To Drink or to Switch
Context-relevant Alcohol Set Shifting Measurement and Training: a Pilot Study
J.G. Quist, 6044832
Daily Supervisor: dhr. drs. W.J. Boendermaker
ResMas Supervisor: dhr. prof. dr. R.W.H.J. Wiers
Objective: As explained by models emphasizing the importance of automatic and regulatory processes in addictive behavior, training executive functions such as working memory can decrease problematic drinking behavior. Although it has been shown that set shifting in general is a trainable construct, effects of set shifting training on alcohol consumption have not yet been investigated. Method: Automatic tendencies towards alcohol during set shifting were measured and a set shifting training in the context of alcohol was developed and tested by means of a pilot study. Results: Participants were slower to respond to a non-alcohol rule and to pictures of non-alcohol compared to an alcohol rule and pictures of alcohol. General set shifting ability increased from the first to the last training session of the training task. Discussion: These results indicate the presence of automatic tendencies towards alcohol during set shifting. Furthermore, preliminary results indicate that the developed training task could prove to be a useful method to retrain (control over) these tendencies. Keywords:
Table of Contents
Introduction 4
The Set Shifting Paradigm 6
The Current Research and Hypotheses 7
Method 9
Participants 9
Materials and measures 10
Procedure 18
Results 20
Differences at Baseline 20
Switch Assessment Task 20
Implicit Associations 23
Alcohol Consumption 23
Energy levels, Mood and Task Rating 24
Output Exit Interview 25
Switch Training task 25
Discussion 27
Limitations of the Training Study 29
Conclusion 32
Introduction
The moment someone is confronted with the decision to have an alcoholic drink or a non-alcoholic drink, motivational as well as cognitive processes are activated. When the tendency to drink alcohol is strong, choosing the non-alcoholic alternative does not only require willpower, but also a shift in mindset. A mindset can be defined as a temporary configuration of the mental system, which is associated with the readiness to carry out a particular activity (Rahamim, Bar-Anan, Shahar, & Meiran, 2013). The ability to shift flexibly between mindsets, also defined as cognitive flexibility, is considered to be one of the core executive functions (Miyake et al., 2000). For individuals who have strong impulses to drink alcohol, it can be difficult to shift away from the mindset of having an alcoholic drink. As a result, maintaining control over drinking behavior could become problematic (Noel et al., 2005).
Many individuals struggle with the urge to drink excessive amounts of alcohol. In the Netherlands, for example, almost 10 % of the people aged above 12 can be considered heavy drinkers. Among 20-30 year olds, this is even 18 % (Trimbos Institute, 2013). Of individuals aged between 18 and 64, 0.3 % to 1.2 % can be considered alcohol addicted (De Graaf, Ten Have, Tuithof, & Van Dorsselaer, 2012). Besides having large negative consequences for physical and mental health, excessive alcohol (mis)use is a major risk factor in causing injury-related disability and premature death (Rehm, Shield, Rehm, Gmel, & Frick, 2012). Therefore, the development of methods aimed at prevention and reduction of alcohol-related problems is highly prioritized.
Over the past two decades, a new approach to decrease problematic drinking has become increasingly popular. In this approach, cognitive and automatic processes are used to explain the mechanisms behind addictive behavior. In a highly influential model proposed by Wiers et al., (2007), two systems are important: an appetitive, approach-oriented system and a regulatory executive system that only becomes fully developed when young adulthood is reached. The appetitive system is very sensitive to learning experiences. As a result, repeated sensitization with alcohol can lead to automatic action tendencies towards alcohol, which can be very long-lasting and persistent. The regulatory system acts largely conscious and under cognitive control. According to the model of Wiers et al., addictive behavior is the outcome of an imbalance between both systems, in which the automatic system can sometimes override the regulatory system1. This imbalance can be the result
of too strongly developed impulsive reactions to addictive stimuli as well as too weakly developed
1 This dual-process model has recently been criticized on its distinction between two separate systems, which have instead been suggested to be more or less overlapping (Gladwin, Figner, Crone, & Wiers, 2011) . Therefore, the terminology of two strictly separated systems was avoided here. Instead, the model should be interpreted in terms of automatic and cognitive processes that interact dynamically (Wiers et al., 2013).
regulatory processes to control these impulses (R W Wiers, Gladwin, Hofmann, Salemink, & Ridderinkhof, 2013).
Strong impulsive reactions to addictive stimuli such as alcohol can include both increased attention aimed at these stimuli (attention bias) and tendencies to approach these stimuli (approach bias). Attention biases and approach biases (cognitive biases) towards alcohol have been shown in a large number of studies (e.g., Field, Kiernan, Eastwood, & Child, 2008; Schoenmakers, Wiers, & Field, 2008; Schoenmakers, Wiers, Jones, Bruce, & Jansen, 2007). In addition to cognitive biases, implicit associations with alcohol have been shown to be important predictors of drinking behavior. For example, it has been shown that strong positive implicit associations with alcohol are related to increased levels of alcohol consumption and alcohol-related problems (e.g., Houben, Rothermund, & Wiers, 2009; Houben & Wiers, 2008). The impact of these automatic associations is even further increased when levels of cognitive control are low (Friese & Hofmann, 2009; Houben & Wiers, 2009).
In order to exert control over (unwanted) tendencies towards alcohol, such as attention biases and approach biases, executive functions are needed. According to Miyake et al. (2000), three main executive functions can be distinguished: working memory, response inhibition and set shifting. The importance of executive functions in regulating drinking behavior has been confirmed in research showing impaired executive functions in alcohol-addicted individuals (Bechara & Martin, 2004; Kamarajan et al., 2005; Noel et al., 2005).
These findings raise the question to what extent executive function training can decrease problematic drinking behavior. Regarding the automatic and cognitive processes involved in addictive behavior, executive function training could decrease problematic drinking behavior for two reasons. First, executive function training in the context of alcohol could result in a decrease in automatic tendencies towards alcohol as the result of trained (non-)responses specific to alcohol or non-alcohol stimuli. Indeed, it has been shown that training response inhibition can result in a decrease in both positive implicit associations with alcohol as well as decreased alcohol consumption, but only when participants are trained to inhibit their responses specifically to alcohol stimuli (Houben, Nederkoorn, Wiers, & Jansen, 2011; Jones & Field, 2013; Houben, Havermans, Nederkoorn, & Jansen, 2012). This type of executive function training, in which alcohol-related stimuli are used instead of neutral stimuli, can be referred to as alcohol context-relevant training. Second, executive function training could increase general cognitive control, thereby increasing the ability to moderate the impact of automatic tendencies on drinking behavior. This domain-general approach has been shown to be effective in working memory training (Houben, Wiers, & Jansen, 2011). Although it has been shown that (either domain-general or alcohol context-relevant) training of working memory as well as
response inhibition can result in decreased consumption of alcohol, similar effects have, to our knowledge, not yet been investigated for set shifting.
The Set Shifting Paradigm
Set shifting can be defined as the ability to shift flexibly between mindsets or attentional sets (Ravizza & Carter, 2008) and has been shown to be an important measure of cognitive control (Monsell, 2003). Most studies on set shifting have focused on the ability to switch between two different cognitive tasks. In a typical task switching2 paradigm, participants have to switch between
two different tasks using the same type of stimuli. For example, in some trials, they have to
categorize a number according to its numerical size (e.g., judging whether it is larger or smaller than 5), and in other trials, they have to categorize a number according to its parity (judging whether it is odd or even). The combination of two different tasks during the same experiment results in a distinction between two sorts of trials. When the same task has to be performed in the current trial as in the previous trial, the task is defined as a repeat trial. When the task changes from the previous trial to the current trial, the current trial is defined as a switch trial. Reaction times of (both non-heavy drinking and non-heavy drinking, Noel et al. [2005]) participants are usually higher in switch trials than in repeat trials. This distinction characterizes the set shifting element of the task: the participant has to disengage from one particular mindset and has to engage in another one, a process that takes time. The main outcome of interest in set shifting research is usually defined as the switch cost, the difference in reaction times between switch trials and repeat trials (Rahamim et al., 2013). During task switching, it is either possible for the participant to know what task has to be performed next (e.g., the task switches after every n number of trials), or not (e.g., the upcoming task is announced by a cue presented on the screen). The first type of task switching can be referred to as predictable switching, and the second type as unpredictable switching (Rahamim et al., 2013).
Instead of switching between two different tasks, it is also possible to let participants perform the same task over trials, but instead change the rule between trials3. With respect to alcohol
context-relevant set shifting, rule switching is an interesting paradigm to measure cognitive biases 2 Some declaration of the terminology used in the current report is in order here. In the literature, both the terms set shifting and task switching are used, often to describe the exact same paradigm (Ravizza & Carter, 2008). In the current report, the term set shifting is used to refer to the general process of shifting between mindsets, rules or tasks, since these processes are broader than only switching between tasks. The term task switching will be used to refer to the specific switching between two distinct tasks in one experimental setting and the verbs to switch and switching will be used accordingly.
3 Although task switching and rule switching are somewhat different processes, they are dependent on the same executive function of set shifting, also called cognitive flexibility. As shown in latent variable analyses, set shifting can be seen as a separable construct that shows correlations with different kind of set shifting tasks (Miyake et al., 2000). Therefore, it can be proposed that different types of set shifting (cognitive flexibility), task switching as well as rule switching, are dependent on the same underlying process. In the current research, the focus lies on rule switching and its relationship with automatic tendencies towards alcohol and alcohol consumption.
towards alcohol. Korucuoglu, Gladwin, & Wiers (in prep.) presented participants with pictures of either alcoholic drinks or non-alcoholic drinks. The task was always to use one response button to select a presented stimulus and another response button to reject a stimulus. According to a cue presented on the screen, participants either had to select pictures of alcoholic drinks and reject pictures of non-alcoholic drinks (alcohol rule task set), or select pictures of non-alcoholic drinks and reject pictures of alcoholic drinks (non-alcohol rule task set). As a result, when the rule in switch trials was non-alcohol, participants had to switch away from the alcohol task set and switch towards the non-alcohol task set. In addition to the switch costs generally found in switching tasks, Korucuoglu et al. found that participants were slower in non-alcohol rule switch trials compared to alcohol switch trials. This suggests that participants found it particularly difficult to disengage from the alcohol rule mindset and to engage in the non-alcohol rule mindset instead.
The finding that switching away from alcohol can be a particularly difficult task (i.e., a cognitive bias towards alcohol) raises the question to what extent this shifting ability is a trainable construct. As shown in earlier training studies, improving executive functions can result in a decrease in alcohol consumption. Although it has been shown that set shifting ability in general can be improved by means of training (Karbach & Kray, 2009; Minear & Shah, 2008), set shifting training in the context of alcohol has not yet been done. Given that set shifting is considered to be one of the core executive functions, this paradigm could be an interesting and relevant addition to the findings already obtained with other paradigms of cognitive control.
The Current Research and Hypotheses
The aims of the current study were threefold. The first aim was to replicate the findings on cognitive biases towards alcohol during set shifting obtained in Korucuoglu et al. (in prep.). The second aim was to develop a new method to train set shifting in the context of alcohol. The third aim was to introduce this training and to measure its effects on alcohol consumption.
An important problem associated with alcohol context-relevant set shifting is that once the switch towards non-alcohol has been made, it is always necessary to switch back in order to reach the beginning state again. To omit any switches towards alcohol would result in a task containing nothing but repeat trials, whereas the switch trials are the training object of interest (Figure 1). Since the goal of the current study was to train switching away from alcohol and not switching towards alcohol, a training condition was included in which the type of switch was manipulated (Alcohol Predictable Condition). For this purpose, the distinction between predictable and unpredictable switching was used. Since unpredictable switching is in general more difficult than predictable switching (Monsell, 2003) and shows larger transfer effects (Minear & Shah, 2008), it was expected
Figure 1. Visual representation of the context-relevant switching problem. The switch from alcohol to non-alcohol is the switch we wish to train. However, to omit any switches towards alcohol would result in a task containing nothing but repeat trials. Stimuli are obtained from Van Schie & Boendermaker (2014).
that the training effect of switching towards non-alcohol would be greater in unpredictable trials than in predictable trials.
In addition to the training condition with manipulated predictability, two additional training conditions were added. The first condition included an alcohol-context relevant switch training in which switches towards alcohol were predictable as often as unpredictable (Alcohol 50/50 Condition). The second condition included an alcohol-context irrelevant switch training in which neutral stimuli were used (Neutral Switch Condition). Altering the design in this way made it possible to test 1) whether training effects could be attributed to the manipulation in predictability of switch direction and 2) whether training effects could be attributed to the alcohol-specific context or to a domain-general training of set shifting. With respect to the alcohol-context relevant condition, it could be argued that set shifting training would still show training effects since training the switch towards non-alcohol could in itself have more effect than training the switch towards alcohol. As a result of strong cognitive biases towards alcohol, frequent drinkers may be very good at switching towards alcohol already, thereby decreasing the possibility of any additional training effects. With respect to the neutral switch condition, it could be argued that domain-general set shifting training would still show training effects as well as a result of a general strengthening of cognitive control. For comparison of training effects, a placebo condition was added in which alcohol stimuli were used but no switching task was performed (Alcohol Non-Switch Condition).
Regarding the effects of the training, it was expected that 1) set shifting training would lead to improved set shifting, defined as a decrease in switch costs and ultimately in a decrease in alcohol consumption, compared to the non-switch condition. In addition, it was expected that 2) the training effects would be the largest in the Alcohol Predictable Condition, second-largest in the Alcohol 50/50 Condition and smallest in Neutral Switch Condition. Since it is hypothesized that executive function training is effective because executive functions are needed to suppress automatic impulses to drink alcohol, it was expected that 3) the effects of set-shifting training would be mediated by implicit associations (cf. Houben, Wiers, et al., 2011). Specifically, it was expected that individuals with the strongest automatic impulses to drink alcohol would benefit the most from set shifting training. See
Table 1 for a schematic overview of the different conditions and expected training effects. Table 1
Schematic Overview of the Different Conditions
Method
The original aim of the study was to investigate the effects of set shifting training on alcohol consumption and to compare these effects between conditions who received different versions of the task. However, the development part of the training task turned out to be far more complex and time-consuming than originally assumed. Given the time frame that was available, it was therefore decided to perform a pilot study to test the training task on its training potential and to perform a replication study on the findings of Korucuoglu et al. (in prep.) in an assessment variant of the task. Participants
To recruit moderate to heavy drinkers, an invitation e-mail was sent to all first-year psychology students of the University of Amsterdam who obtained a score of 6 or higher on the Alcohol Use Disorder Identification Test (AUDIT; Saunders, Aasland, Babor, de la Fuente, & Grant, 1993) during an earlier assessment4. In addition, students were recruited via advertisements and
posters asking for participants who enjoyed drinking alcohol. Students were informed they would receive a computer-based training in which they would have to classify drinks and received course credits in return for their contribution. The resulting sample consisted of 27 first-year psychology students (11 females; Mage = 20.30, SDage = 1.64) of the University of Amsterdam. On average,
participants consumed an average of 17.60 Dutch standard alcoholic drinks (10 g of alcohol per glass) 4 In their first year, all psychology students of the University of Amsterdam have to take part in the mandatory Testweek. During this Testweek, various questionnaires are administered, including the AUDIT. Data from the Testweek can be requested from the University for research and recruitment aims, such as in the current study.
Switching Task Context-relevant (alcohol stimuli) Manipulated Predictability Expected Training Effect Alcohol Predictable + + + +++ Alcohol 50/50 + - - ++ Neutral Switching + - - +
Alcohol Non-Switch - + n/a -
in the week prior to the study, as measured with a shortened version of the Timeline Follow-Back Questionnaire (TLFB; Sobell & Sobell, 1992). Fifty-six percent of the participants had a score of 8 or higher on the AUDIT5, indicating hazardous drinking (Saunders et al., 1993). The average AUDIT score
was 9.78 (SD = 3.71). One participant failed to complete all training sessions and was therefore excluded from all analyses except those on the pretest. Due to technical problems, one participant was excluded from the analyses on the Implicit Association Task, two participants were excluded from the analyses on the pretest-posttest comparison of the assessment task, and two participants were excluded from analyses on the training task. As a result of minor adaptations that were applied to the training task, 8 participants (30.77 %) received 2 different versions of the task across training sessions, and 18 participants (69.23 %) received 3 different versions.
Materials and measures
Alcohol-related problems. To measure problematic drinking behavior, the AUDIT (Saunders et al., 1993) was used. The AUDIT consists of 10 items, in which alcohol consumption as well as drinking-related problems are assessed. The score of the AUDIT ranges between 0 and 40.
Alcohol consumption. Alcohol consumption was measured with a shortened version of the TLFB (Sobell & Sobell, 1992) that was modified by Wiers, Hoogeveen, Sergeant & Gunning (1997). In this questionnaire, participants had to indicate how much alcohol they had consumed on each day during the past week. To measure binge drinking occasions, participants also had to indicate on how many days of this past week, they drank more than 5 (for males) or 4 glasses (for females) on one day. Additionally, participants were asked whether in the past occurring week, any events had occurred that caused them to drink more alcohol than they would regularly do.
Implicit associations. Automatic associations with alcohol were measured with a variant of the Implicit Association Test (IAT; Greenwald, McGhee, & Schwartz, 1998). In the IAT, participants have to classify stimuli into two target categories and two attribute categories using two response keys. In the current study, the target categories were “alcohol” and “soft drink” and the attribute categories were “me” and “others”. The target stimuli were five alcohol-related words (e.g., beer) and five soft drink-related words (e.g., ice tea). The attribute stimuli were five words associated with “me” (e.g., myself) and five words associated with “others” (e.g., themselves). See Appendix A for an overview of all stimuli that were used. The Identity IAT was chosen since this type of IAT has been shown to outperform other types of IATs (e.g., Alcohol Approach IAT and Alcohol Cope IAT) on psychometric qualities as well as the ability to predict unique variance in alcohol consumption and alcohol-related
5 As assessed in the current study, not the Testweek.
problems (Lindgren, Foster, Westgate, & Neighbors, 2013; Lindgren, Neighbors, Teachman & Wiers, 2013).
The IAT consisted of 7 blocks, in which different combinations of attribute categories and target categories were assessed. In the first two blocks (20 trials each); participants practiced the classification of target stimuli into target categories (one block) and the classification of attribute stimuli into attribute categories (other block). In the third (20 trials; practice) and fourth (40 trials; assessment) blocks, participants had to classify target and attribute stimuli in both combinations of target and attribute categories. In the fifth block (20 trials), the target labels were reversed and participants practiced the reversed assignment of response keys to target categories. In the sixth (20 trials; practice) and seventh (40 trials; assessment) blocks, participants had to classify stimuli in the combination of target and attribute categories that was the reverse from the third and fourth blocks. As response keys, the buttons “A” and “L” were used. The assignment of response keys to attribute categories, the order of assignment of response keys to target categories, and the order of the pairings of target and attribute categories, were counterbalanced across participants. Stimuli appeared in the center of the screen and category labels were presented in the upper corners of the screen, dependent on the response key assignment of these categories. The stimuli remained on the screen until a correct response was given. When an error was made, a red ‘X’ appeared beneath the stimulus and the trial had to be repeated until a correct response was given.
Switch assessment task. To measure set shifting ability, as well as cognitive biases towards alcohol during switching, an adapted version of the switch task developed by Korucuoglu et al. (in prep.) was used. As stimuli, seven pictures of alcoholic drinks and seven pictures of non-alcoholic drinks were used. Participants were instructed to select one of both types of stimuli and reject the other of these types of stimuli using the “A” and “L” buttons. Reminders of the response key assignment were shown on the left and right side of the stimuli. Stimuli increased in size when they were selected and decreased in size when they were rejected to visually emphasize the effect of both responses.
The task began with three practice blocks. During the first and second block (7 trials each), every stimulus was shown once. In the first block, the rule was to select alcohol pictures and reject non-alcohol pictures (alcohol rule). In the second block, the rule was to select non-alcohol pictures and reject alcohol pictures (non-alcohol rule). When participants made a correct response, the text “Well done!” was shown on the screen in green letters. When participants made an erroneous response, the text: “Unfortunately, that was incorrect. Press a button to try again” was shown on the screen in red letters and participants had to repeat the trial until a correct response was given. When participants were too slow, the text: “That was not fast enough! Press a button to try again” was
shown. During the third block and the subsequent experimental blocks, the rule switched randomly between trials. The rule that had to be followed in the upcoming trial was announced by a cue presented on the screen. When an “X” was shown, participants had to select pictures of alcohol and reject pictures of alcohol. When an “O” was shown, participants had to select pictures of non-alcohol and reject pictures of non-alcohol. A visual representation two subsequent trials is displayed in Figure 2. During the third block, which consisted of 20 trials, participants had to reach an 80 % accuracy rate in order to proceed to the experimental blocks. There were 4 experimental blocks with 50 trials each. After each block except for the last block, a reminder of the instructions was shown before participants could proceed. The combination between response buttons and the associated response (select or reject), the order of the first two practice blocks and the combination between the cues (X or O) and associated rules (alcohol or non-alcohol) were counterbalanced across participants.
Each trial began with the presentation of a fixation cross in the center of the screen. After 500 ms or 700 ms (at random), either the stimulus (first two practice blocks) or the cue (third practice block and experimental blocks) was shown. Cues preceding the stimuli were presented for 1000 ms. After a response was given, the stimulus increased or decreased in size during 500 ms, resulting in a “zooming” effect. During practice blocks, feedback was shown on the screen for an additional 500 ms.
Figure 2. Visual representation of two consecutive trials of the Switch Measure Task. The first trial is an alcohol stimulus trial and the second trial is a non-alcohol stimulus trial.
Energy levels, mood and task rating. To measure energy levels and mood before, during and after training sessions, a Self-Assessment Manikin (SAM; Bradley & Lang, 1994) was administered. The scales ranked from sleepy to active and from pleasant to unpleasant. Additionally, participants were asked how much they liked performing the training session on a 10-point scale, where a 10 stood for very much and a 1 for not at all.
Exit interview. The Exit Interview contained questions about the rating and difficulty of the training task, the clarity of instructions and the rate of distraction during the training sessions. The complete content of the Exit Interview is displayed in Appendix B.
Switch training task. During training sessions, participants received one of four variants of the switch training task, dependent on condition: the Alcohol Predictable variant(n = 7), the Alcohol 50/50 (n = 6) variant, the Neutral Switch variant (n = 7), or Alcohol Non-Switch variant (n = 7). In all variants, a 2 x 2 grid of squares was presented on the screen, as seen in Figure 3. During each trial, two stimuli were placed separately in two of the four possible locations in the grid. Pictures of alcohol were always placed on the left side of the grid and pictures of non-alcohol were always presented on the right side of the grid. As a result, the stimuli could have two different orientations relative to each other. When both stimuli were placed in the upper half of the grid or when both stimuli were placed in the lower half of the grid, the orientation was horizontal (e.g., Figure 3a). When one stimulus was placed in the upper half of the grid and the other stimulus was placed in the lower half of the grid, the orientation was diagonal (e.g., Figure 3b).
Figure 3. Visual representations of the different stimuli orientations in the training task. (A) shows a horizontal orientation and (B) shows a diagonal orientation.
In the switch conditions, the task was to select or reject one of the stimuli, according to the orientation (horizontal or diagonal) of both stimuli on the screen. The stimulus that had to be responded to (the target stimulus), was indicated by a blue circle placed around the stimulus (the selector). Participants selected stimuli using the “L” button and rejected stimuli using the “A” button”. Reminders of the response key assignment were shown on the left and right side of the grid. When a stimulus was selected, it increased in size and when a stimulus was rejected, it decreased in size. The duration of this visual effect, after which the stimulus returned to its original size, was 300 ms. Response feedback was given during the entire task. The feedback interval was 500 ms; when the response was correct, the selector turned green and when the response was incorrect the selector turned red.
After a response was given, the target stimulus changed position in the grid. When the stimulus was located in the upper half of the screen, it would always move one position down in the grid. During the next trial, it would therefore be located in the lower half of the screen. In this situation, the next trial was predictable: The participant was able to predict whether the orientation of the stimuli would be horizontal or diagonal. As a result, the participant was able to anticipate the required response in the upcoming trial, as can be seen in Figure 4. When the target stimulus had been in one of the lower locations, it “fell off” the grid the moment a response was given. In the next trial, the stimulus returned in the grid and could move to two different locations. The stimulus could either move to the upper half of the screen, or it could move to its previous location in the lower half of the screen. In this situation, the next trial was unpredictable: The participant was not able to predict the orientation of the stimuli in the next trial. Therefore, the participant was not able to anticipate the required response in the upcoming trial, as can be seen in Figure 5. When the next trial was unpredictable, a new stimulus appeared on the screen, whereas the stimulus always remained the same during a predictable trial. In unpredictable trials, it was also possible for the selector to change position between trials. In the current trial, it could either be placed around the previous target stimulus, or it could be placed around the previous non-target stimulus. Therefore, it was also not possible for the participants to predict which stimulus would be the target of response in the upcoming trial. When the orientation of the stimuli in the current trial was the same as the orientation in the previous trial, the rule (whether the target had to be selected or rejected) remained the same. The trial was therefore defined as a repeat trial. When the orientation in the current trial was different from the orientation in the previous trial, the rule changed. The trial was therefore defined as a switch trial.
Figure 4. Example of a predictable switch trial sequence.
The percentage of switch trials was 65 % for all experimental conditions6. The percentage of
horizontal and diagonal trials was both 50 %.
In the Alcohol Predictable and Alcohol 50/50 condition, participants had to select and reject pictures of alcoholic and non-alcoholic drinks, similar to the Switch Assessment task. The required response was dependent on the type of stimulus as well as the orientation of the stimuli relative to each other. When the orientation of the stimuli was horizontal, participants had to select pictures of alcohol and reject pictures of non-alcohol (alcohol rule). When the orientation was diagonal,
participants had to select pictures of non-alcohol and reject pictures of alcohol (non-alcohol rule). In 6 In the current task design, this relatively high percentage of switch trials was needed to maintain equal probabilities of horizontal and diagonal (thus alcohol/non-alcohol) rule trials. The ideal distribution of trial types was calculated using a random trial generator that was built with ActionScript 3.0.
Figure 5. Example of an unpredictable switch trial sequence.
the Alcohol Predictable Condition, the ratio of the different types of switch trials was manipulated so that switches from the non-alcohol-rule to the alcohol-rule were predictable (26 % of all trials) more often than unpredictable (6 % of all trials) and switches from the alcohol-rule to the non-alcohol rule were unpredictable (27 %) more often than predictable (6 %). In the Alcohol 50/50 Condition, each type of switch trial, either predictable or unpredictable, had the same probability to occur for each type of rule (alcohol or alcohol). Thus, participants shifted between the alcohol rule and the non-alcohol rule unpredictable as often as predictable. The Neutral Switching Condition was equal to the Alcohol 50/50 Condition, except that the stimuli were triangles and squares instead of pictures of alcoholic or non-alcoholic drinks. In the Alcohol Non-Switch Condition, the task was visually similar to that of the Alcohol Predictable Condition and the Alcohol 50/50 Condition. However, the task was simply to judge whether the stimuli, again pictures of alcoholic or non-alcoholic drinks, were oriented
horizontally or diagonally relative to each other. When the orientation was horizontal, participants had to press the “A” button and when the orientation was diagonal, participants had to press the “L” button. In this condition, selectors were placed around both stimuli, and both selectors turned green or red as the result of a correct or erroneous response.
In the Alcohol conditions, stimuli were six circle-shaped pictures of alcoholic drinks and six pictures of non-alcoholic drinks. In the Neutral Switch condition, stimuli were triangles and squares in six different colors. A different set of stimuli was used in each different training session. During each block, a progress bar was presented on the right side of the screen, which enabled participants to track the course of time while performing the task. During the entire task, a trial had to be repeated if no response was given for 3000 ms. The order of the practice blocks, the assignment of response keys to responses (select/reject or horizontal/diagonal) and the side on which the different types of stimuli were placed (left/right), was counterbalanced across participants.
In each session, equal for all conditions, the task began with 5 practice blocks of 7 trials each. In the first two blocks, the grid was 1 x 1 instead of 2 x 2. During the first block, the rule was to select alcohol and reject non-alcohol. During the second block, the rule was to select non-alcohol and reject alcohol. All stimuli that would be used later in the experimental blocks were presented to the
participants once during these blocks. During the third block, the rule was to select the target stimulus when the orientation was horizontal and reject the target stimulus when the orientation was diagonal. During the fourth block, the rule was to select the target stimulus when the orientation was diagonal and reject the target stimulus when the orientation was horizontal.
Besides the feedback shown by the color of the selector, participants were presented with additional feedback during these practice trials. Equal to the Switch Measure Task, additional feedback text was placed in the center of the screen (2000 ms). When an erroneous response had been given, participants had to repeat the trial until a correct response was given. After the four practice blocks, a fifth practice block was administered in which participants had acquire an accuracy rate of 80 % in order to proceed to the experimental blocks. The task that had to be performed in the fifth block was identical to the task in the experimental blocks of the training conditions.
There were four experimental blocks, each consisting of 100 trials7. In a later version of the
task, a score counter was added to increase motivation during the task. From that moment, participants received points for each correct response. The number of points they received, was dependent on the speed of their response. After each response, the number of earned points appeared in green text in the center of the stimulus (500 ms). The cumulative score for each block was presented in the left upper corner of the screen. When participants made three consecutive 7 The resulting total amount of trials was based on earlier set shifting studies (Karbach & Kray, 2009; Minear & Shah, 2008; Van der Oord et al., 2012) and was adapted to fit the number and duration of the training sessions.
errors, they were presented with a reminder of the instructions and had to repeat the last trial until a correct response was given. Additionally, when participants had an error rate of 30 % during 10 consecutive trials, they were also presented with a reminder of the instructions and could only proceed to the next trial by pressing a button. In between blocks, participants were presented with their total number of earned points for that block and a reminder of the instructions and could take a break of any preferred size.
During the course of the study, some minor adaptations were applied to the training task. As a result, not all participants received the exact same version of the task during the individual training sessions. On the fourth day of testing, the score counter was added. On the eleventh day of testing, the background color was changed from white to light grey, the font size was increased and the instructions were spread out over more pages.
All tasks were programmed using Adobe Flash ActionScript 3.0. Procedure
The experiment consisted of a pretest, four training sessions, posttest and follow-up. For the pretest and first training session, as well as the posttest and last training session, participants came to the lab. Participants could compete the second and third training session either at home or any alternative location where they had access to a computer with internet connection. After registering on the training website and giving their informed consent, participants were randomly divided between the four conditions (Alcohol Predictable, Alcohol 50/50, Neutral Switch or Non-Switch). During pretest, participants completed the AUDIT, IAT, the TLFB for the week prior to the study and the Switch Assessment Task. Subsequently, they performed the first session of the training task and the SAM questionnaire. Since the current study was a pilot and the focus of interest was especially placed on the usability of the training task, the experimenter accompanied a number of participants in the lab cubicle to see whether they understood the instructions of the task. Also, participants were interviewed on their experiences while performing the task. During the following days, participants completed two more training sessions and accompanying SAM questionnaires; each time they were informed about the new awaiting training session by e-mail. Every new training session would open automatically after 24 hours upon finishing the previous session. Participants were instructed to complete the next session within two days. After two days of inactivity, reminders were sent by e-mail. After seven days of inactivity, participants were excluded from further participation. After a minimum of 24 hours upon completion of the third training session, participants could return to the lab, where they performed the last training session and, immediately upon completing the training, the posttest. During posttest, participants performed another session of the Switch Assessment task,
the IAT, the TLFB and a short exit interview. The same versions of the tasks were used during each session of each individual participant with respect to the counterbalances that were applied to these tasks. One week after posttest, participants were e-mailed with an invitation link to another TLFB and received their course credits. In Figure 4, an overview is given of the procedure of the current study.
Results Differences at Baseline
There were no differences in age, F = .650, p = .591, or gender, χ2(3, N = 27) = .175, p = .981,
between conditions, whereas there was a difference in average AUDIT scores, F(3, 23) = 4.34, p = .015. Games-Howell post-hoc tests revealed that participants in the Non-Switch condition displayed a lower AUDIT score (M = 6.43, SD = 0.98) compared to participants in the Alcohol 50/50 condition (M = 11.17, SD = 2.56) and the Neutral Switching condition (M = 12.14, SD = 4.34). Since the assumption of equal variances was violated, F(3, 23) = 4,08, p = .018, a Kruskal-Wallis Test was performed additionally. This test revealed similar results, χ2(3, N = 27) = 10.70, p = .013, reflecting an overall
difference in AUDIT scores between conditions. Switch costs, IAT scores and TLFB scores at baseline were equal across conditions.
Switch Assessment Task
Pretest Data. Before analyzing the data of the assessment task, erroneous responses (9.70 % of all responses) and responses faster than 200 ms (0.40 % of all responses) were removed from the dataset. Additionally, only responses from experimental blocks were used in which participants made more than 30 correct responses (60 % of all responses) per block. As a result, responses from the first experimental block of two participants were removed from the data since these participants made an error on almost every trial during this block, 94 % and 96 % respectively. Possibly, they
misremembered the cues belonging to the specific required responses. Indeed, when interviewed after pretest, one of these participants confirmed that she had remembered the wrong combination of cues and associated rules.
On the remaining data, a repeated measures ANOVA was performed with trial type (switch vs. repeat), task set (alcohol rule vs. alcohol rule) and stimulus type (alcohol picture vs. non-alcohol picture) as within subjects factors. As expected, reaction times were higher for switch trials (M = 932.17, SD = 39.91) than for repeat trials (M = 843.59, SD = 33.78), F(1, 26) = 27.04, p <.0001, η𝑝2 = .510. Additionally, participants were faster in alcohol rule trials (Select Alcohol/Reject non-alcohol; M = 859.57, SD = 35.14) than in non-alcohol rule trials (Select Alcohol/Reject non-alcohol; M = 916.19, SD = 39.69), F(1, 26) = 7.23, p = .012, η𝑝2 = .218. Furthermore, participants were faster when they had to respond to an alcohol picture (M = 844.30, SD = 34.63) than when they had to respond to a non-alcohol picture (M = 931.45, SD = 38,76), F(1, 26) = 33.63, p <.0001, η𝑝2 =.564, regardless of whether the rule was to select or reject the stimulus. There was a significant interaction effect between trial type and stimulus type, F(1, 26) = 9.945, p = .004, η𝑝2 = .277, where the difference in
reaction times of switch and repeat trials was larger for non-alcohol stimulus trials than for alcohol stimulus trials. Thus, when the presented stimulus was a picture of non-alcohol, the increase in reaction time as a result of rule switching was larger than when the stimulus was a picture of alcohol. There was also an interaction effect between task set and stimulus type, F(1, 26) = 9.809, p = .004, η𝑝2 = .274, where the difference in reaction times of alcohol and non-alcohol stimuli was larger during alcohol rule trials than during non-alcohol task set trials, as can be seen in Figure 7. Put together, participants responded faster to alcohol stimuli than to non-alcohol stimuli, but this effect was stronger during alcohol rule trials than during non-alcohol rule trials.
Figure 7. Mean reaction times during pretest compared for trial type and stimulus type. Displayed error bars represent standard errors.
There was no significant interaction between trial type and task set, F(1, 26) = .001, p = .971. Thus, the difference in reaction times between switch and repeat trials was not larger in non-alcohol rule trials compared to alcohol rule trials. To test whether the reaction times of non-alcohol rule switch trials were still highest compared to the other types of trials, a repeated measures ANOVA was run on the separate reaction times of alcohol rule repeat trials, alcohol rule switch trials, non-alcohol rule repeat trials and non-alcohol rule switch trials. Mauchly’s test indicated that the assumption of sphericity was violated, χ2 = 11.748, p = .039, therefore degrees of freedom were corrected using
Huyn-Feldt estimates of sphericity (ε = .859). There was a significant difference between the four types of trials, F(1, 67.04) = 12.15, p <.0001, η𝑝2 = .319. Specifically, reaction times in non-alcohol rule switch trials were higher than those in alcohol rule repeat trials, F(1, 26) = 23.04, p < .0001, 𝜂𝑝2 = .470, non-alcohol rule repeat trials, F(1, 26) = 15.65, p = .001, ηp2 = .376 and alcohol rule switch trials, F(1, 26) = 5.850, p = .023, η𝑝2 = .184 (Figure 6). Thus, participants were slower in non-alcohol rule trials
700 750 800 850 900 950 1.000 1.050 Alcohol Non-Alcohol M ea n Rea ct io n Ti m es (m s) Stimulus Type Repeat Switch
when they were preceded by alcohol rule trials than in alcohol rule trials preceded by non-alcohol rule trials, especially when the current trial was a switch trial.
Figure 7. Mean reaction times during pretest compared for trial type and task set. Displayed error bars represent standard errors.
To test whether differences in reaction times between non-alcohol rule switch trials and alcohol rule switch trials correlated with the AUDIT, a bias score was computed by subtracting the reaction times of alcohol rule switch trials from those of non-alcohol rule switch trials. There was no significant correlation between the bias score and AUDIT score, r = -.312, p = .138. In order to test for differences in the obtained results between individuals with a low AUDIT score compared to a high AUDIT score, participants were divided into two groups based on a median-split (Med = 9,00) on the AUDIT scores of the total sample. Participants who scored lower than 9 points on the AUDIT were assigned to the low-AUDIT group and participants who scored equal to or higher than 9 points on the AUDIT were assigned to the high-AUDIT group. All performed analyses were repeated with AUDIT group as a between subjects factor. There was no interaction effect between AUDIT group and trial type, F(2, 25) = .032, p =.860, AUDIT group and task set, F(1, 25) = 2.395, p = .134, or AUDIT group and stimulus type, F(1, 25) = 1.233, p = .277. There was, however, a marginal significant three-way interaction between AUDIT group, trial type and stimulus type, F(1, 25) = .363, p = .068, η𝑝2 = .127. To break down this interaction, a difference score was computed between the switch costs in non-alcohol stimulus trials and non-alcohol stimulus trials. On the resulting bias score, a One-Way ANOVA was performed with AUDIT group as a Between Subjects factor. The findings confirmed a marginal significant difference between AUDIT groups, F(1, 25) = 3.300, p = .081, η𝑝2 = .117, where the bias score was higher in the high AUDIT group (M = 152.42, SD = 93.47) compared the low AUDIT group (M = 80.48, SD= 112.42). Thus, on average, higher AUDIT scores were related to a larger bias towards
700 750 800 850 900 950 1.000 Alcohol Non-Alcohol M ea n Rea ct io n Ti m es (m s) Trial Type Repeat Switch
alcohol stimuli during switching, compared to low AUDIT scores.
Preliminary Training Effects. To investigate whether training had any effects on set shifting skills, switch costs were calculated by subtracting the reaction times of repeat trials from those of switch trials for both alcohol rule and non-alcohol rule trials. In a Mixed ANOVA on switch costs with time (pretest vs. posttest) and task set (alcohol rule vs. non-alcohol rule) as within subjects factors and condition as a between subjects factor there was no significant main effect of time, F(1, 20) = .056, p = .816, no significant interaction effect between time and condition, F(3, 20) = .734, p = .544. In none of the experimental conditions, switch costs for alcohol or non-alcohol rule trials decreased significantly between pre- and posttest.
Implicit Associations
D-scores of the IAT were calculated using an adapted version of the D-score algorithm derived from Greenwald, Nosek, & Banaji (2003). One adaptation included the replacement of error latencies by the block mean plus two standard deviations, instead of the block mean plus 600 ms. Additionally, in the IAT of the current study participants had to correct their response after making an error. These corrected trials were excluded from the data. The resulting D-Scores reflected the strength of the IAT effect, where larger D-scores implied stronger associations between alcohol and “Me”. There was a significant correlation between the D-scores at pretest and the D-scores at posttest, r =.696, p <.0001. There was however no significant IAT effect at baseline, t(26) = .604, p = .551. Participants did not perform better on the IAT when alcohol was paired with “Me” than when alcohol was paired with “Others”. In addition, there was no correlation between the AUDIT and the D-scores at pretest (M = .11, SD = .48), r = 0.000, p = 1.000, or posttest (M = .11, SD = .45), r = -.077, p = .713. A Mixed ANOVA with time as a within subjects factor and condition as a between subjects factor revealed no significant difference in D-scores between pretest and posttest, F(1, 21) = .003, p =.958 and no significant interaction between time and condition, F(3, 21) = .951, p = .434.
Alcohol Consumption
Alcohol consumption at pretest and follow-up was defined as the total amount of glasses consumed during the week prior to measurement, as measured with the TLFB. Since the resulting TLFB scores did not follow a normal distribution, all analyses were performed on log-transformed8
scores. To facilitate interpretation, the reported means are non-transformed scores. There was a significant correlation between the total amount of glasses consumed in the week prior to pretest, 8 Since one participant drank no alcohol during the week prior to posttest with a zero value as a result, a constant was added to all scores.
(M = 17.52, SD = 15.20) and the AUDIT r = .537, p = .005 and the week prior to follow-up, (M = 16.57, SD = 11.49) and the AUDIT, r = .633, p = .001. To test for gender differences, average TLFB scores at pretest were compared for males and females. An independent t-test with gender as a between subjects factor, revealed that males (M = 26.01, SD = 4.38), on average, consumed marginally significant more alcohol than females (M = 12.22, SD = 3.46, t(11.61) = 2.12, p = .056. However, no differences were found when comparing the difference score between pretest and follow-up, t (24) = .198, p = .845. Therefore, gender differences were not taken into account when comparing alcohol consumption at pretest and follow-up.
Both at pretest (n = 11) and follow-up (n = 9), a number of participants indicated that they had consumed more alcohol during the past week than they normally did. In a One-Way ANOVA on TLFB difference scores, there was a significant difference between participants who consumed more alcohol than average (M = -0.23, SD = 0.28) and individuals who did not consume more alcohol than average (M = 0.06, SD = 0.17) at pretest, F(1, 24) = 11.858, p = .003, ηp2 = .310, but not at follow-up, F(1, 22) = .935, p = .935. Therefore, above average drinking was included as a covariate in an
ANCOVA on TLFB scores with time as a within subjects factor and condition as a between subjects factor. The results revealed a significant effect of time, F(1, 21) = 13.692, p = .001, ηp2 = .395. At follow-up, participants consumed significantly less alcohol (M = 16.57, SD = 11.49) than at pretest, (M = 17.52, SD = 15.20). There was no significant interaction effect between time and condition, F(3, 21) = 1.704, p = .197. Thus, on average, alcohol consumption decreased in from pretest to follow-up, for participants in both the training conditions and control condition. When analyses were performed with AUDIT group as an additional between subjects factor, no interaction effect between time and AUDIT group, F(1, 18) = .417 p = .527 and no significant interaction effect between time, condition and AUDIT group, F(2, 18) = .746, p = .488. No mediation analyses with IAT scores was performed on the difference scores of the TLFB since there was no significant IAT effect in the current study. Energy levels, Mood and Task Rating
To test whether there were any differences in energy levels and mood during and between sessions, separate Mixed ANOVA’s were performed on the average SAM scores with time (before, during and after session), session (session 1, session 2, session 3, session 4) as within subjects factors and condition as a between subjects factor. There was a significant effect of time on energy level, F(2, 34) = 5.136, p = .011, η𝑝2 = .233. Pairwise comparisons revealed that participants felt less active during sessions (M = 2.70, SD = .120) than before sessions (M = 3.03, SD = .09), p = .021, and less active after sessions (M = 2.650, SD = .117) than before sessions, p = .023. There was also a significant main effect of time on mood, F(2, 34) = 9.829, p < .0001, η𝑝2 = .366. Pairwise comparisons revealed
that mood was lower during sessions (M = 3.150, SD = .103) than mood before sessions (M = 3.584, SD = .097), p = .002 and higher after sessions (M = 3.426, SD = .100) than during sessions. Thus, participants felt worse during sessions than before the sessions, but after sessions, their mood rose back to approximately the same levels again. Since the distributions of the SAM scores deviated significantly from normal, Friedman’s related samples tasks were performed in addition. Findings confirmed the difference of time on energy levels, p = .004 and mood, p = .002.
On rating of the task, there were no differences between the different sessions, F(3, 51) = .606, p = .614 and different conditions, F(3, 17) = 2.089, p = .140.
Output Exit Interview
80.7 % (n = 21) of participants had been distracted at least once during the training sessions. For example, one participant indicated that he had been distracted because a roommate suddenly entered the room while he was performing the task. 46.15 % (n = 12) indicated they did not like the training task. Most of the responses (58.3 %) suggested that this was the result of the duration. 11.5 % of participants (n = 3) indicated the instructions were not clear and 0.08 % (n = 2) indicated that it was not clear what was expected from them9. See Appendix B for all (Dutch) responses to the
open questions of the exit interview.
To test whether there were any differences in scores on the items regarding the training task a MANOVA was performed on all these items with condition as between subjects factor. The
Multivariate Test indicated no differences between condition, F(27, 45) = .999, p = .489. The means and associated standard deviations for each item are displayed in Table 2. Duration of the sessions was, on average, scored a 7.12 (SD = 1.40), where a score of 1 stood for way too short and a score of 10 for way too long.
Switch Training task
Exploratory, it was tested whether reaction times were indeed higher in unpredictable switch trials than in predictable switch trials. Similar to the Switch Measure data, erroneous responses (6.9 % of all data) and responses faster than 200 ms (19.9 % of all data) were excluded from the dataset. Since participants in the control condition performed a control task, analyses on switch costs and reaction times of switch trials could only be computed for the training conditions.
9 Although it was explicitly stated in the Exit Interview that the questions applied to the training task, a number of participants seem to have interpreted that the questions instead applied to the assessment task. In their responses, they for example indicated that they disliked the absence of a progress bar in the assessment task.
Table 2
Mean scores on exit interview items related to the training task.
Item Alcohol Pred Alcohol 50/50 Neutral Switch Control
I liked performing this task 5.83 (2.40) 5.40 (2.07) 4.57 (1.99) 5.14 (2.12)
I found it difficult to perform this task 3.67 (2.34) 4.60 (3.21) 5.42 (2.43) 5.29 (2.43)
I found the instructions provided in the task clear 7.33 (1.97) 8.80 (0.84) 7.29 (1.80) 8.71 (1.11)
It was clear what was expected of me during the task 7.33 (2.25) 9.20 (0.84) 8.29 (0.76) 8.71 (1.11)
As I performed the task longer, I noticed I got better at it 5.67 (3.20) 7.80 (1.30) 7.43 (2.70) 7.00 (2.45)
The practice blocks were of added value for me 7.17 (1.94) 6.60 (2.40) 7.86 (2.91) 5.71 (2.81)
Also after the first session, the practice blocks were of added value for me
5.83 (2.31) 5.40 (3.85) 6.71 (2.43) 6.00 (2.58)
The score count had a motivating effect on me 6.44 (2.88) 8.20 (1.30) 9.00 (1.15) 7.43 (2.51)
Note. Standard deviations are displayed between brackets. Scores ranged from 1 (Not true at all) to 10 (Very true).
To test whether participants responded indeed faster in predictable switch trials than in
unpredictable switch trials, a Repeated Measures ANOVA was run on the reaction times of switch trials with switch type (predictable vs. unpredictable) as a within subjects factor. Participants responded faster in predictable switch trials (M = 470.18, SD = 15.03), than in unpredictable switch trials (M = 754.09, SD = 41.88), F(1, 16) = 74.64, p <.0001., η𝑝2 =.823 (Figure 8).
Figure 8. Mean reaction times for predictable and unpredictable switch trials. Displayed error bars represent standard errors.
To compute switch costs, only reaction times of unpredictable switch trials were used, since repeat trials were unpredictable as well. To test whether switch costs decreased during training, a Repeated Measures ANOVA was run time and task set as within subjects factors. Mauchly’s test indicated that the assumption of sphericity was violated, χ2 = 17.07, p = .004, therefore degrees of
0 100 200 300 400 500 600 700 800 900 Predictable Unpredictable M ea n Rea ct io n Ti m es (m s) Switch Type
freedom were corrected using Huyn-Feldt estimates of sphericity (ε = .708). There was a significant main effect of time on switch costs, F(2.123,48) = 4.931, p = .012, η𝑝2 = .236. Contrasts revealed that switch costs were significantly lower at session 4 (M = 82.45, SD = 15.62) than at session 3 (M = 144.31, SD= 21.71), F(1, 16) = 23.38, p <.0001, η𝑝2 = .594 and session 1 (M = 178.97, SD = 19.38), F(1, 16) = 19.163, p <.0001, η𝑝2 = .545 and marginal significantly lower than at session 2, session 2 (M = 142.15, SD = 23.03), F(1, 16) = 4.105, p = .06, η𝑝2 = .204. Thus, switch costs decreased significantly over time in all training conditions.
To test whether there was a difference in the number of errors made between online and lab sessions, a Repeated Measures ANOVA was run on the total amount of errors per session with setting (lab vs. home) as a between subjects factor. There was a marginal significant effect of setting, where subjects made more errors when they performed the training sessions online (second and third session; M = 7.03, SD = 2.78) than when they performed the sessions in the lab (first and fourth session; M = 6.12, SD =3.14), F(1, 17) = 4.116, p = .058, η𝑝2 = .195. However, contrast tests of the separate sessions revealed that the only significant difference between sessions occurred between session 3 (M = 7.38, SD = .75) and 4 (M = 5.91, SD = .70), F(1, 20) = 5.944, p = .024, η𝑝2 = .229.
Discussion
In the current research a set shifting training method was developed in the context of alcohol and tested by means of a pilot study. In addition, an attempt was done to replicate the findings of Korucuoglu et al. (in prep.) on the pre-measurement data. As expected, it was found that moderate to heavy drinkers were slower to apply a non-alcohol rule compared to an alcohol rule in a switch task where they alternately had to select and reject different types of pictures. This difference in reaction time was larger when the stimuli were alcohol pictures compared to non-alcohol pictures. In addition, individuals were in general slower to respond to non-alcoholic pictures than to alcoholic pictures, independent of the rule (selection or rejection of the stimulus) they had to apply. Taken together, these findings imply the presence of an overall alcohol bias in the results.
Similar to the findings of Korucuoglu et al. (in prep.), there was an interaction effect between trial type and stimulus type. Switch costs were higher in trials in which individuals had to respond to a picture of a non-alcoholic drink, compared to trials in which they had to respond to a picture of an alcoholic drink. This effect was independent of the rule that had to be applied: i.e., whether the presented stimulus had to be selected or rejected. When comparing groups with high and low AUDIT scores, there was a trend indicating that this effect was extra profound in individuals who displayed a high AUDIT score. Although this finding can be considered to represent further evidence for a
to switch towards stimuli of non-alcohol than to switch towards stimuli of alcohol. In fact, switch trials were defined as trials in which the rule (select alcohol/reject alcohol vs. select non-alcohol/reject alcohol) had changed relative to the previous trial. The stimulus type could either be identical to or different from the stimulus type in the previous trial. Instead, this finding shows that switching between rules is particularly difficult when the target stimulus in the switch trial is a picture of non-alcohol compared to a picture of alcohol. Possibly the demands placed on cognitive control as a result of switching between rules caused participants to be more influenced by the visual content of the stimuli during switch trials than during repeat trials. This bias during switching was slightly higher in individuals with a high AUDIT score than in individuals with a low AUDIT score. This is in line with research showing that heavy drinkers show larger cognitive biases than light drinkers (e.g., Field et al., 2008; Field, Mogg, Zetteler, & Bradley, 2004; Tibboel, De Houwer, & Field, 2010). This finding can be considered as evidence that the demands called upon by set shifting cause the cognitive biases towards alcohol already present during shifting to increase even further.
Analyses on the data of the training task revealed that, on average, participants responded faster in predictable switch trials than in unpredictable switch trials. This finding is in line with the notion that unpredictable switching is more difficult than predictable switching. Therefore, the manipulation of predictability of switches in the training task can be considered successful. This is especially relevant for the Alcohol Predictable version of the task, in which it is expected that the unpredictable switches towards non-alcohol have a greater training effect than the predictable switches towards alcohol.
Although manipulation of the predictability of switch direction was successful, no decreases in cognitive biases towards alcohol as a result of training were found, nor was there a decrease in overall switch costs. However, as a result of the small sample size, the current study was strongly underpowered. As a result, the possibility to detect any differences between conditions was very low. Larger sample sizes should be used in follow-up studies to derive more definitive conclusions on training effects. The decrease in switch costs between training sessions that was found in all switch conditions (although not possible to compare with a non-switch condition) does show that set shifting skills increased, at least during training.
In addition to a decrease in switch costs during training, there was a significant decrease in alcohol consumption between pretest and follow-up. However, since this finding was independent of condition, it is not clear whether it was the result of any training effects. An alternative explanation could be that the shortened version of the TLFB is very sensitive to variations in drinking behavior caused factors other than training, compared to measures of alcohol consumption in which the measurement period extends over a longer time period. In follow-up studies, measures of alcohol
consumption could be used that are less sensitive to non-training related variability in drinking behavior on the one hand, and more sensitive to detect short-term training-related changes in drinking behavior on the other hand, such as a taste test (cf. Houben, Nederkoorn, et al., 2011). Another possible explanation for the decrease in alcohol consumption could be found in placebo effects. The aim of the current study was very transparent, since participants were told they would receive a cognitive training that could help them to decrease their alcohol consumption. This in itself could have created the obtained findings.
Contrary to expectation, analyses on the IAT revealed that participants were not faster to classify alcohol stimuli when alcohol was paired with the “Me” category compared to when the alcohol was paired with the “Others” category. Additionally, there was no correlation between IAT and the AUDIT. This is in contrast with findings of other studies in which the Identity version of the IAT was assessed (Lindgren, Foster, Westgate, & Neighbors, 2013; Lindgren, Neighbors, Teachman & Wiers, 2013; Pronk & Wiers, 2014). An important difference between the version of the IAT used in these studies and the IAT used in the current study is represented in the target categories that were employed. In the studies of Lindgren et al., as well as in that of Pronk & Wiers, the target categories were drinker versus abstainer instead of alcohol versus soda. A possible explanation for these different findings lies in the self-identification nature of both types of target categories. Whereas the target categories drinker and abstainer can be strongly related to one’s identity, it could be argued that this is different for alcohol and soft drink. Since the targets have to be classified into the attribute categories me and others during an Identity IAT, this could be an important distinction to make. In terms of identity, it is possible to be a drinker or an abstainer, but it is not possible to be a glass of beer.
Limitations of the Training Study
It is possible to list a few possible limitations regarding the training task and the design of the study. A first point of limitation could be the duration of the training sessions. Virtually all
participants indicated that the duration of the individual training sessions was too long. Therefore, it is possible that levels of motivation, attention and energy decreased during sessions, which could in turn have had negative effects on task performance. Average energy levels of participants were indeed lower during and after sessions when compared to before sessions, although it is not clear whether this difference is the result of session duration; and what influence this may have had on the training outcome. Therefore, it could be argued that session duration should be shortened in follow-up studies. However, to preserve the total number of trials, this would imply that the total number of trials should be distributed over more individual training sessions. More frequent training sessions,
