Content uploaded by Morgan Lyphout-Spitz
Author content
All content in this area was uploaded by Morgan Lyphout-Spitz on Sep 19, 2023
Content may be subject to copyright.
PREPRINT 1
*Corresponding authors: Morgan Lyphout-Spitz and François Maquestiaux
E-mail: morgan.lyphout@gmail.com and francois.maquestiaux@univ-rouen.fr
Uncorking the Central Bottleneck: Even Novel Tasks
Can Be Performed Automatically
MANUSCRIPT IN PRESS AT Journal of Experimental Psychology: Human
Perception and Performance
Morgan Lyphout-Spitz 1, François Maquestiaux 2, Eric Ruthruff 3, & Steeven Chaloyard 1
1Université de Franche-Comté
2Université de Rouen Normandie
3University of New-Mexico
Can people perform two novel tasks in parallel? Available
evidence and prevailing theories overwhelmingly indicate that the
answer is no, due to stubborn capacity limitations in central stages
(e.g., a central bottleneck). Here we propose a new hypothesis,
which suggests otherwise: people are capable of fully parallel
central processing (i.e., bypassing the central bottleneck), yet often
fail to do so mainly due to preparation neglect. This preparation-
neglect hypothesis was evaluated in four dual-task experiments
pairing novel tasks (Task 1 and Task 2) using arbitrary stimulus-
response mappings. Experiment 1, using a classic psychological
refractory period (PRP) procedure, replicated the finding of dozens
of previous PRP studies: none of the participants bypassed the
bottleneck, instead exhibiting large dual-task interference on Task
2 (445 ms). In Experiment 2, the same dual-task PRP trials were
randomly intermixed with single-task trials on Task 2, to boost
preparation on that task. Here, nearly half the sample of
participants bypassed the central bottleneck, exhibiting small dual-
task interference on Task 2 (48 ms). Two additional experiments
showed that initial practice does not by itself enable bottleneck
bypassing, but boosting preparation of Task 2 (via intermixing
single-task trials of Task 2) does. We conclude that, when properly
prepared, people are capable of far more dual-task automaticity
than was previously believed.
Keywords
Attention, automaticity, central bottleneck, dual-task interference,
preparation
Public significance
• This study examined the
source of the processing
limitations that people
generally encounter when
attempting to perform two
tasks at the same time.
• We demonstrated that
novel tasks, when properly
prepared, can be performed
entirely automatically (i.e.,
without recruiting
attentional resources),
thereby bypassing the
processing limitations.
• This finding of dual-task
automaticity is striking
given that the tasks were
neither particularly easy nor
previously trained.
• The key enabler of dual-
task automaticity was not
practice but boosting
advanced preparation of the
to-be-performed tasks.
PREPRINT 2
Dual-task interference, manifested by slowed
reaction times (RTs) or increased mistakes, usually
arises when individuals attempt to simultaneously
perform two novel tasks (i.e., unpracticed tasks with
arbitrary stimulus-response [S-R] mappings).
Despite controversies, almost all prevailing accounts
agree that the problem lies within the central
processes that fall between sensory encoding and
motor execution, such as response selection (e.g.,
Pashler, 1984; Pashler & Johnston, 1989), memory
retrieval (e.g., Carrier & Pashler, 1995; but see
Green, Johnston, & Ruthruff, 2011), or mental
rotation (Ruthruff, Miller, & Lachman, 1995).
Specifically, the problem has been attributed to
capacity limitations in central processing stages
(Navon & Miller, 2002; Pashler, 1994; Tombu &
Jolicoeur, 2003), and/or crosstalk between central
task representations due to shared conceptual
similarities (Logan & Schulkind, 2000; Navon &
Miller, 1987) or due to shared working-memory
subsystems (Halvorson & Hazeltine, 2015, 2019;
Hazeltine et al., 2006; Maquestiaux et al., 2018).
In one dominant account, Pashler (1994)
hypothesized a stubborn cognitive limitation, called
a central bottleneck, that prevents central processing
(e.g., selecting the appropriate response to the
current stimulus) from simultaneously operating on
two tasks. At the other extreme, Meyer and Kieras
(1997a, 1997b) proposed that central limitations
reflect a strategic postponement of central stages.
However, even these authors and their collaborators
implied that central limitations are in force with
unpracticed tasks (the first few sessions), prior to
proceduralization (Schumacher et al., 2001). These
prevailing models agree that, without extensive
practice, dual-task processing is inefficient due to
constraints in central processes. The empirical data
appear to support this claim, as large dual-task costs
are nearly universally obtained with novel tasks (see,
e.g., Janczyk & Kunde, 2020; Koch, Poljac, Müller,
& Kiesel, 2018; Maquestiaux, 2012, 2017).
In the present article, we tested the neglected
possibility of fully parallel central processing with
novel tasks. Fully parallel central processing means
that while one is performing the central operations
for one task, all the central operations required by
another task can operate in parallel. In order words,
the tasks are automatic in that they can operate
“without recruiting attention” (for other meanings of
automaticity, such as uncontrolled or autonomous,
see Moors & De Houwer, 2006). Below, for the sake
of simplicity, we will use the shorthand bottleneck
bypassing.
Bottleneck bypassing has been reported in only a
few rare exceptions, such as highly practiced tasks
(e.g., thousands of trials; Maquestiaux et al., 2008,
2018; Ruthruff et al., 2006) or those involving
specialized systems (e.g., eye movements; Pashler et
al., 1993) or extremely easy tasks that might not
require any central processing (e.g., ideomotor
compatibility, Maquestiaux et al., 2020; semantic
compatibility, Lyphout-Spitz et al., 2022). But
bypassing is currently not believed to be possible for
novel tasks. We use the term novel tasks to refer to
tasks that (a) a participant has not been extensively
practiced and (b) do not benefit from associations
previously established between the stimuli and
responses (as in moving in the direction of an arrow).
A majority of dual-task studies have used novel
tasks.
Of course, numerous studies with novel tasks
have demonstrated partially automatic processing of
certain specific central operations such as visual
word recognition (Ruthruff et al., 2006), response
activation (Hommel, 1998; Janczyk et al., 2018), or
motoric aspects of central operations (Hartley et al.,
2015). But even these studies still reported
substantial dual-task costs indicative of central
interference (e.g., bottlenecks) and did not claim to
have observed bottleneck bypassing (in fact, they
often explicitly assume a central bottleneck).
Studies demonstrating bottleneck bypassing with
novel tasks are essentially non-existent.
As unlikely as bottleneck bypassing may appear
given previous work, a recent accumulation of hints
from studies with easy tasks (Lyphout-Spitz et al.,
2022; Maquestiaux et al., 2020; Maquestiaux &
Ruthruff, 2021) suggests otherwise. Concretely, we
hypothesize that dual-task interference often results
from preparation neglect, rather than a true capacity
limitation in central processing, and that central
bottlenecks can be bypassed merely by boosting task
preparation.
The Preparation-Neglect Hypothesis
The present preparation-neglect hypothesis was
inspired, albeit indirectly, by recent studies
demonstrating unexpectedly high amounts of
parallel central processing with very easy tasks
(Lyphout-Spitz et al., 2022; Maquestiaux et al.,
2020). We began to suspect that this parallel
processing was enabled by one uncommon aspect of
the methodology – intermixing single-task trials
within dual-task trials. These recent studies,
described below, relied on variants of the classic
psychological refractory period (PRP) procedure, in
which participants perform two tasks (Task 1 and
Task 2) as quickly as possible, with the Task-1
stimulus (S1) and Task-2 stimulus (S2) separated by
a stimulus onset asynchrony (SOA) that randomly
PREPRINT 3
varies from very short (e.g., 15 ms) to very long (e.g.,
1,500 ms). The fact that Task 1 always precedes
Task 2 might encourage participants to neglect Task
2. However, this neglect might be counteracted by
intermixing single-task trials of Task 2.
One such study investigated bottleneck
bypassing with ideomotor (IM)-compatible tasks
(i.e., pressing a left key when an arrow points left or
a right key when it points right; Maquestiaux et al.,
2020). This study reported negligible dual-task costs
consistent with fully parallel central processing. An
IM Task 2 could operate in parallel even with a non-
IM Task 1.
IM tasks are very easy due to the very high
conceptual overlap between the stimulus and the
response (e.g., pressing a left key when an arrow
points left). But task easiness could not by itself
explain bottleneck bypassing in Maquestiaux et al.
(2020), since previous PRP studies with IM tasks did
not report such evidence (e.g., Lien et al., 2002).
One salient methodological difference seemed the
most likely culprit: unlike Lien et al., Maquestiaux
et al. randomly intermixed single-task trials on both
Task 1 and Task 2 within the dual-task PRP trials.
The stated purpose of these mixed single-task trials
was to provide an optimal performance baseline for
assessing dual-task costs (see also Hazeltine et al.,
2006). We came to suspect, however, that it may
have inadvertently boosted task preparation, thereby
enabling bottleneck bypassing. If so, it is logically
possible that it would help enable bottleneck
bypassing with novel tasks as well (though this idea
has not been tested).
We later obtained an equally surprising amount
of bottleneck bypassing with highly semantically
compatible tasks such as saying “ping” when hearing
“pong”, and “pong” to “ping” (Lyphout-Spitz et al.,
2022), even though they were not IM compatible.
This led us to speculate that “what matters the most
is task preparation, specifically the ease with which
people can load the S-R mapping into working
memory” (p. 510). We further conjectured that
“intermixing single-task trials on Task 2 with dual-
task PRP trials may have further encouraged pre-
loading of Task-2 S-R mapping into working
memory” (p. 510). However, this study did not
actually manipulate the presence/absence of mixed
single-task trials on Task 2, so this speculation was
not actually tested.
A similar account may explain dual-task
automatization following multiple sessions of
single-task practice (Maquestiaux et al., 2008, 2018;
Ruthruff et al., 2006). Practicing an arbitrary task
may progressively transfer the burden of
representing the S-R mappings from working
memory into long-term memory, thus reducing
preparation neglect.
Another hint that preparation is crucial comes
from Maquestiaux and Ruthruff (2021) who
hypothesized that older adults tend to apply too
much attention. They reported a counterintuitive
finding: age differences in dual-task interference
were much larger when Task 2 was IM (i.e., easy)
than when it was non-IM (i.e., more difficult, due to
the use of arbitrary S-R mappings). According to
Maquestiaux and Ruthruff (2021), older adults
attended to Task 1 and prepared for it exclusively,
thus leaving Task 2 unprepared (for a similar
account, see Hartley & Little, 1999). The lack of
preparation meant that the very easy Task 2, which
could have been performed automatically, was not.
If preparation neglect can harm automaticity in older
adults, then it might sometimes do the same for
younger adults.
Meyer and Kieras (1997a, 1997b) also
anticipated the possibility of parallel central
processing in that their executive process-interactive
control (EPIC) architecture appears to impose no
structural central limitations (see also Meyer &
Kieras, 1999). However, they did not emphasize
specifically the importance of preparation neglect
and did not actually demonstrate bottleneck
bypassing with novel tasks. Later work by Meyer
and colleagues appeared to assume that bottleneck
bypassing was possible only after practice (i.e., after
proceduralization; see also Hazeltine et al., 2002).
This can be seen in Schumacher et al.’s (2001)
important paper entitled “Virtually perfect time
sharing in dual-task performance: Uncorking the
central cognitive bottleneck”, in which they had
participants perform 6 sessions of practice and did
not even examine dual-task performance during the
first session.
The present work also builds on Hazeltine and
Halvorson (2019), who hypothesized that
minimization of dual-task costs is possible when
each task relies on a distinct working memory
subsystem (e.g., one task using verbal working
memory and one using visual working memory).
However, their empirical work focused not on novel
(unpracticed and arbitrary) tasks as conceived in the
present work, but rather on what they called
paramotor tasks, in which the stimulus shares
perceptual features with the response (e.g., seeing a
hand and making a corresponding manual response).
The Apparent Ubiquity of Bottlenecking
There is now a very large literature on dual-task
performance with novel tasks. Given this literature,
bottleneck bypassing seems highly unlikely. Also,
PREPRINT 4
prevailing accounts of dual-task interference (e.g.,
Pashler, 1994) do not predict bypassing. In fact,
recent reviews of multitasking did not mention the
possibility of bypassing with novel tasks (see, e.g.,
Fischer & Plessow, 2015; for a review across several
multitasking paradigms, see Koch et al., 2018). A
few studies have reported bypassing with special
tasks that are not novel, for example those that have
uniquely high semantic compatibility between
stimulus and response (Halvorson & Hazeltine,
2015; Lyphout-Spitz et al., 2022). With novel tasks,
there is also evidence that the Task-2 response code
can be activated to some degree during the central
bottleneck (i.e., the backward crosstalk effect;
Hommel, 1998; Janczyk, 2016; Janczyk et al., 2018).
However, even these studies report large dual-task
costs consistent with massive central interference
(e.g., a central bottleneck). To our knowledge, there
are no claims of bottleneck bypassing with novel
tasks.
Despite seemingly ubiquitous dual-task
interference with novel tasks, we remained
optimistic about the possibility of bypassing for two
main reasons. First, no study ever reported
experiments that were designed to directly boost
preparation of Task 2; so, at the very least, there is
no evidence that boosting preparation will not enable
bottleneck bypassing. A few dual-task studies have
included mixed single-task trials, however they
focused on easy tasks and/or extensive practice
(Hazeltine et al., 2002; Huestegge & Strobach, 2021;
Schumacher et al., 2001; Strobach & Schubert,
2017) and did so with the intention of providing a
performance baseline for each of the two tasks
performed alone, rather than the intention of
boosting task preparation. Studies with novel tasks
have rarely, if ever, done anything to boost task
preparation and this alone might explain why they
always found robust dual-task costs.
Another reason to remain optimistic is that
almost all studies have reported group data only. So
long as the group as a whole shows substantial dual-
task costs, one might not even notice if half the
sample had bypassed the central bottleneck. There
has rarely been any close examination of possible
individual differences in the ability to bypass the
central bottleneck with novel tasks. A near
exception is Brüning et al. (2022) who reported more
automatic activation of Task 2 during Task-1 central
processing (i.e., a larger backward crosstalk effect)
for the subset of participants who tended to prefer a
parallel processing mode, as assessed in a task-
switching paradigm. However, even this study did
not argue that any individual had bypassed the
central bottleneck. In sharp contrast, the literature
on practice effects has routinely reported striking
qualitative differences between individuals
(Maquestiaux et al., 2008, 2010, 2013, 2018;
Ruthruff et al., 2006; Schumacher et al., 2001;
Watson & Strayer, 2010). For instance,
Maquestiaux et al. (2018) argued that their sample of
highly-practiced individuals included a mixture of
bottleneckers (i.e., performing central stages
serially) and bypassers (i.e., fully performing the
central stages of the two tasks in parallel), producing
vastly different amounts of dual-task costs. Given
these striking individual differences following
practice, it certainly makes sense to also look for
them prior to practice.
THE CURRENT STUDY
The aim of the present study was to examine
whether boosting preparation on Task 2 (i.e., the task
that suffers the most in dual-task PRP experiments)
enables people to bypass the central bottleneck that
often constrains dual-task performance. Even
though bypassing without practice is not predicted
by prevailing accounts of dual-task interference
(e.g., Pashler, 1994), this prediction follows from the
preparation-neglect hypothesis (Lyphout-Spitz et al.,
2022; Maquestiaux et al., 2020; Maquestiaux &
Ruthruff, 2021), shown in Figure 1. According to
this new hypothesis, insufficient advance
preparation causes serial processing of the Task-1
and Task-2 central stages (e.g., response selection,
represented by stage B), also known as bottlenecking
(see Figure 1A). The corollary is that boosting task
preparation should enable fully parallel central
processing of Task-1 and Task-2 central stages (i.e.,
bottleneck bypassing; see Figure 1B). Note that,
even without preparation boosting, stimulus
identification (stage A) and response initiation (stage
C) of one task are assumed to proceed in parallel
with any stage of the other task (Pashler, 1984).
To evaluate the preparation-neglect hypothesis,
we looked for evidence of bottleneck bypassing
using two different dual-task procedures: a classic
PRP procedure (Experiment 1 and Experiment 3) vs.
a PRP procedure modified to boost preparation
(Experiment 2 and Experiment 4). The classic PRP
procedure, used in most dual-task experiments,
contained only dual-task PRP trials with the Task-1
stimulus always preceding the Task-2 stimulus by a
variable SOA. For the modified PRP procedure, we
boosted preparation by intermixing into the dual-
task PRP trials some single-task trials on the task
most likely to be neglected (Task 2). We reasoned
that inclusion of single-task trials on Task 2 would
encourage participants to expect Task 2 and prepare
to process it without waiting for Task 1.
PREPRINT 5
Probing for Bottleneck Bypassing
Following previous studies (e.g., Lyphout-Spitz
et al., 2022), we probed for bypassing vs.
bottlenecking by relying on a Task 1 that produced
long RTs (> 500 ms) due to a long central stage.
When pairing this slow Task 1 with Task 2, the
candidate models make highly distinctive
predictions regarding the distributions of inter-
response intervals (IRIs), as well as other behavioral
indicators (i.e., the amount of dual-task interference
on Task 2 also referred to as PRP effect for short, the
correlation between Task-1 RTs and Task-2 RTs). A
slow Task 1, along with the use of multiple SOAs,
sidesteps the problem of a “latent” bottleneck, which
occurs when Task-1 and Task-2 central stages are so
short (e.g., following extensive practice) and/or
misaligned that they are rarely demanded at the same
time (Anderson et al., 2005; Levy & Pashler, 2001;
Ruthruff et al., 2003, 2009).
Most dual-task studies analyze data only for the
entire group. We believe that this is a mistake
because marked individual differences have been
documented in many previous dual-task studies
(e.g., Maquestiaux et al., 2018; Schumacher et al.,
2001; Watson & Strayer, 2010). Therefore, below,
we specifically set out to look for possible individual
differences.
EXPERIMENT 1: CLASSIC PRP
PARADIGM
This experiment aimed to replicate the classic
finding from numerous PRP studies: the presence of
Figure 1
The Preparation-Neglect Hypothesis of Dual-Task Interference
Note. A: Bottlenecking without preparation boosting: central processing is
serial, as in the classic psychological refractory period paradigm. B:
Bottleneck bypassing with preparation boosting: central processing is parallel.
The clouds represent task preparation prior to stimulus onset. A = stimulus
identification, B = response selection, C = response initiation, S = stimulus, R
= response, SOA = stimulus onset asynchrony.
PREPRINT 6
a central bottleneck constraining dual-task
performance. The main purpose is to provide a
baseline against which to compare the results of
Experiments 1 and 4, which will attempt to boost
Task-2 preparation. Given previous studies (e.g.,
Pashler, 1984, 1994), we expect this classic PRP
paradigm to support a central bottlenecking.
However, previous PRP studies of this kind have
rarely considered individual differences, so it is
unclear what percent of participants will be subject
to a central bottleneck.
Twenty-four participants completed one session
of a typical dual-task PRP experiment, lasting under
an hour. Task 1 was the traditional tone
identification task (low-pitched tone vs. high-
pitched tone) used in many previous PRP studies
supporting processing bottlenecks (e.g., De Jong,
1993; Pashler & Johnston, 1989; Ulrich et al., 2006).
Participants said “bas” ([bɑ]; French for low) to a
low-pitched tone or “haut” ([‘o]; French for high) to
a high-pitched tone. Task 2 was a shape
identification task (triangle vs. circle) with an
arbitrary S-R mapping. Because these tasks rely on
distinct sensory and motor modalities (Task 1 is
auditory-vocal and Task 2 is visual-manual), they
minimize peripheral conflicts (Hazeltine et al., 2006;
Maquestiaux et al., 2018; Ruthruff et al., 2001; Van
Selst et al., 1999).
Very few PRP experiments have required
participants to respond to Task 1 before Task 2 (see,
e.g., Mittelstädt et al., 2022; Schumacher et al.,
2001, Experiment 2). We did not do so because this
could create an artificial central bottleneck. Instead,
we simply instructed our participants to “respond as
quickly as possible to the tone and the shape.
Emphasize response speed to the tone. Make as few
errors as possible” (in French: “Répondre le plus vite
possible au son et à la forme. Accentuer la vitesse
de réponse au son. Faire le moins possible
d’erreurs”). These instructions are similar to those
from several previous PRP studies (see, e.g., Pashler
& Johnston, 1989).
Unlike most previous studies that focused on
group-level data only, we specifically considered
individual differences. First, we estimated how
frequently each participant bypassed the central
bottleneck. Second, we classified each individual as
a bypasser or a bottlenecker based on whether they
bypassed the bottleneck on more or less than half of
the dual-task trials (see details in the methods
below). This binary classification is an
oversimplification (because many participants
appeared to use a mixture of processing modes) but
it makes it easy to visualize data patterns (e.g., the
IRI distribution for all bypassers or all
bottleneckers).
Our main tests of bottlenecking vs. bypassing
relied upon an examination of entire distributions of
IRIs: the time that elapsed between the Task-1
response and the Task-2 response. We have
previously shown that it is possible to compare the
observed IRI distributions to that predicted by
bottleneck bypassing (Lyphout-Spitz et al., 2022;
Maquestiaux et al., 2020). As will be seen,
bottleneck bypassing and bottlenecking predict
highly distinct IRI distributions. Bottleneck
bypassing predicts a high propensity to emit the
Task-2 response first (i.e., mostly negative IRIs),
whereas bottlenecking predicts a high propensity to
emit the Task-1 response first (i.e., mostly positive
IRIs).
Given that there was no attempt to boost
preparation in Experiment 1 and the classic PRP
paradigm favors Task 1 over Task 2 (because Task 1
always comes first), the preparation neglect
hypothesis predicts little or no bypassing. As a
consequence, we expected the observed IRI
distributions to match those predicted by
bottlenecking.
Method
Figure 2 shows the S-R mappings (panel A) and the
sequence of events (panel B) for a dual-task PRP
trial in Experiment 1 as well as in all subsequent
experiments.
Participants. Twenty-four volunteers (M = 20.5
years old, SD = 1.9 years; 21 women) took part in
this experiment. The sample size of 24 was fixed in
advance and chosen so that we would have as much
statistical power as the most comparable studies (i.e.,
N = 24 in both Lyphout-Spitz, 2022, as well as in
Maquestiaux et al., 2020). The methods and
procedures of this experiment and subsequent
experiments, both involving human participants,
strictly followed the standards of the Ethics
Committee of the Université de Franche-Comté, as
well as the 1964 Helsinki declaration and its later
amendments.
Apparatus and Stimuli. The experiment (E-
Prime 2.0) was conducted on a laptop computer
using an AZERTY keyboard and a PST Serial-
Response Box. Participants listened to the auditory
tones through headphones and spoke the answer into
a microphone connected to the response box. Voice
onset was detected automatically by the response
box but the response choice (“bas” vs. “haut”) was
manually entered by the experimenter.
PREPRINT 7
The Task-1 tone was presented for 250 ms with
either a low pitch (400 Hz) or a high pitch (1,800
Hz). The Task-2 stimulus was a filled triangle (3 cm
in width and 2 cm in height) or a filled circle (2.2 cm
in diameter), displayed in black in the screen center
against a white background.
Design and Procedure. For Task 1, participants
were asked to respond to the high-pitched tone by
saying “haut” (French for “high”) and to the low-
pitched tone by saying “bas” (French for “low”). For
Task 2, participants responded to the triangle by
pressing the E key with their left index finger and to
the circle by pressing the P key with their right index
finger.
Participants first performed 6 familiarization
trials on Task 1, then 6 familiarization trials on Task
2, followed by 24 familiarization trials on both Task
1 and Task 2 (dual-task PRP trials). They then
performed the 384 experimental trials (dual-task
PRP trials). Each of the 16 distinct trial types (2
Task-1 tones X 2 Task-2 shapes X 4 SOAs [15, 65,
250, and 1,500 ms]) was repeated 24 times, in a
random order. The 384 experimental trials were
broken into 12 blocks of 32 trials each. Participants
were given a 2-minute break between blocks, during
which time they viewed performance feedback on
the preceding block (Task-1 speed and accuracy plus
Task-2 accuracy). To encourage participants to
strive to continually improve their performance, we
asked them to enter their scores into a matrix on a
sheet of paper. Participants received typical PRP
instructions: respond as quickly and accurately as
possible to each task while emphasizing the speed of
Task-1 responses. The instructions did not say
anything about response order or response grouping;
the computer accepted every possible response order
Figure 2
Stimulus-Response Mappings and Dual-Task PRP Trial in Experiments 1-4
Note. Panel A shows the stimulus-response mappings for Task 1 and Task 2.
The finger highlighted in black is the one used to respond to that Task-2
stimulus. Panel B shows the sequence of events in a dual-task psychological
refractory period (PRP) trial. The stimulus onset asynchrony (SOA) varied
randomly: 15, 65, 250, or 1,500 ms. Even though the Task-1 stimulus (S1)
was always followed by the Task-2 stimulus (S2), participants were allowed
to respond in either order. In the depicted example, the response order was
reversed: the Task-2 response (R2) was emitted before the Task-1 response
(R1).
PREPRINT 8
(Task-1 response before Task-2 response, Task-2
response before Task-1 response).1
Each dual-task PRP trial began with a fixation
sign (+) displayed for 500 ms in the screen center.
Then the Task-1 tone sounded. After a randomly
selected SOA (15, 65, 250, or 1,500 ms), the Task-2
shape appeared and remained until response or 2,500
ms had elapsed. The computer then displayed two
consecutive 300-ms messages indicating whether
the Task-1 and Task-2 responses were correct or
incorrect. If participants failed to respond to both
tasks within the 2,500 ms timeout period, then an
additional 200-ms message noted that fact. The next
trial began 800 ms later.
Simulating Bottleneck Bypassing and
Bottlenecking. To simulate the distributions of IRIs
for the candidate models (bottleneck bypassing and
bottlenecking) for each participant, we used the
long-SOA (i.e., 1,500 ms) dual-task PRP trials as a
baseline. On these trials there is rarely any
processing overlap between Task 1 and Task 2, so
they provide a sort of “single-task” baseline without
the possibility of dual-task interference. Concretely,
for each participant, we paired every such response
to Task 1 (96) with every such response to Task 2
(96), so as to create 9,216 simulated dual-task trials.
Bypassing. If participants bypass the bottleneck,
with no central interference, then the simulated dual-
task RTs are simply equal to the original “single-
task” RTs. We then calculate the entire distribution
of IRIs, which are particularly diagnostic of the
bypassing processing mode, as follows:
IRI = SOA + RT2 – RT1
After eliminating outliers (i.e., trials for which
RT was below 100 ms or above 2,500 ms on Task 1
or Task 2) and errors (i.e., error trials), just as for the
real data, we computed the entire IRI distribution
predicted by bypassing. Note that this predicted IRI
distribution has no free parameters, so it is a true
prediction and not a post-hoc fitting exercise (for a
critique of data fitting to test theories with free
parameters, see Roberts & Pashler, 2000).
Bottlenecking. Our simulation of bottlenecking
was based on the finding that, according to a central
bottleneck, the predicted dual-task interference on
Task 2 can be expressed as RT1 – SOA – 1C – 2A
(Ruthruff et al., 2001; Van Selst et al., 1999).
Applying this at the level of individual trials, we
estimated the bottleneck delay (represented by the
horizontal dashed line on Figure 1A) as RT1 – SOA
– 1C – 2A. Van Selst et al., who also used an
auditory-vocal Task-1 and a visual-manual Task 2,
estimated that 1C + 2A took 257 ms. Following this
finding and assuming that 1C (initiating a vocal
response) takes longer than 2A (perceiving a shape),
we set the duration of 1C at 150 ms and the duration
of 2A at 100 ms. Similar to Lien et al. (2005), we
simply forced the coefficient of variation (mean/SD)
to be .2, yielding standard deviations of 30 ms and
20 ms, respectively. We used the normal
distribution, but with the restriction that these stage
times were not allowed to go below 0 ms.
We then added this simulated bottleneck delay to
the RT2 for that simulated trial to obtain the
predicted dual-task RT2. Note that the bottleneck
delay can be zero, for example at long SOAs, in
which case the bottleneck can be said to be latent
(i.e., stage 1B finished before stage 2B is demanded;
see Ruthruff et al., 2003). Although this model
could be considered to have several free parameters,
we did not use them to fit the data. Instead, we
simulated only once, with the parameters above,
specified based only on previous studies.
Initial Classification of Individuals as Bypassers
Vs. Bottleneckers. Bottleneck bypassing predicts
that R2 will frequently be emitted before R1 at short
SOAs, producing a high rate of response reversals
(negative IRIs), especially when baseline Task 2 is
faster than Task 1 (as was the case here).
Bottlenecking, in contrast, predicts that R1 will be
emitted before R2, producing few response reversals
(i.e., mostly positive IRIs). It has also been
established that some participants have the tendency
to synchronize their responses, known as response
grouping, which tends to produce IRIs close to zero
(within +/-50 or +/-100 ms; Ulrich & Miller, 2008).
Note that negative IRIs of less than -100 ms will
rarely ever occur while bottlenecking and/or
response grouping, but routinely occur while
bottleneck bypassing. Thus they are the signature of
bottleneck bypassing.
We quantified each individual’s degree of
bypassing. Specifically, we calculated how often
that individual produced IRIs more negative than -
100 ms as a percentage of that expected from the
bypassing simulation. For the sake of robustness, we
averaged these results across the two shortest SOAs
1 To our knowledge, this precaution is rarely mentioned in the
dual-task literature. Some dual-task procedures even prevented
or discouraged response reversals (Schumacher et al., 1999;
2001). For instance, Schumacher et al. (1999) indicated that “on
dual-task blocks, participants had to make their Task 1 response
first or else both responses were considered to be incorrect” (p.
797). Similarly, Schumacher et al. (2001, Experiment 2)
instructed the participants that “their response for it [an auditory-
vocal task] should always occur before responses for the
secondary VM [visual-manual] task” (p. 104).
PREPRINT 9
(15 and 65 ms); note that long SOAs are not
diagnostic because bottlenecking and bypassing tend
to predict similar IRIs. Even though the percentage
of bypassing is a continuous variable, we then
classified participants with a percentage greater than
50% as “primarily bypassers” and the others as
“primarily bottleneckers.”
A follow-up analysis was then performed to
determine whether the identified bypassers were
truly bypassing the central bottleneck and not simply
performing the central operations of the tasks
serially, but starting with the second task (i.e.,
reverse bottlenecking). The latter would also
produce negative IRIs, but with much more extreme
negative values. This analysis involved comparing
the observed IRI distribution to the distributions
derived from (a) simulations of bypassing and (b)
simulations of reverse bottlenecking2, and then
choosing the one with the lowest chi-squared value.
In the present study, only one participant’s IRI data
(cf. Experiment 2) fit the reverse bottlenecking
distribution better than the bypassing distribution.
This individual was excluded from subsequent
subgroup analyses. Note that this individual could
not simply be added to the bottlenecker subgroup,
because the predictions are nearly the opposite for
bottlenecking and reverse-order bottlenecking.
Confirmation of Bottleneck Bypassing vs.
Bottlenecking. After the initial classification of each
individual as described above, we then compared
each group’s data against classic indicators of
bypassing vs. bottlenecking. Following previous
studies (Lyphout-Spitz et al., 2022; Maquestiaux et
al., 2020), we applied multiple indicators: entire IRI
distributions, the amount of dual-task interference on
Task 2, and the correlation between RT1 and RT2.
Data Analysis. ANOVAs were Greenhouse-
Geisser corrected in case the assumption of
sphericity was violated. Following Lakens (2013),
effect size is reported as Cohen’s dz for within-
subject t tests and Cohen’s ds for between-subject t
tests. Post hoc analyses relied on the Bonferroni
procedure.
Transparency and Openness. The study materials
and raw data of all experiments are openly available
on the Open Science Framework
(https://osf.io/n4jhe/) and on a personal website
(https://sites.google.com/site/frmaquestiaux/home/d
ata-sharing). The Excel files with which we
performed the simulations are available upon
request.
Results
We excluded from analysis trials for which RT
on either task was below 100 ms or above 2,500 ms
(dual-task PRP: 2.44%). In addition, error trials
were excluded from RT analyses (dual-task PRP:
3.60%).
Initial Classification. Figure 3 shows that none
of the 24 participants in Experiment 1 was able to
reach the 50% cutoff of bypassing, with a mean
bypassing percentage of only 7.3% (ranging from
0.0% to 37.0%). Therefore, all 24 were classified as
“bottleneckers”. Note that some of these
bottleneckers might have bypassed the central
bottleneck on some trials, so this classification
should be interpreted as “primarily bottlenecking.”
Confirming Indicators for the Bottlenecker
Group. Several indicators of bypassing vs.
bottlenecking were then applied to the resulting
group data (i.e., the 24 bottleneckers, as no
participant was classified as primarily a bypasser).
Observed vs. Simulated IRI Distributions.
Figure 4 shows, for each SOA, the observed IRI
distribution (using a bin width of 50 ms) along with
the predicted IRI distributions based on
bottlenecking and bypassing (for details, see the
Methods section). When task overlap was high (i.e.,
at the two short SOAs), the observed distributions
were consistently quite well fitted by the
distributions predicted by bottlenecking. The mean
IRIs did not differ statistically either at the 15-ms
SOA (219 ms vs. 223 ms), t(23) = -0.19, p = .854, dz
= -0.04, or at the 65-ms SOA (228 ms vs. 224 ms),
t(23) = 0.17, p = .870, dz = 0.03. At the 250-ms SOA,
however, the observed distribution was shifted right
relative the simulated distribution of bottlenecking
(293 ms vs. 243 ms), t(23) = 2.80, p = .010, dz = 0.57.
In contrast, the simulated distributions predicted
from bottleneck bypassing provided very poor fits.
At the 15-ms SOA, the difference between the
observed and predicted means (219 ms vs. -191 ms)
was large, t(23) = 13.43, p < .001, dz = 2.74. The
difference between the observed and predicted
means was also significant at the 65-ms SOA (226
ms vs. -141 ms), t(23) = 12.11, p < .001, dz = 2.47.
This large difference can still be observed at the 250-
ms SOA (293 ms vs. 44 ms), t(23) = 9.82, p < .001,
dz = 2.00.
2 Our simulation of reverse bottlenecking followed the exact
same steps as those used to simulate bottlenecking. We
estimated the bottleneck delay on Task 1 as RT2 – SOA – 2C –
1A. We set the durations of 1A to 150 ms (SD = 30 ms) and the
duration of 2C to 100 ms (SD = 20 ms). For each participant, the
simulated IRI distributions were divided into five equal bins,
then we compared the expected and observed frequencies in each
bin.
PREPRINT 10
Amount of Dual-Task Interference on Task 1 and
Task 2. Figure 5 shows the mean RTs on Task 1 and
Task 2 as a function of SOA (see also Table 1 for
descriptive statistics). The amounts of dual-task
interference, in the case of Task 2 referred to as PRP
effect, were calculated as the RT difference between
the shortest SOA and the longest SOA. In this
experiment, RT2 sharply increased from the longest
(470 ms) to the shortest SOA (915 ms), F(1.262,
29.037) = 183.13, p < .001, ηp2 = 0.89, resulting in a
PRP effect of 445 ms. Meanwhile, mean RT1
remained relatively constant across SOAs, with RT1
being slightly longer at the two shortest SOA (an
average of 710 ms) compared to the longest SOA (M
= 679 ms), F(2.220, 51.053) = 5.78, p = .004, ηp2 =
0.20. The amount of dual-task interference on Task
1 was small (only 32 ms). This pattern – a very large
PRP effect of 445 ms on Task 2 with modest dual-
task interference on Task 1 – is plainly consistent
with a central bottleneck and classic PRP results
(e.g., Pashler, 1984).
Error Rates. Mean Task-2 error rate was 2.07%
and was not significantly influenced by SOA, F(3,
69) = 2.47, p = .069, ηp2 = 0.10. Mean Task-1 error
rate was 1.61% and was also not significantly
influenced by SOA, F(3, 69) = 2.26, p = .089, ηp2 =
0.09.
RT1:RT2 Correlations. Bottleneck bypassing
predicts weak RT1:RT2 correlations at all SOAs
because random variation in Task-1 processing is not
inherited by Task-2 processing (the tasks are
performed independently). In contrast, bottlenecking
predicts stronger correlations at short SOAs (where
a bottleneck constrains dual-task processing) than at
the long SOA (i.e., where Task 1 is finished before
S2 even occurs), because random variation in Task-
1 processing (in particular in stages 1A and 1B) is
inherited by Task 2 after the bottleneck delay (see
Figure 3
Percentage of dual-task psychological refractory period (PRP) trials consistent with
bottleneck bypassing for each individual participant in Experiments 1-4
Note. The horizontal dashed line represents the percentage cutoff (50%) for
classifying participants as primarily bypassers versus primarily bottleneckers.
PREPRINT 11
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 15
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 65
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 250
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 1500
Observed
Predicted-bypassing
Predicted-bottlenecking
Figure 4
Observed vs. simulated distributions of inter-response intervals (IRIs) at each stimulus onset
asynchrony (SOA) for Experiment 1 (classic PRP)
PREPRINT 12
ting
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 15
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 65
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 250
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 1500
Observed
Predicted-bypassing
Predicted-bottlenecking
Figure 4
Observed vs. simulated distributions of inter-response intervals (IRIs) at each
stimulus onset asynchrony (SOA) for Experiment 1 (classic PRP)
Figure 5
Reaction times on the auditory-vocal (AV) Task 1 and the visual-manual (VM) Task 2 at each
stimulus onset asynchrony (SOA) for Experiment 1
300
400
500
600
700
800
900
1000
1100
1200
1300
0 500 1000 1500
RT (ms)
SOA (ms)
AV Task 1
VM Task 2
Table 1
Mean Reaction Time (RT), Error Rate (ER), and PRP effect as a Function of Stimulus Onset
Asynchrony (SOA) in Experiments 1-4
PREPRINT 13
Figure 1B).3 To evaluate the predictions of
competing models (bypassing vs. bottlenecking), we
calculated the correlations between RT1 and RT2 at
each SOA for each participant, then carried out an
ANOVA. Consistent with bottlenecking, the
coefficient of correlation was moderate overall (.61)
and the main effect of SOA was significant, F(2.160,
49.678) = 67.83, p < .001, ηp2 = 0.75. A planned
comparison showed that correlations were much
stronger at the three shortest SOAs (average of .71)
than at the longest SOA (M = .33), t(69) = 14.17, p
< .001.
Discussion
In this classic PRP experiment, we looked for
evidence of bypassing but did not find any
participant who did so consistently. The percentage
of dual-task PRP trials consistent with bypassing
was below 50% for every individual and very low
overall (only 7.3%; see Figure 3). Additional
analyses at the group level confirmed the presence
of a central bottleneck: there was a very high overlap
at short SOAs between the observed IRI distribution
and that predicted by bottlenecking (see Figure 4),
the dual-task interference on Task 2 was very large
(i.e., 445 ms), and the RT1:RT2 correlations were
strong at short SOAs (an average of .71).
In short, the results of Experiment 1 replicated
the ubiquitous phenomenon of large dual-task costs,
as found in dozens of previous PRP studies with
novel tasks. We also confirmed that, in a classic PRP
paradigm with only dual-task trials, participants
rarely processed the central operations of two tasks
in parallel (see Pashler, 1984; Pashler, 1994; Pashler
& Johnston, 1989).
EXPERIMENT 2: PREPARATION
BOOSTING
According to the preparation-neglect hypothesis,
bottlenecking bypassing (i.e., fully parallel central
processing) without practice is possible in PRP
conditions that boost task preparation. To evaluate
this novel prediction, we used the same PRP
procedure as in Experiment 1 but modified it to boost
preparation on Task 2 (i.e., the task that suffered the
most in Experiment 1 and, more broadly, in virtually
all previous PRP studies). To do so, single-task trials
were randomly intermixed with dual-task PRP trials.
We previously found that intermixing single-task
trials within dual-task trials led to unusually high
rates of bottleneck bypassing (Lyphout-Spitz et al.,
2022; Maquestiaux et al., 2020). However, that was
with tasks that were uniquely easy due to unusually
high S-R compatibility (hearing “ping” and saying
“pong”). So, it is unclear whether bypassing would
be found here, with more difficult tasks with
arbitrary S-R mappings. Also, note that our previous
studies did not have a condition without intermixed
single-task trials. So, although this technique for
preparation boosting seemed promising, it is
impossible to say whether it actually increases
bypassing, or whether the bypassing was merely due
to the extreme ease of the tasks (perhaps not even
requiring any central operations).
Here we used the exact same ratios of mixed
single-task trials and dual-task trials as in our
previous studies (Lyphout-Spitz et al., 2022;
Maquestiaux et al., 2020): 96/192. However, there
was one major difference. Whereas our previous
studies included 96 mixed single-task trials divided
evenly between Task 1 and Task 2, here we used 96
mixed single-task trials on Task 2 only. The reason
is that those studies mixed single-task trials with the
sole purpose of providing an ideal baseline for
measuring dual-task costs, so it was important to
include both tasks. Here the sole purpose is to boost
preparation and enable bypassing. According to the
preparation-neglect hypothesis, boosting preparation
on Task 2 is the key to enabling bottleneck
bypassing.
In Experiment 1 (classic PRP), 24 participants
performed 384 classic dual-task PRP trials. But, in
Experiment 2 (preparation boosting), 24 new
participants first performed 96 single-task trials on
Task 2, and then 192 dual-task PRP trials randomly
intermixed with 96 single-task trials on Task 2.
Although participants in Experiment 2 received only
half as many PRP trials (i.e., 192 vs 384), the total
number of trials on Task 2 was equated between
Experiments 1 and 2 (i.e., 384). By holding constant
the amount of Task-2 practice, we ensure that the
emergence of bypassing cannot be attributed simply
to an increase Task-2 skill. Participants were asked
to respond as quickly and accurately as possible to
Task 2 when performing the 96 single-task trials.
They then received typical PRP instructions when
performing PRP trials intermixed with single-task
trials on Task 2.
We again first classified each individual as
primarily a bypasser or primarily a bottlenecker. We
then applied confirming indicators – IRI
distributions, dual-task interference on Task 2, and
RT1:RT2 correlations – separately for the bypasser
group and the bottlenecker group.
3 Following Navarro and Foxcroft (2022), we will interpret
correlations up to .2 as negligible, between .2 and .4 as weak,
between .4 and .7 as moderate, and above .7 as strong.
PREPRINT 14
Method
Except where noted, the method was identical to
that of Experiment 1.
Participants. A fresh sample of 24 volunteers (M
= 19.6 years old, SD = 1.7 years; 20 women)
participated.
Design and procedure. As in Experiment 1,
participants first performed familiarization trials on
each of the two tasks separately: 6 trials on Task 1
and 6 trials on Task 2. They then performed 96
single-task trials on Task 2, broken into 3 blocks of
32 trials each. They then performed 24 dual-task
familiarization trials (as in Experiment 1). Finally,
they performed 288 experimental trials consisting of
a random mixture of 192 dual-task PRP trials and 96
single Task-2 trials. The 288 experimental trials
were broken into 9 blocks of 32 trials each. As in
Experiment 1, participants were given typical PRP
instructions and received no explicit instructions
regarding response order or response grouping.
The event sequence for single-task trials on Task
2 was identical to that of the dual-task PRP trials,
except there was no Task-1 tone and no Task-1
feedback.
Simulating Bottleneck Bypassing and
Bottlenecking. The simulations of the candidate
models were identical to those carried out in
Experiment 1. The only difference is that there were
fewer simulated trials than in Experiment 1 (2,304
instead of 9,216) because there were half as many
long-SOA (1500 ms) dual-task trials.
Results
We removed trials for which RT on either task
was below 100 ms or above 2,500 ms (dual-task
PRP: 2.80%; mixed single-task: 0.13%; single-task:
0.17%). In addition, error trials were excluded from
RT analyses (dual-task PRP: 5.54%; mixed single-
task: 1.48%; single-task: 2.91%).
Initial Classification. Figure 3 shows that
nearly half the sample (i.e., 9 participants out of 23)
was able to go beyond the 50% cutoff, with on
average 81.1% of dual-task PRP trials consistent
with bottleneck bypassing (ranging from 50.1% to
100%). In contrast, the other participants that fell
below the 50% cutoff were classified as primarily
bottleneckers (M = 21.9%; range: 2.5% to 46.3%).
Note that one participant initially classified as a
bypasser was in fact a reverse bottlenecker, as
revealed by a follow-up analysis (for more details,
see the subsection “Initial Classification of
Individuals as Bypassers Vs. Bottleneckers” of
Experiment 1). As a consequence, this individual
was excluded from subsequent analyses.
Confirming Indicators for the Bypasser
Group. Our main focus is on the existence of
bypassers. We applied the same indicators as before
to the group of participants initially classified as
bypassers (n = 9).
Observed vs. Simulated IRI Distributions. Figure
6 shows, for each SOA, the observed and simulated
IRI distributions for the bypassers (left panel).
When task overlap was high (i.e., at the two short
SOAs), the observed distributions were consistently
quite well fitted by the simulated distributions
predicted by bottleneck bypassing. Notably, there
was a high percentage of negative IRIs (i.e., when
Task-2 response is emitted before Task-1 response),
which can easily arise from bypassing but not from
bottlenecking. The mean IRIs of the observed
distribution were very similar to the distribution
predicted by bottleneck bypassing at the 15-ms SOA
(-322 ms and -339 ms), t(8) = 0.48, p = .646, dz =
0.16, and also at the 65-ms SOA (-246 ms and -289
ms), t(8) = 0.70, p = .504, dz = 0.23. At the 250-ms
SOA, there was a nonsignificant trend towards a
slight shift of the observed distribution to the right of
the simulated distribution of bottleneck bypassing (-
8 ms vs. -104 ms), t(8) = 1.86, p = .100, dz = 0.62,
perhaps because participants find it more difficult to
perform central operations in parallel when the
stimuli do not appear to be simultaneous. In
contrast, the mean of the observed IRI distributions
differed sharply and significantly from the simulated
distributions predicted from bottlenecking: -322 ms
vs. 263 ms at the 15-ms SOA, t(8) = -7.09, p < .001,
dz = -2.36; -246 ms vs. 264 ms at the 65-ms SOA,
t(8) = -5.03, p = .001, dz = -1.68; and -8 ms vs. 267
ms at the 250-ms SOA, t(8) = -3.35, p = .010, dz = -
1.12.
Amount of Dual-Task Interference on Task 1 and
Task 2. Figure 7 shows the RTs on Task 1 and Task
2 as function of SOA (see also Table 1). RT2 was
significantly influenced by SOA, F(3, 24) = 6.53, p
=.002, ηp2 = 0.45. Post hoc analyses revealed that
RT2 increased from the longest SOA (519 ms) to the
250-ms SOA (641 ms)4, t(8) = 4.30, p = .001; no
other comparison between SOAs was significant.
The resulting PRP effect, calculated as the RT2
difference between the shortest SOA and the longest
SOA (567 vs. 519 ms), was only 48 ms. Meanwhile,
RT1 was 100 ms longer at the short SOAs (an
average of 904 ms) than the longest SOA (M = 804
ms), F(3, 24) = 4.75, p = .008, ηp2 = 0.38.
4 Because Task 1 is slower than Task 2, bypassing would often
synchronize responses (i.e., an IRI near zero) at the 250-ms
SOA. Response execution conflicts might therefore be strongest
at this SOA, explaining the RT2 increase.
PREPRINT 15
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 15
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 65
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 250
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 1500
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 15
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 65
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 250
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 1500
Observed
Predicted-bypassing
Predicted-bottlenecking
Figure 6
Observed vs. predicted distributions of inter-response intervals (IRIs) at each stimulus onset asynchrony
(SOA), separately for participants classified as bypassers (left panels) and bottleneckers (right panels),
for Experiment 2 (preparation boosting)
Bypassers (n = 9)
Bottleneckers (n = 14)
PREPRINT 16
Error Rates. Mean Task-2 error rate was 3.76%;
it was not significantly influenced by SOA, F(3, 24)
= 1.99, p = .143, ηp2 = 0.20. Mean Task-1 error rate
was 2.20% and influenced by SOA, F(3, 24) = 6.11,
p = .003, ηp2 = 0.43, with more errors committed at
the 15-ms SOA (3.54%) and 65-ms SOA (3.13%)
than the longer SOAs (1.42% at the 250-ms SOA,
and 0.72% at the 1,500-ms SOA); no other
comparison was significant.
RT1:RT2 Correlations. Consistent with
bottleneck bypassing, the coefficient of correlation
was weak overall (.28) and not significantly
influenced by SOA, F(3, 24) = 0.69, p = .566, ηp2 =
0.08.
Confirming Indicators for the Bottlenecker
Group. Even though our main focus was on the
existence of bypassers, we will briefly summarize
the results for the participants classified as
bottleneckers (n = 14) for the sake of completeness.
Visually, this group’s IRI distribution is best
fitted by the simulated distribution of bottlenecking.
However, note that their data likely also include a
minority of trials with bypassing and possibly also
response grouping. For instance, at the 15-ms SOA,
the mean of the observed distribution differed from
the mean of the simulated distribution of bypassing
(84 ms vs. -321 ms), t(13) = 9.24, p < .001, dz = 2.47,
but also from the mean of the simulated distribution
of bottlenecking (84 ms vs. 259 ms), t(13) = -4.91, p
< .001, dz = -1.31. Thus, we conjecture that these
individuals used a combination of different
processing modes on different trials: bypassing,
grouping, and bottlenecking.
Amount of Dual-Task Interference on Task 1 and
Task 2. RT2 steadily increased from the longest
SOA (M = 512 ms) to the shortest SOA (M = 923
ms), F(1.848, 24.030) = 77.22, p < .001, ηp2 = 0.86,
producing a large PRP effect of 411 ms. Meanwhile,
RT1 was slightly longer by 74 ms at the two short
SOAs (an average of 857 ms) compared to the
longest SOA (M = 783ms), F(3, 39) = 8.34, p < .001,
ηp2 = 0.39.
Error Rates. Mean Task-2 error rate was 2.55%
and was not significantly influenced by SOA, F(3,
39) = 0.34, p = .795, ηp2 = 0.03. Mean Task-1 error
rate was 2.76% and influenced by SOA, F(3, 39) =
4.03, p = .014, ηp2 = 0.24, with more errors
committed at the 15-ms SOA (3.43%) and 65-ms
SOA (3.78%) compared to the longest SOA
(1.37%); no other comparisons between SOAs were
significant.
RT1:RT2 Correlations. The coefficient of
correlation was .50 and influenced by SOA, F(3, 39)
= 7.18, p < .001, ηp2 = 0.36. A planned comparison
showed stronger correlations at the three shortest
SOAs (average of .56) than at the longest SOA (M =
.32), t(39) = 3.86, p < .001. These data suggest that
this group of participants were more often
bottlenecking (or grouping) than bypassing.
Discussion
This experiment tested the new prediction of the
preparation-neglect hypothesis that bottleneck
bypassing (i.e., fully parallel central processing) is
possible without practice. In a PRP experiment
modified to boost preparation on Task 2 (i.e., the
secondary task, known to suffer the most in dual-task
PRP situations), nearly half the sample of
individuals (n = 9) could be classified as primarily
bypassers. For this group, the average percentage of
dual-task PRP trials consistent with bypassing was
81.1% (see Figure 3). Additional analyses
confirmed that these individuals did bypass the
central bottleneck: there was a very high overlap at
300
400
500
600
700
800
900
1000
1100
1200
1300
0 500 1000 1500
RT (ms)
SOA (ms)
Bypassers (n= 9)
AV Task 1
VM Task 2
300
400
500
600
700
800
900
1000
1100
1200
1300
0 500 1000 1500
RT (ms)
SOA (ms)
Others (n= 14)
AV Task 1
VM Task 2
Figure 7
Reaction times on the auditory-vocal (AV) Task 1 and the visual-manual (VM) Task 2 as a function of the
stimulus onset asynchrony (SOA) for the bypassers (left panel) and the bottleneckers (right panel) for
Experiment 2 (preparation boosting)
PREPRINT 17
short SOAs between the observed IRI distribution
and the bypassing simulations (see Figure 6), dual-
task interference on Task 2 was quite small (i.e., a
mean PRP effect of only 48 ms), and RT1:RT2
correlations remained weak across SOAs (an
average of .28).
For those individuals who did not reach the 50%
cutoff of bottleneck bypassing, and henceforth
classified as primarily bottleneckers (n = 14), the
results indicated a preponderance of bottlenecking.
Their mean PRP effect was large (411 ms), and the
RT1:RT2 correlations were moderate at the three
shortest SOAs (.56). Although bottlenecking
predominated in this group, note that the IRI
distributions indicate that some of these participants,
especially those closest to the cutoff of 50%,
bypassed the central bottleneck on some trials.
In sum, in PRP conditions modified to boost
Task-2 preparation, the estimated percentage of
bypassing was high overall (45.1%). We found
evidence of more bypassing than bottlenecking for
nearly half the sample of participants (9 out 24).
This finding is even more striking when one
considers that the two tasks were not particularly
easy: Task 1 was the same tone-pitch identification
task used in numerous previous studies supporting
processing bottlenecks (e.g., Pashler & Johnston,
1989) and Task 2 was a shape identification task
(triangle vs. circle) with an arbitrary S-R mapping.
The finding of bottleneck bypassing with novel tasks
is extremely unusual. It contradicts prevailing
accounts of dual-task interference (e.g., Pashler &
Johnston, 1989), but supports the preparation-
neglect hypothesis.
EXPERIMENT 3: DOES INITIAL
PRACTICE ON TASK 2 ENABLE
BOTTLENECK BYPASSING?
In Experiment 2, we concluded that the key
enabler of bottleneck bypassing was boosting
preparation of Task 2, which we induced by
intermixing single-task trials on Task 2 with the
dual-task PRP trials. Yet, following the procedure
of our previous studies showing bottleneck
bypassing with easy tasks (Lyphout-Spitz et al.,
2022; Maquestiaux et al., 2020), we had also added
96 extra practice trials on Task 2 only, prior to the
dual-task PRP trials. Thus, we do not know whether
bypassing was induced by these initial 96 practice
trials or by the 96 single-task trials intermixed within
the dual-task blocks.
Experiments 3 and 4 addressed this issue by
including only the initial practice trials (Experiment
3) or only the intermixed single-task trials
(Experiment 4). Table 2 shows the differences in
experimental design between the four experiments.
According to the preparation-neglect account, it is
the latter (intermixing) that induces bottleneck
bypassing, not the former (initial practice). So,
Experiment 3 should produce bottlenecking,
whereas Experiment 4 should enable bypassing. As
will be seen, this is exactly what we found.
Experiment 3 started with 96 trials on Task 2
only, followed by 288 dual-task PRP trials (with no
mixed single-task trials). Note that the total number
of trials on Task 2 was identical to that of the
previous experiments (i.e., 384 overall).
Table 2
Overview of the Experimental Designs Used in
Experiments 1-4
Note. Exp. = experiment, PRP = psychological
refractory period.
Method
Participants. A fresh sample of 24 volunteers (M
= 22.1 years old, SD = 4.2 years; 19 women)
participated.
Design and procedure. Participants first
performed familiarization trials on each of the two
tasks separately: 6 trials on Task 1 and 6 trials on
Task 2. They then performed 96 single-task trials on
Task 2, broken into 3 blocks of 32 trials each. They
then performed 24 dual-task familiarization trials (as
in Experiments 1 and 2). Finally, they performed
288 experimental dual-task PRP trials. As in
Experiments 1 and 2, the total number of trials on
Task 2 was 384.
Results
We removed trials for which RT on either task
was below 100 ms or above 2,500 ms (dual-task
PRP: 2.16%; single-task: 0.00%). In addition, error
trials were excluded from RT analyses (dual-task
PRP: 4.79%; single-task: 3.34%).
Initial Classification. As in Experiment 1
(classic PRP), none of the 24 participants was able
No Yes
No
Exp. 1
(classic PRP)
Exp. 4
Yes Exp. 3 Exp. 2
Initial practice
on Task 2
Preparation boosting
(intermixing single-task trials on Task 2)
PREPRINT 18
to reach the 50% cutoff of bypassing; the mean
percentage was only 6.5% and ranged from 0.0% to
32.5%. All 24 participants were thus classified as
“primarily bottleneckers.”
Confirming Indicators for the Bottlenecker
Group.
Observed vs. Simulated IRI Distributions. As can
be seen in Figure 8, the observed IRI distribution
was significantly different from that predicted by
bypassing at each of the short SOAs (ps < .001): 225
vs. -189 ms at the 15-ms SOA, t(23) = 14.08, p <
.001, dz = 2.87, 231 vs. -139 ms at the 65-ms SOA,
t(23) = 11.45, p < .001, dz = 2.34, and 288 vs. 46 ms
at the 250-ms SOA, t(23) = 8.85, p < .001, dz = 1.81.
The observed IRI distributions were much closer to
that predicted by bottlenecking, albeit with
statistically significant deviations in the direction of
greater than expected IRIs (ranging from ~60 to 90
ms). At the 15-ms SOA, the observed IRI
distribution was shifted right relative to the IRI
distribution predicted by bottlenecking, as evidenced
by a statistically significant difference between their
respective means (225 ms vs. 165 ms), t(23) = 2.52,
p = .019, dz = 0.52. A similar discrepancy was
observed at the 65-ms SOA (231 ms vs. 166 ms),
t(23) = 2.82, p = .010, dz = 0.58, and at the 250-ms
SOA (288 ms vs. 200 ms), t(23) = 4.79, p < .001, dz
= 0.98.
Amount of Dual-Task Interference on Task 1 and
Task 2. Figure 9 shows the RTs on Task 1 and Task
2 as function of SOA (see also Table 1 for
descriptive statistics). Similar to Experiment 1,
mean RT2 sharply increased from the longest to the
shortest SOA (412 ms vs. 826 ms), F(1.431, 32.921)
= 136.33, p < .001, ηp2 = 0.86, yielding a large 414-
ms PRP effect. Mean RT1 remained relatively
constant across SOAs, F(3, 69) = 0.17, p = .914, ηp2
= 0.01.
Error Rates. Mean Task-2 error rate was 3.08%;
it was significantly influenced by SOA, F(3, 69) =
3.95, p = .012, ηp2 = 0.15. Post-hoc analyses
revealed that mean Task-2 error rate was
significantly higher at the shortest SOA (M = 4.11%)
than at the longest SOA (M = 1.94%), t(23) = 2.18,
p = .008; no other pairwise comparison was
significant (ps > .05). Mean Task-1 error rate was
1.79%; it was not significantly influenced by SOA,
F(1.967, 45.230) = 2.19, p = .125, ηp2 = 0.09.
RT1:RT2 Correlations. Consistent with
bottlenecking, the coefficient of correlation was
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 15
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 65
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 250
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 1500
Observed
Predicted-bypassing
Predicted-bottlenecking
Figure 8
Observed vs. simulated distributions of inter-
response intervals (IRIs) at each stimulus onset
asynchrony (SOA) for Experiment 3 (initial
practice on Task 2)
5 Our simulation of bottlenecking slightly underestimated the
true observed IRI distributions. This deviation likely occurred
because we underestimated the summed duration of stages 1C
and 2A, which we fixed in advance following Van Selst et al.
(1999).
PREPRINT 19
moderate overall (.55), and significantly influenced
by SOA, F(3, 69) = 25.77, p < .001, ηp2 = 0.53. A
planned comparison showed that correlations were
much stronger at the three short SOAs (average
coefficient of .64) than at the longest SOA (M = .28),
t(69) = 8.77, p < 001.
Discussion
Experiment 3 evaluated whether initial practice
on Task 2 is, by itself, sufficient to enable bottleneck
bypassing. It was identical to the classic PRP
paradigm of Experiment 1, except for the addition of
96 single-task practice trials on Task 2 prior to the
dual-task PRP trials. There was no evidence that the
additional practice had any effect on bypassing. The
overall probability of bypassing was only 6.5%.
Furthermore, none of the 24 participants fell into the
category of bypassers. Confirmatory analyses were
also consistent with bottlenecking: the observed IRI
distributions were closer to the IRI distribution
predicted by bottlenecking than the one predicted by
bypassing, the PRP effect was very large (414 ms),
and RT1:RT2 correlations were stronger at short
SOAs than at the longest SOA.
Overall, this additional experiment replicated the
findings from Experiment 1 (i.e., bottlenecking).
We therefore conclude that 96 initial practice trials
are not sufficient, by themselves, to enable
bottleneck bypassing.
EXPERIMENT 4: CONFIRMING
THAT MIXED SINGLE-TASK
TRIALS ON TASK 2 (I.E.,
PREPARATION BOOSTING) ARE A
KEY ENABLER OF BOTTLENECK
BYPASSING
After having demonstrated that initial practice on
Task 2 is not a key enabler of bottleneck bypassing,
Experiment 4 aimed at confirming that preparation
boosting (via intermixing single-task trials on Task
2) is the actual key. To this end, this experiment
removed the 96 initial practice trials but intermixed
128 single-task trials on Task 2 with 256 dual-task
PRP trials (see Table 2). Again, the total number of
trials on Task 2 was identical to that from previous
experiments (384 overall). Also note that the
proportion of mixed single-task trials within dual-
task PRP trials (one third) was identical to that of
Experiment 2. If preparation boosting is the key to
enabling bypassing, as asserted by the preparation
neglect hypothesis, then bottleneck bypassing
should arise even without initial practice.
Method
Participants. A fresh sample of 24 volunteers (M
= 19.1 years old, SD = 0.9 years; 21 women)
participated.
Design and procedure. Participants first
performed familiarization trials on each of the two
tasks separately: 6 trials on Task 1 and 6 trials on
Task 2. They then performed 24 dual-task
familiarization trials (as in Experiments 1-3).
300
400
500
600
700
800
900
1000
1100
1200
1300
0 500 1000 1500
RT (ms)
SOA (ms)
AV Task 1
VM Task 2
Figure 9
Reaction times on the auditory-vocal (AV) Task 1 and the visual-manual (VM) Task 2 at each stimulus
onset asynchrony (SOA) for Experiment 3 (initial practice on Task 2)
PREPRINT 20
Finally, they performed 384 experimental trials
consisting of 256 dual-task PRP trials randomly
intermixed with 128 single Task-2 trials. The 384
experimental trials were broken into 12 blocks of 32
trials each.
Results
We removed trials for which RT on either task
was below 100 ms or above 2,500 ms (dual-task
PRP: 1.61%; mixed single-task: 0.16%). In addition,
error trials were excluded from RT analyses (dual-
task PRP: 5.64%; mixed single-task: 1.57%).
Initial Classification. Nine participants were
able to reach the cutoff of bypassing, with an average
of 85.9% of trials consistent with bypassing (ranging
from 50.6% to 100.0%, cf. Figure 3). These
participants were thus classified as “primarily
bypassers.” In contrast, the 15 other participants
were classified as “primarily bottleneckers”: their
mean percentage of bypassing was 27.5% (range:
2.3% to 48.6%).
Confirming Indicators for the Bypasser
Group
Observed vs. Simulated IRI Distributions. Figure
10 shows, for each SOA, the observed and simulated
IRI distributions. When task overlap was high (i.e.,
at the two short SOAs), the observed IRI
distributions were well-fitted by the distributions
predicted by bottleneck bypassing: the mean IRIs did
not statistically differ at the 15-ms SOA (-206 ms vs.
-237 ms), t(8) = 1.14, p = .287, dz = 0.38, or the 65-
ms SOA (-144 ms vs. -187 ms), t(8) = 1.48, p = .177,
Figure 10
Observed vs. predicted distributions of inter-response intervals (IRIs) at each stimulus onset
asynchrony (SOA), separately for participants classified as bypassers (left panel) and bottleneckers
(right panel), for Experiment 4 (which included mixed singles on Task 2)
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 15
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 65
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 15
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 250
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 1500
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 65
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 250
Observed
Predicted-bypassing
Predicted-bottlenecking
0
5
10
15
20
25
-1000 -500 0 500 1000 1500 2000
Frequency (%)
IRI (ms)
SOA 1500
Observed
Predicted-bypassing
Predicted-bottlenecking
Bypassers (n = 9)
Bottleneckers (n = 15)
PREPRINT 21
dz = 0.49. At the 250-ms SOA, the observed IRI
distribution was shifted right relative to the IRI
distribution of bottleneck bypassing (94 ms vs. -2
ms), t(8) = 3.50, p = .008, dz = 1.17, as was also
observed in Experiment 2. In contrast, the observed
IRI distributions differed dramatically from the
predicted bottlenecking distributions at the three
short SOA: -206 ms vs. 229 ms at the 15-ms SOA,
t(8) = -9.24, p < .001, dz = -3.08, -144 ms vs. 230 ms
at the 65-ms SOA, t(8) = -7.47, p < .001, dz = -2.49,
and 94 vs. 241 ms at the 250-ms SOA, t(8) = -3.71,
p = .006, dz = -1.24.
Amount of Dual-Task Interference on Task 1 and
Task 2. Figure 11 shows the mean RTs on Task 1
and Task 2 as a function SOA (Table 1 shows
descriptive statistics). RT2 was influenced by SOA,
F(1.394, 11.152) = 12.83, p = .002, ηp2 = 0.62. Post
hoc analyses revealed that RT2 was longer at the
short SOAs (average of 651 ms) than at the longest
SOA (480 ms): 630 ms at the 15-ms SOA vs. 480
ms, t(8) = 4.35, p = .001, 649 ms at the 65-ms SOA
vs. 480 ms, t(8) = 4.90, p < .001, and 673 ms at the
250-ms SOA vs. 480 ms, t(8) = 5.62, p < .001; no
other comparison between SOAs was significant.
The PRP effect was 150 ms. Meanwhile, RT1 was
long overall (average of 816 ms) and influenced by
SOA, F(3, 24) = 12.46, p < .001, ηp2 = 0.61. Post
hoc analyses showed that mean RT1 was slightly
longer at the three short SOAs (average of 846 ms)
relative to the longest SOA (M = 728 ms): 850 ms at
the 15-ms SOA vs. 728 ms, t(8) = 5.08, p < .001, 858
ms at the 65-ms SOA vs. 728 ms, t(8) = 5.38, p <
.001, and 830 ms at the 250-ms SOA vs. 728 ms, t(8)
= 4.23, p = .002; no other comparison between SOAs
was significant.
Error Rates. Mean Task-2 error rate was 5.15%;
it was significantly influenced by SOA, F(3, 24) =
3.60, p = .028, ηp2 = 0.31. Post hoc analyses revealed
that mean Task-2 error rate was significantly higher
at the 250-ms SOA (M = 7.16%) than at the 15-ms
SOA (M = 3.16%), t(8) = 3.02, p =.036; no other
comparison between SOAs was significant. Mean
Task-1 error rate was 1.98% and remained relatively
constant across SOAs, F(3, 24) = 1.36, p = .280, ηp2
= 0.15.
RT1:RT2 Correlations. Consistent with
bottleneck bypassing, the coefficient of correlation
was weak overall (.25) and not significantly
influenced by SOA, F(3, 24) = 0.25, p = .860, ηp2 =
0.03: the correlations were .22, .23, .29, and .25 at
the 15-ms, 65-ms, 250-ms, and 1,500-ms SOAs,
respectively.
Confirming Indicators for the Bottlenecker
Group. Visually, at the two shortest SOAs, this
group’s IRI distributions were closer to the
bottlenecking distribution, but shifted in the
direction of the bypassing distribution; this
presumably reflects that bypassing (and/or response
grouping) occurred on a modest proportion of the
trials. At the 15-ms SOA, the mean of the observed
IRI distribution was located 280 ms right of the
predicted bypassing distribution (72 ms vs. -208 ms),
t(23) = 10.47, p < .001, dz = 2.70, and 141 ms left of
the predicted bottlenecking distribution (72 ms vs.
213 ms), t(23) = -4.32, p < .001, dz = -1.12. The
same pattern also held at the 65-ms SOA: the mean
of the observed distribution was located 283 ms right
of the predicted bypassing distribution (125 ms vs. -
158 ms), t(23) = 8.85, p < .001, dz = 2.29, and 90 ms
left of the predicted bottlenecking distribution (124
ms vs. 214 ms), t(23) = -2.56, p = .023, dz = -0.66.
At the 250-ms SOA, the mean of the observed
distribution significantly differed from that of the
predicted bypassing distribution (233 ms vs. 27 ms),
t(23) = 7.09, p < .001, dz = 1.83, whereas the means
of the observed distribution and the predicted
300
400
500
600
700
800
900
1000
1100
1200
1300
0 500 1000 1500
RT (ms)
SOA (ms)
Bottleneckers (n= 15)
AV Task 1
VM Task 2
300
400
500
600
700
800
900
1000
1100
1200
1300
0 500 1000 1500
RT (ms)
SOA (ms)
Bypassers (n= 9)
AV Task 1
VM Task 2
Figure 11
Reaction times on the auditory-vocal (AV) Task 1 and the visual-manual (VM) Task 2 as a function of the
stimulus onset asynchrony (SOA) for the bypassers (left panel) and the bottleneckers (right panel) for
Experiment 4 (mixed singles on Task 2)
PREPRINT 22
bottlenecking distribution did not statistically differ
(233 ms vs. 222 ms), t(23) = 0.41, p = .686, dz = 0.11.
Amount of Dual-Task Interference on Task 1 and
Task 2. Mean RT2 was significantly influenced by
SOA, F(1.777, 24.882) = 79.45, p < .001, ηp2 = 0.85:
RT2s were 794 ms, 820 ms, 740 ms, and 462 ms at
the 15-ms, 65-ms, 250-ms, and 1,500-ms SOAs,
respectively. The mean PRP effect, calculated as the
RT difference between the most extreme SOAs, was
332 ms. RT1 was also significantly influenced by
SOA, F(3, 42) = 10.94, p < .001, ηp2 = 0.44: RT1s
were 737 ms, 761 ms, 757 ms, and 684 ms at the 15-
ms, 65-ms, 250-ms, and 1,500-ms SOAs,
respectively.
Error Rates. Mean Task-2 error rate was 2.48%;
it was not significantly influenced by SOA, F(3, 42)
= 0.98, p = .410, ηp2 = 0.07. Mean Task-1 error rate
was 2.34%. It was significantly influenced by SOA,
F(3, 42) = 5.03, p = .005, ηp2 = 0.26: Task-1 error
rates were 3.05%, 3.17%, 2.12%, and 0.94% at the
15-ms; 65-ms, 250-ms, and 1,500-ms SOAs,
respectively.
RT1:RT2 Correlations. The coefficient of
correlation was .49 overall and influenced by SOA,
F(3, 42) = 14.78, p < .001, ηp2 = 0.51. A planned
comparison showed significantly stronger
correlation at the three shortest SOAs (average of
.57) than the longest SOA (M = .24), t(42) = 5.93, p
< .001. This pattern suggests that this group of
participants were more often bottlenecking than
bypassing.
Discussion
Experiment 4 tested whether preparation
boosting is the key enabler of bottleneck bypassing.
Thus, this experiment intermixed single-task trials
on Task 2 with dual-task PRP trials, but did not
include the initial 96 practice trials on Task 2 (see
Table 2). Despite the absence of initial practice on
Task 2 (beyond 30 familiarization trials), 49.4% of
the dual-task PRP trials were consistent with
bypassing. Relatedly, 9 participants (out of 24) were
able to bypass the central bottleneck; their mean
percentage of bypassing was high (85.9%). Their
observed IRI distributions at the two short SOAs
were well fitted by the simulated IRI distributions of
bottleneck bypassing. Furthermore, their mean PRP
effect was especially small (150 ms) given their long
mean RT1 (816 ms). Finally, the RT1:RT2
correlations were weak overall (.25).
In sum, the results of Experiment 4 provided a
close replication of the frequent bypassing observed
in Experiment 2. They also confirm that the key
enabler of bottleneck bypassing is not extra initial
practice (which had no apparent effect), but rather
preparation boosting, induced by intermixing single-
task trials on Task 2 within the dual-task PRP trials.
Between-Experiment and Between-
Group Comparisons
To formally demonstrate the impact of our
manipulation to boost preparation (intermixing
single-task trials amongst the dual-task PRP trials),
we statistically compared across experiments the key
indicators of dual-task interference on Task 2 and
RT1:RT2 correlations.
The Effects of Preparation Boosting Via
Mixed Single-Task Trials
For this analysis we pooled Experiments 1 and 3
(without a preparation boost) and compared them to
Experiments 2 and 4 (with a boost). We carried out
independent-samples t tests on the two key
indicators of bottlenecking vs. bottleneck bypassing:
the amount of dual-task interference on Task 2 and
the RT1:RT2 correlations averaged across the three
shortest SOAs. According to the preparation-neglect
hypothesis, preparation enables bottleneck
bypassing. Thus, dual-task interference on Task 2
and the correlations at short SOAs should be
significantly smaller in Experiments 2 and 4 than in
Experiments 1 and 3.
Consistent with preparation neglect, the amount
of dual-task interference on Task 2 was significantly
smaller in Experiments 2 and 4 (M = 266 ms, SD =
185 ms) than in Experiments 1 and 3 (M = 430 ms,
SD = 141 ms), t(93) = -4.86, p < .001 (one-tailed), ds
= -1.00. Also consistent with preparation neglect,
the RT1:RT2 correlations at short SOAs were
significantly smaller in Experiments 2 and 4
(average of .45) than in Experiments 1 and 3
(average of .67), t(93) = -5.39, p < .001 (one-tailed),
ds = -1.11. In sum, the comparisons between
experiments confirmed that preparation is a key
enabler of bottleneck bypassing.
Bottleneckers vs. Bypassers
We classified individuals based on their IRIs. To
verify that the bypassers really did show more
efficient dual-task performance than the
bottleneckers, we compared them statistically. We
first combined the bypassers from Experiments 2
and 4 (n = 18), as well as the bottleneckers (n = 29)
from Experiments 2 and 4.
The results showed much smaller dual-task
interference on Task 2 for the bypassers than the
bottleneckers (99 vs. 370 ms), t(45) = -6.95, p < .001
(one-tailed), ds = -2.09. In addition, the RT1:RT2
correlation at short SOAs was significantly smaller
for the bypassers than the bottleneckers (.27 vs. .57),
t(45) = -5.21, p < .001 (one-tailed), ds = -1.56. These
PREPRINT 23
analyses confirm more efficient and more
independent performance among the bypassers than
the bottleneckers.
General Discussion
In this study, we evaluated the possibility that the
central bottleneck can be bypassed even for novel
tasks, allowing central processes (e.g., selecting a
response) to be fully performed in parallel. Such a
result would be unprecedented and should not be
possible according to prevailing accounts of dual-
task interference and past results (e.g., Pashler,
1994). Indeed, reviews of dual-task interference do
not even mention the possibility of bottleneck
bypassing with novel tasks (e.g., Koch et al., 2018).
The preparation neglect hypothesis, however,
predicts that merely boosting task preparation should
enable bottleneck bypassing.
Experiment 1 confirmed that, in classic PRP
conditions, none of the 24 participants was able to
reliably bypass the central bottleneck. The observed
IRI distributions closely fitted the bottlenecking
simulations but were poorly fitted by the bypassing
simulations. In addition, the amount of dual-task
interference on Task 2 was very large (445 ms) and
the RT1:RT2 correlations were strong at short SOAs
(.71). Thus, this classic PRP experiment confirmed
the presence of a central bottleneck that severely
constrains dual-task performance, thereby
replicating dozens of previous PRP studies.
The results of Experiment 2, however, told a
different story, despite using the same tasks. The
main change was the insertion of additional single-
task trials of Task 2 within the dual-task PRP blocks,
to give participants extra incentive to prepare for that
task. Surprisingly, nearly half the sample of
participants (9 participants out of 24) were able to
bypass the central bottleneck. This was evidenced
by a very high percentage of dual-task PRP trials
consistent with bypassing (81.1%; see Figure 3). For
these 9 bypassers, the observed IRI distributions
closely matched the bypassing simulations, but
poorly matched the bottlenecking simulations. In
addition, the PRP effect was only 48 ms, and
RT1:RT2 correlations were weak (.28 on average).
For the 14 other participants who did not reach the
cutoff of 50%, the results indicated a preponderance
of bottlenecking. Although these individuals
primarily bottlenecked, the IRI distributions suggest
that some were able to occasionally bypass (see
Figure 3).
Experiments 3 and 4 compared the effects of
adding single-task practice of Task 2 prior to the
dual-task trials (Experiment 3) versus intermixing
single-task trials of Task 2 within the dual-task PRP
trials (Experiment 4). According to the preparation
neglect account, the latter should boost preparation
and therefore enable bypassing, whereas the former
should not. That is exactly what we found. Initial
practice on Task 2, prior to a classic PRP design, led
to bottlenecking (Experiment 3). However,
intermixing single-task trials of Task 2 within dual-
task PRP blocks led to frequent bypassing
(Experiment 4). Closely replicating Experiment 2,
we estimated that bypassing occurred on 49% of the
trials and that it predominated for 9 of 24
participants.
As noted above, the PRP effect was significantly
reduced in the experiments with a preparation boost
(266 ms; Experiments 2 and 4), compared to those
without the preparation boost (430 ms; Experiments
1 and 3). Furthermore, the proportion of participants
classified as bypassers was significantly greater for
the preparation boosting Experiments 2 and 4 (18 out
of 47) than for the classic PRP Experiments 1 and 3
(0 out of 48), χ² (1, N = 95) = 22.68, p < .001.
Overall, the present study supports the preparation-
neglect hypothesis of dual-task interference.
Preparation: A Key Enabler of Bottleneck
Bypassing
Previous studies have pointed out that
preparation plays at least some role in dual-task
interference, albeit secondary (Gottsdanker, 1980;
Logan, 1978; Pashler, 1984, 1994). For instance,
Pashler (1984) speculated that, in addition to a
central bottleneck, some Task-2 slowing stems from
insufficient preparation of Task 2. Later, Pashler
(1994) summarized this speculation as follows: “the
inability to select two responses at the same time (the
central bottleneck) is not the only cause of dual-task
slowing… dual-task slowing is probably increased
by the fact that tasks are prepared less effectively
when other tasks must be prepared at the same time”
(p. 323). This view, however, clearly does not
predict any bypassing with increased task
preparation, nor does it note the specific problem
with Task-2 preparation.
According to De Jong (1995), preparation is
needed to schedule in advance the dual-task
processing sequence, in particular the allocation of
attention to Task 1 followed by a subsequent switch
of attention to Task 2. Even though De Jong
speculated that “parallel processing should also be
promoted when one or both tasks require only
minimal preparation, as when one or both tasks are
highly ideomotor compatible” (p. 22), he did not
consider it to be a possibility with novel tasks.
In contrast to Pashler’s (1994) or De Jong’s
(1995) accounts, the present preparation-neglect
hypothesis posits that preparation plays a primary
PREPRINT 24
role in dual-task performance. Consistent with this
new hypothesis, we demonstrated that, in conditions
boosting preparation on Task 2, the central
bottleneck could be entirely bypassed without
practice, on almost half the trials. Preparation
boosting was induced by randomly intermixing
single-task trials on Task 2 with dual-task PRP trials.
In our view, these mixed single-task trials on Task 2
encouraged participants to get ready in advance for
Task 2, by loading the S-R mapping of Task 2 into
working memory. This allowed the Task-2 stimulus
to automatically activate the associated response
code, and the response to be executed, without
recruiting attention. A closely related possibility is
that preparing to (possibly) perform Task 2 alone
causes participants to set up advance permission for
the Task-2 response to execute automatically (e.g.,
when the response code reaches some threshold of
activation).
The Bypassers Really are Bypassing, Not
Reverse Bottlenecking
Before concluding that preparation boosting
enabled 18 individuals to bypass the central
bottleneck most of the time, it is important to show
that they did not merely perform the central
operations of the tasks serially, but starting with the
second task on almost half the trials (i.e., reverse
bottlenecking). To examine this possibility, we
simulated the IRI distributions predicted by reverse
bottlenecking and bottleneck bypassing at the two
shortest SOAs (i.e., those the most sensitive to
discriminate between candidate models), for all 18
individuals identified as “primarily bypassers” in
Experiments 2 (n = 9) and 4 (n = 9). We then
compared their simulated and observed IRI
distributions.
As can be seen in Figure 12, the IRI distribution
predicted by reverse bottlenecking provided a very
poor fit of the bypasser group’s IRI distribution: the
mean of the predicted reverse bottlenecking
distribution was located 309 ms left of the observed
distribution (-538 ms vs. -229 ms), t(17) = -10.21, p
< .001, dz = -2.41. In sharp contrast, the IRI
distribution predicted by bottleneck bypassing
provided an excellent fit of the observed IRI
distribution: the means of the predicted bypassing
distribution (-263 ms) and of the observed
distribution (-229 ms) did not statistically differ,
t(17) = -1.30, p = .211, dz = -0.31.
Additionally, mean RT1 at the two shortest SOAs
should be much longer in case of reverse
bottlenecking (because now Task 1 suffers the PRP
effect) than bottleneck bypassing. This is exactly
what happened: mean RT1 as predicted by the
simulation of reverse bottlenecking (1,076 ms) was
much longer than mean RT1 as observed in the 18
bypassers (880 ms), t(17) = 9.86, p < .001, dz = 2.32.
These analyses confirm the conclusion that the 18
identified bypassers really were able to bypass the
central bottleneck.6,7
Figure 12
Observed vs. predicted distributions of inter-
response intervals (IRIs) at the two shortest stimulus
onset asynchronies (SOAs) for participants
classified as bypassers in experiments boosting
preparation (i.e., Experiments 2 and 4)
Is Preparation Boosting a Stable and Global
Strategy?
After having demonstrated that bottleneck
bypassing on the novel Task 2 was enabled by
boosting preparation on Task 2, it is then only
natural to examine the nature of this boost. Is it a
transient strategy or a more stable strategy.8
Because this boost was induced by mixed single-task
trials on Task 2, we examined whether there was a
local influence of the preceding trial (mixed single-
task trial on Task 2 vs. dual-task trial) on the
6 As rightly pointed out by Iring Koch, our model predicts a
longer mean RT1 in case of bottleneck bypassing than in case of
bottlenecking, because greater preparation of Task 2 implies less
preparation of Task 1 (preparation is a limited resource). To
evaluate this prediction, we compared mean RT1 as a function
of SOA between Experiment 3 (which revealed bottlenecking)
and Experiment 4 (which revealed bottleneck bypassing in
nearly half the sample). We compared these two experiments
because they used roughly similar numbers of dual-task PRP
trials (i.e., 288 in Experiment 3, 256 in Experiment 4).
Consistent with the view that preparation is a limited resource,
mean RT1 was longer in Experiment 4 (producing bypassing)
than in Experiment 3 (producing bottlenecking): 765 ms vs. 620
ms, F(1, 46) = 9.84, p = .003, ηp2 = 0.176.
7 Following the multi-phase framework of temporal response-
order control (Pieczykolan & Huestegge, 2019), bottleneck
bypassing may represent a third response-order control (in
addition to no-reversal response order and reversal response
order): either response order. Note that this response order might
be particularly vulnerable to instructions encouraging
participants to explicitly reverse their responses.
8
We thank Kate Arrington, Lynn Huestegge, Iring Koch, and an
anonymous reviewer for suggesting this analysis.
PREPRINT 25
performance of the subsequent dual-task PRP trial.
If preparation is a transient strategy, then RT2 should
be faster when preceded by a mixed single-task trial
on Task 2 relative to a dual-task PRP trial. But if
preparation is a more stable and global strategy, then
RT2 should not be significantly influenced by the
nature of the preceding trial.
To test these predictions, we analyzed the
combined data of Experiments 2 and 4 (i.e., the two
experiments using mixed single-task trials on Task
2). Consistent with the view that preparation
boosting is a global strategy, Task-2 performance on
dual-task PRP trials was virtually identical
regardless of whether the preceding trial was a
mixed single-task trial on Task 2 or a dual-task PRP
trial (692 vs. 690 ms), F(1, 46) = 0.08, p = .774, ηp2
= 0.002. Also, neither the type of preceding trial X
type of subgroup interaction, F(1, 45) = 0.60, p =
.445, ηp2 = 0.01, the preceding trial X SOA
interaction, F(2.463, 110.823) = 2.44, p = .080, ηp2 =
0.05, nor the type of preceding trial X type of
subgroup x SOA interaction, F(2.463, 110.823) =
1.40, p = .250, ηp2 = 0.03, were significant. In short,
we did not detect any task order effects.
We also checked whether bottleneck bypassing
occurred only because individual bypassers
recognized a stimulus repetition from the preceding
trial. For the bypassers identified in Experiments 2
and 4, we examined whether dual-task RT2 across
SOAs depended on whether the preceding mixed
single-task trial displayed the exact same Task-2
stimulus (e.g., a triangle then another triangle) or
different (e.g., triangle then circle).10 Task-2
stimulus repetition (same vs. different) had no
significant effect (696 vs. 688 ms), F(1, 46) = 0.33,
p = .568, ηp2 = 0.007. Also, Task-2 stimulus
repetition did not significantly interact with SOA,
F(2.101, 96.644) = 0.60, p = .561, ηp2 = 0.01.
Therefore, bottleneck bypassing did not occur
simply because participants recognized an exact
stimulus repetition.
Individual Differences
Another important contribution of the present
study is the demonstration of large individual
differences, with the degrees of bypassing ranging
from near 0% to near 100% of trials (see Figure 3).
Several participants primarily bypassed the
bottleneck (n = 9 in Experiment 2, n = 9 in
Experiment 4), whereas others primarily
bottlenecked (n = 14 in Experiment 2, n = 15 in
Experiment 4). We therefore suggest that future
studies take individual differences into account,
rather than solely analyzing data for the entire
sample together. Note, however, that the amount of
bypassing was not strictly bimodal; rather we
observed many individuals with intermediate
frequencies of bypassing.
Individual differences are even more salient
when plotting PRP effects against bypassing
probabilities for each individual across all four
experiments, as shown in Figure 13. The PRP
effects were linearly related to the individual
probabilities of bypassing across Experiments 1-4 (N
= 95), with a strong negative correlation, r(95) = -
.73, p < .001. Those individuals with near 0%
bypassing probability produced 462 ms of dual-task
interference on Task 2 whereas those with 100%
bypassing probability produced negligible dual-task
interference on Task 2 (only 41 ms). We conclude
that residual PRP effects are largely due to the
percent of trials in which the bottleneck was still in
force (i.e., not bypassed).
What are the causes of these individual
differences? As in Maquestiaux and Ruthruff (2021)
and Lyphout-Spitz et al. (2022), baseline skill at
Task 2 or baseline skill at Task 1, as measured by
RT, did not predict who would bypass. In
Experiment 2, the percent of trials consistent with
bypassing did not significantly correlate with mean
RT2 in the initial single-task blocks, r(23) = -.10, p
= .659. Relatedly, there was also no significant
correlation in Experiments 2 and 4 combined, with
mean RT2 measured at the longest SOA, r(47) =.02,
p = .906. Similarly, there was no correlation with
Task-1 skill as measured by RT1 at the longest SOA
in Experiments 2 and 4, r(47) = .07, p = .632.
Watson and Strayer (2010) reported a similar finding
with practiced tasks: skill level at the individual
tasks (i.e., driving and complex arithmetic
calculations) did not explain why 2.5% of their
participants showed extraordinary dual-tasking
ability.
Because both processing modes (i.e., bypassing
vs. bottlenecking) emerged from the very beginning,
9 It is logically possible that some bypassed more than others
only because they were fortunate to receive an unusually large
number of dual-task trials preceded by a single-task trial.
However, an ad hoc analysis confirmed that the number of such
occurrences did not differ statistically between the bypassers and
the bottleneckers in Experiment 2 (64.1 vs. 66.2 trials), t(21) = -
1.14, p = .266, ds = -0.49, or in Experiment 4 (84.4 vs. 87.4 trials),
t(22) = -1.38, p = .182, ds = -0.58.
10 A post hoc analysis showed that the number of stimulus
repetition from single-task trials to dual-task PRP trials did not
differ statistically between the bypassers and the bottleneckers
in Experiment 2 (33.1 vs. 34.4 trials), t(21) = -0.69, p = .499, ds
= -0.29. It was statistically significant in Experiment 4, but with
fewer stimulus repetitions for the bypassers than the
bottleneckers (39.4 vs. 44.4 trials), t(22) = -2.64, p = .015, ds = -
1.11.
PREPRINT 26
independently of skill level, we tentatively speculate
that it is an individual choice for one strategy over
another, perhaps driven by metacognitive factors
such as confidence in the ability to multitask (for a
similar proposal in the field of associative learning,
see Touron & Hertzog, 2004; for a discussion of
individual differences in multitasking, see Brüning
et al., 2022). Relatedly, some individuals might
choose to prepare Task 2 and others to focus their
preparation on Task 1.
The Classic PRP Paradigm Teaches
Participants to Bottleneck
Does a preparation boost cause bypassing to
gradually develop across trials? Or does the classic
PRP paradigm teach participants to stop bypassing?
Figure 14 shows, for each experiment, the average
probability of bypassing across blocks of 64 dual-
task PRP trials (a more detailed breakdown across
individuals can be found in the online supplemental
material). For the experiments with preparation
boosting (i.e., Experiments 2 and 4), the estimated
probability was fairly high even in the first block and
did not tend to increase or decrease.
Figure 14
Mean probability of bypassing across blocks for
Experiments 1-4 (N = 95)
For the classic PRP experiments with no
preparation boost (i.e., Experiments 1 and 3),
however, the probabilities of bypassing began at a
modest level then dropped markedly across sessions:
from 31.2% in block 1 to 1.1% in blocks 5-6 for
Experiment 1, F(1.239, 28.492) = 16.83, p < .001,
ηp2 = 0.423, and from 14.5% in block 1 to 1.7% in
Figure 13
Psychological refractory period (PRP) effect as a function of the probability of bypassing the central
bottleneck for each individual participant of Experiments 1-4 (N = 95)
Note. For the regression line (represented by the oblique dashed line): PRP effect = -4.21 (probability
of bypassing) + 462.13, r² = .528. The four larger shapes represent the mean values of Experiments
1, 2, 3, and 4.
PREPRINT 27
blocks 4-5 for Experiment 3, F(1.770, 40.721) =
6.07, p = .007, ηp2 = 0.209. Apparently, classic PRP
trials, with a fixed task order, caused participants to
stop bypassing and begin bottlenecking. A
preparation boost, however, counteracted this effect
and participants continued to bypass despite a fixed
task order.
Relation to Previous Work
Broadly speaking, the present findings support
the view that there is no immutable structural central
limitation (Hazeltine et al., 2002; Meyer & Kieras,
1997a, 1997b, 1999; Schumacher et al., 2001).
Meyer and Kieras (1997b) envisioned that the
central limitation is under the strategic control of the
individual. According to these authors, individuals
choosing a daring scheduling strategy can perform
the central operations of two tasks at the same time
(i.e., bottleneck bypassing).
However, Meyer and Kieras (1997a, 1997b)
underestimated the possibility of bottleneck
bypassing without practice. In fact, Meyer and
Kieras (1999) proposed that highly efficient dual-
task processing “occurs when five prerequisite
conditions prevail in combination: (a) participants
are encouraged to give the task equal priority; (b)
participants are expected to perform each task
quickly; (c) there are no constraints on temporal
relations or serial order among responses; (d)
different tasks use different perceptual and motor
processors; and (e) participants receive enough
practice to compile complete production rule sets for
performing each task.” (p. 54). We agree with many
of their prerequisites, especially (c) as well as (d),
which is supported by research demonstrating that
sensory-motor modality compatibility plays a key
role in enabling bottleneck bypassing (Halvorson &
Hazeltine, 2015, 2019; Hazeltine & Ruthruff, 2006;
Hazeltine et al., 2006; Maquestiaux et al., 2018) as
well as in reducing switching costs (Stephan &
Koch, 2011, 2015, 2016). However, Experiments 2
and 4 of the present study clearly do not meet
requisite (e), as we did not provide extensive practice
(unlike Schumacher et al., 2001, who had
participants perform six sessions). Also, although
our manipulation of preparation most closely
resembles prerequisite (a), we did not find it
necessary to provide equal emphasis on Task 1 and
Task 2; rather, it may be sufficient to provide
adequate emphasis on Task 2.
Given that bottleneck bypassing is highly
sensitive to experimental conditions (e.g.,
eliminating input and output conflicts, promoting
preparation on Task 2), one can question whether it
is truly under “strategic control” as advocated by
Meyer and Kieras (1997b). Future studies are
needed to directly address this question.
Constraints on Generality
The target population for our study is younger
adults. We used samples of young adults (average
of 20 years old), mostly students enrolled in various
bachelor programs at the Université de Franche-
Comté, Besançon (France), which is located in a
rural region (Bourgogne-Franche-Comté).
Although we expect our main finding – parallel
central processing, with negligible interference, with
novel tasks – to generalize to samples of younger
adults with different backgrounds, this has not yet
been established.
Older adults generally experience greater dual-
task difficulties and are less likely to automatize
tasks with practice (e.g., Maquestiaux et al., 2004,
2010, 2013; Hartley & Maquestiaux, 2007; Hartley
et al., 2015, 2016; Maquestiaux & Ruthruff, 2021).
Therefore, it remains to be seen whether parallel
central processing will generalize to this population.
We report a very unusual finding (bottleneck
bypassing with novel tasks) in a dual-task PRP
experiment boosting task preparation. We did so by
randomly intermixing single-task trials on that task
along with dual-task PRP trials. However, several
caveats are in order regarding the generality of this
finding. First, many real-world situations might
better resemble the classic PRP paradigm (e.g.,
typing requires a sequence of actions in a correct
order) and thus promote preparation neglect, leading
to ubiquitous central bottlenecks.
Second, even with incentives to prepare Task 2,
we found that not all participants bypassed the
bottleneck consistently; roughly half appeared to be
bottlenecking most of the time. It remains to be seen
whether the proportion of bypassing can be
increased, for example by further incentivizing the
preparation of Task 2.
Third, although we used tasks very similar to
those used in the seminal studies that originally
argued for a central bottleneck, these tasks are
admittedly simple compared to many real-world
tasks. Presumably, central interference will be
strong for particularly complex tasks, such as
planning a vacation or driving a car. Further work is
therefore needed to identify the boundary conditions
of bottleneck bypassing and nearly perfect time-
sharing. Based on the preparation neglect account,
we speculate that the ease with which a task can be
represented in working memory will be particularly
critical, rather than task complexity (the number of
stimuli and responses) and stimulus-response
compatibility, per se.
PREPRINT 28
Conclusions and Contributions
The present study confirmed bottlenecking in
classic PRP experiments (Experiments 1 and 3) but,
most strikingly, found bottleneck bypassing in PRP
experiments modified to boost task preparation
(Experiments 2 and 4). Evidence for bottleneck
bypassing with novel tasks is unprecedented. These
novel findings contradict prevailing models of dual-
task interference, which do not predict bottleneck
bypassing on novel tasks, without extensive practice.
These findings also contradict the view that what
matters the most in improving Task-2 performance
in PRP experiments is dual-task practice (rather than
mere single-task practice), which is supposed to aid
the acquisition of dual-task coordination skills and
efficient instantiation of task sets (Schubert &
Strobach, 2018; Strobach, 2020). Coordination
skills might be critical in classic PRP paradigms.
Yet here we found more bypassing in Experiment 2
even though these participants actually performed
only half as many dual-task trials relative to
Experiment 1 (192 vs. 384). As confirmed in
Experiments 3 and 4, what matters is not practice but
rather boosting task preparation on the task that is
most likely to be neglected (i.e., Task 2).
Third, the present study sheds light on the claim
that the PRP procedure is not valid for studying dual-
task processing (Halvorson et al., 2013; Huestegge
& Hazeltine, 2011). According to Huestegge and
Hazeltine (2011), “the PRP paradigm may not be a
valid model to test to what extent two tasks can
principally be processed simultaneously, because its
inherent structure (i.e., the variable SOA) seems to
impose a task-specific serial processing strategy.
This severely limits the validity of PRP paradigm as
a general model for cognitive processes during
multitasking” (p. 446). On the one hand, our
Experiments 1 and 3 confirm that the standard PRP
paradigm is in fact not conducive to bottleneck
bypassing (and may teach participants to
bottleneck). On the other hand, the variant of the
PRP paradigm used in Experiments 2 and 4 revealed
converging evidence of bottleneck bypassing. What
prevented bypassing in Experiments 1 and 3 was not
the use of multiple SOAs per se but rather the lack
of preparation on Task 2. When boosting Task-2
preparation by intermixing dual-task PRP trials with
single-task trials on Task 2, nearly half of the
participants in Experiments 2 and 4 (i.e., 9 out 24 in
each experiment) were able to simultaneously
process the two tasks without central constraints.
Meanwhile, there are advantages to using multiple
SOAs. For one, it reduces the chances of a
bottleneck being latent (Maquestiaux et al., 2008,
2010; 2013, 2018, 2020; Maquestiaux & Ruthruff,
2021; Ruthruff et al., 2003, 2006, 2009). Another
advantage is that multiple SOAs reduce the
likelihood of participants combining both tasks into
one supratask (in which case there would no longer
be any multitasking). In addition, the PRP paradigm
mimics many real-world situations that have
variable SOAs (e.g., athletics, driving). So, although
we agree that the classic PRP paradigm is not well-
suited for studying bottleneck bypassing, variants of
it are.
Fourth, the theoretical contribution of this study
is the validation of the preparation-neglect
hypothesis. Preparation neglect not only can explain
why dozens of previous experiments have shown
large amounts of dual-task interference, but also why
the present study obtained unprecedented evidence
of dual-task automaticity with novel tasks.
Context of the Research
The main idea of this work, the preparation-
neglect hypothesis to explain dual-task interference,
was inspired by our previous works showing
massive evidence of parallel central processing (i.e.,
bottleneck bypassing) with very easy tasks
(Lyphout-Spitz et al., 2022; Maquestiaux et al.,
2020; Maquestiaux & Ruthruff, 2021). This work
forms parts of the first author’s PhD dissertation, as
well as the four authors’ research program, which
investigates the nature of capacity limitations in
dual-task performance with a focus on the manner
with which tasks are represented in the human mind.
References
Anderson, J. R., Taatgen, N. A., & Byrne, M. D.
(2005). Learning to achieve perfect timesharing:
Architectural implications of Hazeltine, Teague,
and Ivry (2002). Journal of Experimental
Psychology: Human Perception and
Performance, 31, 749-761.
Brüning, J., Koob, V., Manzey, D., & Janczyk, M.
(2022). Serial and parallel processing in
multitasking: Concepts and the impact of
interindividual differences on task and stage
levels. Journal of Experimental Psychology:
Human Perception and Performance, 48, 724-
742.
Carrier, L. M., & Pashler, H. (1995). Attentional
limits in memory retrieval. Journal of
Experimental Psychology: Learning, Memory,
and Cognition, 21, 1339-1348.
De Jong, R. (1993). Multiple bottlenecks in
overlapping task performance. Journal of
PREPRINT 29
Experimental Psychology: Human Perception
and Performance, 19, 965-980.
De Jong, R. (1995). The role of preparation in
overlapping-task performance. Quarterly
Journal of Experimental Psychology, 48A, 2-25.
Fischer, R., & Plessow, F. (2015). Efficient
multitasking: parallel versus serial processing of
multiple tasks. Frontiers in Psychology, 6, 1-11.
Gottsdanker, R. (1980). The ubiquitous role of
preparation. In G. Stelmach and J. Requin (Eds.),
Tutorials in minor behavior (pp. 315-371).
Amsterdam: North-Holland.
Green, C., Johnston, J. C., & Ruthruff, E. (2011).
Attentional limits in memory retrieval –
Revisited. Journal of Experimental Psychology:
Human Perception and Performance, 37, 1083-
1098.
Halvorson, K. M., Ebner, H., & Hazeltine, E. (2013).
Investigating perfect timesharing: The
relationship between IM-compatible tasks and
dual-task performance. Journal of Experimental
Psychology: Human Perception and
Performance, 39, 413-432.
Halvorson, K. M., & Hazeltine, E. (2015). Do small
dual-task costs reflect ideomotor compatibility or
the absence of crosstalk? Psychonomic Bulletin
& Review, 22, 1403-1409.
Halvorson, K. M., & Hazeltine, E. (2019).
Separation of tasks into distinct domains, not set-
level compatibility, minimizes dual-task
interference. Frontiers in Psychology, 10, 1-10.
Hartley, A. A., & Little, D. M. (1999). Age-related
differences and similarities in dual-task
interference. Journal of Experimental
Psychology: General, 128, 417-450.
Hartley, A. A., & Maquestiaux, F. (2007). Success
and failure at dual-task coordination by younger
and older adults. Psychology and Aging, 22, 215-
222.
Hartley, A. A., Maquestiaux, F., Festini, S. B.,
Frazier, K., & Krimmer, P. J. (2016). Backward
compatibility effects in younger and older adults.
Attention, Perception, & Psychophysics, 78,
1337-1350.
Hartley, A. A., Seaman, B., & Maquestiaux, F.
(2015). Ideomotor-compatible tasks partially
escape dual-task interference in both young and
elderly adults. Psychology and Aging, 30, 36-45.
Hazeltine, E., & Ruthruff, E. (2006). Modality
pairing effects and the response selection
bottleneck. Psychological Research, 70, 504-
513.
Hazeltine, E., Ruthruff, E., & Remington, R. W.
(2006). The role of input and output modality
pairings in dual-task performance: Evidence for
content-dependent central interference.
Cognitive Psychology, 52, 291-345.
Hazeltine, E., Teague, D., & Ivry, R. B. (2002).
Simultaneous dual-task performance reveals
parallel response selection after practice. Journal
of Experimental Psychology: Human Perception
and Performance, 28, 527-545.
Hommel, B. (1998). Automatic stimulus-response
translation in dual-task performance. Journal of
Experimental Psychology: Human Perception
and Performance, 24, 1368-1384.
Huestegge, L., & Hazeltine, E. (2011). Crossmodal
action: modality matters. Psychological
Research, 75, 445-451.
Huestegge, L., & Strobach, T. (2021). Structuralist
mental representation of dual-action demands:
Evidence for compositional coding from dual
tasks with low cross-task dimensional overlap.
Acta Psychologica, 103298.
Janczyk, M. (2016). Sequential modulation of
backward crosstalk and task-shielding in dual-
tasking. Journal of Experimental Psychology:
Human Perception and Performance, 42, 631-
647.
Janczyk, M., & Kunde, W. (2020). Dual tasking
from a goal perspective. Psychological Review,
127, 1079-1096.
Janczyk, M., Renas, S., & Durst, M. (2018).
Identifying the locus of compatibility-based
backward crosstalk: Evidence from an extended
PRP paradigm. Journal of Experimental
Psychology: Human Perception and
Performance, 44, 261-276.
Koch, I., Poljac, E., Müller, H., & Kiesel, A. (2018).
Cognitive structure, flexibility, and plasticity in
human multitasking—An integrative review of
dual-task and task-switching research.
Psychological Bulletin, 144, 557-583.
Lakens, D. (2013). Calculating and reporting effect
sizes to facilitate cumulative science: A practical
primer for t-test and ANOVAs. Frontiers in
Psychology, 4, 863.
Levy, J., & Pashler, H. (2001). Is dual-task slowing
instruction dependent? Journal of Experimental
Psychology: Human Perception and
Performance, 27, 862-869.
Lien, M.-C., McCann, R. S., Ruthruff, E., & Proctor,
R. W. (2005). Dual-task performance with
ideomotor-compatible tasks: Is the central
bottleneck intact, bypassed, or shifted in locus?
Journal of Experimental Psychology: Human
Perception and Performance, 31, 122-144.
Lien, M.-C., Proctor, R. W., & Allen, P. A. (2002).
Ideomotor compatibility in the psychological
refractory period effect: 29 years of
PREPRINT 30
oversimplification. Journal of Experimental
Psychology: Human Perception and
Performance, 28, 396–409.
Logan, G. D. (1978). Attention in character-
classification tasks: Evidence for the
automaticity of component stages. Journal of
Experimental Psychology: General, 107, 32-63.
Logan, G. D., & Schulkind, M. D. (2000). Parallel
memory retrieval in dual-task situations. I.
Semantic memory. Journal of Experimental
Psychology: Human Perception and
Performance, 26, 1072-1090.
Lyphout-Spitz, M., Maquestiaux, F., & Ruthruff, E.
(2022). Bypassing the central bottleneck with
easy tasks: Beyond ideomotor compatibility.
Psychonomic Bulletin & Review, 29, 501-511.
Maquestiaux, F. (2012). The simultaneity of mental
acts: Contributions from PRP procedure [La
simultanéité des actes psychiques : apports du
protocole PRP]. Topics in Cognitive Psychology
[L’Année Psychologique], 112, 631-663.
Maquestiaux, F. (2017). Psychologie de l’attention
(2nd ed.). De Boeck Université.
Maquestiaux, F., Didierjean, A., Ruthruff, E.,
Chauvel, G., & Hartley, A. A. (2013). Lost
ability to automatize task performance in old age.
Psychonomic Bulletin & Review, 20, 1206-1212.
Maquestiaux, F., Hartley, A. A., & Bertsch, J.
(2004). Can practice overcome age-related
differences in the psychological refractory period
effect? Psychology and Aging, 19, 649-667.
Maquestiaux, F., Laguë-Beauvais, M., Ruthruff, E.,
& Bherer, L. (2008). Bypassing the central
bottleneck after single-task practice in the
psychological refractory period paradigm:
Evidence for task automatization and greedy
resource recruitment. Memory & Cognition, 36,
1262-1282.
Maquestiaux, F., Laguë-Beauvais, M., Ruthruff, E.,
Hartley, A. A., & Bherer, L. (2010). Learning to
bypass the central bottleneck: Declining
automaticity with advancing age. Psychology
and Aging, 25, 177-192.
Maquestiaux, F., Lyphout-Spitz, M., Ruthruff, E., &
Arexis, M. (2020). Bottleneck bypassing and
ideomotor compatibility: Ideomotor
compatibility enables automatic response
selection. Psychonomic Bulletin & Review, 27,
742-750.
Maquestiaux, F., & Ruthruff, E. (2021). Testing the
over-reliance on central attention (ORCA)
hypothesis: Do older adults have difficulty
automatizing especially easy tasks? Journal of
Experimental Psychology: General, 150, 1722-
1740.
Maquestiaux, F., Ruthruff, E., Defer, A., &
Ibrahime, S. (2018). Dual-task automatization:
The key role of sensory-motor modality
compatibility. Attention, Perception, &
Psychophysics, 80, 752-772.
Meyer, D. E., & Kieras, D. E. (1997a). A
computational theory of executive cognitive
processes and multiple-task performance: Part 1.
Basic mechanisms. Psychological Review, 104,
3–65.
Meyer, D. E., & Kieras, D. E. (1997b). A
computational theory of executive cognitive
processes and multiple-task performance: Part 2.
Accounts of psychological refractory-period
phenomena. Psychological Review, 104, 749–
791.
Meyer, D. E., & Kieras, D. E. (1999). Precis to a
practical unified theory of cognition and action:
Some lessons from EPIC computational models
of human multiple-task performance. In D.
Gopher & A. Koriat (Eds.), Attention and
performance XVII: Cognitive regulation of
performance: Interaction of theory and
application (pp. 17–88). Cambridge, MA: MIT
Press.
Mittelstädt, V., Mackenzie, I. G., & Miller, J. (2022).
Evidence of resource sharing in the
psychological refractory period (PRP) paradigm.
Journal of Experimental Psychology: Human
Perception and Performance, 48, 1279-1293.
Moors, A., & De Houwer, J. (2006). Automaticity:
A theoretical and conceptual analysis.
Psychological Bulletin, 132, 297-326.
Navarro, D. J., & Foxcroft, D. R. (2022). Learning
statistics with jamovi: A tutorial for psychology
students and other beginners. (Version 0.75)
https://doi.org/10.24384/hgc3-7p15
Navon, D., & Miller, J. (1987). Role of outcome
conflict in dual-task interference. Journal of
Experimental Psychology: Human Perception
and Performance, 13, 435-448.
Navon, D., & Miller, J. C. (2002). Queuing or
sharing. A critical evaluation of the single-
bottleneck notion. Cognitive Psychology, 44,
193-251.
Pashler, H. (1984). Processing stages in overlapping
tasks: Evidence for a central bottleneck. Journal
of Experimental Psychology: Human Perception
and Performance, 10, 358-377.
Pashler, H. (1994). Dual-task interference in simple
tasks: Data and theory. Psychological Bulletin,
116, 220-244.
Pashler, H., Carrier, M., & Hoffman, J. (1993).
Saccadic eye movements and dual-task
PREPRINT 31
interference. Quarterly Journal of Experimental
Psychology, 46A, 51-82.
Pashler, H., & Johnston, J. C. (1989). Chronometric
evidence for central postponement in temporally
overlapping tasks. Quarterly Journal of
Experimental Psychology, 41A, 19-45.
Pieczykolan, A., & Huestegge, L. (2019). Action
scheduling in multitasking: A multi-phase
framework of response-order control. Attention,
Perception, & Psychophysics, 81, 1464-1487.
Roberts, S., & Pashler, H. (2000). How persuasive
is a good fit? A comment on theory testing.
Psychological Review, 107, 358-367.
Ruthruff, E., Johnston, J. C., & Remington, R. W.
(2009). How strategic is the central bottleneck:
Can it be overcome by trying harder? Journal of
Experimental Psychology: Human Perception
and Performance, 35, 1368-1384.
Ruthruff, E., Johnston, J. C., & Van Selst, M. (2001).
Why practice reduces dual-task interference.
Journal of Experimental Psychology: Human
Perception and Performance, 27, 3-21.
Ruthruff, E., Johnston, J. C., Van Selst, M., Whitsell,
S., & Remington, R. (2003). Vanishing dual-task
interference after practice: Has the bottleneck
been eliminated or is it merely latent? Journal of
Experimental Psychology: Human Perception
and Performance, 29, 280-289.
Ruthruff, E., Miller, J., & Lachman, T. (1995). Does
mental rotation require central mechanisms?
Journal of Experimental Psychology: Human
Perception and Performance, 21, 552-570.
Ruthruff, E., Van Selst, M., Johnston, J. C., &
Remington, R. (2006). How does practice reduce
dual-task interference: Integration,
automatization, or just stage-shortening?
Psychological Research, 70, 125-142.
Schubert, T., & Strobach, T. (2018). Practice-related
optimization of dual-task performance: Efficient
task instantiation during overlapping task
processing. Journal of Experimental
Psychology: Human Perception and
Performance, 44, 1884-1904.
Schumacher, E. H., Lauber, E. J., Glass, J. M.,
Zurbriggen, E. L., Gmeindl, L., Kieras, D. E., &
Meyer, D. E. (1999). Concurrent response-
selection processes in dual-task performance:
Evidence for adaptive executive control of task
scheduling. Journal of Experimental Psychology:
Human Perception and Performance, 25, 791-
814.
Schumacher, E. H., Seymour, T. L., Glass, J. M.,
Fencsik, D. E., Lauber, E. J., Kieras, D. E., &
Meyer, D. E. (2001). Virtually perfect time
sharing in dual-task performance: Uncorking the
central cognitive bottleneck. Psychological
Science, 12, 101-108.
Stephan, D. N., & Koch, I. (2011). The role of
input–output modality in task switching.
Psychological Research, 75, 491-498.
Stephan, D. N., & Koch, I. (2015). Tactile stimuli
increase effects of modality compatibility in task
switching. Experimental Psychology, 62, 276–
284.
Stephan, D. N., Koch, I. (2016). Modality-specific
effects on crosstalk in task switching: Evidence
from modality compatibility using bimodal
stimulation. Psychological Research, 80, 935–
943.
Strobach, T. (2020). The dual-task practice
advantage: Empirical evidence and cognitive
mechanisms. Psychonomic Bulletin & Review,
27, 3-14.
Strobach, T., & Schubert, T. (2017). No evidence
for task automatization after dual-task training in
younger and older adults. Psychology and Aging,
32, 28-41.
Tombu, M., & Jolicoeur, P. (2003). A central
capacity sharing model of dual-task performance.
Journal of Experimental Psychology: Human
Perception and Performance, 29, 3-18.
Touron, D. R., & Hertzog, C. (2004). Distinguishing
age differences in knowledge, strategy use, and
confidence during strategic skill acquisition.
Psychology and Aging, 19, 452-466.
Ulrich, R., Fernández, S. R., Jentzsch, I., Rolke, B.,
Schröter, H., & Leuthold, H. (2006). Motor
limitations in dual-task processing under ballistic
movement conditions. Psychological Science,
17, 788-793.
Ulrich, R., & Miller, J. (2008). Response grouping
in the psychological refractory period (PRP)
paradigm: Models and contamination effects.
Cognitive Psychology, 57, 75-121.
Van Selst, M., Ruthruff, E., & Johnston, J. C. (1999).
Can practice eliminate the psychological
refractory period effect? Journal of
Experimental Psychology: Human Perception
and Performance, 25, 1268-1283.
Watson, J. M., & Strayer, D. L. (2010).
Supertaskers: Profiles in extraordinary
multitasking ability. Psychonomic Bulletin &
Review, 17, 479-485.
PREPRINT 32
Supplemental Material
Figure 15
Probability of bypassing across blocks for individual participants (each represented by a color line) in
Experiments 1-4