Introduction
When confronting a threatening object or situation, organisms often engage in safety behavior that aims to prevent the occurrence of a threatening outcome. In this regard, safety behavior is adaptive given that it effectively prevents harm. However, safety behavior in anxiety-related disorders is oftentimes maladaptive as it is out of proportion to realistic threat, persistent in the absence of threat, and linked to impairments in everyday functioning (Mendlowicz & Stein,
2000; Olatunji, Cisler, & Tolin,
2007). For instance, an individual with social anxiety disorder may actively avoid pauses when conversing with others to avoid looking anxious, thereby reducing the perceived threat of being negatively evaluated (Kim,
2005).
Fear and avoidance conditioning provide a well-established laboratory model for examining fear-related safety behavior, which usually combines both Pavlovian fear acquisition and instrumental learning. During Pavlovian fear acquisition, an initially neutral conditioned stimulus (CS+) is repeatedly paired with an aversive unconditioned stimulus (US). In a subsequent instrumental learning phase, participants learn that performing a designated response during CS+ presentation prevents the US. This response is referred to as “US-avoidance” given that it prevents the US but does not terminate CS+ presentation (see Pittig, Wong, Glück, & Boschet,
2020). Performing US-avoidance was found to reduce conditioned fear responses to the CS+ (e.g., Lovibond, Saunders, Weidemann, & Mitchell,
2008; Morriss, Chapman, Tomlinson, & van Reekum,
2018; Pittig,
2019), analogous to the reduction in anxiety after engaging in safety behavior when confronting fear-related stimuli or situations. This reduction in anxiety aligns with a propositional Expectancy model (Lovibond,
2006). One prediction this model makes is a decrease in conditioned fear responses to the CS+ once US-avoidance is performed, since the US is not anticipated anymore (cf. Two-factor theory; Mowrer,
1939).
Excessive engagement with safety behavior has been found to preserve maladaptive threat beliefs. Empirical studies found that engagement in US-avoidance during extinction led to heightened fear responses to the CS+ when avoidance availability was removed (Lovibond, Mitchell, Minard, Brady, & Menzies,
2009; Pittig,
2019; Rattel, Miedl, Blechert, & Wilhelm,
2017; Volders, Meulders, Peuter, Vervliet, & Vlaeyen,
2012; Volders, Meulders, Peuter, & Vlaeyen,
2015). This pattern was referred to as ‘protection from extinction’: threat belief to the CS+ was intact, because the absence of an US during extinction was attributed to US-avoidance, therefore, preventing extinction learning to take place. The findings of these studies aligned with clinical observations in which individuals with clinical anxiety excessively engage in safety behaviors showed resistance to exposure therapy or relapse after apparent successful treatment (Helbig-Lang & Petermann,
2010; Salkovskis, Clark, & Gelder,
1996; Wells et al.,
1995).
Although the aforementioned studies provided valuable insights into the mechanisms of avoidance and safety behavior, traditional paradigms have recently been criticized for examining low-cost safety behavior (Krypotos, Vervliet & Engelhard,
2018). Low-cost safety behavior refers to US-avoidance that requires minimal cost or effort (e.g., merely pressing a button). Low-cost US-avoidance arguably does not resemble pathological safety behavior in anxiety-related disorders: safety behavior in anxiety-related disorders often bears a cost. In light of this, more recent laboratory studies have incorporated a competing reward to safety behavior (e.g., Claes, Karos, Meulders, Crombez, & Vlaeyen,
2014; Pittig,
2019; Rattel et al.,
2017; van Damme, van Ryckeghem, Wyffels, van Hulle, & Crombez,
2012). That is, an appetitive outcome that motivates behavioral approach to the fear-related stimulus (i.e., approach-avoidance conflict), rendering avoidance costly (i.e., missing out the competing reward). A common finding from these studies is that healthy individuals showed less US-avoidance with the presence of a competing reward. Furthermore, preliminary studies showed that a decrease in costly US-avoidance did not reduce conditioned fear to the CS+; instead, the competing reward per se acts as an incentive for behavioral approach (fear-opposite actions; Pittig & Dehler,
2019). This incentive for non-avoidance response has been found to facilitate extinction learning (Pittig,
2019; Rattel et al.,
2017) and subsequently alleviated the effect of protection from extinction (Pittig,
2019). Collectively, these findings suggest that incorporating a cost in US-avoidance may provide a useful testbed for investigating pathological safety behavior.
Another criticism refers to safety behavior commonly investigated in a dichotomous manner. That is, it is either performed or not performed. Little research has focused on examining the magnitude or the extent of engagement in safety behavior. It has been suggested that patients sometimes only engage in safety behavior to a certain degree (see Krypotos et al.,
2018; Telch & Lancaster,
2012). For example, an individual with social anxiety may rather converse to a lesser extent in a group discussion. While this safety behavior is believed to prevent the perceived threatening outcome (e.g., appear unintelligent in the conversation) to a certain extent, the individual could still contribute to the group discussion to some extent (obtaining the competing reward; Kashdan, Elhai, & Breen,
2008). In this regard, safety behavior is oftentimes not a dichotomous behavior, but can be seen as a balance of keeping threat at a subjectively acceptable level while limiting its costs (cf. Schlund et al.,
2016). A dimensional measure of safety behavior is thus arguably more sensitive to measure the different degrees of safety behavior. Furthermore, a dichotomous measure usually results in a ceiling effect with little variability in responding (i.e., most participants fully engaging in safety behavior). The lack of variability decreases the sensitivity to examine individual differences or risk factors modulating the acquisition of safety behavior, such as trait anxiety, intolerance of uncertainty, and anxiety sensitivity (see Lonsdorf & Merz,
2017; Pittig et al.,
2020). A dimensional measure may also provide important theoretical insights. For example, it may be a more sensitive test for predictions of the Expectancy model (Lovibond,
2006). Specifically, the model assumes that a higher degree of safety behavior predicts lower conditioned fear to the CS+, or that the degree of US-avoidance during extinction predicts the magnitude of protection from extinction. In sum, a dimensional measure of US-avoidance potentially provides some advantages over the traditional dichotomous measure.
Non-fear conditioning studies have already entertained the idea of measuring defensive behaviors (behaviors that entail escape or avoidance, see Krypotos et al.,
2018) dimensionally. For instance, the “Pac-man” task (Mobbs et al.,
2007,
2009) and the foraging task (Bach et al.,
2014) operationalized defensive behaviors as the distance between an individual’s avatar and a virtual predator. Behavioural Approach Task serves as another paradigm to measure avoidance dimensionally (e.g., Shiban, Pauli & Mühlberger,
2013). In contrast, little research has measured avoidance dimensionally, especially US-avoidance, within a fear conditioning framework. Flores, Lopez, Vervliet and Cobos (
2018) pioneered a novel continuous measure of US-avoidance in a fear conditioning paradigm. Participants were presented with a CS+ of 20 s, in which an US was randomly delivered between 8 and 11 s after CS+ onset. Participants were informed that pressing a designated key during CS+ presentation may prevent the US, but only keypresses within 1 s before US onset could effectively prevent the US. Given the uncertain timing of US administration, participants were encouraged to press the designated key as many times as they wanted during CS+ presentation. Therefore, this study elegantly examined the magnitude of safety behavior as a function of the level of US uncertainty. However, Flores et al. (
2018) did not incorporate a competing reward for US-avoidance, therefore, may not fully tap into the pathological domain of safety behavior. Furthermore, this novel procedure had no room for the measure of conditioned fear to the CS+ after safety behavior had been performed.
In two experiments, we, therefore, developed a dimensional measure of US-avoidance that catered the inclusion of a competing reward and the measurement of conditioned fear after US-avoidance responses, within a fear conditioning framework. We measured US-avoidance using a visual analog scale (0% = complete non-avoidance, 100% = complete avoidance), in which the US-avoidance was negatively proportional to US administration and the competing reward. Therefore, we could examine the role of competing reward in the degree of US-avoidance engagement. Immediately after US-avoidance was made, self-reported US expectancy ratings and skin conductance responses (SCRs) to the CS were measured to reflect the cognitive and physiological aspects of conditioned fear, respectively. This study had two overarching goals: using a dimensional US-avoidance measure to (1) replicate findings on the dynamics between US-avoidance, conditioned fear, and competing reward and (2) testing predictions of the Expectancy model. For the first goal, we aimed to replicate findings on (1a) the dynamics between US-avoidance, conditioned fear, and competing reward during the acquisition of US-avoidance. Specifically, engaging in low-cost US-avoidance decreases conditioned fear, however, an introduction of competing reward reduces US-avoidance, resulting in an increase in conditioned fear; (1b) the dynamics between US-avoidance, competing reward, and extinction learning. Specifically, competing reward reduces US-avoidance, thereby enables extinction learning when US administration is discontinued; and (1c) competing reward reduces protection from extinction. For the second goal, we assumed that a dimensional measure of US-avoidance is more sensitive to predict subsequent levels of conditioned fear. More specifically, the Expectancy model predicts (2a) a negative linear relationship between US-avoidance and conditioned fear. Specifically, an increase in US-avoidance predicts a greater decrease in conditioned fear to the CS+ than the CS–; and (2b) a positive linear relationship between US-avoidance and protection from extinction. This means, an increase in US-avoidance during extinction predicts a stronger protection from extinction effect.
General discussion
Across two experiments using a differential fear and avoidance conditioning procedure, we employed a novel dimensional measure of US-avoidance to examine safety behavior. We expected to replicate findings using a dichotomous measure, in which an introduction of a competing reward would, first, reduce US-avoidance to the CS+, accompanied by an increase in conditioned fear; second, this decrease in US-avoidance would initiate extinction learning to the CS+ when US-administration was discontinued, as indexed by a decrease in US expectancy ratings and SCRs to the CS+; third, would lead to an alleviated protection from extinction effect. Furthermore, we expected that a dimensional measure of US-avoidance to be a sensitive measure to predict subsequent levels of conditioned fear, and that the degree of US-avoidance engagement on the last extinction trial predicted the magnitude of the protection from extinction effect.
As predicted, both experiments showed a decrease in US-avoidance once a competing reward was introduced. This finding is consistent with studies that showed a competing reward acting as an incentive for non-avoidance (e.g., Claes et al.,
2014; Pittig,
2019; Rattel et al.,
2017; van Damme et al.,
2012). Furthermore, the low degree of US-avoidance to the CS+ in the Reward group was accompanied by a high level of US expectancy ratings to the CS+ during the US-avoidance-reward phase. This means that a competing reward encouraged participants to tolerate the US in favor for the reward (i.e., fear-opposite action; Pittig,
2019; Pittig & Dehler,
2019). Second, across both experiments, the Reward group continued to show lower degree of US-avoidance to the CS+ than the Control group throughout the Pavlovian extinction phase. This led to a high level of US expectancies to the CS+ on early extinction trials, which, however, gradually decreased, indicating extinction learning. This suggested that a low degree of US-avoidance induced by a competing reward enabled extinction learning. In contrast, the Control group continued to show a high degree of US-avoidance to the CS+ throughout Pavlovian extinction, accompanied by low levels of US expectancies, suggesting that participants attributed US omission to their engagement in US-avoidance. This means the similarly low levels of US expectancy between groups at the end of the Pavlovian extinction phase were attributed to different causes: In the Reward group, this was mainly attributed to extinction learning to the CS+ , whereas in the Control group, this was mainly attributed to the high degree of US-avoidance. In sum, this finding aligned with studies that showed excessive engagement in safety behavior preventing extinction learning (Lovibond et al.,
2009; Pittig,
2019; Rattel et al.,
2017).
Contrary to our third prediction, both groups showed a similar level of protection from extinction effect in test. We hypothesized that the fixed amount of competing reward in Experiment 1 may be sufficient to encourage low degree of avoidance to some participants while trivial to some participants, therefore decreasing the effect of competing reward on US-avoidance. In light of this, we matched the magnitude of reward to the US intensity for each individual participant in the Reward group in Experiment
2. However, despite our anticipation that this matching procedure would have increased the sensitivity to detect a reduced protection from extinction effect in the Reward group, we still observed no group differences in protection from extinction.
The current results also demonstrated a dimensional measure of US-avoidance sufficiently predicted subsequent levels of conditioned fear and the magnitude of protection from extinction effect. First, an increase in US-avoidance predicted a stronger decrease in US expectancies and SCRs to the CS+ when compared to the CS− during the phases of US-avoidance acquisition. This pattern aligned with the notion of the Expectancy model that avoidance modulates subsequent conditioned fear responses to the CS+ via a process of mental anticipation of an aversive outcome. This means, when a high degree of US-avoidance to the CS+ was made, the chance of US delivery was anticipated to a lesser extent, leading to low levels of conditioned fear to the CS+ (see Mitchell, Lovibond & De Houwer,
2009). In contrast, US-avoidance to the CS− showed little predictiveness to the subsequent conditioned fear, suggesting that participants knew that the CS– would not be followed by an aversive outcome regardless of US-avoidance.
Despite the null group difference in protection from extinction, we found that an increase in US-avoidance on the last extinction trial significantly predicted stronger conditioned fear to the CS+ than the CS− on the first test trial. This finding suggests that an individual engaging in US-avoidance to a greater extent would show a stronger protection from extinction effect. This is in line with the notion of the Expectancy model, suggesting that the more likely one engage in US-avoidance, the more likely one would attribute the absence of an aversive outcome to the avoidance response, despite the aversive outcome no longer follows the CS+ regardless of avoidance responses. Notably, the dimensional measure of US-avoidance allowed US-avoidance on a single trial to predict the magnitude of protection from extinction.
The unexpected null group difference in protection from extinction were presumably driven by other potential factors. One potential candidate is insufficient expectancy violation. According to the inhibitory learning model of extinction (Craske, Treanor, Conway, Zbozinek, & Vervliet,
2014), treatment outcome of exposure-based therapies depends on to what extent a patient’s threat belief has been challenged throughout therapy sessions (Pittig et al., submitted). That is, a mismatch between threat expectancy and the actual outcome. The larger this violation of expectancy, the more extinction learning takes place and therefore the better the treatment outcome. However, a dimensional measure of US-avoidance allowed participants to not fully disengage from US-avoidance during extinction. This means, violation of US expectancy could be partially attributed to genuine extinction learning and partially attributed to US-avoidance. Therefore, expectancy violation was not maximized during extinction, decreasing any observable reductions in the subsequent protection from extinction effect. Another potential factor in play was the change in context between the Pavlovian extinction and test phase. Extinction learning is vulnerable to a change in the spatio-temporal context, that is, an extinguished fear-related stimulus presented in a context different from the extinction context would trigger a return of fear (Bouton,
2002; Bouton & Bolles,
1979; Bouton & King,
1983). In the current study, US-avoidance was available during Pavlovian extinction, but was removed during test. The removal of US-avoidance might have represented a contextual change, resulting in an increase of conditioned fear to the CS+. This increase of conditioned fear might have reduced any observable group differences in protection from extinction. In support, studies showed that the removal of US-avoidance availability from extinction to test resulted in a return of fear (Vervliet & Indeuku,
2015, Pittig,
2019; Rattel et al.,
2017, but see Lovibond et al.,
2009). A final potential factor in play was the use of a strategy. Some participants in the Reward group might have used a strategy to gain the competing reward via non-avoidance to the CS− only. This means that participants could still obtain monetary gain by not avoiding a safety cue, rendering avoidance of CS+ relatively less costly. This strategy might have artificially increased US-avoidance to the CS+ in the Reward group, reduced extinction learning to the CS+ and therefore potentially contributed to the null group difference in protection from extinction. Future studies may reduce the use of this strategy by reducing the amount of CS– trials or employing a single-cue conditioning paradigm, which the latter has been successfully employed in human fear conditioning (e.g., Baas,
2013; Lee, Hayes, & Lovibond,
2018; Pittig,
2019; Wong & Lovibond,
2017).
Both experiments consistently revealed a novel finding regarding the link between low-cost avoidance and US expectancies to a safety cue. The Control group showed higher degree of US-avoidance to the CS– than the Reward group in both US-avoidance-reward and Pavlovian extinction phases. Importantly, this group difference was not due to the Control group exhibiting stronger conditioned fear to the CS– nor a stronger tendency to generally engage with higher degree of US-avoidance (see Supplementary Materials). This strongly suggests that low-cost US-avoidance does not fully reflect the fear-related component, thereby not accurately reflecting how safety behavior is motivated by fear. In fact, the minimal cost of US-avoidance potentially encouraged participants to engage in a higher degree of US-avoidance to the CS– despite a low level of conditioned fear to it. This discrepancy between conditioned fear and low-cost safety behavior is sometimes referred to as a ‘better safe than sorry’ strategy (e.g., Lommen et al.,
2010). However, a ‘better safe than sorry’ strategy implies that participants engage in safety behavior due to the uncertainty of US occurrence. In the current study, participants showed a steady, low level of US expectancies to the CS- prior to US-avoidance acquisition, suggesting a certain absence of an US. Therefore, a ‘why not’ strategy may more suitably describe the discrepancy between conditioned fear and low-cost safety behavior in the current study (i.e., I know the CS– is safe but avoidance response costs nothing, so why not avoid?).
This novel finding has important methodological implications. We should cautiously interpret whether low-cost avoidance is fully motivated by conditioned fear (cf. Krypotos, Effting, Kindt & Beckers,
2015). Specifically, some studies found a high degree of low-cost avoidance to the CS+ after response prevention extinction (e.g., Krypotos & Engelhard,
2019; Vervliet & Indekeu,
2015; but see Krypotos & Engelhard,
2018), presumably caused by a return of Pavlovian fear. However, the current findings favor an alternative interpretation that the heightened avoidance was encouraged by the minimal cost to perform it rather than reflecting a return of fear. In support to this interpretation, incorporating a cost to safety behavior led to little to no avoidance responses to the CS+ after response prevention extinction (Vervliet, Lange & Milad,
2017).
The group differences in US-avoidance to the CS+ and CS− suggested that different processes may modulate the link between conditioned fear and US-avoidance. For instance, the lower degree of US-avoidance to the CS+ in the Reward group reflected a fear-opposite action (e.g., Pittig,
2019; Pittig & Dehler,
2019) modulated by the opportunity cost of safety behavior. In contrast, the elevated US-avoidance to the CS– in the Control group reflected a discrepancy between conditioned fear and US-avoidance (e.g., “why not” strategy) due to the minimal cost of US-avoidance. These findings suggested that a dimensional measure of US-avoidance was sensitive to reveal these different processes linked to safety behavior, providing a promising method for detecting other contributing factors to the acquisition of safety behavior.
Two findings in the current study have important implications in a clinical context. First, the null group difference in protection from extinction was presumably due to expectancy violation not being maximized, given that participants could still engage in safety behavior to a small extent during extinction. This speculation suggests that even partial engagement in safety behavior during exposure sessions is sufficient for one to attribute the absence of an aversive outcome to safety behavior, thus impeding treatment outcome. Second, the Control group showed stronger safety behavior to the safety cue compared to the Reward group, presumably due to the low-cost of safety behavior. This group difference in US-avoidance to the CS– was presumably due to an excessive engagement in low-cost safety behavior (e.g., “why not” strategy). This calls for the attention towards preventing unnecessary low-cost safety behavior to safety cues or situations in individuals with anxiety-related disorders. Past studies showed that clinical samples tended to infer the presence of potential threat via their safety behavior (Gangemi, Mancini, & van den Hout,
2012; van den Hout et al.,
2014; Engelhard, van Uijen, van Seters, & Velu,
2015). Combined with the current findings, this suggests that excessive engagement in low-cost behavior in individuals with anxiety-related disorders may paradoxically increase their fear to other safety cues or situations. Collectively, the current study suggests completely preventing any engagement in safety behavior, if possible, to maximize treatment outcome (see also Blakey & Abramowitz,
2016).
The current study does have some limitations. First, we found only a few significant effects on the skin conductance measure. Importantly, this was not due to a failure of fear acquisition in skin conductance, given that participants developed discriminative skin conductance responding to the CSs in the Pavlovian fear acquisition training phase in both experiments. One reason would be habituation of skin conductance responses to the CSs and the US (Sokolov,
1960). Given the large amount of trials in the current study, habituation of skin conductance may have minimized any expected effects, especially to trials presented late in the experiment. Second, this study did not directly compare a dimensional measure of US-avoidance to a binary measure. Therefore, it remained unclear whether a dimensional measure of US-avoidance fares any better than a binary measure of avoidance. Nonetheless, we see a dimensional measure of US-avoidance as an assessment of the extent of safety behavior engagement. In addition, the continuous nature of this measure allows it to more sensitively delineate the relationship between safety behavior and conditioned fear. Third, the CSs were not counterbalanced across participants, potentially confounding the results. Forth, the questions in the reward matching procedure in Experiment
2 were presented with negative connotation. Some participants might have misinterpreted the question as in they had to pay out of their own pockets. Therefore, they might only answered “yes” to the lower range of monetary value despite the real level of individually matched competing reward might have been higher than that. This might have led to a considerable amount of participants in the Reward group engaging in a high degree of US-avoidance, hence reducing the effect of individually matched competing reward on protection from extinction. Future studies can phrase the reward matching questions in a positive connotation, for instance, “Are you willing to tolerate an electrical stimulation if you are given 0.10€?”.
In conclusion, the current study developed a dimensional measure of US-avoidance that had a negative linear relationship with US administration and the amount of competing reward. The introduction of a competing reward reduced the degree of US-avoidance and initiated extinction learning to the CS+. However, the competing reward had no apparent effect on the alleviation of protection from extinction. The apparent null group difference in protection from extinction may be attributed to the competing reward not strong enough to motivate fear-opposite action (especially in Experiment 1), expectancy violation not being maximized, a change in context from Pavlovian extinction to test, a strategy to not avoid the CS– only, or a combination of these factors. A novel finding was the elevated US-avoidance to the CS– in the Control group in both experiments due to the minimal cost of US-avoidance. This finding further confirmed that low-cost avoidance does not fully reflect the fear-related component (i.e., avoidance is not only motivated by fear to the feared stimulus). Furthermore, combined with past findings (Gangemi et al.,
2012; van den Hout et al.,
2014), this novel pattern suggested that safety behavior with minimal cost may paradoxically increase fear to safety cues in individuals with anxiety-related disorders, therefore clinicians should attend to low-cost safety behavior that may have gone unnoticed.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.