· web viewwhether an appropriate decision is made will depend partly on the criterion that the...
TRANSCRIPT
Running head: SOMATOSENSORY DECISION-MAKING 1
Effects of Learning on Somatosensory Decision-Making and Experiences
Akib-ul-Huque, Ellen Poliakoff and Richard J. Brown
Faculty of Biology, Medicine and Health, University of Manchester, Zochonis Building, Brunswick Street, Manchester M13 9PL
Address for correspondence: Dr Akib-ul-Huque, University of Dhaka, Bangladesh, Dhaka-1000, Bangladesh. E-mail: [email protected]
Accepted for publication in Journal of Experimental Psychology: General, 21 st July 2017
Running head: SOMATOSENSORY DECISION-MAKING 2
Abstract
Operant conditioning has been shown to influence perceptual decision making in the auditory
and visual modalities but the effects of conditioning on touch perception are unknown. If
conditioning can be used to reduce the tendency to misinterpret somatic noise as signal
(tactile false alarms), there may be the potential to use similar procedures in the treatment of
excessive physical symptom reporting in clinical settings. We explored this possibility in four
experiments investigating whether the false alarm (FA) rate in a somatosensory signal
detection task (SSDT) could be altered with operant conditioning, and whether the resultant
learning would transfer to other sensory decisions. In Experiments 1a and 2a, non-clinical
participants were rewarded for hits and punished for misses on the SSDT, with a view to
increasing their FA rate. In Experiments 1b and 2b, participants were rewarded for correct
rejections and punished for FAs, with a view to decreasing their FA rate. Control participants
received no treatment in Experiments 1a and 1b and underwent sham training in Experiments
2a and 2b. As predicted, operant conditioning increased (Experiments 1a, 2a) and decreased
(Experiments 1b, 2b) FAs on the SSDT. Training effects did not transfer to an unrelated
somatosensory task and there was only weak evidence for transfer to an auditory task in
Experiment 2a. Auditory and tactile FAs correlated positively in the baseline phase. The
results indicate that the tactile false alarm rate is trainable, but that the conditioning effect
does not transfer across sensory decisions with this brief training paradigm.
Keywords: signal detection, operant conditioning, transfer of perceptual learning, tactile
perception, medically unexplained symptoms
Running head: SOMATOSENSORY DECISION-MAKING 3
Effects of Learning on Somatosensory Decision-Making and Experiences
Signal detection theory posits that both stimulus and perceiver characteristics determine
perception (Green & Swets, 1966). Sensitivity and response criterion are two observer
characteristics that are independent but which greatly influence perceptual decisions, such
that the same stimulus produces different perceptual experiences in each case. Disturbances
in perceptual decision making are thought to play an important role in the development and
maintenance of certain clinical phenomena, including psychotic experiences, body image
problems and excessive somatic symptom reporting (Albert III, Kelley, & Corlett, 2016;
Brown et al., 2012; Slade, 1994). Signal detection theory provides a useful framework for
understanding and studying these phenomena. In order to maintain our health, for example,
we need to be able to detect sensory signals coming from our body and then decide whether
any variation in these might indicate disease and warrant further investigation. Whether an
appropriate decision is made will depend partly on the criterion that the individual adopts for
judging whether a sensation represents meaningful signal or irrelevant noise. If the individual
sets a very strict criterion, then they may miss a possible threat to their health; if they set a
very liberal criterion, they risk interpreting normal perceptual variations as potentially
dangerous, generating alarm and unnecessary help-seeking as a result.
A variation on this idea is central to so-called somatosensory amplification (Barsky,
1992; Rief & Barsky, 2005), which has been linked to increased health anxiety, excessive
symptom reporting and the experience of “medically unexplained” or functional somatic
symptoms (Witthöft, Fischer, Jasper, Rist, & Nater, 2016). The latter constitute up to three
quarters of the symptoms encountered in hospital settings (Körber, Frieser, Steinbrecher, &
Hiller, 2011) and are a major societal problem. At present, the mechanisms of excessive
symptom reporting and functional symptoms are poorly understood and existing treatments
are inadequate (van Dessel et al., 2014).
Running head: SOMATOSENSORY DECISION-MAKING 4
From a signal detection perspective, functional symptoms are akin to somatosensory
false alarms (cf. Brown, 2004), where bodily noise is misinterpreted as disease signal.
Although numerous factors are likely to contribute to this process (e.g., trait anxiety/negative
affectivity, self-focus, avoidance, rumination; Rief & Barsky, 2005), there is some evidence
that physical symptom reporting correlates with a more general tendency to false alarm
during the detection of somatic signals (Brown, Brunt, Poliakoff, & Lloyd, 2010; Brown et
al., 2012; Katzer, Oberfeld, Hiller, & Witthöft, 2011). Studies also suggest that problems
perceiving the internal bodily state (i.e., interoceptive dysfunction) may influence medical
and psychological problems, such as gastrointestinal and cardiac disease, anxiety disorders,
somatoform disorders and chronic pain (Cameron, 2001). If problems with somatic
perception are central to functional symptoms and excessive symptom reporting, then
ameliorating any perceptual difficulties may be an important part of treatment for these
conditions. Consistent with this, interoceptive accuracy training significantly reduced
symptom reports in patients with functional complaints in one recent study (Schaefer, Egloff,
Gerlach, & Witthöft, 2014).
Numerous studies have found that it is possible to manipulate signal detection
performance using training methods such as operant conditioning (e.g., Johnstone & Alsop,
2000; Lie & Alsop, 2009; Szalma, Hancock, Warm, Dember, & Parsons, 2006). The vast
majority of studies have considered the effects of training in the auditory and visual
modalities, however, and there are few training studies on somatosensory perception (e.g.,
Brown et al., 2010; McKenzie, Lloyd, Brown, Plummer, & Poliakoff, 2012; Mirams,
Poliakoff, Brown, & Lloyd, 2012, 2013). It therefore remains unclear whether it is possible to
train somatic perception, although learning theories (Tazaki & Landlaw, 2006), chronic pain
research (Rief & Broadbent, 2007), and child illness behavior studies (Walker & Zeman,
Running head: SOMATOSENSORY DECISION-MAKING 5
1992) suggest that operant conditioning might contribute to the perception and reporting of
somatic symptoms.
In this paper, we describe four experiments investigating the effect of operant
conditioning on perception in the tactile modality. As a proximal sense with spatial referents
on the body, touch is more directly related to somatic experience than audition or vision, but
is much easier to manipulate than the perception of internally generated stimuli (i.e.,
interoception). We sought to establish whether reward and punishment could be used both to
increase and decrease the false alarm rate on the Somatic Signal Detection Task (SSDT;
Lloyd, Mason, Brown, & Poliakoff, 2008), which has previously been linked to physical
symptom reporting. We also sought to establish whether this training would transfer to other,
unrelated stimuli and tasks (Bratzke, Schröter, & Ulrich, 2014; Liu & Weinshall, 2000),
which would give some indication of the potential clinical utility of the training paradigm..
Experiments 1a and 1b: Method
In Experiments 1a and 1b, we attempted to condition participants to report more or
fewer false alarms on the SSDT respectively using a within-subjects design. We predicted
that rewarding and punishing certain responses on the task would lead to an enduring change
in the false alarm rate post-training, and that this effect would transfer to perception on a
different body-relevant task, the spontaneous sensation (SpS) test (Michael & Naveteur,
2011).
Overview
Experiments 1a and 1b were run in parallel (Figure 1). The control condition was
identical in each case and consisted of eight blocks of the SSDT. In the experimental
condition of Experiment 1a, a conditioning procedure was introduced on blocks 3-6 of the
SSDT aimed at increasing the false alarm rate. In Experiment 1b, blocks 3-6 of the SSDT in
the experimental condition incorporated conditioning aimed at decreasing the false alarm
Running head: SOMATOSENSORY DECISION-MAKING 6
rate. The experimental condition in both cases was otherwise identical to the control
condition, where no training was given. Participants with relatively low (<0.16) and high (>=
0.16) false alarm rates1 were automatically allocated to either Experiment 1a or 1b
respectively, depending on their SSDT performance in the first two blocks of the
experimental condition; this was to eliminate the potential impact of ceiling and floor effects
in Experiments 1a and 1b respectively. Neither the researchers nor the participants knew who
was in which study at the time of participation. Participants completed the two conditions on
separate days at least a week apart.
Design
A mixed design was used for Experiments 1a and 1b, with condition (experimental vs.
control) and phase (baseline vs. manipulation vs. follow-up for SSDT response outcomes;
baseline vs. follow-up for SpS) as within-group variables. Session order (experimental
session first vs. control session first) was included as a between-group variable to test for any
enduring changes in signal detection performance related to the training. The dependent
variables were the SSDT (i.e., hit rate, false alarm rate, response bias [c], and sensitivity [d´])
and SpS response outcomes (total number of SpSs). In Experiment 1a, we predicted that
there would be an increase in both hits and false alarms, resulting in a more liberal response
bias, in the training condition compared to the control condition during both the manipulation
and follow-up phases. We also predicted that there would be an increase in spontaneous
sensations on the SpS test in the training condition compared to the control condition. In
Experiment 1b, we predicted that there would be a decrease in hits, false alarms and
spontaneous sensations in the manipulation and follow-up phases of the experimental
condition compared to the control. We did not expect any impact of the training on perceptual
sensitivity in either study. Given the potential impact of sleepiness and state anxiety on tactile
signal detection (Gillberg & Åkerstedt, 1998; Malow, 1981) we also measured these variables
Running head: SOMATOSENSORY DECISION-MAKING 7
to establish whether the two conditions were comparable in these respects. In addition, we
measured the relationship between somatic symptom reporting and SSDT false alarm rates to
determine the comparability of our sample with previous studies (Brown et al., 2012; Katzer
et al., 2011).
Participants received either 12 academic credits or £15 as compensation. We planned to
recruit 78 participants in total, giving adequate power for both studies assuming the cut-off
false alarm rate of 0.16 resulted in equal allocation in each case. This sample size was
determined using G*Power (Faul, Erdfelder, Lang, & Buchner, 2007) with power = .90,
alpha = .05 (two-tailed), and effect size = .50. The inclusion criteria were: aged 18-40 years
and ability to understand English instructions. The exclusion criterion was having any
medical condition that might affect the sense of touch. Ethical approval was given by the
relevant University committee.
Materials, Measures and Procedures
All study materials, measures and procedures were the same for Experiments 1a and 1b
apart from the conditioning procedure.
SSDT
Thresholding. The vibrotactile perceptual threshold of each participant was first
determined using a forced-choice adaptive procedure. A centrally aligned computer monitor
was used to provide instructions and present a green arrow (962×722 pixels) that cued the
start of each of two 1020ms consecutive intervals that comprised a trial. The arrow appeared
in the middle of the monitor for 250ms, pointing downwards towards the participant’s finger.
The first and second arrows contained the numbers 1a and 1b respectively to designate which
interval was being presented. In each trial, a 20ms tactile pulse (painless vibration) was
delivered through a bone conductor (vibrating device) by amplifying the sound output from a
computer using a custombuilt amplifier (Dancer Design, Merseyside, United Kingdom). A
Running head: SOMATOSENSORY DECISION-MAKING 8
double-sided adhesive circle was used to attach the bone conductor to the pad of the
participant’s non-dominant index finger. The bone conductor was mounted on a foam cube
alongside a 5mm red LED, which was not used during the thresholding. The vibration
appeared randomly in the middle of one of the two intervals. After each trial, participants
were asked to indicate using a computer keyboard which of the two intervals contained a
short tactile vibration. Participants were prompted to rest after 80 trials but could do so at any
time. During both the thresholding and SSDT proper, participants wore headphones that
delivered white noise to mask background noise and the sound of the bone conductor.
A parameter estimation by sequential testing (PEST) computer algorithm was used to
determine the participant’s tactile threshold for use in the SSDT. The PEST began with a
strong, easily perceptible, but painless vibration of 31.16 dB that was equal to a pressure of
274 m/s. A Wald SPRT [N(c) (number of correct responses) - Pt. N (T) (probability threshold
value (0.75) multiplied by current trials completed) W (W’s limits were: 1 to -1)] was used to
change the vibration strength. Selection of the vibration level depended on the responses
given on all trials since it reached its current intensity level. In this process, the threshold set
for each participant was the intensity of vibration that they correctly detected in
approximately 75% of the trials; this is significantly above chance level performance (50%)
but ensures that the stimulus in the main task is sufficiently ambiguous for misperception to
occur. Though the number of trials required for the determination of threshold varied between
individuals, the computer algorithm was programmed to run a maximum of 250 trials. If the
maximum was reached, the average vibration intensity of the last 50 trials was taken as the
participant’s threshold level. E-Prime (Psychology Software Tools, Inc., Pittsburgh, PA) was
used to deliver stimuli and record responses.
Main task. The same apparatus was used for the main SSDT task, which comprised
blocks of 80 trials consisting of four, randomly interspersed trial types (touch only, light only,
Running head: SOMATOSENSORY DECISION-MAKING 9
light and touch and no stimulus), each lasting 1020ms. The start of each trial was signalled by
the green arrow, presented for 250ms. In touch only trials, a threshold level tactile vibration
was presented for 20ms in the middle of the trial interval. In light only trials, the LED light
was flashed for 20ms in the middle of the trial interval. In light and touch trials, both the
vibration and light flash appeared simultaneously. In no stimulus trials, nothing was
presented. After each trial, participants indicated whether they felt any vibration by pressing
1, 2, 3, and 4 on the keyboard number pad corresponding to definitely yes, maybe yes, maybe
no, and definitely no respectively2. The light was included to maximise the false alarm rate
(Lloyd et al., 2008) but was otherwise unrelated to the purpose of the experiment. There were
12 practice trials. Participants were kept naive about the significance of the light and
informed that the vibration would not be present in all trials. Both control and experimental
conditions consisted of eight blocks. The first two blocks were the baseline blocks. The
average false alarm rate in light present trials in these blocks of the experimental condition
was used to determine whether the participant had a high or low false alarm rate and therefore
entered into Experiment 1a or 1b. Blocks 3-6 were the manipulation blocks that delivered
reinforcement and punishment in the training condition but no manipulation in the control
condition. Blocks 7 and 8 were follow-up blocks consisting of regular SSDT trials without
reward or punishment. Participants were encouraged to rest briefly between blocks.
Conditioning. During blocks 3-6 of Experiment 1a, participants in the experimental
condition received reward and punishment with a view to conditioning the “yes” response.
Participants received 10 points in half of the hit trials (i.e., where they correctly indicated a
stimulus was present) and lost 10 points in half of the miss trials (i.e., where they incorrectly
indicated a stimulus was absent). In Experiment 1b, participants in the experimental condition
received reward and punishment during blocks 3-6, with a view to conditioning the “no”
response. Participants received 10 points in half of the correct rejection trials (i.e., where they
Running head: SOMATOSENSORY DECISION-MAKING 10
correctly indicated a stimulus was absent) and lost 10 points in half of the false alarm trials
(i.e., where they incorrectly indicated a stimulus was present).
Before starting the conditioning trials, participants were informed that they would
receive 1p for every point they accumulated during this part of the experiment. A random
selection process (built into the E-Prime programme) was used to select half of the hits and
misses in Experiment 1a, and correct rejections and false alarms in Experiment 1b, to
condition them with reward and punishment respectively. This was done to control
participants’ expectancy about specific trials or responses that might result in consequences.
The random selection process also allowed for a variable ratio training schedule, which tend
to produce more enduring effects (Reynolds, 1975). Participants were told whether they had
won or lost points by 3-second feedback messages on the computer monitor, which also
indicated their running point total (e.g., “You have won 10 points. Your total score is 300”;
“You have lost 10 points. Your total score is 280”). Different colours were used to present the
messages (yellow = win; red = loss). Instructions about the possibility of winning or losing
points and obtaining corresponding feedback messages were given at the start of this phase.
At the end of each experiment, participants received 1p for every point that they accumulated
during this phase (in cases where they had fewer than zero points they did not receive
anything apart from the usual honorarium for taking part). Participants were encouraged to
win as many points as possible to maximise motivation on the task.
Data processing. SSDT data were processed to derive four SSDT variables (hit rate, false
alarm rate, bias and sensitivity) for each block using standard formulae with the log-linear
correction (Snodgrass & Corwin, 1988). For each statistic, the means of Blocks 1a and 1b
were calculated to get baseline measures, the means of Blocks 3, 4, 5, and 6 were calculated
to obtain measures for the manipulation phase, and the means of Blocks 7 and 8 were
calculated to obtain measures for the follow-up phase. Means for the experimental and
Running head: SOMATOSENSORY DECISION-MAKING 11
control conditions were calculated separately, as were those for light present and light absent
trials. We present the data separately for Experiments 1a and 1b below.
SpS test. An adapted version of the protocol and procedure described by Michael and
Naveteur (2011) was used to measure spontaneous sensations. Participants were asked to
relax with their non-dominant hand palm down on a piece of smooth white A4 paper. At this
time, the wrist of their other hand remained at resting position on the corresponding thigh
under the table so that they could not see it. Participants were then asked to focus their
attention on their non-dominant hands for 10 seconds, indicated by verbal start and stop
signals from the experimenter. Immediately after the stop signal, participants were given a
standard picture of a non-dominant hand to mark any areas where they felt sensations during
the 10-second attention period, and to write names (e.g., tingling, throbbing) for these. The
SpS test was repeated twice, making three trials overall in a phase, which were then
combined to get the average number of SpSs reported in the baseline and follow-up phases of
the control and experimental conditions. There was a practice trial on the first administration
of the SpS test to familiarize participants with the task.
Karolinska sleepiness severity scale (KSS). Akerstedt and Gillberg (1990) developed
this one-item scale to measure current level of sleepiness. The scale ranges from 1 (very
alert) to 9 (very sleepy, great effort to keep awake, fighting sleep). A score of 7 or more
indicates excessive sleepiness.
Spielberger state-trait anxiety inventory (STAI) – state short form. This six-item
scale assesses how anxious someone is at that moment (Marteau & Bekker, 1992). There are
four response options for each item ranging from 1 (not at all) to 4 (very much). Total scores
vary between 6 and 24 with higher scores indicating elevated anxiety. The scale is sensitive
to fluctuations in state anxiety and it has acceptable reliability and validity (Marteau &
Running head: SOMATOSENSORY DECISION-MAKING 12
Bekker, 1992; Tluczek, Henriques, & Brown, 2009). The average internal consistency
coefficients were .70 and .75 in Experiments 1a and 1b respectively.
Patient health questionnaire-15 (PHQ-15). The PHQ-15 (Kroenke, Spitzer, &
Williams, 2002) was used to assess somatic symptom severity. Respondents indicated how
bothered they had been about 15 physical symptoms during the last 4 weeks (not bothered,
bothered a little, or bothered a lot corresponding to scores of 0, 1, and 2 respectively). The
scale is known to have excellent psychometric properties (Zijlema et al., 2013). Internal
consistency was .83 and .75 in Experiments 1a and 1b respectively.
Procedure
The order of conditions was determined randomly for each participant. Both sessions
were carried out in a quiet, light-attenuated room. At the start of each session, participants
were asked to remove any jewellery from the fingers and wrists of their non-dominant hand.
Some participants wore wristbands that could not be removed without cutting or breaking
them; in these cases, they were asked to ensure that these bands were worn in both sessions.
Approximately 15 seconds before the tasks, participants were asked to clean their hands with
alcohol-based hand rub. The control condition started with the baseline SpS test followed by
the baseline SSDT (two blocks of 80 trials), manipulation SSDT (four blocks of 80 trials with
no training), follow-up SpS test, and follow-up SSDT (two blocks of 80 trials). The
procedures for the experimental conditions were identical to that of the control condition
apart from the manipulation SSDT, where the conditioning manipulations (i.e., operant
training) were delivered. Participants responded to the PHQ-15 online (at least a week) before
participating in the first session. Sleepiness was measured after the SSDT threshold task, after
the baseline phase, and after the manipulation phase. State anxiety was measured before the
baseline and after the manipulation phase.
Experiment 1a: Results
Running head: SOMATOSENSORY DECISION-MAKING 13
Participants
Thirty three (65.2%) female and 19 (36.53%) male student (n = 50) and staff (n = 2)
volunteers of the University of Manchester, aged between 18 and 38 years (M = 22.15, SD =
4.22) had relatively low false alarm rates and were allocated to Experiment 1a. Most
(84.62%) were right-handed on the Edinburgh Handedness Inventory (EHI; Oldfield, 1971).
Data Preparation
Data were initially examined for consistency with the assumptions of mixed Analysis of
Variance (ANOVA) for the SSDT and SpS test and of dependent t-test for sleepiness and
state anxiety. Standard correction procedures (e.g., replacing extreme outliers, transforming
scores, Huynh-Feldt correction; Field, 2009) were followed where needed. Due to the
violation of parametric assumptions, Wilcoxon signed-rank test was used to examine the
difference in change scores in state anxiety between the conditions. Four participants with an
extremely high (> 95%) baseline hit rate in the experimental condition were excluded from
the analysis on the grounds that the task had been insufficiently ambiguous for these
individuals. Two more participants were excluded for falling asleep during the experiment.
The final ANOVAs were therefore carried out with 46 individuals with a low false alarm rate
on the SSDT, of whom 22 completed the experimental condition first. As changes in false
alarm rates and spontaneous sensations were our primary outcomes, an alpha value of .05 was
used for these variables. A Bonferroni corrected alpha of .01 was adopted for the remaining
analyses. Analyses pertaining to the light were performed to identify any interactions with
possible training effects. As will become apparent, there were very few significant
interactions between the light and training variables in any of the studies in this paper; these
are reported below where relevant. As all other analyses involving the light are peripheral to
our main purpose they are presented as supplementary materials and not discussed further
here.
Running head: SOMATOSENSORY DECISION-MAKING 14
Effects of Conditioning on SSDT
Descriptive statistics and phase by condition interactions for the SSDT outcomes are
presented in Tables 1 and 2 and Figure 2. In each case, the main effects of phase and
condition for false alarms, hits and response bias were subsumed under significant condition
by phase interactions so are not reported here; full details of the analyses are presented in the
supplementary materials. In describing the results, we focus mainly on those that pertain to
likely training effects.
--- INSERT TABLES 1 AND 2 AND FIGURE 2 ABOUT HERE ---
False alarms, hit rate and response bias. Both the false alarm and hit rates were
significantly higher during and after the conditioning procedure than before, and were
significantly higher in the experimental than the control condition in both the manipulation
and follow-up phases. Participants in the training condition were significantly more likely to
say yes during and after the conditioning procedure than before, and significantly more so
than the control participants in the experimental and follow-up phases. Control participants
became significantly less likely to say yes over time, with a corresponding decrease in false
alarms and hits. There was a significant light x phase x condition interaction, F(2, 88) = 4.60,
p = .01, η2p = .10, for the response criterion variable, which was attributable to a minor
variation for the control condition when comparing the manipulation and follow-up phases
(see supplementary analyses); this was not considered theoretically important and is not
discussed further here.
Sensitivity. The main effect of phase was significant, F(1.78, 78.19) = 8.26, p = .001,
η2p = .16. Bonferroni corrected post-hoc tests revealed that participants’ sensitivity
(regardless of condition) did not differ between the baseline and manipulation phases, mean
difference = .07, 95% CI [-.13, .27], p = 1.0. However, sensitivity was lower in the follow-up
Running head: SOMATOSENSORY DECISION-MAKING 15
compared to the baseline, mean difference = -.32, 95% CI [-.57, -.08], p = .007, and
manipulation phases, mean difference = -.25, 95% CI [-.42, -.08], p = .002.
Transfer of Conditioning to SpSs
Descriptive statistics are presented in Table 3. Contrary to our hypothesis, the main
effect of condition, F(1, 44) = 1.46, p = .23, η2p = .03, and interaction between phase and
condition, F(1, 44) = 1.40, p = .24, η2p= .03, were not significant. There was a significant
main effect of phase, F(1, 44) = 5.00, p = .03, η2p = .10, with a Bonferroni corrected post-hoc
test revealing that sensations were significantly more common at follow-up (M = .82, SD
= .47) than at baseline (M = .71, SD = .41), regardless of condition, mean difference = .11,
95% CI [.01, .21], p = .03.
--- INSERT TABLE 3 ABOUT HERE ---
There was a significant interaction between condition and session, F(1, 44) = 6.54, p
= .01, η2p = .13. It was found that the total number of SpSs in the experimental condition (M
= .85, SD = .42) was higher than in the control condition (M = .68, SD = .44) when the
experimental condition was the first session, mean difference = .17, 95% CI [.04, .31], p
= .01. This difference was not found when the control condition was the first session, mean
difference = .06, 95% CI [-.07, .19], p = .34 (Control: M = .79, SD = .44; Experimental: M
= .73, SD = .38).
Sleepiness and State Anxiety
Change in sleepiness (i.e., sleepiness at the follow-up – baseline phase) differed
significantly between the conditions, t(45) = 4.02, p < .001, r = .51. Participants became
sleepier in the control (M = 1.78, SE = .29) than in the experimental (M = .30, SE = .34)
condition. Change in state anxiety (i.e., state anxiety at the follow-up – baseline phase),
however, did not differ between the conditions (Mdn = .00 for both the conditions), z = -.39,
p = .70, r = -.04.
Running head: SOMATOSENSORY DECISION-MAKING 16
Experiment 1b: Results
Participants
Thirty student (n = 27) and staff (n = 3) volunteers from the University of Manchester,
aged between 18 and 39 years (M = 22.97, SD = 5.07), were identified as having a high false
alarm rate and were automatically allocated to Experiment 1b. Of these, 21 (70%) were
female and 27 (90%) were right-handed on the EHI.
Data Preparation and Statistical Analysis
Data were prepared and analysed as for Experiment 1a. None of the participants had an
extremely low (<.05) or high (>0.95) hit rate in the baseline phase of the experimental
condition, indicating that they all understood the task and the vibration level was sufficiently
ambiguous. Data from both Experiments 1a and 1b were used to determine the non-
parametric (Spearman’s) correlation between SSDT false alarms, total SpSs and PHQ-15
scores in the baseline phase of the first session.
Effects of Conditioning on SSDT
Descriptive statistics and phase by condition interactions for the SSDT outcomes are
presented in Tables 1 and 4 and Figure 2.
--- INSERT TABLE 4 ABOUT HERE ---
False alarms, hit rate and response bias. Both the false alarm and hit rates were
significantly lower during and after the conditioning procedure than before, and were
significantly lower in the experimental than the control condition in both the manipulation
and follow-up phases. Participants in the training condition were significantly less likely to
say yes during and after the conditioning procedure than before, and significantly less so than
the control participants in both the experimental and follow-up phases. Participants in the
control condition exhibited more false alarms, and were more likely to say yes, when the
control session was first than when it was second. The difference in false alarms between the
Running head: SOMATOSENSORY DECISION-MAKING 17
two conditions was also significantly greater when the control session was first, as was the
tendency to say yes.
Sensitivity. The main effect of phase was significant, F(2, 56) = 17.64, p = .001, η2p
= .39. Bonferroni corrected post hoc tests indicated that sensitivity in the follow-up phase was
significantly lower than in the baseline, mean difference = -.07, 95% CI [-.10, -.03], p = .001,
and manipulation phases, mean difference = -.06, 95% CI [-.09, -.03], p = .001. Sensitivity
did not differ between the baseline and manipulation phases, mean difference = .01, 95% CI
[-.02, .03], p = 1.00. The remaining main effects and interactions did not reach significance
(Fs < 5, ps > .01).
Transfer of Conditioning to SpSs
Descriptive statistics are presented in Table 3. The main effect of phase was significant,
F(1, 28) = 8.73, p = .01, η2p = .24. Contrary to expectation, a Bonferroni corrected post hoc
test indicated that participants reported significantly more SpSs in the follow-up (M = 1.02,
SD = .71) than in the baseline phase (M = .76, SD = .55), mean difference = .26, 95% CI [.08,
.43], p = .006. There was a significant interaction between condition and session, F(1, 28) =
6.18, p = .02, η2p = .18. Bonferroni corrected post hoc tests indicated that reporting of SpSs
did not differ between the conditions when the control condition was the first session (Ms
= .78 and .60, SDs = .59 and .78 for the control and experimental conditions respectively),
mean difference = .18, 95% CI [-.11, .47], p = .21, but when the experimental condition was
the first session, participants reported significantly more SpSs in the experimental (M = 1.22,
SD = .68) than the control condition (M = .94, SD = .52), mean difference = .28, 95% CI [.03,
.54], p = .03. In the control condition, there was no significant difference between the groups
with regard to the number of SpSs reported, mean difference = -.16, 95% CI [-.60, .29], p
= .47. In the experimental condition, the participants who completed the experimental
condition in the first session reported significantly more SpSs than the participants who
Running head: SOMATOSENSORY DECISION-MAKING 18
completed the control condition in the first session, mean difference = .62, 95% CI [.11,
1.14], p = .02. The remaining main effects and interactions did not reach significance (Fs < 4,
ps > .05).
Sleepiness and State Anxiety
The difference between changes in sleepiness in the two conditions (Ms = 1.90 and
1.27, and SEs = .31 and .46 for the control and experimental conditions respectively) was not
significant, t(29) = 1.27, p = .21, r = .23. Changes in state anxiety also did not differ between
the conditions (Ms = .20 and .00, and SEs = .21 and .30 for the control and experimental
conditions respectively), t(29) = -.61, p = .55, r = .11.
Correlations Between Baseline False Alarms, SpSs and PHQ-15 Scores
The correlation coefficients of SpS with SSDT false alarm rate, rs(76) = .13, p = .28, hit
rate, rs(76) = .09, p = .42, sensitivity, rs(76) = -.004, p = .97, and response bias, rs(76) = -.15,
p = .19, were not significant at baseline. Similarly, there was no significant correlation
between SSDT false alarm rates and PHQ-15 scores, rs(76) = -.04, p = .73.
Discussion: Experiments 1a and 1b
Operant conditioning had the expected effects on false alarms, hits and response
criterion (i.e., their overall tendency to say yes vs. no) in both experiments, with these effects
manifesting both whilst reward and punishment were being delivered in the manipulation
phase and afterwards in the follow-up phase. There was no effect on sensitivity in either
study. Contrary to expectation, neither study revealed a significant effect of conditioning on
the total number of spontaneous sensations, suggesting that the training did not generalize to
subsequent somatosensory decision-making on an unrelated task.
Since false alarm rates are highly reliable (McKenzie, Poliakoff, Brown, & Lloyd,
2010) it was expected that the baseline false alarm rates would be comparable across the
conditions in each study. In both studies, however, it was found that baseline false alarm rates
Running head: SOMATOSENSORY DECISION-MAKING 19
differed significantly between the conditions, with the control condition having significantly
higher rates in Experiment 1a and lower rates in Experiment 1b. Evidently, those participants
who received training in the first session showed a carry-over of training effects to the second
session, with overall baseline response criterion appearing either relatively liberal or
relatively stringent accordingly. Although this was always possible in theory, we did not
anticipate that the effects of a brief conditioning procedure would persist for such a long
period without further training. This unwanted effect of session order suggests that a
between-subjects design may be more appropriate for research of this sort.
Another limitation of both experiments is that the conditioning may have influenced
participants’ mood and motivation during the procedure. The fact that the majority of
participants in the experimental condition won money may explain the sleepiness scores of
Experiment 1a, which indicated that participants were more alert in the experimental than in
the control condition (a similar, although non-significant, effect was also found in
Experiment 1b). Reward-related changes in mood and motivation could have systematically
affected participants’ subsequent experiences of SpSs, which might explain why participants
reported more SpSs following the training condition when it was presented first and
motivation was likely at its highest. Though no study has yet investigated possible
relationships between affect-related physiological arousal and the reporting of SpSs on this
paradigm, it is well established that affective states influence whether we notice and attend to
normal body sensations (Watson & Pennebaker, 1989). One way of dealing with this possible
confounding effect would be to introduce rewards into the control condition to make the
conditions more comparable, in a manner that creates a pleasant experience of winning
without introducing any form of learning.
One possibility is that the use of an alcohol-based hand-rub immediately prior to the
SpS procedure obscured many of the more subtle sensations in the participants’ hand by
Running head: SOMATOSENSORY DECISION-MAKING 20
inducing much less ambiguous feelings of coldness and/or tingling3. The SpS test is also
potentially problematic because the focus for participants’ responses on this task (i.e., the
non-dominant hand) is also the site of stimulation on the SSDT task. As the SSDT involves
attending to and detecting tactile stimulation in the index finger over many trials, it is likely
that prolonged attention to the hand will result in the perception of more SpSs (Michael et al.,
2012; Michael & Naveteur, 2011), consistent with the main effect of phase observed in both
studies. Fatigue in the non-dominant hand may also have contributed to this effect. An
alternative approach would be to ask participants to focus on the entire body instead of just
on the non-dominant hand to identify and report SpSs.
It is also possible that the SpS test is not the most appropriate test to detect whether
SSDT conditioning generalises. Compared to the SSDT, it is much less clear what
participants should attend to during the SpS test and when such experiences might arise. With
this in mind, a task more similar to the SSDT in a different sensory modality might be more
suitable to study the presumed transfer of any conditioning effects.
Contrary to the findings of previous SSDT studies (Brown et al., 2012; Katzer et al.,
2011), the correlation between the baseline false alarm rate and physical symptom reporting
(as measured by the PHQ-15) was not significant. Katzer, Oberfeld, Hiller, Gerlach, and
Witthöft (2012) also reported a non-significant relationship between the variables. One
possible explanation for the absence of the relationship in Experiments 1a and 1b might be
that participants answered the PHQ-15 online at least one week before performing the SSDT,
and that the previously observed correlations are a context rather than a trait effect.
Experiments 2a and 2b
To address the limitations of Experiments 1a and 1b, replicate the SSDT findings, and
further investigate the possible transfer effects of SSDT conditioning, two further
experiments were carried out. In Experiment 2a, we sought to replicate the findings of
Running head: SOMATOSENSORY DECISION-MAKING 21
Experiment 1a by conditioning participants with a low baseline false alarm rate to exhibit
more false alarms. In Experiment 2b, we sought to replicate the findings of Experiment 1b by
conditioning participants with a high baseline false alarm rate to exhibit fewer false alarms.
In both experiments, we adopted a between-subjects design, attempted to control for possible
differences in motivation by providing sham reinforcement and punishment in the control
condition, and replaced the original SpS with a modified version of the task focusing on the
whole body. We also included an auditory signal detection (voice-hearing) task to investigate
whether somatosensory training generalizes to a different modality.
Method
Design
A mixed design was used for both experiments, with condition (control vs.
experimental) as the between-subject variable and phase (baseline vs. manipulation vs.
follow-up) as the within-subject variable. False alarm rate, hit rate, response bias, and
sensitivity on the SSDT, total number of SpSs, and false alarm rate, hit rate and total number
of voice responses on the voice hearing task were the dependent variables. We also measured
sleepiness, state anxiety and symptom reporting. As before, both experiments were carried
out in parallel, with participants being automatically allocated to studies according to their
baseline false alarm rate. Participants with low (< 0.15) and high false alarm rates (>= 0.15)
in light present trials in the experimental condition were allocated to Experiments 2a and 2b
respectively. The new false alarm criterion was approximate to the median false alarm rate
found in Experiments 1a and 1b combined and was expected to ensure approximately equal
numbers of participants in each study. A power calculation based on the effect sizes observed
in Experiments 1a and 1b (1.44 and .80 respectively) indicated that 18 participants (9 in each
group) in Experiment 2a and 52 participants (26 in each group) in Experiment 2b would give
80% power to detect the expected changes in false alarm rate with alpha = .05 (two-tailed).
Running head: SOMATOSENSORY DECISION-MAKING 22
Inclusion criteria were being aged between 18 and 40 years and having a good understanding
of English instructions. Exclusion criteria were non-corrected visual impairment, having a
medical condition that might affect the sense of touch or hearing, and participation in
previous SSDT studies. Participants received an honorarium of £10 or eight experimental
credits (psychology students only) for taking part.
Materials, Measures and Procedures
The SSDT setup and procedure and the questionnaires were the same as those in
Experiments 1a and 1b.
SpS test. The SpS test used in Experiments 1a and 1b was modified to address the
limitations identified previously. Participants were asked to relax and focus on their whole
body (instead of the non-dominant hand) for 20 seconds (compared to 10 seconds previously;
the duration was increased to allow participants to have adequate time to attend to the entire
body). There was a practice trial followed by three main trials, each indicated by a verbal
start and stop signal. After each trial, participants were given a printed body figure to circle
the areas where they felt the sensations. The three trials were averaged to obtain mean
baseline and follow-up SpSs for each participant.
Voice hearing task (VHT). The VHT was built using E-prime and consisted of a
continuous, 4.5 minute stream of white noise over which nonsense speech stimuli of different
amplitudes were randomly presented. Each of the speech stimuli consisted of seven random
English letters (e.g., oppvqsc) generated by PassMaker ( version 1.2; Rohr, 2013). These
were then converted into IVONA voice Brian (2014; WAV speech file) using a text-to-
speech software program (Balabolka version 2.9; Morozov, 2014) with volume = 100, rate =
0, pitch = 0, and duration = 800ms. Participants’ task was to press the spacebar every time
they thought they heard speech. The amplitude of the voices was determined by a pilot study
and ranged between three standard deviations of the mean auditory threshold of the pilot
Running head: SOMATOSENSORY DECISION-MAKING 23
sample; this range was selected on the assumption that it would result in a set of stimuli that
ranged from barely perceptible to clearly audible for most young adult participants. This
made the VHT sufficiently ambiguous (and thereby increased the likelihood of false alarms)
whilst remaining face valid to participants. We opted not to use a formal thresholding
procedure for this task to minimise the apparent similarity between the VHT and SSDT. The
main VHT was preceded by a one-minute practice version of the task using the same
structure and similar stimuli, as well as high amplitude sounds that were used to calculate
each participant’s mean reaction time (RT) to clearly audible voices. The RT was then used
to determine whether a given spacebar press was classed as a hit (within two standard
deviations of the mean practice RT excluding outliers) or a false alarm (beyond two standard
deviations of the mean practice RT excluding outliers) in the main task. Pseudorandom
intervals (1-2 seconds in the practice phase and 3-10 seconds in the main task) were
maintained between each presentation of the voice to ensure that they were spread over the
entire duration without overlapping or becoming predictable. False alarm rates on this task
are highly reliable over a period of 3 weeks (Huque, Heaney, Poliakoff, & Brown, 2016).
Procedure
The procedure was the same as that in Experiments 1a and 1b except that the VHT was
included in the task sequence and each participant took part in a single (randomly
determined) testing session only. As before, the only difference between the control and
experimental conditions was in the manipulation phase when conditioning was delivered
according to study allocation. The control condition started with the baseline VHT followed
by the baseline SpS test, baseline SSDT (two blocks of 80 trials), manipulation SSDT (four
blocks of 80 trials), follow-up VHT, follow-up SpS test, and follow-up SSDT (two blocks of
80 trials). As before, the short-form of the STAI (alpha coefficients were .68 and .66 in
Experiments 2a and 2b respectively) and KSS were administered before and after the SSDT
Running head: SOMATOSENSORY DECISION-MAKING 24
manipulation phase. In addition, participants answered the PHQ-15 (Cronbach’s alphas = .68
and .56 in Experiments 2a and 2b respectively) and MPPS at the end of the experiments. At
the start of the manipulation phase in the control condition, participants were informed that
they would get regular feedback about their performance after a certain number of trials and
told that this would improve their tactile perception and decision making. After every 40
trials, they saw a message on a computer screen stating how many points they had won or lost
and what the cumulative score was (e.g., after the 160th trial, “You have won 140 points; you
have lost 130 points; your cumulative total point is 80”). These messages were the same for
all participants in the control condition and were unrelated to their actual performance on the
task. At the end of this phase, participants were informed that their total score was 250 points
for which they received £2.50 (the average amount of money won in Experiments 1a and 1b)
in addition to their usual honorarium. These messages were included as a general motivator
but were not expected to train particular responses.
Conditioning. In Experiment 2a, the procedure for the experimental condition was the
same as Experiment 1a, which sought to increase the false alarm rate by rewarding 50% of
the hit trials and punishing 50% of the miss trials in blocks 3-6 of the SSDT. In Experiment
2b, the procedure for the experimental condition was the same as Experiment 1b, which
sought to decrease the false alarm rate by rewarding 50% of the correct rejection trials and
punishing 50% of the false alarm trials in blocks 3-6 of the SSDT. In this study, participants
won or lost five points (i.e., five pence), with the accompanying messages now including
either a yellow smiley-face emoticon (win) or a red sad-faced emoticon. It was expected that
the use of happy and sad emoticons would strengthen the reward and punishment and would
also act as a visual aid to the feedback message (Derks, Bos, & von Grumbkow, 2007;
Huang, Yen, & Zhang, 2008). Participants received one penny for each point they won in the
manipulation phase of the experimental condition plus their usual honorarium.
Running head: SOMATOSENSORY DECISION-MAKING 25
Data preparation and statistical analysis. Like Experiments 1a and 1b, data from the
SSDT, SpS test, VHT, sleepiness, and state anxiety were examined to identify outliers and
violation of mixed ANOVA assumptions, and standard procedures were followed to correct
problems. As before, a significance level of .05 was adopted for hypothesis testing
concerning the false alarms on the SSDT and VHT, and total SpS; a level of .01 was adopted
for all other comparisons.
Experiment 2a: Results
Participants
Seventy five (Female = 41, 54.67%) student (n = 65, 86.67%) and staff volunteers from
the University of Manchester took part. Age ranged between 19 and 39 years (M = 23.92. SD
= 4.96). All but four (94.67%) were right handed on the EHI. None of the participants’ hit
rates in the baseline phase was excessively high or low; data from all participants were
therefore used to analyse the SSDT responses. Though we originally intended to recruit a
much smaller sample (N = 18), the cut-point of 0.15 for a high false alarm rate proved to be
unexpectedly stringent, meaning we ended up testing many more participants in Experiment
2a to allow for adequate recruitment to Experiment 2b.
Effects of Conditioning on SSDT
Descriptive statistics on the SSDT response outcomes and phase by condition
interactions are presented in Tables 5 and 6 and Figure 3.
---INSERT TABLE 5 AND 6 AND FIGURE 3 ABOUT HERE---
False alarms, hit rate and response bias. Both the false alarm and hit rates were
significantly higher during and after the conditioning procedure than before, and were
significantly higher in the experimental than the control condition in both the manipulation
and follow-up phases. Participants in the training condition were significantly more likely to
Running head: SOMATOSENSORY DECISION-MAKING 26
say yes during and after the conditioning procedure than before, and significantly more so
than the control participants in the manipulation and follow-up phases.
Sensitivity. The main effect of phase was significant, F(1.84, 134.38) = 6.26, p = .003,
η2p = .08, but the main effect of condition, F(1, 73) = .22, p = .64, η2
p = .003, and the phase
by condition interaction, F(2, 146) = .39, p = .66, η2p = .01, were not. Bonferroni post-hoc
tests indicated that baseline sensitivity did not differ from sensitivity in the manipulation
phase, mean difference = .004, 95% CI [-.01, .02], p = 1.00, but dropped significantly in the
follow-up phase, mean difference = -.02, 95% CI [-.03, -.004], p = .01. Sensitivity in the
manipulation phase was significantly higher than that in the follow-up phase, mean difference
= .01, 95% CI [.002, .02], p = .01.
Transfer of Conditioning to SpSs and Auditory Modality
Descriptive statistics are presented in Table 7.
---INSERT TABLE 7 ABOUT HERE---
SpS. Contrary to expectation, the main effects of phase, F(1, 73) =.39, p = .54, η2p
= .01, condition, F(1, 73) = 1.56, p = .22, η2p = .02, and the interaction between them, F(1,
73) =.25, p = .62, η2p = .003, were not significant.
Voice false alarms. The main effect of phase was significant, F(1, 73) = 57.49, p
= .001, η2p= .44, but the main effect of condition, F(1, 73) = 3.25, p = .08, η2
p= .04, and the
interaction between phase and condition, F(1, 73) = 2.51, p = .12, η2p= .03, were not (see
Figure 4). Regardless of condition, participants produced significantly more false alarms in
the follow-up than in the baseline phase, mean difference = .92, 95% CI [.68, 1.16], p = .001.
Total voices. The main effect of phase was significant, F(1, 73) = 25.4, p = .001, η2p
= .26, but the main effect of condition was not, F(1, 73) = 1.58, p = .21, η2p= .02. Regardless
of condition, participants reported significantly more voices in the follow-up than in the
Running head: SOMATOSENSORY DECISION-MAKING 27
baseline phase, mean difference = .48, 95% CI [.29, .67], p = .001. The interaction between
phase and condition (see Figure 4) was close to significant at the Bonferroni corrected alpha
value, F(1, 73) = 4.50, p =.037, η2p= .06 .
Total hits. None of the main or interaction effects reached significance (F < 3, p > .05).
Sleepiness and State Anxiety
There was no significant difference in changes in sleepiness between the conditions (Ms
= 1.18 and 1.00, SEs = .28 and .42 for the control and experimental conditions respectively),
t(73) = 1.78, p = .08, r = .20. Changes in state anxiety also did not differ between the
conditions (Ms = -.05 and .26, SEs = .33 and .31 for the control and experimental conditions
respectively), t(73) = -.68, p = .50, r = .08. Main and interaction effects for the SSDT
variables remained unchanged when the influences of sleepiness and state anxiety were
controlled for.
Results: Experiment 2b
Participants
Seventy six student (n = 67, 88.16%) and staff volunteers from the University of
Manchester participated (female = 39, 51.31%), aged between 19 and 37 years (M = 22.89,
SD = 3.66); nine (11.84%) were left-handed on the EHI.
Effect of Conditioning on the SSDT
Descriptive statistics and phase by condition interactions for the SSDT outcomes are
presented in Tables 5 and 8 and Figure 3.
---INSERT TABLE 8 ABOUT HERE---
False alarms, hit rate and response bias.
Both the false alarm and hit rates were significantly lower during and after the
conditioning procedure than before, and were significantly lower in the experimental than the
control condition in both the manipulation and follow-up phases. Participants in the training
Running head: SOMATOSENSORY DECISION-MAKING 28
condition were significantly less likely to say yes during and after the conditioning procedure
than before, and significantly less so than the control participants in both the experimental
and follow-up phases. The tendency to say yes decreased further in the follow-up phase
compared to the manipulation phase in the experimental condition. Participants in the control
condition were less likely to say yes over time, with a corresponding descrease in their false
alarms and hit rate. In addition, there was a significant light × condition interaction, F(1, 74)
= 9.36, p < .01, ηp2 = .11, such that the hit rate was higher in the control than the experimental
condition but only for light present trials (light absent mean difference = .07, 95% CI
[-.04, .18], p = .20; light present mean difference = .15, 95% CI [.05, .25], p < .005). As this
has no obvious theoretical importance in this context it is not discussed further here.
Sensitivity. The main effects of phase, F(1.72, 127.60) = 2.60, p = .09, η2p= .03, and
condition, F(1, 74) = .26, p = .61, η2p= .003, were not significant and neither was the
interaction between them, F(2, 148) = 2.00, p = .15, η2p= .03.
Transfer of Conditioning to SpSs and Auditory Modality
Descriptive statistics are presented in Table 7.
SpS. Contrary to expectation, the main effects of phase, F(1, 74) = .46, p = .50, η2p
= .01, and condition, F(1, 74) = .04, p = .83, η2p= .001, and the interaction between them, F(1,
74) = 2.24, p = .14, η2p= .03, were not significant.
Voice false alarms. The main effect of phase was significant, F(1, 74) = 34.28, p
= .001, η2p= .32. Bonferroni corrected post-hoc tests revealed that participants produced
significantly more false alarms in the follow-up than in the baseline phase, mean difference
= .82, 95% CI [.54, 1.09], p = .001. The main effect of condition was not significant, F(1, 74)
= .24, p = .63, η2p= .003, and neither was the phase by condition interaction (see Figure 4),
F(1, 74) = 2.16, p = .15, η2p= .03.
Running head: SOMATOSENSORY DECISION-MAKING 29
Total voices. The main effect of phase was significant, F(1, 74) = 6.98, p = .01, η2p
= .09. A Bonferroni-corrected post-hoc test revealed that participants reported more voices in
the follow-up than in the baseline phase, mean difference = .28, 95% CI [.07, .49], p = .01.
The main effect of condition, F(1, 74) = .00, p = 1.00, η2p= .00, and the phase by condition
interaction were not significant, F(1, 74) = .28, p = .42, η2p= .01 (see Figure 4).
Voice hits. The main effect of phase was significant, F(1, 74) = 23.30, p = .001, η2p
= .24. Bonferroni post-hoc tests revealed that participants made fewer hits in the follow-up
than in the baseline phase, mean difference = -2.44, 95% CI [-3.45, -1.44], p = .001. The
main effect of condition, F(1, 74) = .19, p = .66, η2p= .003, and the phase by condition
interaction, F(1, 74) = .45, p = .50, η2p= .01, were not significant.
Sleepiness and State Anxiety
There was no significant difference between the conditions in changes in sleepiness (Ms
= 1.43 and .38, SEs = .37 and .39 for the control and experimental conditions respectively),
t(74) = 1.94, p = .06, r = .22, and state anxiety (Ms = .62 and -.08, SEs = .36 and .46 for the
control and experimental conditions respectively), t(74) = 1.18, p = .24, r = .14. Main and
interaction effects for the SSDT variables remaind unchanged when the influences of
sleepiness and state anxiety were controlled for.
Correlations Between Baseline False Alarms, SpSs and PHQ-15 Scores
As predicted, SSDT false alarm rate correlated positively with voice false alarms, rs
(149) = .21, p = .01. The correlation between SSDT hit rate and total SpSs was close to
significant, rs (149) = .14, p = .08. Contrary to expectation, the relationship between PHQ-15
and SSDT false alarm rates, was not significant, rs (149) = .09, p = .25. All the correlation
coefficients are presented in Table 5.
Discussion: Experiments 2a and 2b
Running head: SOMATOSENSORY DECISION-MAKING 30
Operant conditioning had the expected effects on false alarm rate in both studies,
replicating the findings of Experiments 1a and 1b using a more robust design that controlled
for order effects and between group differences in motivation and sleepiness. It is noteworthy
that the same effects pertained even though the monetary value of the reward and punishment
was half that used in Experiments 1a and 1b. However, neither study showed a significant
effect of SSDT conditioning on SpS. There was a near-significant trend for a larger increase
in voice false alarms in the conditioning group, suggesting that training may transfer between
tasks; there was no evidence of transfer on the VHT in Experiment 2b, however.
It is possible that the between-groups difference at VHT follow-up in Experiment 2a
would become significant with additional SSDT training. Indeed, our training was brief in
comparison to other studies on cross-modal transfer, which can involve up to 1000 training
trials over several days (e.g., Bratzke et al., 2014; Nagarajan, Blake, Wright, Byl, &
Merzenich, 1998). In the present experiment, there was a maximum of 80 trials resulting in
reward and punishment depending on participant responses. The strength of the effect on
SSDT performance and the apparent transfer to VHT responses is striking given the small
number of training trials used. It is less likely that a lack of training trials alone could account
for the absence of transfer on the VHT in Experiment 2b. It may be that the general increase
in voice false alarms over time, which was apparent in both studies and unrelated to
condition, concealed any countervailing training effects.
General Discussion
The primary aim of these experiments was to investigate whether it is possible to train
the tendency to report somatosensory misperceptions (i.e., tactile false alarms). As predicted,
we found reliable evidence that operant conditioning can both increase and decrease the
tendency to report false alarms on the SSDT. These findings are consistent with previous
perceptual training studies focusing on other sensory modalities (e.g., Johnstone & Alsop,
Running head: SOMATOSENSORY DECISION-MAKING 31
1996, 2000; Lie & Alsop, 2010) and, to our knowledge, are the first showing that
conditioning can also be used to manipulate tactile perception. Although we did not study
patients and our vibrotactile stimuli are not clinically relevant per se, the results of our
experiments provide proof of concept that learning can contribute to somatosensory
misperception and therefore have important implications for models of symptom reporting
and functional symptoms that draw on this notion (e.g., Brown, 2004). Our findings are
consistent with studies demonstrating the effect of classical conditioning on symptom reports
(e.g., Klinger, Soost, Flor, & Worm, 2007; Van den Bergh et al., 1999), whilst showing that
somatosensory false alarms can effectively be changed in both directions using operant
condition, and that this persists over time (for as long as a week, as suggested by the carry
over to the following session in Experiments 1a and 1b). They also raise the possibility that
operant conditioning could be used to reduce excessive symptom reporting. Future studies
should investigate whether it is possible to use a similar procedure to that studied here to
reduce the perceived unpleasantness of aversive stimuli (e.g., painful laser or electric pulses),
and thereby its potential as a clinical intervention in this sphere.
Contrary to expectation, we found no evidence that perceptual training on the SSDT
transfers to an unrelated task in which participant are asked to report SpSs in their hand or
their body. There was some weak evidence that training participants to say “yes” more on the
SSDT (i.e., exhibit more false alarms and hits) increased their tendency to make perceptual
errors (i.e., hear illusory voices) on the VHT. Although modest, this effect is noteworthy
given the extent of the training and the fact that the VHT and SSDT were strucuturally and
procedurally distinct, meaning that our findings are more likely to reflect common perceptual
mechanisms than shared-method variance. This is in marked contrast to other studies that
have used different versions of the same task to investigate possible cross-modal training
transfer effects (e.g., Bratzke et al., 2014; Lapid, Ulrich, & Rammsayer, 2009; Nagarajan et
Running head: SOMATOSENSORY DECISION-MAKING 32
al., 1998). We also found a significant correlation between false alarm rates on the SSDT and
VHT across Experiments 2a and 2b, suggesting some overlap between the two tasks despite
their procedural differences. Given the modest nature of these effects, however, it is evident
that further research is required before firm conclusions can be drawn regarding perceptual
transfer following operant conditioning in the tactile modality.
We have suggested previously that variations in a common perceptual decision process
could confer vulnerability to medically unexplained symptoms (MUS) as well as other
perceptual disturbances seen in clinical settings (e.g., hallucinations, body image
disturbance), with the precise nature of the experience depending on the individual’s beliefs,
preoccupations and focus of attention (Brown et al., 2012). Although some previous studies
have found support for this idea in the form of a correlation between false alarm rate on the
SSDT and physical symptom reporting (Brown et al., 2012; Katzer et al., 2011), no such
correlations were found in the studies reported here. Indeed, the relationship between these
variables appears to be highly inconsistent (Katzer et al., 2012) and presumably dependent on
other, as yet unidentified, factors. Any apparent variation in this mechanism also had little or
no effect on judgements about the presence of spontaneous bodily sensations in our studies.
We have already noted the small number of training trials in the conditioning phase, which
may have been inadequate to bring about a significant transfer effect and which should be
considered in future studies. It may also be that the SpS task, which asks participants to
describe all emerging sensations in their hand/body, is too non-specific to show what may be
a relatively small training effect. If so, categorising qualitatively similar sensations or asking
participants to detect and report specific sensations may address this issue.
One limitation of the present study is that we do not know whether our training
procedure changed the quality of perception or simply decision processes. To disentangle
this, the activation of sensory areas and higher-level processing areas of the brain in false
Running head: SOMATOSENSORY DECISION-MAKING 33
alarm trials can be examined with brain imaging techniques before, during, and after SSDT
training.
In summary, this is the first published study to demonstrate the effects of operant
conditioning on tactile perception and misperception. Our findings agree with previous
studies on signal detection training and provide proof of concept that similar training
procedures could have utility as treatments for excessive symptom reporting and other related
phenomena. Further research using more intensive training methods is required to establish
whether these effects transfer to other perceptual decisions and therefore have true clinical
potential.
Running head: SOMATOSENSORY DECISION-MAKING 34
References
Akerstedt, T., & Gillberg, M. (1990). Subjective and objective sleepiness in the active
individual. The International Journal of Neuroscience, 52(1-2), 29–37.
Albert III, R. P., Kelley, M., & Corlett, P. R. (2016). Hallucinations as top-down effects on
perception. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging.
doi:10.1016/j.bpsc.2016.04.003
Barsky, A. J. (1992). Amplification, somatization, and the somatoform disorders.
Psychosomatics, 33(1), 28–34.
Bratzke, D., Schröter, H., & Ulrich, R. (2014). The role of consolidation for perceptual
learning in temporal discrimination within and across modalities. Acta Psychologica,
147, 75–79. doi:10.1016/j.actpsy.2013.06.018
Brown, R. J. (2004). Psychological mechanisms of medically unexplained symptoms: An
integrative conceptual model. Psychological Bulletin, 130(5), 793–812.
doi:10.1037/0033-2909.130.5.793
Brown, R. J., Brunt, N., Poliakoff, E., & Lloyd, D. M. (2010). Illusory touch and tactile
perception in somatoform dissociators. Journal of Psychosomatic Research, 69(3),
241–248. doi:10.1016/j.jpsychores.2009.11.010
Brown, R. J., Skehan, D., Chapman, A., Perry, E. P., McKenzie, K. J., Lloyd, D. M., …
Poliakoff, E. (2012). Physical symptom reporting is associated with a tendency to
experience somatosensory distortion. Psychosomatic Medicine, 74(6), 648–655.
doi:10.1097/PSY.0b013e3182595358
Cameron, O. G. (2001). Interoception: The inside story—A model for psychosomatic
processes. Psychosomatic Medicine, 63(5), 697–710.
Running head: SOMATOSENSORY DECISION-MAKING 35
Derks, D., Bos, A. E. R., & von Grumbkow, J. (2007). Emoticons and online message
interpretation. Social Science Computer Review, 26(3), 379–388.
doi:10.1177/0894439307311611
Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G* Power 3: A flexible statistical
power analysis program for the social, behavioral, and biomedical sciences. Behavior
Research Methods, 39(2), 175–191.
Field, A. P. (2009). Discovering statistics using SPSS. London, England: SAGE.
Gillberg, M., & Åkerstedt, T. (1998). Sleep loss and performance: No “safe” duration of a
monotonous task. Physiology & Behavior, 64(5), 599–604. doi:10.1016/S0031-
9384(98)00063-8
Green, D. M., & Swets, J. A. (1966). Signal detection theory and psychophysics. New York,
NY: Wiley.
Huang, A. H., Yen, D. C., & Zhang, X. (2008). Exploring the potential effects of emoticons.
Information & Management, 45(7), 466–473. doi:10.1016/j.im.2008.07.001
Huque, A. U., Heaney, A., Poliakoff, E., & Brown, R. J. (2016). Development and validation
of a voice-hearing task for research on auditory verbal hallucinations and auditory
misperception. Manuscript submitted for publication.
IVONA Voice Brian [Computer software]. (2014). Retrieved from
http://www.ivona.com/us/about-us/voice-portfolio/
Johnstone, V., & Alsop, B. (1996). Human signal-detection performance: Effects of signal
presentation probabilities and reinforcer distributions. Journal of the Experimental
Analysis of Behavior, 66(2), 243–263. doi:10.1901/jeab.1996.66-243
Johnstone, V., & Alsop, B. (2000). Reinforcer control and human signal-detection
performance. Journal of the Experimental Analysis of Behavior, 73(3), 275–290.
Running head: SOMATOSENSORY DECISION-MAKING 36
Katzer, A., Oberfeld, D., Hiller, W., Gerlach, A. L., & Witthöft, M. (2012). Tactile perceptual
processes and their relationship to somatoform disorders. Journal of Abnormal
Psychology, 121(2), 530–543. doi:10.1037/a0026536
Katzer, A., Oberfeld, D., Hiller, W., & Witthöft, M. (2011). Tactile perceptual processes and
their relationship to medically unexplained symptoms and health anxiety. Journal of
Psychosomatic Research, 71(5), 335–341. doi:10.1016/j.jpsychores.2011.03.009
Klinger, R., Soost, S., Flor, H., & Worm, M. (2007). Classical conditioning and expectancy
in placebo hypoalgesia: A randomized controlled study in patients with atopic
dermatitis and persons with healthy skin. Pain, 128(1-2), 31–39.
doi:10.1016/j.pain.2006.08.025
Körber, S., Frieser, D., Steinbrecher, N., & Hiller, W. (2011). Classification characteristics of
the Patient Health Questionnaire-15 for screening somatoform disorders in a primary
care setting. Journal of Psychosomatic Research, 71(3), 142–147.
doi:10.1016/j.jpsychores.2011.01.006
Kroenke, K., Spitzer, R. L., & Williams, J. B. W. (2002). The PHQ-15: Validity of a new
measure for evaluating the severity of somatic symptoms. Psychosomatic Medicine,
64(2), 258–266.
Lapid, E., Ulrich, R., & Rammsayer, T. (2009). Perceptual learning in auditory temporal
discrimination: No evidence for a cross-modal transfer to the visual modality.
Psychonomic Bulletin & Review, 16(2), 382–389. doi:10.3758/PBR.16.2.382
Lie, C., & Alsop, B. (2009). Effects of point-loss punishers on human signal-detection
performance. Journal of the Experimental Analysis of Behavior, 92(1), 17–39.
doi:10.1901/jeab.2009.92-17
Running head: SOMATOSENSORY DECISION-MAKING 37
Lie, C., & Alsop, B. (2010). Stimulus disparity and punisher control of human signal-
detection performance. Journal of the Experimental Analysis of Behavior, 93(2), 185–
201. doi:10.1901/jeab.2010.93-185
Liu, Z., & Weinshall, D. (2000). Mechanisms of generalization in perceptual learning. Vision
Research, 40(1), 97–109.
Lloyd, D. M., Mason, L., Brown, R. J., & Poliakoff, E. (2008). Development of a paradigm
for measuring somatic disturbance in clinical populations with medically unexplained
symptoms. Journal of Psychosomatic Research, 64(1), 21–24.
doi:10.1016/j.jpsychores.2007.06.004
Malow, R. M. (1981). The effects of induced anxiety on pain perception: A signal detection
analysis. PAIN, 11(3), 397–405. doi:10.1016/0304-3959(81)90639-4
Marteau, T. M., & Bekker, H. (1992). The development of a six-item short-form of the state
scale of the Spielberger State—Trait Anxiety Inventory (STAI). British Journal of
Clinical Psychology, 31(3), 301–306.
McKenzie, K. J., Lloyd, D. M., Brown, R. J., Plummer, F., & Poliakoff, E. (2012).
Investigating the mechanisms of visually-evoked tactile sensations. Acta
Psychologica, 139(1), 46–53. doi:10.1016/j.actpsy.2011.09.012
McKenzie, K. J., Poliakoff, E., Brown, R. J., & Lloyd, D. M. (2010). Now you feel it, now
you don’t: How robust is the phenomenon of illusory tactile experience? Perception,
39(6), 839–850.
Michael, G. A., Dupuy, M.-A., Deleuze, A., Humblot, M., Simon, B., & Naveteur, J. (2012).
Interacting effects of vision and attention in perceiving spontaneous sensations arising
on the hands. Experimental Brain Research, 216(1), 21–34. doi:10.1007/s00221-011-
2901-y
Running head: SOMATOSENSORY DECISION-MAKING 38
Michael, G. A., & Naveteur, J. (2011). The tickly homunculus and the origins of spontaneous
sensations arising on the hands. Consciousness and Cognition, 20(3), 603–617.
doi:10.1016/j.concog.2010.11.013
Mirams, L., Poliakoff, E., Brown, R. J., & Lloyd, D. M. (2012). Interoceptive and
exteroceptive attention have opposite effects on subsequent somatosensory perceptual
decision making. The Quarterly Journal of Experimental Psychology, 65(5), 926–938.
doi:10.1080/17470218.2011.636823
Mirams, L., Poliakoff, E., Brown, R. J., & Lloyd, D. M. (2013). Brief body-scan meditation
practice improves somatosensory perceptual decision making. Consciousness and
Cognition, 22(1), 348–359. doi:10.1016/j.concog.2012.07.009
Morozov, I. (2014). Balabolka [Computer software]. Retrieved from http://www.cross-plus-
a.com/balabolka.htm
Nagarajan, S. S., Blake, D. T., Wright, B. A., Byl, N., & Merzenich, M. M. (1998). Practice-
related improvements in somatosensory interval discrimination are temporally
specific but generalize across skin location, hemisphere, and modality. The Journal of
Neuroscience, 18(4), 1559–1570.
Oldfield, R. C. (1971). The assessment and analysis of handedness: The Edinburgh inventory.
Neuropsychologia, 9(1), 97–113. doi:10.1016/0028-3932(71)90067-4
Reynolds, G. S. (1975). Simple schedules of positive reinforcement. In A primer of operant
conditioning (pp. 59–78). Oakland, N.J.: Scott, Foresman & Company.
Rief, W., & Barsky, A. J. (2005). Psychobiological perspectives on somatoform disorders.
Psychoneuroendocrinology, 30(10), 996–1002. doi:10.1016/j.psyneuen.2005.03.018
Rief, W., & Broadbent, E. (2007). Explaining medically unexplained symptoms-models and
mechanisms. Clinical Psychology Review, 27(7), 821–841.
doi:10.1016/j.cpr.2007.07.005
Running head: SOMATOSENSORY DECISION-MAKING 39
Rohr, K. (2013). PassMaker [Computer software]. Retrieved from
https://itunes.apple.com/gb/app/passmaker/id423229019?mt=12
Schaefer, M., Egloff, B., Gerlach, A. L., & Witthöft, M. (2014). Improving heartbeat
perception in patients with medically unexplained symptoms reduces symptom
distress. Biological Psychology, 101, 69–76. doi:10.1016/j.biopsycho.2014.05.012
Slade, P. D. (1994). What is body image? Behaviour Research and Therapy, 32(5), 497–502.
Snodgrass, J. G., & Corwin, J. (1988). Pragmatics of measuring recognition memory:
Applications to dementia and amnesia. Journal of Experimental Psychology: General,
117(1), 34–50.
Szalma, J. L., Hancock, P. A., Warm, J. S., Dember, W. N., & Parsons, K. S. (2006).
Training for vigilance: Using predictive power to evaluate feedback effectiveness.
Human Factors: The Journal of the Human Factors and Ergonomics Society, 48(4),
682–692.
Tazaki, M., & Landlaw, K. (2006). Behavioural mechanisms and cognitive-behavioural
interventions of somatoform disorders. International Review of Psychiatry, 18(1), 67–
73. doi:10.1080/09540260500467046
Tluczek, A., Henriques, J. B., & Brown, R. L. (2009). Support for the reliability and validity
of a six-item state anxiety scale derived from the State-Trait Anxiety Inventory.
Journal of Nursing Measurement, 17(1), 19-28.
Van den Bergh, O., Stegen, K., Van Diest, I., Raes, C., Stulens, P., Eelen, P., … Nemery, B.
(1999). Acquisition and extinction of somatic symptoms in response to odours: A
Pavlovian paradigm relevant to multiple chemical sensitivity. Occupational and
Environmental Medicine, 56(5), 295–301.
van Dessel, N., den Boeft, M., van der Wouden, J. C., Kleinstäuber, M., Leone, S. S., Terluin,
B., … van Marwijk, H. (2014). Non-pharmacological interventions for somatoform
Running head: SOMATOSENSORY DECISION-MAKING 40
disorders and medically unexplained physical symptoms (MUPS) in adults. Cochrane
Database of Systematic Reviews, 11. doi.10.1002/14651858.CD011142.pub2
Walker, L. S., & Zeman, J. L. (1992). Parental response to child illness behavior. Journal of
Pediatric Psychology, 17(1), 49–71.
Watson, D., & Pennebaker, J. W. (1989). Health complaints, stress, and distress: Exploring
the central role of negative affectivity. Psychological Review, 96(2), 234–254.
Witthöft, M., Fischer, S., Jasper, F., Rist, F., & Nater, U. M. (2016). Clarifying the latent
structure and correlates of somatic symptom distress: A bifactor model approach.
Psychological Assessment, 28(1), 109–115. doi:10.1037/pas0000150
Zijlema, W. L., Stolk, R. P., Löwe, B., Rief, W., White, P. D., & Rosmalen, J. G. M. (2013).
How to assess common somatic symptoms in large-scale studies: A systematic review
of questionnaires. Journal of Psychosomatic Research, 74(6), 459–468. doi:
10.1016/j.jpsychores.2013.03.093
Running head: SOMATOSENSORY DECISION-MAKING 41
Footnotes
1Determined by the median false alarm rate in previous SSDT studies (Katzer, Oberfeld,
Hiller, & Witthöft, 2011; Lloyd, Mason, Brown, & Poliakoff, 2008; McKenzie, Lloyd,
Brown, Plummer, & Poliakoff, 2012; McKenzie, Poliakoff, Brown, & Lloyd, 2010).
2For the purposes of analysis, definitely yes and maybe yes responses were combined as
were definitely no and maybe no responses.
3We are grateful to an anonymous reviewer for this suggestion.
Running head: SOMATOSENSORY DECISION-MAKING 42
Table 1 Estimated Marginal Means (Standard Deviations) of the SSDT Outcomes in the Baseline,
Manipulation, and Follow-Up Phases of the Experimental and Control Conditions in
Experiments 1a and 1b
Condition FA rate Hit rate Bias Sensitivity
Experiment 1a (LFA)
Control condition
Baseline .36 (.12)a .63 (.24) 1.53 (.16)a 1.57 (.83)
Manipulation .34 (.14)a .54 (.27) 1.60 (.16)a 1.35 (.94)
Follow-up .32 (.14)a .47 (.26) 1.64 (.15)a 1.20 (.94)
Experimental condition
Baseline .29 (.09)a .54 (.22) 1.62 (.13)a 1.55 (.71)
Manipulation .46 (.22)a .75 (.22) 1.40 (.20)a 1.63 (1.13)
Follow-up .41 (.18)a .58 (.26) 1.52 (.19)a 1.27 (1.07)
Experiment 1b (HFA)
Control condition
Baseline -.88 (.37)b .63 (.22) .32 (.62) .55 (.11)b
Manipulation -1.03 (.45)b .51 (.26) .59 (.70) .54 (.09)b
Follow-up -1.01 (.45)b .40 (.25) .75 (.70) .49 (.11)b
Experimental condition
Baseline -.64 (.18)b .62 (.20) .15 (.38) .54 (.09)b
Manipulation -1.27 (.39)b .37 (.25) .99 (.58) .54 (.12)b
Follow-up -1.47 (.37)b .17 (.15) 1.45 (.58) .47 (.11)b
Note. SSDT = somatosensory signal detection task, FA = false alarm, LFA = low false alarm
participants, HFA = high false alarm participants.a Square root transformed score; bLog transformed score.
Running head: SOMATOSENSORY DECISION-MAKING 43
Table 2 Effects of Training Designed to Increase the False Alarm Rate in Experiment 1a
Significant interactions Source of interaction
False alarm rate
Phase × Condition: F(2, 88) = 29.32, p = .001, η2
p = .40
Baseline phaseExperimental < Control (diff. = .08, 95% CI [.04, .11], p = .001)
Manipulation phaseExperimental > Control (diff. = .12, 95% CI [.06, .19], p = .001)
Follow-up phaseExperimental > Control (diff. = .09, 95% CI [.04, .14], p = .001)
Control conditionFollow-up < Baseline (diff. = .04, 95% CI [.01, .07], p = .003)
Experimental conditionManipulation > Baseline (diff.= .18, 95% CI [.10, .25], p = .001)Follow-up > Baseline (diff. = .12, 95% CI [.06, .19], p = .001)Manipulation > Follow-up (diff. = .05, 95% CI [.00, .10], p = .05)
Hit rate
Phase × Condition: F(1.90, 83.80) = 19.81, p = .001, η2
p = .31
Manipulation phaseExperimental > Control (diff. = .21, 95% CI [.11, .32], p = .001)
Follow-up phaseExperimental > Control (diff. = .11, 95% CI [.003, .22], p = .04)
Experimental conditionManipulation > baseline (diff. = .21, 95% CI [.13, .29], p = .001)Manipulation > follow-up (diff. = .17, 95% CI [.10, .24], p = .001)
Control conditionBaseline > Manipulation (diff. = .09, 95% CI [.03, .15], p = .001)Baseline > Follow-up (diff. = .16, 95% CI [.07, .24], p = .001)
Response bias (tendency to say ‘yes’)a
Phase × Condition:F(2, 88) = 36.92, p = .001, η2
p = .46
Baseline phaseControl > Experimental (diff. =-.09, 95% CI [-.15, -.04], p = .001)
Manipulation phaseExperimental > Control (diff. = -.20, 95% CI [-.28, -.13], p = .001)
(continued)Significant interactions
Source of interaction
Running head: SOMATOSENSORY DECISION-MAKING 44
Follow-up phaseExperimental > Control (diff. = -.12, 95% CI [-.18, -.05], p = .001)
Control conditionBaseline > Manipulation (diff. = -.07, 95% CI [-.11. -.04], p = .001)Baseline > Follow-up (diff. = -.11, 95% CI [-.16, -.06], p = .001)
Experimental conditionManipulation > baseline (diff. = -.23, 95% CI [-.30, -.15], p = .001)Manipulation > follow-up (diff. = -.13, 95% CI [-.19, -.06], p = .001)Follow-up > baseline (diff. = -.10, 95% CI [-.17, -.03], p = .002)
Note. Only significant interaction effects are presented; see supplementary materials for full
details of analysis. aTendency to say yes is associated with lower scores of c (i.e. response bias).
Running head: SOMATOSENSORY DECISION-MAKING 45
Table 3
Mean Number (Standard Deviation) of Spontaneous Sensations in the Baseline and Follow-
up SpS Tests in the Experimental and Control Conditions Across Different Session Orders in
Experiments 1a and 1b.
First session
Experiment Condition Phase Control condition Experimental condition
Experiment 1a (LFA)a Experimental Baseline .67 (.44) .76 (.47)
Follow-up .79 (.48) .95 (.47)
Control Baseline .74 (.38) .67 (.48)
Follow-up .84 (.36) .70 (.64)
Experiment 1b (HFA) Experimental Baseline .59 (.63) .98 (.67)
Follow-up .62 (.62) 1.47 (.92)
Control Baseline .67 (.54) .80 (.62)
Follow-up .90 (.67) 1.08 (.79)
Note. SpS = spontaneous sensation, LFA = low false alarm participants, HFA = high false
alarm participants.aSquare root transformed data.
Running head: SOMATOSENSORY DECISION-MAKING 46
Table 4
Effects of Training Designed to Decrease the False Alarm Rate in Experiment 1b
Significant interactions Source of interaction
False alarm rate
Phase × Condition: F(2, 56) = 50.60, p = .001, η2
p = .64
Condition × Session: F(1, 28) = 18.30, p = .001, η2
p = .40
Baseline phaseExperimental > Control (diff. = .24, 95% CI [.07, .40], p = .006)
Manipulation phaseExperimental < Control (diff. = .24, 95% CI [.06, .41,], p = .01)
Follow-up phaseExperimental < Control (diff. = .47, 95% CI [.33, .60], p = .001)
Control conditionManipulation < Baseline (diff. = .15, 95% CI [.03, .27], p = .01)
Experimental conditionFollow-up < Manipulation (diff. = .21, 95% CI [.07, .35], p = .003) < Baseline (diff.= .62, 95% CI [.44, .80], p = .001)
Control session firstControl > Experimental (diff. = .44, 95% CI [.24, .65], p = .001)
Control conditionControl first > Experimental first (diff. =.48, 95% CI [.18, .78], p = .003)
Hit rate
Phase × Condition: F(2, 56) = 12.45, p = .001, η2
p = .31
Manipulation phaseExperimental < Control (diff. = .14, 95% CI [.04, .24], p = .007)
Follow-up phaseExperimental < Control (diff. = .23, 95% CI = [.13, .33], p = .001)
Response bias (tendency to say ‘yes’)a
Phase × Condition: F(2, 56) = 29.95, p = .001, η2
p = .52
Condition × Session: F(1, 28) = 6.27, p = .02, η2
p = .18
Manipulation phaseExperimental < Control (diff. = -.40, 95% CI [-.62, -.18], p = .001)
Follow-up phaseExperimental < Control (diff. = -.70, 95% CI [-.95, -.45], p = .001)
Control session firstControl > Experimental (diff. = -.57, 95% CI [-.89, -.25], p = .001)
(continued)Control condition
Running head: SOMATOSENSORY DECISION-MAKING 47
Significant interactions Source of interactionControl second < Control first (diff. = -.60, 95% CI [-1.08, -.11], p = .02)
Note. Only significant interaction effects are presented; see supplementary materials for full
details of analysis. aTendency to say yes is associated with lower scores of c (i.e. response bias).
Running head: SOMATOSENSORY DECISION-MAKING 48
Table 5
Estimated Marginal Means (Standard Deviations) of the SSDT Outcomes in the Baseline,
Manipulation, and Follow-Up Phases of the Experimental and Control Conditions in
Experiments 2a and 2b
Condition FA rate Hit rate Bias Sensitivity
Experiment 2a (LFA)
Control condition
Baseline -1.24 (.33)b .72 (.13)c 1.65 (.15)a .88 (.05)b
Manipulation -1.15 (.46)b .71 (.12)c 1.62 (.17)a .87 (.06)b
Follow-up -1.30 (.42)b .68 (.12)c 1.70 (.18)a .86 (.05)b
Experimental condition
Baseline -1.19 (.33)b .70 (.13)c 1.66 (.15)a .87 (.05)b
Manipulation -.68 (.46)b .83 (.12)c 1.35 (.17)a .87 (.06)b
Follow-up -.70 (.42)b .80 (.12)c 1.43 (.18)a .85 (.05)b
Experiment 2b (HFA)
Control condition
Baseline -.64 (.20)b .61 (.19) 1.46 (.12)a 1.40 (.27)a
Manipulation -.74 (.30)b .57 (.25) 1.51 (.15)a 1.39 (.30)a
Follow-up -.80 (.38)b .51 (.26) 1.54 (.18)a 1.34 (.32)a
Experimental condition
Baseline -.68 (20)b .59 (.20) 1.49 (.12)a 1.38 (.27)a
Manipulation -1.25 (.30)b .41 (.25) 1.70 (.15)a 1.46 (.29)a
Follow-up -1.34 (.38)b .36 (.26) 1.75 (.18)a 1.40 (.32)a
Note. SSDT = somatosensory signal detection task, FA = false alarm, LFA = low false alarm
participants, HFA = high false alarm participants.a Square root transformed score; bLog transformed score; cReciprocal transformed score.
Running head: SOMATOSENSORY DECISION-MAKING 49
Table 6
Effects of Training Designed to Increase the False Alarm Rate in Experiment 2a
Significant interactions Source of interactionFalse alarm rate
Phase × Condition: F (2, 146) = 18.21, p = .001, η2
p = .20
Manipulation phaseExperimental > Control (diff. = .47, 95% CI [.26, .68], p = .001)
Follow-up phaseExperimental > Control (diff. = .59, 95% CI [.40, .79], p = .001)
Control conditionManipulation > Follow-up (diff. = .15, 95% CI [.04, .26], p = .003)
Experimental conditionManipulation > Baseline (diff.= .51, 95% CI [.31, .72], p = .001)Follow-up > Baseline (diff. = .49, 95% CI [.31, .67], p = .001)
Hit ratePhase × Condition: F(2, 146) = 24.54, p = .001, η2
p = .25
Manipulation phaseExperimental > Control (diff. = .13, 95% CI [.08, .18], p = .001)
Follow-up phaseExperimental > Control (diff. = .12, 95% CI [.06, .17], p = .001)
Experimental conditionManipulation > baseline (diff. = .13, 95% CI [.09, .18], p = .001)Follow-up > baseline (diff. = .10, 95% CI [.05, .15], p = .001)Manipulation > Follow-up (diff. = .03, 95% CI [.00, .07], p = .05)
Response bias (tendency to say ‘yes’)a
Phase × Condition:F(2, 146) = 33.28, p = .001, η2
p= .31
Manipulation phaseExperimental > Control (diff. = -.27, 95% CI [-.35, -.19], p = .001)
Follow-up phaseExperimental > Control (diff. = -.27, 95% CI [-.35, -.19], p = .001)
Control conditionManipulation > Follow-up (diff. = -.08, 95% CI [-.13, -.03], p = .001)
Experimental conditionManipulation > baseline (diff. = -.30, 95% CI [-.38, -.22], p = .001)Follow-up > baseline (diff. = -.22, 95% CI [-.30, -.15], p = .001)Manipulation > follow-up (diff. = -.08, 95% CI [-.13, -.02], p = .002)
Note. Only significant interaction effects are presented; see supplementary materials for full
details of analysis.
Running head: SOMATOSENSORY DECISION-MAKING 50
aTendency to say yes is associated with lower scores of c (i.e. response bias).
Table 7
Mean Number (Standard Deviation) of Spontaneous Sensations, Voice False Alarms, Total
Voices, and Voice Hits in the Baseline and Follow-up Phases of the Experimental and
Control Conditions in Experiments 2a and 2b.
Control condition Experimental condition
Variable Baseline Follow-up Baseline Follow-up
Experiment 2aSpontaneous sensationsa 1.45 (.25) 1.46 (.28) 1.37 (.26) 1.40 (.24)
Voice false alarmsa 2.44 (.99) 3.17 (1.24) 2.71 (1.12) 3.82 (1.50)
Total voicesa 4.50 (.79) 4.78 (1.19) 4.55 (.74) 5.24 (1.07)
Voice hits 13.93 (5.82) 13.15 (6.77) 12.68 (6.07) 11.74 (6.81)
Experiment 2bSpontaneous sensationsb .33 (.17) .31 (.19) .29 (.16) .33 (.15)
Voice false alarmsa 2.59 (1.14) 3.61 (1.53) 2.92 (1.09) 3.53 (1.24)
Total voicesa 4.78 (.85) 5.15 (1.30) 4.87 (.82) 5.06 (.98)
Voice hits 15.62 (6.38) 12.84 (6.31) 14.72 (5.88) 12.62 (5.60)
Note. aSquare root transformed data. b Log transformed data.
Running head: SOMATOSENSORY DECISION-MAKING 51
Table 8
Effects of Training designed to Decrease the False Alarm Rate in Experiment 2b
Significant interactions Source of interactionFalse alarm rate
Phase × Condition: F(2, 148) = 32.62, p = .001, η2
p= .31
Manipulation phaseExperimental < Control (diff. = -.51, 95% CI [-.65, -.37], p = .001)
Follow-up phaseExperimental < Control (diff. = .55, 95% CI [-.72, -.37], p = .001)
Control conditionFollow-up < Baseline (diff. = -.15, 95% CI [-.30, -.01], p = .05)
Experimental conditionManipulation < Baseline (diff.= -.57, 95% CI [-.67, -.47], p = .001)Follow-up < Baseline (diff.= -.66, 95% CI [-.80, -.53], p = .001)
Hit ratePhase × Condition: F(2, 148) = 7.67, p = .001, η2
p= .09
Manipulation phaseExperimental < Control (diff. = -.16, 95% CI [-.27, -.04], p = .01)
Follow-up phaseExperimental < Control (diff. = -.15, 95% CI [-.27, -.03], p = .02)
Control conditionFollow-up < Baseline (diff. = -.10, 95% CI = [-.18, -.02], p = .01)Follow-up < Manipulation (diff. = -.06, 95% CI [-.11, -.001], p = .05)
Experimental conditionManipulation < Baseline (diff.= -.18, 95% CI [-.25, -.12], p = .001)Follow-up < Baseline (diff.= -.23, 95% CI [-.31, -.15], p = .001)
Response bias (tendency to say ‘yes’)a
Phase × Condition: F(2, 148) = 22.23, p = .001, η2
p= .23
Manipulation phaseControl > Experimental (diff. = -.19, 95% CI [-.26, -.12], p = .001)
Follow-up phaseControl > Experimental (diff. = -.21, 95% CI [-.29, -.13], p = .001)
Control conditionBaseline > Follow-up (diff. = -.08, 95% CI [-.14, -.01], p = .01)
Experimental conditionBaseline > Manipulation (diff. = -.21, 95% CI [-.27, -.16], p = .001)Baseline > Follow-up (diff. = -.26, 95% CI [-.32, -.20], p = .001Manipulation > Follow-up (diff. = -.05, 95% CI [-.09, -.01], p = .01)
Running head: SOMATOSENSORY DECISION-MAKING 52
Note. Only significant interaction effects are presented; see supplementary materials for full
details of analysis. aTendency to say yes is associated with lower scores of c (i.e. response bias).
Running head: SOMATOSENSORY DECISION-MAKING 53
Table 9
Summary of Medians, Ranges, and Intercorrelations for the SSDT and VHT Outcomes, and Total SpSs in the Baseline Phase of Experiments 2a
and 2b Combined (N = 151)
Variables Mdn Range 1 2 3 4 5 6 7
1. False alarm rate .13 .74 --
2. Hit rate .57 .94 .21** --
3. Sensitivity 1.27 4.01 -.41*** .76*** --
4. Bias .40 2.81 -.68*** -.83*** -.30*** --
5. Total SpSs 1 4.06 .02 .14 .08 -.126 --
6. Voice false alarms 7 41 .21* .10 -.04 -.21* -.05 --
7. Total voices 22 41 .21* .02 -.09 -.14 -.06 .51*** --
Note. SSDT = Somatosensory signal detection task; VHT = Voice hearing task; SpS = Spontaneous sensation.
*p < .05. **p < .01. ***p < .001.
Running head: SOMATOSENSORY DECISION-MAKING 54
Figure 1. Experimental condition of Experiments 1a and
1b. The same sequence of tasks was followed in the control condition except that SSDT
responses did not produce any consequence. SpS = spontaneous sensation; SSDT =
somatosensory signal detection task; FA = false alarm; CR = correct rejection.
Follow-up SpS test
Experiment 1a (Low FA participants)
Baseline SSDT (2 X 80) trials
Baseline SpS test
Experiment 1b (High FA participants)
Manipulation phase SSDT (4 X 80) trials Won 10 points for a CR Lost 10 points for a FA
Manipulation phase SSDT (4 X 80) trials Won 10 points for a hit Lost 10 points for a miss
Follow-up SSDT(2 X 80) trials
Running head: SOMATOSENSORY DECISION-MAKING 55
Control conditionExperimental condition
Figure 2. Phase by condition interactions for the SSDT outcomes of Experiments 1a and 1b. False alarm rate, hit rate, response bias, and
sensitivity are depicted in 1a-i, 1a-ii, 1a-iii, 1a-iv respectively for Experiment 1a, and in 1b-i, 1b-ii, 1b-iii, and 1b-iv respectively for Experiment
1b. Error bars are standard errors.
Control conditionExperimental condition
*
* ****
*
*
1a-iv1a-iii1a-ii1a-i
**
***
*
2a-iv2a-iii2a-ii2a-i
*
*
****
*
1b-iv1b-iii1b-ii1b-i
Running head: SOMATOSENSORY DECISION-MAKING 56
Figure 3. Phase by condition interactions for the SSDT outcomes of Experiments 2a and 2b. False alarm rate, hit rate, response bias, and
sensitivity are depicted in 2a-i, 2a-ii, 2a-iii, 2a-iv respectively for Experiment 2a, and in 2b-i, 2b-ii, 2b-iii, 2b-iv respectively for Experiment 2b.
Error bars are standard errors.
**
**
**
2b-iv2b-iii2b-ii2b-i
Running head: SOMATOSENSORY DECISION-MAKING 57
Figure 4. Interactions between phase and condition for voice false alarms and total voices in
Experiment 2a (i, ii) and Experiment 2b (iii, iv). Error bars are standard errors.
Baseline Follow-up2
2.36
2.71
3.07
3.43
3.79
4.14Control condition
Phase
Squ
are
root
tran
sfor
med
voi
ce
fals
e al
arm
s
Baseline Follow-up4.3
4.625
4.95
5.275
5.6
Phase
Squ
are
root
tran
sfor
med
tota
l vo
ices
i. ii.
iii.
Baseline Follow-up2.2
2.65
3.1
3.55
4
Phase
Squ
are
root
tran
sfor
med
voi
ce
fals
e al
arm
s
Baseline Follow-up4.6
4.8
5
5.2
Phase
Squ
are
root
tran
sfor
med
tota
l vo
ices
iv.