Avoidance habits and cognitive distrust : a novel experimental paradigm

(1)

Avoidance Habits and

Cognitive Distrust: A

novel experimental

paradigm

Research Master Internship Report

Zara Bamdad (10620842)

June 2014

Word count : 9493

Supervised by Sanne de Wit

(2)

2

Introduction

Many of us can relate to the feeling that you may not have locked the door properly when you left the house this morning. When you return to the scene you find that indeed you had locked the door, as you do every morning. We can explain this phenomenon fairly intuitively; the fact that we do this action every morning means that it has become an automatic, efficient action; we initiate it without having to think about it too much. However, the downside to this can be that we don’t properly encode the memory of the behaviour. Hence, the thoughts of ‘did I definitely do that?’ and ‘did I do it properly?’ creating a need to return and check it again.

A more severe version of this ‘cognitive distrust’ phenomenon is seen in compulsive checking behaviour in Obsessive Compulsive Disorder (OCD) – a person will repeatedly check something (such as a lock) to a pathological degree. This compulsive checking can severely impair the quality of an individuals’ life and thus it is crucial to understand the mechanisms underpinning this behaviour. Compulsive checking is traditionally explained as behaviour aimed at avoiding an intrusive thought (obsession) (American Psychiatric Association, 2013). The present study introduces a potential alternative explanation to the traditional account through investigating the novel hypothesis that memory uncertainty arises as a consequence of habit formation, which is perpetuated through an automaticity-cognitive distrust cycle.

The traditional idea that compulsive checking behaviour is ‘aimed’ at achieving an outcome (i.e. avoiding a thought) assumes that the behaviour is goal-directed. Goal-directed behaviour is characterised by meeting the belief and desire criteria; the person must believe that the behaviour (in this case, checking) will lead to a certain outcome, and the person must desire this outcome (de Wit & Dickinson, 2009). This goal-directed explanation of compulsive checking is plausible in part. For instance, people with OCD may experience intrusive thoughts of being burgled and check their locks because they believe that this will reduce their security concerns, and this outcome is desirable. Nevertheless, the goal-directed account falls short of explaining some other aspects of compulsive checking. Patients can report feeling compelled to perform checking behaviour without the intrusive thought (e.g. checking the door without the anxiety of being burgled). Similarly, patients often acknowledge that there is no rational relationship between their compulsions and the thought (for example, someone may repeatedly check the front door to relieve anxiety of intrusive thoughts of killing their children) – presenting a problem with the belief criteria. Further, checkers can report that the act of checking in itself can be more unpleasant than their anxiety – thus the checker may not desire the outcome of the behaviour.

(4)

4

Thus, a goal-directed explanation of compulsive checking does not account for all features of compulsive checking.

In contrast, a habitual account of compulsive checking can account for such limitations to the goal-directed account. This is because habitual behaviour is a result of

environmental stimuli (S) being so strongly with associated responses (R) that the presence of a stimuli will automatically activate the response in the motor system without the need to consider the outcome of the behaviour (de Wit & Dickinson, 2009). This automatic and

outcome-insensitive nature of habit means that it is able to occur independently of the belief and desire criteria, thus accounting for these limitations to the goal-directed account. A behaviour may begin as goal directed; a person behaves a certain way to obtain some valued outcome. However, in accordance with Thorndike’s Law of Effect (Thorndike, 1911), when a behaviour (R) leads to a positive outcome (O) this strengthens the association between the behaviour (R) and the environmental cues present at the time (S). Thus, every time the person obtains the outcome of their goal-directed behaviour, this acts as a reward which strengthens the S-R association. In this way as a goal-directed behaviour is repeated it becomes a habit (Killcross & Coutureau, 2003; Tricomi, Balleine, & O’Doherty, 2009).

As behaviour repetition is a prominent feature of compulsive checking, it is clear that checking behaviour could begin as a goal-directed avoidance behaviour which is repeated frequently enough that it becomes a habit. In the same way as S-R associations are positively reinforced by an appetitive outcome in an approach behaviour, the sense of relief from non-occurrence of an aversive event can positively reinforce avoidance behaviour (de Wit & Dickinson, 2009). In OCD for instance, an intrusive thought of burglary in a certain

environment (S) may be followed by checking the locks (R) and the relief of removing the threat of burglary represents the outcome (O). The thought of burglary arises (S); the person desires removal of the threat (O) and knows that making sure the door is locked (R) will achieve this. The sense of relief (O) from pairing the thought of burglary with the checking the locks acts as a positive reinforcer, which increases the association between the thought environment (S) and checking response (R). After repeatedly reinforcing this S-R association, it becomes strong enough that behaviour is now automatic. The environmental context in which the intrusive thought arises (e.g. parts of the checkers home) (S) now automatically evokes the checking (R) without need for the thought of burglary or sense of safety. The person may no longer desire the

(5)

5

sense of relief, nor know that they will obtain such of relief from checking the locks; nevertheless, the behaviour will occur as is now a habit and thus outcome insensitive.

There is substantial experimental evidence for such outcome insensitivity after behaviour repetition through ‘revaluation’ task paradigms. These paradigms assign outcome value to a response through rewards or punishments for that specific behaviour. This outcome is later manipulated (‘revalued’) to examine whether behaviour changes in accordance with the changed outcome value. When outcome value of the behaviour is reduced/increased but a participant continues to respond in the same way, this suggests that the behaviour is insensitive to outcome value and thus a habit has been formed (e.g. Valentin, Dickinson, & O’Doherty, 2007). Tricomi et al. (2009) rewarded participants with food for responding to images with specific key presses. After a training phase, participants satiated themselves on the food to reduce the outcome value of the response. Participants who had received short training responded less to the images which had been paired with the devalued outcome. However, participants who had received longer training continued to respond in the same way after the images were devalued, despite the reduced outcome value of the behaviour. This suggested that their behaviour had moved from goal-directed to habitual as a result of repetition. Using a similar experimental paradigm in a clinical setting, Gillan et al. (2013) trained healthy and OCD participants for either long or short periods. Participants would learn to avoid electric shocks to either wrist by pressing a

corresponding foot pedal in response to a light. When the threat of shock was removed (i.e. the outcome was revalued), participants with OCD continued to respond despite the absence of a threat. This effect was stronger after longer training, suggesting increased reliance on avoidance habits after more behaviour repetition compared to controls (Gillan et al., 2013; for related study see Gillan et al., 2011). This experiment demonstrated a transition from goal-directed to habitual behaviour in avoidance behaviour. The present study investigates similar avoidance habit

formation in the context of compulsive-like checking behaviour.

In our original example of returning to check your front door after leaving in the morning, it is the door-locking habit that underlies the ease with which you lock the door, however it is the resulting meta-cognitive lack of confidence in attention and memory (i.e. “did I do that properly?”) that drives you to return and check the lock. These thought processes occur because repeated behaviour becomes more and more automatic as habits are formed

(Verplanken & Aarts, 1999); we begin to rely on stimulus-response associations rather than conscious processing (Miller & Escobar, 2004) so we do not need to explicitly attend to these

(6)

6

well-rehearsed behaviours (e.g. we don’t need to think much about locking the door). This process is adaptive as it reduces the cognitive load of everyday tasks (Aarts & Dijsterhuis, 2000) – for instance, not having to concentrate on locking our door frees up our cognitive resources so that we are able to talk to our neighbor at the same time. Without such habits, daily life would require considerably more effort. However, such automaticity means that the agent pays less attention to performing an action, which can cease to be adaptive when they are unable to recall the action later on (e.g. “Did I remember to lock the door when I left this morning?”). Thus, habit formation from repetition may play a large role in compulsive checking, with the resultant attention mistrust and memory uncertainty exacerbating the repetitive nature of the checking. This leaves an unfortunate paradox for the compulsive checker, whose repeated checking makes them more uncertain about the event, which in turn drives them to check again, thus

perpetuating the cycle.

Indeed, experimental evidence has suggested that well-practised actions can lead to uncertainty around memory of the action as well as distrust in how much attention the checker paid to their action (described collectively as ‘cognitive distrust’). For instance, Hermans et al. (2008) examined the effects of repeating compulsive and neutral behaviours in OCD participants and matched healthy controls. The compulsive behaviours were actions known to provoke anxiety in the OCD participant and the neutral behaviours were not expected to provoke anxiety or avoidant responding in the OCD participant. Repeated checking led to distrust and less certainty in how much attention the participant paid to the act and less certainty in the memory of the act. This effect was heightened in participants with OCD and was specifically related to checking behaviour rather than neutral behaviours. Van den Hout & Kindt (2003) demonstrated a similar effect in participants who repeatedly turned gas stoves on and off in a computer

simulated checking task. Thus, repeating a behaviour has been demonstrated to lead to cognitive distrust, which may necessitate and thus perpetuate compulsive checking.

Attention mistrust and memory uncertainty from behaviour repetition is typically explained as a switch from episodic to semantic processing (Hermans et al., 2008; van den Hout & Kindt, 2003), however an increased reliance on stimulus-response associations from forming habits could equally explain this cognitive distrust phenomena. Previous research has yet to address this question. So far separate experimental paradigms have been used to demonstrate the effects of repeated behaviour on cognitive distrust (e.g. van den Hout & Kindt, 2004) and habit formation (e.g. Gillan et al., 2013). Therefore, the present study aims to establish an experimental

(7)

7

paradigm that allows us to investigate the effect of behaviour repetition on memory uncertainty, attention mistrust and habit formation. We expect that behavioural repetition will lead to habit formation and cognitive distrust. To investigate these constructs, the present study used a computerised operant avoidance task.

Thus far, operant avoidance tasks have required participants to choose between two responses – for example Gillan et al. (2013) required participants to respond to lights using one of two foot pedals. Arguably this does not reflect reality, where an individual may be faced with multiple response options in the presence of a given stimulus. Therefore, to better capture the real-life decision making process, the present study introduces a novel method of responding to the aversive stimuli: namely a dial with 36 possible positions (or response options), four of which are correct for the corresponding stimuli.

In the task itself participants acted as a ‘spider guard’ and learned to lock spiders in different cages in order to avoid losing points from a total score (corresponding to a real

financial reward). Participants were presented with an on-screen picture of a cage with their score above and circular dial below. This image represented the stimulus (S). They had to respond to this by clicking around the dial to find the correct position that would ‘lock’ that specific cage (R). They lost one point for every mouse click, but three points if they failed to locate the correct position in time. Furthermore, on these incorrect trials, an amplified image of a spider appeared on the screen to symbolise its escape from the cage, this was accompanied by a loud aversive noise. In contrast, if participants succeeded in locking the cage in time, they heard the sound of a lock falling in to place, no points were deducted from their total score and they were shown a small image of the spider inside the cage. This feedback represented the outcome (O) of

responses. After this training phase we asked participants to report attention mistrust. Similarly, participants were asked to respond to cages with no feedback and asked memory uncertainty questions (see ‘Method’ for detail). Following this, to test for habit we looked at participants’ outcome-insensitivity. We devalued the outcome of the learned responses by telling participant that there was an insect infestation and that they should now leave one kind of spiders cages unlocked. If they locked this spider up they would now loose points. Inability to inhibit responding in accordance with this changed outcome suggests outcome insensitivity and suggests that the responses are habitual. We expected that this habit formation would be inversely related to attention mistrust and memory uncertainty. Finally, participants were told that another spider guard may have left some cages unlocked and they should check them but

(8)

8

only when necessary. This gave participants an opportunity to check locks unnecessarily, offering an exploratory measure of compulsive-like checking behaviour.

We aimed to perform exploratory analyses of learning during the task as well as subsequent measures of goal-directed and habitual responding, attention mistrust, memory uncertainty, and a measure of compulsive-like checking behaviour. This enables us to establish whether our experimental paradigm is suitable for carrying forward to further investigate how the putative automaticity-cognitive distrust cycle may underlie compulsive checking.

(9)

9

Method

Design

The present study used within-subjects design. All participants underwent an identical four-phase task. This began with 30 blocks of 8 trials of training in a computerised operant avoidance task, followed by three further phases measuring subsequent memory certainty and attention trust, R-O, S-O contingency knowledge, habit formation and an exploratory measure of compulsive-like checking.

Participants

31 participants (ages 18-30, 10 males) were recruited from within the University of Amsterdam. Participants received ten Euros for participating as well as the financial reward earned during the experimental task (1 euro for every 200 points retained).

Stimuli and Materials

Questionnaires

Sample characteristics were examined prior to participants’ arrival at the lab, through

questionnaires administered online through Qualtrics. The Fear of Spiders Questionnaire (FSQ)

(Szymanski & O’Donohue, 1995)1_{was used to measure spider fear and the}

Obsessive-Compulsive Inventory – Revised (OCI-R) (Foa et al., 2002)2_{was used to measure obsessive}

compulsive traits. Both questionnaires were in Dutch. These data are analysed in a separate report by Casper de Wit.

Task stimuli

A computerised operant avoidance task was administered through Neurobehavioural Systems Presentation (version 17). This task required participants to act as a ‘spider guard’, where they learned the correct responses on an on-screen dial to avoid an aversive outcome. Examples of stimuli presented during the task are illustrated in Figure 2.

Participants were presented with an image of a cage with a score above and a circular dial below. The number appearing as the score varied depending on a participant’s performance (see ‘Procedure’ for details).

1_{Cronbach’s alpha = .91(Muris & Merckelbach, 1996)}

(10)

10

Four different coloured cages (yellow, red, blue and green) were used. In each trial participants were able to respond to the stimuli with a mouse by clicking on the dial image beneath the cage. The dial was split into quadrants by thin lines, with one thick line stemming from the centre to the 12 o’clock position on the outer rim of the dial. This thick line would immediately jump to the dial position that participants clicked on. There were 36 possible positions spread equally across the dial. Each cage had a corresponding dial position that was randomly assigned by the program at the start of the task.

Positive feedback in the task was a locking sound delivered to participants through headphones, an image of a spider in a cage and no change in the score. Negative feedback was a 105DB startle probe delivered through headphones, a three-point reduction in score and an augmented image of a spider coming towards them (see Figure 1). Spider images were either the forked pirate spider (spider A) jumping tan spider (spider B). Each spider was contained in two of the four cages. The only exception to this was during the demo phase, where a black cage containing a wasp spider was presented.

Visual Analogue Scales

Several 100mm visual analogue scales (VAS) ranging from “not at all” to “very much” were used during the task. Participants were able to respond to these by using the mouse to drag a slider along the scale. Participants reported their stress (adapted from Schwabe & Wolf (2009)) at the start and end of the task. Similarly, during the cognitive distrust and R-O, S-O phases, a VAS was presented after each cage for participants to rate how certain they were that they had selected the correct cage position (questions adapted from van den Hout & Kindt, 2003; 2004) (see Figure 3 for example). Immediately following the final acquisition trial, participants were asked to rate the statements about attention mistrust (Hermans et al., 2008) as well as additional measures of perceived cognitive effort, perceived task difficulty and a subjective measure of perceived automaticity. Finally, after the checking phase a subjective measure of urge to check was used.

(11)

11

Procedure

Participants were given an information sheet which explained that the purpose of the experiment was to investigate learning processes and described the task. This said that the participant would be required to lock on-screen cages with a mouse. If they were unable to do this in time the spider would escape and they would lose points from the ‘salary’ that they would receive at the start of the task. This ‘salary’ corresponded to a real financial rewards that they would receive at the end of the task. Having given written consent, the task began (see Table 1 for summary of phases). Before receiving instructions to continue, participants were asked to report how stressed they felt on a VAS (see ‘Visual Analogue Scales’). Following this, participants were informed on-screen that they would receive a starting salary of 2000 points (worth 10 euros). To keep this they should learn to lock cages correctly. They would know they had done this as they would hear a locking sound. Because ‘time is money’, their boss would take one point from their salary for every lock attempt they made and a further three points if they did not identify the correct lock position. Further, they were told that they would first have a chance to practice, and should use the mouse to click on the rim on the dial to find the correct position is.

Table 1 - Task phases of the 'Spider Guard Game'

Task Phase Summary Feedback Duration

1) Acquisition Participants learned to respond to four different colour cages to avoid aversive outcome.

Positive and negative

30 blocks 2) Cognitive

distrust

Participants asked series of questions about attention mistrust, perceived cognitive effort, perceived task difficulty and perceived automaticity during acquisition phase. Following this, cages were presented with no feedback. Participants respond accordingly and report subsequent memory certainty for each cage

None 1 block

3) R-O, S- O knowledge

Based on the contingencies of the acquisition phase, participants are asked to match each dial quadrants (R) to its accompanying spider (O) and each cage to its accompanying spider (O). Participants reported certainty after each selection.

None 1 block

4) Slip-of-Action (SOA)

Participants are asked to respond as normal to cages containing one spider, but refrain from responding to other spider. Then vice versa.

None 8 blocks - each spider is devalued for 4 blocks. 5) Checking Participants are told that they should respond to dials that are

not already in the correct position. Unnecessary responses give an exploratory measure of compulsive-like checking.

(12)

12

Demo Phase

Participants first underwent a demo phase to become familiar with the task. The instructions informed participants that they should not be concerned about making mistakes during this phase as their salary would reset afterwards. Participants pressed the ‘space’ bar when they were ready to begin.

A blank screen ITI was presented for a random time interval between 500ms and 1000ms (this varied in each trial), followed by an image of a black cage with the dial in the centre of the screen and the ‘salary’ at the top of the screen (S). At the start of the task this salary read “2000”. Participants clicked on various positions on the dial to learn through trial-and-error which position on the dial would ‘lock’ the sider up (R). Participants had 2000ms to do this and lost one point for every click. If participants clicked on the correct position they received positive feedback: a locking sound, a small image of a spider in a cage, and no further loss of points. If participants could not find the position in time they received negative feedback: a startle probe sound, an augmented image of a spider on the screen and a three-point reduction in points (see ‘Task stimuli’ for details). This feedback represented the outcome of the response (O). In this phase, the same cage was presented for one block (eight trials) and the correct dial position was at 2 o’clock for all participants. To avoid biasing future learning, the wasp spider and black cage used in the demo were not used in the following phases of the task. This phase lasted two blocks. The score at the end of the demo phase was not carried forward to the remaining phases.

Acquisition Phase

Participants received additional instructions at the start of this phase informing them that the real task would be more difficult and take longer as there are four different coloured cages, each with its own unique lock position. Their job would now be to lock as many cages as possible by learning the correct positions through trial-and-error. Two kinds of spiders would be contained in

the cages (two with one kind of spider, two with another) and they should learn which spider

(13)

13

belongs with which cage. Again, they would lose three points for every escaped spider and one point for every locking attempt.

This phase was similar to the demo phase, with the exception that participants were now expected to learn the correct dial positions for four different coloured cages which contained two different spiders: Spider A and B. These spiders were randomly assigned to two of the cages by the program at the start of the task. Participants’ salary was reset to 2000 points at the start of this phase and decreased across the trials.

As in the demo phase, an ITI was presented, followed by a cage with a dial and remaining points displayed above (See Figure 2). Participants had 2000ms to respond to the cages. One coloured cage was presented per trial and there were eight trials in one block, thus each cage appeared twice in a block. The order of presentation was randomised by the program at the start of the task. Participants underwent 30 blocks of eight trials with a two-minute break after completing 10 blocks and 20 blocks of training. They were able to skip these breaks by pressing the space bar.

Memory Phase

Immediately following the final trial of the acquisition phase participants were asked to rate the following VAS statements (see ‘Visual Analogue Scales’) measuring attention trust, perceived cognitive effort, perceived task difficulty and perceived automaticity:

Attention trust

1) “While I was making sure the cages were locked I was not distracted” 2) “If I made mistakes during my job as a spider guard, I noticed them”

Perceived Cognitive Effort

1) “I did my best as a spider guard”

Perceived Task Difficulty

1) “I found my job as a spider guard difficult” 2) “I had to concentrate to lock the cages correctly”

Perceived Automaticity

1) “I locked the cages automatically”

After answering these questions participants were instructed that they would be presented

with more cages that they should continue to lock, however they would no longer receive feedback. They were presented with the same images of a cage and dial, however this time the score read ‘????’ (see Figure 3), there was no time limit and they received no feedback for their responses. Participants first choice of position was accepted and they were immediately asked to

(14)

14

rate the statement “How sure are you of your answer?” in order to measure memory certainty. This phase lasted for one block.

Response-outcome, stimulus-outcome knowledge phase (R-O, S-O)

Following the memory phase, the R-O, S-O phase measured participants’ certainty in their memory of response-outcome (R-O) and stimulus-outcome contingencies (S-O). First

participants were shown a correct dial response (R) on screen next to images of both spiders (O). One-by-one they were shown each response in this way and asked to click the correct spider corresponding to that response. After selecting a spider in each trial, participants were asked to rate the VAS statement: “How sure are you of your answer?” to establish memory certainty around R-O contingency knowledge. After matching each response to an outcome, participants were presented with each individual cage (S) and asked to match it to a spider (O), rating the certainty VAS each time in order to measure memory certainty of S-O contingency knowledge.

Slip-of-action (SOA) phase

As habit is characterised by outcome insensitivity, to test for habit formation this phase devalued the outcome of certain responses to check for ‘slips of action’ – continued responding despite reduced outcome value. Participants were told that there had been an outbreak of insects and they should now refrain from locking this spider up as it should be free to eat the insects. If they continued to lock the cages associated with this spider they would now loose three points. Continued responding to the spider after the outcome value is reduced in these test trials suggests habitual responding. Further, participants were told that their boss would be testing them and would occasionally give them a cage that was already in the lock-position. If this happened they had to click on the lock – regardless of which spider belonged with that cage. The purpose of these active baseline trials was to keep responses active, so participants could not rely on completely avoiding certain dial quadrants to inhibit responding to devalued trials.

(15)

15

There were eight randomly ordered test trials within each block of this phase (see Figure 4 for examples). As with the memory phase,

participants were presented with a cage, a dial, and a score reading ‘?????’. Participants had 1500ms to click the correct dial

position. In each block, four active baseline trials were interspersed within these eight slip-of-action test trials were. In these trials the cage and dial would appear as normal, however after 150ms the dial would jump to the correct position and participants would have to click on it, regardless of whether the spider was revalued. Each colour of cage was used in one active baseline trial. The order of trials was randomised by the program.

Checking phase

Participants were told that another sloppy guard may have left some cages unlocked and they have the opportunity to check these. They would not see their score, and would not receive any feedback, but this time they would hear the locking sound if they clicked the correct position. They would still lose three points if a spider escaped, and they would still lose one point for every response made. A ‘check’ would be if a participant clicked on a dial in its position, and further and ‘unnecessary check’ was when they did this when the cage was already locked. This offered an exploratory measure of compulsive-like checking behaviour. This phase lasted for two blocks, with each cage presented four times in total. In half of these trials the cage was presented with the dial already in the lock position. In the other half of trials the dial was in the position immedietly neighbouring the correct position; once to the left and once to the right.

Having responded to these trials, participants reported their subjective urge to check by rating the following statement on a VAS: “During the final stage of the task, did you feel an urge to check that the cage was locked, even when you knew it was locked correctly?”.

Participants were also asked if they had adjusted their reactions to spiders based on the change in the number of points for the spiders in the SOA phase.

Following this, participants were asked to report how stressed they were and how motivated they

Figure 4 - Slip of action phase trials. From top to bottom: devalued trial, valued trial, active baseline trial.

(16)

16

felt during the task on a VAS. They were then presented with their final score which was converted into a financial reward.

(17)

17

Results

The data were examined phase-by-phase to investigate the constructs individually and assess whether the task phases do indeed reflect the constructs put forward in the proposed paradigm. Subsequently, the relationships between variables were examined in relation to our automaticity-cognitive distrust hypothesis. Initially descriptive findings are presented, followed by exploratory correlational analyses of variables. Statistical tests and correlational analyses were assessed with an alpha level of 0.05.

Data Preparation

One participant was completely omitted from analyses due to erroneously having only received 10 blocks of training during the acquisition phase. Thus, 30 participants were included in analyses.

Distributions

No behavioural performance variables were normally distributed, with the exception of reaction times during the acquisition phase. These normal variables were examined using non-parametric tests.

Perceived cognitive effort, perceived task difficult, perceived automaticity and memory certainty were all normally distributed and examined using parametric tests.

Acquisition Phase

To examine participant learning we assessed accuracy and reaction time (RT) per block during this training phase. Participants were able to make as many responses as they wanted within the trial time limit. Accuracy was measured in terms of whether the participants’ last response in a trial was correct or not, regardless of the total number of responses made. For each block, we averaged the number of correct trials out of the total number of trials that participants made responses for. This was the accuracy percentage per block. RT was measured as the time taken for participants to make their first response after trial onset. As with accuracy, we averaged RTs over trials that participants made responses for, giving us a mean RT for each block.

First of all, Figure 5 shows the mean number of trials that participants responded to per block. Participants tended to respond to a large proportion of trials per block (M = 98.9%, SD =

(18)

18

3.7). There appeared to be a slight upward trend in the first 3 blocks, however this fluctuated between a mean of 97.5% and 100% for the remainder of the phase.

Figure 5 - Percentage of trials responded to per block. Error bars indicate standard error of the mean.

Figure 6 - Percentage accuracy across acquisition phase. Error bars indicate standard error of the mean.

As illustrated in Figure 6, there was an upward trend in the percentage of correct trials.

Mean accuracy increased with training, suggesting that participants were learning to find the correct lock positions for cages in time. Similarly, standard deviations appear to decrease across trials, suggesting less variation in accuracy across participants with more training. A Wilcoxon

90 95 100 105 110 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 % t ri als resp on de d to Block

Acquisition: Mean percentage of trials responded to per block

0 10 20 30 40 50 60 70 80 90 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 P ercen tag e of C orrect R es po ns es Block

(19)

19

signed rank test suggested that accuracy was significantly better in the final block (M = 91.96, SD = 14.52) compared to the first (M = 8.48, SD = 12.75); Z = -4.834, p<.001.

There was a similar upward trend in RTs (see Figure 7). This is surprising in light of theory suggesting automaticity, and presumably more speed in responses, with better learned actions. However, here participants appeared to be taking longer to respond with more training and better accuracy. A paired samples t-test suggested that reaction times were indeed

significantly longer during the final block of acquisition (M = 90.298.5, SD = 15.423) than the first block of acquisition (M = 8.333, SD = 12.428); t = -4.834, p<.001.

Figure 7 - Mean RTs across the acquisition phase. Error bars indicate standard error of the mean.

Memory phase

Participants’ perceived cognitive effort, perceived task difficulty and attention mistrust during the acquisition phase was measured by responses to a number of VAS scales, ranging from 0 to 100 (see ‘Method’ for question details). For the four trials with no feedback that followed these VASs, we measured the number of trials where participants gave the correct response as an accuracy percentage, the participants RT (while in the acquisition phase the participants last response was used, in this phase participants were only able to make one response). Subsequent VAS responses to these trials were averaged for a measure of memory certainty.

800 900 1000 1100 1200 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 M ea n R T ( ms) Block

(20)

20

VAS response data were missing for two participants, and thus 28 participants were included for analyses of attention trust, memory certainty, perceived cognitive effort and perceived task difficulty.

We obtained perceived cognitive effort and perceived automaticity scores from participants’ responses on these single items. Perceived task difficulty score was created from averaging the responses of the two difficulty items. A higher score corresponded to higher endorsement of these constructs.

To transform attention mistrust responses so that a higher score also indicated higher attention mistrust, responses to the attention mistrust VAS’ were subtracted from 100. An average score for the two questions was then established. Thus again, a higher score indicated higher attention mistrust.

In a similar way, memory uncertainty score was calculated from participants VAS responses after each no-feedback trial. To change the scale to a measure of memory uncertainty rather than certainty, each VAS score was subtracted from 100, then aggregated and averaged. Thus, on this scale a score of 0 indicated memory certainty and 100 indicated memory

uncertainty.

We also measured participant accuracy in responding to the no-feedback memory trials. We did this through calculating the percentage of these trials in which participants selected the correct response. Reaction time was averaged across trials in this phase, regardless of whether the responses made were correct or incorrect. Compared to the acquisition phase, accuracy dropped in this phase (M = 70.5, SD = 23.6) and reaction time increased (M = 1533.3, SD = 383.3).

Table 2- Means and standard deviations of VAS responses

Construct Mean (SD)

Perceived Cognitive Effort 75.9 (15.0)

Perceived Task Difficulty 64.6 (17.2)

Attention Trust 38.6 (15.0)

Memory Certainty 25.7 (18.7)

(21)

21

Table 2 shows that participants did report some attention mistrust and memory

uncertainty following behaviour repetition. Overall, perceived task difficulty and cognitive effort seemed fairly high on average, suggesting that participants were challenged by the task. However, it does not appear that the task was too difficult or demanding as accuracy scores demonstrate that participants were learning and responding correctly during the acquisition phase.

Interestingly, participants reported perceiving some automaticity in their responses (M = 49.6,

SD = 23.8). All of these VAS measures showed some variation in responses, as expected from

such a task.

As additional exploratory analyses, we looked at correlations between the variables at this phase. As attention mistrust and accuracy during the memory phase were not normally

distributed, these were examined using Spearman rank correlations. Remaining variables were examined using Pearson correlations.

Memory uncertainty correlated negatively with mean accuracy during the acquisition

phase (rs[26] = -.50, p = .007) and mean percentage accuracy during the memory phase (rs[26] =

-.763, p = .000). This mean percentage accuracy during the memory phase was also inversely

related to reported task difficulty (rs[26] = -.526, p = .04). As one would expect, mean percentage

accuracy during the memory phase was positively related to accuracy during the acquisition

phase (rs[28] = .433, p = .014). These relationships are likely due to better knowledge of the

correct cage positions; or deeper S-R-O contingency knowledge.

We expected that both attention mistrust and memory uncertainty would be related; paying less attention to an event would likely lead to poorer recollection of the event. However

no association was found between attention mistrust and memory uncertainty (rs[26] = -.78, p =

.692). Further, we hypothesised that both attention mistrust and memory uncertainty would arise from habit formation (i.e. behavioural repetition). In light of this we expected to see a positive correlation between both of these variables and perceived automaticity, a subjective measure of habit formation. However, perceived automaticity was not significantly related to either attention

mistrust (rs[26] = -.252 , p =.197) nor memory uncertainty (rs[26] = -.331 , p =.085).

This perceived automaticity did, on the other hand, demonstrate relationships with other variables. In fitting with theory suggesting that repeating behaviour leads to habit formation, and thus increased automaticity, our measure of perceived automaticity correlated positively with

(22)

22

mean accuracy across the acquisition phase (rs[26] = .540, p = .003). This suggests, as expected,

as participants were learning during the acquisition phase, they felt that their responses were become more automatic. Similarly, perceived automaticity was inversely related to perceived cognitive effort (r[26] = -.502, p = .07) and perceived task difficulty (r[26] = -.602, p=.001). This is also expected in light of theory suggesting that automaticity reduces the cognitive load of tasks, thereby reducing the effort required to perform an action and so the agent perceives the action as easier.

Response-Outcome/Stimulus-Outcome knowledge phase (R-O, S-O)

The R-O/O phase was in place in order to test participants’ explicit knowledge of R-O and S-O contingencies. Due to a programming error, R-S-O and S-S-O accuracy was not recorded, however we were able to examine participant certainty in their memory of R-O and S-O

contingencies. We aggregated and averaged memory certainty for S-O and R-O contingencies as separate constructs. A high score indicated more certainty in contingency memory, on a scale from 0 (no certainty) to 100 (high certainty). Participants were slightly more certain in their memory of S-O contingencies (M = 85.2, SD = 23.7) than R-O contingencies (M = 83.0, SD = 24.7), however these were very close.

Neither R-O nor S-O contingency memory certainty were normally distributed, thus non-parametric Spearman correlational tests were used for exploratory analyses. R-O and S-O

contingency memory were strongly associated with each other (rs[26] = .90, p = .00). This was to

be expected, as the acquisition phase involved learning S-R-O contingencies, which form the basis of both S-R and R-O contingency knowledge. In line with this was the findings that mean accuracy percentage during the acquisition phase was positively associated with both R-O

contingency memory certainty (rs[26] = .534, p= .003) and S-O contingency memory certainty

(rs[26] = .599, p = .001). And further, this underlying contingency knowledge may also explain

why there was a positive relationship between R-O contingency certainty and mean percentage

accuracy during the memory phase (rs[26] = .441, p = .019)

Slip-of-Action (SOA) Phase

Five participants who had demonstrated less than 75% accuracy in final two blocks of the acquisition stage were not included in any SOA phase analysis, due to potentially poorly learned

(23)

23

S-R-O contingencies which could influence responding in this phase. Thus, 25 participants were included in initial descriptive analysis of SOA variables.

In the SOA phase, participants were required to inhibit responding to certain spiders cages (devalued trials) and continue to respond to the other spiders cages (valued trials). Further, this phase required participants to respond to active baseline trials by clicking on the dial when it jumped into position, regardless of the value of the spider. This trial type was primarily in place to keep responses active, however we also included responses to these trials in analyses. All participants were presented with 32 valued trials, 32 devalued trials and 32 active baseline trials. This was used to create response percentages for the different trial types; for instance, the number of valued trials where participants made responses, as a proportion of the total number of valued trials they were presented with.

We were able to examine participants’ responding to valued and devalued trials to

examine relative habitual control, as reflected in a failure to inhibit responding on devalued trials. In other words, these ‘slips of action’ suggest habit formation. Furthermore, to give a

proportional measure of goal-directed behaviour, we calculated a balance index by subtracting percentage of responses to devalued trials (slips of action), from the percentage of responses to valued trials. A high score suggests an ability to inhibit responding to devalued trials – suggesting goal directed behaviour; a low score suggests inability to inhibit responding to devalued trials (i.e. that the person is making ‘slips of action’) – suggesting habitual responding. The variables measured during this phase were not normally distributed and thus non-parametric tests were used.

As shown in Table 3, participants were making responses to all trial types, and further, were responding more to valued trials (M = 85.5, SD = 15.1) than devalued trials (M = 15.4, SD = 26.2). In line with this observation, a Wilcoxon-signed-rank test showed that participants responded significantly more to valued than devalued trials (Z = -4.256, p <.001). Aside from indicating that participants understood the task instructions, as they were able to distinguish between the different trial types, this also suggests that participants still had some goal-directed control – as reflected in largely the positive balance indices (M = 70.1, SD = 36.3). However participants were still making some slips of action, showing possible outcome insensitivity, as expected after such behaviour repetition.

(24)

24

Table 3 - Means and standard deviations of SOA phase trials

Exploratory Spearman rank correlational analyses were conducted to examine relationships amongst variables within the paradigm. An additional participant who had responded more for devalued than valued trials was removed from these analyses due to the possibility that they had not understood the task correctly. This participants VAS data was also missing. Thus, 24 participants were included in correlational analyses of behavioural variables and due to missing data, 23 participants were included in the correlational analyses which involved VAS variables.

Percentage of responses to valued trials were positively associated with mean accuracy

during the acquisition phase (rs[22] = .63, p = .001) and self-reported automaticity (rs[22] = .478,

p = .021). Considering self-reported automaticity as a subjective measure of task performance,

and accuracy during acquisition as a more objective measure, these findings are as one might expect; those with better knowledge of responses made more responses to valued trials. In

addition, balance index was positively related to S-O contingency memory certainty (rs[22] =

.420, p = .046).

Percentage of responses to valued trials was also inversely related to percentage of

responses to devalued trials (rs[22] = -.431, p = .035). Participants who made more responses to

valued trials tended to make less responses to devalued trials and vice versa. This may reflect a tendency towards either goal-directedness or habit, however this is not clear as this finding may also be due to such factors as task understanding. It may be worth noting however, that there was no significant positive relationship between active baseline responding and responding for devalued trials. Such a relationship might have suggested responding for devalued trials was in fact more of a general willingness to respond; participants were simply clicking more, regardless of trial type. However, having found no such relationship might suggest that responses to devalued trials are indeed slips of action rather than a by-product of willingness to respond.

Mean percentage (SD)

Valued Responses 85.5 (15.1)

Devalued Responses 15.4 (26.2)

Active Baseline Responses 77.9 (17.5)

(25)

25

We performed correlational analyses to investigate the hypothesis that habit formation would give rise to cognitive distrust. Responses to valued trials were not significantly associated

with memory uncertainty (rs[21] = .130, p = .553) or attention mistrust (rs[21] = -.357, p = .094).

Responses to devalued trials were also not related to memory uncertainty (rs[22] = .227, p = .297)

or attention mistrust (rs[22] = .127, p = .562). Thus, no relationship was found between habit and

cognitive distrust.

Checking Phase

In this phase we measured unnecessary checks on a dial following the avoidance learning in prior phases; this was an exploratory measure of compulsive-like checking. Participants had been presented with the dial which was already in the lock position in half of the trials. An unnecessary check was when a participant responded to a dial which was already in the lock position by clicking on that same position. Thus, for each participants, the number of correct first responses made on an already-locked trials were used as ‘number of checks’. As there were eight already-locked trials, participants were able to make a maximum of eight unnecessary checks. Additionally, participants VAS rating for the checking question (see ‘Methods’ section) was used as an exploratory self-report measure of urge to check unnecessarily, a higher score corresponded to more of an urge to check. Neither checking variable was normally distributed and thus Spearman rank correlation tests were used.

The number of unnecessary checks that participants made ranged from 0-8, with a mean of 3.2 (SD = 2.7) of a possible maximum of 8, suggesting that some participants were indeed engaging in unnecessary checking. Similarly, participants also reported experiencing an urge to check (M = 59.1, SD = 30.3). This suggests that indeed the paradigm may be sufficient in eliciting unnecessary checking behaviour. To investigate whether these unnecessary checks may be a reflection of lack of S-R knowledge (i.e. whether participants clicked because they did not know if the dial was already in the correct position), we checked for a significant relationship between number of unnecessary checks and variables which required good S-R-O knowledge. However, unnecessary checks showed no significant relationship with accuracy during the training phase, the memory phase and S-O memory confidence. On the other hand however,

number of unnecessary checks was positively associated with reported urge to check (rs[26] =

.614, p = .001), this could indicate that the number of unnecessary checks were due to an urge, rather than lack of S-R knowledge.

(26)

26

Exploratory correlational analyses of unnecessary checks and urge to check with other variables were used to further examine the nature of this unnecessary checking. In light of theory, we had expected that both measures of checking would be related to habit formation and cognitive distrust. However no relationship was found between habit measures (slips of action and self-reported automaticity) or cognitive distrust measures (memory uncertainty and attention mistrust).

(27)

27

Discussion

We presented the novel hypothesis that avoidance habit formation from repeating a behaviour leads to cognitive distrust, and that it is this cycle that underlies compulsive checking. We aimed to establish a new experimental paradigm to examine this which is now discussed in light of our findings.

In the initial acquisition phase participants appeared to show learning. They were able to distinguish between stimuli, locate lock positions and respond accordingly. This was

demonstrated by the increase in accuracy across training blocks, with near-perfect performance by the final block. In this respect, our task paradigm appeared successful.

In light of theory suggesting that increased learning leads to increased automaticity, it seems intuitive that participants would get faster at responding to stimuli with more practice, and thus we did not expect the finding that reaction times increased across acquisition blocks. This may be due to the fact that at the start of training participants must identify one of 36 possible cage positions through trial-and-error, with the risk of the aversive outcomes if they do not manage this. From this perspective it is understandable that the initial responses would be rather frantic. Every time participants identified a correct position they were able to remove a quadrant of the dial (and thus a quarter of responses) from the array of response possibilities, and thus responding would become more systematic and less frantic. After identifying all correct responses (or at least the quadrants containing the positions) perhaps the reduced pressure to hunt for a response as well making the response meant that there was less need for rushing to respond, and thus they were slower in responding. This complicates measuring automaticity using the experimental paradigm. Traditionally, automaticity is linked to faster reaction times, however in this paradigm a fast reaction time with multiple meaningless responses may not indicate automaticity from learning, nor does it parallel the specific behaviour patterns seen in compulsive checking. A solution may be to combine reaction time with the number of responses made during each trial to create a new measure of automaticity. This is worth considering for developing the paradigm further. Reaction times aside, the learning seen during the acquisition phase suggests that the paradigm may be suitable for testing habit.

We test for habit by devaluing the outcome of an S-R-O relationship. As habit is S-R reliance, continued S-R responding after devaluing the outcome (O) (i.e. a ‘slip of action’) suggests that habit has been formed. In our task, participants responded more to trials where O

(28)

28

was still valuable (‘valued’ trials) than when O was not valuable (‘devalued’ trials), suggesting that they understood the task and that their responding was primarily goal-directed. However,

participants were also making slips of action. This suggests that the paradigm is successful in testing for habit formation, however should be treated with caution. Firstly, if a participant did not learn the O correctly to begin with, then participants would not be able to respond

accordingly when O is devalued (regardless of goal-directedness); thus, the participant may make slips of action due to a lack of contingency knowledge rather than because they over-rely on S-R associations. Indeed, we found that better contingency knowledge (S-O) was associated with more goal-directed responding. However, this does not necessarily mean that slips-of-action were due to lack of contingency knowledge. Further, the number of slips of action made in the task was not associated with any other variable that might suggest impaired contingency knowledge (such as accuracy during acquisition).

Secondly, if slips of action made by participants do indeed indicate that they formed habits, we cannot be sure of how to interpret the scale of this. In other words: how many slips correspond to how much habit? Further, based on balance index data, participants still had some goal-directed control over their responses, and yet also reported feeling that they were

‘automatically’ locking the cages. However, this self-reported automaticity was not associated with slips-of-action.

A second, longer, training condition may bring us closer to resolution. This would give the opportunity for more S-R-O learning and build to stronger S-R associations. Training participants across multiple days, such as in the design used by Tricomi et al., (2009), may have the added benefit of allowing for learning consolidation and thus potentially deeper learning. This would enable us to compare current findings with the number of slips of action made as a result of more behaviour repetition and thus stronger S-R associations. This would give a more firm idea of whether participants are indeed undergoing S-R-O learning, as well as the scale of habit formation. Furthermore, as existing experimental paradigms include both long and short training periods (e.g., Gillan et al., 2013), including an additional condition would allow us to compare findings with established paradigms, offering a stronger measure of the utility of our proposed paradigm.

Bringing together research from multiple domains within psychology, we put forward the idea of the automaticity-cognitive distrust cycle; that habit formation from behaviour repetition

(29)

29

would lead to attention mistrust and memory uncertainty. Thus, we expected that memory uncertainty and attention mistrust would arise during our learning task, and that these variables would be associated with habit formation. Indeed, participants did report some uncertainty in memory and mistrust in attention, suggesting some utility of our task paradigm. However, these variables were not significantly associated with habit measures, and thus we were unable to provide support for this hypothesis. This may be because the habit formation was not sufficient to elicit attention mistrust or memory uncertainty (as discussed earlier, we cannot derive

definitive conclusions from the number of slips of action). However, it is also possible that our cognitive distrust measures were flawed. As performance was not perfect during the memory phase it raises the possibility that participants may have interpreted “how sure are you of your answer?” as a question of whether their response was correct, rather than how well they

remembered the response. This might also explain association between memory uncertainty and accuracy during both acquisition and memory phases. The aforementioned suggestion of

creating an additional task condition may resolve these issues with the cognitive distrust

constructs. Assuming that accuracy will improve with more training, perfect performance during the memory phase would bring us closer to establishing whether participants are interpreting the memory question as an accuracy question. Further, assuming that habit formation will increase with more training, due to increased reinforcement of S-R associations, this additional condition would help establish whether the automaticity-cognitive-distrust hypothesis was not upheld due to insufficient habit formation.

We introduced the automaticity-cognitive distrust cycle hypothesis to explain the process underpinning the development of compulsive checking. For this reason we included an

exploratory compulsive-like checking measure in the paradigm. This involved presenting participant with a number of trials with the dials below already in a position, giving participants the opportunity to check if the cages were locked. If these dials were already in the correct lock position and a participant clicked directly on the dial in its position, this suggested that they were checking that it was locked. Many participants did indeed engage in such unnecessary checking, this checking was positively associated with self-reported urge to check and this checking did not appear to be associated with variables that would indicate lack of contingency knowledge.

However, we expected that this compulsive-like checking would be linked with habit formation (slips of action) or cognitive distrust (memory uncertainty and attention mistrust) and this was not the case. As discussed earlier, however, habit formation and cognitive distrust may not have arisen during the task. This suggests that the checking phase may be useful to include in the

(30)

30

paradigm, however more investigation is needed into the nature of the unnecessary checks in this phase. Again, it is possible that including an additional longer training condition in the task will shed more light on the utility of the checking phase, and its relationship with other variables included in the paradigm.

Despite the uncertainty in the paradigms ability to measure the desired constructs, the proposed paradigm has the advantage of flexibility; the task can easily be adapted for

improvements. Elements of the task used in the present study have already been adjusted and improved based on piloting. For example, initially participants were not making slips-of-action as they were able to inhibit responses based on avoiding clicking on specific quadrants. However introducing active baseline trials to keep participant responses active during the SOA phase gives a more dependable measure of S-R reliance. The current acquisition phase length was based on an initial study, however as discussed above, an additional condition with an extended training length during the acquisition phase may improve the task further. Currently, we are unable to closely compare findings from the present paradigm with other studies, as the present study used only one condition whereas other studies tend to include both a long and short training

condition. Including an additional condition may enable this. Moreover, it may be of interest for the task to investigate the effect of using different outcomes on habit formation and cognitive distrust. An image of a spider on a screen is perhaps not as aversive as, for instance a live spider in a jar. This may affect S-R-O learning and have implications for habit formation. This may better parallel the aversive nature of the intrusive thoughts that clinical compulsive checkers experience.

The paradigm put forward in the present study in its current state is suitable for broadly measuring the desired constructs; S-R-O learning, devaluation and outcome-insensitivity, as well as measures of attention mistrust and memory uncertainty. It builds on previous experimental paradigms habit by creating a more complex set of response options which better parallel S-R-O learning in real life. Furthermore, it is flexible enough to introduce and remove elements as needed. With some refining, it will be suitable for testing with a clinical population. In

conclusion, we feel that this paradigm in its current state offers a solid base to develop and use to further explore the automaticity-cognitive-distrust cycle hypothesis.

(31)

31

References

Aarts, H., & Dijsterhuis, A. (2000). The Automatic Activation of Goal-Directed Behaviour: The Case of Travel Habit. Journal of Environmental Psychology, 20(1), 75–82.

American Psychiatric Association (2013). Diagnostic and Statistical Manual of Mental Disorders (Fifth ed.). Arlington, VA: American Psychiatric Publishing.

De Wit, S., & Dickinson, A. (2009). Associative theories of goal-directed behaviour: a case for animal-human translational models. Psychological Research, 73(4), 463–476.

Foa, E. B., Huppert, J. D., Leiberg, S., Langner, R., Kichic, R., Hajcak, G., & Salkovskis, P. M. (2002). The Obsessive-Compulsive Inventory: Development and validation of a short version. Psychological Assessment, 14(4), 485.

Gillan, C. M., Papmeyer, M., Morein-Zamir, S., Sahakian, B. J., Fineberg, N. A., Robbins, T. W., & de Wit, S. (2011). Disruption in the balance between goal-directed behavior and habit learning in obsessive-compulsive disorder. The American Journal of Psychiatry, 168(7), 718–26. Gillan, C. M., Morein-Zamir, S., Urcelay, G. P., Sule, A., Voon, V., Apergis-Schoute, A. M., …

Robbins, T. W. (2013). Enhanced Avoidance Habits in Obsessive-Compulsive Disorder.

Biological Psychiatry, 1–8.

Hajcak, G., Huppert, J. D., Simons, R. F., & Foa, E. B. (2004). Psychometric properties of the OCI-R in a college sample. Behaviour Research and Therapy, 42(1), 115–123.

Hermans, D., Engelen, U., Grouwels, L., Joos, E., Lemmens, J., & Pieters, G. (2008). Cognitive confidence in obsessive-compulsive disorder: distrusting perception, attention and memory.

Behaviour Research and Therapy, 46(1), 98–113.

Killcross, S., & Coutureau, E. (2003). Coordination of actions and habits in the medial prefrontal cortex of rats. Cerebral Cortex, 13(4), 400–8.

Muris, P., & Merckelbach, H. (1996). A comparison of two spider fear questionnaires. Journal of

(32)

32

Schwabe, L., & Wolf, O. T. (2009). Stress prompts habit behavior in humans. The Journal of

Neuroscience : The Official Journal of the Society for Neuroscience, 29(22), 7191–8.

Szymanski, J., & O’Donohue, W. (1995). Fear of Spiders Questionnaire. Journal of Behavior Therapy

and Experimental Psychiatry, 26(1), 31–34.

Thorndike, E. L. (1911). Animal Intelligence: Experimental Studies. New York: Macmillan.

Tricomi, E., Balleine, B. W., & O’Doherty, J. P. (2009). A specific role for posterior dorsolateral striatum in human habit learning. The European Journal of Neuroscience, 29(11), 2225–32. Valentin, V. V, Dickinson, A., & O’Doherty, J. P. (2007). Determining the neural substrates of

goal-directed learning in the human brain. The Journal of Neuroscience : The Official Journal of the

Society for Neuroscience, 27(15), 4019–26.

Van den Hout, M., & Kindt, M. (2003). Repeated checking causes memory distrust. Behaviour

Research and Therapy, 41(3), 301–316.

Van den Hout, M., & Kindt, M. (2004). Obsessive-compulsive disorder and the paradoxical effects of perseverative behaviour on experienced uncertainty. Journal of Behavior Therapy and

Experimental Psychiatry, 35(2), 165–81.

Verplanken, B., & Aarts, H. (1999). Habit, Attitude, and Planned Behaviour: Is Habit an Empty Construct or an Interesting Case of Goal-directed Automaticity? European Review of Social

Avoidance habits and cognitive distrust : a novel experimental paradigm