Processing Speed, Working Memory, and Executive Functions:

Independent or inter-related predictors of general intelligence

Gidon T. Frischkorn*, Anna-Lena Schubert & Dirk Hagemann

Heidelberg University, Institute of Psychology

Author Note

We want to thank our research assistants – Ben Riemenschneider, Christoph Löffler,

Isabel Gebhardt and Larissa Kunoff – who supported us with the data collection and preparation

for this project. This research did not receive any specific grant from funding agencies in the

public, commercial, or not-for-profit sectors.

* Correspondence concerning this article should be addressed to Gidon Frischkorn,

Psychologisches Institut, Ruprecht-Karls-Universität Heidelberg, Hauptstrasse 47-51, D-

69117 Heidelberg, Germany, email: gidon.frischkorn@psychologie.uni-heidelberg.de,

telephone: +49-6221-54-7746
PS, WM, AND EF AS PREDICTORS OF G 2

Abstract

Both working memory capacity (WMC) and processing speed (PS) have been discussed

as important covariates of individual differences in intelligence. Recent results indicated that

especially latencies of ERP components associated with higher-order processing (P2, N2, and

P3) may share up to 80% of variance with individual differences in intelligence. WMC has a

similar predictive power and thus these two processes cannot explain individual differences in

intelligence independently. The current study explores in how far individual differences in

executive functions (EFs) may bridge the gap between WMC and PS as predictors of

intelligence. We recruited 101 participants who completed three EF tasks – one for each of the

three executive functions shifting, updating, and inhibition – while an EEG was recorded.

Additionally, we assessed participants’ intelligence, WMC, and PS. Results showed that only

variance of behavioral RTs consistent across manipulations in the EF tasks was related to WMC,

PS, and intelligence. While P3 latencies were not associated with intelligence, they showed

significant correlation with WMC and PS, and N1 latencies showed no correlation with any of

the three covariates. The variance specific to the manipulations in EF tasks was small for both

behavioral RTs and ERP latencies and showed no consistent correlations with each other or with

any of the three covariates. These results suggest that EF tasks capture mostly manipulation-

unspecific cognitive processes. Hence, individual differences in the impairment due to additional

executive processing demands cannot explain why WMC and PS are related predictors of

individual differences in intelligence.

Keywords: Intelligence; Processing Speed; Working Memory; Executive Functions;

EEG;
PS, WM, AND EF AS PREDICTORS OF G 3

Processing Speed, Working Memory, and Executive Functions: Independent or inter-

related predictors of general intelligence

The currently most discussed cognitive processes underling individual differences in general

intelligence (g) are speed of information processing (Jensen, 2006; Schubert, Hagemann, &

Frischkorn, 2017; Sheppard & Vernon, 2008), working memory capacity (Ackerman, Beier, &

Boyle, 2005; Conway, Cowan, Bunting, Therriault, & Minkoff, 2002; Kane, Hambrick, &

Conway, 2005), and executive functions (Friedman et al., 2006; Jewsbury, Bowden, & Strauss,

2016; Miyake, Friedman, Rettinger, Shah, & Hegarty, 2001). These three processing domains

were often discussed separately or as independent predictors of individual differences in

intelligence, focusing on the question which of the processes shows the largest relationship to

individual differences in intelligence. Especially regarding processing speed and working

memory capacity as predictors of g, some researchers argued for them being independent

predictors (Colom, Abad, Quiroga, Shih, & Flores-Mendoza, 2008), while others showed

considerable correlations between these two processes (Ackerman, Beier, & Boyle, 2002;

Kyllonen & Christal, 1990; Schmiedek, Oberauer, Wilhelm, Süß, & Wittmann, 2007; Schmitz &

Wilhelm, 2016).

Recent results indicated that the neural speed of information processing of higher order

cognitive processes explains up to 80% of variance in intelligence (Schubert, Hagemann, &

Frischkorn, 2017), matching the amount of variance in intelligence often explained by working

memory capacity (Kyllonen & Christal, 1990; Oberauer, Schulze, Wilhelm, & Süß, 2005). This

finding strongly suggests that these two processes cannot be independent predictors of individual

differences in intelligence. Hence, it follows that speed of information processing and the
PS, WM, AND EF AS PREDICTORS OF G 4

capacity of working memory have to be inter-related and might be constrained by similar

features of the neuro-cognitive system (Dang, Braeken, Colom, Ferrer, & Liu, 2015). The aim of

the present study is to bridge the gap between working memory capacity (WMC) and processing

speed (PS) as predictors of g. On the basis of current theories of working memory (Barrouillet,

Portrat, & Camos, 2011; Cowan, 2017; Oberauer, 2002; Oberauer & Kliegl, 2006) that

emphasize the role of attentional processes the present study explores in how far executive

functions (Miyake et al., 2000) may explain the inter-relation between WMC and PS as

predictors of g.

The relationship of processing speed with intelligence

Across a variety of different measures there is a consistent negative relationship between

speed of information processing and general intelligence (Jensen, 2006). A review of a broad

variety of tasks reported an average correlation of r = -.24 of single task measures of PS and g

(Sheppard & Vernon, 2008). These correlations tend to increase (r = -.40 to -.50) when reaction

times from different tasks are aggregated (Kranzler & Jensen, 1991; Schmitz & Wilhelm, 2016;

Schubert, Hagemann, & Frischkorn, 2017), indicating that foremost variance shared between

different measures for PS is related to g. Moreover, the correlation between PS and g increases

even further, if measures specifically representing the speed of the decision process, such as the

drift rate from the drift-diffusion model, are used to represent processing speed (Schmiedek et

al., 2007).

Separating encoding processes and motor execution from the actual decision process via

cognitive modeling is a promising approach to further investigate the relationship between speed

of information processing and intelligence (Frischkorn & Schubert, 2018). Specifically, the drift-

diffusion model (DDM; Ratcliff, 1978) has often been successfully used to separate the speed of
PS, WM, AND EF AS PREDICTORS OF G 5

information accumulation, represented by the drift rate, from other processes included in

reaction times, such as encoding, motor execution or response caution. Results indicated that

drift rates show trait-like properties (Schubert, Frischkorn, Hagemann, & Voss, 2016) and show

consistent relationships with intelligence ranging from r = .50 to .90 (Ratcliff, Thapar, &

McKoon, 2010; Schmiedek et al., 2007; Schubert, Hagemann, Voss, Schankin, & Bergmann,

2015). This indicates that it is precisely the speed of information accumulation that is related to

general intelligence and not speed of motor execution or encoding.

However, PS can not only be measured via behavioral indicators such as reaction times,

but also with neurophysiological indicators such as latencies of event-related potential

components (Verleger, 1997). The event-related potential (ERP) decomposes the stream of

neural information processing from stimulus onset until the response into distinct components

that can be linked to specific cognitive functions. Specifically, individual differences in the

latency of an ERP component may reflect individual differences in the neural speed of

information processing, meaning that a higher speed of information processing results in shorter

latencies of an ERP components. Moreover, the onset of an ERP component may also serve as an

indicator of its functionality. While ERP components occurring early after stimulus onset are

mostly related to stimulus encoding, ERP components with a later onset are foremost connected

to higher-order processing.

Despite rather weak and inconsistent results on the relationship between latencies of ERP

components and intelligence (Schulter & Neubauer, 2005), a number of empirical results showed

consistent negative relations between the latency of the P3 component or the mismatch

negativity (MMN) and intelligence (Bazana & Stelmack, 2002; McGarry-Roberts, Stelmack, &

Campbell, 1992). In detail, more intelligent people displayed shorter latencies of the P3 (Bazana
PS, WM, AND EF AS PREDICTORS OF G 6

& Stelmack, 2002; McGarry-Roberts et al., 1992; Troche, Houlihan, Stelmack, & Rammsayer,

2009), which is an ERP component that is associated with stimulus evaluation and

categorization, context updating, and context closure. Furthermore, this relationship of latencies

of ERP components with intelligence is mediated via behavioral RTs (Schubert et al., 2015)

suggesting that neural processing speed may functionally underlie faster information processing

on a behavioral level.

The mostly inconsistent relationship of ERP latencies with intelligence can be attributed

to different problems, such as (1) small sample sizes (N < 50), (2) questionable selection of

electrode sites for the measurement of ERP latencies, and (3) quantification of relationships with

single task measures that confound task-related fluctuations with the trait-like neural processing

speed of a person. A recent study addressing these issues by measuring ERP latencies for three

different tasks at two measurement occasions could demonstrate that the shared variance of later

ERP components (P2, N2, & P3) explained up to 65% of variance in general intelligence

(Schubert, Hagemann, & Frischkorn, 2017). Moreover, the results of this study indicated that the

variance of single task latencies of ERP components included a large proportion of task- and

condition-specific variance, which may be irrelevant to the relationship of ERP latencies with

general intelligence, as well as substantial unsystematic error variance. This finding may explain

the inconsistent results from earlier studies using only single task measures. Taken together, all

these results from behavioral and neuro-psychological studies indicate that there is a strong

relationship between speed of higher order information processing and intelligence.

The relationship of working memory capacity with intelligence

In addition to speed of information processing, also working memory capacity has been

closely linked to individual differences in intelligence (Ackerman et al., 2005; Kane et al., 2005;
PS, WM, AND EF AS PREDICTORS OF G 7

Oberauer et al., 2005). Comparable to cognitive ability tasks, measures of working memory

capacity (WMC) are highly correlated with each other (Engle, Tuholski, Laughlin, & Conway,

1999; Kane et al., 2004; Oberauer, Süß, Wilhelm, & Wittman, 2003; Unsworth, Fukuda, Awh, &

Vogel, 2014) and resemble a hierarchical structure, with a broad single factor at the highest level,

and more domain and process specific factors on the lower level (Bayliss, Jarrold, Gunn, &

Baddeley, 2003; Kane et al., 2004; Oberauer et al., 2003; Shah & Miyake, 1996). Correlations

between the broad factor of WMC and g are very high, ranging from r = .70 to .90 (Conway et

al., 2002; Kane et al., 2005; Kyllonen & Christal, 1990; Oberauer et al., 2005). These high

correlations have led to a vivid discussion in how far WMC and intelligence may be isomorphic

(Ackerman et al., 2005; Kane et al., 2005; Oberauer et al., 2005), ultimately resolved by

concluding WMC explains a large proportion of individual differences in intelligence.

In addition to these correlational studies there are results from an experimental study

suggesting that overloading working memory while performing a test of fluid intelligence

affected performance in the intelligence measure (Rao & Baddeley, 2013). In detail, participants

were asked to remember a set of three digits and count backwards while working on a matrix

reasoning task. Results showed that especially the time needed to solve an item increased

compared to a silent control and an articulatory suppression condition. Altogether these results

suggest that working memory is not only strongly related to individual differences in intelligence

but may actually causally underlie variations in g.

While this strong relationship between working memory and intelligence is undisputed,

researchers do not agree which process within working memory is central to the relationship of

working memory and intelligence. Some researchers argue that the relationship is best explained

by similar demands on short-term memory storage (Colom et al., 2008; Colom, Flores-Mendoza,
PS, WM, AND EF AS PREDICTORS OF G 8

Quiroga, & Privado, 2005; Shahabi, Abad, & Colom, 2014), and others argue that processes

specific to complex span tasks such as attention regulation are the reason for the strong

association between working memory and intelligence (Conway et al., 2002; Unsworth et al.,

2014). As there is robust evidence that the capacity of working memory is related to attentional

processes (Chuderski, Taraday, Nęcka, & Smoleń, 2012; Kane & Engle, 2003; McVay & Kane,

2009, 2012; Meier & Kane, 2013) and current theories of working memory assume that attention

plays a major role in maintenance of memory items regardless of concurrent processing

(Oberauer, Farrell, Jarrold, & Lewandowsky, 2016; Souza & Vergauwe, 2018), it is plausible

that both the capacity of short-term memory storage and additional demands in complex span

tasks are strongly related to the same attentional processes within working memory (Barrouillet

et al., 2011; Wilhelm, Hildebrandt, & Oberauer, 2013).

Executive Functions: Bridging the gap between processing speed and working memory?

Candidates for the attentional processes underlying both maintenance of items in short-

term memory as well as additional demands of complex span tasks are executive functions.

Executive functions (EFs) are defined as attentional control mechanisms (Karr et al., 2018;

Miyake et al., 2000) that are used to (a) focus attention on relevant information while ignoring

irrelevant information (i.e. inhibition), (b) encode new information to memory while removing

outdated and no longer relevant information (i.e. updating), or (c) switch between different tasks

(i.e. shifting). While it is still under debate in how far these different EFs have to be separated or

share common variance (Friedman & Miyake, 2017; Miyake & Friedman, 2012), the majority of

results suggests that there is considerable overlap between the three EFs (Karr et al., 2018).

Moreover, EFs have recently been subsumed within the hierarchical structure of intelligence

(Jewsbury et al., 2016). In detail, updating was integrated within a general memory factor gm,
PS, WM, AND EF AS PREDICTORS OF G 9

while shifting and inhibition were integrated in the general speed factor gs, pointing towards

relations of EFs with both memory and processing speed.

With respect to speed of information processing, the results from Schubert et al. (2017)

support the proposal that executive functions may underlie the relationship of processing speed

with intelligence as well. In detail, the latency of the P3 component showed the strongest

association with general intelligence. The P3 component has often been interpreted as an

indicator of the efficiency of context-updating (Donchin, 1981; Polich, 2007), and thus shorter

latencies of the P3 may reflect a faster inhibition of nonessential processes that in turn ease the

transmission of information from attention and working memory regions located frontally in the

brain to parietal memory storage processes (Polich, 2007). There is additional support for this

hypothesis from behavioral studies showing strongest relations between inhibition and updating

with intelligence (Wongupparaj, Kumari, & Morris, 2015).

Beyond that, cognitive as well as neural theories of intelligence are in line with this

theoretical perspective. The process-overlap theory (POT; Kovacs & Conway, 2016) assumes

that attentional control mechanisms are among the domain-general processes that act as a

bottleneck constraining performance in a broad range of cognitive tasks. Moreover, the parieto-

frontal integration theory (P-FIT; Jung & Haier, 2007) proposes that individual differences in the

efficiency of information transmission from frontal association cortices and parietal brain regions

may explain individual differences in g. P-FIT has been widely supported by results from

structural and functional neuroimaging studies (Burgess, Gray, Conway, & Braver, 2011; Colom,

Jung, & Haier, 2007; Colom & Thompson, 2013; Gläscher et al., 2010). Altogether, attention

regulation mechanisms such as EFs may provide a theoretical account to bridge the gap between

processing speed and working memory as predictors of individual differences in g.

PS, WM, AND EF AS PREDICTORS OF G 10

The present study

We conducted the present study to investigate in how far individual differences in

executive functions (EFs) may underlie the relationship of processing speed (PS) and working

memory capacity (WMC) with intelligence (g). To that end, we administrated three different EF

tasks each tapping one of the executive functions (i.e. shifting, updating, and inhibition). To

further differentiate between behavioral and neurophysiological indicators of executive functions

we recorded the EEG while participants worked on these three EF tasks. Additionally, we

assessed participants’ general intelligence, their working memory capacity, and speed of

information processing to investigate in how far individual differences in EF tasks explain the

relationship between these three constructs.

More specifically, we aimed to address two major points: First, we wanted to investigate

in how far performance in the different experimental conditions of EF tasks measure

performance specific to the manipulations that are related to the respective EFs, or performance

that is unspecific with respect to the experimental manipulations. And second, we were

interested how these two components of performance in EF tasks are related to WMC, PS, and g.

All in all, joining the individual differences constructs of intelligence, WMC, and PS with

executive functions may provide insights in how far individual differences in EFs may explain

the relationship between WMC and PS as predictors of intelligence.

PS, WM, AND EF AS PREDICTORS OF G 11

Methods

Sample

We recruited a community sample of 107 participants via newspaper ads and flyers. The

101 participants1 who attended both sessions of the experiment were on average 39.1 years

(SDage = 14.5, Minage = 18, Maxage = 61), and 52.5 % were female. Participants had a

heterogeneous educational background (42.6% university degree, 42.6% college degree, 10.9 %

high school degree, 3.9% did not report educational background) and were rewarded with 50€

for their participation. While the overall sample size is comparably small for structural equation

modeling, we secured that it had sufficient power (1- > .80) to assess model fit with the

RMSEA (H0RMSEA =.05, H1RMSEA = .10, dfModel = 50, α = .05, NMin = 97).

General Procedure

The study consisted of two sessions that were approximately four months apart. In the

first occasion, participants completed three executive functioning tasks – a Switching task, an N-

Back task, and the Attention Network Test – while an EEG was recorded. For this occasion,

participants were seated in a dimly lit, sound-attenuated EEG cabin and tested individually. In

the second session, participants’ intelligence, working memory capacity, and processing speed

were measured with paper-pencil tests and computerized tasks. In addition, participants

completed three knowledge tests and two personality questionnaires not reported here. This

session was conducted in groups of up to four participants. Both sessions took approximately 3

hours, and the sequence of tasks within the two occasions was the same for all participants.

1
Only data from the participants who attended both sessions was analyzed and reported in the manuscript.
PS, WM, AND EF AS PREDICTORS OF G 12

Measures

Executive Functions. The three executive functions – shifting, updating, and inhibition –

were each assessed with one task. To additionally asses in how far all executive functions rely on

the same attention process conceptualized as executive attention by Posner and Petersen (1990),

a flanker manipulation was implemented in each of the three tasks (Eriksen & Eriksen, 1974).

Switching Task. The Switching task was adapted from Sauseng et al. (2006). An

illustration of the trial procedure can be found in Figure 1A (p. 50). In this task, participants saw

a digit from 1 to 9 (except the number 5) colored red or green that was presented centrally on a

black screen. Participants either had to decide whether the digit was smaller or larger than five or

whether it was odd or even, depending on the color of the presented number. Prior to the onset of

the target stimulus, a light grey fixation cross was presented centrally on the screen for 400-

600ms. Between onset of the target stimulus and the fixation cross, an inter-stimulus interval

(ISI) consisting of a blank screen was shown for 400-600ms that was presented until participants

responded and stayed on the screen for another 500ms to avoid offset potentials in the EEG due

to perceptual changes on the screen. Participants responded via keypresses on the keyboard by

pressing a left key “d” if the digit was smaller than five or odd, and a right key “l” if the number

was larger than five or even. Before the next trial started, there was an inter-trial interval (ITI) of

1000-1500ms.

The switching task consisted of four different blocks. In the first two blocks, participants

had to decide whether the number was less or more than 5 in one block, or odd or even in the

other block, irrespective of the stimulus color. These control blocks consisting of 48 trials (8

digits x 2 colors x 3 repetitions) did not require any task switching and were used to quantify

global switch costs in comparison to the two switching blocks. In the third block, participants
PS, WM, AND EF AS PREDICTORS OF G 13

were instructed to decide whether the digit was smaller or bigger than 5 for red stimuli and odd

or even for green stimuli. This shifting block consisted of 96 trials (8 digits x 2 tasks x 2 shifting

x 3 repetitions) of which the to-be-conducted task switched in half of the trials. In the last block,

additional flanker stimuli that were congruent, neutral, or incongruent to the target stimulus were

added to the task and participants were instructed to ignore the flankers while completing the

same task as in the shifting block. This shifting flanker block consisted of 288 trials (8 digits x 2

tasks x 2 shifting x 3 flanker x 3 repetitions). The color of flanker and target stimuli was always

the same, therefore congruency of the flankers was only manipulated on the content level (i.e. the

numerical information), but not on the task cue level. Incongruent flankers were always a

different digit than the target stimulus, but could indicate the same response as the target

stimulus (e.g., red 8 as flankers and a red 6 as target are both larger than 5).

All participants completed 16 practice trials per block. The experimental trials within all

blocks were pseudo-randomly sorted following some constraints: In all blocks, digits were not

allowed to repeat more than three times in a row. Likewise, stimulus color, and thus tasks in the

switching blocks, and responses were not allowed to repeat more than three times in a row.

N-Back Task. Participants completed a 2-Back task that was adapted from Scharinger et

al. (2015). The trial procedure of the N-Back task can be seen in Figure 1B (p. 50). Participants

saw a series of light grey letters (H, C, F, or S) shown centrally on the screen one after the other.

For each letter, participants decided whether or not it was identical to the letter presented two

steps before. Between subsequent letters, there was an ISI consisting of a blank screen that was

shown for 1000-1500ms. The letters were always shown for 2500ms irrespective of the time

participant needed to respond. This way we ensured that all participants had the same time to

encode the new letter and decide whether it matched the letter two steps back. Additionally, by
PS, WM, AND EF AS PREDICTORS OF G 14

not changing the perceptual input after participants’ responses, we avoided offset potentials in

the EEG. Participants responded via keypress, pressing a left key “d” if the current letter did not

match the letter two steps back, and pressing a right key “l” if the current letter matched the letter

two steps back.

The N-Back task consisted of two blocks. In the 2-back block, participants completed 96

trials (4 letters x 2 match x 12 repetitions) of the 2-Back task, preceded by two introductory trials

requiring no response as there were no letters two steps prior to presentation. In the 2-back

flanker block, participants completed 384 trials (4 letters x 2 match x 4 flanker x 12 repetitions)

of the 2-Back task with additional flanker stimuli. Unlike in the Shifting Task, there were four

levels of the flanker manipulation within this block. There either were no flanker stimuli – like in

the 2-back block – or flanker stimuli that were either congruent, neutral or incongruent.

Moreover, the flanker block was separated into three sub-blocks consisting of 128 trials, in order

to give participants short breaks. Like in the 2-back block, the experimental trials of all sub-

blocks were preceded by two introductory trials requiring no response.

Participants completed 16 practice trials that were repeated until participants’ average

accuracy in these 16 practice trials was above chance. Experimental trials were pseudo-randomly

sorted with the constraint that responses and thus the match conditions were allowed to repeat a

maximum of two times. Additionally, flanker congruency was not allowed to repeat more than

two times in the flanker block.

Attention Network Test (ANT). In the Attention Network Test (Fan et al.), participants

had to decide whether an arrow pointed left or right (see Figure 1C for the trial procedure). The

arrow (i.e., the target stimulus) could appear above or below a fixation cross that was located

centrally on the screen. Furthermore, the centrally presented arrow was flanked by two more
PS, WM, AND EF AS PREDICTORS OF G 15

arrows to the left and right. The flanking arrows were either pointing in the same direction

(congruent), in the other direction (incongruent), or were without arrow heads indicating any

direction (neutral).

Each trial started with a light grey fixation cross presented centrally on the black screen

for 400-1600ms, followed by a short cue stimulus presented for 100ms. There were four different

cue options: (1) There was no cue and the fixation cross remained on the screen, (2) there was a

central cue at the position of the fixation cross, (3) there was a double cue above and below of

the fixation cross, or (4) there was a spatial cue located either above or below the fixation cross

validly cueing the position of the following target stimulus. Between the cue stimulus and the

target stimulus there was an ISI of 400ms with the fixation cross being presented centrally on the

screen. Then the target stimulus, i.e. the central arrow, and flanker arrows appeared above or

below the fixation cross on the screen and participant had to decide whether the central arrow

pointed right or left. Participants responded via key press, pressing the left key “d” if the arrow

pointed left, and the right key “l” if the arrow pointed right. The target stimulus and flanker

stimuli remained on-screen for 1700ms, irrespective of the speed of the response. Before the next

trial started, there was an ITI again consisting of the light grey fixation cross presented centrally

on the screen that lasted 1700ms.

The ANT consisted of three blocks of 96 trials each (2 direction x 2 location x 4 cue x 3

flanker x 2 repetitions) that were pseudo-randomly sorted. Specifically, all of the four

experimental factors were allowed to repeat a maximum of three times in subsequent trials. 24

practice trials were conducted prior to the first experimental block. In between blocks,

participants took short breaks and read a short reminder of the task instructions.
PS, WM, AND EF AS PREDICTORS OF G 16

Intelligence (Gf). Intelligence was measured with the short-version of the Berlin

Intelligence Structure test (BIS; Jäger, Süß, & Beauducel, 1997). The BIS is based on the

bimodal Berlin Intelligence Structure model (Jäger, 1982). According to this model, the 15 tasks

of the BIS short version are grouped into four operation-related (processing capacity, memory,

processing speed, and creativity) and three content-related (verbal, numerical, and figural)

components of fluid intelligence. Each task combines one operation-related component with one

content-related component of intelligence. The standard scores of the five tasks with verbal,

numerical, and figural content were aggregated across operations and used as separate indicators

of fluid intelligence.

Working Memory Capacity (WMC). Working memory capacity was measured with

four tasks from the working memory test battery by Lewandowsky et al. (2010). Specifically, we

used the memory updating task, two complex span tasks, and a spatial short-term memory task.

Following the scoring script provided by Lewandowsky et al. (2010), performance was measured

by the mean proportion of correctly remembered items for each task separately.

Processing Speed (PS). We measured participants’ processing speed with two

elementary cognitive tasks (ECTs), the Posner letter matching task and the Sternberg memory

scanning task. These tasks are commonly used as an indicator of basic information processing

speed.

Posner Letter Matching Task. In the Posner Letter Matching task (Posner & Mitchell,

1967), participants decided whether two letters were physically or semantically identical. The

participants completed two different blocks, first a physical identical and second a name identity

block. Each of the two blocks consisted of 40 trials preceded by 10 practice trials, in half of

which the two letters matched physically or semantically corresponding to the block instructions.
PS, WM, AND EF AS PREDICTORS OF G 17

The two letters comprising the target stimulus were selected from a pool of five letters (a, b, f, h,

q) that could be capitalized or not. Each trial started with a fixation cross presented centrally on

the screen for 1000-1500ms. Immediately after that, the letter pair was shown on the screen.

Participants then responded via key press, pressing either a right or left key with their index

fingers, indicating whether the two letters were physically or semantically identical. Response

mapping of the keys was counterbalanced across participants. After the response, the trial ended

and the next trial started after an ITI of 1000-1500ms. We used the mean logarithmized RT of

correct responses in the two blocks as a measure of processing speed.

Sternberg Memory Scanning Task. In the Sternberg Memory Scanning task (Sternberg,

1969), participants saw a memory set of digits from zero to nine and had to decide whether a

subsequently presented probe digit was contained in the memory set or not. Participants

completed two blocks, first one block with memory set size three and second one block with set

size five, each consisting of 40 trials, preceded by 10 practice trials. In each block, the probe

digit was contained in the memory set in half of the trials. Each trial started with a fixation cross

presented centrally on the screen for 1000-1500ms. Then the digits comprising the memory set

were presented sequentially on the screen for 1000ms with an ISI of 400-600ms. After the last

digit, a question mark was presented for 1800-2200ms, followed by the probe digit. Participants

then responded via key press, pressing either a left or right key with their index finger. The

response mapping of keys was counterbalanced across participants. After the response, the trial

ended and the next trial started after an ISI of 1000-1500ms. Again, we used the mean

logarithmized RT of correct responses in the two experimental blocks as an indicator of

processing speed.
PS, WM, AND EF AS PREDICTORS OF G 18

EEG Recording

In the first session, we recorded the participants’ EEG during the three EF tasks

(Switching Task, N-Back task, and ANT) with 32 equidistant Ag–AgCl electrodes. In addition,

we used the aFz as ground electrode, and Cz as a common online reference. The signal of all 32

channels was offline re-referenced to an average reference. We kept all electrode impedances

below 5 k, and recorded the EEG signal continuously with a sampling rate of 1000 Hz (band-

pass 0.1–100 Hz). The data was filtered offline with a low-pass filter of 12 Hz.

Data Analyses

Behavioral data. To ensure that intra-individual outliers in reaction times measured in

EF tasks and ECTs did not distort our results, we discarded trials with RTs shorter than 150ms or

longer than 3000ms. Then, we discarded any incorrect trials and trials in which the logarithmized

reaction times of correct responses deviated more than 3 SD from the mean logarithmized

reaction time of each participant within the different conditions in each task. Finally, we

calculated the mean logarithmized reaction time as the dependent variable. As accuracies were

very high (M > .90) and showed little to no variation in all EF tasks and ECTs (see Table 1), we

refrained from analyzing accuracy measures.

For all measures, we conducted additional uni- and multi-variate outlier analyses on the

between-person level. First, univariate outliers deviating more than 3 SDs from the mean were

deleted, resulting in 0.0 % to a maximum of 2.3 % of subjects being excluded across the different

measures. Second, multi-variate outliers on the different measures within each cognitive process

were identified. Only in the ANT did the Mahalanobis distance for one participant exceed the

critical value of 2(12) = 39.9. This subject was excluded case wise for the ANT. Finally,

multivariate outliers across the measures of different cognitive processes were again identified
PS, WM, AND EF AS PREDICTORS OF G 19

via the Mahalanobis distance. As no subject exceeded the critical value of 2(23) = 49.7, no data

was discarded in this final step.

Electrophysiological data. The event-related potentials (ERPs) were calculated

separately for each EF tasks and conditions with the ERPLAB toolbox (Lopez-Calderon & Luck,

2014). All ERPs were time-locked to the stimulus onset with a baseline of 200ms before stimulus

onset for the Switching and the N-Back task and 700ms to 500ms for the ANT. The baseline for

the ANT was earlier due to the cue stimulus 500ms before stimulus onset that elicited

considerable activity in the EEG. After stimulus onset epochs continued for 1000ms, resulting in

1200ms epochs for the Switching and the N-Back task and 1700ms epochs for the ANT. First,

channels and epochs with gross artifacts were rejected based on the standard settings

implemented in EEGLAB. Second, ocular artifacts and generic discontinuities were corrected via

ICA and artifact ICs were identified using the ADJUST algorithm (Delorme & Makeig, 2004;

Mognon, Jovicich, Bruzzone, & Buiatti, 2011).

As we were interested in the neural speed of information processing we determined

latencies of EPR components instead of evaluating amplitudes that are associated with

processing capacity. Furthermore, to differentiate between the neural speed of earlier versus later

processes in the neural stream of information processing we analyzed the N1 and P3 latency (see

Figure 2 for grand average ERPs). The latency of the N1 was determined at the frontal electrode

site over midline and the latency of the P3 was determined at the parietal electrode over midline

for all EF tasks. Peak latencies were determined separately for each EF task and the conditions

within each task. For participants that did not show a N1 or P3 component in the average ERP,

peak latencies were coded as missing in the respective condition of the respective EF task.

Finally, univariate-outliers within each condition of the three EF tasks exceeding ±3SDs from the
PS, WM, AND EF AS PREDICTORS OF G 20

mean latency were discarded. We did not analyze amplitudes of ERP components, as recent

research suggest that the latency of ERP components is strongly associated with intelligence

(Schubert, Hagemann, & Frischkorn, 2017)

Statistical analyses. As an initial manipulation check, we analyzed whether the

experimental manipulations in the EF tasks showed the usual effects on behavioral response

times. This ensured that the EFs supposed to be required by specific experimental manipulations

were actually demanded within the respective task. In addition, we reported the corresponding

results for latency measures of ERP components. For all ANOVAs, we corrected violations of

sphericity by adjusting the degrees of freedom with the Greenhous-Geisser correction. For post-

hoc comparisons, p-values were corrected with the Tukey method.

Following these experimental analyses, we ran structural equations models for behavioral

and EEG measures of the EF tasks separately. First, we established separate measurement

models for the EF tasks. All of these models were set up as bi-factor models with all indicators

across the different blocks and experimental manipulations loading on a general behavioral or

neural speed factor, and factors specific to the experimental manipulations or blocks within each

EF task. This approach allowed us to separate task-general and manipulation-specific variance in

the EF tasks, with the manipulation-specific factors capturing individual differences in executive

functions associated with specific experimental manipulations. These bi-factor models will

answer the first of the two major points we wanted to investigate within the present study,

namely, in how far performance within one condition of an EF task represents general

performance, or performance specific to the manipulation that is linked to the respective EF.

The best fitting bi-factor models for the EF tasks were then merged and covariates were

entered into the model. In a first step, we analyzed the three covariates – general intelligence,
PS, WM, AND EF AS PREDICTORS OF G 21

processing speed, and working memory capacity – separately; in a second step, we included all

covariates simultaneously to additionally assess the inter-relations between covariates and

answer the second of the two major points we wanted to investigate in the present study: How is

general and manipulation-specific variance from EF task related to WMC, PS, and intelligence,

and do these relationships provide evidence that individual difference in EF might represent the

missing link between WMC and PS as a predictor of intelligence?

We assessed model fit of all structural equation models using the comparative fit index

(CFI; Bentler, 1990) and the root mean square error of approximation (RMSEA; Browne &

Cudeck, 1992). We considered model fit as acceptable with CFI > .90 and RMSEA < .10

(Bentler, 1990; Browne & Cudeck, 1992; Hu & Bentler, 1999). When model evaluation diverged

between the two fit criteria, we evaluated model fit with the more favorable fit index, because

previous research has shown that goodness-of-fit statistics tend to underestimate absolute model

fit in small samples (Kenny, Kaniskan, & McCoach, 2015; Schubert, Hagemann, Voss, &

Bergmann, 2017). For comparisons of two models, we required more complex models to show a

lower AIC than more parsimonious models with an AIC difference > 10 to retain the more

complex model (Burnham & Anderson, 2002). Finally, we assessed statistical significance of

model parameters with the two-sided critical ratio test. If parameters did not differ significantly

from zero, we fixed them to zero and estimated the SEM again. Thus, only parameters

significantly different from zero are reported in the results section.

PS, WM, AND EF AS PREDICTORS OF G 22

Results

Manipulation Check: EF tasks

To ensure that experimental manipulations within the EF tasks demanded the respective

attentional control mechanisms, we ran within-subject ANOVAs for the three EF tasks. The

mean reaction time and proportion of correct responses across the different experimental

conditions within the three tasks are displayed in Table 1, descriptive statistics for the latencies

of ERP components are displayed in Table 2. Grand Averages of the ERPs are displayed in

Figure 2. For brevity, we only report the effect size estimates of the critical manipulations, the

full statistical results of the ANOVAs can be found in the supplementary material (osf.io/6trne).

For descriptive plots that display the effects of experimental manipulations on behavioral RT in

the three EF tasks see Figure 3.

Switching. For reaction times there were substantial global switch costs as indicated by

the difference between control and both the shifting and shifting flanker blocks, p2 = .94. Within

shifting blocks, responses times were faster in trials with task repetition than task switches, p2 =

.69, indicating large local switch costs over all conditions (see left part of Figure 3A). While

there was a small difference in N1 latency across blocks, p2 = .11, the direction of the effect

contradicted the usual global switch costs (i.e. longer latencies for shifting than for control

blocks). In addition, we obtained no local switch cost on the N1 latency, p2 = .01. P3 latencies

showed no global switch costs, as P3 latency did not vary across blocks, p2 = .02. However, P3

latency was slightly shorter for switch than for repeat trials, p2 = .04, contradicting the usual

direction for local switch costs. Altogether, the Shifting manipulation was successful for

behavioral reaction times.

PS, WM, AND EF AS PREDICTORS OF G 23

With respect to the flanker manipulation in the shifting flanker block, response times

were slowest in trials with incongruent flankers and response times did not differ between trials

with neutral and congruent flankers, p2 = .31 (see right part of Figure 3A). In contrast, neither

N1, p2 = .00, nor P3 latency, p2 = .00, differed between flanker conditions. Similar to the

shifting manipulations, the flankers showed an effect for behavioral reaction that resembles the

standard inhibition effect of flanker stimuli (c.f. Eriksen & Eriksen, 1974)

N-Back. Neither behavioral reaction times nor N1 latencies varied between the 2-back

and the 2-back flanker block, both p2 = .00. Only P3 latencies were slightly shorter for the 2-

back than for the 2-back flanker block, p2 = .09. In contrast, only response times were faster for

match than for no-match trials in both blocks, p2 = .87 (see left part of Figure 3B). This is a

common difference between match and no match retrievals from memory. While N1 did not

differ between the two match conditions, p2 = .00, P3 latencies were slightly shorter for no

match than for match trials, p2 = .07, contradicting the usual direction of this effect.

All three dependent variables varied between flanker conditions in the flanker block, with

the largest effect for behavioral response times, p2 = .60 (see right part of Figure 3B), and

slightly smaller effects for N1, p2 = .21, and P3 latency, p2 = .14. However, this flanker effect

was mostly due to differences between the neutral and no flanker condition with congruent or

incongruent flankers. The critical inhibition effect – i.e. longer reaction times in incongruent than

in congruent trials – was, however, only present for behavioral reaction times in match trials. In

sum, there was no consistent inhibition effect of flanker stimuli in the N-Back task on behavioral

RTs.

ANT. Overall, participants’ response time as well as ERP latencies varied across cue

conditions (see Figure 3C). Again this effect was largest for behavioral response time, p2 = .87,
PS, WM, AND EF AS PREDICTORS OF G 24

and smaller for N1, p2 = .49, as well as P3 latency, p2 = .14. While all three dependent

variables showed an alerting effect (i.e. longer reaction times/latencies for no cues than for

double cues), an orienting effect (i.e. shorter latencies for spatial than for central cues) was only

present for behavioral reaction times. More importantly, behavioral reaction times showed a

strong inhibition effect, p2 = .93, with slowest response times for trials with incongruent

flankers compared to congruent or neutral flankers. However, neither N1, p2 = .04, nor P3

latency, p2 = .05, showed this inhibition effect. In sum, the manipulation of inhibition again was

successful for behavioral reaction times.

SEM Analysis: Bi-factor models for the EF tasks

Switching task. The bi-factor model capturing general and condition-specific variance of

behavioral RTs in the Shifting task (see Figure 4A) fitted well to the data, 2(53) = 41.1, p <

.884, CFI = 1.00, RMSEA = .00, 90% CI = [.00, .03]. The general processing speed factor

explained between 43 to 76% of the variance of the manifest indicators, while all condition

specific factors together explained between 36 to 51% of the variance in the manifest indicators.

In detail, the global shifting factor explained between 6 to 51%, the local shifting factor 9 to

10%, and the flanker factor between 21 to 23% of the variance in the manifest indicators.

Additional factors for inhibition or facilitation effects of flanker stimuli had non-significant

variances and were thus not included in the final model. Taken together, between 76 to 95% of

variance in manifest indicators was explained by both manipulation-specific factors and the

general factor.

The bi-factor model for the N1 latency in the Shifting task (see Figure 5A) showed a

good fit to the data, 2(60) = 51.7, p < .767, CFI = 1.00, RMSEA = .00, 90% CI = [.00, .05]. In

detail, the general factor of neural information processing speed captured 71 to 88 % of variance
PS, WM, AND EF AS PREDICTORS OF G 25

of manifest N1 latencies across the different conditions. Furthermore, only the factor capturing

individual differences in the flanker effect had a variance significantly different from zero,

capturing 8 to 9 % of variance in manifest N1 latencies. The other manipulation-specific factors

had variances not significantly different from zero and were not included in the model. In sum,

manipulation-specific factors and the general factor explained between 77 to 89% of variance in

manifest N1 latencies.

Likewise, the bi-factor model for P3 latencies (see Figure 6A) showed a good fit to the

data, 2(56) = 47.3, p < .983, CFI = 1.00, RMSEA = .00, 90% CI = [.00, .05]. In specific, the

overall speed factor explained between 61 to 93 % of variance in manifest P3 latencies, while all

condition specific factors together captured between 11 to 25% of variance in manifest P3

latencies. However, there was only a condition-specific factor for global switch cost that

captured between 9 to 11%, and a flanker-specific factor that captured between 13 to 14% of

variance in manifest P3 latencies. Variances of manipulation-specific factors for local switch

costs, facilitation or inhibition had non-significant variances. In this, all these factors together

explained between 81 to 93% of variance in manifest P3 latencies.

N-Back task. The bi-factor model for the behavioral RTs in the N-Back task (see Figure

4B) fitted well to the data, 2(56) =60.2, p < .326, CFI = 1.00, RMSEA = .03, 90% CI = [.00,

.07]. The general processing speed factor again explained the largest proportion of variance in

manifest RTs in the N-Back task with 43 to 81%. All condition-specific factors together

explained 18 to 52% of variance. Specifically, the flanker factor explained between 27 to 29%,

the no match factor explained between 18 to 21% and the facilitation factor 3% of the variance in

manifest indicators. The factor for inhibition effects of flanker stimuli showed a non-significant
PS, WM, AND EF AS PREDICTORS OF G 26

variance and was thus not included in the model. All these factors together explained between 81

to 95% of variance in manifest response times.

For the N1 latency in the N-Back task (see Figure 5B), the bi-factor model showed a

good fit to the data as well, 2(56) = 52.2, p < .619, CFI = 1.00, RMSEA = .00, 90% CI = [.00,

.06]. The general speed factor captured between 44 to 80 % of variance in manifest N1 latencies,

and only the flanker-specific factor had a variance significantly different from zero and captured

8 to 9 % of variance in manifest N1 latencies. Apart from that, none of the other manipulation-

specific factor had a variance significantly different from zero. Thus, the two factors with

significant variances explained between 52 to 92% of variance in manifest N1 latencies.

For the P3 latency, the bi-factor model (see Figure 6B) fit the data acceptably, 2(55) =

72.0, p < .062, CFI = .98, RMSEA = .06, 90% CI = [.00, .09]. The general neural speed factor

explained between 62 to 81% of variance, while all condition-specific factors explained between

8 to 22 % of variance in manifest P3 latencies. In detail, both the factor for no match trials and

the factor for the flanker manipulation explained 8 to 10% of variance and the factor for the

facilitation effect of flanker 7% of variance in manifest P3 latencies. The factors for inhibition

effects of the flanker stimuli did not differ from zero and was not included in the final model.

Together these factors explained between 80 to 89% of variance in manifest P3 latencies.

ANT. Consistent with results of the two other EF tasks, the bi-factor model for

behavioral RTs in the ANT (see Figure 4C) showed a good fit as well, 2(83) = 91.8, p < .238,

CFI = 1.00, RMSEA = .03, 90% CI = [.00, .07]. Again, consistent with results from the other two

EF tasks, the general processing speed factor explained between 86 to 96% of variance in

manifest indicators. Contrary to results from the other two EF tasks, the condition-specific

factors explained a lower amount of variance in manifest variables with only 2 to 9%.
PS, WM, AND EF AS PREDICTORS OF G 27

Specifically, the inhibition factor explained 7% and the no cue and spatial cue factor about 2% of

variance in manifest variables. Condition-specific factors for the other cue conditions and for

facilitation showed non-significant variances and were not included in the model. Taken

together, all factors explained between 93 and 96% of variance in manifest response times.

The bi-factor model for N1 latencies (see Figure 5C) fit the data acceptable, 2(83) =

114.7, p < .012, CFI = .942, RMSEA = .06, 90% CI = [.03, .09]. In this model, the general factor

of N1 latencies across all experimental manipulations captured between 36 to 62% of variance in

manifest N1 latencies, while the only experimental factor significantly differing from zero, the

factor for spatial cues, captured 17 to 18% of variances in the respective manifest N1 latencies.

All other manipulation-specific factors did not have variance significantly different from zero.

Altogether these factors explained between 47 to 64% of variance in manifest N1 latencies.

For the P3 latency in the ANT, the bi-factor model (see Figure 6C) showed a good fit to

the data as well, 2(83) = 79.7, p < .581, CFI = 1.00, RMSEA = .00, 90% CI = [.00, .06]. With

50 to 72% of variance the general factor capturing variance consistent across all experimental

manipulation explained most of the variance in manifest P3 latencies. In this model, the factor

for the no cue condition capturing 13% and the factor for the spatial cue condition capturing 19

to 20% of variance had variances significantly different from zero. The other manipulation-

specific factors did not have a variance significantly different from zero. In sum, these factors

explained between 62 to 75% of variance in manifest P3 latencies.

SEM Analysis: Relationship of EFs with WMC, PS, and Intelligence

Joint Models for the three EF tasks. A joint model of behavioral reaction times in the

three EF tasks indicated only correlations between the three general factors measured in the three

EF tasks (rs = .37 – .76). Additionally estimating correlations between manipulation-specific

PS, WM, AND EF AS PREDICTORS OF G 28

factors of the three EF tasks did not improve model fit, AIC = -30.8, 2(27) = 23.2, p = .672.

Moreover, joining the three general processing speed factors into one factor did not impair model

fit, AIC = -1.8, 2(1) = 0.2, p = .643, and represented a more parsimonious account of the

covariance structure. Specifically, the task-general factor explained all variance of the general

factor from the shifting task, 56 % of variance of the general ANT factor, and 27 % of variance

of the general N-Back factor. As this model showed a good fit to the data, 2(528) = 647.8, p <

.001, CFI = .98, RMSEA = .05, 90% CI = [.0533, .06], it was retained for further analyses with

the three covariates.

Similarly, the joint model for N1 latencies indicated correlations between the general N1

factors in all three EF tasks (rs = .62 - .93). Exploratory analysis revealed that additionally

estimating a correlation between the flanker factor from the shifting task with the flanker factor

from the N-back task (r = .81) and with the spatial cue factor from the ANT (r = .50) improved

the model fit, AIC = 16.4, 2(2) = 20.4, p < .001. In general, there was however no consistent

pattern of correlations between manipulation-specific factors that could have indicated shared

variance between the different manipulations. In addition, merging the general N1 latency factors

from the three different EF tasks in one task-general N1 factor did not impair model fit and

represented a more parsimonious account for the data, AIC = -1.3, 2(2) = 2.7, p = .261. This

task-general N1 factor captured all variance of the general N1 factor for the N-Back task, 87% of

variance of the general N1 factor for the shifting task, and 47% of variance of the general N1

factor for the ANT. While this model still showed only a mediocre fit to the data, 2(536) =

953.5, p < .001, CFI = .864, RMSEA = .09, 90% CI = [.08, .10], it still was retained for further

analysis as it was the best fitting model.

PS, WM, AND EF AS PREDICTORS OF G 29

The joint model for P3 latencies indicated correlations between the general P3 factors (rs

= .48 - 67) as well. Again exploratory analysis revealed that additionally estimating four

correlations between manipulation-specific factors improved the model fit, AIC = 19.6, 2(4)

= 27.6, p < .001. Specifically, correlations between the global shifting factor and the flanker

factor in the N-Back task (r = .73), between the flanker factor in the shifting task and the double

cue factor from the ANT (r = -.45), between the flanker factor in the N-back task and the no cue

factor from the ANT (r = -.59), and between the facilitation factor from the N-Back task and the

double cue factor from the ANT (r = .53) were significantly different from zero. Still, there was

no consistent pattern of correlations between manipulation-specific factors that would indicate a

general factor of the different manipulations across EF tasks. Moreover, joining the three general

P3 factors in one overarching factor for the P3 latency in the three EF tasks did not impair model

fit and represented a more parsimonious account for the data, AIC = -3.7, 2(2) = 0.3, p =

.845. In this model, the task-general P3 factor explained all variance in the general P3 factor for

the shifting task, 42% of variance in the general P3 factor for the N-Back task, and 47% of

variance of the general P3 factor for the ANT. Albeit, this model still had only a mediocre fit,

2(529) = 908.2, p < .001, CFI = .867, RMSEA = .09, 90% CI = [.08, .09]. It still was retained

for further analysis because it represented the best solution of all estimated models.

Altogether, the joint modeling of the three EF tasks indicated that general performance in

the three EF tasks was consistently correlated and could be merged into one factor. Although

there were some significant correlations between manipulation-specific factors, there were no

consistent patterns within these correlations suggesting that individual differences with respect to

specific manipulations were divergent rather than unitary.

PS, WM, AND EF AS PREDICTORS OF G 30

Bi-variate models of EF tasks and covariates. Detailed results of the bi-variate models

between EF tasks and the three covariates can be found in the analysis scripts available at:

osf.io/6trne. There were no consistent correlations between the three covariates (i.e. WMC, PS,

and intelligence) and manipulation-specific factors in the EF tasks. Interestingly, this was the

case both for behavioral RTs and latencies of ERP component measures in the EF tasks. Thus,

these correlations were not estimated in the joint model with all three covariates and the three EF

tasks. In contrast, the general factors that represented similar individual differences across the

three EF tasks showed considerable correlations and were thus included in the model joining all

three covariates and the three EF tasks. Detailed results will be reported in the next section.

Joint modeling of all covariates and EFs. A combined model with all three covariates

and behavioral RTs in the three EF tasks showed a good fit to the data, 2(940) = 1168.6, p <

.001, CFI = .97, RMSEA = .05, 90% CI = [.04, .06]. The path diagram of this model is shown in

the top part (A) of Figure 7. Specifically, the factor merging behavioral performance in the three

EF tasks showed a large positive correlation with processing speed in the ECTs (r = .77), and

slightly lower and negative correlations with both Gf (r = -.55) and WMC (r = -.49). In addition,

results indicated a very high correlation between Gf and WMC (r = .95), and medium

correlations of Gf and WMC with PS in the ECTs (r = -.46 to -.55). Due to the strong association

between PS in ECTs and the general performance factor of EFs, we simplified the model by

estimating one general processing speed factor consisting of EFs and ECTs and one factor

tentatively named higher cognition summarizing Gf and WMC (see bottom part of Figure 7 for

the path diagram). These simplifications did not impair model fit, AIC = -4.0, 2(7) = 10.0, p

= .191, and the model itself fit the data well, 2(947) = 1178.6, p < .001, CFI = .97, RMSEA =
PS, WM, AND EF AS PREDICTORS OF G 31

.05, 90% CI = [.04, .06]. This model indicated a medium correlation between the factor for

higher cognition and general processing speed (r = -.54).

The SEM combining N1 latencies from the EF tasks with all three covariates showed

only a mediocre fit to the data, 2(947) = 1926.6, p < .001, CFI = .77, RMSEA = .10, 90% CI =

[.09, .11]. All correlations between the task general N1 factor and covariates were low (rs = -.16

to .06) and non-significant (all ps > .160). In additions, assuming no covariance between the task-

general N1 latency factor and the three covariates did not impair model fit, AIC = -1.7, 2(3)

= 4.3, p = .230, further indicating that there was no correlation between the N1 latencies in the

three EF tasks and any of the three covariates (see Figure 8A for a path-diagram of this model).

Finally, the joint model of P3 latencies in the EF tasks with all three covariates showed a

mediocre fit to the data, 2(941) = 2366.2, p < .001, CFI = .68, RMSEA = .12, 90% CI = [.11,

.13]. Because the size of our sample did not allow for a more complex model to be estimated, we

provisionally retained this model. It must be noted that no strong conclusions can be drawn due

to unsatisfactory model fit. In detail, the task-general P3 factor showed significant correlations

with WMC (r = -.38), and PS from ECTs (r = .45), while the correlation with intelligence (r = -

.16) was non-significant (p = .179). Setting the correlation between intelligence and the task-

general P3 factor to zero did not impair model fit, AIC = -0.1, 2(1) = 1.9, p = .173, but

correlations with the other two covariates changed (see Figure 8B for a path diagram).
PS, WM, AND EF AS PREDICTORS OF G 32

Discussion

The present study aimed to disentangle the relationship between individual differences in

processing speed, working memory capacity and executive functions with general intelligence.

Specifically, we were interested in two different points: (1) in how far performance in EF tasks

represented general or manipulation-specific aspects. And (2) which of these two different

aspects of performance in EF tasks was related to intelligence, working memory capacity

(WMC), and processing speed (PS). Overall, performance in specific conditions within EF tasks

seemed to largely capture general variance rather than variance specific to an experimental

manipulation. Furthermore, manipulation-specific variance in EF tasks did not show any

consistent relationships among the different EF tasks and with the three covariates, while general

variance in behavioral RTs was related to all three covariates, and general variance in P3

latencies was related to WMC and PS.

Performance in EF tasks: What does it measure?

Before taking on the question in how far performance in EF tasks is related to WMC, PS

and intelligence, we addressed the question what is measured by reaction times as performance

measures in EF tasks. This is an important point as former studies investigating this relationship

have used various indicators for individual differences in EFs. Some studies have used

performance from specific conditions in an EF task that should require one specific executive

function (e.g., RTs for incongruent conditions in a Stroop task; Wongupparaj et al., 2015) or

average performance across conditions (e.g., mean proportion correct in updating tasks; Miyake

et al., 2000), while differences between specific experimental conditions in EF tasks were used

in other studies (e.g. the difference between RTs in switch versus repeat trials in a shifting task;

Friedman et al., 2006). Interestingly, these different measures have often been mixed within
PS, WM, AND EF AS PREDICTORS OF G 33

studies (Friedman & Miyake, 2017; Miyake et al., 2000; Wongupparaj et al., 2015), and in the

most cases difference measures have been used for inhibition and shifting tasks, while average

performance across conditions has been used for updating tasks.

From a theoretical perspective, the use of either measure is not entirely unproblematic.

Using performance from a single condition or average performance across conditions may

confound different variance sources. In this case, such measures may contain variance specific to

experimental manipulations that require specific executive functions and more general variance

linked to processing speed or memory capacity (Frischkorn & Schubert, 2018). In contrast,

difference measures assume that cognitive processes are additive and that experimental

conditions vary in all but a single cognitive process (Donders, 1868, 1969). Yet, it is likely that

cognitive processes are not additive and that inserting an additional cognitive process interacts

with other cognitive processes required in the task (Alexander, Trengove, & van Leeuwen, 2015;

Friston et al., 1996; Schubert et al., 2015).

Our results from the bi-factor models of the three EF tasks indicate that behavioral RTs and

ERP latencies from one condition within any EF task represented mostly general performance.

Specifically, the general factors summarizing the variance consistent across all manipulations did

capture the largest proportion of variance in manifest indicators across all EF tasks (on average

68%), while each manipulation-specific factor captured considerably smaller variance

proportions (on average 14%). Hence, behavioral or neural measures from a single condition or

average performance across all conditions will mostly represent individual differences in general

rather than manipulation-specific cognitive processes.

As manipulation-specific factors can be interpreted as latent difference scores, the small

amount of variance of manipulation-specific factors might also explain why difference scores in
PS, WM, AND EF AS PREDICTORS OF G 34

experimental paradigms often show low reliabilities (Hedge, Powell, & Sumner, 2018). In detail,

when calculating the difference between two correlated experimental conditions, the small

amount of systematic variance in this difference (i.e. the variance of manipulation-specific

factors) gets outweighed by unsystematic error variance that gets amplified when calculating the

difference between highly correlated variables. One very recently proposed solution to this may

be to account for trial-to-trial noise by adopting a hierarchical modeling approach (Rouder &

Haaf, 2018). However, it remains to be seen whether this approach solves the reliability issues of

difference measures, and provides an increment above the here used method of latent difference

scores that are virtually error free.

In conclusion, researchers have to consider that the selection of a specific measure such

as difference scores or performance in a single experimental condition may change both the

interpretation of the measure and the relationship with covariates. Specifically, results from

studies that have not used difference scores as indicators of EFs could also be interpreted as

indication that general processing speed (when using RTs in shifting or inhibition tasks) or

memory/processing capacity (when using accuracies in updating tasks) are related to general

intelligence instead of individual differences specific to executive functions. Since difference

measures are not unproblematic either (Friston et al., 1996), developing theoretically founded

measures for executive functions is a critical step towards accurately assessing the relationship

between EFs and individual differences in other cognitive processes such as intelligence, WMC,

and PS.

Executive Functions: Still no bridge across the gap.

While theoretically and empirically executive functions are supposed to underlie working

memory capacity and thus to be related to intelligence as well (Kane, Conway, Hambrick, &
PS, WM, AND EF AS PREDICTORS OF G 35

Engle, 2007; Kane & Engle, 2003; Unsworth, 2010; Unsworth et al., 2014), the results of the

present study indicated that variance specific to experimental manipulations is consistently

unrelated to working memory capacity, processing speed, and general intelligence. Moreover,

manipulation-specific factors did not show any consistent correlational pattern with each other,

indicating that executive functions required by the different experimental manipulations are

divergent rather than unitary. This is in line with recent results suggesting that individual

differences in executive functions, specifically inhibition, may not be as unitary as suggested

(Rey-Mermet, Gade, & Oberauer, 2018; Stahl et al., 2014). In sum, correlations between

difference measures from executive functioning tasks both with each other and with external

criteria seem to be small and largely inconsistent (Hedge et al., 2018; Rey-Mermet et al., 2018;

Stahl et al., 2014), calling into question (1) whether individual differences in executive functions

are unitary at all and (2) whether they underlie the relationship between WMC, PS and

intelligence.

Instead, only variance consistent across experimental manipulations showed relationships

with the three covariates. Specifically, behavioral RTs in EF tasks showed medium-sized

negative relationships with working memory capacity and general intelligence (r = -.49 to -.55),

and a large positive relationship with PS from ECTs (r = .77). P3 latencies showed medium

correlations with processing speed (r = .40), a small negative correlation with WMC (r = -.28),

but no relationship with intelligence. Finally, N1 latencies showed no significant relationships

with any of the three covariates. While smaller relationships of ERP latencies are often observed

due to their tendency for lower reliability (Cassidy, Robertson, & O’Connell, 2012), the present

results do not suggest that this is a problem, because residual error variance captured only a

small proportion of variance in manifest ERP latencies. Nevertheless, it is important to note that
PS, WM, AND EF AS PREDICTORS OF G 36

model fit was unsatisfactory for all structural equation models including ERP latencies,

especially P3 latencies, and the covariates (i.e. intelligence, WMC, and PS). Thus, the results for

ERP latencies need to be replicated with tasks better suited for the EEG and a larger sample.

Until then they should be interpreted with caution.

Although relationships of ERP latencies with intelligence were smaller and not as

consistent as suggested by results from a recent study (Schubert, Hagemann, & Frischkorn,

2017), the present results replicated the finding that the speed of higher-order cognitive processes

occurring later in the stream of neural processing is related to cognitive processes. In detail,

latencies of the N1 component showed no significant relationship with general intelligence,

working memory capacity, and processing speed, while latencies of the P3 component were

positively correlated with processing speed, and negatively correlated with WMC. These results

are in line with the theoretical interpretation by Schubert et al. (2018) and further support the

idea that specific neural or cognitive processes underlie the relationship between general

intelligence and information processing (Kievit, Davis, Griffiths, Correia, & Henson, 2016;

McVay & Kane, 2012; van Ravenzwaaij, Brown, & Wagenmakers, 2011).

Speed of higher-order information processing: Basis of general intelligence?

Although behavioral results indicated that processing speed measured in both EF tasks

and ECTs is related to intelligence and WMC, the present results did not replicate the large

relationship between latencies of later ERP components (P2, N2, and P3) with general

intelligence (Schubert, Hagemann, & Frischkorn, 2017). In fact, the present results did not show

any correlation of P3 latency with general intelligence. Nevertheless, there was a small

correlation of P3 latency with working memory capacity (r = -.28) and a medium correlation

with basic information processing speed (r = .40). Still, the non-significant correlation of P3
PS, WM, AND EF AS PREDICTORS OF G 37

latency with general intelligence in the present study and the overall smaller correlations demand

an explanation.

One important aspect is that the present study used considerably different tasks. On the

one hand the present tasks were more complex resulting in longer reaction times and more errors.

And on the other hand, the present tasks integrated multiple cognitive processes to be performed

simultaneously. The study by Schubert et al. (2017) used elementary cognitive tasks such as the

Posner Letter matching task or the Sternberg task, whereas the present study measured latencies

of ERP components in more complex executive functioning tasks. In detail, ECTs are designed

to systematically vary the demand on one specific cognitive process (Hick, 1952; Posner &

Mitchell, 1967; Sternberg, 1969). Thus, individual differences in latencies of ERPs computed in

ECTs may specifically capture individual differences in the speed of this specific process (e.g.,

memory retrieval) the ECTs are designed to tap.

In contrast, the tasks used in the present study required at least two different cognitive

processes while processing the target stimulus. In all three tasks, a decision with respect to the

target stimulus had to be made, while additionally irrelevant information had to be ignored (i.e.,

inhibition in the ANT, and in flanker blocks in the Shifting and N-Back task), the decision task

to be conducted had to be determined (i.e., shifting), or the target stimulus had to be encoded in

memory while outdated information was being removed (i.e., updating in the N-back task). It is

likely that these different processes run at least partly in parallel. As soon as there is a latency

jitter between this two processes this will smear out the ERP (Ouyang, Herzmann, Zhou, &

Sommer, 2011) and render the two processes inseparable. Thus, individual differences in average

ERP latencies may mix up individual differences with respect to the decision process and other

processes such as inhibition, updating, and shifting.

PS, WM, AND EF AS PREDICTORS OF G 38

Results from both behavioral and neural measures suggested that individual differences

specific to executive functions were not related to each other and other cognitive processes.

However, the variance stemming from these EF processes might still be reflected in latencies of

ERP components in addition to variance due to individual differences of the decision process.

This uncorrelated additional variance may have masked the strong relationship of purely

decision-specific variance in the latency of later ERP components with general intelligence.

Processing speed, working memory capacity and general intelligence: The missing link

Neural speed of information processing showed small and inconsistent relationships with

general intelligence, working memory capacity, and processing speed. Nevertheless, the present

results replicate the negative relationship (r = -.54) of behavioral processing speed with both

intelligence and WMC and the strong relationship between the latter processes (Ackerman et al.,

2005; Kyllonen & Christal, 1990; Schmiedek et al., 2007; Schmitz & Wilhelm, 2016; Schubert et

al., 2015). Moreover, results still showed that individuals with higher working memory capacity

and faster behavioral processing speed have shorter latencies of the P3 component. Taken

together, we observed a considerable overlap between individual differences in speed of

information processing, working memory and general intelligence that requires a theoretical

explanation.

Even though researchers have often argued for a causal relationship between basic

cognitive processes such as processing speed or working memory and general intelligence, there

may also be a confounding variable that affects all these different cognitive processes. For

instance, a recent study using a psychopharmacological manipulation of processing speed with

nicotine indicated that the speed of neural information processing might not causally underlie

individual differences in general intelligence (Schubert, Hagemann, Frischkorn, & Herpertz,

PS, WM, AND EF AS PREDICTORS OF G 39

2018). In detail, nicotine administration did increase neural as well as behavioral processing

speed, while not showing any effect on performance in a matrix reasoning task. Hence,

processing speed may not causally underlie individual differences in intelligence despite being

correlated with intelligence.

Conclusion

Altogether, the present results further emphasize the important role of processing speed

and working memory for individual differences in general intelligence. In contrast, executive

functions did not underlie either individual differences in processing speed or working memory

capacity and thus did not explain why these two domains are related to general intelligence.

Nevertheless, it is reasonable to assume that individual differences in both processing speed and

working memory capacity arise due to similar limitations in the cognitive system (Meiran &

Shahar, 2018; Wilhelm & Oberauer, 2006). A promising approach to further investigate this idea

might lie in joining theoretically grounded measures for behavioral indicators of processing

speed and working memory capacity (e.g. cognitive models; Frischkorn & Schubert, 2018), with

biological indicators of neural processing related to these two processes (c.f. Schubert, Nunez,

Hagemann, & Vandekerckhove, 2018). More comprehensive insights on the basic cognitive

processes underlying individual differences in general intelligence may thus be gained by

associating structural and function architectural features of the brain related to intelligence

(Hilger, Ekman, Fiebach, & Basten, 2017; Menon & Uddin, 2010) with cognitive process

domains such as working memory capacity and processing speed. Ultimately, this may provide

the integration of working memory capacity and processing speed as related predictors of

intelligence that could not be reached by executive functions.

PS, WM, AND EF AS PREDICTORS OF G 40

References

Ackerman, P. L., Beier, M. E., & Boyle, M. D. (2002). Individual differences in working
memory within a nomological network of cognitive and perceptual speed abilities.
Journal of Experimental Psychology: General, 131(4), 567–589. https://doi.org/10/bjtjnz
Ackerman, P. L., Beier, M. E., & Boyle, M. O. (2005). Working Memory and Intelligence: The
Same or Different Constructs? Psychological Bulletin, 131(1), 30–60.
https://doi.org/10/drqmhx
Alexander, D. M., Trengove, C., & van Leeuwen, C. (2015). Donders is dead: cortical traveling
waves and the limits of mental chronometry in cognitive neuroscience. Cognitive
Processing, 16(4), 365–375. https://doi.org/10/f7zgts
Barrouillet, P., Portrat, S., & Camos, V. (2011). On the law relating processing to storage in
working memory. Psychological Review, 118(2), 175–192. https://doi.org/10/bq7rsv
Bayliss, D. M., Jarrold, C., Gunn, D. M., & Baddeley, A. D. (2003). The complexities of
complex span: Explaining individual differences in working memory in children and
adults. Journal of Experimental Psychology: General, 132(1), 71–92.
https://doi.org/10/d6hwsp
Bazana, P. G., & Stelmack, R. M. (2002). Intelligence and information processing during an
auditory discrimination task with backward masking: an event-related potential analysis.
Journal of Personality and Social Psychology, 83(4), 998–1008.
Bentler, P. M. (1990). Comparative fit indexes in structural models. Psychological Bulletin,
107(2), 238–246. https://doi.org/10/dbj
Browne, M. W., & Cudeck, R. (1992). Alternative Ways of Assessing Model Fit. Sociological
Methods & Research, 21(2), 230–258. https://doi.org/10/dbn
Burgess, G. C., Gray, J. R., Conway, A. R. A., & Braver, T. S. (2011). Neural mechanisms of
interference control underlie the relationship between fluid intelligence and working
memory span. Journal of Experimental Psychology: General, 140(4), 674–692.
https://doi.org/10/cvxsd7
Burnham, K. P., & Anderson, D. R. (2002). Model Selection and Multimodel Inference: A
Practical Information-Theoretic Approach (2nd ed.). New York: Springer-Verlag.
Retrieved from //www.springer.com/de/book/9780387953649
Cassidy, S. M., Robertson, I. H., & O’Connell, R. G. (2012). Retest reliability of event-related
potentials: Evidence from a variety of paradigms. Psychophysiology, 49(5), 659–664.
https://doi.org/10/gd3vr9
Chuderski, A., Taraday, M., Nęcka, E., & Smoleń, T. (2012). Storage capacity explains fluid
intelligence but executive control does not. Intelligence, 40(3), 278–295.
https://doi.org/10/f3zwhw
Colom, R., Abad, F. J., Quiroga, M. Á., Shih, P. C., & Flores-Mendoza, C. (2008). Working
memory and intelligence are highly related constructs, but why? Intelligence, 36(6), 584–
606. https://doi.org/10/dn58z8
Colom, R., Flores-Mendoza, C., Quiroga, M. Á., & Privado, J. (2005). Working memory and
general intelligence: The role of short-term storage. Personality and Individual
Differences, 39(5), 1005–1014. https://doi.org/10/fvckcg
Colom, R., Jung, R. E., & Haier, R. J. (2007). General intelligence and memory span: Evidence
for a common neuroanatomic framework. Cognitive Neuropsychology, 24(8), 867–878.
https://doi.org/10/bn8rqz
PS, WM, AND EF AS PREDICTORS OF G 41

Colom, R., & Thompson, P. M. (2013). Understanding Human Intelligence by Imaging the
Brain. In The Wiley-Blackwell Handbook of Individual Differences (pp. 330–352). Wiley-
Blackwell. https://doi.org/10.1002/9781444343120.ch12
Conway, A. R. ., Cowan, N., Bunting, M. F., Therriault, D. J., & Minkoff, S. R. . (2002). A latent
variable analysis of working memory capacity, short-term memory capacity, processing
speed, and general fluid intelligence. Intelligence, 30(2), 163–183.
https://doi.org/10/frs9t8
Cowan, N. (2017). The many faces of working memory and short-term storage. Psychonomic
Bulletin & Review, 24(4), 1158–1170. https://doi.org/10/gbvchk
Dang, C.-P., Braeken, J., Colom, R., Ferrer, E., & Liu, C. (2015). Do processing speed and short-
term storage exhaust the relation between working memory capacity and intelligence?
Personality and Individual Differences, 74, 241–247. https://doi.org/10/f6xc38
Delorme, A., & Makeig, S. (2004). EEGLAB: an open source toolbox for analysis of single-trial
EEG dynamics including independent component analysis. Journal of Neuroscience
Methods, 134(1), 9–21. https://doi.org/10/bqr2f2
Donchin, E. (1981). Surprise!… Surprise? Psychophysiology, 18(5), 493–513.
https://doi.org/10/ff2cmm
Donders, F. C. (1868). Over de snelheid van psychische processen. Onderzoekingen gedaan in
het Physiologisch Laboratorium der Utrechtsche Hoogeschool (1868–1869), 2, 92–120.
Retrieved from
https://pure.mpg.de/pubman/faces/ViewItemOverviewPage.jsp?itemId=item_2316352
Donders, F. C. (1969). On the speed of mental processes. Acta Psychologica, 30, 412–431.
https://doi.org/10/cjj95t
Engle, R. W., Tuholski, S. W., Laughlin, J. E., & Conway, A. R. (1999). Working memory,
short-term memory, and general fluid intelligence: A latent-variable approach. Journal of
Experimental Psychology. General, 128(3), 309–331. https://doi.org/10/ddp3gg
Eriksen, B. A., & Eriksen, C. W. (1974). Effects of noise letters upon the identification of a
target letter in a nonsearch task. Perception & Psychophysics, 16(1), 143–149.
https://doi.org/10/d23822
Friedman, N. P., & Miyake, A. (2017). Unity and diversity of executive functions: Individual
differences as a window on cognitive structure. Cortex, 86, 186–204.
https://doi.org/10/f9sqwm
Friedman, N. P., Miyake, A., Corley, R. P., Young, S. E., DeFries, J. C., & Hewitt, J. K. (2006).
Not All Executive Functions Are Related to Intelligence. Psychological Science, 17(2),
172–179. https://doi.org/10/bmb68s
Frischkorn, G. T., & Schubert, A.-L. (2018). Cognitive Models in Intelligence Research:
Advantages and Recommendations for Their Application. Journal of Intelligence, 6(3),
34. https://doi.org/10/gd3vqn
Friston, K. J., Price, C. J., Fletcher, P., Moore, C., Frackowiak, R. S., & Dolan, R. J. (1996). The
trouble with cognitive subtraction. NeuroImage, 4(2), 97–104. https://doi.org/10/drwjs5
Gläscher, J., Rudrauf, D., Colom, R., Paul, L. K., Tranel, D., Damasio, H., & Adolphs, R.
(2010). Distributed neural system for general intelligence revealed by lesion mapping.
Proceedings of the National Academy of Sciences, 107(10), 4705–4709.
https://doi.org/10/b9h5df
PS, WM, AND EF AS PREDICTORS OF G 42

Hedge, C., Powell, G., & Sumner, P. (2018). The reliability paradox: Why robust cognitive tasks
do not produce reliable individual differences. Behavior Research Methods, 50(3), 1166–
1186. https://doi.org/10/gddfm4
Hick, W. E. (1952). On the Rate of Gain of Information. Quarterly Journal of Experimental
Psychology, 4(1), 11–26. https://doi.org/10/fc97vw
Hilger, K., Ekman, M., Fiebach, C. J., & Basten, U. (2017). Efficient hubs in the intelligent
brain: Nodal efficiency of hub regions in the salience network is associated with general
intelligence. Intelligence, 60, 10–25. https://doi.org/10/f9jg3q
Hu, L.-T., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure
analysis: Conventional criteria versus new alternatives. Structural Equation Modeling,
6(1), 1–55. https://doi.org/10/dbt
Jäger, A. O. (1982). Mehrmodale Klassifikation von Intelligenzlestungen: Experimentell
kontrollierte Weiterenwicklung eines deskriptiven Intelligenzstrukturmodells.
[Multimodal classification of intelligence achievement: Experimentally controlled,
further development of a descriptive intelligence structure model.]. Diagnostica, 28(3),
195–225.
Jäger, A. O., Süß, H.-M., & Beauducel, A. (1997). Berliner Intelligenzstruktur-Test: Form 4.
Göttingen: Hogrefe.
Jensen, A. R. (2006). Clocking the Mind: Mental Chronometry and Individual Differences.
Amsterdam: Elsevier.
Jewsbury, P. A., Bowden, S. C., & Strauss, M. E. (2016). Integrating the switching, inhibition,
and updating model of executive function with the Cattell—Horn—Carroll model.
Journal of Experimental Psychology: General, 145(2), 220–245.
https://doi.org/10/f79zsq
Jung, R. E., & Haier, R. J. (2007). The Parieto-Frontal Integration Theory (P-FIT) of
intelligence: Converging neuroimaging evidence. Behavioral and Brain Sciences, 30(02),
135. https://doi.org/10/d8ddm9
Kane, M. J., Conway, A. R. A., Hambrick, D. Z., & Engle, R. W. (2007). Variation in Working
Memory Capcacity as Variation in Executive Attention and Control. In A. R. A. Conway
& M. Jarrold (Eds.), Variation in Working Memory (pp. 21–46). Oxford University Press.
Kane, M. J., & Engle, R. W. (2003). Working-memory capacity and the control of attention: The
contributions of goal neglect, response competition, and task set to Stroop interference.
Journal of Experimental Psychology: General, 132(1), 47–70. https://doi.org/10/fwhz92
Kane, M. J., Hambrick, D. Z., & Conway, A. R. A. (2005). Working memory capacity and fluid
intelligence are strongly related onstructs: Comment on Ackerman, Beier and Boyle
(2005). Psychological Bulletin, 131, 66–71. https://doi.org/10/fgtzfp
Kane, M. J., Hambrick, D. Z., Tuholski, S. W., Wilhelm, O., Payne, T. W., & Engle, R. W.
(2004). The Generality of Working Memory Capacity: A Latent-Variable Approach to
Verbal and Visuospatial Memory Span and Reasoning. Journal of Experimental
Psychology: General, 133(2), 189–217. https://doi.org/10/cntzrs
Karr, J. E., Areshenkoff, C. N., Rast, P., Hofer, S. M., Iverson, G. L., & Garcia-Barrera, M. A.
(2018). The unity and diversity of executive functions: A systematic review and re-
analysis of latent variable studies. Psychological Bulletin. https://doi.org/10/gd3vsx
Kenny, D. A., Kaniskan, B., & McCoach, D. B. (2015). The Performance of RMSEA in Models
With Small Degrees of Freedom. Sociological Methods & Research, 44(3), 486–507.
https://doi.org/10/f7kdvc
PS, WM, AND EF AS PREDICTORS OF G 43

Kievit, R. A., Davis, S. W., Griffiths, J., Correia, M. M., & Henson, R. N. (2016). A watershed
model of individual differences in fluid intelligence. Neuropsychologia, 91, 186–198.
https://doi.org/10/f892mx
Kovacs, K., & Conway, A. R. A. (2016). Process Overlap Theory: A Unified Account of the
General Factor of Intelligence. Psychological Inquiry, 27(3), 151–177.
https://doi.org/10/gd3vr6
Kranzler, J. H., & Jensen, A. R. (1991). The nature of psychometric g: Unitary process or a
number of independent processes?. Intelligence, 15(4), 397–422.
https://doi.org/10/bz2p63
Kyllonen, P. C., & Christal, R. E. (1990). Reasoning ability is (little more than) working-
memory capacity?! Intelligence, 14(4), 389–433. https://doi.org/10/bxmdv4
Lewandowsky, S., Oberauer, K., Yang, L.-X., & Ecker, U. K. H. (2010). A working memory test
battery for MATLAB. Behavior Research Methods, 42(2), 571–585.
https://doi.org/10/c7646z
Lopez-Calderon, J., & Luck, S. J. (2014). ERPLAB: an open-source toolbox for the analysis of
event-related potentials. Frontiers in Human Neuroscience, 8. https://doi.org/10/gdz7rc
McGarry-Roberts, P. A., Stelmack, R. M., & Campbell, K. B. (1992). Intelligence, reaction time,
and event-related potentials. Intelligence, 16(3), 289–313. https://doi.org/10/fcrtwf
McVay, J. C., & Kane, M. J. (2009). Conducting the train of thought: Working memory capacity,
goal neglect, and mind wandering in an executive-control task. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 35(1), 196–204.
https://doi.org/10/d2bmhv
McVay, J. C., & Kane, M. J. (2012). Drifting From Slow to “D’oh!”: Working Memory Capacity
and Mind Wandering Predict Extreme Reaction Times and Executive Control Errors.
Journal of Experimental Psychology: Learning, Memory and Cognition, 38, 525–549.
https://doi.org/10/c945qk
Meier, M. E., & Kane, M. J. (2013). Working memory capacity and Stroop interference: Global
versus local indices of executive control. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 39(3), 748–759. https://doi.org/10/gd59t9
Meiran, N., & Shahar, N. (2018). Working memory involvement in reaction time and its
contribution to fluid intelligence: An examination of individual differences in reaction-
time distributions. Intelligence, 69, 176–185. https://doi.org/10/gdxqtm
Menon, V., & Uddin, L. Q. (2010). Saliency, switching, attention and control: a network model
of insula function. Brain Structure and Function, 214(5), 655–667.
https://doi.org/10/fmz3ch
Miyake, A., & Friedman, N. P. (2012). The Nature and Organization of Individual Differences in
Executive Functions: Four General Conclusions. Current Directions in Psychological
Science, 21(1), 8–14. https://doi.org/10/fzdfvs
Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A., & Wager, T. D.
(2000). The Unity and Diversity of Executive Functions and Their Contributions to
Complex “Frontal Lobe” Tasks: A Latent Variable Analysis. Cognitive Psychology,
41(1), 49–100. https://doi.org/10/bkksp2
Miyake, A., Friedman, N. P., Rettinger, D. A., Shah, P., & Hegarty, M. (2001). How are
visuospatial working memory, executive functioning, and spatial abilities related? A
latent-variable analysis. Journal of Experimental Psychology: General, 130(4), 621–640.
https://doi.org/10/cp4w8m
PS, WM, AND EF AS PREDICTORS OF G 44

Mognon, A., Jovicich, J., Bruzzone, L., & Buiatti, M. (2011). ADJUST: An automatic EEG
artifact detector based on the joint use of spatial and temporal features: Automatic spatio-
temporal EEG artifact detection. Psychophysiology, 48(2), 229–240.
https://doi.org/10/cwhtpn
Oberauer, K. (2002). Access to information in working memory: Exploring the focus of
attention. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28(3),
411–421. https://doi.org/10/bpqg2q
Oberauer, K., Farrell, S., Jarrold, C., & Lewandowsky, S. (2016). What limits working memory
capacity? Psychological Bulletin, 142(7), 758–799. https://doi.org/10/f8t2hx
Oberauer, K., & Kliegl, R. (2006). A formal model of capacity limits in working memory.
Journal of Memory and Language, 55(4), 601–626. https://doi.org/10/bm923b
Oberauer, K., Schulze, R., Wilhelm, O., & Süß, H.-M. (2005). Working Memory and
Intelligence--Their Correlation and Their Relation: Comment on Ackerman, Beier, and
Boyle (2005). Psychological Bulletin, 131(1), 61–65. https://doi.org/10/cfjxgw
Oberauer, K., Süß, H.-M., Wilhelm, O., & Wittman, W. W. (2003). The multiple faces of
working memory: Storage, processing, supervision, and coordination. Intelligence, 31(2),
167–193. https://doi.org/10/fs4vfj
Ouyang, G., Herzmann, G., Zhou, C., & Sommer, W. (2011). Residue iteration decomposition
(RIDE): A new method to separate ERP components on the basis of latency variability in
single trials. Psychophysiology, 48(12), 1631–1647. https://doi.org/10/bqvh3b
Polich, J. (2007). Updating P300: An integrative theory of P3a and P3b. Clinical
Neurophysiology, 118(10), 2128–2148. https://doi.org/10/hmf
Posner, M. I., & Mitchell, R. F. (1967). Chronometric analysis of classification. Psychological
Review, 74(5), 392–409. https://doi.org/10/bm5nqh
Posner, M. I., & Petersen, S. E. (1990). The attention system of the human brain. Annual Review
of Neuroscience, 13, 25–42. https://doi.org/10/b486gb
Rao, K. V., & Baddeley, A. (2013). Raven’s matrices and working memory: A dual-task
approach. The Quarterly Journal of Experimental Psychology, 66(10), 1881–1887.
https://doi.org/10/gd59vd
Ratcliff, R. (1978). A theory of memory retrieval. Psychological Review, 85(2), 59–108.
https://doi.org/10/fjwm2f
Ratcliff, R., Thapar, A., & McKoon, G. (2010). Individual differences, aging, and IQ in two-
choice tasks. Cognitive Psychology, 60(3), 127–157. https://doi.org/10/fndcr6
Rey-Mermet, A., Gade, M., & Oberauer, K. (2018). Should we stop thinking about inhibition?
Searching for individual and age differences in inhibition ability. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 44(4), 501–526.
https://doi.org/10/gcx8pf
Rouder, J., & Haaf, J. M. (2018). A Psychometrics of Individual Differences in Experimental
Tasks. https://doi.org/10/gfdbw2
Sauseng, P., Klimesch, W., Freunberger, R., Pecherstorfer, T., Hanslmayr, S., & Doppelmayr,
M. (2006). Relevance of EEG alpha and theta oscillations during task switching.
Experimental Brain Research, 170(3), 295–301. https://doi.org/10/dm39wk
Scharinger, C., Soutschek, A., Schubert, T., & Gerjets, P. (2015). When flanker meets the n-
back: What EEG and pupil dilation data reveal about the interplay between the two
central-executive working memory functions inhibition and updating: When flanker
meets the n-back. Psychophysiology, 52(10), 1293–1304. https://doi.org/10/f7qn24
PS, WM, AND EF AS PREDICTORS OF G 45

Schmiedek, F., Oberauer, K., Wilhelm, O., Süß, H.-M., & Wittmann, W. W. (2007). Individual
differences in components of reaction time distributions and their relations to working
memory and intelligence. Journal of Experimental Psychology: General, 136(3), 414–
429. https://doi.org/10/dwr688
Schmitz, F., & Wilhelm, O. (2016). Modeling Mental Speed: Decomposing Response Time
Distributions in Elementary Cognitive Tasks and Correlations with Working Memory
Capacity and Fluid Intelligence. Journal of Intelligence, 4(4). https://doi.org/10/gd3vrp
Schubert, A.-L., Frischkorn, G. T., Hagemann, D., & Voss, A. (2016). Trait Characteristics of
Diffusion Model Parameters. Journal of Intelligence, 4(3), 7. https://doi.org/10/gd3vs3
Schubert, A.-L., Hagemann, D., & Frischkorn, G. T. (2017). Is general intelligence little more
than the speed of higher-order processing? Journal of Experimental Psychology:
General, 146(10), 1498–1512. https://doi.org/10/gch83n
Schubert, A.-L., Hagemann, D., Frischkorn, G. T., & Herpertz, S. C. (2018). Faster, but not
smarter: An experimental analysis of the relationship between mental speed and mental
abilities. Intelligence, 71, 66–75. https://doi.org/10/gffjb9
Schubert, A.-L., Hagemann, D., Voss, A., & Bergmann, K. (2017). Evaluating the model fit of
diffusion models with the root mean square error of approximation. Journal of
Mathematical Psychology, 77, 29–45. https://doi.org/10/f943hk
Schubert, A.-L., Hagemann, D., Voss, A., Schankin, A., & Bergmann, K. (2015). Decomposing
the relationship between mental speed and mental abilities. Intelligence, 51, 28–46.
https://doi.org/10/f7mtv3
Schubert, A.-L., Nunez, M. D., Hagemann, D., & Vandekerckhove, J. (2018). Individual
differences in cortical processing speed predict cognitive abilities: a model-based
cognitive neuroscience account. BioRxiv, 374827. https://doi.org/10/gfhzdz
Schulter, G., & Neubauer, A. C. (2005). Zentralnervensystem und Persönlichkeit [Central
nervous system and personality]. In J. Henning & P. Netter (Eds.), Biopsychologische
Grundlagen der Persönlichkeit (pp. 35–190). Munich, Germany: Elsevier.
Shah, P., & Miyake, A. (1996). The separability of working memory resources for spatial
thinking and language processing: An individual differences approach. Journal of
Experimental Psychology: General, 125(1), 4–27. https://doi.org/10/cgvnhr
Shahabi, S. R., Abad, F. J., & Colom, R. (2014). Short-term storage is a stable predictor of fluid
intelligence whereas working memory capacity and executive function are not: A
comprehensive study with Iranian schoolchildren. Intelligence, 44, 134–141.
https://doi.org/10/f576d3
Sheppard, L. D., & Vernon, P. A. (2008). Intelligence and speed of information-processing: A
review of 50 years of research. Personality and Individual Differences, 44(3), 535–551.
https://doi.org/10/fn3gm2
Souza, A. S., & Vergauwe, E. (2018). Unravelling the intersections between consolidation,
refreshing, and removal. Annals of the New York Academy of Sciences, 1424(1), 5–7.
https://doi.org/10/gdxvb7
Stahl, C., Voss, A., Schmitz, F., Nuszbaum, M., Tüscher, O., Lieb, K., & Klauer, K. C. (2014).
Behavioral components of impulsivity. Journal of Experimental Psychology: General,
143(2), 850–886. https://doi.org/10/f62q5t
Sternberg, S. (1969). Memory-scanning: Mental processes revealed by reaction-time
experiments. American Scientist, 57(4), 421–457. Retrieved from
https://www.jstor.org/stable/27828738
PS, WM, AND EF AS PREDICTORS OF G 46

Troche, S. J., Houlihan, M. E., Stelmack, R. M., & Rammsayer, T. H. (2009). Mental ability,
P300, and mismatch negativity: Analysis of frequency and duration discrimination.
Intelligence, 37(4), 365–373. https://doi.org/10/dm4nz7
Unsworth, N. (2010). On the division of working memory and long-term memory and their
relation to intelligence: A latent variable approach. Acta Psychologica, 134(1), 16–28.
https://doi.org/10/bmxh4j
Unsworth, N., Fukuda, K., Awh, E., & Vogel, E. K. (2014). Working memory and fluid
intelligence: Capacity, attention control, and secondary memory retrieval. Cognitive
Psychology, 71, 1–26. https://doi.org/10/f54shj
van Ravenzwaaij, D., Brown, S., & Wagenmakers, E.-J. (2011). An integrated perspective on the
relation between response speed and intelligence. Cognition, 119(3), 381–393.
https://doi.org/10/dwf8d6
Verleger, R. (1997). On the utility of the P3 latency as an index of mental chronometry.
Psychophysiology, 34, 131–156. https://doi.org/10/fvv22s
Wilhelm, O., Hildebrandt, A., & Oberauer, K. (2013). What is working memory capacity, and
how can we measure it? Frontiers in Psychology, 4, 433. https://doi.org/10/gd3vs7
Wilhelm, O., & Oberauer, K. (2006). Why are reasoning ability and working memory capacity
related to mental speed? An investigation of stimulus–response compatibility in choice
reaction time tasks. European Journal of Cognitive Psychology, 18(1), 18–50.
https://doi.org/10/dgsf8g
Wongupparaj, P., Kumari, V., & Morris, R. G. (2015). The relation between a multicomponent
working memory and intelligence: The roles of central executive and short-term storage
functions. Intelligence, 53, 166–180. https://doi.org/10/gd3vsr
PS, WM, AND EF AS PREDICTORS OF G 47

Tables
Table 1
Descriptive Statistics for the Executive Functioning tasks
Task Block Task Shifting Flanker M RT (SDRT) M Pc (SDPc )
Control LM 583.23 (94.12) 0.93 (0.01)
Control OE 679.28 (130.57) 0.80 (0.02)
Switch 1'383.33 (306.83) 0.95 (0.09)
Shifting (SH)
Repeat 1'181.51 (245.18) 0.97 (0.08)
congruent 1'176.39 (249.79) 0.98 (0.03)
Shifting
Switch neutral 1'181.13 (246.60) 0.98 (0.03)
incongruent 1'215.83 (251.07) 0.98 (0.03)
SH Flanker
congruent 1'072.24 (234.69) 0.99 (0.02)
Repeat neutral 1'080.16 (242.77) 0.99 (0.02)
incongruent 1'132.30 (248.04) 0.99 (0.03)
Block Match Flanker
False 1'108.53 (186.25) 0.91 (0.10)
2-Back
True 911.32 (190.45) 0.95 (0.06)
no 1'059.88 (189.14) 0.95 (0.08)
congruent 1'144.80 (202.73) 0.94 (0.09)
False
neutral 1'106.65 (190.48) 0.93 (0.10)
N-Back
2-Back incongruent 1'133.33 (202.27) 0.94 (0.08)
Flanker no 843.02 (169.66) 0.97 (0.04)
congruent 912.05 (163.71) 0.97 (0.05)
True
neutral 859.21 (163.17) 0.96 (0.05)
incongruent 934.95 (165.75) 0.97 (0.05)
Cue Flanker
congruent 703.76 (93.69) 1.00 (0.00)
no neutral 684.50 (85.34) 1.00 (0.00)
incongruent 825.63 (103.92) 0.98 (0.05)
congruent 669.78 (97.38) 1.00 (0.00)
central neutral 660.10 (90.57) 1.00 (0.00)
incongruent 809.02 (102.68) 0.98 (0.04)
ANT
congruent 653.44 (92.16) 1.00 (0.00)
double neutral 652.69 (90.99) 1.00 (0.00)
incongruent 790.15 (105.19) 0.98 (0.04)
congruent 593.61 (93.37) 1.00 (0.00)
spatial neutral 593.61 (88.29) 1.00 (0.00)
incongruent 706.48 (113.98) 0.99 (0.03)
Note. RT = Reaction Time in ms; Pc = Proportion correct responses; LM = Less More; OE = Odd-Even
PS, WM, AND EF AS PREDICTORS OF G 48

Table 2
Descriptive statistics for the ERP latencies in the Executive Functioning tasks
Task Block Task Shifting Flanker N1 Lat P3 Lat
M (SD) M (SD)
Control LM 117.41 (24.27) 465.39 (89.93)
Control OE 118.01 (23.88) 467.83 (91.32)
Switch 120.43 (22.39) 473.87 (92.62)
Shifting (SH)
Repeat 120.59 (21.54) 466.81 (96.82)
congruent 114.84 (22.81) 492.48 (88.66)
Shifting
Switch neutral 114.16 (24.46) 497.88 (105.80)
incongruent 116.44 (23.67) 489.48 (82.99)
SH Flanker
congruent 112.72 (22.91) 487.65 (87.63)
Repeat neutral 114.52 (23.57) 480.70 (96.00)
incongruent 115.01 (22.13) 481.13 (85.33)
Block Match Flanker
False 126.10 (25.49) 505.05 (97.14)
2-Back
True 130.29 (25.88) 478.47 (101.20)
no 121.49 (28.32) 510.05 (97.96)
congruent 122.86 (25.26) 512.43 (101.48)
False
neutral 122.86 (28.14) 495.34 (91.48)
N-Back
2-Back incongruent 131.90 (27.12) 505.58 (109.58)
Flanker no 121.50 (24.79) 508.64 (84.91)
congruent 121.32 (26.31) 501.34 (85.81)
True
neutral 122.84 (24.23) 464.29 (76.57)
incongruent 131.78 (24.34) 469.60 (99.87)
Cue Flanker
congruent 144.05 (30.91) 431.58 (52.40)
no neutral 148.33 (28.00) 440.05 (56.18)
incongruent 150.61 (31.07) 448.16 (51.45)
congruent 140.80 (28.54) 425.77 (62.13)
central neutral 142.43 (29.97) 419.55 (55.21)
incongruent 149.54 (31.73) 436.70 (61.62)
ANT
congruent 135.93 (24.54) 416.18 (60.74)
double neutral 137.48 (20.83) 408.56 (51.77)
incongruent 136.59 (24.30) 425.86 (49.24)
congruent 127.67 (17.61) 422.12 (69.08)
spatial neutral 131.81 (17.70) 418.43 (70.76)
incongruent 130.47 (22.34) 431.10 (61.16)
Note . Lat = latency, LM = Less More, OE = Odd-Even;
PS, WM, AND EF AS PREDICTORS OF G 49

Table 3
Descriptive Statistics for the elementary cognitive tasks
Task Block Match MRT (SDRT ) MPc (SDPc)
False 673.90 (118.19) 0.99 (0.03)
PI
True 623.83 (105.12) 0.99 (0.02)
Posner
False 753.08 (141.04) 0.99 (0.02)
NI
True 693.93 (120.58) 0.98 (0.03)
False 885.29 (193.78) 0.99 (0.02)
S3
True 882.70 (202.77) 0.98 (0.03)
Sternberg
False 1051.46 (291.60) 0.98 (0.04)
S5
True 1017.01 (250.26) 0.97 (0.05)
Note. RT = Reaction time in ms; Pc = Proprotion correct; PI = Physical
Identity; NI = Name Identity; S3 = Set Size 3; S5 = Set Size 5.
PS, WM, AND EF AS PREDICTORS OF G 50

Figures

Figure 1. Trial Procedure of the three EF tasks for (A) Task Shifting with red (dark) as cue for
less/more decision and green (light) for odd/even decision (Switching Task), (B) Updating (N-
Back task), and (C) Inhibition (Attention Network Test, ANT). Presentation times are given
below the different screens in the trial procedure. In the Shifting task and the N-Back task, the
flanker stimuli as shown above were only presented in the flanker blocks. The other blocks in
these two tasks did not include flanker stimuli and only showed the central target stimulus.
PS, WM, AND EF AS PREDICTORS OF G 51

Figure 2. Grand Averages of the ERPs in the three executive function tasks across the
experimental blocks. More detailed differences between specific experimental functions were
omitted for readability. Time is displayed on the x-axis in milliseconds and the potential on the
y-axis in µV. N1 latency was determined at Fz, and P3 latency at Pz.
PS, WM, AND EF AS PREDICTORS OF G 52

Figure 3. Descriptive Plots displaying the effects of experimental manipulations on behavioral

RTs in the three EF tasks. The top panel (A) displays the effects in the Shifting task, the mid
panel (B) the effects in the N-Back task, and the bottom panel (C) the effects in the ANT.
PS, WM, AND EF AS PREDICTORS OF G 53

Figure 4. Path-diagrams of the Bi-factor models for behavioral RT in the three executive
function tasks. The top part (A) shows the model for the shifting task (Shift), the middle part (B)
shows the model for the N-Back task, and the lower part (C) the model for the Attention
Network Test (ANT). All loadings that are not explicitly stated were fixed to one and
unstandardized parameters are reported.
PS, WM, AND EF AS PREDICTORS OF G 54

Figure 5. Path-diagrams of the Bi-factor models for N1 latency in the three executive function
tasks. The top part (A) shows the model for the shifting task, the middle part (B) shows the
model for the N-Back task, and the lower part (C) the model for the ANT. All loadings that are
not explicitly stated were fixed to one and unstandardized parameters are reported.
PS, WM, AND EF AS PREDICTORS OF G 55

Figure 6. Path-diagrams of the Bi-factor models for P3 latency in the three executive function
tasks. The top part (A) shows the model for the shifting task, the middle part (B) shows the
model for the N-Back task, and the lower part (C) the model for the ANT. All loadings that are
not explicitly stated were fixed to one and unstandardized parameters are reported.
PS, WM, AND EF AS PREDICTORS OF G 56

Figure 7. Path-diagrams for SEM of behavioral reaction times (RTs) in three EF tasks with the
three covariates processing speed (PS), intelligence (Gf), and working memory capacity (WMC).
Gf indicators are: verbal (Verb), numerical (Num), and figural (Fig) score from the BIS.
Indicators for WMC are: proportion correct in a memory updating (MU), a sentence span (SS),
an operation span (OS), and a spatial short-term memory (SSTM) task. Processing Speed
indicators are: name identity (NI), and physical identity (PI) RTs from the Posner task (P), and
set size 3 (S3) and set size 5 (S5) RTs from the Sternberg task (S). The top part (A) shows a
correlational model, whereas the bottom part (B) shows a simplified model joining PS and EF
performance into one general processing speed factor and Gf and WMC into a factor for higher
cognitive abilities. Manipulation-specific factors of EF tasks are not depicted as all relationships
of these factors with other factors were fixed to zero. All loadings that are not explicitly stated
were fixed to one. Parameters are unstandardized except for correlations and differ all
significantly from zero.
PS, WM, AND EF AS PREDICTORS OF G 57

Figure 8. Path-diagrams for SEM of ERP latencies in three EF tasks with the three covariates
processing speed (PS), intelligence (Gf), and working memory capacity (WMC). Gf indicators
are: verbal (Verb), numerical (Num), and figural (Fig) score from the BIS. Indicators for WMC
are: proportion correct in a memory updating (MU), a sentence span (SS), an operation span
(OS), and a spatial short-term memory (SSTM) task. Processing Speed indicators are: name
identity (NI), and physical identity (PI) RTs from the Posner task (P), and set size 3 (S3) and set
size 5 (S5) RTs from the Sternberg task (S). The top part (A) shows the model for N1 latencies,
and the bottom part (B) shows the model for P3 latencies. Manipulation-specific factors of EF
tasks are not depicted as all relationships of these factors with other factors were fixed to zero.
All loadings that are not explicitly stated were fixed to one. Parameters are unstandardized
except for correlations and differ all significantly from zero.