Professional Documents
Culture Documents
Verhaeghencerellabasak (Efficiencymodes)
Verhaeghencerellabasak (Efficiencymodes)
Verhaeghencerellabasak (Efficiencymodes)
Aging, task complexity, and efficiency modes: The influence of working memory involvement
Syracuse University
Abstract
adults for two verbal and one visuo-spatial task; each task was implemented in a baseline and a
an additive or a multiplicative fashion. The processing efficiency of older adults was defined as
the old-young ratio of the slopes of the load functions. Three levels of efficiency could be
distinguished. The first level, with an age-related slowing factor of about 1.2, consisted of low-
complexity verbal processing and additive-complexity verbal processing. The second level,
multiplicative complexity. The results go against any common-factor theory of aging. Instead,
they suggest that a shift from a higher to a lower mode of efficiency is triggered by a greater
Aging, task complexity, and efficiency modes: The influence of working memory involvement
One key question in the study of cognitive aging is the number of theoretical factors necessary to
explain the observed deficits. Although there is no doubt that age differences in cognition are
many and diverse (see Kausler, 1991, and Salthouse, 1991, for an enumeration of deficits raised
in the literature), one of the striking results from the past quarter century of cognitive aging
research is how much regularity can be detected amid the large variety of data. In fact,
correlational research in the 1990s has led to a theoretical account of aging as governed by a
single mechanism, identified as either processing speed (e.g., Salthouse, 1996), the
neuromodulation effects underlying it (Braver & Barch, 2002; Li, Lindenberger, & French,
2000), or even a common cause linking cognitive change to perceptual/motor processes (e.g.,
three scenarios are true. A: The cognitive system breaks down in a global fashion in all
individuals (but not necessarily at identical rates). Under this scenario, change in one ability
would be highly correlated with change in another, because they indeed change together. B: The
cognitive system breaks down randomly. In scenario B, some components become defective
earlier than others, but which components break down is a matter of chance (Gavrilov &
Gavrilova, 1991). When averaged at the group level, the random factor leads to perceived global
changes in cognitive functioning. C: The cognitive system breaks down along predictable lines
(called ‘dissociations’ by Perfect & Maylor, 2000). Under scenario C, we expect correlated
change within each of the levels of dissociation. (One such dissociation that has been
investigated extensively is that between lexical/verbal tasks and visuospatial tasks – the former
Scenario A follows if all elements of the system are equally vulnerable; scenario B
components is compromised differently due to life history (e.g., vascular damage due to tobacco
use); scenario C follows if some components are particularly vulnerable or resistent to the effects
of aging. Despite the deep differences among these alternatives, correlational research is not very
helpful in distinguishing between them (Hofer & Sliwiniski, 2001). If change is correlated, that
is, if the rank ordering of individuals remains largely identical over time, then by necessity a
common factor will emerge; this is true under either scenario A or scenario C. Hofer and
Sliwinski have demonstrated analytically that even scenario B can lead to the emergence of a
common factor. Experimental research, however, can help in detecting predictable dissociations
(scenario C).
reaction-time data by using Brinley plots (e.g., Brinley, 1965; Cerella, Poon, & Williams, 1980).
A Brinley plot is a scatter plot with mean performance of younger adults on the X-axis and mean
performance of older adults on the Y-axis. The early plots typically showed that a single straight
line (with a small intercept), and hence a single linear equation, could describe data drawn from
slower than younger adults by a certain constant multiplicative factor, regardless of task (see
Cerella, 1990, for a theoretical model that derives slowing factors from the linear slopes of
Brinley functions). This factor defines what we (Verhaeghen et al., 2002) have termed the
In more recent studies, however, distinct age-related slowing functions have been found
for different types or ‘domains’ of tasks, bringing us closer to scenario C. Two types of findings
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 5
have emerged. Hale, Myerson and colleagues have consistently found distinct age-related
slowing factors for lexical versus visuospatial processing (e.g., Hale & Myerson, 1996; Jenkins,
Myerson, Joerding, & Hale, 2000; Lima, Hale, & Myerson, 1991; Myerson & Hale, 1993; see
also Sliwinski & Hall, 1998). Kliegl, Mayr and colleagues have consistently found different
slowing factors for tasks that do not require coordination within working memory versus tasks
that do (e.g., Kliegl, Mayr, & Krampe, 1994; Mayr, Kliegl, & Krampe, 1996; Verhaeghen,
Kliegl, & Mayr, 1997). These results appear to be true bifurcations, that is, the stratification of
points in the Brinley plot suggested qualitative, all-or-none jumps in processing speed, to the
exclusion of intermediary values. To underscore the quantal character of the deficits, Verhaeghen
Hale/Myerson and the Kliegl/Mayr bifucations. We included a large set of verbal and spatial
tasks (roughly corresponding to the two Hale/Myerson modes) requiring low or high degrees of
executive control (roughly corresponding to the two Kliegl/Mayr modes) and tested for
dissociations. What we observed was a trifurcation, that is, we found evidence for the existence
of three distinct age-related slowing factors. The first mode had an age-related slowing factor of
about 1.0 (i.e., no slowing) and consisted of low-complexity lexical processing. The second
mode, associated with a slowing factor of about 1.7, consisted of a mixture of verbal-high-
slowing factor of about 2.2, consisted of visuospatial processing of high complexity. This study
affirmed the existence of a manifold of efficiency modes, each governed by a distinct age-related
slowing factor, coupled with general age-related effects within the modes.
The Verhaeghen et al. (2002) study raises two questions. The first concerns the status of
the intermediate efficiency mode. This mode contained a mixture of verbal tasks requiring
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 6
permutations (anagram solving and anagram solving after letter-by-letter transformation) and of
simple spatial tasks (pattern detection). Although these tasks all yielded the same slowing factor,
it is still possible that the efficiency modes for the two types are theoretically distinct, and were
conflated only empirically. Alternately, the two sets of tasks may truly form a unitary mode -- a
more provocative possibility. In the present study, we tested whether the high-complexity verbal
efficiency mode indeed coincides with the low-complexity visuospatial efficiency mode, using a
more sensitive paradigm. By implementing multiple levels of difficulty, we were able to measure
linear functions relating response time to difficulty in both low- and high-complexity contexts.
The parameters from these information-processing functions (slopes and intercepts), rather than
the point estimates of the previous studies, should enhance the reliability and sensitivity of our
A second question concerns the status of the complexity effect. We found that one
complexity manipulation left intact the age-related slowing factor that governed the baseline
(e.g., does the letter string ‘atx’ after transformation – resulting in the string ‘buy’ – form a
word?) and to anagram solving (e.g., does the letter string ‘txa’ after transformation – resulting in
the string ‘uyb’ – form a scrambled word?). Another complexity manipulation, namely a
concurrent spatial transformation in a pattern-detection task (the ‘matrices’ task included in the
present study; for more details, see below), led to a marked increase in the slowing factor that
How can these different complexity outcomes be understood? As we define the term,
making a task more complex must lead to some sort of latency increase in the central processing
component. We thereby exclude manipulations at the sensory or motor levels. One possibility is
that the manipulation will add an extra processing stage to a task or prolong an existing stage –
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 7
perhaps by imposing a fixed overhead cost or ‘set-up charge’. We (Verhaeghen & Cerella, 2002;
Verhaeghen, Steitz, Cerella, & Sliwinski, in press) have labeled this type of complexity additive,
because the manipulation will induce additive effects between the baseline and experimental
conditions: Latencies in the complex conditions will be equal to latencies in the simpler
conditions plus a constant (see below for more details). A second possibility is that the
complexity manipulation may modulate all of the iterative processing of a task, prolonging or
‘inflating’ each step in the chain of baseline computation. We label this type of complexity
complex conditions will be a fixed ratio (larger than unity) of central-processing latencies in the
simpler conditions. Combinations of both complexity effects are also possible (and likely): the
complexity manipulation may add a processing stage to the stream and additionally modulate the
iterative process.
jump in efficiency is that the manipulation be multiplicative, that is, it must interfere with the
central/computational processes involved in the low-complex version of the task (see also Mayr
& Kliegl, 1993). For instance, the one-letter-up alphabet transformation likely adds a simple
translation stage to the chain of processes, without interfering with the lexical decision
component itself. Other tasks might interfere with central processing, for instance by introducing
from Verhaeghen et al. (1997). Solving relatively complex problems with brackets, such as [6 +
(2 + 1)] – [8 – (4 + 2)], not only adds a translation stage, it also forces the participant to store
intermediate solutions in working memory and access them while calculating intermediate
answers. This requirement did lead to a multiplicative effect (as well as an additive effect).
Another example of a task necessitating such operations may be the complex version of the
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 8
matrices task used in Verhaeghen et al. and in the present study (see below). Under our
hypothesis, a jump from one efficiency mode to another will be triggered only if the complexity
To address these questions, our experiment utilized three baseline tasks, each subjected to
a complexity manipulation. Two tasks received a multiplicative manipulation, one of them verbal
(mental arithmetic) and the other visuospatial (the matrices task from Verhaeghen et al., 2002 --
in essence a pattern-detection task). A third task was also verbal ( counting), and was subject to
an additive manipulation. For each task, we derived full information-processing functions from
both the baseline version and the complex version. Regarding the alleged domain of these tasks,
note that the cognitive literature has maintained that mental arithmetic relies mostly on direct
retrieval from a semantic store, e.g., Ashcraft, Donley, Halas, & Vikali, 1992; Butterworth, Zorzi,
Girelli, & Jonckheere, 2001; Siegler & Shipley, 1995. Several aging studies on mental arithmetic
are also compatible with the conceptualization of this task as a verbal one (Salthouse & Coon,
1994; Verhaeghen, Kliegl, & Mayr, 1999). The cognitive literature describes counting as a
deliberate verbal process (see, amongst others, Butterworth, 1999, and Dehaene, 1997); see
Basak & Verhaeghen (2003), and Watson, Maylor, & Manson (2002), for aging studies
manipulation of task difficulty. Our study employed “iterative” tasks, that is, tasks for which the
difficulty manipulation represents the addition of successive (presumably identical) steps in a set
explanation is that the stimuli are processed serially and that each item adds (on average) a fixed
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 9
increment of time to the total task latency (e.g., Neisser, 1963; Shepard & Metzler, 1970). On
this interpretation, the slope of the information-processing function measures the central, or
computational, processing rate, that is, the time to complete one processing step.
As stated above, the three tasks in our study were counting, mental arithmetic, and the
matrices task from Verhaeghen et al. (2002). Each task was implemented at four levels of
difficulty. For counting, we varied the number of elements to be enumerated; for arithmetic, the
number of operations; and for the matrices task, the number of elements in the display. Each task
was also implemented at two complexity levels: a baseline level, and a level necessitating
additional processing. For counting, we added distractors; for arithmetic, brackets; and for
matrices, arrowheads signifying spatial transformations. We expected that response times in the
baseline version of each task would be described by a simple linear function. The slope of the
function indicates the time needed to complete a single step in the iterative chain; it is therefore a
direct estimate of the rate of processing of the iterative process. These rates can be calculated for
each individual separately, and pooled within age groups to yield an estimate of the rate of
processing of younger and older adults, respectively. To assess the presence or absence of age
differences in the rate of processing, the rates can be compared directly using standard mean-
difference techniques like t-tests or ANOVA. To assess the size of the age differences in the rate
of processing, that is, to derive the efficiency modes, we can calculate the age-related slowing
factor by taking the ratio of the slopes of older adults over the slopes of younger adults.
The intercept of the information processing function is an indicator of the duration of all
task processes that are independent of the rate-of-processing that governs the slope. We should
note that a complication arises here. Although this intercept is typically evaluated at N = 0, this is
generally assumed that when the number of items is small participants do not count in a serial
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 10
fashion, but process all items in parallel (the ‘subitizing’ range, Trick & Pylyshyn, 1994). Serial
counting processes are engaged once the number of items is greater than can be processed in
parallel. Therefore, when counting is the relevant process, the intercept should be evaluated at
the largest number of items the individual can subitize, rather than at zero. We call the intercept
value evaluated at the correct location on the abscissa the ‘true intercept’, RT0, and label its
location N0. N0 can be expected to vary across tasks, and, even worse, for certain tasks (such as
counting) across individuals and age groups (Basak & Verhaeghen, 2003). This has
consequences for the interpretation of age differences under particular types of complexity
Complexity manipulations
additive-factors logic. The difficulty manipulation creates four levels of a Factor A, which are
crossed with two levels of a Factor B, the complexity manipulation. Sternberg has elaborated
three possible outcomes of this factorial design: Factor A and Factor B may be independent, in
which case two parallel lines will emerge in the information-processing plot; or there may be an
underadditive interaction or an overadditive interaction, in which case the two lines will either
converge or diverge, respectively. As stated above, we (see Verhaeghen, Steitz, Sliwinski, &
Cerella, 2003, Appendix B and C, for a complete formal mathematical treatment) have focused
on the parallel and diverging outcomes, and have labeled these two cases as additive and
When the complexity manipulation adds an extra processing stage to a task or prolongs
an existing stage, it yields additive effects: Latencies in the complex conditions will be equal to
functions will then be parallel, and the complexity cost or set-up charge is given directly by the
When the complexity manipulation modulates all of the central processing of a task,
prolonging or ‘inflating’ each step in the chain of baseline computation, the effect will be
multiplicative: Central-processing latencies in the complex conditions will be a fixed ratio (larger
than unity) of central-processing latencies in the simpler conditions. This will lead to a divergent
interaction between the two factors. The fixed ratio or inflation factor is given directly by the
ordinarily be unchanged. This will be true, however, only if the true intercept location N0 is the
same as the default intercept built into the regression, N = 0. Consider the case of counting,
where the true intercept marks the upper limit of the subitizing range, a value close to N0 = 3 in
typical subjects (Verhaeghen & Basak, 2003). Let the response time at N0 be designated RT0 (if
not measured directly, this value can be obtained by extrapolation from the baseline function).
Under the action of multiplicative complexity, the two information-processing functions will
diverge or “fan-out” from the point (N0, RT0). Because the function pivots around this point in
the first quadrant of the N x RT plane, the zero-intercept of the high-complexity line will be
reduced. This distortion of the true intercept could be corrected for, except that the true intercept
location N0 is unknown for any particular dataset. In practice, only a general statement can be
A third possibility exists, the “full” complexity model, in which both additive and
multiplicative effects are present. In general, the full complexity model and the multiplicative
model will share a similar signature: a change in slope, and, possibly, a change in intercept
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 12
(evaluated at zero) of the information-processing function. Thus the two models are potentially
indistinguishable. Note, however, that if the zero-intercept increases, this implicates the full
complexity model and rules out a simple multiplicative model. If, on the other hand, the intercept
is constant or reduced, either model could be responsible, and the data are inconclusive.
Research questions
multiplicative effects by lines that diverge upwards. The full complexity model (i.e., a model
with both multiplicative and additive effects) cannot be distinguished reliably from the
multiplicative model, unless a positive intercept difference between complex and baseline
conditions is found, which points at the effects of the full model. Efficiency modes (i.e., age-
related slowing factors) can be calculated by taking the ratio of the slopes of the information-
We examined the following hypotheses. First, we expected two distinct efficiency modes
to emerge from the baseline tasks, one for the verbal tasks (counting and mental arithmetic) and
one for the visuospatial task (matrices). Second, we expected additive complexity effects for
counting with distractors, and multiplicative or full complexity effects for bracket arithmetic and
arrowhead matrices. Third, we expected that the latter two tasks would entail a jump in
efficiency mode; the multiplicative complexity demand would inflate age-related slowing factors
in the iterative process only in those two tasks. An open question was the number of age-related
slowing factors, or efficiency modes, that would be uncovered. It could be either three (low
verbal, high verbal and low visuospatial combined, and high visuospatial) or four (low verbal,
Method
Participants
Participants were 36 younger adults (30 females), all undergraduate students at Syracuse
University who received course credit for participating; and 32 older adults (20 females)
recruited through newspaper advertisements and recruitment at community centers, who received
$20 for participating. The younger adults were on average 18.69 years old (SD = 0.77), and had
completed 12.55 years of education (SD = 0.61); the older adults were on the average 72.53
years old (SD = 5.76), and had completed 14.84 years of education (SD = 2.21).
Tasks
In the baseline counting task, each stimulus consisted of a number of Xs, displayed in
random off-set locations within a virtual 10-by-10 grid. Four conditions representing different
levels of difficulty were implemented, by varying the target numerosity between 6 and 9, using a
Yes/No verification procedure. Numerosity levels were blocked. For half of the trials within each
difficulty level, the number of Xs shown was identical to the target numerosity, and the
participant pressed the ‘yes’ button. On the other half of the trials, the number of items shown
was either one item smaller or larger than the target numerosity, and the participant pressed the
‘no’ button. The complex version of the counting task was identical to the baseline version,
except that on each trial, between 4 and 7 distractors (+ signs) were interspersed with the X-
targets.
In the baseline arithmetic task, participants had to verify the end result of a series of
arithmetic operations projected in the middle of the screen (e.g., 4 + 5 – 3 + 1 – 2 = 7). All
numbers presented were larger than 0 and smaller than 10; the operations were either
subtractions or additions. All intermediate solutions, as well as the end solution, were larger than
0 and smaller than 10. Four conditions representing different levels of difficulty were
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 14
implemented, by varying the number of operations between 4 and 7. Conditions were blocked.
For half of the trials within each level of problem size, the provided answer matched the correct
answer, and the participant pressed the ‘yes’ button. On the other half of the trials, the solution
shown was either one number smaller or larger than the correct solution, and the participant
pressed the ‘no’ button. The complex version of the arithmetic task was created by introducing
brackets in the series of operations (e.g., [(4 + (5 – 3)] + (2 – 1) = 7). Brackets were placed such
that at least one imbedded bracketed operation occurred in every problem. Again, all
intermediate solutions and the end solution were constrained to be between 1 and 9.
In the baseline matrices task (also used in Verhaeghen et al., in press), each stimulus
consisted of a number of Xs in a 10-by-10 grid. In positive items, four of the Xs formed the
corner points of a virtual square three, four, or five cells wide (for an example, see the top part of
Figure 1); in negative items none of the Xs could be configured as the corner points of a square.
The participant’s task was to decide whether such a virtual square was present or not. Four
conditions representing different levels of difficulty were implemented: zero distractors, three
distractors, five distractors, and seven distractors. The distractors were Xs; thus there was a total
of 4, 7, 9 and 11 Xs in the four levels respectively. The zero-distractor or ‘copy’ condition gave
participants a preview of the possible configurations of Xs, before attempting the distractor
conditions. The complex version of the task was similar to the baseline version, with two
differences. Instead of Xs, arrowheads were printed (the symbols ‘<’, ‘>’, ‘V’ or an upside down
‘V’ — see the bottom row of Figure 1 for examples). The participant was required to take into
account not the location of the arrowhead, but the location it was pointing to, and was instructed
to “Imagine an X in the box that the arrow is pointing to”. The arrow grids also differed in the
order not to overtax participants. For an example, see the bottom part of Figure 1.
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 15
Procedure
All tasks were presented on computer, and response times were recorded to the nearest
millisecond using a timing routine for Turbo Pascal developed by Brysbaert, Bovens,
d’Ydewalle, and Van Calster (1989). Participants positioned the monitor at their most
comfortable viewing distance. For each task at each difficulty level, 40 items were presented.
The first six were considered practice items; response times and error rates were calculated from
the last 34 trials. All items were presented in a verification (yes/no) format; chance level was
50%. Decisions were indicated by pressing the ‘/’ key for ‘yes’ (marked with green tape) or the
‘z’ key for ‘no’ (marked with red tape). Subjects were instructed to be both fast and accurate.
Trials were subject-initiated by pressing the space bar, and the stimulus remained on the screen
Testing was conducted individually in quiet and light-controlled testing rooms. Order of
presentation of tasks was randomized across participants. Within each task, the baseline version
was presented first, with its conditions presented in increasing level of difficulty, and then the
complex version, again with the conditions presented in increasing level of difficulty. We
adopted this increasing order of difficulty and complexity in order to make certain that
participants would be well practiced on the easier and the baseline versions before moving on the
more difficult levels and especially to the more complex version of the task (see also Verhaeghen
et al., 1997). If age-related slowing observed in the baseline versions of the task were partially
due to disuse in the older sample, this procedure would work against finding age by complexity
interactions.
The alpha level for statistical testing was set at p < .05.
Results
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 16
A first pass over the data indicated that some participants were overtaxed by the more difficult
conditions of some tasks, as indicated by very low accuracy, sometimes coupled with very short
latencies. To ensure that data were used only from participants who performed a task in a
meaningful way, we applied task-wise deletion of data from participants who scored lower than
60% accuracy (with 50% being chance level) on any of the conditions within each of the tasks
(i.e., within counting, within arithmetic, or within matrices). The final sample size was: for
counting, 33 younger and 29 older adults; for arithmetic, 25 younger and 24 older adults; and for
matrices, 16 younger and 23 older adults. All results are depicted in Figure 2 (response times in
Error analyses
Error data were analyzed within each task using repeated measures analysis of variance
(ANOVA), with age group as the between-subject variable, and level of difficulty and level of
complexity as the within-subject variables. These analyses are mainly control analyses; we
expect the dissociations to occur in latency, not accuracy. Previously, we (Cerella, 1990;
Verhaeghen et al., 2002) have stressed that latency dissociations can be meaningfully interpreted
only when no age by complexity interactions are present; otherwise, age by condition
across conditions.
For counting (Figure 2B), only the main effects of difficulty, F(3, 180) = 7.07, and
complexity, F(1, 60) = 28.19, were significant, indicating that increased difficulty led to higher
error rates, and so did adding distractors. The main effect of age was marginally significant, F(1,
60) = 3.50, p = .066; older adults tended to be more accurate than younger adults.
For arithmetic (Figure 2D), the main effects of difficulty, F(3, 141) = 11.84, and
complexity, F(1, 47) = 53.34, were significant, indicating that increased difficulty led to higher
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 17
error rates, and so did adding brackets to the problems. The interaction between complexity and
difficulty was also significant, F(3, 141) = 6.26, indicating that the effects of difficulty were
more pronounced in the complex condition. The main effect of age was marginally significant,
F(1, 47) = 3.97, p = .052; older adults tended to be more accurate than younger adults.
Because the difficulty level is implemented differentially for the two complexity
conditions in the matrices task (Figure 2F), a standard ANOVA on the full range of the difficulty
manipulation cannot be applied. Instead, we conducted a 2-by-2 ANOVA, using the two number-
of-distractor levels for which data are available for both the baseline and the arrow version of the
matrices task (zero and three distractors). The main effects of difficulty, F(1, 37) = 199.04, and
complexity, F(1, 37) = 13.77, were significant, indicating that increased difficulty led to higher
error rates, and so did the arrowhead manipulation. The interaction between complexity and
difficulty was significant, F(1, 37) = 68.77, indicating that the effects of difficulty were more
pronounced in the complex condition. A significant three-way interaction was also found, F(1,
37) = 5.99. Inspection of Figure 2F suggests that this is due to one data point: an age difference
(favoring the older adults) emerged in the three-distractor baseline version of the task; age
Our response time analyses were conducted using each individual’s median response time
(for correct trials only) within each difficulty level within each complexity level within each task.
We used median values rather than average response time, because medians are less influenced
by outlying values. Figure 2 (panels A, C, and E) displays the averaged medians per difficulty
level within complexity level within task, as well as the fitted regression lines.
The regression analysis of the response time data was done using multilevel modeling
(random coefficients model), so that nested sources of variability could be taken into account.
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 18
For all three tasks, random coefficients models were found to be a better fit to the data than
random intercepts models; therefore, only results from the former type of analysis are reported.
Random effects and their variances were used to model individual differences in intercepts and
models: (a) a dummy variable indicating the participant’s age group (0 = younger and 1 = older);
(b) a dummy variable indicating the complexity level (0 = baseline; 1 = complex); and (c) a
dummy variable (ranging from 1 to 4) indicating the difficulty level (for counting: number of
items to be counted; for arithmetic: the number of arithmetic operations; for matrices: the
number of distractors). The full model used for counting and arithmetic was:
where Yij is the median response time for each individual measured for the type of task,
separated by complexity of the task. U0j is the random effect of intercept for person j, U1j is the
random effect of slope for person j, and Rij is the residual for person j at difficulty level i.
For the counting task (see Table 1), two interaction terms failed to reach significance,
namely the complexity by difficulty interaction (indicating that the slope of the information-
processing function did not differ between complexity levels), and the age by difficulty by
complexity interaction (indicating that age differences in the slope of the information-processing
function are identical across complexity levels). Therefore, a new model was estimated (the
constrained model, see Table 2) in which the coefficients carrying these effects were dropped. In
this model, a significant main effect of difficulty was found, F(1, 60) = 543.41, indicating that
response time increased with the number of items to be enumerated, as well as a significant main
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 19
effect of complexity, F(1, 370) = 2261.27, indicating that the presence of distractors slowed
participants down. Age yielded only a trend towards a main effect on the intercept, F(1, 370) =
3.08, p = .08, but there was a significant age by difficulty interaction, F(1, 370) = 8.46, indicating
that the slopes of the information-processing functions were steeper for older adults than for
younger adults. The age by complexity interaction was also significant, F (1, 370) = 202.92,
indicating that the effect of the complexity manipulation on the intercept of the information-
processing function was larger for older than for younger adults.
For the arithmetic task, we found a significant main effect of difficulty, F(1, 47) = 330.33,
indicating that response time increased with the number of operations, as well as a significant
effect of complexity, F(1, 290) = 10.81, indicating that the bracket manipulation slowed
participants down. There was a significant difficulty by complexity interaction, F(1, 290) =
67.49, indicating that the slopes of the information-processing functions were steeper for the
bracket condition than for the baseline condition. Age yielded no significant main effect on the
intercept, F(1, 291) = 2.18, p = .14, nor did it interact significantly with complexity, F(1,290) =
0.67, p=.41, indicating that there are no significant age differences in the intercept of the
information-processing function in any of the conditions. Age did, however, interact significantly
with difficulty, F(1, 290) = 9.59, indicating that the slopes of the information-processing
functions were steeper for older adults than for younger adults. A significant three-way
interaction was also obtained, F(1, 290) = 10.44, indicating that the effect of the bracket
manipulation on the slope of the information-processing function was larger for older than for
younger adults.
Because different difficulty manipulations were used for the baseline and complex
version of the matrices task, they could not be included in a single model. Therefore, two
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 20
separate models were fitted to this task, one for the baseline condition, one for the arrow
All effects were significant in both conditions. Significant main effects were found for
difficulty, F(1, 39) = 145.97 and 74.32, for the baseline and arrow condition, respectively,
indicating that response time increased with the number of distractors. The age by difficulty
interaction was significant in both conditions, F(1, 78) = 10.48 and 26.88, for the baseline and
arrow condition, respectively, indicating that the slope of the information-processing function
was steeper for older adults than for younger adults. The F value associated with the interaction
effect for the arrow condition is approximately two-and-a-half times larger for the arrow
condition than for the baseline condition, suggesting that the effect of the arrow condition is
larger in older than in younger adults. In fact, the ratio of the slope of the arrow condition over
the baseline condition is more than twice as large in older adults than in younger adults (4.02 vs.
1.73). As a more formal test, we conducted a repeated-measures ANOVA on slope values, and
found indeed a significant age by complexity effect, F (1, 37) = 11.50, MSE = 633,740.04,
confirming that the age differences favoring the younger adults were larger in the more complex
arrowhead condition.
Table 4 summarizes the results from the multilevel analysis, by reporting the intercepts
(evaluated at difficulty = 0) and slopes for the information-processing functions, as well as the
effects of age on the slopes (i.e., the ratio of slopes for older over slopes of younger adults) for
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 21
both the simple and complex versions of the tasks, and the effects of complexity on the slopes
(i.e., the ratio of slopes for complex over slopes of simple tasks) for both young and older adults.
With regard to the age effects, we expected either three or four age-related slowing
factors to emerge. These slowing factors are operationalized as the ratio of the slopes of the
functions of younger adults. More specifically, we expected that a first efficiency mode would
complex arithmetic; a third by baseline matrices; and a fourth by complex matrices. The main
question was whether the second and third mode would coincide or not.
slowing factors within an efficiency mode and non-equivalence of slowing factors across modes.
More specifically, we can formulate five hypotheses that, by transitivity, imply all possible
relevant contrasts about the efficiency modes. First, is the age-related slowing factor for baseline
counting equal to the age-related slowing factor for baseline arithmetic? This tests for
equivalence of slowing factors within the first mode, and we expect an affirmative answer.
Second, is the age-related slowing factor for baseline counting equal to the age-related slowing
factor for complex counting? The complexity effect in counting was found to be additive, so this
contrast again tests for the equivalence of slowing factors within the first mode, and we expected
an affirmative answer. Third, is the age-related slowing factor for complex arithmetic larger than
the age-related slowing factor for baseline arithmetic? This tests for equivalence of slowing
factors between the first and second mode, and we expect the answer to be negative. Fourth, is
the age-related slowing factor for baseline matrices larger than or equal to the age-related
slowing factor for complex arithmetic? This tests for equivalence of slowing factors between the
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 22
second and third mode, and we had no clear expectation regarding the answer. Finally, is the age-
related slowing factor for complex matrices larger than the age-related slowing factor for
baseline matrices? This tests for equivalence of slowing factors between the third and fourth
Age-related slowing factors are calculated at the group level (i.e., they are calculated as
the average slope of one group over another). Therefore, no variance is associated with an age-
related slowing factor, and therefore, they cannot be implicated in significance testing. To derive
statistically testable hypotheses, we rearranged each of the equations implicit in the five contrasts
such that the contrasts could be evaluated as age contrasts of ratios of information-processing
slopes. For instance, the first contrast above translates into the equation:
To test the latter equality, we can calculate the slope of the information-processing
function for baseline counting and baseline arithmetic for each individual, calculate the ratio of
those two slopes, and then use an independent samples t-test to test for age differences in this
ratio.
Results of t-tests for all five contrasts are provided in Table 5. The results of the t-tests
confirm the presence of three, not four, efficiency modes, with the tasks grouped as follows (in
ascending order of age-related slowing factors: (a) baseline counting, complex counting, and
baseline arithmetic (average slowing factor of 1.24); (b) complex arithmetic and baseline
matrices (average slowing factor of 1.65); and (c) complex matrices (slowing factor of 4.02).
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 23
Discussion
The present study examined age differences in response time and error rates in three
cognitive tasks, each presented at four difficulty levels. Each task was presented in both a low-
complexity and a high-complexity version. The main analysis concerned the effects of age on the
difficulty). More specifically, we were interested in whether the interaction between task
complexity and task type (verbal or visuospatial) would induce either three or four distinct age-
related slowing factors. Before we present the results concerning this overall analysis, we first
The first task we examined was counting. The information-processing functions were
highly linear (see Figure 2A). In a baseline version of the task, in which participants enumerated
between 6 and 9 Xs, we found slight age differences. Younger adults counted at a rate of 292
ms/item, older adults needed 376 ms/item, yielding an age-related slowing factor of 1.3. In a
complex version of the task, between 4 and 7 distractor + signs were included in the display. This
increased overall response time by 936 ms in the younger adults, and 1,737 ms in the older
adults, but it did not reliably affect the counting rate itself in either age group. We have labeled
this type of complexity effect, which elevates the information-processing function without
altering its slope, an additive complexity effect. Such effect is compatible with the addition of a
new stage in the information-processing chain, namely the stage of identifying and inhibiting the
distractors. This stage does not interfere with the counting process itself, and its duration is given
directly by the difference in intercept between the two conditions. This identification/inhibition
stage took much longer in older adults than in younger adults: The age-related slowing factor
was 1.9 (i.e., 1,737 ms/936 ms). The counting process itself was not affected by the presence of
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 24
distractors and therefore, the age-related slowing factor for this process was identical for the
The second task was mental arithmetic, again yielding highly linear information-
processing functions (Figure 2C). In the baseline version, participants calculated the end result of
ms/operation, and older adults 1,533 ms/operation, resulting in a small age-related slowing factor
of 1.2. In the complex version, brackets were introduced. As in previous research (e.g., Salthouse
& Coon, 1994; Verhaeghen et al., 1997), this slowed the rate of processing. Younger adults
needed 2,067 ms/operation in the bracket condition, and older adults 3,247 ms/operation,
inflating the age-related slowing factor to 1.6. The complete result we obtained (a decrease in the
with both a multiplicative and a full (i.e., simultaneously additive and multiplicative) complexity
model, as explained in the Introduction. (A small additive effect is almost certain of occurrence,
because the brackets need to be perceived and interpreted). Either model is compatible with the
notion that the insertion of brackets interferes with the mental arithmetic processes themselves,
and that this effect is more pronounced in older than in younger adults.
The final task was the matrices task (Verhaeghen et al., 2002), again yielding highly
task. In the baseline version, participants searched for the presence of absence of the corner
needed 392 ms/item, older adults 679 ms/item, resulting in an age-related slowing factor of 1.7.
In the complex condition, the Xs were replaced by arrowheads, pointing at the location of virtual
Xs. This array of virtual Xs was then to be searched for the presence or absence of the corner
points of a virtual square. The grid contained between 0 and 3 additional arrowheads. This
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 25
complexity manipulation affected the rate of processing: Younger adults needed 603 ms/item in
the arrowhead condition, and older adults 2,424 ms/item, inflating the age-related slowing factor
of 4.0. It also increased the intercept of the information-processing function. The increase in
intercept shows that the full complexity model is necessary to account for the data. We assume
that each arrow transformation by itself is an additional process that takes a fixed amount of
added time, and that the need for concurrent storage and processing may be the source of the
multiplicative effect (e.g., Verhaeghen et al., 1997). As explained in the Introduction, we cannot
calculate age effects in the additive component directly because the true intercept value is
unknown.
Bringing the results for the three tasks together, we found clear evidence for the existence
of three distinct tiers, or levels, of age-related slowing factors. This is evident in the clustering of
the old-young ratios of the slopes of the information processing functions around three values;
these slope ratios denote the age-related slowing factors. Baseline counting, complex counting,
and baseline arithmetic showed the smallest age-related effect, with an average slowing factor of
1.24. A second tier, with moderate age effects, was formed by complex arithmetic and baseline
matrices (average slowing factor of 1.65). The largest age effects were noted for complex
matrices (slowing factor of 4.02). These tiers in response times occur in the absence of such tiers
in error rates. Therefore, we can interpret these levels as indicators of age differences in
efficiency of processing, that is, of age differences in the speed at which a stable age difference
in accuracy is reached. These levels have been labeled ‘efficiency modes’ (Verhaeghen et al.,
2002). Statistical analysis showed that the age-related slowing factors were identical within those
groupings, and different between them. As in the previous study, the existence of statistically
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 26
distinct values in the old/young slope ratios suggests qualitative, all-or-none jumps in processing
These results go against any single-parameter model of aging. Cerella (1994) has shown
that if a single information-processing mechanism drives the performance of both age groups,
then the resulting plot of points in Brinley space is by necessity monotonic. Dunn and Kirsner
(1988) have proven analytically that the inverse is also true: Monotonicity in two variables (older
adults’ mean latency and younger adults’ mean latency) implies that a third variable (general
cognitive efficiency) can be constructed to which both are monotonically related. Conversely, a
Note that the age-related effects obtained here run counter to what has often been termed
the ‘complexity effect’ (e.g., Cerella, 1990), that is, the finding that age differences are often
monotonic with the degree of task difficulty, as derived (in the original conception) from
latencies of younger adults. (Unfortunately, Cerella’s original use of the term ‘complexity’
coincides with what we – following standard practice in the psychometric literature – have here
labeled as ‘task difficulty’.) Within Cerella’s framework, efficiency mode jumps might be
accommodated by positing threshold effects: Maybe older adults perform quite adequately on a
task if it is not too difficult, but once difficulty exceeds a particular threshold, age differences
magnify. This rationale does not hold here. If we rank-order the tasks by the best measure of
across-task difficulty in central processing that is available in the present data set, namely by the
slope of the information-processing function in younger adults, the rank order is: baseline and
complex counting (292 ms/item), baseline matrices (392 ms/item), complex matrices (603
ms/item), baseline arithmetic (1,321 ms/item) and complex arithmetic (2,067 ms/item). This
ordering does not correspond to the ordering of age effects as given by the efficiency modes. In
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 27
fact, the smallest age effect (age-related slowing of 1.16) was observed in the next-to-most
The pattern we obtained replicates the pattern of results found in the Verhaeghen et al.
(2002) study, although we included different tasks and used a different methodology to derive
age-related slowing factors. The first mode in the previous study consisted of lexical decision
alphabetical transformation to be applied to each letter; e.g., does the letter string ‘atx’ after
transformation – resulting in the string ‘buy’ – form a word?), and category judgment. The
medium efficiency mode consisted of anagram solving with and without concurrent
also included here). The least efficient mode consisted of visual search for Ys amid X distractors
and of pattern detection with concurrent transformations (a task also included here).
The present data shed some light on the interpretation of the intermediate efficiency mode
as found in Verhaeghen et al. Previously, taking a lead from other studies on permutations (Li &
Lewandowsky, 1993, 1995), we argued that the intermediate level could be interpreted as a
purely visuospatial level if it is assumed that anagrams rely on some form of visuospatial
representation/manipulation of the letters. The present study seems to rule out that explanation: It
is unlikely that mental arithmetic with brackets can be construed as a spatial task. Therefore, the
most plausible explanation for the intermediate level is that it is indeed a mixture of two
Pooling the results from both studies, the four efficiency modes (two of which are
coinciding) can now be interpreted as: (a) a verbal/low-complexity mode (counting with and
without distractors; arithmetic; lexical decision with and without alphabetical transformations;
identical in its slowing factor to the verbal/high-complexity mode (pattern detection); and (d) a
An additional point of interest in the study was the effect of different types of complexity.
On the one hand, we hypothesized that the imposition of multiplicative complexity is a necessary
condition for a shift in efficiency mode to occur. Two of our tasks (arithmetic and matrices)
yielded multiplicative complexity effects (and the latter task, an additional additive effect), in
that the slope of the information-processing function increased from the low-complexity to the
high-complexity version. As expected, these were the tasks that showed a jump in efficiency
mode. On the other hand, we hypothesized that merely additive complexity would not result in
an efficiency-mode shift. One of our tasks (counting) yielded a purely additive complexity effect,
in that the intercept of the information-processing function increased from the low-complexity to
the high-complexity version (but not the slope); as predicted, the slope here remained the same.
Study limitations
A limitation of the present study was due in the unexpected difficulty of the tasks. We
applied an admission criterion of 60% accuracy to ensure data quality (50% accuracy was
chance), and this resulted in the taskwise deletion of a sizeable portion of our sample.
Proportionally more younger than older adults were deleted from the sample. It is difficult to
assess the consequences of this differential retention rate on the measured age differences in
response times; likely the procedure led to a somewhat inflated estimate of age-related slowing
factors (after removal of the poorer young adults in each task). Likewise, the criterion may have
been partially responsible for the absence of age by condition interactions in error rates.
Although the present study replicates and extends the findings of Verhaeghen et al.
(2002), some questions remain open, such as the nature of the mechanism that drives the
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 29
manipulation is related to the working memory requirements of the task (Kliegl, Mayr, &
Krampe, 1993; Verhaeghen et al., 1997). But this is still a very imprecise statement. For instance,
Verhaeghen et al. found that concurrent processing and storage (as in the lexical decision with
Tuholski, Laughlin, & Conway, 1999) is not sufficient to create a complexity jump. Recent work
on working memory (Miyake, Friedman, Emerson, Witzki, & Howerter, 2000) suggests that
different types of executive processes can be reliably distinguished; perhaps the complexity jump
is restricted to the involvement of a subset of those processes. Another possibility is that age
differences may be situated in the capacity of working memory rather than in the control
processes, and may be triggered by buffer overflow, or the concomitant switching of the focus of
attention (see also Verhaeghen & Cerella, 2002; Verhaeghen, Cerella, Bopp, & Basak, in press).
Also open is the question of why there is a dissociation between verbal and visuospatial
processing.
We are not claiming here that ‘three’ (at the observational level) or ‘four’ (at the
conceptual level) is the final answer to the question of the dimensionality of cognitive aging.
First, while the number of tasks explored here and in Verhaeghen et al. (2002) is large, they
hardly exhaust the repertoire. It is possible that inclusion of a larger number of tasks will expose
intermediate efficiency modes, or slowing factors larger than 4.0. A second limitation is that
efficiency modes are by definition tied to response times or other speedlike measures, whereas
many cognitive tasks (such as working memory span, episodic memory, reasoning ability) are
measured predominantly in the accuracy domain. One way to open such traditionally non-
speedlike tasks to efficiency-mode analysis is to examine the joint time-accuracy continuum for
these tasks (Kliegl, 1996; Kahana & Loftus, 1999; Verhaeghen, 2000).
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 30
References
Ashcraft, M.H., Donley, R.D., Halas, M.A., & Vakali, M. (1992). Working memory,
automaticity, and problem difficulty. In J.I.D. Campbell (Ed.), The nature and origins of
mathematical skills (pp. 301-329). Amsterdam: Elsevier Science. Basak, C., & Verhaeghen, P.
(2003). Subitizing speed, subitizing range, counting speed, the Stroop effect, and aging: Capacity
Braver, T. S., Barch, D. M. (2002). A theory of cognitive control, aging cognition, and
Brinley, J. F. (1965). Cognitive sets, speed and accuracy of performance in the elderly.
In A. T. Welford & J. E. Birren (Eds.), Behavior, aging and the nervous system (pp. 114-149).
Brysbaert, M., Bovens, N., D'Ydewalle, G., Van Calster, J. (1989). Turbo Pascal timing
routines for the IBM microcomputer family. Behavior Research Methods, Instruments, &
facts: the role of number comparison. Quarterly Journal of Experimental Psychology, 54A, 1005-
1029.
Cerella, J. (1990). Aging and information processing rate. In J. E. Birren & K. W. Schaie
(Eds.), Handbook of the psychology of aging (3rd. ed., pp. 201-221). San Diego, CA: Academic
Press.
Cerella, J., Poon, L. W., & Williams, D. H. (1980). Age and the complexity hypothesis. In
L. W. Poon (Ed.), Aging in the 1980’s (pp. 332-340). Washington, DC: American Psychological
Association.
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 31
Engle,R. W., Tuholski, S. W., Laughlin, J. E., Conway, A. R. A. (1999). Working memory,
Gavrilov, L. A., & Gavrilova, N. S. (1991). The biology of life span: A quantitative
Hale, S., & Myerson, J. (1996). Experimental evidence for differential slowing in the
Jenkins, L., Myerson, J., Joerding, J. A., & Hale, S. (2000). Converging evidence that
visuospatial cognition is more age-sensitive than verbal cognition. Psychology and Aging, 15,
157-175.
Kahana, M, & Loftus, G. (1999). Response time versus accuracy in human memory. In R.
J. Sternberg (ed.) The nature of cognition (pp. 323-384). Cambridge, MA: MIT Press.
Kausler, D. H. (1991). Experimental psychology, cognition, and human aging (2nd ed.).
competencies: Issues in growth and development (pp. 89-110). Mahwah, NJ: Erlbaum.
Kliegl, R., Mayr, U., & Krampe, R. T. (1994). Time-accuracy functions for determining
process and person differences: An application to cognitive aging. Cognitive Psychology, 26,
134-164.
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 32
Li, S.-C., & Lewandowsky, S. (1993). Intralist distractors and recall direction:
Li, S.-C., & Lewandowsky, S. (1995). Forward and backward recall: Different retrieval
processes. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 837-847.
Li, S.-C., Lindenberger, U., & Frensch, P. (2000). Unifying cognitive aging: From
Lima, S. D, Hale, S., Myerson, J. (1991). How general is general slowing? Evidence
Lindenberger, U., & Baltes, P. B. (1997). Intellectual functioning in old and very old age:
Cross-sectional results from the Berlin Aging Study. Psychology and Aging, 12, 410-432.
Mayr, U., & Kliegl, R. (1993). Sequential and coordinative complexity: Age-based
Mayr, U., Kliegl, R., & Krampe, R. (1996). Sequential and coordinative processing
dynamics in figural transformation across the life span. Cognition, 59, 61-90.
Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., & Howerter, A. (2000). The
unity and diversity of executive functions and their contributions to complex “frontal lobe” tasks:
Myerson, J., & Hale, S. (1993). General slowing and age invariance in cognitive
processing: The other side of the coin. In J. Cerella, J. Rybash, W. Hoyer, & M. L. Commons
(Eds.), Adult information processing: Limits on loss (pp. 115-141). San Diego, CA: Academic
Press.
Perfect, T. J., & Maylor, E. A. (2000). Rejecting the dull hypothesis: The relation between
method and theory in cognitive aging research. In T. J. Perfect & E. A. Maylor (Eds.), Models of
Erlbaum.
Salthouse, T. A., & Coon, V. E. (1994). Interpretation of differential deficits: The case of
aging and mental arithmetic. Journal of Experimental Psychology: Learning, Memory, and
Shepard, R.N, & Metzler, J. (1971). Mental rotation of three dimensional objects.
Siegler, R. S., & Shipley, C. (1995). Variation, selection, and cognitive change. In T.
Simon & G. Halford (Eds.), Developing cognitive competence: New approaches to process
modeling (pp. 31-76). Hillsdale, NJ: Erlbaum. Sliwinski, M., & Hall, C. B. (1998).
Constraints on general slowing: A meta-analysis using hierarchical linear models with random
factors. In D. Scarborough & S. Sternberg (Eds.), An invitation to cognitive science (Vol. 4):
Methods , models, and conceptual issues (pp. 703-863). Cambridge, MA: MIT Press.
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 34
Trick, L. M., & Pylyshyn, Z. W. (1994). Why are small and large numbers enumerated
Verhaeghen, P. (2000). The parallels in beauty’s brow: Time-accuracy functions and their
implications for cognitive aging theories. In T. J. Perfect and E. A. Maylor (Eds.), Models of
Verhaeghen, P., & Cerella, J. (2002). Aging, executive control, and attention: A review of
Verhaeghen, P., Cerella, J., Bopp, K. L., & Basak, C. (in press). Aging and varieties of
focus switching. In R. W. Engle, G. Sedek, U. von Hecker, & D. N. McIntosh (Eds.), Cognitive
Verhaeghen, P., Cerella, J., Semenec, S. C.; Leo, M. E., Bopp, K. L., &. Steitz, D. W.
(2002). Cognitive efficiency modes in old age: Performance on sequential and coordinative
Verhaeghen, P., Kliegl, R., & Mayr, U. (1997). Sequential and coordinative complexity in
time-accuracy function for mental arithmetic. Psychology and Aging, 12, 555-564.
Verhaeghen, P., Steitz, D. W., Sliwinski, M. J., & Cerella, J. (2003). Aging and dual-task
Watson, D. G., Maylor, E. A., & Manson, N. J. (2002). Aging and enumeration: A
selective deficit for the subitization of targets among distractors. Psychology and Aging, 17, 496-
504.
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 35
Author Notes
Paul Verhaeghen, John Cerella, and Chandramallika Basak, Department of Psychology and
This research was supported in part by a grant from the National Institute on Aging (AG-
16201). Correspondence concerning this article should be addressed to Paul Verhaeghen, john
Table 1.
Parameter estimates for fixed effects and variance components for random effects for the random
coefficients model (full model) (for model description, see text) for arithmetic and counting.
Difficulty refers to the number of elements to be counted and the number of arithmetic
Counting Arithmetic
Fixed Effects Coefficient SE p Coefficient SE p
Intercept 1167.03 196.61 <.0001 -4262.53 848.85 <.0001
Difficulty 380.90 27.80 <.0001 3247.13 192.87 <.0001
Complexity -1656.86 278.50 <.0001 3450.30 1200.46 0.0044
Agegroup -665.77 269.50 0.0139 1931.02 1188.40 0.1053
Difficulty*Complexity -10.63 36.67 0.7720 -1713.84 213.89 <.0001
Difficulty*Agegroup -92.51 38.11 0.0157 -1180.21 270.02 <.0001
Complexity*Agegroup 663.07 381.13 0.0827 -1375.87 1680.64 0.4137
Difficulty*Complexity*Ag 18.34 50.26 0.7154 967.56 299.45 0.0014
egroup
Within-person Random
Effect
Variance(Random) 97483 6617.58 2744962 209606
Note: Coefficient refers to the parameter estimate, Var refers to Variance, Cov refers to Covariance, and SE refers to
Standard Error.
Yij = γ00 + γ10 (Difficulty) + γ01 (agegroup) + γ20 (complexity) + γ30(Difficulty*complexity) + γ11 (Difficulty*agegroup)
+ γ21 (complexity*agegroup) + γ31(Difficulty*complexity*agegroup) + U0j + U1j + Rij
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 37
Table 2.
Parameter estimates for fixed effects and variance components for random effects for the
constrained model for counting (for model description, see text). Difficulty refers to the number
Within-person Random
Effect
Variance(R) 97513 6619.64
Yij = γ00 + γ10 (Difficulty) + γ01 (agegroup) + γ20 (complexity) + γ11 (Difficulty*agegroup) + γ21 (complexity*agegroup)
+ + U0j + U1j + Rij
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 38
Table 3.
Parameter estimates for fixed effects and variance components for random effects for the random
coefficients model for the matrices task (for model description, see text). Difficulty refers to the
number of distractors.
Baseline Arrow
Fixed Effects Coefficient SE p Coefficient SE p
Intercept 1478.04 138.53 <.0001 6552.51 422.31 <.0001
Difficulty 679.30 58.65 <.0001 2424.48 232.15 <.0001
Agegroup -676.84 211.25 0.0019 -2575.58 640.00 0.0001
Difficulty*Agegroup -287.15 88.68 0.0017 -1820.99 351.25 <.0001
Within-person Random
Effect
Variance(R) 510252 81706 1482976 232011
Note. Model used is: Yij = γ00 + γ10 (Difficulty) + γ01 (agegroup) + γ11 (Difficulty*agegroup) + U0j + U1j + Rij
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 39
Table 4.
Intercepts and slopes for the information-processing functions, separated by age groups, along
with the age and complexity effects in slope
Complexity effect in
Younger Older slope
Intercept Slope Intercept Slope Age effect in Younger Older
slope
Counting 0.97 1.03
Simple -492.53 296.10 -489.83 370.27 1.25
Complex 501.26 288.39 1167.03 380.90 1.32
Counting (constrained) 1 1
Simple -463.63 292.25 -529.71 375.59 1.29
Complex 472.36 292.25 1206.91 375.59 1.29
Arithmetic 1.57 2.12
Simple -257.08 1,320.64 -812.23 1,533.29 1.16
Complex -2331.51 2,066.92 -4,262.53 3,247.13 1.57
Matrices 1.53 3.57
Simple 801.20 392.15 1,478.04 679.30 1.73
Complex 3,976.93 603.49 6,552.51 2,424.48 4.02
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 40
Table 5. Tests for the hypothesis that 3 or 4 efficiency modes (i.e., separate age-related slowing factors) are present in the data: Original
hypotheses regarding efficiency modes, hypotheses translated into testable predictions about age differences in the ratios of information-
processing slopes, and results from the independent samples t-test. (For full explanation, see text.)
Figure 1. Schematic representation of stimuli for the matrices (top) and matrices-plus task
(bottom). The task requires deciding whether or not there are four points, denoted by X symbols,
that form the corner points of a virtual square. For the matrices-plus tasks, the participant has to
imagine X symbols in each of the boxes the arrows are pointing to. The sample stimuli come
from the positive set (i.e., there are four such points).
Figure 2. Means of response time and accuracy for the baseline and complex version of the three
tasks, separated by age group: counting (A and B), arithmetic (C and D), and matrices (E and F),
along with the best-fitting linear regression lines for the information-processing functions.
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 42
X X
X X X
X X X
X
X
X
< >
V
> V
> <
AGING, COMPLEXITY, AND EFFICICIENCY MODES – p. 43
5000 40
A B
4500 38
3000 32
2500 30
2000 28 Baseline, younger
1500 26 Baseline, older
1000 24 Complex. Younger
0 20
5 6 7 8 9 10 5 6 7 8 9 10
Difficulty level Difficulty level
20000 40
C D
18000 38
16000
Number correct (out of 40)
36
14000 34
Response time
12000 32
10000 30
8000 28 Baseline, younger
6000 26 Baseline, older
4000 24 Complex. Younger
0 20
3 4 5 6 7 8 3 4 5 6 7 8
Difficulty level Difficulty level
16000 40
E F
14000 38
Number correct (out of 40)
36
12000
34
Response time
10000 32
8000 30