Introduction
Both impulsivity and compulsivity are psychological concepts that have
increasingly gained research interest over the last years, as they are
driving human behavior and are presumably involved in dysfunctional or
even psychopathological behavior. While initially, they were often
considered opposing ends of one spectrum , impulsivity and compulsivity
are now conceptualized as two distinct, albeit correlated, personality
traits .
Impulsivity can be characterized as a “predisposition toward rapid,
unplanned reactions to internal or external stimuli without regard to
the negative consequences of these reactions to the impulsive individual
or to others” . While there are different conceptualizations of
impulsivity, it is commonly described as multidimensional and
encompassing cognitive (e.g., attention, planning), behavioral (e.g.,
motor control, risk taking) and sometimes affective (dependency on
strong emotions) domains. The concrete operationalization mostly depends
on the impulsivity scales used, some of which will be further described
below. Compulsivity does not match the mere opposite of this definition.
Instead, compulsive behavior comprises repetitive acts carried out while
under a subjective loss of control thereover . Their implementation
serves the purpose of preventing a negative outcome (i.e., specific
feared consequences or stress and anxiety), although the agent is also
on some level aware that it goes against their longer-term goals or the
behavior is not suitable for reaching or even related to the intended
outcome. Thus, compulsivity itself can be described as the trait
facilitating this behavior.
Due to their dimensional nature, both impulsivity and compulsivity are
represented to varying degrees in the general population and can have
substantial effects, since they are by definition associated with
potential negative outcomes: Heightened impulsivity is correlated with
substance use, aggressive behavior, delinquency, gambling and risky
sexual behavior, potentially leading to personal as well as societal
costs . As of such, heightened impulsivity is associated with disorders
like substance use disorder (SUD; , borderline personality disorder and
eating disorders , primarily those associated with purging and/or
bingeing. Compulsivity is predominantly examined in the context of
psychopathology, as it is one of the key diagnostic criteria of
obsessive-compulsive disorder (OCD; , which makes non-clinical findings
scarce. A longitudinal study in Switzerland found a one-year prevalence
of about 15% of subclinical obsessive-compulsive symptomatology in 591
subjects . Compulsivity can, among others, influence eating
(facilitating obesity) and addictive behavior (e.g., substance use,
internet use) even in healthy people, thus leading to distress.
Moreover, the healthcare and socioeconomic costs of compulsive disorders
are likely transferable to subclinical compulsive behavior.
Nevertheless, there is no clear distinction between “impulsive” and
“compulsive” disorders. An overlap can be seen on a neuronal level, as
alterations in inhibitory cortical-striatal-thalamic-cortical circuits
have been found in both OCD and attention-deficit-hyperactivity disorder
, which is mostly associated with impulsivity . In both OCD and various
substance and behavioral addictions, compulsivity is associated with
dysfunctional reward and punishment processing, learning and flexibility
in favor of symptom-related stimuli that can be attributed to altered
frontostriatal and limbic activation . Phenontypically, patients often
show increases in both impulsivity and compulsivity in different
disorders such as OCD and addictions . Furthermore, some etiological
models propose e.g. substance use to originate in impulsive behavior,
which then turns compulsive as the SUD chronifies . Both characteristics
have been found to drive dysfunctional behavior such as compulsive
exercise, overeating and alcohol use in parallel , but their interaction
has so far not been resolved. Thus, transdiagnostic research is highly
warranted .
Both impulsivity and compulsivity can not only be observed in complex
behavior, but also on a basic behavioral level, as they involve weakened
motor inhibition and are defined by premature or repetitive responses,
respectively. Of particular interest here is their relationship to
response inhibition, which describes the act of withholding a prepotent
response in order to adapt to the present context . As such, response
inhibition plays an important role in goal-directed behavior . The most
prominent tasks to assess response inhibition are the Go/Nogo and the
stop-signal-task (SST). Although the paradigms have been found to
measure slightly different aspects of response inhibition (i.e., action
selection vs. preparation of stopping) and associated neuronal processes
, both are used to examine the underlying inhibitory performance. In
Go/Nogo paradigms, participants are asked to respond as fast as possible
to a specific stimulus (Go) but withhold their response to a different
stimulus of the same modality (Nogo). Similarly, in SST, participants
are asked respond quickly to a stimulus (Go) but stop the initiated
response if another stimulus is presented subsequently (stop signal).
Thereby response inhibition has been found to activate fronto-striatal
areas, primarily the dorsal anterior cingulate cortex, right inferior
frontal cortex (rIFG), presupplementary motor areas (pre-SMA), the
dorsal and ventral prefrontal cortex and basal ganglia . further propose
successful motor inhibition to be a top-down process driven by signaling
from the rIFG to the pre-SMA. Using electroencephalographic (EEG)
methods, several event-related potentials (ERP) have been linked to
cognitive processes connected to response inhibition. Firstly, the N2 is
a negative deflection peaking 200 ms after stimulus onset and presumably
generated frontally in the midcingulate cortex and ventral and
dorsolateral prefrontal cortex . In Go/Nogo tasks, it is mostly
associated with novelty, or surprise (Wessel 2012), and conflict
monitoring . The N2 is followed by a positive peak in the time-window
around 300 ms (P3) which can be further disentangled into the
frontocentral P3a and the parietal P3b. For the former, some assume the
signal to mirror bottom-up attentional orienting to potentially
significant or salient events , while others associate it with the
pre-SMA and motor response inhibition . The P3b is assumed to reflect
top-down processes of attention allocation and updating of working
memory to facilitate response selection .
Combining the aforementioned measures, healthy individuals have been
found to make more errors in Go/Nogo tasks when scoring higher on
impulsivity measures , while other researchers have not found such an
association . These findings translate to the neuropsychological
correlates of response inhibition in respective tasks: Higher
impulsivity is often associated with reduced P3 and enlarged N2
amplitudes . As for compulsivity, higher scores are related to more Nogo
commission errors in subclinical samples . However, most findings are
reported in relation to OCD, where patients are found to show longer
reaction times , lower task performance and altered N2 and P3 signals in
Go/Nogo paradigms. Similarly, performance on SST appears to be related
to OCD and its symptom scores . However, it is unclear whether this is
related to compulsivity or to other confounding measures, such as
impulsivity, or if alterations rely on clinical impairments.
As stated before, impulsivity and compulsivity are two distinct
features, yet share significant overlap in their neurocircuitry,
associated neurocognitive abnormalities and influences on dysfunctional
behaviors or even clinical impairments . Thus, further investigation of
their interaction is needed to understand their impact. Moreover, the
effect of impulsivity and compulsivity might have non-linear qualities
(see for example ) which could contribute to the partial null findings
reported above, as a potential relationship may not be apparent in
linear analyses.
Accordingly, this study aimed to determine how impulsivity and
compulsivity are dimensionally related to response inhibition and its
electrocortical correlates in a visual Go/Nogo task in a non-clinical
sample of adults. We explored the questions whether higher self-reported
impulsivity is linked to lower task performance, higher amplitude of
Nogo N2, and reduced amplitude of Nogo P3. Second, as the association
between compulsivity and response inhibition in Go/Nogo tasks has rarely
been studied outside of clinical populations, we opted for an
exploratory analysis of the effect of self-reported compulsivity on task
performance and event-related potentials (Nogo N2 and Nogo P3). Third,
we examined how self-reported impulsivity and compulsivity interact in
influencing task performance as well as event-related potentials (Nogo
N2 and P3). These associations were studied in a linear as well as a
non-linear fashion.
Methods
Participants
Data was collected from 253 participants from the general Dresden area.
Inclusion criteria were: age 18-45 years, normal or corrected-to-normal
vision and German as a native language. Participants were excluded if
they reported history of neurological disease or head trauma; lifetime
diagnosis of bipolar disorder, borderline personality disorder,
psychotic episodes or severe alcohol use disorder; acute eating disorder
or severe episode of major depression; taking psychotropic substances
within the last 3 months; lifetime use of illicit substances of more
than twice a year or lifetime use of cannabis of more than twice a
month. One participant was excluded from further analyses due to poor
task compliance (multiple responses) and two participants due two
subsequent detection of exclusion criteria (regular Cannabis use in the
past and German as a second language). Thus, the final sample consisted
of 250 participants (49% female; age M = 25.16, SD =
5.07; education level: 94% high school or higher; self-reported
lifetime diagnosis of any mental disorder: 12%).
The fit of the models obtained in the subsequent analyses was tested on
a set of 43 subjects from the ongoing follow-up project (43% female;
age M = 24.70, SD = 5.96; education level: 89% high
school or higher). Participants did not meet diagnostic criteria for any
lifetime or current mental disorders; other inclusion and exclusion
criteria were identical to the subject group above.
Data analysis for this report was preregistered under
https://osf.io/d4ezm/, where data and analysis routines are also
accessible. The project has been approved by the ethics committee at the
University Hospital Carl Gustav Carus, TUD (EK 372092017) and was
conducted in accordance with the ethical guidelines of the Declaration
of Helsinki. All participants gave informed consent and received
financial compensation or course credit for participation. The study is
part of a larger research project which assessed different cognitive
control functions in relation to impulsivity and compulsivity.
Procedure and measures
Go/Nogo task
All participants completed the Go/Nogo task as part of an EEG session in
the lab. In addition to the Go/Nogo task, they completed other EEG tasks
as well as a neuropsychological test battery at a first lab appointment
and ecological momentary assessment, which will not be reported here.
The Go/Nogo task consisted of 256 trials and was divided into two
blocks. Each trial started with a white circle presented on a grey
background for 200-500 ms. At the center of the circle either a green
square was presented as go stimulus (75% of all trials) and
participants were instructed to respond as quickly as possible with
their right index finger, or a red square was shown as nogo stimulus
(25% of all trials) and participants were asked to withhold their
response. Stimuli were presented for 500 ms and were separated by a
variable inter-stimulus interval of 400-1000 ms. Nogo trials could
immediately follow each other or were separated with up to five go
trials.
Task performance in the Go/Nogo task was measured by nogo accuracy,
reaction time (RT), and the inverse efficiency score (IES) as first
introduced by : For every participant, the mean RT for correct responses
was normalized by the proportion of correct responses (PC), i.e.,IES = \(\frac{\text{RT}}{\text{PC}}\). This allows for inclusion
of speed (RT) as well as accuracy (PC) in a combined score.
Personality scales
Impulsivity.
Impulsivity was measured with a German translation of the 11th version
of the Barratt Impulsiveness Scale (BIS-11; . Its 30 self-report items
comprise attentional, motor, and non-planning impulsiveness. We used its
sum score as it has shown good internal consistency (α = .83; . Further
the UPPS Impulsive Behavior Scale was used . Its 59 items yield separate
scores for the traits urgency, lack of premeditation, lack of
perseverance and sensation seeking. As Cronbachs’s α is only reported
for the separate subscales, lying between .8 and .85, we refrained from
using the UPPS sum score.
Compulsivity.
The Obsessive-Compulsive Inventory-Revised (OCI-R; is a self-report
measure of obsessive-compulsive symptoms, namely washing, checking,
doubting, ordering, obsessing (i.e., having obsessional thoughts),
hoarding, and mental neutralizing. Its 18 items result in sum score with
good internal consistency (α = .85), which was used here.
EEG recording and data reduction
EEG was recorded with Ag/AgCl electrodes from 61 sites of equidistant
electrode montage (Easycap GmbH, Breitbrunn, Germany) as well as from
three external positions: approximately 2 cm below each eye to record
eye movements and at the lower back to record the electrocardiogram. The
EEG was amplified with two 32-channel BrainAmp amplifiers (Brain
Products GmbH, Munich, Germany), a sampling rate of 500 Hz, and
referenced to an electrode next to FCz. Offline data analysis was
performed with MATLAB R2021a and EEGlab (Delorme & Makeig, 2004) using
the high performance computing system (HPC) at the TU Dresden.
Continuous data was filtered (0.1 – 30 Hz) and subjected to artifact
removal in an adaptive fashion, removing between a single and 10 % of
trials in epochs of 1.5 s. Data were submitted to an adaptive mixture
independent component analysis (AMICA) and components containing
eye-movement and cardioballistic artifacts were removed. Behaviorally,
trials with reaction times outside the range of 70 – 600 ms were
rejected. EEG data was then referenced to average reference.
Stimulus-locked event-related potentials (ERP; N2, P3a, and P3b) were
analyzed for correct go and nogo trials. The N2 and P3a were determined
in a frontocentral electrode cluster (FCz, FC1, FC2, Cz). The N2 was
defined as the most negative value in the time window 230 to 270 ms
after stimulus onset and P3a as the average amplitude from 300 to 380 ms
after stimulus onset. The P3b was obtained as average amplitude in the
time window from 320 to 450 ms in a parietal electrode cluster (CPz,
CP1, CP2, Pz, P1, P2). Baseline correction was applied in the 200 ms
prior to stimulus onset.
Statistical analysis
Linear analyses
All further analyses were computed with R . The effects of impulsivity
and compulsivity on the dependent variables IES, RT and Nogo accuracy as
well as amplitudes of the Nogo N2, Nogo P3a and Nogo P3b were analyzed
via robust linear regression. For BIS-11 and OCI-R, the respective sum
scores were used as regressors in a multivariate model to uncover
possible interaction effects. The UPPS scales were analyzed in a
backwards stepwise fashion. Subscales were selected on their influence
on the robust final prediction errors (RFPE), where those reducing RFPE
were then analyzed in a robust linear regression model. The results of
all linear correlation and regression analyses were tested for
significance with correction of the false discovery rate as recommended
by .
Gradient boosting regression trees
Gradient boosting regression trees (GBRT) are a type of decision tree, a
machine-learning method used to uncover non-linear relationships between
variables . Decision trees perform binary splits in the predictor space
in a way that best minimizes residuals. Every split creates new segments
(so called nodes), wherein the mean of the response variable serves as
the predicted value. GBRT then, rather than computing a single tree,
combine an ensemble of various trees based on the same training data
set. The individual trees are grown sequentially, each modifying the
previous version (boosting). Ultimately, model performance is enhanced
by focusing on the observations that proved difficult to predict in the
previous iterations. GBRT have several tuning parameters: Firstly, the
number of trees B and the number of splits (interaction depth)d , which can lead to overfitting if too high. This risk is
controlled by the learning rate or shrinkage parameter λ. We
optimized the tuning parameters for each analysis through a grid search
(see the provided R code for the range of values).
Model fit
Both the linear and the non-linear models assess the influence of
impulsivity and compulsivity on different ERP and the IES. To test their
accuracy, we employed two methods: Firstly, each model was created via
nested cross-validation. Here, two cross-validation processes are nested
into each other to allow model selection and assessment of prediction
performance on the same data set. In order to do this, we divided the
data into k subsets or outer loops. Nested into each of these
were inner loops, which were used via cross-validation to optimize GBRT
tuning parameters. The fit of the optimized model was then assessed in
the outer layer. This should prevent an overly optimistic generalization
error as model training and validation are separated. Second, the models
were run on a test set of 43 completely separate participants. We
estimated model performance through the root square mean error
(RMSE ) both in the nested cross-validation process as well as for
the test set prediction. These were then compared to the RMSE of
the linear regression analyses. Lastly, the regression tree analyses
were repeated in permutation tests with randomly assigned outcome
variables to test whether the models perform above chance level.P values for the permutation test were calculated as the
proportion of RMSE resulting from the permutations (N =
1000) that were lower than the actual RMSE .
Results
Behavioral results
Participants achieved an average Nogo accuracy of 0.86 (SD =
0.11) and a Go reaction time of 282 ms (SD = 30), resulting in an
IES of 293 (SD = 28). See Table 1 for more details and Table 2
for questionnaire data.
Table 1. Behavioral performance