Consequence-Based Approach-Avoidance Training: A New and Improved Method for Changing Behavior

Abstract

The repeated performance of approach or avoidance actions in response to specific stimuli (e.g., alcoholic drinks) is often considered a most promising type of cognitive-bias modification that can reduce unwanted behavior (e.g., alcohol consumption). Unfortunately, approach-avoidance training sometimes fails to produce desired outcomes (e.g., in the context of unhealthy eating). We introduce a novel training task in which approach-avoidance actions are followed by affective consequences. Four experiments (total N = 1,547) found stronger changes in voluntary approach-avoidance behavior, implicit and explicit evaluations, and consumer choices for consequence-based approach-avoidance training in the food domain. Moreover, this novel type of training reduced self-reported unhealthy eating behavior after a 24-hr delay and unhealthy snacking in a taste test. Our results contrast with dominant (association-formation) accounts of the effects of approach-avoidance training and support an inferential explanation. They further suggest that consequence-based approach-avoidance training, and inference training more generally, holds promise for the treatment of clinical behavior.

Keywords

approach-avoidance training consequences unhealthy food consumption cognitive-bias modification inferential theory open data open materials preregistered

Approach and avoidance represent two fundamental classes of behaviors that organisms have at their disposal when interacting with the environment. When one is faced with a positive (or appetitive) stimulus, it is often beneficial to approach that stimulus, whereas it is usually beneficial to avoid negative (or aversive) stimuli. Many theories assume that, as a result of this evolutionary benefit, evaluative processing is closely tied to approach-avoidance behavior (Lang, Bradley, & Cuthbert, 1990; Solarz, 1960; Strack & Deutsch, 2004). More specifically, these theories postulate that affective evaluation of a stimulus automatically predisposes people to approach or avoid that stimulus (Chen & Bargh, 1999; but see Rotteveel & Phaf, 2004). It is further assumed that these approach-avoidance tendencies can represent a cognitive bias that mediates unwanted or maladaptive behavior. For instance, a strong tendency to approach appetitive but unhealthy stimuli (e.g., unhealthy foods) can facilitate unhealthy behavior (e.g., consuming such foods; Hofmann, Gschwendner, Friese, Wiers, & Schmitt, 2008).

A growing number of researchers have tried to modify maladaptive behaviors such as unhealthy eating by establishing changes in approach-avoidance tendencies via approach-avoidance training. In a typical approach-avoidance training task, participants repeatedly perform approach or avoidance actions in response to specific stimuli. For instance, in studies on addiction, participants consistently avoid appetitive stimuli that they might normally approach (e.g., alcoholic drinks) by moving away from the stimuli or by moving the stimuli away from them, and they approach control stimuli (e.g., water). Several studies have indicated that approach-avoidance training can be effective in the treatment of clinical conditions. For instance, avoidance training might have beneficial effects in the treatment of addiction (e.g., alcohol dependence: Wiers, Eberl, Rinck, Becker, & Lindenmeyer, 2011; or smoking addiction: Wittekind, Feist, Schneider, Moritz, & Fritzsche, 2015), and approach training might be effective for treating social anxiety (Taylor & Amir, 2012), spider phobia (C. R. Jones, Vilensky, Vasey, & Fazio, 2013), or depression (E. S. Becker et al., 2016). Yet several studies have also failed to find such effects. For instance, training people to avoid unhealthy foods is often ineffective for changing unhealthy eating behavior (e.g., D. Becker, Jostmann, Wiers, & Holland, 2015). Two meta-analyses even concluded that there is little reliable evidence for the effectiveness of approach-avoidance training interventions on many clinical outcomes (Cristea, Kok, & Cuijpers, 2015, 2016; but see Kakoschke, Kemps, & Tiggemann, 2017).

A recent review further established that, overall, data do not support dominant explanations of approach-avoidance training effects that assume that behavioral approach-avoidance tendencies reflect mental stimulus–response associations that are gradually changed on the basis of repeated stimulus–action pairings (A. Jones, Hardman, Lawrence, & Field, 2017; see also Spruyt et al., 2013). In light of this conclusion, we recently developed an alternate model that postulates that inferential processes underlie approach-avoidance training effects (Van Dessel, Hughes, & De Houwer, 2018). From this perspective, repeated performance of approach-avoidance actions in response to a stimulus (e.g., avoidance of alcoholic drinks) leads to the formation of inferences about evaluative properties of the stimulus (e.g., that alcohol is to be avoided). This inferential learning can then affect subsequent stimulus-related actions (e.g., alcohol consumption).

The inferential account assumes that approach-avoidance training effects depend on specific boundary conditions. One such condition is whether participants infer information about the consequences that result from responding in a particular way. It is well established that learning about positive or negative action consequences determines the performance of related actions (Thorndike, 1905) and that stimulus-based actions such as approach-avoidance are typically facilitated when more positive action consequences are anticipated (Eder & Hommel, 2013). It is possible that affective consequences of approach-avoidance actions are sometimes learned during approach-avoidance training. For instance, participants who approach feared stimuli (e.g., spiders; C. R. Jones et al., 2013) might learn that the approach responses do not lead to the anticipated negative consequences. Critically, however, typical approach-avoidance training does not specify in a clear manner what action consequences participants should learn, which might explain null effects and even contrast effects in past work. For instance, repeated avoidance of desired stimuli such as chocolate might sometimes be unpleasant (e.g., because chocolate looks tasty), which could hinder rather than facilitate future avoidance of the stimuli (D. Becker et al., 2015, Experiment 3).

With this in mind, we performed four experiments that tested the effectiveness of a novel type of approach-avoidance training in which approaching or avoiding specific stimuli consistently led to positive or negative consequences. In Experiment 1, one group of participants performed typical approach-avoidance training in which they consistently approached products of one unknown food brand and avoided products of another by moving either a mannequin or an avatar representing themselves toward or away from the products. Another group performed consequence-based approach-avoidance training in which they approached and avoided products from both brands with an avatar. Importantly, for one brand, approach always produced positive consequences, and avoidance always produced negative consequences, whereas the actions produced opposite consequences for the other brand. Action consequences were chosen so they would facilitate evaluative learning in the food context (i.e., improve or decline the avatar’s general health). In Experiment 2, we further supercharged consequence-based approach-avoidance training by making action consequences relevant to task goals (i.e., participants were instructed to try to maximize avatar health). On the basis of our inferential account, we predicted that this would facilitate the inferential step from action performance to evaluative learning and therefore enhance approach-avoidance training effects. Both experiments probed for effects on consumer choices, voluntary approach-avoidance responses, and implicit (i.e., automatic) and explicit (i.e., controlled) stimulus evaluations.

Experiments 3 and 4 extended our investigation to familiar healthy and unhealthy foods and examined the impact of consequence-based approach-avoidance training on (a) self-reported healthy eating behavior and intentions (completed 24 hr after the intervention; Experiment 3) and (b) the amount of unhealthy food that participants consumed in an ad libitum snack task (Experiment 4).

Method

Participants and design

A total of 600, 525, and 420 volunteers participated online in Experiments 1, 2, and 3, respectively, via the Prolific Academic website (https://prolific.ac). Participants in Experiment 4 were 184 undergraduate students from Ghent University. The sample size of all experiments was determined on the basis of an a priori power analysis, which determined that the sample sizes we recruited would provide sufficient power (> .80) to detect a small- to medium-sized effect. Prior to data collection, the target sample size for each experiment was preregistered, together with the study design, data-analysis plans, and experimental hypotheses, on the Open Science Framework. For all experiments and all measures, we predicted that consequence-based approach-avoidance training would lead to stronger effects than typical approach-avoidance training or control training. Raw data and experimental and analytic scripts are available via the Open Science Framework. After exclusions on the basis of preregistered criteria (for details, see the Supplemental Material available online), we retained the data of 519 (300 women; age: M = 34 years, SD = 12), 455 (288 women; age: M = 34 years, SD = 12), 389 (219 women; age: M = 34 years, SD = 13), and 184 (59 women; age: M = 20 years, SD = 2) participants. A total of 307 participants (78.9%) completed the second part of Experiment 3 (mean delay = 29 hr, SD = 4).

Procedure of Experiments 1 and 2

Approach-avoidance training task

After providing informed consent and completing demographic information, each participant performed one of three different versions of the approach-avoidance training task.

Mannequin task (Experiments 1 and 2)

The mannequin task was adopted from Woud, Maas, Becker, and Rinck (2013). It was selected as a typical approach-avoidance training task for this study because it has produced robust effects in the past (see the Supplemental Material). In this task, participants performed 80 trials in which they saw a stick figure (mannequin) that represented themselves along with a product from one of two novel food brands (named Vekte and Empeya; see Fig. 1). Depending on the color of the frame, participants approached the product by moving the mannequin toward it or avoided the product by moving the mannequin away from it. Products from one brand (e.g., Vekte) were always surrounded by the colored frame that had to be approached, and products from the other brand (e.g., Empeya) were always surrounded by the colored frame that had to be avoided. Whenever the participant made a correct response by pressing the up or down key on the keyboard, the mannequin moved toward (up) or away from (down) the food product.

Fig. 1.

Illustration of a trial with an approach response to a Vekte brand product in the four task conditions of Experiments 1 and 2.

Avatar task (Experiment 1)

The avatar task was designed for this study and served as a typical approach-avoidance training control for the consequence-based approach-avoidance training task described below. Before starting the task, participants selected whether a male or female avatar would represent them in the task. Training consisted of 80 trials in which participants first saw the avatar standing in front of a refrigerator. The refrigerator then gradually opened until a brand product with a colored frame appeared. A correct response (pressing the up or down key) resulted in the avatar moving toward or away from the brand product.

Avatar-consequences task (Experiments 1 and 2)

On each trial of this task, participants saw the avatar, the refrigerator, and the product with the frame but also a health bar that was presented above the avatar. After a correct response and resulting avatar movement, action consequences were presented: (a) The health bar gradually depleted, the sentence “I feel sick” appeared, and the avatar had an unhealthier appearance (negative consequences) or (b) the health bar filled, the sentence “I feel healthy” appeared, and the avatar had a healthier appearance (positive consequences). Products from both brands were presented equally often with blue and green frames (and, thus, were approached and avoided an equal number of times). Crucially, however, approaching one brand always produced positive consequences and avoiding it produced negative consequences, whereas approaching the other brand always produced negative consequences and avoiding it produced positive consequences.

Goal-relevant avatar-consequences task (Experiment 2)

This task was designed to supercharge the avatar-consequences task by making action consequences relevant for participants’ task goals. Participants were told that each time they approach or avoid the products, they would see the avatar become more healthy or sick and that their task would be to make the avatar as healthy as possible by performing these actions. During the task, there were no colored frames surrounding the brand products, and participants freely selected whether to move the avatar toward or away from the brand product. Contingencies among products, actions, and action consequences were the same as in the avatar-consequences task.

Outcome measures

After the approach-avoidance training, participants completed a question that probed consumer behavior. They then completed an implicit association test (IAT; Greenwald, McGhee, & Schwartz, 1998) that measured implicit evaluations of the two food brands and an explicit-rating task that measured explicit evaluations of those same brands. Finally, they completed a task that measured their voluntary approach-avoidance behavior.

Consumer choices

Participants were informed that we would be willing to send them a free sample of products from the two food brands, and they were then asked to indicate whether they would prefer products from Vekte, Empeya, both, or neither.

Implicit evaluations (IAT)

The IAT for Experiment 1 was constructed following the recommendations of Nosek, Greenwald, and Banaji (2005). Participants categorized eight attribute words (e.g., wonderful, evil) as positive or negative and four different versions of the brand logos as their respective names (Vekte or Empeya). In two experimental blocks of 56 trials each, stimuli related to one brand and “positive” shared a response key, and stimuli related to the other brand and “negative” shared a second response key. The IAT for Experiment 2 was personalized, with the category labels “I like” and “I dislike” used to categorize the attribute words (see Han, Olson, & Fazio, 2006).

Explicit evaluations

Participants indicated how positive or negative they considered each of the two brands by using a Likert-type scale ranging from 1 (very negative) to 9 (very positive).

Approach-avoidance behavior

Participants were told that they would perform a final task in which they would again see the brand products and the mannequin (or avatar). They were asked to imagine that they were now at home and that they were free to choose which action to make when they encountered the products (i.e., approach or avoid). Participants then completed 10 trials of the same approach-avoidance training task they had completed before. However, there were no colored frames surrounding the food products, and participants were free to either approach or avoid the products without any consequences of doing so.

Exploratory questions

We also probed the extent to which participants (a) had learned the correct contingencies between food brands and approach-avoidance actions (and action consequences in the consequence task conditions), (b) had imagined that they were the mannequin (or avatar), (c) liked the action of approaching or avoiding in general, and (d) had provided evaluative responses in order to comply with experimental demands (demand compliance) or to react against these demands (reactance).

Procedure of Experiments 3 and 4

Phase 1

Procedures were similar to those in Experiment 2, with the following exceptions. First, we used known healthy and unhealthy food products (e.g., carrots and cookies) as stimuli rather than products from novel food brands. Second, when participants started the experiment, they were asked to indicate the extent to which they (a) had the goal to eat healthy (healthy eating intention), (b) felt hungry at that moment (hunger), (c) often ate healthy (healthy eating behavior), and (d) found it difficult to cut down or stop eating unhealthy foods (healthy eating behavior difficulty). Third, participants performed either (a) a mannequin task in which they approached and avoided healthy and unhealthy foods an equal number of times (control condition) or (b) a goal-relevant avatar-consequences task in which approaching healthy foods and avoiding unhealthy foods always led to positive health outcomes, whereas avoiding healthy foods and approaching unhealthy foods always led to negative health outcomes (goal-relevant avatar-consequences-task; see Fig. 2). Experiment 3 additionally included a typical approach-avoidance training task in which participants always approached healthy foods and avoided unhealthy foods with a mannequin (mannequin-task condition).

Fig. 2.

Illustration of a trial with an approach response to a healthy food product in the three task conditions of Experiments 3 and 4.

Fourth, the outcome measures in Experiment 3 included a consumer-choice task, an explicit-evaluation-rating task, personalized IAT, and an approach-avoidance task that consisted of 12 free-choice approach-avoidance trials. Participants in Experiment 4 did not perform the free-choice approach-avoidance task but completed an ad libitum snack task, adapted from Haynes, Kemps, and Moffitt (2015), in which participants were presented with four full bowls of preweighed popular energy-dense snack foods: mixed candy (3.3 kcal/g), potato chips (5.3 kcal/g), M&Ms (4.7 kcal/g), and Cheese Flips (5.1 kcal/g). Participants were instructed to rate the foods on different sensory characteristics for use in an unrelated study. After providing these ratings, participants were informed that they could consume as much of these snacks as they wanted. Unbeknownst to participants, bowls were weighed at the end of the experiment. Fifth, the consumer choice task consisted of six trials in which participants indicated which of two food products they would prefer to receive a coupon for (Experiment 3) or actually receive (Experiment 4) after completing the experiment. Trials presented either trade-off pairs involving an unhealthy food and a less attractive healthy food (e.g., chocolate cookie and rice cake) or controlled pairs involving an unhealthy food and an equally attractive healthy food (e.g., banana and waffle). Finally, at the end of Experiment 4, participants reported task engagement for the approach-avoidance training phase (short version of the Dundee Stress State Questionnaire; Helton & Näswall, 2015) and temptation to eat the foods presented in the snack task (7-point Likert-type scale: 1 = not at all, 7 = extremely).

Phase 2 (Experiment 3)

Participants were contacted via the Prolific Academic website 1 day after completing the first experimental phase. They returned and completed four questions that probed healthy and unhealthy eating behavior, healthy eating intentions, and difficulty to stop unhealthy eating in the period between the first and second study phases. A final question asked participants to what extent they currently had the goal to eat healthily (healthy eating goal). All answers were provided on Likert-type scales ranging from 1 to 9.

Results

Results of Experiments 1 and 2

Consumer choices

Behavioral choice scores were computed by recoding responses to the consumer behavior question such that −1 indicated that participants selected only the negative brand (i.e., the brand that was consistently avoided in typical approach-avoidance training or for which approach produced negative and avoidance produced positive consequences in consequence-based approach-avoidance training), 0 indicated that participants selected both or neither brands, and 1 indicated that participants selected only the positive brand (i.e., the brand that was consistently approached in typical approach-avoidance training or for which approach produced positive and avoidance produced negative consequences in consequence-based approach-avoidance training). Scores were subjected to a Kruskal-Wallis test, which revealed significant main effects of approach-avoidance task condition for both experiments—Experiment 1: χ²(2, N = 519) = 14.66, p < .001; Experiment 2: χ²(2, N = 415) = 33.69, p < .001. Participants selected the positive brand more often than the negative brand in all conditions of Experiment 1 (see Table 1) and Experiment 2 (see Table 2), ps < .003, d_zs > 0.41. Crucially, participants selected the positive brand more often in the avatar-consequences condition than in the mannequin or avatar condition, ps < .003. Moreover, the positive brand was selected more often in the goal-relevant avatar-consequences condition of Experiment 2 than in both other conditions, ps < .035. For more detailed results of this and all other analyses, see the Supplemental Material.

Table 1.

Means and Effect Sizes for All Outcome Measures in the Three Task Conditions of Experiment 1

	Mannequin task(n = 173)			Avatar task(n = 171)			Avatar-consequences task(n = 175)
Measure	M (SD)	95% CI	d_z	M (SD)	95% CI	d_z	M (SD)	95% CI	d_z
Consumer choice	0.31 (0.54)	[0.22, 0.39]	0.56	0.33 (0.60)	[0.24, 0.42]	0.54	0.50 (0.68)	[0.40, 0.60]	0.73
Explicit rating	1.76 (2.90)	[1.28, 2.25]	0.61	1.99 (3.03)	[1.50, 2.48]	0.66	3.31 (3.75)	[2.83, 3.80]	0.88
IAT score	0.28 (0.45)	[0.22, 0.35]	0.62	0.28 (0.47)	[0.20, 0.33]	0.59	0.20 (0.48)	[0.13, 0.27]	0.42
Approach-avoidance	1.55 (2.80)	[1.13, 1.97]	0.55	1.71 (2.90)	[1.27, 2.16]	0.59	2.29 (3.21)	[1.85, 2.73]	0.71

Note: CI = confidence interval; IAT = implicit association test.

Table 2.

Means and Effect Sizes for All Outcome Measures in the Three Task Conditions of Experiment 2

	Mannequin task(n = 157)			Avatar-consequences task(n = 146)			Goal-relevant avatar-consequences task(n = 152)
Measure	M (SD)	95% CI	d_z	M (SD)	95% CI	d_z	M (SD)	95% CI	d_z
Consumer choice	0.24 (0.58)	[0.15, 0.33]	0.42	0.47 (0.62)	[0.36, 0.57]	0.75	0.63 (0.51)	[0.54, 0.71]	1.22
Explicit rating	1.75 (2.38)	[1.29, 2.22]	0.74	3.49 (3.68)	[3.01, 3.97]	0.95	5.60 (2.71)	[5.13, 6.07]	2.07
IAT score	0.28 (0.38)	[0.22, 0.33]	0.72	0.23 (0.44)	[0.16, 0.29]	0.52	0.40 (0.37)	[0.34, 0.46]	1.07
Approach-avoidance	1.53 (2.61)	[1.13, 1.93]	0.59	2.58 (3.09)	[2.16, 2.99]	0.83	4.09 (1.77)	[3.69, 4.50]	2.32

Note: CI = confidence interval; IAT = implicit association test.

Implicit and explicit evaluations

IAT scores indicated an implicit preference for the positive over the negative brand in all task conditions of both experiments, ps < .001, d_zs > 0.42. Analyses of variance (ANOVAs) revealed a main effect of approach-avoidance task condition in Experiment 2, F(2, 443) = 9.02, p < .001, but not in Experiment 1, F(2, 507) = 2.07, p = .13. In Experiment 2, IAT scores were higher in the goal-relevant avatar-consequences condition than in the mannequin condition, t(307) = 2.76, p = .006, or avatar-consequences condition, t(296) = 3.64, p < .001. IAT scores did not differ significantly between the latter two conditions, t(246) = 1.12, p = .27.

Explicit ratings indicated a preference for the positive over the negative brand in all task conditions of both experiments, ps < .001, d_zs > 0.74. ANOVAs revealed main effects of approach-avoidance task condition in both Experiment 1, F(2, 513) = 11.61, p < .001, and Experiment 2, F(2, 449) = 65.89, p < .001. In Experiment 1, scores were higher in the avatar-consequences condition than in the mannequin or avatar condition, t(346) = 4.46, p < .001, and t(344) = 3.79, p < .001, respectively. Experiment 2 also found higher scores in the avatar-consequences condition than in the mannequin condition, t(246) = 4.85, p < .001. Furthermore, scores were higher in the goal-relevant avatar-consequences condition than in either of the other two conditions, t(307) = 13.27, p < .001, and t(296) = 5.64, p < .001, respectively.

Approach-avoidance behavior

Participants more often approached the positive than the negative brand in all task conditions of both experiments, ps < .001, d_zs > 0.54. ANOVAs on approach-avoidance behavior scores revealed main effects of approach-avoidance task condition for both Experiment 1, F(2, 513) = 3.02, p = .049, and Experiment 2, F(2, 449) = 40.33, p < .001. In both experiments, scores were higher in the avatar-consequences condition than in the mannequin or avatar condition—Experiment 1: avatar-consequences vs. mannequin condition: t(346) = 2.34, p = .020, avatar-consequences vs. avatar condition: t(344) = 1.82, p = .070; Experiment 2: avatar-consequences vs. mannequin condition: t(301) = 3.58, p < .001. In Experiment 2, scores were also higher in the goal-relevant avatar-consequences condition than in both other conditions, t(307) = 8.86, p < .001, and t(296) = 5.15, p < .001, respectively.

Results of Experiments 3 and 4

Consumer choices

Participants in all task conditions selected healthy foods more often than unhealthy foods, ps < .001, d_zs > 0.35, except for the control condition in Experiment 4 (see Table 3). Behavioral-choice scores were subjected to an ANOVA with approach-avoidance task condition and choice (trade-off, control) as between-subjects factors. For this and all subsequent analyses, we included prerated health behavior, health intention, health behavior difficulty, and hunger as covariates if they significantly improved model fit. We observed main effects of choice—Experiment 4: F(1, 182) = 16.32, p < .001—health behavior—Experiment 3: F(1, 382) = 8.70, p = .003; Experiment 4: F(1, 179) = 5.31, p = .022—health intention—Experiment 3: F(1, 382) = 46.54, p < .001; Experiment 4: F(1, 179) = 52.39, p < .001—and hunger—Experiment 3: F(1, 382) = 6.40, p = .012; Experiment 4: F(1, 179) = 5.31, p = .022—and, crucially, also a main effect of approach-avoidance task condition—Experiment 3: F(2, 382) = 11.94, p < .001; Experiment 4: F(1, 179) = 8.91, p = .003. In both experiments, behavioral-choice scores were higher for the goal-relevant avatar-consequences condition compared with the control condition—Experiment 3: t(257) = 2.71, p = .007; Experiment 4: t(182) = 2.59, p = .005—and mannequin condition—Experiment 3: t(254) = 4.55, p < .001.

Table 3.

Means and Effect Sizes for All Phase 1 Outcome Measures in the Three Task Conditions of Experiments 3 and 4

Measure and experiment	Mannequin task (Experiment 3: n = 133)			Control task (Experiment 3: n = 130; Experiment 4: n = 92)			Goal-relevant avatar-consequences task(Experiment 3: n = 126; Experiment 4: n = 92)
Measure and experiment	M (SD)	95% CI	d_z	M (SD)	95% CI	d_z	M (SD)	95% CI	d_z
Consumer choice
Experiment 3	0.47 (0.83)	[0.33, 0.60]	0.57	0.28 (0.80)	[0.15, 0.42]	0.36	0.75 (0.84)	[0.60, 0.90]	0.90
Experiment 4				0.09 (0.68)	[−0.05, 0.23]	0.14	0.38 (0.80)	[0.21, 0.54]	0.47
Explicit rating
Experiment 3	2.82 (2.53)	[2.42, 3.21]	1.08	2.57 (2.63)	[2.17, 2.96]	0.98	3.27 (2.71)	[2.87, 3.68]	1.23
Experiment 4				4.15 (2.09)	[3.78, 4.53]	1.95	4.22 (2.15)	[3.84, 4.59]	1.99
IAT score
Experiment 3	0.80 (0.34)	[0.74, 0.85]	2.38	0.78 (0.37)	[0.72, 0.84]	2.11	0.90 (0.35)	[0.85, 0.96]	2.58
Experiment 4				0.83 (0.37)	[0.77, 0.89]	2.12	0.92 (0.25)	[0.86, 0.99]	3.50
Approach-avoidance
Experiment 3	2.56 (2.70)	[2.13, 2.99]	0.92	1.86 (2.82)	[1.43, 2.29]	0.67	2.84 (2.52)	[2.41, 3.28]	1.16
Snack eating
Experiment 4				46.7 (33.5)	[40.8, 52.1]		36.5 (25.2)	[31.2, 42.4]

Note: CI = confidence interval; IAT = implicit association test.

Implicit and explicit evaluations

Participants in all task conditions exhibited an implicit preference for healthy foods over unhealthy foods, ps < .001, d_zs > 1.95. The ANOVA on IAT scores revealed main effects of health intention, F(1, 180) = 28.42, p < .001, and IAT block order, F(1, 180) = 9.40, p = .003, and, crucially, also a main effect of approach-avoidance task condition—Experiment 3: F(2, 379) = 3.83, p = .023; Experiment 4: F(1, 180) = 4.96, p = .027. In both experiments, IAT scores were higher in the goal-relevant avatar-consequences condition than in the control and mannequin conditions—Experiment 3: t(257) = 2.16, p = .031; Experiment 4: t(254) = 2.58, p = .010. Scores did not differ significantly between the mannequin and control conditions (Experiment 3), t(261) = 0.41, p = .68.

Explicit ratings indicated an explicit preference for healthy foods over unhealthy foods in all task conditions, ps < .001, d_zs > 0.98. The ANOVA on explicit-rating scores revealed main effects of health intention—Experiment 3: F(1, 382) = 29.77; Experiment 4: F(1, 180) = 18.80, p < .001—health behavior—Experiment 3: F(1, 382) = 5.84, p = .016; Experiment 4: F(1, 180) = 8.10, p = .005—and hunger—Experiment 3: F(1, 382) = 7.17, p = .008. We also observed a main effect of approach-avoidance task condition in Experiment 3, F(2, 382) = 3.08, p = .047, but not in Experiment 4, F(1, 180) = 0.06, p = .81. In Experiment 3, scores were higher in the goal-relevant avatar-consequences condition compared with the control condition, t(254) = 2.45, p = .015, but not compared with the mannequin condition, t(257) = 1.57, p = .12, and scores did not differ significantly between the mannequin and control conditions, t(261) = 0.89, p = .38.

Approach-avoidance behavior (Experiment 3)

Participants in all task conditions approached more healthy than unhealthy foods, ps < .001, d_zs > 0.67. The ANOVA on approach-avoidance behavior scores revealed main effects of health behavior, F(1, 382) = 6.19, p = .013, and health intention, F(1, 382) = 10.84, p = .001, as well as a main effect of approach-avoidance task condition, F(2, 382) = 5.25, p = .006. Compared with scores in the control condition, scores were higher for both the goal-relevant avatar-consequences condition, t(261) = 2.24, p = .026, and the mannequin condition, t(254) = 3.14, p = .002. Scores did not differ significantly between goal-relevant consequences and the mannequin condition, t(257) = 0.91, p = .36.

Snack eating (Experiment 4)

Mean snack intake (in grams) was subjected to an ANOVA with approach-avoidance task condition as the between-subjects factor and prerated health intention, and hunger as covariates. We observed main effects of health intention, F(1, 180) = 27.08, p < .001, and hunger, F(1, 180) = 5.30, p = .022, and, crucially, also a main effect of approach-avoidance task condition, F(1, 180) = 5.67, p = .018. Participants’ snack intake was lower in the goal-relevant avatar-consequences condition compared with the control condition.

Phase 2 questions (Experiment 3)

A multivariate ANOVA on Phase 2 ratings revealed effects of health behavior, F(5, 297) = 21.11, p < .001; health behavior difficulty, F(5, 297) = 14.83, p < .001; and health intention, F(5, 297) = 29.07, p < .001, and, most importantly, also a main effect of approach-avoidance task condition, F(10, 301) = 4.46, p < .001 (see Table 4). Follow-up analyses that included health behavior, health behavior difficulty, and health intention as covariates revealed a significant effect of approach-avoidance task condition for ratings of unhealthy eating behavior, F(2, 301) = 3.63, p = .028, and ratings of healthy eating intentions, F(2, 301) = 3.31, p = .038, and a marginally significant effect for ratings of healthy eating goals, F(2, 301) = 2.76, p = .065. Participants in the goal-relevant avatar-consequences condition provided lower ratings for unhealthy eating behavior compared with participants in the control, t(206) = 2.23, p = .026, and mannequin conditions, t(197) = 2.43, p = .016, and higher ratings for healthy eating intentions and healthy eating goals compared with participants in the control condition—intentions: t(197) = 2.55, p = .011; goals: t(197) = 2.29, p = .023. Ratings did not differ significantly between the mannequin and control conditions—unhealthy eating behavior: t(205) = 0.20, p = .85; healthy eating intentions: t(205) = 0.98, p = .32; healthy eating goals: t(205) = 0.70, p = .48.

Table 4.

Mean Ratings for the Phase 2 Questions in the Three Task Conditions of Experiment 3

	Mannequin task(n = 108)		Control task(n = 99)		Goal-relevant avatar-consequences task(n = 100)
Item	M (SD)	95% CI	M (SD)	95% CI	M (SD)	95% CI
Healthy eating behavior	5.24 (1.93)	[4.93, 5.54]	5.28 (1.88)	[4.97, 5.60]	5.53 (2.02)	[5.21, 5.84]
Unhealthy eating behavior	4.68 (1.89)	[4.37, 4.98]	4.63 (1.78)	[4.32, 4.94]	4.13 (1.79)	[3.82, 4.45]
Healthy eating intention	6.02 (2.29)	[5.68, 6.35]	5.78 (2.28)	[5.43, 6.12]	6.41 (2.07)	[6.06, 6.76]
Difficulty of stopping unhealthy eating	4.88 (2.27)	[4.52, 5.23]	4.80 (2.21)	[4.43, 5.16]	4.98 (2.28)	[4.62, 5.35]
Healthy eating goal	6.67 (1.97)	[6.42, 6.92]	6.54 (1.96)	[6.28, 6.80]	6.96 (1.56)	[6.71, 7.22]

Note: CI = confidence interval.

Discussion

Four experiments examined the effects of a novel type of approach-avoidance training in which approach and avoidance responses to food products were consistently followed by positive or negative consequences. In Experiment 1, consequence-based approach-avoidance training had a bigger impact on consumer choices, approach-avoidance behavior, and explicit (but not implicit) evaluation of novel food brands than typical approach-avoidance training. Experiment 2 replicated these findings and showed that goal-relevant action consequences enhanced the effects of consequence-based approach-avoidance training (also on implicit evaluation). Experiment 3 found that consequence-based approach-avoidance training also produced bigger effects than typical approach-avoidance training in the context of healthy and unhealthy foods. Moreover, compared with a control training, consequence-based approach-avoidance training (but not typical approach-avoidance training) reduced self-reported unhealthy eating behaviors and increased healthy eating intentions 24 hr after training. Experiment 4 further showed that consequence-based approach-avoidance training reduces actual unhealthy eating as measured in a snack task.

The fact that consequence-based approach-avoidance training produced robust effects on consumer choices, approach-avoidance behavior, and implicit and explicit evaluations is important given the inconsistent effects that typical approach-avoidance training often produces (Cristea et al., 2015, 2016), especially in the context of food products (D. Becker et al., 2015). Although we also observed clear effects of typical approach-avoidance training (mainly in the context of novel foods), the effects of consequence-based approach-avoidance training were consistently stronger. Importantly, consequence-based approach-avoidance training also seemed to reduce actual unwanted behavior such as the consumption of unhealthy foods (i.e., participants reported less unhealthy eating behavior in the consequence-based approach-avoidance training condition of Experiment 3 and consumed less unhealthy foods in the snack task of Experiment 4).

The current findings fit with an inferential account of approach-avoidance training effects that asserted that these effects result from inferences related to goal-directed action (Van Dessel, Hughes, & De Houwer, 2018). In consequence-based approach-avoidance training, participants directly learn about the consequences of approach-avoidance responses to certain (food) stimuli, and this may have caused them to anticipate similar outcomes for similar actions (e.g., actual unhealthy eating). Responses for which positive outcomes are anticipated generally have a higher value and are therefore facilitated in comparison with responses for which negative outcomes are anticipated (Eder, Rothermund, De Houwer, & Hommel, 2015). Because inferential learning is assumed to be determined by activated goals, approach-avoidance training effects are further enhanced when action consequences are goal relevant. Typical approach-avoidance training also allows participants to learn about the consequences of approach-avoidance responses. For instance, avoiding unhealthy foods might facilitate retrieval of information consistent with the idea that unhealthy foods are to be avoided (e.g., negative consequences of unhealthy eating). Yet unlike consequence-based approach-avoidance training, typical approach-avoidance training does not specify or require that action consequences are learned, which might explain why the effects of typical approach-avoidance training were smaller than the effects of consequence-based approach-avoidance training.

Our results contrast with dominant accounts of approach-avoidance training effects that assume that experienced contingencies between stimuli and approach-avoidance responses cause an automatic rewiring of cognitive biases based on mental stimulus–response associations that mediate these effects (e.g., Wiers et al., 2011; Wiers, Gladwin, Hofmann, Salemink, & Ridderinkhof, 2013). Because typical approach-avoidance training involves stronger stimulus–response contingencies than consequence-based approach-avoidance training, these accounts predict stronger changes in cognitive biases and resulting effects for typical compared with consequence-based approach-avoidance training. Of course, the current results do not preclude the possibility that noninferential (e.g., associative) mechanisms contribute to approach-avoidance training effects. Some have argued that only automatic (e.g., unintentional) approach-avoidance training effects depend on associative mechanisms (e.g., Kawakami, Phills, Steele, & Dovidio, 2007). However, automaticity should not be conflated with underlying processes (e.g., inferential reasoning can produce automatic effects). In fact, our inferential account assumes that approach-avoidance training involves an important automatization component to the extent that the repeated nature of the task facilitates more automatic inferences. Moreover, there was no indication that our consequence-based approach-avoidance training effects are more likely to be based on controlled processes than typical approach-avoidance training effects. For instance, although it is possible that demand characteristics biased approach-avoidance training effects (see Sharpe & Whelton, 2016), we found that typical approach-avoidance training effects correlated more strongly with demand compliance ratings than consequence-based approach-avoidance training effects (see the Supplemental Material). Moreover, Experiment 4 showed approach-avoidance training effects (e.g., on IAT scores) in the absence of effects on more controlled, explicit liking ratings.

Note that caution is still warranted when interpreting the current findings. First, it is possible that other factors than learned consequences might explain observed dissociations between typical and consequence-based approach-avoidance training effects. For instance, general task attention or task engagement might be enhanced in consequence-based approach-avoidance training tasks, which could strengthen effects. However, this explanation does not fit with the observation that (a) participants’ overall performance on the approach-avoidance training task was better in the typical approach-avoidance training-task conditions and (b) self-reported task engagement did not mediate approach-avoidance training effects (see the Supplemental Material). Second, effects on actual (unwanted) behavior were established only in the snack-eating task of Experiment 4 (compared with a control condition). However, the fact that four high-powered experiments showed strong effects of consequence-based approach-avoidance training on many outcomes (including self-reported unhealthy eating after a 1-day delay) does lead us to believe that consequence-based approach-avoidance training has practical use.

We therefore hope that future research will examine whether beneficial results of consequence-based approach-avoidance training can also be obtained in clinical samples and in other clinical domains (e.g., alcohol consumption, depression). Such studies could also examine moderators of consequence-based approach-avoidance training effects. We already found evidence that goal relevance of action consequences might be one important moderator, as predicted by our inferential account (Van Dessel, Hughes, & De Houwer, 2018). However, this account also predicts that other task adaptations may further improve effects (e.g., including a higher number of training trials per sessions, using more lifelike or personally relevant affective consequences, or training across multiple contexts). Our results also open the door for designing interventions that aim to facilitate adaptive inferences (inference training) on the basis of actions other than approach and avoidance. This approach bears resemblance to current “nudging” interventions that aim to modify behavior by providing subtle environmental cues (Benartzi et al., 2017) and to effective therapies used in clinical practice that target beliefs underlying maladaptive behavior (i.e., cognitive behavior therapy; Beck & Dozios, 2011). However, the fact that participants need to derive new information themselves and repeatedly act on it is different from current treatments and might facilitate automatic effects (for evidence in the context of approach-avoidance training, see Wiers et al., 2011). Inference training (via approach-avoidance training) might also be easier to distribute (e.g., in mobile apps) and to incorporate into existing initiatives (e.g., health-promotion interventions).

Supplemental Material

VanDesselOpenPracticesDisclosure – Supplemental material for Consequence-Based Approach-Avoidance Training: A New and Improved Method for Changing Behavior

Supplemental material, VanDesselOpenPracticesDisclosure for Consequence-Based Approach-Avoidance Training: A New and Improved Method for Changing Behavior by Pieter Van Dessel, Sean Hughes and Jan De Houwer in Psychological Science

Supplemental Material

VanDesselSupplementalMaterial – Supplemental material for Consequence-Based Approach-Avoidance Training: A New and Improved Method for Changing Behavior

Supplemental material, VanDesselSupplementalMaterial for Consequence-Based Approach-Avoidance Training: A New and Improved Method for Changing Behavior by Pieter Van Dessel, Sean Hughes and Jan De Houwer in Psychological Science

Footnotes

Action Editor

Ayse K. Uskul served as action editor for this article.

Author Contributions

All the authors were involved in developing the study concept and contributed to the design. P. Van Dessel collected and analyzed the data and drafted the manuscript. J. De Houwer and S. Hughes provided critical revisions. All the authors approved the final manuscript for submission.

ORCID iD

Pieter Van Dessel

Declaration of Conflicting Interests

The author(s) declared that there were no conflicts of interest with respect to the authorship or the publication of this article.

Funding

P. Van Dessel was supported by a postdoctoral fellowship from the Scientific Research Foundation, Flanders (FWO-Vlaanderen). J. De Houwer was supported by Methusalem Grant BOF16/MET_V/002 from Ghent University.

Supplemental Material

Additional supporting information can be found at

Open Practices

All data and materials have been made publicly available via the Open Science Framework and can be accessed at https://osf.io/3anqx/. The design and analysis plans for Experiments 1 through 4 are also available at https://osf.io/3anqx/. The complete Open Practices Disclosure for this article can be found at https://journals-sagepub-com.web.bisu.edu.cn/doi/suppl/10.1177/0956797618796478. This article has received the badges for Open Data, Open Materials, and Preregistration. More information about the Open Practices badges can be found at .

References

Beck

A. T.

Dozios

D. J. A.

(2011). Cognitive therapy: Current status and future directions. Annual Review of Medicine, 62, 397–409. doi:10.1146/annurevmed-052209-100032

Becker

Jostmann

N. B.

Wiers

R. W.

Holland

R. W.

(2015). Approach avoidance training in the eating domain: Testing the effectiveness across three single session studies. Appetite, 85, 58–65. doi:10.1016/j.appet.2014.11.017

Becker

E. S.

Ferentzi

Ferrari

Möbius

Brugman

Custers

. . . Rinck

(2016). Always approach the bright side of life: A general positivity training reduces stress reactions in vulnerable individuals. Cognitive Therapy and Research, 40, 57–71.

Benartzi

Beshears

Milkman

K. L.

Sunstein

C. R.

Thaler

R. H.

Shankar

. . . Galing

(2017). Should governments invest more in nudging? Psychological Science, 28, 1041–1055.

Chen

Bargh

J. A.

(1999). Consequences of automatic evaluation: Immediate behavioral predispositions to approach or avoid the stimulus. Personality and Social Psychology Bulletin, 25, 215–224. doi:10.1177/0146167299025002007

Cristea

I. A.

Kok

R. N.

Cuijpers

(2015). Efficacy of cognitive bias modification interventions in anxiety and depression: Meta-analysis. British Journal of Psychiatry, 206, 7–16.

Cristea

I. A.

Kok

R. N.

Cuijpers

(2016). The effectiveness of cognitive bias modification interventions for substance addictions: A meta-analysis. PLOS ONE, 11(9), Article e0162226. doi:10.1371/journal.pone.0162226

Eder

A. B.

Hommel

(2013). Anticipatory control of approach and avoidance: An ideomotor approach. Emotion Review, 5, 275–279.

Eder

A. B.

Rothermund

De Houwer

Hommel

(2015). Directive and incentive functions of affective action consequences: An ideomotor approach. Psychological Research, 79, 630–649.

10.

Greenwald

A. G.

McGhee

D. E.

Schwartz

J. L.

(1998). Measuring individual differences in implicit cognition: The implicit association test. Journal of Personality and Social Psychology, 74, 1464–1480. doi:10.1037/0022-3514.74.6.1464

11.

Han

H. A.

Olson

M. A.

Fazio

R. H.

(2006). The influence of experimentally created extrapersonal associations on the implicit association test. Journal of Experimental Social Psychology, 42, 259–272.

12.

Haynes

Kemps

Moffitt

(2015). Inhibitory self-control moderates the effect of changed implicit food evaluations on snack food consumption. Appetite, 90, 114–122.

13.

Helton

W. S.

Näswall

(2015). Short Stress State Questionnaire: Factor structure and state change assessment. European Journal of Psychological Assessment, 31(1), 20–30. doi:10.1027/1015-5759/a000200

14.

Hofmann

Gschwendner

Friese

Wiers

R. W.

Schmitt

(2008). Working memory capacity and self-regulatory behavior: Toward an individual differences perspective on behavior determination by automatic versus controlled processes. Journal of Personality and Social Psychology, 95, 962–977. doi:10.1037/a0012705

15.

Jones

Hardman

C. A.

Lawrence

Field

(2017). Cognitive training as a potential treatment for overweight and obesity: A critical review of the evidence. Appetite, 124, 50–67. doi:10.1016/j.appet.2017.05.032

16.

Jones

C. R.

Vilensky

M. R.

Vasey

M. W.

Fazio

R. H.

(2013). Approach behavior can mitigate predominately univalent negative attitudes: Evidence regarding insects and spiders. Emotion, 135, 989–996. doi:10.1016/S0896-6273(02)00858-9

17.

Kakoschke

Kemps

Tiggemann

(2017). Approach bias modification training and consumption: A review of the literature. Addictive Behaviors, 64, 21–28.

18.

Kawakami

Phills

C. E.

Steele

J. R.

Dovidio

J. F.

(2007). (Close) distance makes the heart grow fonder: Improving implicit racial evaluations and interracial interactions through approach behaviors. Journal of Personality and Social Psychology, 92, 957–971.

19.

Lang

P. J.

Bradley

M. M.

Cuthbert

B. N.

(1990). Emotion, attention, and the startle reflex. Psychological Review, 97, 377–395. doi:10.1037/0033-295X.97.3.377

20.

Nosek

B. A.

Greenwald

A. G.

Banaji

M. R.

(2005). Understanding and using the implicit association test: II. Method variables and construct validity. Personality and Social Psychology Bulletin, 31, 166–180.

21.

Rotteveel

Phaf

R. H.

(2004). Automatic affective evaluation does not automatically predispose for arm flexion and extension. Emotion, 4, 156–172.

22.

Sharpe

Whelton

W. J.

(2016). Frightened by an old scarecrow: The remarkable resilience of demand characteristics. Review of General Psychology, 20, 349–368.

23.

Solarz

A. K.

(1960). Latency of instrumental responses as a function of compatibility with the meaning of eliciting verbal signs. Journal of Experimental Psychology, 59, 239–245. doi:10.1037/h0047274

24.

Spruyt

De Houwer

Tibboel

Verschuere

Crombez

Verbanck

. . . Noël

(2013). On the predictive validity of automatically activated approach/avoidance tendencies in abstaining alcohol-dependent patients. Drug and Alcohol Dependence, 127, 81–86. doi:10.1016/j.drugalcdep.2012.06.019

25.

Strack

Deutsch

(2004). Reflective and impulsive determinants of social behavior. Personality and Social Psychology Review, 8, 220–247. doi:10.1207/s15327957pspr0803_1

26.

Taylor

C. T.

Amir

(2012). Modifying automatic approach action tendencies in individuals with elevated social anxiety symptoms. Behaviour Research and Therapy, 50, 529–536. doi:10.1016/j.brat.2012.05.004

27.

Thorndike

E. L.

(1905). The elements of psychology. New York, NY: A. G. Seiler.

28.

Van Dessel

Eder

A. B.

Hughes

. (2018). Mechanisms underlying effects of approach-avoidance training on stimulus evaluation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 44, 1224–1241.

29.

Van Dessel

Hughes

De Houwer

. (2018). How do actions influence attitudes? An inferential account of the impact of action performance on stimulus evaluation. Manuscript submitted for publication.

30.

Wiers

R. W.

Eberl

Rinck

Becker

Lindenmeyer

(2011). Retraining automatic action tendencies changes alcoholic patients’ approach bias for alcohol and improves treatment outcome. Psychological Science, 22, 490–497. doi:10.1177/0956797611400615

31.

Wiers

R. W.

Gladwin

T. E.

Hofmann

Salemink

Ridderinkhof

K. R.

(2013). Cognitive bias modification and cognitive control training in addiction and related psychopathology: Mechanisms, clinical perspectives, and ways forward. Clinical Psychological Science, 1, 192–212.

32.

Wittekind

C. E.

Feist

Schneider

B. C.

Moritz

Fritzsche

(2015). The approach-avoidance task as an online intervention in cigarette smoking: A pilot study. Journal of Behavior Therapy and Experimental Psychiatry, 46, 115–120. doi:10.1016/j.jbtep.2014.08.006

33.

Woud

M. L.

Maas

Becker

E. S.

Rinck

(2013). Make the manikin move: Symbolic approach–avoidance responses affect implicit and explicit face evaluations. Journal of Cognitive Psychology, 25, 738–744. doi:10.1080/20445911.2013.817413

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.16 MB

0.51 MB