other variables. Studies
should still use continuous age designs and investigate both linear
and nonlinear effects of age. Testing the generalizability of our find-
ings in longitudinal designs would be of particular interest, since
age-related differences found in cross-sectional designs could also
be due to other factors, and age effects in cognition can display dif-
ferent magnitudes in cross-sectional and longitudinal studies22,196.
For discussions of comparisons between cross-sectional and longi-
tudinal designs in ageing, see refs. 196–198.
Given our hypothesis that the orienting and executive improve-
ments are largely explained by lifelong experience with these net-
works, future research should attempt to tease apart age and the
amount (and perhaps quality) of this experience. Investigations
that reveal the neural bases of the experience-based gains are par-
ticularly warranted. Indeed, underlying neurobiological gains could
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles
Nature HumaN BeHaviour
occur even if they do not lead to observable performance improve-
ments, given the counteracting neurobiological declines1,38. Studies
should also examine whether and to what extent the gains observed
in orienting and executive function might share common mecha-
nisms, including in aspects of (perhaps spatial-related) skills of
selective attention. It may additionally be worthwhile to further
probe whether orienting and executive improvements are found
irrespective of whether processing speed is controlled for, as was
observed here, or whether these improvements may be moderated
by processing speed, as prior studies seem to suggest (Table 1).
Understanding the nature of individual differences in age-related
orienting and executive improvements, including the possibility of
‘super-agers’, also seems desirable. Critically, future research should
investigate likely downstream effects of orienting and executive
improvements on higher cognitive functions. These empirical con-
tributions should lead to theoretical accounts of attention/executive
functions in ageing that robustly integrate experiential, cognitive
and neurobiological factors25,71,199. Finally, given the likely effect of
experience on improvements in both orienting and executive func-
tion, our findings may open doors to translational research and
practical benefits, perhaps beyond healthy ageing.
In sum, our findings suggest that whereas alerting efficiency
shows age-related declines from middle age to older adulthood,
both orienting and executive inhibitory efficiency improve during
these years, apparently until quite old age for orienting and until
one’s mid-to-late 70s for executive function. This underscores the
notion that, even though ageing is widely viewed as leading to cog-
nitive declines, it in fact yields multifaceted outcomes, including a
range of benefits43,47,48,50,55,66,74,195. Importantly, the evidence presented
here demonstrates that these benefits are not limited to crystallized
cognition but rather extend to fluid processes, in particular aspects
of attention and executive function.
Methods
Participants. is study complied with all relevant ethical regulations and was
approved by the Institutional Review Boards at Georgetown University, Princeton
University and the Ministry of Health in Taiwan. Informed consent was obtained
from all participants.
The study was part of the Social Environment and Biomarkers of Aging
Study (SEBAS), conducted in Taiwan200,201. SEBAS was in turn part of the Taiwan
Longitudinal Study of Aging, which examines a nationally representative sample
of middle-aged and older adults in Taiwan200,201. During (only) the 2011 SEBAS
collection wave, three computer-based cognitive tasks were included: a working
memory task109, a declarative memory task73 and the ANT presented here. See
Extended Data Fig. 1 for a summary of results from the working memor y and
declarative memory tasks. The computer-based tasks were given to all available
consenting individuals in the health examination component of SEBAS, which
excluded participants if they lived in an institution, were seriously ill or had some
other health condition that precluded examination. The age range in the present
study was determined by the age range of these participants in 2011 (refs. 109,200).
The ANT was given to 948 participants, of whom 92 were excluded owing to
participant identifier coding errors (3 participants), not performing the full task
(18 participants) or a diagnosis of a neurological, psychiatric, learning, cognitive
or other brain-related disorder (for more information, see ref. 109), including
dementia, brain atrophy/degeneration, stroke, intracranial haemorrhage, brain
tumour, schizophrenia and epilepsy, among others (71 participants). An additional
154 participants were excluded because of low accuracy on the ANT (lower than
75%, following ref. 140). Note that the same pattern of significance in the findings
was obtained in a sensitivity analysis with a more stringent inclusion criterion of
at least 90% accuracy (213 participants excluded rather than 154), as well as in
another sensitivity analysis with a more lenient inclusion criterion of at least 50%
accuracy (122 participants excluded); see Results. The exclusion of all participants
who had diagnoses of brain-related disorders, as well as those who did not perform
the full task or who had low accuracy, together with the sensitivity analyses
showing that the same pattern of results was obtained with both more and less
stringent accuracy inclusion criteria, suggests that the results represent healthy
ageing and were not due to the inclusion of participants with cognitive impairment.
After these exclusions, statistical analyses were conducted on the data from
the remaining 702 participants (mean age, 67.69 years; s.d., 8.30; range, 58–98;
for the age distribution, see Extended Data Fig. 1). (Note that because there were
few participants in their 90s, we also performed a sensitivity analysis with these
individuals excluded; the same pattern of significance was obtained (Results)). The
much larger sample size compared with previous ANT ageing studies increases
statistical power, yields more precise estimates that are probably closer to true
effects189–192 and facilitates the detection of participant variability (for example,
across ages) relative to trial noise202,203. To control for the possible confounding
effects of education (mean years of education, 7.76; s.d., 4.52; range, 0–17+)
and sex (323 female), both of which were associated with age (age–education
correlation, with higher ages associated with lower education levels: r = −0.33
(−0.39, −0.26), t(700) = −9.14, P < 0.001; males older than females, mean ages
68.39 versus 66.88: difference = 1.51 (0.28, 2.74), t(700) = 2.41, P = 0.016), these
variables were included as covariates in the analyses73,109; the same pattern of results
was obtained in a sensitivity analysis without these covariates included (Results).
For the sensitivity analysis controlling for visual acuity, we obtained visual acuity
measures (Snellen fraction) for 547 of our 702 participants, from the 2006 SEBAS
collection wave, as such measures were not available in the 2011 wave. The
validity of this measure was confirmed by its significant effect on RTs in the ANT
(Results)—a finding that was not surprising, since participants were given the ANT
only a few years after visual acuity was measured.
Materials and design. The ANT task was based on the version employed by Costa
and collaborators162. In each of the 96 trials, the participants saw a ‘target’ row of
five horizontally aligned images of arrows and had to indicate the direction of the
central arrow, which faced either the same or the opposite direction as the four
surrounding ‘flanker’ arrows (Supplementary Fig. 1). The target rows appeared
either below or above the centre of the screen. In each trial, the target row was
preceded by one of four different cue types: (1) no cue; (2) an asterisk in the centre
of the screen (central cue); (3) an asterisk either above or below the centre of the
screen (spatial cue), which was always consistent with the spatial location of the
subsequent target row; and (4) two asterisks presented both above and below
the central fixation point (double cue). Thus, there were 32 possible cue–target
sequences (4 cue types × 2 target locations × 2 central arrow directions × 2 flanker
arrow directions), each presented three times to each participant. Note that
although Costa and collaborators162 also included a neutral condition of dashes as
flankers, this is often omitted from ANT tasks (as in the present study) or from
analyses, including in studies of older adults140,144,151, because responses to neutral
and congruent flanker targets are highly correlated122, and because this omission
simplifies and shortens the task. The images were enlarged for all participants to
minimize visual difficulties for the older population (see ‘Linear effects of age:
sensitivity analyses’ in Results).
Procedure. Each trial began with a central fixation cross (400 ms), which was
followed by one of the four types of cue (100 ms)162: the continuing fixation cross
accompanied by one asterisk either above or below it (spatial cue; Supplementary
Fig. 1); two asterisks both above and below the continuing cross (double cue); no
asterisk—that is, just the continuing fixation cross (no cue); or a central asterisk
in lieu of the fixation cross (central cue). The fixation cross was then shown alone
for 400 ms, after which it was accompanied by the target row of arrows (until a
response or the timeout of 1,700 ms; Supplementary Fig. 1). After an interstimulus
interval of 400 ms (blank screen), the next trial began. The participants were
instructed to report the direction of the central arrow as quickly and accurately as
possible by pressing one of two buttons on a response box (SRBOX, Psychology
Software Tools). The task was presented on a Windows XP laptop with E-Prime
version 2.0. Both the participants and the experimenters were blind to the aims of
the study, in particular the examination of age effects on the efficiency of
each network.
Data exclusions and analysis. Before the analyses, trials with timeouts (1.08%
of all trials), presses of inappropriate response box buttons (0.02%) and incorrect
responses (1.62%) were excluded, since (following the ANT literature) our analyses
focus on RTs of correct responses; for accuracy analyses, see Extended Data Fig.
2. Extremely fast responses (that is, below 100 ms) were discarded (0.009% of the
remaining data) because they are likely to constitute ‘short outliers’204. Responses
to double cue trials were additionally excluded (following refs. 140,143,149,150), owing
to their redundancy with central cue trials127,140; this enabled the calculation of the
alerting and orienting effects within a single statistical model by comparing no cue
and spatial cue trials against the same baseline (central cue trials)140. A sensitivity
analysis including both central and double cue trials yielded the same pattern of
results (Results).
Analyses were conducted on the natural logarithms of RTs. The linear statistical
models reported thus assume linear effects on the log scale and so correspond to
proportional effects on the millisecond scale (for example, two distinct differences
in milliseconds are treated similarly if they correspond to the same percentage
differences). The log transformation is commonly applied to RT distributions
because it provides several advantages relative to analyses of raw (untransformed)
RTs: (1) it reduces the high skewness of RTs, normalizing the distribution of
residuals and thus satisfying the assumptions of linear statistical models205,206; (2)
it minimizes the influence of ‘long outliers’, which are common in button-pressing
tasks204; and, perhaps most importantly, (3) it reduces the large differences that
exist across items, conditions or individuals in their processing speed, a particular
problem in studies of ageing15,158,207. Nevertheless, a sensitivity analysis was
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles Nature HumaN BeHaviour
conducted on raw (untransformed) RTs; this yielded the same pattern of findings
as the main analysis (Results).
The data were analysed with linear mixed-effects regression models in
R, using lme4 package version 1.1-23208]. Fixed effects were cue type (no cue,
central cue or spatial cue), flanker type (congruent or incongruent) and age (in
years; continuous). The two-way interactions between these predictors were also
included as predictors—namely, cue type × flanker type (which is standard in
ANT analyses122,140,146) and the critical interactions of interest: age × cue type and
age × flanker type, which allow estimating the effect of age on the three attentional
networks (alerting, orienting and executive). Additionally, sex (male or female) and
education (in years; continuous) were included as covariates (without interactions)
to control for correlations between age and these variables. Trial position in
the experiment was also included as a covariate without interactions to control
for practice/fatigue effects and to reduce the autocorrelation of residuals205. A
sensitivity analysis without any covariates yielded the same pattern of findings
(Results). To avoid complexity, higher-level interactions were not included in the
statistical model.
The predictors were coded to allow the assessment of the three attentional
network effects and the main effect of age within a single statistical model.
Specifically, we used ‘sum contrasts’ for the two-level factors—that is, for flanker
(congruent = −0.5, incongruent=0.5) and sex (female = −0.5, male = 0.5); we
employed ‘sliding difference contrasts’ for the three-level factor cue (to estimate
the alerting effect: central = −1/3, no cue = 2/3, spatial = −1/3; to estimate the
orienting effect: central = 1/3, no cue = 1/3, spatial = −2/3)209; and we centred all
continuous predictors at their respective means (age, education and trial position).
With these regression contrasts included in the same model, it was possible to
obtain estimates comparing central with no cue conditions (alerting), spatial
with central cue conditions (orienting) and incongruent with congruent flanker
conditions (executive)209. Each of these comparisons was estimated at the mean of
all other predictors, while controlling for all covariates. Follow-up comparisons
(for example, for specific conditions or at specific values of age) were obtained
by dummy-coding predictors with different reference levels (that is, using 0, 1
contrasts) or by centring age at different values and then refitting the model.
The primary model, as well as all further models, included a random
by-participant intercept. To further reduce the probability of committing false
positive errors, a ‘random slope’ for flanker type by participant (which captures the
variation of flanker effects across participants) was included in all models (except
in cases of non-convergence; see the caption for Extended Data Fig. 2 and notes
to Supplementary Tables 1 and 2), as it produced models with greater goodness
of fit (lower Akaike information criterion) than the intercept-only models210,211. A
random slope for cue type by participant was not included because models with
this random structure did not converge. Item as a random effect was not included
since all possible combinations of item types (cue types × target locations × central
arrow directions × flanker arrow directions) are included in the study design,
and thus items were not randomly sampled from a population. Assumptions of
additivity and normality as well as homogeneity of variances were broadly satisfied
in our models, as ascertained by visualizations of residuals212,213.
Reporting Summary. Further information on research design is available in the
Nature Research Reporting Summary linked to this article.
Data availability
The anonymized data (with accompanying documentation) have been uploaded to
the Open Science Framework. They can be found at https://osf.io/59er2/.
Code availability
Commented analysis scripts (in the R programming language) for all statistical
models reported in this paper have been uploaded to the Open Science Framework.
They can be found at https://osf.io/59er2/.
Received: 7 July 2020; Accepted: 22 June 2021;
Published: xx xx xxxx
References
1. Nyberg, L., Lövdén, M., Riklund, K., Lindenberger, U. & Bäckman, L.
Memory aging and brain maintenance. Trends Cogn. Sci. 16, 292–305 (2012).
2. Old, S. R. & Naveh-Benjamin, M. Dierential eects of age on item and
associative measures of memory: a meta-analysis. Psychol. Aging 23,
104–118 (2008).
3. Rönnlund, M., Nyberg, L., Bäckman, L. & Nilsson, L.-G. Stability, growth,
and decline in adult life span development of declarative memory:
cross-sectional and longitudinal data from a population-based study.
Psychol. Aging 20, 3–18 (2005).
4. Ratcli, R., apar, A., Gomez, P. & McKoon, G. A diusion model analysis
of the eects of aging in the lexical-decision task. Psychol. Aging 19,
278–289 (2004).
5. Wierenga, C. E. et al. Age-related changes in word retrieval: role of bilateral
frontal and subcortical networks. Neurobiol. Aging 29, 436–451 (2008).
6. Greve, A., Cooper, E. & Henson, R. N. No evidence that ‘fast-mapping’
benets novel learning in healthy older adults. Neuropsychologia 60,
52–59 (2014).
7. Whiting, E., Chenery, H. J. & Copland, D. A. Eect of aging on learning
new names and descriptions for objects. Aging Neuropsychol. Cogn. 18,
594–619 (2011).
8. Howard, D. V. et al. Implicit sequence learning: eects of level of structure,
adult age, and extended practice. Psychol. Aging 19, 79–92 (2004).
9. Howard, J. H. & Howard, D. V. Aging mind and brain: is implicit learning
spared in healthy aging?. Front. Psychol. 4, 817 (2013).
10. Shea, C. H., Park, J.-H. & Braden, H. W. Age-related eects in sequential
motor learning. Phys. er. 86, 478–488 (2006).
11. Bo, J., Borza, V. & Seidler, R. D. Age-related declines in visuospatial
working memory correlate with decits in explicit motor sequence learning.
J. Neurophysiol. 102, 2744–2754 (2009).
12. Iachini, T., Iavarone, A., Senese, V., Ruotolo, F. & Ruggiero, G. Visuospatial
memory in healthy elderly, AD and MCI: a review. Curr. Aging Sci. 2,
43–59 (2009).
13. Monge, Z. A. & Madden, D. J. Linking cognitive and visual perceptual
decline in healthy aging: the information degradation hypothesis. Neurosci.
Biobehav. Rev. 69, 166–173 (2016).
14. Naveh-Benjamin, M. & Kilb, A. Age-related dierences in associative
memory: the role of sensory decline. Psychol. Aging 29, 672–683 (2014).
15. Salthouse, T. A. e processing-speed theory of adult age dierences in
cognition. Psychol. Rev. 103, 403–428 (1996).
16. Jenkins, L., Myerson, J., Joerding, J. A. & Hale, S. Converging evidence that
visuospatial cognition is more age-sensitive than verbal cognition. Psychol.
Aging 15, 157–175 (2000).
17. Lindenberger, U., Scherer, H. & Baltes, P. B. e strong connection between
sensory and cognitive performance in old age: not due to sensory acuity
reductions operating during cognitive assessment. Psychol. Aging 16,
196–205 (2001).
18. Park, D. C. et al. Mediators of long-term memory performance across the
life span. Psychol. Aging 11, 621–637 (1996).
19. Verhaeghen, P. & Salthouse, T. A. Meta-analyses of age–cognition relations
in adulthood: estimates of linear and nonlinear age eects and structural
models. Psychol. Bull. 122, 231–249 (1997).
20. Cabeza, R., Nyberg, L. & Park, D. C. Cognitive Neuroscience of Aging:
Linking Cognitive and Cerebral Aging (Oxford Univ. Press, 2016).
21. Carstensen, L. L. et al. Emotional experience improves with age:
evidence based on over 10 years of experience sampling. Psychol. Aging 26,
21–33 (2011).
22. Schaie, K. W. & Willis, S. L. e Seattle Longitudinal Study of adult
cognitive development. ISSBD Bull. 2010(1), 24–29 (2010).
23. Anderson, N. D. & Craik, F. I. M. 50 years of cognitive aging theory.
J. Gerontol. B 72, 1–6 (2017).
24. Hartshorne, J. K. & Germine, L. T. When does cognitive functioning peak?
e asynchronous rise and fall of dierent cognitive abilities across the life
span. Psychol. Sci. 26, 433–443 (2015).
25. Hedden, T. & Gabrieli, J. D. E. Insights into the ageing mind: a view from
cognitive neuroscience. Nat. Rev. Neurosci. 5, 87–96 (2004).
26. Park, D. C. et al. Models of visuospatial and verbal memory across the adult
life span. Psychol. Aging 17, 299–320 (2002).
27. Piolino, P., Desgranges, B., Benali, K. & Eustache, F. Episodic and semantic
remote autobiographical memory in ageing. Memory 10, 239–257 (2002).
28. Schaie, K. W. Intellectual Development in Adulthood: e Seattle
Longitudinal Study (Cambridge Univ. Press, 1996).
29. Cansino, S. et al. e decline of verbal and visuospatial working memory
across the adult life span. Age (Dordr.) 35, 2283–2302 (2013).
30. Ikier, S., Yang, L. & Hasher, L. Implicit proactive interference, age,
and automatic versus controlled retrieval strategies. Psychol. Sci. 19,
456–461 (2008).
31. Jacoby, L. L. Ironic eects of repetition: measuring age-related dierences in
memory. J. Exp. Psychol. Learn. Mem. Cogn. 25, 3–22 (1999).
32. Laver, G. D. Adult aging eects on semantic and episodic priming in word
recognition. Psychol. Aging 24, 28–39 (2009).
33. Campbell, K. L. et al. Robust resilience of the frontotemporal syntax system
to aging. J. Neurosci. 36, 5214–5227 (2016).
34. Cohen-Shikora, E. R. & Balota, D. A. Visual word recognition across the
adult lifespan. Psychol. Aging 31, 488–502 (2016).
35. Reifegerste, J., Elin, K. & Clahsen, H. Persistent dierences between native
speakers and late bilinguals: evidence from inectional and derivational
processing in older speakers. Biling. Lang. Cogn. 22, 425–440 (2019).
36. Royle, P., Steinhauer, K., Dessureault, É., Herbay, A. C. & Brambati, S. M.
Aging and language: maintenance of morphological representations in older
adults. Front. Commun. 4, 16 (2019).
37. Shao, M. A. & Tyler, L. K. Language in the aging brain: the
network dynamics of cognitive decline and preservation. Science 346,
583–587 (2014).
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles
Nature HumaN BeHaviour
38. Cabeza, R. et al. Maintenance, reserve and compensation: the cognitive
neuroscience of healthy ageing. Nat. Rev. Neurosci. 19, 701–710 (2018).
39. Horton, S., Baker, J. & Schorer, J. Expertise and aging: maintaining skills
through the lifespan. Eur. Rev. Aging Phys. Act. 5, 89–96 (2008).
40. Horn, J. L. in Aging and Cognitive Processes (eds Craik, F. I. M. & Trehub,
S.) 237–278 (Plenum, 1982).
41. Horn, J. L. & Cattell, R. B. Age dierences in uid and crystallized
intelligence. Acta Psychol. (Amst.) 26, 107–129 (1967).
42. Li, S.-C. in International Encyclopedia of the Social and Behavioral Sciences
(eds Smelser, N. J. & Baltes, P. B.) 310–317 (Elsevier, 2001).
43. Baltes, P. B. & Kliegl, R. in Neurology (eds. Poeck, K. et al.) 1–17 (Springer,
1986).
44. Ben-David, B. M., Erel, H., Goy, H. & Schneider, B. A. “Older is always
better”: age-related dierences in vocabulary scores across 16 years. Psychol.
Aging 30, 856–862 (2015).
45. Cattell, R. B. Abilities: eir Structure, Growth, and Action (Houghton
Miin, 1971).
46. Goral, M., Spiro, A., Albert, M. L., Obler, L. K. & Connor, L. T. Change in
lexical retrieval skills in adulthood. Ment. Lex. 2, 215–238 (2007).
47. Horn, J. L. & Donaldson, G. On the myth of intellectual decline in
adulthood. Am. Psychol. 31, 701–719 (1976).
48. Spreng, R. N. & Turner, G. R. e shiing architecture of cognition and
brain function in older adulthood. Perspect. Psychol. Sci. 14, 523–542 (2019).
49. Verhaeghen, P. Aging and vocabulary score: a meta-analysis. Psychol. Aging
18, 332–339 (2003).
50. Baltes, P. B. & Smith, J. e fascination of wisdom: its nature, ontogeny, and
function. Perspect. Psychol. Sci. 3, 56–64 (2008).
51. Lim, K. T. K. & Yu, R. Aging and wisdom: age-related changes in economic
and social decision making. Front. Aging Neurosci. 7, 120 (2015).
52. Happé, F. G. E., Winner, E. & Brownell, H. e getting of wisdom: theory
of mind in old age. Dev. Psychol. 34, 358–362 (1998).
53. Burr, D. A., Castrellon, J. J., Zald, D. H. & Samanez-Larkin, G. R. Emotion
dynamics across adulthood in everyday life: older adults are more
emotionally stable and better at regulating desires. Emotion https://doi.
org/10/gg3kg2 (2020).
54. Carstensen, L. L., Pasupathi, M., Mayr, U. & Nesselroade, J. R. Emotional
experience in everyday life across the adult life span. J. Pers. Soc. Psychol.
79, 644–655 (2000).
55. Carstensen, L. L., Fung, H. H. & Charles, S. T. Socioemotional selectivity
theory and the regulation of emotion in the second half of life. Motiv. Emot.
27, 103–123 (2003).
56. Urry, H. L. & Gross, J. J. Emotion regulation in older age. Curr. Dir. Psychol.
Sci. 19, 352–357 (2010).
57. Li, Y., Baldassi, M., Johnson, E. J. & Weber, E. U. Complementary cognitive
capabilities, economic decision making, and aging. Psychol. Aging 28,
595–613 (2013).
58. Mata, R., Schooler, L. J. & Rieskamp, J. e aging decision maker: cognitive
aging and the adaptive selection of decision strategies. Psychol. Aging 22,
796–810 (2007).
59. Strough, J., Schlosnagle, L. & DiDonato, L. Understanding decisions about
sunk costs from older and younger adults’ perspectives. J. Gerontol. B 66B,
681–686 (2011).
60. Strough, J., Mehta, C. M., McFall, J. P. & Schuller, K. L. Are older adults less
subject to the sunk-cost fallacy than younger adults? Psychol. Sci. 19,
650–652 (2008).
61. Kanfer, R., Beier, M. E. & Ackerman, P. L. Goals and motivation related to
work in later adulthood: an organizing framework. Eur. J. Work Organ.
Psychol. 22, 253–264 (2013).
62. Kooij, D. T. A. M., De Lange, A. H., Jansen, P. G. W., Kanfer, R. & Dikkers,
J. S. E. Age and work-related motives: results of a meta-analysis. J. Organ.
Behav. 32, 197–225 (2011).
63. Donnellan, M. B. & Lucas, R. E. Age dierences in the big ve across
the life span: evidence from two national samples. Psychol. Aging 23,
558–566 (2008).
64. Ibanez, A. et al. Empathy, sex and uid intelligence as predictors of theory
of mind. Pers. Individ. Dier. 54, 616–621 (2013).
65. Opitz, P. C., Lee, I. A., Gross, J. J. & Urry, H. L. Fluid cognitive ability is a
resource for successful emotion regulation in older and younger adults.
Front. Psychol. 5, 609 (2014).
66. Kanfer, R. & Ackerman, P. L. Aging, adult development, and work
motivation. Acad. Manage. Rev. 29, 440–458 (2004).
67. Lövdén, M., Ghisletta, P. & Lindenberger, U. Cognition in the Berlin
Aging Study (BASE): the rst 10 years. Aging Neuropsychol. Cogn. 11,
104–133 (2004).
68. Ng, T. W. H. & Feldman, D. C. e relationship of age to ten dimensions of
job performance. J. Appl. Psychol. 93, 392–423 (2008).
69. Ardelt, M. Wisdom as expert knowledge system: a critical review of a
contemporary operationalization of an ancient concept. Hum. Dev. 47,
257–285 (2004).
70. Buchler, N. E. G. & Reder, L. M. Modeling age-related memory decits: a
two-parameter solution. Psychol. Aging 22, 104–121 (2007).
71. Ramscar, M., Hendrix, P., Shaoul, C., Milin, P. & Baayen, R. H. e myth of
cognitive decline: non-linear dynamics of lifelong learning. Top. Cogn. Sci.
6, 5–42 (2014).
72. Ramscar, M., Sun, C. C., Hendrix, P. & Baayen, R. H. e mismeasurement
of mind: life-span changes in paired-associate-learning scores reect the
‘cost’ of learning, not cognitive decline. Psychol. Sci. 28, 1171–1179 (2017).
73. Reifegerste, J. et al. Early-life education may help bolster declarative
memory in old age, especially for women. Aging Neuropsychol. Cogn. 28,
218–252 (2021).
74. Stine-Morrow, E. A. L. e Dumbledore hypothesis of cognitive aging.
Curr. Dir. Psychol. Sci. 16, 295–299 (2007).
75. Umanath, S. & Marsh, E. J. Understanding how prior knowledge inuences
memory in older adults. Perspect. Psychol. Sci. 9, 408–426 (2014).
76. Zaval, L., Li, Y., Johnson, E. J. & Weber, E. U. in Aging and Decision Making
(eds Hess, T. M. et al.) 149–168 (Elsevier, 2015).
77. Fernandez-Duque, D. & Posner, M. I. Relating the mechanisms of orienting
and alerting. Neuropsychologia 35, 477–486 (1997).
78. Miyake, A. et al. e unity and diversity of executive functions and their
contributions to complex ‘frontal lobe’ tasks: a latent variable analysis.
Cogn. Psychol. 41, 49–100 (2000).
79. Zanto, T. P. & Gazzaley, A. in e Oxford Handbook of Attention (eds
Nobre, A. C. & Kastner, S.) 927–971 (Oxford Univ. Press, 2014).
80. Kinsella, G., Storey, E. & Crawford, J. R. in Neurology and Clinical
Neuroscience (ed. Schapira, A. H. V.) 83–95 (Elsevier, 1998).
81. Vaughan, L. & Giovanello, K. Executive function in daily life: age-related
inuences of executive processes on instrumental activities of daily living.
Psychol. Aging 25, 343–355 (2010).
82. Del Missier, F., Mäntylä, T. & Bruine de Bruin, W. Executive functions in
decision making: an individual dierences approach. ink. Reason. 16,
69–97 (2010).
83. García-Madruga, J. A., Gómez-Veiga, I. & Vila, J. Ó. Executive functions
and the improvement of thinking abilities: the intervention in reading
comprehension. Front. Psychol. 7, 58 (2016).
84. Martin, R. & Allen, C. A disorder of executive function and its role in
language processing. Semin. Speech Lang. 29, 201–210 (2008).
85. Mazuka, R., Jincho, N. & Oishi, H. Development of executive control and
language processing. Lang. Linguist. Compass 3, 59–89 (2009).
86. Schiebener, J. et al. Among three dierent executive functions, general
executive control ability is a key predictor of decision making under
objective risk. Front. Psychol. 5, 1386 (2014).
87. Braver, T. S. & Barch, D. M. A theory of cognitive control, aging cognition,
and neuromodulation. Neurosci. Biobehav. Rev. 26, 809–817 (2002).
88. Buckner, R. L. Memory and executive function in aging and AD. Neuron
44, 195–208 (2004).
89. Fjell, A. M., Sneve, M. H., Grydeland, H., Storsve, A. B. & Walhovd, K. B.
e disconnected brain and executive function decline in aging. Cereb.
Cortex 27, 2303–2317 (2016).
90. Hasher, L., Lustig, C. & Zacks, R. T. in Variation in Working Memory (eds
Conway, A. R. A. et al.) 227–249 (Oxford Univ. Press, 2007).
91. Hasher, L. & Zacks, R. T. in e Psychology of Learning and Motivation (ed.
Bower, G. H.) Vol. 22, 193–225 (Academic, 1988).
92. Hasher, L., Stoltzfus, E. R., Zacks, R. T. & Rypma, B. Age and inhibition. J.
Exp. Psychol. Learn. Mem. Cogn. 17, 163–169 (1991).
93. MacPherson, S. E., Phillips, L. H. & Della Sala, S. Age, executive function
and social decision making: a dorsolateral prefrontal theory of cognitive
aging. Psychol. Aging 17, 598–609 (2002).
94. Posner, M. I., Rothbart, M. K. & Ghassemzadeh, H. Restoring attention
networks. Yale J. Biol. Med. 92, 139–143 (2019).
95. West, R. L. An application of prefrontal cortex function theory to cognitive
aging. Psychol. Bull. 120, 272–292 (1996).
96. West, R. L. In defense of the frontal lobe hypothesis of cognitive aging. J.
Int. Neuropsychol. Soc. 6, 727–729 (2000).
97. Goh, J. O., Beason-Held, L. L., An, Y., Kraut, M. A. & Resnick, S. M.
Frontal function and executive processing in older adults: process and
region specic age-related longitudinal functional changes. NeuroImage 69,
43–50 (2013).
98. Kennedy, K. M. & Raz, N. Aging white matter and cognition: dierential
eects of regional variations in diusion properties on memory, executive
functions, and speed. Neuropsychologia 47, 916–927 (2009).
99. Mather, M. & Harley, C. W. e locus coeruleus: essential for maintaining
cognitive function and the aging brain. Trends Cogn. Sci. 20, 214–226 (2016).
100. Pardo, J. V. et al. Where the brain grows old: decline in anterior cingulate
and medial prefrontal function with normal aging. NeuroImage 35,
1231–1237 (2007).
101. Raz, N., Ghisletta, P., Rodrigue, K. M., Kennedy, K. M. & Lindenberger, U.
Trajectories of brain aging in middle-aged and older adults: regional and
individual dierences. NeuroImage 51, 501–511 (2010).
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles Nature HumaN BeHaviour
102. Raz, N. & Rodrigue, K. M. Dierential aging of the brain: patterns,
cognitive correlates and modiers. Neurosci. Biobehav. Rev. 30,
730–748 (2006).
103. Bopp, K. L. & Verhaeghen, P. Aging and verbal memory span: a
meta-analysis. J. Gerontol. B. 60, P223–P233 (2005).
104. Bopp, K. L. & Verhaeghen, P. Aging and n-back performance: a
meta-analysis. J. Gerontol. B https://doi.org/10/ggjwwt (2020)
105. Burke, D. M. & Osborne, G. in Inhibition in Cognition (eds Gorfein, D. S. &
MacLeod, C. M.) 163–183 (American Psychological Association, 2007).
106. Glisky, E. L. in Brain Aging: Models, Methods, and Mechanisms (ed. Riddle,
D. R.) Ch. 1 (Taylor & Francis, 2007).
107. McDonough, I. M., Wood, M. M. & Miller, W. S. A review on the trajectory
of attentional mechanisms in aging and the Alzheimer’s disease continuum
through the Attention Network Test. Yale J. Biol. Med. 92, 37–51 (2019).
108. McDowd, J. M. & Shaw, R. J. in e Handbook of Aging and Cognition (eds
Craik, F. I. M. & Salthouse, T. A.) 221–292 (Lawrence Erlbaum, 2000).
109. Pliatsikas, C. et al. Working memory in older adults declines with age,
but is modulated by sex and education. Q. J. Exp. Psychol. 72,
1308–1327 (2019).
110. Rey-Mermet, A. & Gade, M. Inhibition in aging: what is preserved? What
declines? A meta-analysis. Psychon. Bull. Rev. 25, 1695–1716 (2018).
111. Verhaeghen, P. Aging and executive control: reports of a demise greatly
exaggerated. Curr. Dir. Psychol. Sci. 20, 174–180 (2011).
112. Verhaeghen, P. e Elements of Cognitive Aging: Meta-analyses of
Age-Related Dierences in Processing Speed and eir Consequences (Oxford
Univ. Press, 2013).
113. Verhaeghen, P. & Cerella, J. Aging, executive control, and attention: a
review of meta-analyses. Neurosci. Biobehav. Rev. 26, 849–857 (2002).
114. Friedman, N. P. & Miyake, A. Unity and diversity of executive functions:
individual dierences as a window on cognitive structure. Cortex 86,
186–204 (2017).
115. Karr, J. E. et al. e unity and diversity of executive functions: a systematic
review and re-analysis of latent variable studies. Psychol. Bull. 144,
1147–1185 (2018).
116. Miyake, A. & Friedman, N. P. e nature and organization of individual
dierences in executive functions: four general conclusions. Curr. Dir.
Psychol. Sci. 21, 8–14 (2012).
117. Petersen, S. E. & Posner, M. I. e attention system of the human brain: 20
years aer. Annu. Rev. Neurosci. 35, 73–89 (2012).
118. Salthouse, T. A. Is anker-based inhibition related to age? Identifying
specic inuences of individual dierences on neurocognitive variables.
Brain Cogn. 73, 51–61 (2010).
119. Uttl, B. & Graf, P. Color–Word Stroop test performance across the adult life
span. J. Clin. Exp. Neuropsychol. 19, 405–420 (1997).
120. Posner, M. I. & Petersen, S. E. e attention system of the human brain.
Annu. Rev. Neurosci. 13, 25–42 (1990).
121. Staub, B., Doignon-Camus, N., Després, O. & Bonnefond, A. Sustained
attention in the elderly: what do we know and what does it tell us about
cognitive aging? Ageing Res. Rev. 12, 459–468 (2013).
122. Fan, J., McCandliss, B. D., Sommer, T., Raz, A. & Posner, M. I. Testing the
eciency and independence of attentional networks. J. Cogn. Neurosci. 14,
340–347 (2002).
123. Fan, J. et al. Testing the behavioral interaction and integration of attentional
networks. Brain Cogn. 70, 209–220 (2009).
124. Posner, M. I., Sheese, B. E., Odludaş, Y. & Tang, Y. Analyzing and shaping
human attentional networks. Neural Netw. 19, 1422–1429 (2006).
125. Wang, H., Fan, J. & Johnson, T. R. A symbolic model of human attentional
networks. Cogn. Syst. Res. 5, 119–134 (2004).
126. Fan, J., Fossella, J., Sommer, T., Wu, Y. & Posner, M. I. Mapping the genetic
variation of executive attention onto brain activity. Proc. Natl Acad. Sci.
USA 100, 7406–7411 (2003).
127. Fan, J., Mccandliss, B., Fossella, J., Flombaum, J. & Posner, M. I. e
activation of attentional networks. NeuroImage 26, 471–479 (2005).
128. Konrad, K., Neufang, S., Hanisch, C., Fink, G. R. & Herpertz-Dahlmann, B.
Dysfunctional attentional networks in children with attention decit/
hyperactivity disorder: evidence from an event-related functional magnetic
resonance imaging study. Biol. Psychiatry 59, 643–651 (2006).
129. Niogi, S., Mukherjee, P., Ghajar, J. & McCandliss, B. D. Individual
dierences in distinct components of attention are linked to anatomical
variations in distinct white matter tracts. Front. Neuroanat. https://doi.
org/10/cghhr4 (2010).
130. Beane, M. & Marrocco, R. T. in Cognitive Neuroscience of Attention
(ed. Posner, M. I.) 313–325 (Guilford, 2004).
131. Posner, M. I. & Rothbart, M. K. Research on attention networks as a
model for the integration of psychological science. Annu. Rev. Psychol. 58,
1–23 (2007).
132. Witte, E. A., Davidson, M. C. & Marrocco, R. T. Eects of altering brain
cholinergic activity on covert orienting of attention: comparison of monkey
and human performance. Psychopharmacol. (Berl.) 132, 324–334 (1997).
133. Witte, E. A. & Marrocco, R. T. Alteration of brain noradrenergic activity in
rhesus monkeys aects the alerting component of covert orienting.
Psychopharmacol. (Berl.) 132, 315–323 (1997).
134. Fossella, J. et al. Assessing the molecular genetics of attention networks.
BMC Neurosci. 3, 14 (2002).
135. Pozuelos, J. P., Paz-Alonso, P. M., Castillo, A., Fuentes, L. J. & Rueda, M. R.
Development of attention networks and their interactions in childhood.
Dev. Psychol. 50, 2405–2415 (2014).
136. Posner, M. I. Orienting of attention. Q. J. Exp. Psychol. 32, 3–25 (1980).
137. Eriksen, B. A. & Eriksen, C. W. Eects of noise letters upon the
identication of a target letter in a nonsearch task. Percept. Psychophys. 16,
143–149 (1974).
138. Ishigami, Y. et al. e Attention Network Test-Interaction (ANT-I):
reliability and validity in healthy older adults. Exp. Brain Res. 234,
815–827 (2016).
139. Ishigami, Y. & Klein, R. M. Repeated measurement of the components of
attention of older adults using the two versions of the Attention Network
Test: stability, isolability, robustness, and reliability. Front. Aging Neurosci. 3,
17 (2011).
140. Mahoney, J. R., Verghese, J., Goldin, Y., Lipton, R. & Holtzer, R. Alerting,
orienting, and executive attention in older adults. J. Int. Neuropsychol. Soc.
16, 877–889 (2010).
141. MacLeod, J. W. et al. Appraising the ANT: psychometric and theoretical
considerations of the Attention Network Test. Neuropsychology 24,
637–651 (2010).
142. Dash, T., Berroir, P., Joanette, Y. & Ansaldo, A. I. Alerting, orienting, and
executive control: the eect of bilingualism and age on the subcomponents
of attention. Front. Neurol. 10, 1122 (2019).
143. Deiber, M.-P., Ibañez, V., Missonnier, P., Rodriguez, C. & Giannakopoulos,
P. Age-associated modulations of cerebral oscillatory patterns related to
attention control. NeuroImage 82, 531–546 (2013).
144. Fernandez-Duque, D. & Black, S. E. Attentional networks in normal aging
and Alzheimer’s disease. Neuropsychology 20, 133–143 (2006).
145. Gamboz, N., Zamarian, S. & Cavallero, C. Age-related dierences in the
Attention Network Test (ANT). Exp. Aging Res. 36, 287–305 (2010).
146. Jennings, J. M., Dagenbach, D., Engle, C. M. & Funke, L. J. Age-related
changes and the Attention Network Task: an examination of alerting,
orienting, and executive function. Aging Neuropsychol. Cogn. 14,
353–369 (2007).
147. Kaufman, D. A. S., Sozda, C. N., Dotson, V. M. & Perlstein, W. M. An
event-related potential investigation of the eects of age on alerting,
orienting, and executive function. Front. Aging Neurosci. 8, 99 (2016).
148. Knight, M. & Mather, M. Look out—it’s your o-peak time of day! Time of
day matters more for alerting than for orienting or executive attention. Exp.
Aging Res. 39, 305–321 (2013).
149. Lu, H., Fung, A. W. T., Chan, S. S. M. & Lam, L. C. W. Disturbance of
attention network functions in Chinese healthy older adults: an
intra-individual perspective. Int. Psychogeriatr. 28, 291–301 (2016).
150. Westlye, L. T., Grydeland, H., Walhovd, K. B. & Fjell, A. M. Associations
between regional cortical thickness and attentional networks as measured
by the Attention Network Test. Cereb. Cortex 21, 345–356 (2011).
151. Williams, R. S. et al. Age dierences in the attention network test: evidence
from behavior and event-related potentials. Brain Cogn. 102, 65–79 (2016).
152. Young-Bernier, M., Tanguay, A. N., Tremblay, F. & Davidson, P. S. R. Age
dierences in reaction times and a neurophysiological marker of cholinergic
activity. Can. J. Aging 34, 471–480 (2015).
153. Zhou, S., Fan, J., Lee, T. M. C., Wang, C. & Wang, K. Age-related
dierences in attentional networks of alerting and executive control in
young, middle-aged, and older Chinese adults. Brain Cogn. 75,
205–210 (2011).
154. Baayen, R. H. A real experiment is a factorial experiment? Ment. Lex. 5,
149–157 (2010).
155. Jacobsen, G. M., de Mello, C. M., Kochhann, R. & Fonseca, R. P. Executive
functions in school-age children: inuence of age, gender, school type and
parental education. Appl. Cogn. Psychol. 31, 404–413 (2017).
156. Kamkar, N. H. & Morton, J. B. CanDiD: a framework for linking executive
function and education. Front. Psychol. 8, 1187 (2017).
157. Liu, G., Hu, P., Fan, J. & Wang, K. Gender dierences associated with
orienting attentional networks in healthy subjects. Chin. Med. J. (Engl.) 126,
2308–2312 (2013).
158. Faust, M. E., Balota, D. A., Spieler, D. H. & Ferraro, F. R. Individual
dierences in information-processing rate and amount: implications for
group dierences in response latency. Psychol. Bull. 125, 777–799 (1999).
159. Friedman, L. & Wall, M. Graphical views of suppression and
multicollinearity in multiple linear regression. Am. Stat. 59, 127–136 (2005).
160. Wurm, L. H. & Fisicaro, S. A. What residualizing predictors in regression
analyses does (and what it does not do). J. Mem. Lang. 72, 37–48 (2014).
161. Gittings, N. S. & Fozard, J. L. Age related changes in visual acuity. Exp.
Gerontol. 21, 423–433 (1986).
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles
Nature HumaN BeHaviour
162. Costa, A., Hernández, M. & Sebastián-Gallés, N. Bilingualism aids conict
resolution: evidence from the ANT task. Cognition 106, 59–86 (2008).
163. Rey-Mermet, A., Gade, M. & Oberauer, K. Should we stop thinking about
inhibition? Searching for individual and age dierences in inhibition ability.
J. Exp. Psychol. Learn. Mem. Cogn. 44, 501–526 (2018).
164. Sekuler, A. B., Bennett, P. J. & Mamelak, M. Eects of aging on the useful
eld of view. Exp. Aging Res. 26, 103–120 (2000).
165. Baayen, R. H. Analyzing Linguistic Data: A Practical Introduction to
Statistics Using R (Cambridge Univ. Press, 2008).
166. Neter, J., Kutner, M., Nachtsheim, C. & Wasserman, W. Applied Linear
Statistical Models (Irwin, 1996).
167. Bosch, S., Veríssimo, J. & Clahsen, H. Inectional morphology in bilingual
language processing: an age-of-acquisition study. Lang. Acquis. 26, 339–360
(2019).
168. Veríssimo, J., Heyer, V., Jacob, G. & Clahsen, H. Selective eects of age of
acquisition on morphological priming: evidence for a sensitive period.
Lang. Acquis. 25, 315–326 (2018).
169. Hassing, L., Wahlin, Å. & Bäckman, L. Minimal inuence of age, education,
and gender on episodic memory functioning in very old age: a population-
based study of nonagenarians. Arch. Gerontol. Geriatr. 27, 75–87 (1998).
170. Lindenberger, U., Singer, T. & Baltes, P. B. Longitudinal selectivity in aging
populations: separating mortality-associated versus experimental
components in the Berlin Aging Study (BASE). J. Gerontol. B. 57,
P474–P482 (2002).
171. Singer, T., Verhaeghen, P., Ghisletta, P., Lindenberger, U. & Baltes, P. B. e
fate of cognition in very old age: six-year longitudinal ndings in the Berlin
Aging Study (BASE). Psychol. Aging 18, 318–331 (2003).
172. Chen, Y. et al. Testing a cognitive control model of human intelligence.
Sci. Rep. 9, 2898 (2019).
173. Madden, D. J. & Gottlob, L. R. Adult age dierences in strategic and
dynamic components of focusing visual attention. Aging Neuropsychol.
Cogn. 4, 185–210 (1997).
174. Mathewson, K. J., Dywan, J. & Segalowitz, S. J. Brain bases of error-related
ERPs as inuenced by age and task. Biol. Psychol. 70, 88–104 (2005).
175. Pettigrew, C. & Martin, R. C. Cognitive declines in healthy aging: evidence
from multiple aspects of interference resolution. Psychol. Aging 29,
187–204 (2014).
176. Sullivan, M. P. e functional interaction of visual-perceptual and response
mechanisms during selective attention in young adults, young-old adults,
and old-old adults. Percept. Psychophys. 61, 810–825 (1999).
177. Wright, L. L. & Elias, J. W. Age dierences in the eects of perceptual noise.
J. Gerontol. 34, 704–708 (1979).
178. Friedman, N. P. & Miyake, A. e relations among inhibition and
interference control functions: a latent-variable analysis. J. Exp. Psychol.
Gen. 133, 101–135 (2004).
179. Nigg, J. T. On inhibition/disinhibition in developmental psychopathology:
views from cognitive and personality psychology and a working inhibition
taxonomy. Psychol. Bull. 126, 220–246 (2000).
180. Rouder, J. N. & Haaf, J. M. A psychometrics of individual dierences in
experimental tasks. Psychon. Bull. Rev. 26, 452–467 (2019).
181. Stahl, C. et al. Behavioral components of impulsivity. J. Exp. Psychol. Gen.
143, 850–886 (2014).
182. Borella, E., Delaloye, C., Lecerf, T., Renaud, O. & de Ribaupierre, A. Do age
dierences between young and older adults in inhibitory tasks depend on
the degree of activation of information? Eur. J. Cogn. Psychol. 21,
445–472 (2009).
183. Hsieh, S., Liang, Y.-C. & Tsai, Y.-C. Do age-related changes contribute to
the anker eect? Clin. Neurophysiol. 123, 960–972 (2012).
184. Kawai, N., Kubo-Kawai, N., Kubo, K., Terazawa, T. & Masataka, N. Distinct
aging eects for two types of inhibition in older adults: a near-infrared
spectroscopy study on the Simon task and the anker task. NeuroReport 23,
819–824 (2012).
185. Waszak, F., Li, S.-C. & Hommel, B. e development of attentional
networks: cross-sectional ndings from a life span sample. Dev. Psychol. 46,
337–349 (2010).
186. Wild-Wall, N., Falkenstein, M. & Hohnsbein, J. Flanker interference in
young and older participants as reected in event-related potentials. Brain
Res. 1211, 72–84 (2008).
187. Zeef, E. J. & Kok, A. Age-related dierences in the timing of stimulus and
response processes during visual selective attention: performance and
psychophysiological analyses. Psychophysiology 30, 138–151 (1993).
188. Lambrechts, A., Karolis, V., Garcia, S., Obende, J. & Cappelletti, M. Age
does not count: resilience of quantity processing in healthy ageing. Front.
Psychol. 4, 865 (2013).
189. Ioannidis, J. P. A. Issues in comparisons between meta-analyses and large
trials. JAMA 279, 1089 (1998).
190. Slavin, R. & Smith, D. e relationship between sample sizes and eect
sizes in systematic reviews in education. Educ. Eval. Policy Anal. 31,
500–506 (2009).
191. Sterne, J. A. C., Gavaghan, D. & Egger, M. Publication and related bias in
meta-analysis. J. Clin. Epidemiol. 53, 1119–1129 (2000).
192. Vadillo, M. A. Ego depletion may disappear by 2020. Soc. Psychol. 50,
282–291 (2019).
193. Park, D. C. & McDonough, I. M. e dynamic aging mind: revelations from
functional neuroimaging research. Perspect. Psychol. Sci. 8, 62–67 (2013).
194. Reuter-Lorenz, P. A. & Cappell, K. A. Neurocognitive aging and the
compensation hypothesis. Curr. Dir. Psychol. Sci. 17, 177–182 (2008).
195. Staudinger, U. M., Cornelius, S. W. & Baltes, P. B. e aging of intelligence:
potential and limits. Ann. Am. Acad. Pol. Soc. Sci. 503, 43–59 (1989).
196. Salthouse, T. A. Selective review of cognitive aging. J. Int. Neuropsychol. Soc.
16, 754–760 (2010).
197. Homan, L., Hofer, S. M. & Sliwinski, M. J. On the confounds among retest
gains and age-cohort dierences in the estimation of within-person change in
longitudinal studies: a simulation study. Psychol. Aging 26, 778–791 (2011).
198. Salthouse, T. A. All data collection and analysis methods have limitations:
reply to Rabbitt (2011) and Raz and Lindenberger (2011). Psychol. Bull.
137, 796–799 (2011).
199. Wul, D. U., De Deyne, S., Jones, M. N., Mata, R. & e Aging Lexicon
Consortium. New perspectives on the aging lexicon. Trends Cogn. Sci.
https://doi.org/10/gf4x6z (2019).
200. Cornman, J. C. et al. Cohort prole: the Social Environment and Biomarkers
of Aging Study (SEBAS) in Taiwan. Int. J. Epidemiol. 45, 54–63 (2016).
201. Weinstein, M. et al. Social Environment and Biomarkers of Aging Study
(SEBAS) in Taiwan, 2000 and 2006 (Inter-university Consortium for
Political and Social Research, 2014).
202. Haaf, J. M. & Rouder, J. N. Some do and some don’t? Accounting for
variability of individual dierence structures. Psychon. Bull. Rev. 26,
772–789 (2019).
203. Rouder, J. N., Kumar, A. & Haaf, J. M. Why most studies of individual
dierences with inhibition tasks are bound to fail. Preprint at https://doi.
org/10.31234/osf.io/3cjr5 (2019).
204. Ratcli, R. Methods for dealing with reaction time outliers. Psychol. Bull.
114, 510–532 (1993).
205. Baayen, R. H. & Milin, P. Analyzing reaction times. Int. J. Psychol. Res. 3,
12–28 (2010).
206. Kliegl, R., Masson, M. E. J. & Richter, E. M. A linear mixed model analysis
of masked repetition priming. Vis. Cogn. 18, 655–681 (2010).
207. Cerella, J., Poon, L. W. & Williams, D. M. in Aging in the 1980s:
Psychological Issues (ed. Poon, L. W.) 332–340 (American Psychological
Association, 1980).
208. Bates, D., Mächler, M., Bolker, B. & Walker, S. Fitting linear mixed-eects
models using lme4. J. Stat. Sow. 67, 1–48 (2015).
209. Schad, D. J., Vasishth, S., Hohenstein, S. & Kliegl, R. How to capitalize on
a priori contrasts in linear (mixed) models: a tutorial. J. Mem. Lang. 110,
104038 (2020).
210. Barr, D. J., Levy, R., Scheepers, C. & Tily, H. J. Random eects structure for
conrmatory hypothesis testing: keep it maximal. J. Mem. Lang. 68,
255–278 (2013).
211. Matuschek, H., Kliegl, R., Vasishth, S., Baayen, R. H. & Bates, D. Balancing
type I error and power in linear mixed models. J. Mem. Lang. 94,
305–315 (2017).
212. Fox, J. Applied Regression Analysis and Generalized Linear Models
(SAGE, 2015).
213. Zuur, A. F., Ieno, E. N., Walker, N., Saveliev, A. A. & Smith, G. M. Mixed
Eects Models and Extensions in Ecology with R (Springer, 2009).
214. Bäckman, L., Nyberg, L., Lindenberger, U., Li, S.-C. & Farde, L. e
correlative triad among aging, dopamine, and cognition: current status and
future prospects. Neurosci. Biobehav. Rev. 30, 791–807 (2006).
215. Mann, D. M. A. e locus coeruleus and its possible role in ageing and
degenerative disease of the human central nervous system. Mech. Ageing
Dev. 23, 73–94 (1983).
216. Mather, M. in e Cognitive Neurosciences (eds Poeppel, D. et al.) 91–104
(MIT Press, 2019).
217. Schliebs, R. & Arendt, T. e cholinergic system in aging and neuronal
degeneration. Behav. Brain Res. 221, 555–563 (2011).
218. Walhovd, K. B. et al. Consistent neuroanatomical age-related volume
dierences across multiple samples. Neurobiol. Aging 32, 916–932 (2011).
219. Braskie, M. N. et al. Relationship of striatal dopamine synthesis capacity to
age and cognition. J. Neurosci. 28, 14320–14328 (2008).
220. Karrer, T. M., Josef, A. K., Mata, R., Morris, E. D. & Samanez-Larkin, G. R.
Reduced dopamine receptors and transporters but not synthesis capacity in
normal aging adults: a meta-analysis. Neurobiol. Aging 57, 36–46 (2017).
221. Kelley, T. A. & Yantis, S. Learning to attend: eects of practice on
information selection. J. Vis. 9(7), 16 (2009).
222. Rueda, M. R., Rothbart, M. K., McCandliss, B. D., Saccomanno, L. &
Posner, M. I. Training, maturation, and genetic inuences on the
development of executive attention. Proc. Natl Acad. Sci. USA 102,
14931–14936 (2005).
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles Nature HumaN BeHaviour
223. Rueda, M. R., Checa, P. & Cómbita, L. M. Enhanced eciency of the
executive attention network aer training in preschool children:
immediate changes and eects aer two months. Dev. Cogn. Neurosci. 2,
S192–S204 (2012).
224. Bialystok, E., Craik, F. I. M., Klein, R. & Viswanathan, M. Bilingualism,
aging, and cognitive control: evidence from the Simon task. Psychol. Aging
19, 290–303 (2004).
225. Bialystok, E., Martin, M. M. & Viswanathan, M. Bilingualism across
the lifespan: the rise and fall of inhibitory control. Int. J. Biling. 9,
103–119 (2005).
226. Bialystok, E., Craik, F. I. M. & Luk, G. Bilingualism: consequences for mind
and brain. Trends Cogn. Sci. 16, 240–250 (2012).
227. Costa, A., Hernández, M., Costa-Faidella, J. & Sebastián-Gallés, N. On the
bilingual advantage in conict processing: now you see it, now you don’t.
Cognition 113, 135–149 (2009).
228. Hayakawa, S. & Marian, V. Consequences of multilingualism for neural
architecture. Behav. Brain Funct. 15, 6 (2019).
229. Pelham, S. D. & Abrams, L. Cognitive advantages and disadvantages
in early and late bilinguals. J. Exp. Psychol. Learn. Mem. Cogn. 40,
313–325 (2014).
230. Schroeder, S. R., Marian, V., Shook, A. & Bartolotti, J. Bilingualism
and musicianship enhance cognitive control. Neural Plast. 2016,
4058620 (2016).
231. Woumans, E., Ceuleers, E., Van der Linden, L., Szmalec, A. & Duyck, W.
Verbal and nonverbal cognitive control in bilinguals and interpreters. J. Exp.
Psychol. Learn. Mem. Cogn. 41, 1579–1586 (2015).
232. Todd, M. et al. Apolipoprotein E, cognitive function, and cognitive decline
among older Taiwanese adults. PLoS ONE 13, e0206118 (2018).
233. Chung, Y., Rabe-Hesketh, S., Dorie, V., Gelman, A. & Liu, J. A
nondegenerate penalized likelihood estimator for variance parameters in
multilevel models. Psychometrika 78, 685–709 (2013).
234. Baayen, R. H., Davidson, D. J. & Bates, D. M. Mixed-eects modeling
with crossed random eects for subjects and items. J. Mem. Lang. 59,
390–412 (2008).
235. Callejas, A., Lupiáñez, J. & Tudela, P. e three attentional networks: on
their independence and interactions. Brain Cogn. 54, 225–227 (2004).
Acknowledgements
This work was supported by the Deutsche Forschungsgemeinschaft (German Research
Foundation) Project ID 317633480 (SFB 1287, Project Q) (University of Potsdam, to J.V.);
NIH R01 AG016790 (Princeton University, to N.G.), with a subcontract to M.T.U. at
Georgetown University; NIH R01 AG016661 (Georgetown University, to M.W.); NSF
BCS 1940980 (Georgetown University, to M.T.U.); and the Graduate School of Arts and
Sciences, Georgetown University (to M.W.). The funders had no role in study design,
data collection and analysis, decision to publish or preparation of the manuscript.
We thank the Health Promotion Administration at the Ministry of Health in Taiwan for
their support of this project; M. Pullman and L. Babcock for task preparation and testing;
and M. Posner, M. Riesenhuber, M. Rugg, D. Fernandez-Duque and especially D. Glei for
helpful comments.
Author contributions
The study was conceived by M.T.U. and M.W. and designed by M.T.U., as part of the
larger SEBAS project led by N.G. and M.W. J.V. performed the data preparation and
analysis. J.V. and M.T.U. wrote the paper, with contributions from P.V. as well as N.G. and
M.W. All authors read and approved the final manuscript.
Competing interests
The authors declare no competing interests.
Additional information
Extended data is available for this paper at https://doi.org/10.1038/s41562-021-01169-7.
Supplementary information The online version contains supplementary material
available at https://doi.org/10.1038/s41562-021-01169-7.
Correspondence and requests for materials should be addressed to J.V. or M.T.U.
Peer review information Nature Human Behaviour thanks Gregory Samanez-Larkin and
the other, anonymous, reviewer(s) for their contribution to the peer review of this work.
Reprints and permissions information is available at www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
© The Author(s), under exclusive licence to Springer Nature Limited 2021, corrected
publication 2021
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles
Nature HumaN BeHaviour Articles
Nature HumaN BeHaviour
Extended Data Fig. 1 | Demographic and cognitive information for participants, presented in 5-year age brackets. This table is for informational purposes
only; we remind readers that all analyses were performed with age as a continuous variable. For each age bracket, the columns show first of all the sample
size (total, with number of females in parentheses) and the mean (and SD) of years of education. Thus, these two columns display the age distributions
of sex and education, the two variables covaried out in our analyses; for discussion of these distributions see refs. 73,109. The subsequent columns display
means (and SDs) of the previously published cognitive measures of working memory (n-back task; mean d’ over 1-back and 2-back109) and declarative
memory (recognition memory task; mean d’ over real and novel objects73) for this sample. Note that the sample sizes in each 5-year age bracket are
slightly different for the working memory scores (Supplementary Table 1 in ref. 109) and the declarative memory scores (Table 1 in Ref. 73) than for the data
in the present paper (for example, due to slightly different subsets of participants having valid performance measures in the respective tasks). For a more
general cognitive measure obtained in this sample, see Ref. 232. N: number of participants; NA: not available; SD: standard deviation.
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles Nature HumaN BeHaviour
Articles Nature HumaN BeHaviour
Extended Data Fig. 2 | Results from the mixed-effects logistic regression model on the accuracy of responses. We examined accuracy using generalized
linear mixed-effects regression with a logit link function. We analysed correct/incorrect responses produced prior to the timeout period (number of data
points: 49,980), for the same 702 participants as in the main analysis on RTs. The same fixed-effect predictors were included as in the main analysis.
Models included a random by-participant intercept only, because models with random slopes did not converge. Model convergence was reached only
by applying a weakly informative Bayesian prior on the effects233 (priors on fixed effects were normal distributions with mean=0 and SD = 3 for the
intercept, and mean=0 and SD = 0.4 for slopes). Effect sizes are reported as unstandardized estimates (b-values) in the logit scale with 95% confidence
intervals, together with z-values; p-values are reported as two-tailed, with exact values to three digits. The significant interaction between age and
the executive effect indicated that age was associated with increasing executive efficiency, parallel to the finding of increasing efficiency in the main
analysis on RTs. Follow-up analyses indicated that there was an interference cost on accuracy for incongruent flankers at the minimum age of 58 years
(back-transformed accuracy in percent correct for congruent, 99.70% vs. incongruent, 99.38%; b = -0.7302 [-0.9756, -0.4847], z = -5.83, p < .001), but
no significant difference between incongruent and congruent flankers at the maximum age of 98 years (back-transformed accuracy for congruent, 98.10%
vs. incongruent, 97.94%; b = -0.0825 [-0.4693, 0.3043], z = -0.42, p = .676). Likewise, there was no significant executive effect at age 90 (b = -0.2127
[-0.5150, 0.0897], z = -1.38, p = .168). The executive effect was almost nine times larger at the minimum than maximum age, as revealed by the regression
coefficients in the logit scale (b values: 58 years: -0.7302 vs. 98 years: -0.0825), and twice as large in back-transformed accuracy (0.32% vs. 0.16%).
Education effect: higher education was associated with higher accuracy (across all ages and all cues and flankers).
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles
Nature HumaN BeHaviour Articles
Nature HumaN BeHaviour
Extended Data Fig. 3 | Results from the linear mixed-effects regression model on log RTs. P-values were obtained from t-tests with 49,163 degrees of
freedom, calculated as the number of data points (that is, 49,176) minus the number of fixed effect estimates (that is, 13)234. Here and elsewhere, effect
sizes of linear mixed-effects models are reported as unstandardized estimates (b-values) with 95% confidence intervals, together with t-values; p-values
are reported as two-tailed, with exact values to three digits. Education effect: higher education was associated with faster responses (across all ages
and all cues and flankers). Trial effect: later trials were associated with faster responses. Follow-up analyses to the two network interactions (Alerting X
Executive, Orienting X Executive) were performed. Note that both of these interactions, and the general patterns found in their follow-up analyses, are
commonly reported for the ANT in both younger and older adults122,138,140,144–146,152 ,235. First, the alerting effect was significant in trials with congruent flankers
(b = 0.0127 [0.0070, 0.0184], t = 4.37, p < .001), but not in trials with incongruent flankers (b = -0.0032 [-0.0089, 0.0026], t = -1.08, p = .279). Second,
the orienting effect was larger in trials with incongruent flankers (b = 0.0227 [0.0169, 0.0284], t = 7.76, p < .001) than in those with congruent flankers
(b = 0.0094 [0.0037, 0.0151], t = 3.24, p = .001), but was significant in both. We also followed up on both interactions by examining the executive effect
in the different cue types involved in alerting and orienting. The executive effect was larger in trials with a central cue (b = 0.0777 [0.0713, 0.0841],
t = 23.79, p < .001) than in trials with no cue (b = 0.0618 [0.0554, 0.0682], t = 18.95, p < .001) (Alerting X Executive). The executive effect was also larger
in trials with a central cue (see just above) than in trials with a spatial cue (b = 0.0644 [0.0580, 0.0708], t = 19.75, p < .001) (Orienting X Executive). CI:
confidence interval.
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles Nature HumaN BeHaviour
Articles Nature HumaN BeHaviour
Extended Data Fig. 4 | Mean untransformed RTs for each condition, and mean attentional network effects, in milliseconds, presented in 5-year age
brackets. This table displays mean untransformed RTs, by flanker and cue condition (with SDs in parentheses), together with mean attentional network
effects (computed as differences between the mean untransformed RTs in each pair of relevant conditions for example, between the central and no cue
conditions for alerting), in 5-year age brackets. This table is for informational purposes only; we remind readers that all analyses were performed on
log-transformed RTs with age as a continuous variable. Congruent and incongruent flanker types are computed over all three cue types, and each cue type
is computed over congruent and incongruent flankers. RTs: response times; N: number of participants; SDs: standard deviations.
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles
Nature HumaN BeHaviour Articles
Nature HumaN BeHaviour
Extended Data Fig. 5 | Linear effects of age on the efficiencies of the three attentional networks, showing network effects for each of the 702
participants. The linear age effects are displayed for (a) the alerting network, (b) the orienting network, and (c) the executive network. For each network,
each data point reflects the difference between mean log-transformed RTs in the two relevant conditions (for example, between no cue and central cue
for alerting) for each participant. The y-axis ranges (maximum minus minimum) of the three panels are identical, while their numerical values differ;
specifically, because the executive effect is larger than the other two attentional effects, the numerical values for the y-axis in panel (c) are shifted
upwards.
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles Nature HumaN BeHaviour
Articles Nature HumaN BeHaviour
Extended Data Fig. 6 | Age effects on RTs shown separately for each of the cue and flanker conditions. Though age-related slowdowns were observed for
all conditions, the RT increases differed among the relevant conditions, that is, among the cue or flanker types. These interactions yielded the observed age
effects on efficiency for the three attentional networks. First (panel a), the RT increase with aging was greater for trials preceded by a central cue (solid
line; b = 0.0087 [0.0072, 0.0103], t = 11.12, p < .001) than for those with no cue (line with short dashes; b = 0.0080 [0.0064, 0.0095], t = 10.15, p < .001),
leading to the observed decrease of alerting efficiency with aging. (The reasons for the detrimental effect of the central cue relative to no cue at later
ages remain to be determined; perhaps once the alerting benefit has decreased past a certain point, processing costs associated with the presentation of
the cue become predominant.) Second (also in panel a), the RT increase with age was smaller for trials preceded by spatial cues (line with long dashes;
b = 0.0081 [0.0065, 0.0096], t = 10.29, p < .001) than for those preceded by a central cue (see just above), yielding the reported increase in orienting
efficiency with age (that is, older participants benefited particularly from spatial cues as compared to central cues). Third (panel b), the age-related
RT increase in incongruent flanker trials (solid line; b = 0.0077 [0.0061, 0.0092], t = 9.79, p < .001) was smaller than for congruent trials (dashed line;
b = 0.0089 [0.0073, 0.0104], t = 11.29, p < .001), leading to the observed increase in executive efficiency with aging.
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles
Nature HumaN BeHaviour Articles
Nature HumaN BeHaviour
Extended Data Fig. 7 | Density plots for the distributions of untransformed RTs of correct trials. The plots show these RTs (from the trials analysed in the
main regression model) for incongruent flankers (solid lines) and congruent flankers (dashed lines), for younger participants (n = 592; panel a) and older
participants (n = 110; panel b), split at the midpoint of the age range (age 78). As can be seen, very few responses for either the incongruent or congruent
trials approached the timeout of 1,700ms for either group of participants, arguing against a ‘timeout’ alternative explanation for the age-related increase in
efficiency of the executive network (see ‘Linear effects of age: sensitivity analyses’, in Results).
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles Nature HumaN BeHaviour
Articles Nature HumaN BeHaviour
Extended Data Fig. 8 | Discovery of the optimal breakpoint model. This was achieved by comparing model goodness-of-fit (AIC) for
regression-with-breakpoints models with breakpoints at successive ages. The AIC of the optimal model (with a breakpoint at age 76) is indicated with the
gray arrow.
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

Articles
Nature HumaN BeHaviour Articles
Nature HumaN BeHaviour
Extended Data Fig. 9 | The nonlinear effect of age on the efficiency of the executive network, showing the executive effect for each of the 702
participants. The nonlinear age effect is displayed for (a) the model with a quadratic term for age, and (b) the breakpoint model with the optimal
breakpoint (age 76). Each data point reflects the difference between the mean log-RTs for incongruent and congruent flankers, for each participant.
NATURE HUMAN BEHAVIOUR | www.nature.com/nathumbehav

1
nature research | reporting summary April 2020
Corresponding author(s): João Veríssimo and Michael Ullman
Last updated by author(s): Jun 9, 2021
Reporting Summary
Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency
in reporting. For further information on Nature Research policies, see our Editorial Policies and the Editorial Policy Checklist.
Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.
n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement
A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly
The statistical test(s) used AND whether they are one- or two-sided
Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested
A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons
A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient)
AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals)
For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted
Give P values as exact values whenever suitable.
For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings
For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes
Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated
Our web collection on statistics for biologists contains articles on many of the points above.
Software and code
Policy information about availability of computer code
Data collection E-Prime (version 2.0)
Data analysis R (version 3.6.3), lme4 package (version 1.1-25), blme package (version 1.0-4). Our commented analysis code has been uploaded to Open
Science Foundation, at https://osf.io/59er2/.
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and
reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.
Data
Policy information about availability of data
All manuscripts must include a data availability statement. This statement should provide the following information, where applicable:
- Accession codes, unique identifiers, or web links for publicly available datasets
- A list of figures that have associated raw data
- A description of any restrictions on data availability
The anonymized data (with accompanying documentation) has been uploaded to Open Science Foundation, at https://osf.io/59er2/.

2
nature research | reporting summary April 2020
Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.
Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf
Behavioural & social sciences study design
All studies must disclose on these points even when the disclosure is negative.
Study description This was a quantitative experimental study, employing a cross-sectional continuous age design.
Research sample This study was part of the Social Environment and Biomarkers of Aging Study (SEBAS), conducted in Taiwan. SEBAS was in turn part of
the Taiwan Longitudinal Study of Aging, which examines a nationally representative sample of middle-aged and older adults in
Taiwan. During (only) the 2011 SEBAS collection wave, the Attention Network Test (ANT) reported here was included. It was given to
all available consenting participants of the health examination component of SEBAS. The main analyses reported here were based on
data from 702 participants (323 female), with a mean age of 67.69 years, and mean years of education of 7.76.
Sampling strategy As indicated above, we tested all available consenting participants from the health examination component of SEBAS in the 2011
collection wave. The sample size was considered to provide adequate statistical power, as it was far larger than previous ANT studies
that have reported age effects.
Data collection The task was presented on a Windows XP laptop with E-Prime Version 2.0. Participants responded on a response box (SRBOX,
Psychology Software Tools). Participants were tested in the presence of a single experimenter. Both the participants and the
experimenters were blind to the aims of the study, in particular the examination of age effects on the efficiency of each attentional
network. Informed consent was obtained from all participants.
Timing Participants were tested between October 24 2011 and May 6 2012.
Data exclusions Subject and trial exclusions proceeded in four steps. First, the ANT was given to all available consenting individuals of the health
examination component of SEBAS, which excluded respondents if they lived in an institution, were seriously ill, or had some other
health conditions that precluded examination. This yielded a sample size of 948 participants. Second, of these participants 92 were
excluded owing to: participant identifier coding errors (3 participants), not performing the full task (18 participants), or a diagnosis of
a neurological, psychiatric, learning, cognitive, or other brain-related disorder (71 participants). Third, an additional 154 participants
were excluded because of low accuracy on the ANT (lower than 75%). The same pattern of significance in the findings was obtained
in a sensitivity analysis with a more stringent inclusion criterion of at least 90% accuracy (213 participants excluded, rather than 154),
as well as in another sensitivity analysis with a more lenient inclusion criterion of at least 50% accuracy (122 participants excluded).
Fourth, prior to analyses, trials with timeouts (1.08% of all trials), presses of inappropriate response box buttons (0.02%), and
incorrect responses (1.62%) were excluded. Extremely fast responses (below 100ms) were discarded (0.009% of the remaining data)
because they are likely to constitute ‘short outliers’. Responses to double cue trials were additionally excluded, owing to their
redundancy with central cue trials; this enabled the calculation of the alerting and orienting effects within a single statistical model
by comparing no cue and spatial cue trials against the same baseline (central cue trials). A sensitivity analysis including both central
and double cue trials yielded the same pattern of results.
Non-participation 18 of the 948 participants did not complete the task (see just above).
Randomization Participants were not allocated into experimental groups.
Reporting for specific materials, systems and methods
We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material,
system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.
Materials & experimental systems
n/a Involved in the study
Antibodies
Eukaryotic cell lines
Palaeontology and archaeology
Animals and other organisms
Human research participants
Clinical data
Dual use research of concern
Methods
n/a Involved in the study
ChIP-seq
Flow cytometry
MRI-based neuroimaging

3
nature research | reporting summary April 2020
Human research participants
Policy information about studies involving human research participants
Population characteristics See above for age, sex, and education population characteristics. In order to control for the possible confounding effects of
sex and education, both of which were significantly associated with age, these variables were included as covariates in the
main analysis; the same pattern of results was obtained in a sensitivity analysis without these covariates included.
Recruitment We tested all available consenting participants from the health examination component of SEBAS in the 2011 collection
wave; see above. The pattern of results does not appear to be explained by survivor effects or the self-selection of fitter
adults in old age; see second paragraph of Discussion.
Ethics oversight This study complied with all relevant ethical regulations, and was approved by the Institutional Review Boards at Georgetown
University, Princeton University, and the Ministry of Health in Taiwan.
Note that full information on the approval of the study protocol must also be provided in the manuscript.