Intelligence 50 (2015) 196–208

Contents lists available at ScienceDirect

Intelligence

Neural efficiency in working memory tasks: The impact of

task demand
Daniela Nussbaumer a,⁎, Roland H. Grabner b, Elsbeth Stern a
a
Institute for Behavioral Sciences, ETH Zurich, Clausiusstr. 59, 8092 Zurich, Switzerland
b
Department of Psychology, University of Graz, Universitätsplatz 2, 8010 Graz, Austria

a r t i c l e i n f o a b s t r a c t

Article history: Studies of human intelligence provide strong evidence for the neural efficiency hypothesis, which
Received 17 April 2014 suggests more efficient brain functioning (i.e., less or more focused activation) in more intelligent
Received in revised form 31 March 2015 individuals. Recent studies have specified the scope of the neural efficiency hypothesis by
Accepted 9 April 2015 suggesting that the relationship between brain activation and intelligence only holds true for
Available online xxxx
problems of moderate difficulty and can be altered through training and is only found in frontal
brain regions. We investigated the moderating roles of task difficulty and training on the neural
Keywords: efficiency phenomenon in the context of working memory (WM) training.
Intelligence In two studies of 54 participants (study 1) and 29 participants (study 2), cortical activation was
Working memory task
assessed by means of electroencephalography (EEG), or more precisely by means of event-related
Training
desynchronization (ERD) in the upper alpha band. ERD was assessed during the performance of
Task difficulty
Event-related desynchronization (ERD) WM tasks in a pre-test – training – post-test design, comparing groups of lower and higher
intelligence.
We found supportive evidence for the neural efficiency hypothesis only in moderately difficult
WM tasks in frontal brain regions, even in the absence of performance differences. There was no
effect of intelligence on the simple or highly demanding, adaptive WM tasks. In the latter task,
however, an intelligence-related difference emerged at the behavioral level, but training did not
modulate the relationship between intelligence and brain activation.
These results corroborate the moderating role of task difficulty in the neural efficiency hypothesis
in the context of WM demands and suggest that training does not impact the neural efficiency
phenomenon in the context of WM demands.
© 2015 Elsevier Inc. All rights reserved.

1. Theoretical background which showed less brain glucose metabolism in more intelligent
individuals while solving cognitive tasks. Haier and colleagues
According to the neural efficiency hypothesis, differences in stated, “Intelligence is not a function of how hard the brain works
intelligence become apparent in the degree of brain activation but rather how efficiently it works … This efficiency may derive
that occurs during problem solving, i.e., for more intelligent from the disuse of many brain areas irrelevant for good task
individuals, the correct answer comes with less brain activation performance as well as the more focused use of specific task-
than for less intelligent individuals (Haier et al., 1988). This relevant areas” (Haier, Siegel, Tang, Abel, & Buchsbaum, 1992b,
original hypothesis of neural efficiency was introduced in a pp. 415–416). In addition, with electroencephalography (EEG), it
positron emission tomography (PET) study, the results of was shown that event-related desynchronization (ERD) in the
upper alpha band, considered an index of cortical activation
⁎ Corresponding author. (Klimesch, Doppelmayr, Pachinger, & Ripper, 1997; Pfurtscheller
E-mail addresses: [email protected] (D. Nussbaumer), & Aranibar, 1977), is negatively related to intelligence (for a
[email protected] (R.H. Grabner), [email protected] (E. Stern). review, cf. Neubauer & Fink, 2009). However, although the

http://dx.doi.org/10.1016/j.intell.2015.04.004
0160-2896/© 2015 Elsevier Inc. All rights reserved.
 D. Nussbaumer et al. / Intelligence 50 (2015) 196–208 197

neural efficiency hypothesis has often been confirmed, moder- cortical activation discriminates between subjects with higher
ating factors have been identified, in particular, task difficulty and lower intelligence, which is in accordance with findings of a
and practice or learning (Neubauer & Fink, 2009). high involvement of frontal areas while solving WM tasks (Smith
Various studies have demonstrated that the relationship & Jonides, 1997). A discrepancy between frontal and parietal
between neural efficiency and intelligence may be altered by brain areas in the context of neural efficiency has also been
task difficulty (for an overview, see Neubauer & Fink, 2009). For revealed in three studies which reported less frontal activation
instance, Neubauer, Sange, and Pfurtscheller (1999) did not find for more intelligent participants and a tendency for more parietal
differences in brain activation between individuals with higher activation in the same participants (Gevins & Smith, 2000;
and lower IQ for simple (i.e., elementary cognitive) problems. Jaušovec & Jaušovec, 2004 and Rypma et al., 2006). Thus, even
The authors therefore concluded that a certain level of task though an interplay of frontal and parietal brain areas is
difficulty is required for a corroboration of the neural efficiency discussed to be important for intelligence (cf. the parieto-
effect. A different picture emerged in a study EEG measures frontal integration theory by Jung & Haier, 2007); neural
while solving the Advanced Progressive Matrices Test (RAPM; efficiency in terms of a negative brain–intelligence relationship
Raven, 1990). Specifically, a negative relation between brain has predominantly been found in frontal brain regions.
activation and intelligence was found for the easier items only, To summarize, several studies have provided support for the
while for the more difficult ones, the opposite relationship was neural efficiency hypothesis mainly for frontal brain areas, but
observed (Doppelmayr et al., 2005a). According to Neubauer and have also shown that task difficulty and training can moderate
Fink (2009), these results do not necessarily contradict each the relationship between intelligence and brain activation. There
other. The authors conclude, rather, that when more effort is is, however, a paucity of studies in which task difficulty and
required, more intelligent participants invest their available training were combined in a comprehensive design. We
resources, resulting in both higher cortical activation and better conducted such a study involving WM training.
achievement. Thus, it seems that in complex tasks, more There is a wide agreement that WM is a core of human
intelligent individuals invest more cortical resources, resulting intelligence. Numerous studies have demonstrated substantial
in a positive correlation between cortical activation and correlations between achievement on WM tasks and IQ
performance. In contrast, for moderate tasks, more intelligent (e.g., Conway, Cowan, Bunting, Therriault, & Minkoff, 2002;
individuals require less cortical resources to achieve the same Conway, Kane, & Engle, 2003; Engle, Tuholski, Laughlin, &
performance as less intelligent individuals, resulting in a negative Conway, 1999; Kane et al., 2004; Kyllonen & Christal, 1990). It
relation between cortical activation and performance. can therefore be expected that the use of brain imaging while
Individual task difficulty, however, can be altered by practice, solving WM tasks will particularly highlight the impact of
and based on the neural efficiency hypothesis, practice-related intelligence on neural activation. Moreover, the difficulty of
changes in brain activation may also be a function of intelligence. WM tasks can be varied in a systematic and transparent way,
This has, in fact, been confirmed in two studies, which found a for instance, by modulating demands for interference resolu-
stronger decrease in activation after training for individuals with tion or the amount of load. This allows the study of how the
higher intelligence (Haier et al., 1992b; Neubauer, Grabner, relationship between brain activation and intelligence may be
Freudenthaler, Beckmann, & Guthke, 2004). The role of practice moderated by task difficulty. Lastly, there is overwhelming
in the neural efficiency phenomenon has also become salient in evidence for the trainability of many types of WM tasks. As a
investigations of experts in different domains who had achieved result of repeated practice, the solution rate increases while the
their expertise level through long-term training (Grabner, solution time goes down. Whether training effects transfer to
Neubauer, & Stern, 2006; Grabner, Stern, & Neubauer, 2003). other WM tasks, thereby demonstrating the malleability of
These studies revealed that neural efficiency (in terms of more WM functions, is hotly debated in psychology, and reviews and
focused brain activation) is a function not only of intelligence but meta-analyses have provided reasons to doubt broader transfer
also of expertise. For instance, Grabner et al. (2006) compared effects (Chein & Morrison, 2010; Melby-Lervåg & Hulme, 2013;
the brain activation of individuals with lower and higher Shipstead, Redick, & Engle, 2012). Taken together, the advan-
intelligence as well as with lower and higher expertise in tages of WM tasks in investigations of neural efficiency are
tournament chess while solving chess-related tasks. They found threefold: First, WM is seen as a basis of human intelligence.
independent impacts of intelligence and expertise level on brain Second, the difficulty level can be manipulated gradually in that
activation. As expected, brighter individuals (independently of more or less WM load is incorporated into tasks. Third, WM
their expertise) displayed lower overall brain activity than their activities are well represented in cortical activation, i.e., it is
less intelligent peers. In addition, experts showed a lower frontal known that while solving WM tasks, there is a high involve-
and more focused brain activation pattern compared to novices ment of frontal areas (Smith & Jonides, 1997). The present
(i.e., individuals with lower degree of expertise). study consists of two training studies in which we assess brain
Also with regard to brain areas only partial support for activation (in terms of alpha ERD) in frontal areas before and
the neural efficiency hypothesis has been found. Neubauer after a three-week WM training in adult students differing in
and Fink (2009) summarize that effects of neural efficiency, intelligence. The two studies differ in the level of task difficulty
i.e. the expected negative brain–intelligence relationship or WM load.
has been observed for frontal (but not for parietal) brain In study 1, we administered a WM-training with moderate
areas. For instance, Neubauer et al. (2004) found the strongest complexity focusing on interference resolution. Interference
intelligence-related differences during reasoning tasks in frontal resolution, which is the ability to select information among
areas, more specifically in the prefrontal cortex, an area most competing alternatives, is seen as a key function of WM
strongly associated with reasoning processes. Similarly, Gray, (e.g., Nee, Wager, & Jonides, 2007). Specifically, the participants
Chabris, and Braver (2003) reported that for WM tasks prefrontal in the interference group (i.e., experimental group) practiced
198 D. Nussbaumer et al. / Intelligence 50 (2015) 196–208

WM tasks requiring the resolution of proactive interference. problems of the training software on their home computers. Two
These participants were contrasted with a control group who other participants were excluded due to non-adherence to the
worked on parallelized but very simple tasks that were not training paradigms, resulting in 54 participants for subsequent
expected to challenge WM. Based on current evidence, we analysis. All of the participants were right-handed and without
expected that the more intelligent participants would show any medical or psychological diseases (both determined by self-
higher neural efficiency (less brain activation) in frontal areas report). The participants were paid for their participation in the
than those with lower intelligence while solving these WM tasks study.
of moderate complexity. In addition, we hypothesized a decrease
in brain activation in the course of the training that is expected to 2.1.2. Procedure
be larger for the more intelligent participants. In the control Participants were randomly assigned to either the interfer-
group, in contrast, we did not expect a relationship between ence group or the control group. The interference group trained
intelligence and brain activation before or after training or that at three different non-adaptive tasks with a high number of
the changes in brain activation would depend on the individuals' interference trials and a moderate level of WM load, whereas
intelligence level. In other words, we hypothesized that task the control group trained at three different tasks for which only
difficulty influences the corroboration of the neural efficiency 1 item had to be memorized at a time. The participants trained
hypothesis in WM tasks in that the negative relationship 5 days a week during a 3-week period for half an hour daily on
between intelligence and frontal brain activation can be found their home computer. To check for the plausibility of training
for moderately difficult tasks only. In study 2, we employed a gains, the participants were required to send their training
different WM training with high complexity to investigate the results immediately after completing the training session at
neural efficiency hypothesis in an adaptive task that maximized home. The first and last training session took place at the first
the individual task difficulty. Specifically, the training in study 2 author's institution, and the tasks were performed while
had an identically high demand for resolution of proactive measuring EEG.
interference as in the interference group of study 1 but varied in
two aspects. First, the training was adaptive and therefore 2.1.3. Material
ensured that WM load was constantly high and that the
participants trained at their limits at all times. The task ensured 2.1.3.1. Training paradigms. Both of the groups trained at three
the continuous engagement of executive processes and allowed non-adaptive WM tasks, with each task taking approximately
only minimal development of task-specific strategies but 10 min on each training day in a counterbalanced order. The
encouraged solution strategies that could be applied in varying solution time and the proportion of correct answers (i.e., the
situations. Second, a dual task component was incorporated so correct answers divided by the number of trials) were measured
that executive processes were required for managing the two for all tasks.
tasks simultaneously as well as each task separately. By this task The interference group practiced two recognition tasks (one
composition we attempted to ensure that automatic processes with letters and one with faces) and one n-back task. All of the
were minimized and that the dual task group trained with tasks of the interference group were characterized by a focus on
maximal WM load and with a maximal demand for interference resolution of proactive interference and by moderate WM load.
resolution. The participants had to allocate relevant information In the two recognition tasks, a fixation point (1500 ms) was
to one of both tasks (dual task) and to the correct trial (resolution followed by a target set of four items that were arranged in a
of proactive interference) while inhibiting irrelevant informa- square configuration (faces were presented for 1500 ms and
tion. In light of the adaptive nature of the task, we did not expect letters were presented for 500 ms). After a 3000 ms retention
differences in neural activation in the pre-test between partic- interval, during which only a fixation point was presented, a
ipants of differing intelligence levels. Intelligence differences, single probe appeared for a maximum of 1500 ms (see Fig. 1).
however, should emerge at the behavioral level in that the more On 50% of the trials, the probes were of the current target set;
intelligent individuals should display better performance com- on the other 50% of the trials, the probes were new, i.e., non-
pared to their less intelligent peers. A similar result pattern can matching. Of these nonmatching probes, two-thirds were
be expected to hold true for the post-test. Finding no training- interference trials in that they appeared in the target set of
related change (and no differences between the more and less the previous trial or the previous two trials, and only one-third
intelligent individuals) in brain activation would be in line with of the probes were non-recent. The participants had to press
our expectation, as the individual level of cognitive demand two different buttons to indicate a match vs. a mismatch. The
remains constantly high throughout training. solution times were measured from probe onset to the button
press, and the proportion of correct answers was calculated.
2. Study 1 The stimuli in verbal tasks consisted of 19 consonants (without
l and y). As face stimuli, 20 digitized gray-scale portraits (50%
2.1. Method of study 1 males, 50% females) of ordinary people were used. In addition,
the interference group practiced one n-back task in a 3-back
2.1.1. Participants version in which a series of letters was shown at a rate of
A total of 54 healthy students of science- and humanities- 2000 ms per letter, with a 2000 ms interstimulus interval (see
related fields from three Swiss universities completed the Fig. 2). When the current letter matched the letter that
study, which included a pre-and a post-test session at the first appeared three positions earlier, the participants were required
author's institution and training sessions at home (n = 54, to press a button as quickly and accurately as possible. Half of
Mage = 23.4, SD = 3.5, 24 males, 30 females). From the initial 61 the letters were target trials (matched the letter three positions
participants, five participants were excluded due to installation earlier), and half were non-target trials. Of the non-target trials,
 D. Nussbaumer et al. / Intelligence 50 (2015) 196–208 199

a) b)
time fixation point time
(1,500 ms)

stimuli

(1,500 ms for faces,

500 ms for letters)

retention interval

(3,000 ms)

comparison stimulus

(until key press,

max. 1,500 ms)

Fig. 1. a) Schematic view of the letter recognition task (4-letters task). Note that the comparison stimulus is written in capital letters to minimize purely perceptual
processing. The letter-matching task (1-letter task) is analogous, except that the stimulus only consists of one letter. b) Schematic view of the face recognition task (4-
face task). The face-matching task (1-face task) is analogous, except that the stimulus only consists of one face.

a)
time fixation point

(2,000 ms)

stimulus
k
(2,000 ms)

fixation point

(2,000 ms)

b)
Sequence a f f a m m s m

3-back x x x hit x x x hit

1-back x x hit x x hit x x

Fig. 2. Schematic view of the n-back training tasks. a) The same timing applied for the 3-back and the 1-back task b) possible sequence of the 3-back and the 1-back task
(upper line), with hits in the 3-back version (middle line) and the 1-back version (lower line).
200 D. Nussbaumer et al. / Intelligence 50 (2015) 196–208

three quarters were interference trials, i.e., a letter presented

0.6 (0.09)

0.60 (0.09)

0.91 (0.06)
0.91 (0.03)
0.91 (0.07)
0.59 (0.1)
matched the letter two, four, or five trials before. All of the tasks
of the control group were tightly parallelized to the tasks of the

Post
interference group but induced only marginal WM load. In all

Item recognition
(4-letter task)
3 tasks, only 1 item had to be memorized at a time, and no

0.64 (0.07)
0.7 (0.11)
0.67 (0.09)

0.84 (0.14)
0.89 (0.05)
0.86 (0.10)
interference trials were incorporated into the task. The face-
and letter-matching tasks used the same stimuli, timing and

Pre
response format as the face- vs. letter-recognition task in the
interference group (see Fig. 1). However, the memory set
consisted of one item only, so the participants simply indicated

0.75 (0.13)
0.73 (0.12)

0.70 (0.09)
0.76 (0.14)
0.73 (0.12)
0.7 (0.1)
whether or not the probe matched an item. Analogous to the 3-
back task, the participants solved a 1-back task with the same

Post
timing. In this task, the participants had to indicate whether

Item recognition
a letter was repeated or not (see Fig. 2). In agreement with

(4-face task)
the WM literature (e.g., Shipstead, Redick, & Engle, 2010), a

0.86 (0.11)
0.96 (0.16)
0.92 (0.15)

0.65 (0.07)
0.7 (0.11)
0.68 (0.10)
task with a WM load of three items or less is not an actual
WM task; this group was therefore considered a control group.

Pre
The tight parallelization of the interference and the control
group allowed us to straightforwardly attribute potential

Mean and standard deviation. a) For solution time per task for pre- and post-testing. b) For proportion of correct answers per task for pre- and post-testing.

0.47 (0.13)
0.46 (0.11)
0.47 (0.11)

0.94 (0.08)
0.93 (0.09)
differences developing over the course of the training to the

0.93 (0.1)
training itself.

Post
Interference group
2.1.3.2. Paper–pencil measures. As a fluid intelligence test, the
well-established Advanced Progressive Matrices Test (RAPM,

3-back task

0.6 (0.12)
0.6 (0.13)
0.60 (0.12)

0.71 (0.12)
0.81 (0.15)
0.77 (0.15)
Set II) by Raven (1990) was administered. Half of the items
were presented in the first session, and the other half were

Pre
presented in the session after the training. As all participants
solved the RAPM twice, once at pre-testing before and once
at post-testing after training, an even-odd split version was

0.48 (0.07)
0.45 (0.07)
0.47 (0.07)

0.96 (0.03)
0.95 (0.06)
0.96 (0.05)
presented (participants were randomly assigned to the specific
order). As only half the items were presented at each
Post
measurement point, no IQ-value could be calculated.
Item matching
(1-letter task)

0.56 (0.08)
0.54 (0.11)
0.55 (0.09)

0.96 (0.05)
0.97 (0.03)
0.96 (0.04)
2.1.3.3. Mental effort rating. Participants had to judge the
subjective mental effort of each training task on the mental
Pre

effort rating scale (Paas, 1992; Paas, Tuovinen, Tabbers, & Van
Gerven, 2003) in order to determine the subjectively perceived
task difficulty. This measure allows to judge whether the level
0.64 (0.13)
0.58 (0.09)
0.61 (0.11)

0.95 (0.06)
0.96 (0.05)
0.95 (0.05)
of task complexity in the two training groups differs as
intended. Specifically, the differentiation between moderate
Post

and high WM load has been established via this rating scale
a) Solution time (measured in seconds) per task for pre- and post-testing

(Paas, 1992, p. 431; minimum 1, maximum 9). Values between

Item matching
(1-face task)

4 and 6 are qualified as moderate, values below as low and

0.66 (0.12)
0.70 (0.11)

0.97 (0.02)
0.96 (0.02)
0.97 (0.02)
0.72 (0.1)

b) Proportion of correct answers per task for pre- and post-testing

values above as high.

Pre

2.1.3.4. EEG. The EEG assessment was conducted using an

ActiveTwo-System (BioSemi, Amsterdam, The Netherlands).
0.42 (0.06)
0.4 (0.06)
0.41 (0.06)

0.97 (0.04)
0.97 (0.03)
0.97 (0.04)

Event-related desynchronization/synchronization (ERD/ERS)

was calculated for the upper alpha band (10–13 Hz) (Klimesch,
Post

1999; Neubauer, Fink, & Grabner, 2006). Sixty-four scalp

electrodes were placed according to the extended 10–20 system,
Control group

and four additional electrodes were placed horizontally and

1-back task

0.48 (0.05)
0.46 (0.08)
0.47 (0.06)

0.86 (0.21)
0.99 (0.01)
0.91 (0.17)

vertically around the eyes to measure the electrooculogram

(EOG). The EEG and EOG were sampled at 256 Hz. The EEG was
Pre

recorded during resting state for 3 min and during deliberate,

instructed eye movement to allow for the automatic correction
of eye movement artifacts while performing tasks. To eliminate
High IQ-group

High IQ-group
Low IQ-group

Low IQ-group

contamination artifacts, a band-pass filter was administered

(between 0.5 and 45 Hz), EOG artifacts were reduced (automatic
Overall

Overall
Table 1

regression method; Schloegl et al., 2007), and all sequences were

visually inspected for artifacts. For a detailed description of data
 D. Nussbaumer et al. / Intelligence 50 (2015) 196–208 201

Table 2

1.71 (10.85)
–0.18 (18.94)
1.81 (15.09)
–4.51 (16.48)
MANOVA results for task performance in study 1 with the between-subject

Item recognition (4-letter

factor intelligence-group (IQ-group: lower vs. higher intelligence) and the
within-subject factor time (pre- and post-testing) for all trained tasks.

T13
Main effect Main effect Interaction
time IQ-group time ∗ IQ-group

–5.48 (13.99)

–3.72 (20.93)
–3.41 (17.21)
–6.73 (17.5)
Solution time
Control group F(2,20) = 10.52 F(2,20) = 0.45 F(2,20) = 0.18

task)
p b .001 n.s. (p = .72) n.s. (p = .90)

T1
η2p = .62 η2p = .07 η2p = .03
Interference F(2,21) = 22.34 F(2,21) = 2.08 F(2,21) = 2.52

–7.76 (12.05)
–0.24 (11.63)
p b .001

–7.41 (9.54)
–0.67 (5.96)
group n.s. (p = .13) n.s. (p = .09)
η2p = .76 η2p = .23 η2p = .27

Item recognition (4-face

Proportion of correct answers

T13
Control group F(2,20) = 0.96 F(2,20) = 0.91 F(2,20) = 2.07
n.s. (p = .43) n.s. (p = .45) n.s. (p = .14)

–5.88 (11.5)
0.27 (7.88)
–9.33 (9.33)
–1.12 (5.78)
η2p = .13 η2p = .13 η2p = .25
Interference F(2,21) = 11.17 F(2,21) = 0.97 F(2,21) = 1.20
group p b .001 n.s. (p = .43) n.s. (p = .34)

task)
η2p = .62 η2p = .12 η2p = .15

T1

5.92 (10.67)
4.11 (10.18)
–1.21 (7.94)
–0.85 (6.31)
analyses, see Grabner and De Smedt (2011) and De Smedt,

T13
Grabner, and Studer (2009). The ERS/ERD ratio was calculated

Interference group
for correctly solved trials for the upper alpha frequency band

–2.02 (10.82)
6.34 (11.51)
Mean and standard deviation. a) For ERD frontal per task for pre- and post-testing. b) For ERD parietal per task for pre- and post-testing.

–2.85 (8.23)
2.81 (7.42)
(10–13 Hz). A reference interval (R) comprised the time during

3-back task
the fixation interval (from 500 to 2500 ms after trial onset). The
activation interval (A) is defined as the period from problem

T1
presentation the response. The 125 ms before response were
excluded to eliminate motor artifacts. The band powers of A and

14.46 (26.72)
–7.79 (15.79)
–2.43 (16.66)
–15.92 (14.37)
R were obtained by squaring and averaging the artifact-free EEG
signal separately for the sequences R and A. The ERS/ERD ratio Item matching (1-letter task)

was computed as follows: %ERS/ERD = [(A ∗ R) / R]100. Negative

values (ERD) indicate desynchronization and a decrease in T13
power. Positive values (ERS) indicate synchronization and an

–5.75 (15.68)
–11.82 (11.48)
0.07 (23.13)

–2.45 (20.3)
increase in power.
For the statistical analyses, the %ERS/ERD was topograph-
ically aggregated by averaging electrodes. In light of the crucial
T1

role of frontal brain areas in the context of neural efficiency

(Neubauer & Fink, 2009) our hypotheses focus on frontal areas,
1.09 (15.09)
–1.29 (11.06)
1.21 (13.68)
–1.57 (10.15)

i.e. we expect differences to be limited to these parts of the

Item matching (1-face task)

brain. We therefore calculated a frontal %ERS/ERD-measure by

averaging the following electrodes: Fp1, AF7, AF3, F7, F5, F3, F1,
T13

FC5, FC3, FC1 Fp2, AF8, AF4, F8, F6, F4, F2, FC6, FC4, and FC2 (for
the distribution of the electrodes see, e.g., Grabner & De Smedt,
–9.64 (14.43)

–0.04 (12.06)
–8.23 (15.25)
–0.02 (8.81)

2011). Since parietal areas are also critical in WM tasks (Gevins

& Smith, 2000; Jaušovec & Jaušovec, 2004 and Rypma et al.,
2006) and have been considered to be important in the
T1

parieto-frontal integration theory of intelligence (Jung &

b) ERD parietal per task for pre- and post-testing
a) ERD frontal per task for pre- and post-testing

Haier, 2007), we also calculate a parietal %ERS/ERD-measure

2.96 (14.84)
0.46 (8.83)
–3.47 (8.48)

–0.41 (7.2)

by averaging the following electrodes: CP5, CP3, CP1, P7, P5, P3,
P1, PO7, PO3, O1 CP6, CP4, CP2, P8, P6, P4, P2, PO8, PO4, and O2.
T13

Finding no differences between more and less intelligent

participants in the parietal areas, however, is compatible with
Control group

–4.01 (10.05)

the neural efficiency hypothesis. Therefore, running separate

–7.07 (6.99)

–3.07 (9.94)
1-back task

–2.06 (8.9)

analyses for both brain areas without correction for multiple

comparisons is justified.
T1

2.1.3.5. Statistical analysis. Solution time on the three training

IQ-group

tasks was first analyzed in a multivariate analyses of variance for

Table 3

High
High

Low
Low

repeated measures (MANOVA, separately for the interference

group and the control group, as tasks did not overlap), with
202 D. Nussbaumer et al. / Intelligence 50 (2015) 196–208

intelligence group (lower intelligence, higher intelligence) as the However, the goal of finding tasks of medium difficulty for
between-subjects factor and solution time as the dependent the interference group was only partly approached. For the 4-
variable. In the case of a significant MANOVA, we further letter task, ceiling effects could be seen to emerge already in the
analyzed each training task separately with univariate analyses pre-test for solution rates (N .80) as well as for solution times,
(ANOVAs for repeated measures, with intelligence group as the which only decreased from 0.67 s in the pre-test to 0.60 in the
between-subjects factor and solution time as the dependent post-test. While the 4-letter task turned out to be too easy
variable). The same procedure was chosen to analyze the already in the pre-test, the 3-back task appeared to be
proportion of correct answers, frontal cortical activation (ERD- particularly amendable by practice and therefore less challeng-
frontal) and parietal cortical activation (ERD-parietal). ing in the post-test than assumed. The mean solution rate
exceeded .90 in the post-test. A comparison of the solution
2.2. Results of study 1 times for the 1-back task (control group) and the 3-back task
(intervention group) underlined the effects of practice: In the
2.2.1. Behavioral data pre-test, it took clearly longer to solve the 3-back task (0.60 s)
The interference group and the control group did not differ than the 1-back task (0.47 s), d = 1.4. In the post-test, however
in their initial intelligence level (t(52) = –.82, p = .42), and solution times grew closer together (1-back task: 0.41 s, 3-back
their intelligence level after completion of the training also did task 0.46 s, d = .7). Comparing the solution times between the
not differ (t(52) = –.46, p = .65). Taking into account all 1-face task and the 4-face task revealed effect sizes of d N 1 for
participants of the two groups, the median of the RAPM raw the pre- as well as the post-test. The same was true for a
scores at pre-testing was (n = 54, M = 12.26, SD = 2.32) and comparison of the 1-letter task and the 4-letter task.
at post-testing was (n = 54, M = 12.65, SD = 2.38). Therefore, The proportion of correct answers and the solution times for
two intelligence groups were formed by a median split of the the 4-face task, however, showed that we were able to create
RAPM raw scores1 at pre-testing (splitting at 12.26 points; problems of medium difficulty for the entire intelligence range
lower intelligence-group: n = 28, M = 10.46, SD = 1.45; that can be expected to highlight neural efficiency effects. It is
higher intelligence-group: n = 26, M = 14.19 SD = 1.27; therefore justified to investigate whether differences in brain
resulting in an effect size of d = 2.74). It should be noted that activation can be traced back to intelligence and training, which
our sample consisted of university students and that, with a will be done in the next section.
mean of 14 out of 18 maximal points for the higher intelligence
group, they performed at a high level. In addition, the lower 2.2.2. Neural efficiency
intelligence group, with a mean performance of 10.5 out of 18, The MANOVA on frontal activation of the interference group
performed at a respectable level, which was lower than the revealed differences in cortical activation between the intelli-
other group but still at an average level. gence groups (main effect of intelligence-group (F(3,20) = 3.37,
Table 1 depicts descriptive statistics for solution times and p = .039, η2p = .336), see Table 4; for descriptive statistics, see
proportion of correct answers, and Table 2 presents the results Table 3). Subsequent ANOVAs showed that these differences
of MANOVAs. All of the simple tasks of the control group occurred due to a significant main effect of intelligence group in
(1-letter, 1-face, 1-back) show ceiling effects for the proportion the recognition task 4-faces (F(1, 24) = 7.38, p = .012, η2p =
of correct answers already in the pre-test. This result indicates .24); for the other tasks (4-letters and 3-back we found p N .10.).
that these tasks were – as intended – not challenging for the As we are doing multiple comparisons at the level of the follow-
entire sample. For the proportion of correct answers in the up ANOVAs we are taking this into consideration by a Bonferroni
interference group, the MANOVA revealed a main effect of adjustment. This lowers the accepted p-value from p b 0.05 to
time, and follow-up analyses showed that this was due to gains p b 0.016, which is higher than the obtained value of p = .012 in
in the 3-back and the 4-face tasks. Contrary to our intention, the ANOVA of the 4-face task.
ceiling effects also partially appeared in the interference group: Finding neural efficiency effects for both the 4-letter and the
For the 4-letter task, the proportion of correct answers 3-back tasks was not very likely, given the ceiling effects
exceeded .80 in the pre-test and .90 in the post-test. A similar discussed in the previous section. It was therefore thoroughly
ceiling effect was observed for the 3-back task in the post-test. plausible that the effect solely occurred in the 4-face task. In
For the solution times, a MANOVA revealed a main effect of order to ensure that we do not rely on an artifact we decided to
time for the control group as well as for the interference group corroborate this finding by two additional analyses.
(see Table 2; for descriptive statistics, see Table 1). Follow-up First, an extreme group approach was chosen for running an
ANOVAs for each task revealed that the mean solution time analysis of variance. The groups were formed by including the
decreased significantly for all tasks of the interference and the third (33%) with highest and with lowest intelligence instead of
control group as a result of practice. No significant effects of forming the groups by a median split. The findings we got from
“intelligence” and no interactions between time and intelligence median split were confirmed in a MANOVA with a considerably
appeared. Thus, as intended, in neither group were intelligence higher effect size (F(3, 12) = 6.33, p = .008, η2p = .613).
differences related to any behavioral measure. Therefore, no Subsequent ANOVAs showed again differences due to a
trade-off between performance and neural efficiency can be said significant main effect of intelligence group in the recognition
to be responsible for significant differences in ERD between task 4-faces (F(1, 15) = 12.63, p = .003, η2p = .46); for the
intelligence groups. other tasks (4-letters and 3-back) we found p N .10 Again, the
result holds true for Bonferroni corrections which require a
1
For the method of median split to form intelligence groups see e.g.
p b 0.016.
Neubauer, Freudenthaler, and Pfurtscheller (1995) for the RAPM and Neubauer Second, due to the high standard deviations of the EEG-data
and Fink (2003) or Doppelmayr et al. (2005b) for other intelligence tests. (which are comparable to other studies; e.g. Neubauer et al.,
 D. Nussbaumer et al. / Intelligence 50 (2015) 196–208 203

-2

ERD-frontal -4

-6 lower intelligence
higher intelligence
-8

-10

-12

-14
pre post

Fig. 3. Graph of the face recognition task (4-faces). Values for ERD-frontal on the y-axis, time on the x-axis. An ANOVA for the face recognition task revealed a main effect
of group (F(1, 24) = 7.38; p = .012; η2p = .24) with no main effect of time and no interaction. The error bars represent the standard error of the mean.

2004) we decided to double-check the findings by running a With respect to the parietal cortical activation of the
nonparametric test. The Kruskal–Wallis test revealed a signifi- interference group, multivariate analyses revealed no main effect
cant result for the face recognition task (H(1) = 5.85, p = . 016 of intelligence, no main effect of time and no interaction.
with a rank of 9.6 for the lower intelligence group and with 16.9 Therefore, for parietal activation, no differences between the
for the higher intelligence group). Thus we confirmed a robust participants of different intelligence levels were found, and there
neural efficiency effect for the 4-face task, which is highlighted in was no progress over time and no difference in the progress
Fig. 3. between the groups. Thus, while the intelligence groups showed
However, while more intelligent participants showed less different levels of cortical activation in the frontal regions, we did
cortical activation while solving the 4-face task than the less not observe any differences in parietal areas. In the control group,
intelligent ones, these activation differences between the again, no differences between the intelligence groups or any
intelligence groups did not change after training. They rather progress over time occurred for parietal activation.
remained stable from pre- to post-testing, as there was no
interaction between intelligence group and time. There also 2.3. Discussion of study 1
was no main effect of time. For the control group, none of the
effects in the multivariate analyses reached significance (see In study 1, we sought to confirm findings that the neural
Table 3 and for descriptive statistics see Table 4), which means efficiency hypothesis holds true for moderately difficult tasks
there was no difference between the intelligence groups and no but not for simple ones. We chose WM tasks because they can
training-related change. be expected to be particularly strongly related to intelligence.
In fact, we show that more and less intelligent participants did
not differ with respect to neural activation when they were
processing tasks that only slightly stressed WM. For tasks that
Table 4
MANOVA results for alpha ERD in study 1 with the between-subject factor moderately strained WM, however, we confirmed the neural
intelligence-group (lower vs. higher intelligence) and the within-subject factor efficiency hypothesis. Specifically, the more intelligent partic-
time (pre- and post-testing), including all trained tasks. ipants showed less frontal brain activation than the less
Main effect Main effect Interaction time * intelligent participants, even though there were no perfor-
time IQ-group IQ-group mance differences between groups. It is worth noting that our
ERD frontal
entire sample was above average on the IQ scale. Therefore, we
Control group F(2,19) = 1.36 F(2,19) = 0.99 F(2,19) = 2.23 compared highly intelligent individuals against moderately
n.s. (p = .28) n.s. (p = .41) n.s. (p = .19) high intelligent individuals. Support for the neural efficiency
η2p = .18 η2p = .14 η2p = .26 hypothesis in this sample could be considered as strong
Interference F(2,20) = 0.70 F(2,20) = 3.37 F(2,20) = 1.01
evidence of its validity.
group n.s. (p = .57) p b .05 n.s. (p = .41)
η2p = .09 η2p = .37 η2p = .13 In line with the review by Neubauer and Fink (2009) we
found frontal but not parietal differences between the higher
ERD parietal
and lower intelligent participants.
Control group F(2,17) = 2.27 F(2,17) = 2.22 F(2,17) = 1.70
n.s. (p = .08) n.s. (p = .12) n.s. (p = .20) Contrary to our expectation, we found neither training
η2p = .33 η2p = .28 η2p = .23 effects nor interactions with intelligence for any of the tasks on
Interference F(2,20) = 2.20 F(2,20) = 1.00 F(2,20) = 2.38 the neural level. This stands in contrast to the results of
group n.s.(p = .12) n.s. (p = .41) n.s. (p = .10) Neubauer et al. (2004), who found a reduction in cortical
η2p = .25 η2p = .13 η2p = .26
activation even after a short training, and to the results of Haier
204 D. Nussbaumer et al. / Intelligence 50 (2015) 196–208

a) visual stimuli

time
fixation point (2,500 ms)

m simultaneous presentation of visual

+
and auditory stimuli (500 ms)

fixation point (2,500 ms)

simultaneous presentation of visual

+ k
and auditory stimuli (500 ms)

b)
Upper left / a Middle left / b Down right / a Middle left / m Upper right / s

x/x x/x x / hit hit / x x/x

Fig. 4. Schematic view of the n-back training tasks a) dual n-back task b) possible sequence of the dual n-back task (upper line) with hits (lower line) in the version of a
2-back task.

et al. (1992a) and Haier et al. (1992b), who found very strong 3.1.2. Procedure and EEG
behavioral training effects, as well as a reduction in cortical All of the participants of study 2 were assigned to a dual task
activation, after a four to eight weeks training with Tetris. training in which they trained in a dual n-back task 5 days a
However, we found clear learning effects on the behavioral week during a 3-week period for half an hour daily on their
level in terms of increased solution rates and reduced solution home computer. The procedure, as well as the EEG assessment
times. It seems that after three weeks of daily training, the and analyses, were parallel to study 1.
burden on WM had not been changed sufficiently to affect our
EEG measures. Thus, there may be a difference between our WM 3.1.3. Material
tasks and tasks with a clear demand on pattern recognition and
chunking as applied by Neubauer et al. (2004) and Haier et al. 3.1.3.1. Training paradigms. The participants trained with an
(1992b). adaptive dual n-back task with high WM load and a large
number of interference trials (similar to Jaeggi, Buschkuehl,
3. Study 2 Jonides, & Perrig, 2008). The participants were presented with a
sequence of simultaneous visual and auditory stimuli (see
3.1. Method of study 2 Fig. 4). The visual stimuli consisted of squares arranged in a
square configuration, resulting in eight different locations. The
3.1.1. Participants auditory stimuli were presented through headphones and
A total of 29 healthy students of science- and humanities- consisted of one of eight consonants. The stimuli were
related fields from three Swiss universities completed the study, presented at a rate of 3000 ms, with a stimulus length of
including all training sessions as well as pre-and post-testing (for 500 ms and an interstimulus interval of 2500 ms. Every block
study 2: n = 29, Mage = 23.7, SD = 2.7, 17 males, 12 females). began with a central fixation cross. The participants had to
From the initial 30 participants, one participant had to be indicate whenever one or both of the presented stimuli
excluded due to non-adherence to the training paradigms and matched the one that was presented n positions before. Each
transfer sessions at the institute, resulting in the 29 participants block consisted of 20 plus n trials. Of these, five trials were no
for subsequent analysis. Again, all participants were right- hits, five trials were targets in the auditory modality, five trials
handed and without any medical or psychological diseases. were targets in the visual modality and five trials were targets
These individuals were also paid for their participation in the for both modalities simultaneously. This task is characterized
study. by a high demand for interference resolution in that lures were
 D. Nussbaumer et al. / Intelligence 50 (2015) 196–208 205

Table 5
Means and standard deviations for the n-back level, ERD-frontal, and ERD-parietal for pre- and post-testing.

Behavioral measure: n-back level ERD frontal ERD parietal

Pre Post Pre Post Pre Post

Low 1.61 (0.30) 3.02 (0.64) –12.65 (12.88) –7.31 (18.94) –11.95 (14.91) –6.22 (23.74)
High 1.87 (0.31) 3.34 (1.36) –16.71 (7.54) –12.26 (11.24) –20.05 (12.67) –13.39 (16.32)

presented, i.e., items with a high familiarity due to recent between the intelligence groups. At pre-test for both areas, the
presentation but not actually hits. Of the non-target trials, three results were non-significant (ERD-frontal t(26) = 1.06, n.s. (p =
quarters were interference trials, i.e., a letter was presented .30) and ERD-parietal t(26) = 1.54, n.s. (p = .14)). In addition,
matched the letter shown two, four, or five trials before. The correlations between intelligence level and cortical activation
participants had to indicate when the current stimulus were non-significant and low (RAPM raw score and ERD frontal
matched the stimulus from n steps earlier in the sequence. at pre-testing: r = –.20, n.s. (p = .30); RAPM raw score and ERD
The responses were required for targets only. The visual targets parietal at pre-testing: r = –.29, n.s. (p = .14)). This result is
had to be indicated by a button press with their left index consistent with those of Neubauer and Fink (2009) and Dunst,
finger, auditory targets by a button press of their right index Benedek, Jauk, and Neubauer (2014), who showed that more
finger and double targets by two button presses with each intelligent students invested an equal amount of cortical
hand. According to the participants' performance, the value of n resources but attained a higher achievement level. Consequently,
could change after each block but always remained the same our data support the neural efficiency hypothesis, such that
for visual and auditory stimuli. The level of n increased by 1 if higher intelligence should manifest either as a lower level of
the participant made fewer than three mistakes per modality cortical activation or as better performance while displaying an
and decreased by 1 if more than five mistakes were made; for 3, equal amount of cortical activation as less intelligent individuals.
4 or 5 mistakes per modality, the level of n stayed the same. Analyses of the post-test time point revealed that the
Each training session consisted of 24 blocks, resulting in a total intelligence groups did not differ regarding their behavioral
daily training time of ca. 30 min. measure or concerning cortical activation (n-back level
t(27) = 0.11, n.s. (p = .44); ERD-frontal t(26) = 0.87, n.s.
3.2. Results and discussion of study 2 (p = .39); t-test ERD-parietal t(26) = 0.95, n.s. (p = .35)). The
correlations of intelligence with n-back level (r = .32, n.s., p =
The two intelligence groups were again formed by a median .10) and intelligence with cortical activation (RAPM raw score
split of the RAPM raw scores at pre-testing (lower intelligence- with ERD frontal at post-testing: r = -.20, n.s. (p = .30) and
group: n = 12, M = 10.00, SD = 1.41; higher intelligence- RAPM raw score with ERD parietal at post-testing: r = –.29, n.s.
group: n = 17, M = 14.88 SD = 1.54; resulting in an effect size (p = .14)) were not significant.
of d = 3.31). This sample consisted of university students who We lastly analyzed the impact of the training using ANOVAs
were performing at a comparably high level as the participants for the intelligence group (i.e., lower vs. higher intelligence) as
of study 1. a between-subject variable and time (pre- vs. post-testing) as a
In a first step, we tested whether more or less intelligent within-subject variable. Concerning the n-back level, the
students differed in the initial achievement on the n-back task. ANOVA showed a significant main effect of time but no group
This was indeed found to be the case. A t-test revealed that the differences or interactions. Thus, all of the participants were
n-back level for the lower intelligence group (n-back level: able to improve their n-back level from pre- to post-testing
M = 1.61, SD = 0.33) was lower than for the higher intelligent irrespective of their intelligence level (see Table 6 and for
group (n-back level: M = 1.87, SD = 0.31), (t(27) = –2.19, descriptive values Table 5). With respect to frontal cortical
p = .037, Cohen's d = 0.88). The impact of intelligence on the activation, none of the values derived from the ANOVA reached
initial n-back level was also reflected in a significant correlation significance. Therefore, that was no progression over time, no
of r = .39, p = .037. differences between the intelligence groups, and no interactions
Next, we performed t-tests to determine if there were any (see Table 6). Analyses of parietal cortical activations again did
initial differences in frontal and parietal cortical activation not show any significant values. Thus, for parietal activation as

Table 6
Training results for study 2: Reporting main effects and interactions for an ANOVA with the between-subject factor of intelligence group (lower vs. higher intelligence)
and the within–subject factor of time (pre– and post-testing) for the n-back level, ERD-frontal, and ERD-parietal.

Measure Main effect time Main effect IQ-group Interaction time ∗ IQ-group

Behavioral measure: n-back Level F(1, 27) = 48.61 F(1,27) = 0.86 F(1,27) = 0.04
p b .001 n.s.(p = .22) n.s.(p = .85)
η2p = .64 η2p = .06 η2p = .001
ERD-frontal F(1, 26) = 3.14 F(1, 26) = 1.28 F(1, 26) = 0.03
n.s. (p = .09) n.s. (p = .27) n.s. (p = .87)
η2p = .11 η2p = .05 η2p = .001
ERD-parietal F(1, 26) = 2.81 F(1, 26) = 2.03 F(1, 26) = 0.02
n.s. (p = .11) n.s. (p = .17) n.s. (p = .90)
η2p = .10 η2p b .07 η2p = .001
206 D. Nussbaumer et al. / Intelligence 50 (2015) 196–208

well, there was no progression over time, no differences between interactions in which the IQ-group is involved. For example,
the intelligence groups, and no interactions (see Table 6). Grabner et al. (2003) found no main effect of IQ-group but a
Before the n-back task training began, there was a substantial significant interaction between IQ-group and task (d = .96).
correlation between performance in this task and intelligence, Also, Doppelmayr, Klimesch, Hödlmoser, Sauseng, and Gruber
indicating that individuals with higher intelligence started on a (2005b) found no significant main effects of IQ-group but only
higher n-back level. Over the course of the training, all of the interactions between intelligence and task (d around 1).
participants improved their n-back level by advancing to a more Grabner et al. (2006) reported a three-way interaction between
difficult task version, with no evident impact of intelligence on hemisphere, area and IQ-group (F(3.12, 130.91) = 3.02, p b .05,
learning gains. Moreover, no differences between the intelli- η2 = .07), which equates d = .55. Taken together, the size of
gence groups were apparent at post-testing. the neural efficiency effect we found for the 4-face task is
Concerning cortical activation no differences were observed relatively large and comparable to other studies. It is worth
between the intelligence groups neither before nor after the mentioning that we found direct evidence for intelligence
training. This result is in line with the reformulation of the differences in terms of a main effect of intelligence on cortical
neural efficiency hypothesis by Neubauer and Fink (2009) and activation rather than interactions.
the specifications by Dunst et al. (2014), who found that with While we found strong neural-efficiency effects for the 4-face
the same (high) subjective task difficulty – in the case of the n- task, neither the 3-back task nor the 4-letter task reached
back task at maximum – no intelligence difference was found in significance, indicating that more and less intelligent participants
cortical activation. While we observed achievement gains in did not differ in their cortical activation. Given the high solution
the n-back level between pre- and post-testing, there was no rates and the low solution times for these two tasks depicted in
change in cortical activation and no differences between the Table 1, it seems that these problems were not challenging
intelligence groups. Better performance without a change in enough for our sample of university students to produce
cortical activation could also be interpreted as a relative gain of differences in cortical activation. Remembering letters and
efficiency. distinguishing them from each other, as it was required in the
3-back task and the 4-letter task is a highly familiar activity for
4. General discussion university students, and they reach high performance levels
without great effort. Therefore, intelligence differences within
We measured cortical activation while solving WM tasks our already highly selected sample do not significantly
to test the scope of the neural efficiency hypothesis. More moderate neural processing. In contrast, remembering faces
specifically, we aimed at replicating the specification of the presented in a context free situation and distinguishing
neural efficiency hypothesis, i.e., that whether frontal cortical between them is a relatively unfamiliar activity which requires
activation is moderated by intelligence depends on the the development of new strategies. More intelligent individ-
difficulty level of the tasks. In addition, we sought to determine uals apparently need less neural resources for mastering this
how cortical activation changes as a result of training. For this novel task.
purpose, we conducted two studies. In study 1, we compared Indirect support for the neural efficiency hypothesis was
participants who underwent a three-week moderate WM found in study 2. Specifically, intelligence affected the
training period (interference group) with an active control performance level of the n-back task but not neural
group. In study 2, we investigated cortical activation in a group activation, presumably because participants were expected
that trained with a high WM load. to invest their entire cognitive resources in improving
As the two training groups in study 1 were fully parallelized performance. The dual task group trained with an adaptive
regarding stimulus presentation, timing, and task demands, WM task that placed high WM load during training by
and differed solely in the amount of WM load, differences including a high number of interference trials and due to the
between these groups can readily be attributed to the amount dual task characteristics. The adaptivity ensured that the
of WM-load during training. In line with our expectations, our task had a high WM load and that each participant trained at
results revealed that the neural efficiency effect appeared only its limit at all times. Therefore, we compared tasks of
in the interference group but not in the control group. It has to different difficulty. Our result is in line with Dunst et al.
be emphasized that the frontal activation difference between (2014), who found that the neural efficiency hypothesis is only
the intelligence groups in the face-recognition task was true when all of the participants work on tasks with the same
of considerable effect size of η2p = .24, which corresponds to difficulty irrespective of their intelligence level. It can be
d = 1.12. To compare this effect with those of similar studies, supposed that neural efficiency reflects that more intelligent
we calculated Cohen's d based on the reported statistical participants face less of a challenge to solve the tasks and that
information. The Tetris-training study of Haier et al. (1992a) less intelligent participants face a greater challenge. When
described correlations between RAPM and frontal activation controlling for subjective task difficulty (i.e., when applying
around .6, which corresponds to d = 1.25. In the training study tasks with a difficulty level according to one's intelligence level),
of Neubauer et al. (2004) non-significant correlations between the participants with lower versus higher intelligence no longer
ERD-frontal and results in a reasoning test were found at pre- show differences in brain activation (see also Larson, Haier,
testing whereas significant correlations of .41 emerged in the LaCasse, & Hazen, 1995). The adaptivity of the training in the
post-test (d = .89). high WM load group in the current study is comparable to the
In addition to these training studies, there is a variety of situation in the study of Dunst et al. (2014). Therefore, our
studies calculating ANOVAs with the between subject factor IQ- finding that there were no intelligence differences in cortical
groups formed through applying a median split (as was done in activation while solving this adaptive task is in agreement with
the current study) showing no main effects of IQ-group but the finding of Dunst et al. (2014).
 D. Nussbaumer et al. / Intelligence 50 (2015) 196–208 207

The brain activation patterns observed in the interference De Smedt, B., Grabner, R.H., & Studer, B. (2009). Oscillatory EEG correlates of
arithmetic strategy use in addition and subtraction. Experimental Brain
group of Study 1 supported the neural efficiency hypothesis Research, 195(4), 635–642.
for frontal but not parietal areas. This result is in concordance Doppelmayr, M., Klimesch, W., Hödlmoser, K., Sauseng, P., & Gruber, W.
with various studies reporting less frontal activation for more (2005a). Intelligence related upper alpha desynchronization in a semantic
memory task. Brain Research Bulletin, 66(2), 171–177.
intelligent participants (Gevins & Smith, 2000; Jaušovec & Doppelmayr, M., Klimesch, W., Sauseng, P., Hödlmoser, K., Stadler, W., &
Jaušovec, 2004 and Rypma et al., 2006) and with the review by Hanslmayr, S. (2005b). Intelligence related differences in EEG-bandpower.
Neubauer and Fink (2009). In the present study we also did not Neuroscience Letters, 381(3), 309–313.
Dunst, B., Benedek, M., Jauk, E., & Neubauer, A.C. (2014). Neural efficiency
find a tendency for a simultaneous higher parietal activation, as
depending on subjective and objective level of task difficulty — An fMRI
did e.g. Rypma et al. (2006). study. Personality and Individual Differences, 60, 36–37.
While most previous studies on the neural efficiency Engle, R.W., Tuholski, S.W., Laughlin, J.E., & Conway, A.R. (1999). Working
memory, short-term memory, and general fluid intelligence: A latent-
hypothesis used either intelligence test items or elementary
variable approach. Journal of Experimental Psychology. General, 128(3), 309.
cognitive tasks, few have shown the effect for WM tasks Gevins, A., & Smith, M.E. (2000). Neurophysiological measures of working
(e.g., Grabner, Fink, Stipacek, Neuper, & Neubauer, 2004). Our memory and individual differences in cognitive ability and cognitive style.
findings also should be viewed in light of our sample of Cerebral Cortex, 10(9), 829–839.
Grabner, R.H., & De Smedt, B. (2011). Neurophysiological evidence for the
university students with above-average intelligence scores. validity of verbal strategy reports in mental arithmetic. Biological
That we still found differences in cognitive activation when Psychology, 87(1), 128–136.
WM tasks of medium difficulty were processed confirms the Grabner, R.H., Fink, A., Stipacek, A., Neuper, C., & Neubauer, A.C. (2004).
Intelligence and working memory systems: Evidence of neural efficiency in
strength of the neural efficiency hypothesis. alpha band ERD. Cognitive Brain Research, 20(2), 212–225.
Contrary to our expectation, we did not observe a decrease in Grabner, R.H., Neubauer, A.C., & Stern, E. (2006). Superior performance and
cortical activation after WM task training in any group, neural efficiency: The impact of intelligence and expertise. Brain Research
Bulletin, 69(4), 422–439.
irrespective of participants' intelligence level. The amount of Grabner, R.H., Stern, E., & Neubauer, A.C. (2003). When intelligence loses its
cortical activation after the three-week-training remained at the impact: Neural efficiency during reasoning in a familiar area. International
same level, and no training-induced changes in neural efficiency Journal of Psychophysiology, 49(2), 89–98.
Gray, J.R., Chabris, C.F., & Braver, T.S. (2003). Neural mechanisms of general fluid
depending on participants' intelligence level were found. intelligence. Nature Neuroscience, 6(3), 316–322.
Learning effects, however, were obvious on the behavioral Haier, R.J., Siegel, B.V., MacLachlan, A., Soderling, E., Lottenberg, S., &
level, as evidenced by increasing solution rates and decreasing Buchsbaum, M.S. (1992a). Regional glucose metabolic changes after
learning a complex visuospatial/motor task: A positron emission tomo-
solution times. The observation of no reduction in cortical
graphic study. Brain Research, 570, 134–143.
activation after the training is in contrast to two other training Haier, R.J., Siegel, B.V., Nuechterlein, K.H., Hazlett, E., Wu, J.C., Paek, J., et al.
studies. Haier et al. (1992a), Haier et al. (1992b) found a (1988). Cortical glucose metabolic rate correlates of abstract reasoning and
reduction in cortical activation after intensive practice of the attention studied with positron emission tomography. Intelligence, 12,
199–217.
computer game Tetris, which fosters spatial pattern matching Haier, R.J., Siegel, B.V., Tang, C., Abel, L., & Buchsbaum, M.S. (1992b). Intelligence
and fine motor skills. Neubauer et al. (2004) even found and changes in regional cerebral glucose metabolic rate following learning.
training cortical effects after a two-hour training session for Intelligence, 16, 415–426.
Jaeggi, S. M., Buschkuehl, M., Jonides, J., & Perrig, W. J. (2008). Improving fluid
elementary cognitive tasks. Given that participants in study 2 intelligence with training on working memory. Proceedings of the National
only worked on a single task during the entire training, a lack Academy of Sciences, 105(19), 6829–6833.
of a decrease in cortical activation can hardly be traced to Jaušovec, N., & Jaušovec, K. (2004). Differences in induced brain activity during
the performance of learning and working-memory tasks related to
insufficient practice. The tasks used in our studies still draw on intelligence. Brain and Cognition, 54(1), 65–74.
WM capacity even after intensive training. Future research Jung, R.E., & Haier, R.J. (2007). The Parieto-Frontal Integration Theory (P-FIT) of
should develop trainable tasks that allow the detection of intelligence: Converging neuroimaging evidence. Behavioral and Brain
Sciences, 30(02), 135–154.
changes in the brain that reflect learning trajectories modulat- Kane, M.J., Hambrick, D.Z., Tuholski, S.W., Wilhelm, O., Payne, T.W., & Engle,
ed by the intelligence of the individuals. R.W. (2004). The generality of working memory capacity: A latent-variable
In conclusion, the present studies corroborate the moderating approach to verbal and visuospatial memory span and reasoning. Journal of
Educational Psychology: General, 133(2), 189–217.
role of task difficulty for the neural efficiency hypothesis. Our
Klimesch, W. (1999). EEG alpha and theta oscillations reflect cognitive
results extend previous findings by demonstrating this mod- and memory performance: A review and analysis. Brain Research Reviews,
erating effect in the context of WM demands and by revealing 29(2-3), 169–195.
that in this task domain, training did not impact on the Klimesch, W., Doppelmayr, M., Pachinger, T., & Ripper, B. (1997). Brain
oscillations and human memory: EEG correlated in the upper alpha and
relationship between intelligence and brain activation. These theta band. Neuroscience Letters, 28(238), 8–12.
insights contribute to a better understanding of the neurophys- Kyllonen, P.C., & Christal, R.E. (1990). Reasoning ability is (little more than)
iological correlates of individual differences in intelligence. working-memory capacity?! Intelligence, 14(4), 389–433.
Larson, G.E., Haier, R.J., LaCasse, L., & Hazen, K. (1995). Evaluation of a “mental
effort” hypothesis for correlations between cortical metabolism and
intelligence. Intelligence, 21(3), 267–278.
Melby-Lervåg, M., & Hulme, C. (2013). Is working memory training effective? A
References metaanalytic review. Developmental Psychology, 49(2), 270–291.
Nee, D.E., Wager, T.D., & Jonides, J. (2007). Interference resolution: insights from
Chein, J.M., & Morrison, A.B. (2010). Expanding the mind's workspace: Training a meta-analysis of neuroimaging tasks. Cognitive, Affective, & Behavioral
and transfer effects with a complex memory span task. Psychonomic Neuroscience, 7(1), 1–17.
Bulletin & Review, 17(2), 193–199. Neubauer, A.C., & Fink, A. (2003). Fluid intelligence and neural efficiency: Effects
Conway, A.R.A., Cowan, N., Bunting, M.F., Therriault, D.J., & Minkoff, S.R.B. (2002). of task complexity and sex. Personality and Individual Differences, 35(4),
A latent variable analysis of working memory capacity, short-term memory 811–827.
capacity, processing speed, and general fluid intelligence. Intelligence, 30(2), Neubauer, A.C., & Fink, A. (2009). Intelligence and neural efficiency: Measures of
163–183. brain activation versus measures of functional connectivity in the brain.
Conway, A.R., Kane, M.J., & Engle, R.W. (2003). Working memory capacity and Intelligence, 37(2), 223–229.
its relation to general intelligence. Trends in Cognitive Sciences, 7(12), Neubauer, A.C., Fink, A., & Grabner, R.H. (2006). Sensitivity of alpha band ERD to
547–552. individual differences in cognition. In C. Neuper, & W. Klimesch (Eds.),
208 D. Nussbaumer et al. / Intelligence 50 (2015) 196–208

Progress in brain research: Event-related dynamics of brain oscillations. Vol. Pfurtscheller, G., & Aranibar, A. (1977). Event-related cortical desynchronization
159. (pp. 167–178). Elsevier. detected by power measurements of scalp EEG. Electroencephalography and
Neubauer, A., Freudenthaler, H.H., & Pfurtscheller, G. (1995). Intelligence and Clinical Neurophysiology, 42, 817–826.
spatiotemporal patterns of event-related desynchronization (ERD). Raven, J.C. (1990). Advanced progressive matrices sets 1 and 2. Oxford: Oxford
Intelligence, 20(3), 249–266. Psychologists Press.
Neubauer, A.C., Grabner, R.H., Freudenthaler, H.H., Beckmann, J.F., & Guthke, J. Rypma, B., Berger, J.S., Prabhakaran, V., Martin Bly, B., Kimberg, D.Y., Biswal, B.B.,
(2004). Intelligence and individual differences in becoming neurally et al. (2006). Neural correlates of cognitive efficiency. NeuroImage, 33(3),
efficient. Acta Psychologica, 116(1), 55–74. 969–979.
Neubauer, A.C., Sange, G., & Pfurtscheller, G. (1999). Psychometric intelligence and Schloegl, A., Keinrath, C., Zimmermann, D., Scherer, R., Leeb, R., & Pfurtscheller,
event-related desynchronization during performance of a letter matching G. (2007). A fully automated correction method of EOG artifacts in EEG
task. Handbook of Electroencephalography and Clinical Neurophysiology, 6, recordings. Clinical Neurophysiology, 118(1), 98–104.
219–231. Shipstead, Z., Redick, T.S., & Engle, R.W. (2010). Does working memory training
Paas, F. (1992). Training strategies for attaining transfer of problem-solving skill generalize? Psychologica Belgica, 50(3–4), 245–276.
in statistics: A cognitive load approach. Journal of Educational Psychology, Shipstead, Z., Redick, T.S., & Engle, R.W. (2012). Is working memory training
84, 429–434. effective? Psychological Bulletin, 138(4), 628–654.
Paas, F., Tuovinen, J.E., Tabbers, H., & Van Gerven, P.W. (2003). Cognitive load Smith, E.E., & Jonides, J. (1997). Working memory: A view from neuroimaging.
measurement as a means to advance cognitive load theory. Educational Cognitive Psychology, 33(1), 5–42.
Psychologist, 38(1), 63–71.