Can cognitive bias modification simultaneously target two behaviors? Approach bias retraining for alcohol and condom use

Abstract

This study tested the effectiveness of a cognitive bias modification (CBM) intervention to simultaneously reduce approach biases toward alcohol and increase approach biases toward condoms among high-risk young adults. Participants (N = 102) were randomly assigned to either a training condition or a sham-training condition. Participants in the training condition were trained to make avoidance movements away from alcohol stimuli and approach movements toward condom stimuli over four training sessions. Approach biases and behavior were assessed at pretest, posttest, and 3-month follow-up. Approach biases changed for both stimulus categories in accordance with training condition. Condom behavior and attitudes also changed as a function of training condition, such that participants in the training condition reported fewer instances of condom nonuse and a more positive attitudes toward condoms at a three-month follow-up. Participants in both conditions had significant reductions in alcohol consumption following the intervention and did not differ by training condition.

Can cognitive bias modification simultaneously target two behaviors? Approach bias retraining for risky sex and alcohol use

Cognitive bias modification (CBM) interventions are emerging as effective intervention tools for treating psychiatric disorders, reducing problematic behaviors, and changing attitudes. CBM interventions are effective because they target implicit cognitions, which underlie the behavioral decision-making process. Namely, if an individual is trained to attend towards or attend away from specific information, behavioral patterns may also change. CBM interventions have been successful in reducing stress (MacLeod, Rutherford, Campbell, Ebsworthy, & Holker, 2002), anxiety (Amir et al., 2009; Beard, Weisberg, & Amir, 2011), and alcohol use (Eberl et al., 2013, 2014; Wiers, Rinck, Kordts, Houben, & Strack, 2010; Wiers, Eberl, Rinck, Becker, & Lindenmeyer, 2011) through targeting implicit automatic biases.

One approach to modifying implicit automatic biases is through targeting specific motor movements. Social psychologists have linked automatic approach and avoidance tendencies with attitudes, behaviors, and emotions (Cacioppo, Priester, & Berntson, 1993; Chen & Bargh, 1999), which dates back to William James (1884), who suggested that approach motor movements (i.e., muscle flexion) would be associated with positive emotions and evaluations and avoidance motor movements (i.e., muscle extension) would be associated with negative emotions and evaluations. Indeed, a growing amount of literature supports James’s theory finding that approach motor movements (e.g., pulling a computer joystick) are associated with increased positive evaluations of stimuli (e.g., words, images, etc.), while avoidance motor movements (e.g., pushing a computer joystick) are associated with increased negative evaluation of stimuli (Cacioppo et al., 1993; Chen & Bargh, 1999; Neumann & Strack, 2000). Cacioppo and colleagues (1993) suggest that these associations are a product of classical conditioning in which humans consistently approach (e.g., grab or reach for) preferred objects (e.g., foods, beverages, etc.) and avoid (e.g., push away) aversive stimuli. Thus, over time, humans develop learned associations between arm flexion and positive evaluations and, conversely, arm extension and negative evaluations.

Recent work has applied the aforementioned approach and avoidance movement paradigm to computerized assessment and intervention approaches. Stimuli presented on a computer screen can elicit a motivational orientation and subsequent behavioral response similar to a physical object (Strack & Deutsch, 2004). The Approach-Avoidance Task (AAT; Rinck & Becker, 2007), is a computerized program in which participants push or pull a joystick in response to the format of an image presented on a computer screen (e.g., push when in portrait, pull when in landscape). The AAT features a zooming effect to simulate the sensation of approaching when pulling the joystick and avoiding when pushing the joystick, such that the images increase in size when the joystick is pulled and decreases in size when the joystick is pushed. The AAT is an effective tool for measuring behavioral response to a category of stimuli via reaction time measurements of approach (i.e., pulling/arm flexion) and avoidance (i.e., pushing/arm extension) movements with a joystick in response to the stimuli.

Approach bias is the behavioral action tendency to be faster to approach rather than avoid cues for a stimulus category. The AAT is effective in assessing implicit approach biases for various stimuli, including alcohol stimuli (Wiers et al., 2009), cannabis (Cousijn, Goudriaan, & Wiers, 2011), gambling (Boffo et al., 2018), sexual stimuli (Hofmann, Friese, & Gschwendner, 2009; Simons, Maisto, Wray, & Emery, 2016), and condom stimuli (Simons et al., 2016). A modified version of this task is also used to retrain participants’ implicit biases toward or away from stimuli by presenting the target category of stimuli predominantly in one format (e.g., push or pull; Eberl et al., 2013; 2014; Wiers et al., 2010; 2011). Given the associations between arm flexion and positive evaluations and arm extension and negative evaluations, training an individual to respond to certain stimuli with arm flexion or arm extension will subsequently change their approach or avoidance biases, respectively.

Generally speaking, there have been two primary approaches used to evaluate the AAT in approach bias retraining. First, proof-of-principle studies have been conducted to evaluate the hypothesized relationships between bias and behavior among participants who are not necessarily motivated to change their behavior (e.g., training one group toward alcohol and another group away from alcohol, cf. Field & Eastwood, 2005; Wiers et al., 2010). For example, Wiers and colleagues (2010) successfully modified participant action tendencies and the effects of the training generalized to subsequent drinking behavior, such that individuals who were trained to avoid alcohol drank less alcohol, while those trained to approach alcohol showed increases in alcohol consumption. The second approach to evaluating the effects of AAT retraining is via randomized controlled trials (RCTs) among clinical samples in which participants have an objective to change behavior. RCT studies have demonstrated success in retraining participants’ implicit action tendencies away from alcohol in alcohol-dependent participants at an inpatient treatment facility (Eberl et al., 2013, 2014; Manning et al., 2016; Rinck et al., in press; Wiers et al., 2011). Among large scale RTCs using this methodology (e.g., five training sessions) one found a medium effect (f² = 0.16) with significant effects at one-year follow-up (Wiers et al., 2011) and two others found 8.5% less relapse one year post-treatment (Eberl et al., 2013; Rinck et al., in press).

Although there have been recent proof-of-principle studies that have failed to replicate these findings (Lindgren et al., 2015; see Cristea, Kok, & Cuijpers, 2016 for meta-analysis), the aforementioned findings indicate that AAT training programs can have significant effects on subsequent behavior regardless of whether they are increasing or decreasing implicit approach tendencies. Although recent research investigated the impact of a CBM training program on decreasing maladaptive behaviors (e.g., Eberl et al., 2013, 2014; Wiers et al., 2010; 2011) and increasing adaptive behaviors (Taylor & Amir, 2012) independently, no studies have systematically examined the effect of a CBM training that simultaneously decreases maladaptive behaviors (i.e., alcohol use) and increases healthy protective behaviors (i.e., condom use). Furthermore, only one study has examined two active experimental categories in the same joystick AAT task for assessment purposes (i.e., sexual stimuli and condom stimuli; Simons et al., 2016), and no study to date has used CBM in the domain of condom use.

Alcohol Consumption and Condom Nonuse

Alcohol use is very prevalent among college campuses and the majority of college students consume alcohol, making it the most abused drug among college students (SAMHSA, 2013). According to SAMHSA (2013) over 60% of college students drank alcohol in the past month and nearly 40% are current binge drinkers. Not surprisingly, nearly one third of college students meet criteria for an alcohol use disorder (Blanco et al., 2008; Knight et al., 2002; Wechsler & Nelson, 2008). Thus, the college environment is strongly associated with extreme levels of alcohol use and related problems, one of which is sexual risk taking (e.g., condom nonuse).

Condom nonuse is associated with many negative outcomes including unplanned pregnancies and STIs. According to the American College Health Association (2013), 70% of college students in the United States engaged in oral, vaginal, or anal sex during the past year. Among college students who have engaged in vaginal sexual intercourse during the past 30 days, 50% reported not using a condom and 44% did not use any form of contraceptive. Consequently, sexually active individuals under the age of 25 account for nearly half of all new sexually transmitted infections (STIs; Satterwhite et al., 2013; CDC, 2014) and approximately one of every four sexually active adolescent females have a STI (Forhan et al., 2009). Consequences associated with condom nonuse are severe and costly on both an individual and societal level, with STIs costs at approximately $16 billion in direct medical costs per year (CDC, 2014).

Young adults are at elevated risk for both problematic drinking and condom nonuse (ACHA, 2013; Certain, Harahan, Saweyc, & Fleming, 2009). Alcohol use, and the disinhibition association with alcohol consumption, has been identified as a significant predictor of sexual risk taking (Elifson, Klein, & Sterk, 2006; Hipwell, Stepp, Chung, Durand, & Kennan, 2012; Turchik, Garske, Probst, & Irvin, 2010). Although these outcomes can share some common psychosocial etiology, it is also possible that elevated drinking in part contributes to the observed increases in risky sexual behavior in this population. Research suggests that intoxication has a causal effect on risky sexual behavior (George et al., 2014; Maisto, Carey, Carey, Gordon, Schum, 2004; Purdie et al., 2011). Global association studies, experimental studies, and event-level research suggest that alcohol intoxication may increase the likelihood of unsafe sex. Specifically, alcohol use is commonly associated with greater number of sexual partners and lower probability of condom and other contraceptive use (Abbey, Saenz, & Buck, 2005; Bailey, Pollock, Martin, & Lynch, 1999; Brown & Vanable, 2007; Guo et al., 2002; Hipwell et al., 2012; Patrick & Maggs, 2009; Scott-Sheldon, Carey, & Carey, 2010). However, among studies examining alcohol intoxication and risky sexual behavior, results have not been consistent and indicate a less clear event-level association (Cooper, 2002; Lewis, Kaysen, Rees, & Woods, 2010). This inconsistency suggests that the relationship between intoxication and risky sex is a complex interplay between interpersonal, intrapersonal, and contextual factors. Indeed, recent event-level research indicates that the relationship between intoxication and condom nonuse was an accelerated curve, where as the likelihood of condom nonuse is significantly greater at higher levels of intoxication (Simons, Simons, Maisto, Hahn, & Walters, 2018). Other event-level research identified that the associations between alcohol use and sexual risk taking is attenuated among heavy drinkers (Neal & Fromme, 2007; Simons, Wills, & Neal, 2011), indicating that less experienced drinkers may be at a greater risk for engaging in sexual risk taking when intoxicated. These findings highlight the complexity of the relationship between alcohol and sexual behavior and underscore the need for continued investigation of these behaviors.

Alcohol use is associated with sexual risk via intoxication’s impact on motivation and behavioral skills. Specifically, intoxicated individuals exhibit negative attitudes and motivation toward condom use (Gordon & Carey, 1996), especially among those with greater sexual-related alcohol expectancies (Gordon, Carey, & Carey, 1997). Young adults who are both intoxicated and aroused also have a lower implicit approach bias for condoms, thus leading to the increased risk for unprotected sex (Simons et al., 2016). Moreover, young adults who are intoxicated have diminished behavioral skills to initiate and/or negotiate condom use (Gordon et al., 1997; Maisto et al., 2004). Correspondingly, studies investigating the effectiveness of alcohol-reduction interventions also found significant subsequent reductions in sexual risk taking (Avins et al., 1997; Carey et al., 2004), further emphasizing the complex interplay between intoxication and sexual risk. High levels of alcohol use and lack of condom use are problematic behaviors that often co-occur and can result in significant consequences. As such, it is vital to develop innovative approaches that simultaneously target both sexual risk taking and problematic alcohol use.

Implicit Processes in Risky Sex and Alcohol Use

Behavioral decision making is impacted by a dynamic interaction between prior learning, reflective processing, and affective-motivational processes (Gladwin, Figner, Crone, & Wiers, 2011; Wiers & Gladwin, 2017). Further, time-dependent processes are theorized to play an important role in the balance between reflective processing and automatic biases (Cunningham, Zelazo, Packer, & Van Bavel, 2007; Gladwin et al., 2011; Gladwin & Figner, 2014). Specifically, automatic approach tendencies decay after delayed stimuli presentation (Gladwin, Mohr, & Wiers, 2014). This time-specific decision making may be especially relevant to risky sexual behavior because decisions about sexual behavior (e.g., using a condom) are made in the context of ongoing sexual stimuli. That is, the time a person needs to activate prior learning and to evaluate potential consequences may play a key role in the decision whether or not to use a condom, particularly when other competing contextual factors might also be present (e.g., sexual arousal, intoxication, etc.). Indeed, studies investigating action tendency biases for sexual images indicate a positive association between implicit approach biases for sexual stimuli and increased risk behavior (Hofmann et al., 2009; Simons et al., 2016). Sexual risk taking may occur when an individual’s implicit approach bias toward sexual stimuli is stronger than the subsequent approach bias toward condoms, especially in situations where sexual cues are more frequent than cues for condoms.

Correspondingly, the attentional myopia model posits that other influences, such as competing cognitive pressures, elicit a myopic effect on attention (Mann & Ward, 2007). Due to attentional myopia, an individual may inaccurately assess the costs and benefits of sexual risk taking. Engaging in sexual risk taking results in immediate satisfaction of sexual urges and desires. Conversely, the negative costs associated with sexual risk taking, such as contraction of STIs and unwanted pregnancy, occur later, if at all. Thus, the immediate benefits of engaging in sexual risk taking may overpower the costs of latent and probabilistic negative outcomes. Similarly, alcohol has a myopic effect on an attention, such that immediate and salient cues have a greater impact on an individual’s behavior and decision-making during acute alcohol intoxication (Steele & Josephs, 1990). The alcohol-myopia theory suggests that a response conflict occurs when individuals are faced with competing alternative options in a behavioral decision, especially with regard to decisions that may require further evaluation and inhibition (e.g., the decision of whether or not to use a condom with an unknown partner). Intoxication is theorized to inhibit the response conflict process, leading to more extreme behavioral responses. Individuals may be at increased risk for engaging in sexual risk taking while intoxicated because the perceived benefits are immediate and more salient compared to the cues that may inhibit these behaviors. Thus, modifying implicit biases for both alcohol and condoms can decrease problematic alcohol use subsequently reducing the myopic effects of alcohol, while also increasing the salience of condoms.

Current Study

The current study was a proof-of-principle investigation that tested a CBM intervention that aimed to simultaneously target alcohol-approach/condom-avoidance implicit processes. To date, no studies have systematically examined the effect of a combined CBM training to simultaneously decrease one behavior and increase another, and only one previous study (Simons et al., 2016) has examined the AAT as an assessment tool with two active stimulus categories in the same joystick AAT task. The hypotheses for the current study were: (1) The CBM training would be associated with changes in implicit approach and avoidance tendencies to alcohol and condoms, such that individuals in the training group would have reductions in approach bias for alcohol stimuli and an increased approach bias for condom stimuli. (2) The CBM training would decrease alcohol use and condom nonuse at the 3- month follow-up. (3) Cognitive biases post-intervention would mediate changes in corresponding behavior at 3-month follow-up. (4) Changes in alcohol use would mediate changes in condom nonuse.

Method

Design

This study employed a 2 (Training: training/sham) × 3 (Time: pretest/posttest/3-month follow-up) mixed design. Participants were randomized to a training (i.e., experimental) or sham (i.e., control) condition. Following Wiers and colleagues (2010) procedure, the intervention and sham-intervention (i.e., control) occurred over the course of four training sessions. Repeated measures of the implicit approach assessments as well as self-report assessments of drinking and sexual risk behavior were conducted at baseline, 1-week post intervention, and at a 3-month follow-up.

A priori power analyses were conducted using G*Power 3.1 (Faul, Erdfelder, Buchner, & Lang, 2009). Previous research investigating the effectiveness of CBM trainings, using similar methodology, found a moderate-to-large effect size (f₂ = .16; Wiers et al., 2011). However, due to the novel nature of simultaneously training participants to avoid and approach different stimuli, a power analysis was conducted for a 2 (Training: experimental/sham) × 3 (Time: pretest/posttest/3-month follow-up) repeated measures ANOVA to detect a small-to-moderate effect (f = .08, α = .05) with a power of .80. Results indicated that 70 participants would be needed.

Participants

Participants were 102 students between the ages of 18 and 24 (M = 19.98; SD = 1.46) from a public university. Participants completed a brief telephone screen to determine eligibility and an online baseline questionnaire that assessed demographic information, alcohol use history, and sexual behavior history. The inclusion criteria for study was consuming alcohol at least moderately (e.g., at least seven drinks per week for women and at least 14 drinks per week for men; NIAAA, 2017) or engaging in binge drinking episodes more than three times during the three months preceding data collection and had unprotected sexual intercourse with a casual partner during the three months preceding data collection.

A total of 1,507 participants were formally screened online and approximately 500 participants were informally screened via telephone in order to recruit the full experimental sample. One hundred and two participants were randomized. Eleven people participated in some aspect of the study but were removed from the final analyses. Two participants, one from each condition, attended the first session but did not complete the training or follow-up sessions and thus were excluded from the final sample. Nine additional participants were initially screened via telephone and provided verbal responses that fit within the inclusion criteria. However, upon completing the baseline assessments, these participants did not meet the inclusion criteria (e.g., eight participants did not engage in sexual intercourse with a casual partner prior to data collection and one participant did not meet the inclusion criteria for alcohol use). Of the nine participants who did not meet criteria, four were from the training condition and five were from the control condition. Thus, the data from the eleven aforementioned participants was not included in the analyses and 91 participants (46 in the training condition; 45 in the control condition) were included in the final sample. The final sample was comprised of 31 males (34%) and 60 females (66%). The majority of the participants were white (93%), 2% were Native American, 2% were Asian, and 2% were multiracial. Four percent of the sample was Hispanic. Of the 91 participants included in the final analysis, 85 completed the three-month follow-up and six participants were unable to be contacted for follow-up (two in the training condition; four in the control condition). Thus, there was a 93% retention rate for participants in this study.

Measures

Daily Drinking Questionnaire – Modified (DDQ-M; Dimeff, Baer, Kivlahan, & Marlatt, 1999).

The DDQ-M was used to assess participants’ daily alcohol use during an average week. The DDQ-M provides participants with a 7-day grid in which participants record the average amount of drinks consumed over the past 3 months. The DDQ-M has been shown to be a valid measure of alcohol consumption with multiple samples of college students (Baer, Kivlahan, Blume, McKnight, & Marlatt, 2001; Larimer et al., 2001; Marlatt et al., 1998). Total drinks per average week were used in the analyses.

Alcohol Use Disorders Identification Test (AUDIT; Saunders, Aasland, Babor, de la Fuente, & Grant, 1993).

The AUDIT is a 10-item self-report measure that is used to identify problematic drinking. The AUDIT is a widely-used alcohol screening assessment that has been validated among both college students (Fleming, Barry, & MacDonald, 1991) and the general population (Saunders et al., 1993). The Cronbach’s alpha was .82 for both the baseline assessment and the three-month follow-up.

HIV Risk Measure (HRM; National Institute of Drug Abuse [NIDA], 2013).

The HRM is a measure included in the “harmonized” instruments used for the Seek, Test, Treat, and Retain (STTR) initiative sponsored by the National Institute on Drug Abuse (NIDA). The HRM is a self-report measure of sexual risk behaviors that is based on the Women’s Health CoOp Baseline Questionnaire (Wechsberg, 1998). For the current study, the HRM assessed sexual risk behavior during the three months preceding the questionnaire administration. The HRM measures sexual behavior with a main partner, sexual behavior with casual partners, and sexual behavior under the influence of drugs and/or alcohol. Moreover, this measure allows for the assessment of number of sexual partners, number of times engaging in sexual intercourse, instances of condom use, and percentage of condom use. The HRM has been used extensively within NIDA-funded research and has been demonstrated to be a reliable measure of risk behavior (e.g., test-retest reliability >.75; Wechsberg, 1998; Wechsberg et al., 2003). Condom nonuse was calculated by subtracting instances of condom use from total number of times engaging in sexual intercourse. Percentage of condom use was calculated by dividing instances of condom use by total instances of sexual intercourse. Only participants who reported engaging in condom nonuse with a casual partner qualified for the study. Moreover, if a participant did not have sexual intercourse with a casual partner during the follow-up period, they were coded as missing on condom use variables.

Multi-Factor Attitudes Towards Condoms Scale (MFACS; Reece, Herbenick, Hollub, Hensel, & Middlestadt, 2010).

The MFACS is a 14-item scale that was used to assess participants’ attitudes toward condoms. The scale consists of three factors, affective, perceived effectiveness, and manageability. Each item is scored on a 7-point semantic differential scale using polar adjectives in assessing condom attitudes. Each item has the prompt, “I would describe condoms as:”. Example adjectives phrases “Effective at preventing sexually transmitted infections” versus “Not effective at preventing sexually transmitted infections” and “Comfortable” versus “Uncomfortable.” Total scores were calculated and higher scores indicate a more negative attitude towards condoms. The Cronbach’s alpha was .78 for the baseline assessment and .80 at the three-month follow-up. The MFACS has been validated for assessment of condom attitudes among college students (Hollub, Reece, Herbinick, Hensel, & Middlestadt, 2011).

Implicit Approach Avoidance Task

The approach avoidance task utilized the Alcohol Approach Avoidance Task in the Inquisit programming environment based on the task developed and modified by Wiers et al., (2009; 2010; 2011). For the current study, 20 alcohol-related images and 20 condom-related images were used as stimuli. The instructions for completing the Approach-Avoidance Task (AAT) were automated.

The assessment AAT consisted of 80 trials of a computerized task in which participants push or pull a joystick in response to the orientation (landscape or portrait) of a picture presented on a computerized screen. The pictures consisted of alcohol images (e.g., beer, liquor, and/or wine) and condom images (e.g., condoms, condom packaging, etc.). Research using joystick tasks indicates that positive and negative stimuli elicit pulling and pushing motions, respectively (Cacioppo et al., 1993). Thus, the automated instructions stated that when pushing or pulling the joystick, participants should imagine pulling the image toward them and pushing the image away. The task featured a zooming effect to simulate the sensation of approaching when pulling the joystick and avoiding when pushing the joystick. The image zoomed-in and increased in size when the joystick was pulled, and the image decreased in size when the joystick was pushed. During the assessment AAT images were presented in landscape and portrait format, such that participants approached and avoided equally for both alcohol and condom stimuli.

The CBM training task for this study followed that of Wiers et al. (2011). However, all of the alcohol pictures were in landscape (avoid) format and all of the condom stimuli were in portrait (approach) format. The task was designed to pair alcohol with an avoidance movement and the condom stimuli with an approach movement. The training task included 200 trials and took approximately ten minutes with a brief break in between. Individuals in the sham (control) condition also completed 200 trials. However, for the sham condition, half of each stimulus-category (e.g., condoms and alcohol) was presented in portrait and 50% in landscape format. Thus, the sham-training task did not attempt to modify response tendencies though manipulating the pairings and alcohol and condom images were pushed and pulled an equal number of times. The sham-training condition controls for exposure effects of the task. Participants completed four training sessions, or sham-training sessions, depending on group randomization (Wiers et al., 2011). Spearman-Brown split-half reliabilities for reaction times at pretest, posttest, and three-month follow-up were .90, .96, and .96, respectively; reliabilities after D score calculations were .62, .33, and .52.

Procedure

Participants completed a telephone screen and an online screening questionnaire to determine eligibility for the study. The experimental portion of the study consisted of six appointments. Participants were told that they were participating a study about reaction time and were unaware that this study was examining behavior modification. The first four appointments were attempted to be scheduled on consecutive days. However, there was occasionally a one- or two-day gap between training sessions if a participant started the protocol midweek and could not present to the laboratory on Saturday and/or Sunday. During the initial appointment, participants completed a baseline assessment AAT and, depending on their group assignment, either the sham-training or CBM-training AAT. The first appointment lasted approximately 15 minutes. The second through fourth appointments consisted only of the sham-training or CBM-training AAT, and lasted approximately ten minutes. The fifth and sixth appointments lasted approximately ten minutes and were a one-week and three-month follow-up appointment, respectively. During the follow-up appointments, participants completed the assessment AAT and an online outcome questionnaire. All appointments took place in a private research room and were conducted by two trained graduate student members of the research team. Participants were compensated $50 for participating.

Results

Statistical Approach

The AAT data were examined for potential outliers or incorrect responses. Response latencies on the AAT assessments that were faster than 300 ms or slower than 2000 ms were removed. The proportion of correct responses for each participant was examined. Three participants, two from the training group and one from the control group, responded correctly <75% of the time and were omitted from the analysis. Mean response latencies were calculated for each sub-block of responses (i.e., approach alcohol, avoid alcohol, approach condoms, avoid condoms), using only correct responses for each sub-block. Next, penalties were applied for incorrect responses, such that incorrect responses were coded as the mean reaction time for each sub-block plus two standard deviations (6.54% of responses; Barkby, Dickson, Roper, & Field, 2012). The number of incorrect responses did not vary as a function of condition (t = −0.160, p = .873 in two-sample t-test). Finally, D scores were calculated for each participant based on the procedures of Greenwald, Nosek, and Banaji (2003). Participants’ mean reaction times during the approach alcohol trials were subtracted from the mean reaction times during the avoid alcohol trials. These scores were then divided by the participant SD across all alcohol trials (i.e., ([Mean Alcohol Approach – Mean Alcohol Avoid]/SD across all alcohol trials)). This was repeated for the condom trials. Positive D scores indicate an approach bias, while negative scores indicate an avoidance bias. The advantage of using D scores (compared to simple differences scores or median reaction times) is that they are less vulnerable to biases that occur as a result of differences in average reaction time (Sriram, Greenwald, & Nosek, 2010).

Descriptive Statistics

Table 1 presents descriptive statistics across time by treatment condition. Two-sample t-tests were analyzed and there were no significant differences between group on any baseline variables or percentage of correct trials. Figure 1 presents biases, alcohol use, and condom use by group across time. Associations between the implicit approach-avoidance biases and drinking and condom-related behaviors at baseline and 3-month follow-up were also tested, See Table 2.

Table 1.

Descriptive Statistics Across Time by Group

	Baseline M(SD)		Posttest M(SD)		3-month Follow-up M (SD)
	Training n=46	Control n=45	Training n=45	Control n=44	Training n=44	Control n=41
Alcohol Use	21.45 (15.40)	18.50 (10.90)			12.28 (9.38)	10.85 (7.78)
Binge	6.27 (3.67)	7.09 (4.58)			3.87 (2.94)	3.49 (3.07)
AUD-C	7.45 (2.21)	6.90 (2.07)			6.33 (2.22)	5.80 (2.16)
Frequency of Intercourse	12.81 (14.82)	12.45 (13.57)			12.52 (11.75)	13.94 (15.33)
Condom Use	3.62 (7.65)	2.52 (3.20)			5.00 (7.66)	1.68 (3.45)
Condom Nonuse	9.19 (12.79)	9.93 (13.52)			7.52 (11.22)	12.26 (15.63)
Condom Use (%)	25.73 (31.05)	29.27 (29.97)			43.39 (41.59)	26.07 (38.09)
Negative Condom Attitudes	45.64 (13.81)	46.36 (13.14)			43.43 (11.58)	50.67 (9.92)
Alcohol Bias	0.18 (0.43)	0.09 (0.46)	−0.21 (0.42)	−0.01 (0.40)	0.11 (0.40)	0.04 (0.49)
Condom Bias	0.07 (0.33)	0.10 (0.40)	0.39 (0.40)	−0.01 (0.34)	0.16 (0.35)	−0.08 (0.39)

Note. Boldface represents significant group differences at given time point at p < .05. n’s differ due to missing data. Alcohol Use = DDQ-M (i.e., self-report of drink consumed during a typical week over the prior 3 months). Binge Drinking = Number of days over past 30 days that participant engaged in binge drinking (e.g., ≥ 4 standard drinks for women; ≥5 standard drinks for men). AUD-C = Consumption subscale of AUDIT. Frequency of intercourse = number of instances of vaginal or anal intercourse during prior three months. Condom Use = Frequency of condom use during instances of vaginal or anal intercourse over prior three months. Condom Nonuse = Frequency of vaginal or anal intercourse instances without using a condom. Condom Use % = Percentage of times using a condom during vaginal or anal intercourse instances. Alcohol and Condom Bias = D scores on the alcohol and condom AAT. Positive values indicate approach bias, negative value indicates avoidance bias.

Figure 1.

D scores on the alcohol and condom AAT for participants in training and control groups at baseline and follow-up. D scores were derived at pretest and posttest. Positive values indicate approach bias, negative value indicates avoidance bias. Error bars: ± 1 standard error. *p<.05.

Table 2.

Correlations Between Observed Variables at Baseline and 3-Month Follow-Up

Note. Baseline correlations are above the diagonal, follow-up values are below the diagonal. Sex (Men = 1, Women = 0). Alcohol Use = Self-report of drink consumed during a typical week over the past 3 months. Condom Nonuse = Frequency of intercourse without the use of a condom over the three-months prior to data collection. % Condom Use = Percentage of times a condom was during intercourse. Condom Attitudes = Participant attitudes toward condoms, higher scores indicate a negative attitude toward condoms. Alcohol and Condom Bias = D scores on the alcohol and condom AAT. Positive values indicate approach bias, negative value indicates avoidance bias. Treatment Condition (Training = 1, Control = 0).

^***p<.001,

^**p<.01,

^*p<.05.

Effect of Intervention on Implicit Biases

The effect of the training on implicit approach and avoidance biases for alcohol and condom stimuli was tested with 2 (Condition: experimental/sham) × 3 (Time: pretest/posttest/3-month follow-up) repeated measures ANOVAs. First, the effect of the training on alcohol bias was tested. The overall model was significant F(89, 243) = 2.39, p < .001. There were significant main effects of condition, F(1, 243) = 9.52, p = .002, η² = .06 and time, F(2, 243) = 10.52, p = .001, η² = .12. Moreover, the Time × Condition interaction was also significantF(2, 243) = 3.95, p = .021, η² = .05. Contrast analyses indicated that the only significant simple effect was for the treatment condition between pretest and posttest F(1, 242) = 16.61, p < .001, and no significant simple effects for the control condition. These results indicated a reduction in approach bias for alcohol stimuli among participants in the training condition. The effect size for the mean differences in alcohol bias was a medium to large sized effect from pretest to posttest between groups (d_pp2 = −.65) and very small effect between pretest and three-month follow up (d_pp2 = −.05).

For condom biases, the overall model was significant F(89, 243) = 1.61, p = .005. There was a significant main effect of time, F(2, 243) = 3.82, p = .020, η² = .05, but the main effect of condition was not significant, F(1, 243) = 2.05, p = .154. The Time × Condition interaction was also significant, F(2, 243) = 7.55, p = .001, η² = .09. These results indicated an increase in approach bias among participants in the training condition. Contrast analysis indicated a significant simple effect for the treatment condition between pretest and posttest, F(1, 243) = 15.30, p < .001. Hence, for both alcohol and condom biases, there were expected treatment effects at posttest that diminished at the three month follow-up. Unexpectedly, there was also a simple effect in the control condition between pretest and follow-up, F(1, 243) = 4.86, p = .028, such that participants in the control condition had an approach bias for condoms at pretest and an avoidance bias for condoms at three-month follow-up. The effect size for the mean differences in condom bias was a large sized effect from pretest to posttest between groups (d_pp2 = 1.17) and medium to large sized effect between pretest and three-month follow up (d_pp2 = .74).

Effect of Intervention on Alcohol Use

The effect of the intervention on alcohol use was tested using 2 (Condition: training/sham) × 2 (Time: pretest/3-month follow-up) repeated measures ANOVAs. For average weekly drinking, the overall model was significant F(87, 166) = 2.18, p < .001 and there was a significant effect of time, F(1, 166) = 29.78, p < .001, η² = .28. However the main effect of condition was not significant F(1, 166) = 0.01, p = .916, η = .00. The Time × Condition effect was also not significant, F(1, 166) = 0.13, p < .718, η² = .00. These findings indicate that both groups had significant reductions in alcohol use between baseline and three-month follow-up assessments. The effect size for the mean differences in alcohol use between baseline and three-month was a small effect (d_pp2 = −.11).

Effect of Intervention on Condom-Related Outcomes

Condom outcomes were also tested using 2 (Condition: training/sham) × 2 (Time: pretest/3-month follow-up) ANOVAs. First the effect of the intervention on percentage of condom use (i.e., proportion of times using a condom/times not using a condom) was tested. The independent effects of conditional F(1, 155)= 1.83, p = .178 and time F(1, 155)= 1.34, p = .249 were not significant. The Time × Condition interaction was significant F(2, 155) = 3.93, p = .049, η² = .03. However, the overall model was not significant F(3, 155) = 2.05, p = .109, η² = .04. As such, the interaction effect should be interpreted with caution. The effect size for the mean differences in condom use percentage from pre to post assessment between groups was a significant medium to large sized effect (d_pp2 = .68). The effect of condition on condom attitudes was also assessed using a 2 × 2 repeated measures ANOVA. The overall model was significant F(87, 157) = 4.65, p < .001, η² = .83. The main effect of condition on condom attitudes was not significant, F(1, 157) = 3.68, p = .059, η² = .05, nor was the main effect of time F(1, 157) = 0.00, p = .950, η² = .00. However, the Time × Condition interaction was significant, F(1, 157) = 7.39, p = .008, η² = .10. Moreover, there was a medium to large effect for the mean differences in condom attitudes from pre to post assessment between conditions (d_pp2 = −.48).

Path Model of Bias on Condom Nonuse

The remaining hypotheses were tested with an autoregressive path model in Mplus 7.4 (Muthén & Muthén, 2015). Pretest implicit approach biases, alcohol use, and condom nonuse (i.e., number of instances of condom nonuse at baseline) were included as covariates with autoregressive paths to the respective follow-up variables. Frequency of intercourse during the follow-up period was controlled for in the path analysis as an exposure variable. Direct paths were specified from the intervention indicator (i.e., training condition) to both posttest bias scores and condom nonuse at follow-up. Lastly, direct paths were specified from both posttest approach bias scores and alcohol use at three-month follow-up to condom nonuse at three-month follow-up. Additionally, to examine whether participant sex moderated treatment outcomes, we included a condition × sex interaction term with paths to posttest biases and condom nonuse at follow-up. However, the interaction term did not have a significant effect on any of the endogenous variables and thus was excluded from the model to reduce the number of parameters. The path model was tested with the maximum likelihood robust estimator using full-information maximum likelihood estimation and Monte Carlo integration. Due the count outcome (i.e., condom nonuse), traditional fit statistics could not be calculated for the model. As hypothesized, the training condition was significantly associated with posttest biases for both condom and alcohol stimuli. Moreover, condition had a significant effect on condom nonuse at the three-month follow-up. However, posttest biases were not significantly condom-related outcomes, but the relationships were in the anticipated direction (i.e., negative association between condom approach and condom nonuse; positive association between alcohol approach and condom nonuse). As such, contrary to hypotheses, posttest biases did not mediate changes in alcohol or condom use.¹ There were no significant indirect effects in the model (i.e., p values ranging from .143-.747). See Figure 2.

Figure 2.

Path model. Values are non-standardized effects, standard error in parentheses. Solid lines indicate significant effects. Dashed lines indicate non-significant effects. Sex was originally included as a covariate to all exogenous variables but had no significant effects and was thus removed. T1 = Pretest scores. T2 = Posttest scores. T3 = 3-month Follow-Up scores. Alcohol use is the number of standard alcoholic drinks during average week over past three months. Frequency of intercourse was included as an exposure variable.

Discussion

The overall objective of the current study was to examine the efficacy of a cognitive bias modification intervention to simultaneously retrain approach biases for two stimulus categories, alcohol and condoms. Specifically, the effect of the intervention on automatic approach tendencies for alcohol and condom related behavior was tested, as well as the effect of the intervention on subsequent behavior at a three-month follow-up. It was hypothesized that the intervention would significantly decrease participants’ automatic approach tendencies for alcohol-related stimuli and increase participants’ automatic approach tendencies for condom-related stimuli. Moreover, it was hypothesized that the intervention would significantly improve participants’ attitudes regarding condoms. The intervention was also predicted to significantly decrease alcohol use and instances of condom nonuse during the three months following the intervention. Lastly, it was hypothesized that automatic approach biases post-intervention would mediate changes in corresponding behavior at 3-month follow-up, and that changes in alcohol use would mediate changes in condom nonuse.

The intervention was successful in modifying automatic approach tendencies for both alcohol and condom-related stimuli. Specifically, individuals in the training condition had significant reductions in alcohol approach bias and significant increases in condom approach bias. This finding is consistent with previous proof-of-principle research with undergraduates that have aimed to modify implicit approach/avoidance tendencies (e.g., Wiers et al., 2010), as well as RCTs with clinical samples (Eberl et al., 2013, 2014; Wiers et al., 2011). However, the current study was the first to simultaneously modify implicit approach/avoidance biases for two stimulus categories using the same task.

There was also a significant effect of the training on condom use and condom attitudes at the three-month follow-up, indicating that the intervention was successful in modifying both behavior and attitudes related to condom use. In contrast, consistent with some recent findings, there was not a significant effect of the training on alcohol-related outcomes (see Cristea, Kok, & Cuijpers, 2016; Lindgren et al., 2015). This lack of effect could be due to a number a factors. First, the participants in this study were not treatment seeking, nor were the participants informed that they were participating in an intervention. Although the CBM targets implicit associations, participants with a motivation to reduce their alcohol use may be more likely to benefit from this type of intervention. Moreover, although college students consume alcohol at high levels, the high levels of consumption could be due to environmental factors, rather than a strong implicit bias.

Lastly, there were no significant associations between changes in biases and subsequent behavior at follow-up. In other words, the intervention was successful in modifying implicit biases and partially successful in modifying behavior and attitudes (i.e., accounted for change in condom behavior and attitudes, but not alcohol-behavior), but post-intervention biases were not associated with corresponding behavior at the follow-up.

Automatic Approach/Avoidance Tendencies

At baseline, all participants exhibited implicit approach biases for both alcohol and condom stimuli. However, at posttest, individuals in the training condition had mean scores indicating an avoidance bias for alcohol stimuli and a stronger approach bias for condom stimuli. The mean scores for the control condition indicated neither an approach nor avoidance bias at the posttest assessment, likely due to practice effects of the task. Interestingly, implicit biases for alcohol and condoms were not significantly associated with respective baseline behavior. In fact, only the combined bias (i.e., approach alcohol/avoid condoms) had a significant positive association with condom use, which is in the opposite direction as one would suspect. Similarly, the posttest implicit biases were also not significantly associated with behavior at the three-month follow-up appointment. The lack of relationship between implicit biases and baseline behavior could be due to the inclusion criteria, such that all individuals in the sample consumed relatively high levels of alcohol and engaged in unprotected sexual intercourse. As such, detecting an association among this particular sample could be more difficult than among a heterogenous population-based sample.

Consistent with previous research on the modification of implicit alcohol biases (Wiers et al., 2010; 2011), the intervention significantly reduced alcohol approach bias with a large effect between pretest and posttest (Cohen’s d = .86), and a small effect size pretest and follow-up (Cohen’s d = .15). Similarly, the intervention had a large effect on the training group’s condom approach bias between pretest and posttest (Cohen’s d = .69) and a small effect on condom approach bias between pretest and follow-up (Cohen’s d = .22). These findings are consistent with hypotheses, such that the intervention was successful in reducing approach bias for alcohol stimuli and increasing approach bias for condom stimuli. However, the decreased effect at follow-up indicates that the treatment effects eroded during that time period.

The Approach/Avoidance Task (AAT) is being used at an increasing rate since first implemented to examine automatic action tendencies for spider stimuli by Rinck and Becker (2007). However, nearly all studies using this intervention included one active stimulus category that is paired with an inactive, or control stimulus category. To date, only one published study has utilized the AAT as an assessment with two active target categories (Simons et al., 2016). This study successfully assessed approach biases for two conditions (erotic and condom stimuli). The current study extended these findings by, not only using two active categories (i.e., alcohol and condoms) in the assessment of approach biases, but also in the modification of biases. Moreover, these findings highlight the ability to simultaneously modify implicit approach biases for two active stimulus categories, without reducing the effect sizes that have been found in previous AAT CBM studies that used only one active condition. Having two active conditions, however, is not without its own limitations. Specifically, studies that utilize only one active condition provide an unambiguous means to interpret approach and avoidance biases. The use of two active conditions in the current study does not allow for the differentiation between actual approach biases for the stimulus categories versus a faster approach reaction irrespective of the category.

Alcohol Outcomes

Both the training and the control group had significant decreases in alcohol use from baseline to three-month follow-up on all measures of alcohol use (i.e., Daily Drinking Questionnaire, AUDIT, and binge drinking). These significant reductions did not differ by treatment condition. As such, the decreases in alcohol consumption cannot be attributed to the intervention. However, it is important to note that all participants demonstrated an approach bias for alcohol stimuli at baseline, regardless of treatment condition. Due to the effect of the intervention, the mean bias for the training condition changed from an approach bias for alcohol to an avoidance bias for alcohol (i.e., D-score = −0.21). Similarly, due to repeated exposure to the sham-training, the control group exhibited practice effects, such that the mean alcohol bias changed from an approach bias for alcohol to no bias for alcohol (i.e., D-score = −0.01). The repeated exposure to the alcohol-stimuli in half of the sham trials and subsequent decrease in approach bias in the control condition could be a potential explanation for the significant decrease in alcohol consumption across both the training and control conditions. Indeed, the decrease in alcohol use is consistent with a recent meta-analysis of CBM RTCs, which found that sham-controlled conditions might exhibit larger effects than training conditions (Boffo et al., in press). Altogether, these findings underscore the pitfalls associated with sham training as a control condition and highlight the need for alternative techniques for comparing against the CBM treatment.

It is also important to note that neither pretest alcohol bias nor posttest alcohol bias was significantly associated with drinking behavior. This is finding is consistent with previous studies using the Alcohol AAT in which there were no significant associations between changes in approach biases and subsequent follow-up behavior (Wiers et al., 2010; 2011; Wiers et al., 2017). Although the aforementioned studies did not find significant associations between postintervention alcohol bias and drinking behavior, those studies did find significant differences between treatment conditions in drinking behavior at follow-up.

Condom Outcomes

The most promising findings from this study were those associated with condom-related outcomes. Specifically, the Time × Condition interaction was significant for both condom attitudes and the percentage of condom use, indicating that the intervention was significant in not only increasing participants’ condom use, but also in improving their explicit attitudes with regard to condoms. The significant effect of condition on condom behavior and condom attitudes supports the theory that physical motor movement can influence motivation and behavior (Strack & Deutsch, 2004).

The path analysis examined the effect of the treatment condition on implicit biases and behavior, and also examined the mediating effect of posttest biases on follow-up behavior. There was a significant effect of condition on condom nonuse at the three-month follow-up, over and above the effects of posttest biases, alcohol use, baseline condom nonuse, and frequency of intercourse. However, the effect of alcohol use was not significant in the path model. The significant effect of condition on condom nonuse indicates that the training did significantly decrease rate of condom nonuse. However, the lack of indirect effect from condition to condom nonuse via condom approach biases does not support the hypothesized mechanism of action (i.e., increasing approach bias for condoms would decrease instances of condom nonuse). This finding is consistent with previous CBM studies using the AAT (Wiers et al., 2010; 2011; Wiers et al., 2017), such that the training was associated with post-intervention biases and follow-up behavior, without significant associations between biases and behavior. However, mediation was found in a larger study (Eberl et al., 2013) and re-analysis of the Wiers et al. (2011), with estimates of associations and control parameters (Gladwin et al., 2015). The findings of the current study accentuate a need for further investigation to better understand and measure how the approach avoidance task changes behavioral outcomes.

Although this study has many important strengths, it is not without limitations. First, motivation to change and participant awareness were not assessed. Participants may have been more motivated to engage in safe sex, than for alcohol, which could explain the significant effects on condom behavior but not alcohol. Second, two-thirds of the sample was female and the sample lacked racial and ethnic diversity thus limiting generalizability. Moreover, an undergraduate sample was used. Although problematic alcohol consumption and risky sexual behavior are prevalent among college students (ACHA, 2013; Certain et al., 2009), the effects that were demonstrated in the current study may not generalize to other populations. In addition, because participants were trained with condom and alcohol stimuli at the same time, it was not feasible to detect a causal relationship between drinking behavior and condom behavior. Future studies would benefit from using a more rigorous design that includes more treatment conditions. An additional limitation was that the same AAT stimuli was used during the assessment and training phases of the study. Therefore, we were unable to discern if participants learned the task with the specific stimuli or if there was an overall shift in evaluations for the target stimulus category. Finally, the relatively small sample size limited the statistical approaches. Future investigations using a larger sample size could be able to detect effects that were less clear with the current sample.

The present findings raise a number of questions in this line of research that could address unresolved issues and improve clinical applications. First, although the present findings are promising, it remains to be established whether the combined training (i.e., approach condoms/avoid alcohol) outperforms modifications targeting each independently. A second question is, to what extent does motivation impact the treatment effect and, relatedly, should participants be informed about the intended purpose of the training? The inconsistency between proof-of-principle studies and RCTs with clinical samples has been attributed to motivation to change, or lack thereof (Wiers et al., 2018). Systematically examining the roles of motivation and awareness could be fruitful. Finally, given the positive effect on condom outcomes, it would be interesting to explore retraining procedures for increasing other positive health behaviors. In conclusion, this brief intervention can simultaneously modify implicit biases for two target categories in the same task while improving health behavior outcomes.

What is the substantive, conceptual, or methodological contribution to knowledge that this work provides to the literature?

Results from cognitive bias modification studies using the approach-avoidance task (AAT) have been promising across a range of behaviors (see Kakoschke, Kemps, & Tiggemann, 2017 for review). However, no study to date has examined the efficacy of this type of intervention to simultaneously target two behavioral outcomes, nor has this intervention been systematically studied as a means for increasing health behaviors (e.g., condom use). As such, the current study not only provides a “proof of concept” that CBM using the AAT can modify biases for two discrete stimulus categories within the same training, but also demonstrated the efficacy of this intervention for increasing a healthy behavior (i.e., condom use).

How does this contribution and any conclusions or recommendations derived from it add to clinical psychological science?

This contribution directly informs the next steps in clinical psychological science research examining risk reduction, as well as behavior change in general. First, this study demonstrated the clinical utility of modifying biases for two behavioral outcomes, whereas previous studies have only targeted one. The ability to target two distinct behaviors through a brief computerized intervention could have a profound impact on increasing efficiency and reducing costs associated with changing potentially harmful behavior. Second, this study found that four brief training sessions can improve condom use among high risk college students, subsequently reducing the likelihood of interpersonal, intrapersonal, and public health consequences. Finally, the results from this study are promising, especially those with regard to increasing condom use, in that this intervention could also be implemented as a strategy to increase other healthy and positive behaviors.