ArticlePDF Available

How Do Actions Influence Attitudes? An Inferential Account of the Impact of Action Performance on Stimulus Evaluation



Over the past decade, an increasing number of studies have shown that the performance of specific actions (e.g., approach and avoidance) in response to a stimulus can lead to changes in how that stimulus is evaluated. In contrast to the reigning idea that these effects are mediated by the automatic formation and activation of associations in memory, we describe an inferential account that specifies the inferences underlying the effects and how these inferences are formed. We draw on predictive processing theories to explain the basic processes underlying inferential reasoning and their main characteristics. Our inferential account accommodates past findings, is supported by new findings, and leads to novel predictions as well as concrete recommendations for how action performance can be used to influence real-world behavior.
How Do Actions Influence Attitudes?
An Inferential Account of the Impact of Action Performance on Stimulus Evaluation
Pieter Van Dessel, Sean Hughes, and Jan De Houwer
Department of Experimental Clinical and Health Psychology
Ghent University, Belgium
Author Note
PVD, SH, JDH, Department of Experimental Clinical and Health Psychology, Ghent University.
PVD is supported by a Postdoctoral fellowship of the Scientific Research Foundation, Flanders
(FWO-Vlaanderen). JDH is supported by Methusalem Grant BOF16/MET_V/002 of Ghent
University. Correspondence concerning this article should be sent to
This paper is not the copy of record and may not exactly replicate the final, authoritative version
of the article as published in Personality and Social Psychology Review. Please do not copy or
cite without authors permission. The final article will be available, upon publication, via its DOI.
Over the past decade, an increasing number of studies have shown that the performance of
specific actions (e.g., approach and avoidance) in response to a stimulus can lead to changes in
how that stimulus is evaluated. In contrast to the reigning idea that these effects are mediated by
the automatic formation and activation of associations in memory, we describe an inferential
account that specifies the inferences underlying the effects and how these inferences are formed.
We draw on predictive processing theories to explain the basic processes underlying inferential
reasoning and their main characteristics. Our inferential account accommodates past findings, is
supported by new findings, and leads to novel predictions as well as concrete recommendations
for how action performance can be used to influence real-world behavior.
Keywords: attitudes; action effects; approach-avoidance; inferential account; predictive
How do actions influence attitudes?
An inferential account of the effects of action performance on stimulus evaluation
It is almost axiomatic to claim that attitudes exert an important impact on behavior (Ajzen
& Fishbein, 2005). Interestingly, however, there is also a substantial amount of research showing
that behavior can have an impact on attitudes (Olson & Stone, 2005). For instance, nodding one’s
head while listening to a message can improve liking of that message (Wells & Petty, 1980),
selecting one object from two equally attractive alternatives can lead to more favorable
evaluations of the object (Gawronski, Bodenhausen, & Becker, 2007), and making approach
movements when viewing Chinese ideographs can result in more positive ratings of those stimuli
(Cacioppo, Priester, and Berntson, 1993). In this paper, we focus on situations where the
performance of a specific action in relation to a stimulus influences the subsequent evaluation of
that stimulus (i.e., evaluative stimulus-action effects).
There has recently been a surge in interest in evaluative stimulus-action effects, triggered
in part by the seminal work of Kawakami, Phills, Steele, and Dovidio (2007) who found that
repeated performance of approach or avoidance movements in response to images of Black and
White individuals altered evaluations of in- and out-groups. This approach-avoidance (AA)
training procedure has now been adopted in many studies with the typical outcome that repeated
approach leads to more positive stimulus evaluations while repeated avoidance leads to more
negative stimulus evaluations (Van Dessel, De Houwer, & Gast, 2016). The fact that AA training
effects have been found with difficult to change behaviors (e.g., implicit prejudice: Kawakami et
al. 2007; addictive behaviors: Wiers, Eberl, Rinck, Becker, & Lindenmeyer, 2011) and that
effects seem to occur under some of the conditions of automaticity (e.g., unintentionally: Van
Dessel, De Houwer, Gast, Smith, & De Schryver, 2016; unconsciously: Kawakami et al., 2007,
but see: Van Dessel, De Houwer, Roets, & Gast, 2016) are important reasons for the increasing
popularity of AA training studies. AA training effects have now been observed across many
different domains in psychology, including social psychology (e.g., racial evaluations: Phills,
Kawakami, Tabi, Nadolny, & Inzlicht, 2011), clinical psychology (e.g., alcohol: Wiers et al.,
2011; cigarettes: Wittekind, Feist, Schneider, Moritz, & Fritzsche, 2015; social anxiety: Taylor &
Amir, 2012; depression: Becker et al., 2016) and educational psychology (e.g., mathematics:
Kawakami, Steele, Cifa, Phills, & Dovidio, 2008).
Although a growing number of studies have shown that evaluative stimulus-action effects
such as AA training effects are both robust and widely applicable, others have sometimes failed
to obtain such findings (e.g., Becker, Jostmann, Wiers, & Holland, 2015; Krypotos, Arnaudova,
Effting, Kindt, & Beckers, 2015; Van Dessel, De Houwer, Roets et al., 2016; Vandenbosch & De
Houwer, 2011). These contradictory findings suggest that (a) there are important boundary
conditions for these effects, (b) many of these conditions have yet to be discovered, and (c)
establishing such conditions has so far proven difficult. One possible reason for these limitations
is that the dominant theoretical accounts offered to explain this phenomenon are inaccurate.
These accounts assume that a change in mental associations due to (repeated) pairings of stimuli
and actions leads to changes in liking (association formation accounts: e.g., Kawakami et al.,
2007). As we describe below, these dominant accounts provide an intuitive explanation of many
evaluative stimulus-action effects, yet they also have important limitations. We therefore propose
that the time is ripe for considering alternative explanations to association formation accounts
that might provide a better understanding of evaluative stimulus-action effects (see Hughes,
Barnes-Holmes, & De Houwer, 2011, Boddez, De Houwer, & Beckers, 2017, for similar
arguments in the context of learning research in general). With this in mind, we offer a new
perspective on evaluative stimulus-action effects that diverges from traditional accounts and
builds on the idea that inferential (rather than associative) processes underlie changes in liking
due to stimulus-action relations. Our model of evaluative stimulus-action effects is in accordance
with the idea that people often infer stimulus evaluations from their actions (as specified in self-
perception-theory: Bem, 1972) and with the increasingly more popular view that, in contrast to
what is often assumed, human cognition does not necessarily depend on two different types of
mental processes such as automatic, associative and controlled, propositional processes (see
Melnikoff & Bargh, 2018). Furthermore, our model is consistent with recent evidence indicating
that propositional processes play an important role in several psychological phenomena that have
long been considered as associative in nature (e.g., automatic evaluation: see De Houwer, 2014a;
Mann & Ferguson, 2015; or intuitive judgments: see Kruglanski & Gigerenzer, 2011).
In the remainder of this paper, we first discuss the strengths and limitations of association
formation accounts of evaluative stimulus-action effects. We then describe the core concepts that
make up our inferential account and outline the general processes that operate on these concepts
in order to produce evaluative stimulus-action effects. Thereafter, we delineate the inference
steps involved in these effects and potential moderators according to our model. In the General
Discussion, we highlight the added explanatory, predictive, and influence value of our model. We
close with a discussion of the potential limitations and future directions offered by our account.
Association Formation Accounts
According to dominant accounts of evaluative stimulus-action effects, the pairing of
stimuli and actions leads to a co-activation of their corresponding mental representations which
automatically creates an association between the two representations (Kawakami et al., 2007).
This association is typically conceived of as an unqualified link that transmits activation from one
representation to another (Shanks, 2007). Once a strong enough association has been established,
presentation of the stimulus will result in activation of the stimulus representation, which will
then increase activation of the action representation. If the action representation contains
evaluative components, this can lead to an evaluative response to the stimulus that is in-line with
the valence of the action. This explanation is similar to associative explanations of evaluative
conditioning (EC) effects (i.e., evaluative changes resulting from the pairing of a stimulus with
other, valenced, stimuli) (Baeyens, Eelen, Crombez, & Van den Bergh, 1992). Pairing a stimulus
with a valenced event (i.e., performance of a valenced action or presentation of a valenced
stimulus) creates a link between valenced representations and stimulus representations which can
- in turn - lead to automatic changes in stimulus evaluation. Several associative accounts have
been proposed to explain evaluative stimulus-action effects, and these differ mainly in the
specific action representations that are assumed to become associated with stimulus
representations. The three most popular and well-described accounts assume that associations are
formed between representations of the evaluative stimuli and (a) representations of evaluative
action attributes such as the valenced words used to describe the actions (common-coding
account of evaluative stimulus-action effects: Eder & Klauer, 2009), (b) positive representations
of the self (self-anchoring account of AA training effects: Phills et al., 2011), or (c) motivational
systems of approach and avoidance (motivational-systems accounts of AA training effects: Wiers
et al., 2011).
In accordance with EC research (see Hofmann, De Houwer, Perugini, Baeyens, &
Crombez, 2010, for an overview), the fact that evaluative stimulus-action effects can occur under
conditions of automaticity (e.g., in the absence of stimulus awareness: Kawakami et al., 2007, or
without an intention to let action performance influence stimulus evaluations: Van Dessel, De
Houwer, Gast, Smith, et al., 2016) has been considered as strong support for the idea that
automatic association formation mechanisms underlie these effects. Specifically, these results are
in-line with the view that evaluative stimulus-action effects are mediated by processing in
automatic systems that operate on the basis of association formation (Strack & Deutsch, 2004).
Further support for this idea has come from studies showing that stimulus-action pairings can
alter difficult to change or spontaneous behaviors that are assumed to be tied into these systems
(e.g., implicit prejudice: Kawakami et al., 2007; addictive behavior: Wiers et al., 2011). One
study also showed that evaluative stimulus-action effects can be enhanced when there are a
higher number of stimulus-action pairings, consistent with the prediction made by associative
accounts that the number of pairings determines the strength of stimulus-action associations, and
as a result, the evaluative effect (Woud, Becker, & Rinck, 2011).
Importantly, however, the findings described in the previous paragraph, such as the
observation that evaluative stimulus-action effects can occur under conditions of automaticity,
are no guarantee that association formation processes mediate these changes in liking (see
Mitchell, De Houwer, & Lovibond, 2009). In fact, evidence suggests that current associative
accounts of evaluative stimulus-action effects have difficulty explaining many of the observed
evaluative stimulus-action effects. First, most associative accounts imply that there are few (if
any) boundary conditions for evaluative stimulus-action effects. Repeated performance of
valenced actions such as approach and avoidance in response to stimuli should automatically lead
to association formation (e.g., formation of associations between stimulus representations and
representations of valenced words: Eder & Klauer, 2009, representations of the self: Phills et al.,
2011; or motivational-systems: Wiers et al., 2011) and resulting changes in evaluation. Hence,
these models have difficulty explaining why some, but not other, studies produce AA training
effects (see Mertens, Van Dessel, & De Houwer, 2018, for a discussion). Note that some
associative accounts that were developed outside of the literature on AA training effects have
specified assumptions about possible boundary conditions of association formation (e.g.,
attention for the pairings: Wagner, 1981). This could potentially explain specific null effects in
previous studies on evaluative stimulus-action effects. Nevertheless, current accounts of AA
training effects provide little information about the assumed boundary conditions of association
Second, recent studies have shown that evaluative stimulus-action effects are moderated
by specific variables that are not easily explained by associative accounts. For instance,
awareness of stimulus-action contingencies has been found to be an important moderator of AA
training effects (see Van Dessel, De Houwer, & Gast, 2016). This does not fit well with current
associative accounts, which assume that pairings lead to automatic associative changes (e.g., in
the absence of contingency awareness: Kawakami et al., 2007).
Third, other studies have shown that instructions about future actions can also create
changes in stimulus evaluations and that these effects share important similarities with
experience-based effects (e.g., unintentionality: Van Dessel, De Houwer, Gast, Smith, et al.,
2016). Similarly, mere observation of approach-avoidance actions can also lead to evaluative
stimulus-action effects (Van Dessel, Eder, & Hughes, in press). Because instructions and
observations do not involve pairings of stimuli and valenced actions, it is unclear how these
effects might occur on the basis of associative processes (see Lovibond, 2003 for a detailed
discussion of the limitations of associative learning models in accounting for learning via
instructions). It is also unclear how associative accounts can explain evaluative stimulus-action
effects that only involve single pairings of stimuli with valenced actions (e.g., Centerbar, Schnall,
Clore, & Garvin, 2008) or pairings of stimuli with actions of neutral valence (Bem, 1972).
In sum, although it is difficult if not impossible to refute association formation models of
evaluative stimulus-action effects as a class of models, there are several findings that do not
readily fit with the association formation models currently available in the literature. With this in
mind, the current paper explores the merits of another type of model of evaluative stimulus-action
effects that differs in important ways from the associative accounts outlined above.
The Inferential Account
Basic Building Blocks
Propositions. The core conceptual unit of our inferential account is the propositional
representation. A propositional representation is a mental representation that constitutes a
statement about the world (De Houwer, 2014a). Propositional representations contain relational
information (i.e., information about how concepts are related; see Lagnado, Waldmann,
Hagmayer, & Sloman, 2007) which distinguishes them from typical (unqualified) associative
representations. Note that propositions could - in principle - be implemented in associative
networks, provided that those networks are capable of encoding relational information (De
Houwer, 2014a). Moreover, and unlike what is often assumed, propositions are not necessarily
verbal, but can involve embodied or grounded representations (De Houwer, 2014b). Hence, non-
human and non-verbal (e.g., infants) organisms can also store propositional information.
We argue that relational information is at the core of the inferences that mediate
evaluative stimulus-action effects. For instance, there is an important difference between the
acquisition of propositional information that “I am approaching stimulus A” compared to
“Stimulus A approaches me”. “I”, “approach”, and “stimulus A” are present in both cases but the
relation between the three concepts – that is, the specified role of each of the concepts - is
fundamentally different. A recent study indicated that this difference in relational information can
moderate evaluative learning: instructions stating that participants would perform an AA action
in relation to a stimulus produced bigger changes in evaluations than instructions stating that the
stimulus would perform an AA action in relation to participants (see Van Dessel, De Houwer, &
Smith, 2018). Such (difference in) relational information is difficult to capture by models that
operate on the basis of associative representations (Hummel, 2010; Gentner, 2016).
Inferences. We define an inference as a specific sub-type of propositional representation:
it is a proposition (thus it is relational rather than associative in nature), but one that is
constructed on the basis of other propositional information. The construction process that leads to
the inference can be seen as an information generation procedure that involves the application of
information generation (i.e., inference) rules to information that is currently entertained. Note that
we use the term ‘inference’ to describe the outcome of the computation process rather than the
computation process itself. We refer to the computation process as ‘making an inference’ or
‘inferential reasoning’. Our definition of an inference is broad in the sense that inferential
reasoning can occur on the basis of a multitude of different inference rules. These rules can
encode general statements about the world (e.g., if-then rules) but they can also constitute mere
similarity metrics (e.g., analogical mapping rules: Gentner & Smith, 2013) (Hahn & Chater,
1998). As we explain below, however, we make specific predictions about the processes
underlying inferential reasoning and the inference rules that people use under specific
circumstances, constraining our account.
Our definition of an inference implies that not all propositions are necessarily inferences
(but all inferences are propositions). However, in the current model we draw on the assumption
of predictive processing theories that the activation of propositional information constitutes an
inferential process that involves the prediction of information (i.e., construction of information
that is compatible with activated information, see the following section). From this perspective,
any activated propositional information can be construed as an inference. We will therefore
always use the term inference (rather than proposition) in the continuation of this manuscript.
Note that in our model, all inferences can also be seen as ‘predictions’ in the sense that they
constitute information that is predicted (i.e., constructed on the basis of probabilistic
Evaluations. The outcome of the processing steps described in our account is a change in
stimulus evaluation. In accordance with De Houwer, Gawronski, and Barnes-Holmes (2013) we
use the term “evaluation” to refer to a behavioral phenomenon, that is, the impact of stimuli on
evaluative responses. Note that this definition of evaluation avoids conflation of the to-be-
explained behavior (i.e., the evaluative response) with the mental construct that is used to explain
the behavior (i.e., the attitudinal representation). Stimulus evaluations can occur under the
various conditions of automaticity (e.g., uncontrolled, unconscious, efficient, or fast; see Moors,
2016) (i.e., implicit evaluation) or arise in a more deliberate and controlled manner (i.e., explicit
evaluation). Our inferential model describes how the performance of actions in response to a
stimulus can produce changes in both types of evaluations.
The Inferential Process
To clarify the basic mental processes underlying evaluative stimulus-action effects, we
will first describe the nature of inferential reasoning in general. In doing so, we draw on the
notion that predictive processing can provide the basis for inferential reasoning.
On the basis of
this idea, we make three key assumptions that help explain evaluative stimulus-action effects,
namely that inferential reasoning (1) strongly depends on momentary goals, (2) is highly
contextual, and (3) is learning-dependent.
Predictive processing. Our approach is based on an idea that is at the core of many recent
theories in various areas of psychological science (e.g., psychophysiology: George & Hawkins,
2009; perceptual psychology: Proulx, 2014; psychopathology: Fletcher & Frith, 2009; see
Metzinger & Wiese, 2017, for an overview), namely the idea that predictive processing is the
basis of cognitive processing. In this view, the mental system is seen as a “prediction machine”
that is constantly anticipating events in the world around it in order to be able to respond to them
quickly and accurately (Helmholtz, 1962). Bayesian approaches to cognitive processing assume
that this comprises the continuous updating of a person’s generative model of the world through a
process that involves computing probabilities - on the basis of Bayes’ theorem (1958) - with the
aim of integrating and updating prior evidence for stored information (see Penny, 2012). This
predictive inference mechanism is considered of great evolutionary importance because it helps
optimize the use of energy expenditure (by reducing prediction error) and avoid entropy (i.e.,
disorder) while allowing organisms to respond to the environment in an optimal fashion (Friston,
2010). Importantly, this mechanism might also fit the architecture of the brain and its various
substrates (see George & Hawkins, 2009; Bastos et al. 2012).
From this perspective, inferential reasoning essentially involves the construction of
information that is compatible with activated information on the basis of a person’s mental model
Note that inferential models - as a class of models - do not necessarily postulate that inferences are based on
predictive processing. We do, however, add this assumption to our own account because it adds precision and leads
to many novel predictions.
of the world. This generative model can be seen as an information network that represents
information in a hierarchical manner such that information at higher levels can be used to predict
compatible information at lower levels (Friston, 2008). The information is essentially
propositional as it has a specific truth value (i.e., a probabilistic index for retrieval). Inferential
reasoning thus involves drawing probabilistic samples from a pool of propositional information
by applying inference rules to currently entertained information (Sanborn & Chater, 2016). For
instance, when information is entertained that one has performed an approach action in response
to a certain stimulus, rules of analogical mapping may be applied to this information to compute
compatible information (e.g., information about stimulus valence) with a certain precision.
It is important to note that predictive processing accounts assume that inferential
reasoning is not necessarily slow and effortful, which deviates from certain other inferential
reasoning theories). Rather, inferential reasoning (i.e., probabilistic construction of compatible
information on the basis of activated information) can occur in a manner that is to a greater or
lesser extent automatic. Specifically, when certain information is entertained, this will facilitate
fast and easy activation of compatible information when the inference rule that supports this
inference is well-practiced or more effortful activation of compatible information when
application of the required inference rule is more difficult. Note that this account proposes the
same general mechanism for automatic and controlled mental processes (i.e., the prediction of
compatible information on the basis of prior evidence via the application of inference rules that
are more or less difficult to apply). This is consistent with recent recommendations to explore
alternatives to dual-process theories of human cognition (e.g., Melnikoff & Bargh, 2018).
Inferential reasoning is goal-dependent. A first essential question for any inferential
reasoning theory is: “Why do organisms engage in inferential reasoning in the first place?”.
Predictive processing theories assume that an essential feature of organisms is that they strive for
homeostasis or the maintenance of optimal internal states (e.g., for the sake of survival). To
achieve homeostasis, organisms have evolved such that they are able to represent desired states
(i.e., goals). Inferential reasoning may serve the function to construct information that is
compatible with goals, allowing for adaptive action (i.e., actions that allow one to achieve desired
states: Pezzulo, Rigoli, & Friston, 2015). As a result, inferential reasoning may be critically
dependent on the goals that one entertains.
An important feature of predictive processing theories is that actions are considered
‘active inferences’ or inferences that act on the environment (Friston, 2010). Specifically, it is
assumed that, when a desired state is activated (e.g., to be satiated), organisms will make
predictions about available actions and their respective outcomes. When the desired end state is
predicted with sufficient precision on the basis of a specific action (e.g., opening the fridge), this
will cause the activation of information about proprioceptive states necessary for the action,
which will lead to action execution (Cisek & Pastor-Bernier, 2014; Chetverikov & Kristjansson,
2016). From this perspective, actions are considered to be essentially goal-directed in that they
are emitted on the basis of a person’s activated goals. We assume that active inferences share this
feature with inferences that involve the construction of other information than proprioceptive
information (e.g., sensory or verbal information). For instance, not only inferences involved in
action planning (e.g., information about proprioceptive states necessary for AA actions) but also
inferences involved in action interpretation (e.g., information about the valence of AA actions)
will depend on active representations of desired states.
Inferential reasoning is context-dependent. A second essential question for an
inferential theory is: “What inferential reasoning will organisms engage in?”. The answer to this
question is based on the assumption that organisms strive to minimize energy expenditure
(Friston, 2010). Inferential reasoning that takes into account large amounts of information will
therefore be unserviceable. Instead, inferential reasoning is considered to be highly contextual
such that it involves the sampling of information on the basis of momentarily entertained
information (Sanborn & Chater, 2016). This can explain why inferential reasoning is not
necessarily optimal even though it might depend on optimizing rules (e.g., Bayesian updating of
information). It is a common misconception that inferential reasoning is ‘cold’, rational, and
error-free (Moors, 2014). Rather, inferential reasoning can be irrational, (in part) because
currently entertained information strongly biases inferential reasoning. Depending on the context,
people might entertain information that leads them to make inferences that are not logical in
nature. For instance, recent evidence suggests that contextual retrieval of information can lead to
belief biases (see Banks, 2013) or probabilistic reasoning errors (see Sanborn & Chater, 2016).
The context-dependence of inferential reasoning might also explain why people engage in sub-
optimal behavior. Specifically, action selection may depend on what information about desired
end states is contextually available, such that only a subset of all relevant goals will inform a
person’s actions (see also Moors, Boddez, & De Houwer, 2017). From this perspective,
inferential reasoning (and action selection) can be seen as a satisfying process: if activated
information is good enough to achieve specific (contextually-activated) goals, the mental system
will stop sampling information and save its energy.
Inferential reasoning is learning-dependent. A third important assumption of many
predictive processing accounts (but also other inferential reasoning accounts) that might help
explain what inferences people make, is that inferential reasoning involves the application of
inference rules that have their roots in both phylogenetic and ontogenetic development. For
instance, a person may become more inclined to apply a logical modus ponens rule (if “A implies
B” and “A” both hold, then we can deduce “B) to specific information if the application of this
rule to similar information has led to good outcomes in the past (i.e., accurate predictions for
achieving specific goals). The fulfillment of goals should trigger the updating of probabilities (on
the basis of Bayes rule) such that application of certain inference rules will be more likely to be
repeated in the future. Note that this mechanism can also lead to irrational inferences. For
instance, a previously learned rule that often leads to good outcomes may be applied even when
application of this rule is not suited in the current environment. Hence, inferential processes can
sometimes be biased because the inference rules that support them are not logically correct or are
incorrectly applied. In fact, inferential reasoning may often depend on (heuristic) rules that do not
always lead to optimal behavior (Evans, 2010; Kahneman, Slovic, & Tversky 1982; Kruglanski
& Gigerenzer, 2011). For instance, people frequently use an availability heuristic (i.e., they give
disproportionate weight to easily available information) when making inferences. These
(heuristic) rules may be strongly integrated in a person’s cognitive system (and hence, easily
used) because they are ‘ecologically’ rational (i.e., they often lead to good outcomes in a person’s
typical environment, see Brighton & Gigerenzer, 2012, for an elaborate discussion).
Processing Steps: from Stimulus-Action Relations to Evaluations
Now that we have specified the nature of inferences and how these inferences are brought
about in general, we will provide more detailed information about the inferential processing steps
involved in evaluative stimulus-action effects specifically. We assume that the impact of action
performance on stimulus evaluations is mediated by a four-step inferential process (see Figure 1)
and that stimulus-action effects only arise when all steps are completed. For each step, we first
describe the specific inference that is formed. Next, we clarify how this inference is formed and
what factors moderate this inferential processing step (and subsequent evaluative stimulus-action
effects) by drawing on the assumptions about the properties of inferential reasoning that we
outlined above.
Step 1: Inference about a Stimulus-Action relation (I
). The inferential process is
initiated when one or more actions are performed in response to a stimulus in the environment.
Under certain conditions (see below), this will involve the formation of an inference that the
stimulus and action are related to one another (e.g., Stimulus A is approached” or “I always
avoid Stimulus B: Inference about a stimulus-action relation, I
). This first step inference can
be about different facets of the relation (e.g., it can specify a single co-occurrence of stimulus and
response or the strength of a statistical stimulus-response contingency; De Houwer, 2009, pp. 6-
Inferential Process. In-line with predictive processing theories, we assume that people
are very good at detecting relations in the environment because they continuously make
predictions about the world around them (to support appropriate action) and readily update these
predictions based on their interactions with the environment. However, people will not predict or
detect every environmental contingency (this would be too energy consuming). Rather, a
person’s activated goals will determine the amount of attention that an environmental regularity
(e.g., stimulus-action relation) will receive (allowing for more precise predictions: Feldman &
Friston, 2010). Hence, performance of an action in response to a stimulus will lead to the
formation of an inference about a relation between the action and the stimulus (I
) when this
information is in line with one’s current goals. Once registered, the activation level of this
inference will depend on its inferred relevance to current goals, and this will determine the extent
to which the inference biases the generation of other information. When the activation level of I
is sufficiently strong, this will provide the basis for evaluative stimulus-action effects.
Moderators. Our account assumes that any contextual factor that either facilitates or
impedes the goal to detect stimulus-action relations will moderate Step 1. We therefore predict
moderation of evaluative stimulus-action effects on the basis of manipulations of the external
context (the state of the external environment) that achieve this aim. For instance, effects should
be facilitated when participants are informed that there are specific (stimulus-action) relations in
the action task or that it is important to detect them. Enhanced effects should also be observed on
the basis of more indirect instructions that facilitate the goal to detect stimulus-action relations
(e.g., instructions that it is important to perform the experiment in a thoughtful manner) or by
providing (performance-related) incentives. Furthermore, effects should be affected by contextual
manipulations that either facilitate or impede a person’s goal to attend to the stimulus, the action,
or the relation between the two, because this should influence the goal to detect stimulus-action
relations. For instance, we predict a facilitation of evaluative stimulus-action effects when
stimulus identity is task-relevant (see Van Dessel, De Houwer, & Gast, 2016; Van Dessel, De
Houwer, Gast, Roets et al., 2016), when the action task involves more trials (facilitating the goal
to accurately predict the correct action on the basis of stimulus features: see Woud et al., 2011),
and when the relation between action and stimulus is stronger (e.g., deterministic rather than
probabilistic stimulus-response contingencies).
We also assume that internal context factors (internal states of the organism) can
moderate Step 1. First, we predict moderation of evaluative stimulus-action effects by transient
internal states of the organism. For instance, when a person has more task motivation, this should
facilitate the goal to detect stimulus-action relations and therefore enhance effects (see Laham,
Kashima, Dix, Wheeler & Levis, 2014; Zogmaister, Perugini, & Richetin, 2016). Second, we
predict moderation by more stable internal states, in-line with the assumption that inferences
strongly depend on a person’s pre-existing beliefs (which are the result of their prior learning
history). For instance, a person who has learned that it is usually beneficial to register stimulus-
action contingencies should show stronger effects. Individual difference factors such as general
processing fluency or need for cognition should also facilitate effects because they improve the
efficiency or extensiveness of inferential reasoning in general.
Note that our account assumes that the inference about a stimulus-action relation (I
) is a
more proximal determinant of evaluative stimulus-action effects than the actual regularity in the
environment. Hence, the subjective representation of this relation should have a stronger impact
on evaluative stimulus-action effects than objective experiences of the stimulus-action relation.
This accords with studies showing that (1) instructions that specify I
can lead to a change in
stimulus evaluation even in the absence of actual action performance (Van Dessel, De Houwer,
Gast, & Smith, 2015) and (2) evaluative stimulus-action effects are stronger when participants are
able to report I
(moderation by awareness of stimulus-action contingencies: Van Dessel, De
Houwer, & Gast, 2016). As we noted above, these two findings are difficult to explain on the
basis of association formation theories. Note that, from our perspective, the observation that
awareness of stimulus-action contingencies moderates AA training effects, is due to the ease of
retrieval of the inference about the relation between stimulus and action rather than a causal
effect of awareness per se. Contingency awareness provides a good indication of whether I
been formed in a sufficiently strong manner such that I
is able to support evaluative stimulus-
action effects. As argued by Cleeremans (2014), awareness of acquired information may occur
when activation of this information has acquired sufficient strength such that the information is
predicted by the mental system itself (i.e., information is represented on a meta-level). Thus,
although awareness does not cause the formation of inferences, inferences that are formed in the
absence of awareness may be much more weakly represented such that they are less easily
retrieved and therefore are less likely to be used during the inferential process.
Step 2: Inference about an Action-Evaluation relation (I
). The second step involved
in evaluative stimulus-action effects is the construction of an inference that relates the performed
action to a certain valence (Inference about an action-evaluation relation, I
). This inference can
refer to the valenced properties of the performed action (e.g., “approaching IS positive”) but it
can also relate the action to valence in other ways (e.g., Pleasant stimuli are typically
Inferential Process. When participants perform an action in response to a stimulus or
when participants make inferences about a stimulus-action relation (I
) (completion of Step 1
can influence Step 2), they will construct compatible information on the basis of their generative
model of the world. Depending on a person’s activated goals, this can involve the construal of
information that refers to the performed action and that relates the performed action to a certain
valence. For instance, participants may retrieve detailed episodes of previous positive or negative
experiences with this action (e.g., other moments when positive stimuli were approached) when
this information is easily generated on the basis of current goals (e.g., the goal to evaluate). When
activation of such information is sufficiently strong, it will provide the necessary input for Step 3
and hence determine evaluative stimulus-action effects.
Moderators. Our account assumes that any contextual factor that either facilitates or
impedes a person’s goal to draw inferences about a relation between the performed action and
valence will moderate Step 2. We therefore predict facilitation of evaluative stimulus-action
effects on the basis of instructions that provide information about a relation between the
performed action and valence (see Van Dessel, Hughes, De Houwer, & Smith, 2018) or when
providing instructions or other incentives to retrieve such information. We also predict
facilitation when participants are incentivized to attend to (a) the action (see Step 1), (b) valence
in general (e.g., by making valence task relevant) or (c) a relation between the action and valence
(e.g., informing participants about evaluative properties of the action). For instance, we recently
found that using valenced terms such as ‘up/down’ or ‘approach/avoid’ to describe an action
(which might communicate the importance of action valence) strongly moderates evaluative
stimulus-action effects (Van Dessel, Eder, & Hughes, in press).
It is important to note that attention can be oriented towards evaluative properties of
specific actions that are not normally considered. For instance, even though a person’s history of
approach and avoidance might be overwhelmingly related to good (approach) or bad (avoid)
things, respectively, activation of incongruent information can be facilitated in specific contexts
(e.g., in the context of an electric shock, avoidance can be positive). A recent experiment
provided evidence that activation of such non-typical I
can lead to contrastive evaluative
stimulus-action effects. In this study, participants evaluated neutral stimuli that were repeatedly
avoided more positively when other, feared, stimuli also had to be avoided rather than
approached (Mertens et al., 2018).
We assume that Step 2 strongly depends on a person’s learning history such that a person
who has had more (salient) experiences in their life history that relate specific actions to valence
will more easily construct information about a relation between this action and valence and will
therefore show stronger evaluative stimulus-action effects. Accordingly, some people might more
easily retrieve information that approaching is positive whereas others might more easily retrieve
information that approaching can sometimes be negative (e.g., they might have learned during
their lifetime that approaching stimuli can be scary; Hsee, Tu, Lu, & Ruan, 2016) and this should
strongly moderate approach-avoidance (AA) training effects.
Step 3: Inference about a Stimulus-Evaluation relation (I
). The third step involves
the construal of an inference about the evaluative properties of the target stimulus (e.g., ‘Stimulus
A is pleasant’: Inference about a stimulus-evaluation relation, I
). We assume that this
inferential process will only lead to evaluative stimulus-action effects when it incorporates both
inferences about a stimulus-action relation (I
) and
an action-evaluation relation (I
). For
instance, when a person has inferred that (1) they approached stimulus A (I
) and (2) they
typically approach stimuli they like (I
), they might apply an “affirm the consequent” inference
rule, allowing them to infer that they like stimulus A (I
). Note that I
does not necessarily
represent an unambiguous relation between stimulus and evaluation. For instance, after a person
retrieves information that (1) they approached stimulus A and (2) approaching is somehow
related to positive valence, they might make the inference that stimulus A is somehow related to
positive valence (transitive inference: Burt, 1911).
Inferential Process. Depending on a person’s activated goals, they may construct
information about a relation between a stimulus and evaluation (I
) on the basis of available
information. Importantly, the generation of I
is insufficient for Step 3. I
and I
need to be
integrated in the inferential process such that they determine the activation level of I
. This will
depend on (1) the activation level of I
and I
and (2) the availability of an inference rule that
facilitates activation of I
on the basis of I
and I
. When I
is activated on the basis of I
and I
this will trigger the integration of I
and I
in evaluation (Step 4), allowing for
evaluative stimulus-action effects.
Moderators. We assume that the third inference step is dependent on (1) a person’s goal
to generate I
and (2) integration of I
and I
in this inferential process. With regard to the first
determinant, we predict stronger evaluative stimulus-action effects when participants learn that it
is important to retrieve evaluative information about the target stimulus (e.g., on the basis of
instructions). Note that this can occur not only during the action task but also during the
evaluation task (at which time the retrieval of evaluative information about the stimulus is task-
relevant). We also predict facilitation of evaluative stimulus-action effects when an incentive is
provided to attend to (a) the stimulus (see Step 1), (b) evaluation in general (see Step 2), or (c) the
relation between the stimulus and evaluation.
Regarding the second determinant (integration of I
and I
), we predict enhanced
evaluative stimulus-action effects when the formation of I
and I
is facilitated (see Step 1 and
2) and when facilitating the use of inference rules that allow constructing I
on the basis of I
and I
. For instance, stronger evaluative stimulus-action effects should be observed when
participants are given experience in applying a rule to I
and I
to infer I
(e.g., via a pre-
training during which participants are required to learn that relations between stimuli and
valenced actions can inform them about stimulus valence). This manipulation should be most
effective if the training occurred in a similar context (facilitating retrieval of the inference in the
current context) and if making the inference led to positive outcomes (facilitating future use of
the inference rule). Importantly, moderation by the first determinant will depend on the second
determinant (i.e., whether participants integrate I
and I
in the inferential process) because
merely generating I
is not sufficient for evaluative stimulus-action effects (i.e., changes in
evaluation that result from performance of a specific action in relation to a stimulus). Hence, we
predict smaller evaluative stimulus-action effects when contextual factors facilitate construal of
(e.g., ‘Person A is positive’) on the basis of other information than I
and I
(e.g., salient
information about positive behaviors of that person). In-line with this latter prediction, a recent
study showed that presentation of information that is highly diagnostic about stimulus valence
(i.e., ‘Niffites are peaceful, civilized, benevolent, and law-abiding; Luupites are violent, savage,
malicious, and lawless’) before AA training can prevent AA training effects on evaluations of
these groups (Van Dessel, De Houwer, Gast, Smith, et al., 2016).
Individual difference factors should moderate Step 3 because one can be more or less
fluent in the application of inference rules (and integration of specific information in evaluative
inferences). We predict that participants who have more experience in using information about
their own actions for making evaluative inferences (e.g., because they previously engaged in
experiments where such integration was useful) should show enhanced effects. Furthermore,
because integration of I
and I
in I
requires rather elaborate inference rules, we predict
reduced (or even non-existent) evaluative stimulus-action effects in organisms with less
developed abilities to follow such inference rules. For instance, the ability to use language may
be an important factor that strongly determines the inferences an individual can make (Gentner,
2016) and integration of I
and I
in I
may require this ability. Hence, non-human animals
that cannot integrate semantic information in their inferences should not exhibit evaluative
stimulus-action effects. Note, however, that the inference of I
on the basis of I
and I
is not
logically valid. We therefore predict that participants who have received formal logic training
will exhibit reduced effects (given the opportunity and motivation to be accurate in evaluation -
see our discussion of Step 4) because these participants might more easily infer that it is not
logically valid to infer stimulus evaluation on the basis of stimulus-dependent actions.
Step 4: Stimulus evaluation. In the fourth and final step, the activation of the inference
about a stimulus-evaluation relation (I
) mediates a subsequent change in stimulus evaluation.
When the stimulus that was involved in the action task is encountered, an evaluative response to
the stimulus is emitted on the basis of the activation of I
. For instance, a person who is asked to
indicate their liking of a stimulus for which they have information available that relates this
stimulus to positive valence will provide a more positive evaluation of the stimulus.
Inferential Process. We assume that inferential processes underlie all cognitive processes,
including stimulus evaluation. In-line with the assumption that action selection is inherently goal-
directed, we postulate that evaluative responses are selected on the basis of their estimated
potency to produce specific (contextually activated) desired outcomes. Hence, information that is
constructed in Step 3 (I
) will inform action selection depending on a person’s goal to use this
information for evaluation. For instance, when a person is asked to indicate their liking of a
stimulus, I
can bias the generation of response-related information (e.g., representation of a
button press that indicates ‘strong liking for a stimulus’) because this accords with the goal to
provide a ‘good enough’ response (and complete the experiment/communicate their feelings/…).
The response will be emitted when this desired outcome is predicted with sufficient precision on
the basis of the activated action representation.
In contrast to associative or dual-process accounts of evaluation (e.g., Gawronski &
Bodenhausen, 2006), our account postulates that both evaluations that are emitted under certain
conditions of automaticity (i.e., implicit evaluations) as well as more controlled (i.e., explicit)
evaluations are the result of a single, inferential, process. Importantly, however, effects on
implicit and explicit evaluations can dissociate because elements of specific procedures that are
used to capture implicit and explicit evaluations can facilitate the activation of distinct goals.
Procedures for measuring implicit evaluations typically require speeded responding and thus
facilitate the corresponding goal to provide fast responses. As a result, we assume that implicit
evaluation strongly depends on inferences that are readily available under those conditions (i.e.,
automatic activation of propositional information: see De Houwer, 2014a). For instance, in an
Implicit Association Test (IAT; Greenwald, McGhee, & Schwartz, 1998), I
can influence
prediction (and resulting execution) of a categorization response to the evaluative stimulus with a
response key that is valence-congruent (i.e., it is also used to categorize stimuli of congruent
valence) in accordance with the goal to emit a fast and accurate response. In contrast, explicit
evaluation measures more strongly facilitate activation of other goals such as the goal to be
accurate in evaluation or the goal to present a positive image of themselves to the experimenter
(i.e., self-presentation goals). As a result, information that requires multiple steps for retrieval
might be more regularly contacted and integrated in explicit evaluation (see also Cunningham,
Zelazo, Packer, & Van Bavel, 2007). For instance, the goal to be accurate in one’s explicit
evaluations may facilitate retrieval of information that is less readily available but is considered
more diagnostic about stimulus valence. This should especially be the case when more easily
retrieved information is not considered very diagnostic about stimulus valence.
Moderators. The translation of I
into stimulus evaluation is assumed to strongly depend
on a person’s current goals. Hence, evaluative stimulus-action effects should typically be
strongest when a goal to evaluate the stimulus is available (e.g., by instructing participants to
evaluate the stimulus in an explicit evaluation task). Importantly, however, evaluative stimulus-
action effects might also arise when participants are instructed to evaluate other stimuli than the
specific stimulus that was involved in the action task because this evaluation goal can be used to
activate I
and integrate this information in their response. Hence, our model can explain
evaluative stimulus-action effects on both explicit and implicit evaluations (see Van Dessel, De
Houwer, Gast, Smith, et al., 2016).
Because implicit and explicit evaluation tasks lead to the activation of different goals, we
can predict specific conditions under which implicit and explicit evaluations can dissociate. First,
we expect that dissociations between implicit and explicit evaluations might depend on the
registered relation between stimulus and evaluation in I
. When participants contact information
that represents an identity relation between stimulus and valence (e.g., Stimulus A IS good”),
this should have a stronger impact on explicit evaluation than information that links stimulus and
valence in a less well-specified manner (e.g., “Stimulus A is somehow related to good”) because
it is more diagnostic about stimulus valence. In contrast, implicit evaluation may be strongly
influenced by both types of information. For instance, there is evidence suggesting that the
propositions “I am good” and “I want to be good” both strongly impact self-evaluations as
measured with an IAT (Remue, Hughes, De Houwer, & De Raedt, 2014) even though only the
former reflects actual self-esteem. Second, we assume that explicit measures more strongly
facilitate the goal to be accurate in the evaluation of the stimulus. Because the inference of
evaluation on the basis of performed actions is not logically valid, participants who contact this
conclusion should show reduced evaluative stimulus-action effects on explicit evaluations
whereas the reduction of effects on implicit evaluations should be less pronounced. In
accordance, one recent study found a dissociation between evaluative stimulus-action effects on
implicit and explicit evaluation (Van Dessel, De Houwer, Gast, Smith, et al., 2016). In this study,
participants first received highly diagnostic information about the valence of the stimuli before
performing an AA training task. Notably, AA training influenced implicit evaluations measured
with an IAT but not explicit evaluations measured with a self-report rating scale. This pattern
might be observed because participants relied on information indicating that approaching and
avoiding a stimulus is not a good basis for explicit evaluation when compared to more diagnostic
information. In contrast, implicit evaluations might reflect AA training contingencies because the
contingency information was easily available and facilitated quick responding (it was a ‘good
enough’ response). Finally, our account also predicts that effects on explicit and implicit
evaluation should be more in agreement with each other when the measurement context
facilitates adoption of similar goals (e.g., when one is asked to “go with their gut” when proving
explicit liking ratings: Ranganath, Smith, & Nosek, 2008; or when information is provided before
IAT administration that the IAT is used to measure attitudes: e.g., Echabe, 2013).
We also assume that integration of an inference about a stimulus-evaluation relation (I
in the evaluative response should depend on individual difference factors that impact a person’s
goal to integrate I
in evaluation. For instance, trait reactance should reduce evaluative stimulus-
action effects because a person’s activation of the goal to be reactant can interfere with the goal
to integrate the learned information in Step 3 in evaluation. As preliminary support for this idea, a
recent study found a strong correlation between AA training effects on evaluations of well-known
social groups and personal reactance measured with a trait reactance scale (Van Dessel, De
Houwer, Gast, Roets, & Smith, 2018). In contrast, demand compliance should increase evaluative
stimulus-action effects because the goal to comply with experimenter’s demands should facilitate
the goal to integrate ISE in their evaluative response. Accordingly, a recent study found that
demand compliance positively correlated with AA training and AA instruction effects in the
context of novel stimuli (Van Dessel, Smith, et al., 2018). One could argue that changes in
stimulus evaluation that are due to demand compliance are less important because they do not
reflect changes in a person’s ‘genuine’ liking of the target stimuli. However, it is difficult to
establish what constitutes as ‘genuine’ liking. Moreover, it is important to note that studies have
found both instruction-based and experience-based evaluative stimulus-action effects for
participants who do not show demand compliance, indicating that these effects do not necessarily
depend on controlled, non-automatic processes that involve the intentional use of the acquired
information for evaluation (e.g., Van Dessel, De Houwer, Gast, Smith, et al., 2017).
The Merits and Limitations of the Inferential Account
Heuristic Value
In the current paper, we have argued that people show evaluative stimulus-action effects
because they make a specific inference about the evaluation of a stimulus (I
) on the basis of
constructed inferences about stimulus-action relations (I
and action-evaluation relations (I
and integrate this inference in stimulus evaluation. This model has heuristic value. First, it can
account for the known characteristics of evaluative stimulus-action effects that are often
considered as support for association formation models. Specifically, our account can explain
why approach-avoidance (AA) training effects sometimes occur under automaticity conditions
(e.g., an unintended impact of training on liking; see Van Dessel, De Houwer, Gast, Smith, et al.,
2016) and why effects can be observed on difficult to change or spontaneous behaviors such as
implicit stimulus evaluations (e.g., Kawakami et al., 2007). Our model encompasses these results
because it acknowledges that inferences can occur in a more or less automatic manner and
because it specifies the conditions under which inferences can lead to changes in implicit
evaluation (see moderation of Step 4).
Second, the heuristic value of the inferential account exceeds that of current association
formation models in that it can also explain the characteristics of evaluative stimulus-action
effects that do not fit well with current associative accounts of these effects. First, it can explain
why studies sometimes fail to find changes in stimulus evaluation on the basis of pairings of
stimuli and valenced actions (e.g., following AA training: Vandenbosch & De Houwer, 2011;
Becker et al., 2015). In contrast to the idea that effects are driven by the automatic installation of
associative links between stimuli and actions, we assume that (a) an inferential process chain is
required that (b) depends on a number of important moderators that either enhance or impede
evaluative stimulus-action effects under specific circumstances. For instance, the activation of
non-dominant information about the relation between an action and evaluation (I
) might
explain some of the observed null findings (Mertens et al., 2018) (see moderation of Step 3).
Second, our account can explain why stronger evaluative stimulus-action effects are observed
under specific conditions that are difficult to explain from an associative perspective, such as
when participants are aware of stimulus-action contingencies (Van Dessel, De Houwer, & Gast,
2016) (see moderation of Step 1). Third, our account can explain recent observations that effects
can arise even in the absence of actual performance of stimulus-based actions but on the basis of
mere instructions or observation of these actions (Van Dessel et al., 2015). These effects are
explained by incorporating into our model a core assumption of propositional theories, namely
the assumption that propositional information about regularities in the environment can be
generated on the basis of instructions, observations, imagination, or experiences with those
regularities (De Houwer, 2009). Moreover, because the acquisition of propositional information
is a more proximal mediator of effects than action performance, our inferential account can
further explain why instruction effects can be stronger than training effects (Hughes, Van Dessel,
Smith, & De Houwer, 2018): it is easier to form inferences about stimulus-action relations when
the contingencies are instructed compared to when one has to discover those contingencies
through trial-and-error.
Finally, the heuristic value of our inferential account extends beyond effects involving
repeated performance of actions in response to a stimulus that we have focused upon so far (i.e.,
training-based procedures such as AA training). In contrast to most association formation
accounts of evaluative stimulus-action effects, our account does not postulate that changes in
evaluation require a large number of pairings of actions and stimuli. Rather, our account can also
explain findings in which actions influence evaluations that arise at the time of the action (e.g.,
Cacioppo et al., 1993). For instance, stimuli viewed during arm flexion might be rated more
positively than stimuli viewed during arm extension because participants infer a stimulus-action
relation (I
) (e.g., that they used an approach movement when viewing the stimulus) and an
action-evaluation relation (I
) (e.g., that approaching is positive) and they use this information
to infer a stimulus-evaluation relation (I
) and their own evaluative responses. Moreover, our
account can also explain evaluative stimulus-action effects that do not involve valenced actions
(e.g., approach-avoidance) such as effects that were obtained in the context of the self-perception
hypothesis (Bem, 1972). Our account accords with the original idea that participants infer
stimulus evaluations from their actions (e.g., smiling, approaching). For instance, when a person
is asked to smile when watching a cartoon., they may infer that the cartoon is funny (I
) on the
basis of two inferences: (1) that they are smiling in the presence of the cartoon (I
) and (2) that
they tend to smile at things they like (I
) (e.g., Laird, 1974).
Similarly, our account can explain findings in the field of emotion research showing that
arousing actions (e.g., doing exercise) can influence evaluations (e.g., of an attractive opposite-
sex confederate: White, Fishbein, and Rutsein, 1981). As argued by predictive processing
theories, organisms need to be good at monitoring internal states (Seth, 2013). Information about
an organism’s own physiological state in the presence of a specific stimulus (I
) and the
positivity or negativity associated with this state in the past (I
) may bias information generation
about evaluative features of the stimulus in the present (I
), which might influence stimulus
evaluation. Related to this idea, our account can also accommodate findings that provided support
for cognitive dissonance theory (Festinger, 1957), which states that a person’s motive to maintain
cognitive consistency can give rise to irrational and sometimes maladaptive behavior. For
instance, in a study by Festinger and Carlsmith (1959), participants rated boring tasks as more
likeable if they had to persuade others that the task was fun, and especially when they received a
small relative to a big sum of money for doing so. In this instance, I
may also be inferred on the
basis of information about a person’s actions (e.g., ‘I must like the task because I talked
positively about it’). In the high money condition, however, there is a second reason available for
participants’ actions (large sum of money) (and hence the evidence for liking the task -I
- is
considered less valid). Note, however, that the explanation of cognitive dissonance, emotional
arousal, or self-perception effects was not the primary purpose of this manuscript and these
explanations are therefore preliminary and require further investigation.
Predictive Value
On the basis of our proposed inferential reasoning framework we can not only explain
established moderators of evaluative stimulus-action effects but also predict new moderators.
Each of the four inferential processing chain steps described above involves inferential reasoning
that follows the described characteristics of inferential reasoning (goal-dependent, context-
dependent, learning-dependent). Hence, in the previous section outlining the processing steps, we
specified many new testable predictions regarding the moderation of evaluative stimulus-action
effects by specific (external and internal) contextual variables. In the studies that we recently
conducted in our lab, we have already tested several predictions that were derived from the basic
ideas of this account. For instance, we recently found that repeated performance of actions other
than approach-avoidance can also lead to changes in stimulus evaluations when these actions are
described in valenced terms and that these changes depend on how positive or negative
participants considered these actions (I
) (Van Dessel, Eder, & Hughes, in press). However, the
inferential account also leads to several interesting predictions that still need to be tested. Hence,
our model will stimulate new research that is bound to increase our understanding of evaluative
stimulus-action effects. In the next sections, we briefly discuss two sets of predictions that we are
currently testing in ongoing research.
Moderation of evaluative stimulus-action effects by the external context. According
to our account, there are many ways to contextually moderate evaluative stimulus-action effects.
On the one hand, each of the proposed steps can be facilitated based on contextual factors that
should enhance effects. For instance, one assumption that we are currently investigating is
whether verbally providing I
(e.g., the inference that people typically approach positive stimuli
and avoid negative stimuli) can lead to stronger AA training effects (because this should facilitate
Step 2; Van Dessel, Hughes, De Houwer, & Smith, 2018). On the other hand, each of the steps
can also be impeded based on contextual factors that should reduce effects. In another study, we
are investigating whether evaluative stimulus-action effects are reduced when participants are
informed that inferring I
on the basis of I
and I
is not in accordance with formal logical
rules (which should impede Step 3; Van Dessel, Hughes, De Houwer, & Smith, 2018).
Moderation of evaluative stimulus-action effects by the internal context. Our account
assumes that characteristics of the organism also strongly influence evaluative stimulus-action
effects. One key assumption is that these effects should depend on the information network that
people bring with them to the experimental context. As we mentioned above, individual
differences in the activation of an inference about an action-evaluation relation (I
) should
moderate evaluative stimulus-action effects in specific directions. We are currently investigating
whether participants’ belief in I
(e.g., whether they consider approaching positive) moderates
evaluative stimulus-action effects. We have also started to examine in a systematic way the role
of individual differences in evaluative stimulus-action effects as they relate to the different
inferential steps in our model (e.g., motivation, task experience).
Influence Value
Changing stimulus evaluations. The inferential account also has important implications
for influencing real-world behavior. Many clinical interventions (e.g., exposure therapies)
involve (approach-avoidance) actions toward stimuli (e.g., Jones, Vilensky, Vasey, & Fazio,
2013). Our account highlights new ways of improving the impact of those interventions.
Specifically, we predict that the manipulation of contextual factors can change the type, number,
evaluative direction, and confidence with which inferences about evaluative stimulus properties
are held, which will influence stimulus evaluations (and resulting changes in real-life behavior:
Ajzen & Fishbein, 2005). In a recent study, we took a first step in applying our inferential
account to improve AA training effects on evaluations of food products (Van Dessel, Hughes, &
De Houwer, in press). We predicted that a training task in which AA actions produced positive or
negative consequences (i.e., an avatar representing themselves became more or less healthy)
when performed in response to specific food products would lead to stronger effects on
evaluations of the food products than typical AA training paradigms (which only manipulate
stimulus-action contingencies). The rationale for this prediction is that consequences would help
participants more easily generate inferences that certain foods are good and other foods are bad
) because negative consequences that follow approaching of a stimulus (and positive
consequence that follow avoidance) are more likely to lead to inferences about the valence of the
stimulus than the mere fact that one consistently approaches or avoids a stimulus. The results
were in-line with our prediction: consequence-based AA training effects on implicit and explicit
evaluations were stronger than effects of typical AA training. These findings also imply that AA
training effects can occur even when participants approach and avoid each stimulus an equal
number of times (contingencies in consequence-based AA training involved stimulus, response,
AND action consequence: e.g., approaching one food product always led to positive outcomes
and avoiding it led to negative consequences), a result that is difficult to explain on the basis of
current association formation accounts. Further in-line with the inferential account, we found that
AA training effects were stronger when the consequences were relevant for participants’ tasks
goals (i.e., when participants were instructed to make an avatar as healthy as possible). Overall,
these results are consistent with our prediction that (small) changes to existing AA training
procedures can help participants to make better inferences and thereby facilitate changes in
stimulus evaluations. On the basis of our account we could therefore have the potential to
improve existing therapies that already use these procedures for changing evaluations (e.g.,
Taylor & Amir, 2012).
Note that our account stops at the translation from inferences into evaluations and does
not specify how changes in evaluation can lead to changes in real-life behavior (Ajzen &
Fishbein, 2005). Our account could be extended to this additional aim by specifying the
inferences underlying the change in behavior that results from evaluations (as well as the
necessary conditions for this inferential process). We believe that this extension could be useful
and might even be feasible (given the abundance of research on this topic). However, it extends
beyond the aims of the current paper and will be an important future endeavor.
Changing unwanted behavior. Importantly, it is also possible that inferences formed on
the basis of stimulus-action contingencies can impact real-life behavior without requiring
mediation via changes in evaluation. Though our model focuses on changes in evaluations, it can
also be easily adapted to account for (and potentially improve) behavior that is not mediated by
changes in evaluation (i.e., the effect of stimuli on evaluative responses). The inferential account
postulates that stimulus-action procedures are effective at changing (pathological) behavior
because they facilitate the installation and retrieval of relevant inferences (rather than installing
stimulus-action associations or ‘response tendencies’
). For instance, when alcoholic patients
repeatedly avoid alcoholic drinks (e.g., Wiers et al., 2011), they might make specific inferences
(e.g., they infer that these drinks are to-be-avoided or that they are able to avoid alcoholic drinks).
Once established, this propositional (rather than associative) knowledge might be contacted in
other contexts, allowing these patients to refrain from drinking. Similarly, when people
consistently avoid unhealthy foods, they might infer that these stimuli are bad for them and this
may inform food choices. In the study described in the previous paragraph (Van Dessel, Hughes,
& De Houwer, in press), we found evidence for this idea. Participants who had performed the
consequence-based AA training in which approaching unhealthy foods led to negative outcomes
and avoiding unhealthy foods led to positive outcomes, reported eating less unhealthily in the
days after the intervention (but not participants who had performed typical AA training without
consequences) and actually ate less unhealthy snacks in an ad libitum snack task. It is possible
Note that automatic response tendencies often do not logically relate to the pathological behaviors under
investigation in cognitive bias modification research (Spruyt et al., 2013; Snelleman, Schoenmakers, de Mheen,
2015) which contrasts with predictions of association formation accounts. In contrast to these accounts, our
inferential account does not assume that the action tendencies produce the unwanted behavior. Rather, action
tendencies (i.e., fast responses in an approach-avoidance measurement tasks) might be determined by the speed of
predictions of stimulus-based actions. This can be based on learned contingencies in Step 1 of the process chain or
other learned information (e.g. ‘I’m required to avoid alcohol’) and should not directly cause the unwanted behavior.
that these effects occurred because the consequence-based AA training more strongly facilitated
the (adaptive) inference that it is good to refrain from unhealthy foods.
These preliminary results point to an important promise for training-based action
interventions that has been claimed ever since these effects were first observed: that action
training (especially AA training) may influence not only difficult-to-change implicit evaluations
(e.g., implicit prejudice: Kawakami et al., 2007) but also difficult-to-change behaviors (e.g.,
addictive behavior: Wiers et al., 2011). Unlike the traditional idea that such change occurs
because these tasks lead to associative changes via repeated pairings, we assume that these
procedures actually help people to make specific inferences. Because these inferences are self-
generated (i.e., a person needs to infer the information on the basis of the contingency
information themselves), the information may be more strongly integrated in a person’s
information network than information that is merely provided to a person (see also research on
behavioral nudging: Benartzi et al., 2017). The person has already made the inference (and may
have done so several times, depending on the extensiveness of the training) which might allow
for the inferences to be easily repeated.
Note that current training procedures might not only be improved by including
consequences of stimulus-action relations and making those consequences goal-relevant but also
by changing other procedural details that promote the installation of novel (and adaptive)
inferences. Such an inference-based therapy is much closer to therapies often used in clinical
practice (e.g., cognitive behavioral therapy) than typical cognitive bias modification therapies
that aim to establish stimulus-action tendencies. However, they might add an automatization
component to the typical therapies used in clinical practice (because they involve training or
repeated generation of certain inferences). This ‘inference training’ could be an important new
method for establishing important clinical effects (especially seeing as current ‘cognitive bias
modification trainings’ to change clinical behavior often do not produce beneficial effects: e.g.,
Cristea, Kok, & Kuijpers, 2015; Jones, Hardman, Lawrence, & Field, 2017).
Relation to other accounts
Our inferential account of evaluative stimulus-action effects is the first to provide an
elaborate account of evaluative learning effects from an inferential (and propositional)
perspective (i.e., an account that makes underlying processes explicit). In doing so, we have
drawn on a number of more general accounts of human learning and mental processing. First, our
account makes many of the same assumptions as single-process propositional accounts of
learning (e.g., Mitchell et al., 2009). Stimulus-action effects are explained with reference to a
single memory system that involves the activation of propositional information and use of this
information in inferential reasoning. These inferential processes operate on the basis of
information that encodes not only co-variation but also the relational properties of concepts.
Second, our model draws on predictive processing models (e.g., Feldman & Friston, 2010).
Although these models currently enjoy widespread appeal elsewhere in psychological science,
this idea has yet to find its way into evaluative learning research. Our model integrates the idea
that the making and updating of predictions (on the basis of Bayesian rules) might underlie
(action) effects on stimulus evaluation and makes assumptions on the basis of dominant ideas in
the predictive processing literature. For instance, our account builds on the idea that activation
level of information is an important concept for explaining human behavior in general and
reasoning in particular (Sanborn & Chater, 2016; see also Atomic Components of Thought-
Rational theory: Banks, 2013).
Our model combines single-process propositional accounts of learning with predictive
processing accounts and applies this specifically to evaluative stimulus-action effects. The idea
that people infer stimulus evaluations from their actions is in-line with self-perception theory
(Bem, 1972). However, our account provides more detail about the underlying conditions.
Moreover, it is broader to the extent that it not only explains effects that arise when emotional
responses are ambiguous but also the abundance of recent findings on training-based effects.
Furthermore, it provides an explanation for effects on implicit and explicit evaluation. In doing
so, we build on recent findings and recent theorizing indicating that propositional processes
underlie not only explicit but also implicit evaluation (De Houwer, 2014a; Mann & Ferguson,
2015). Again, however, we formalize this idea in an inferential model by building on general
accounts of human cognition, specifying clear assumptions, and providing testable predictions.
Our inferential account also bears similarity to the Theory of Event Coding (TEC;
Hommel, Müsseler Aschersleben & Prinz, 2001), to the extent that both theories assume that
action performance is critically dependent on anticipated action consequences. Crucially,
however, the theory of event coding (and the common-coding account of AA training effects of
Eder & Klauer, 2009, which is derived from this theory) assumes that the automatic formation of
associations between action representations and perceptual consequences mediates action
performance. In contrast, our inferential account combines principles from predictive processing
theory and TEC to explain evaluative behavior on the basis of inferential reasoning (see Butz,
2016, for an integrative theory of human cognition in general). A recent study that pitted
predictions of the common-coding and inferential account of AA training effects against each
other provided stronger support for the inferential account which predicted AA effects on the
basis of mere action observation (Van Dessel, Eder, & Hughes, in press).
Limitations of the Inferential Account
The most important limitation of our account may be that it is still a relatively general
account at this point (though this can also be seen as a strength of theories: Gawronski &
Bodenhausen, 2015; Meiser, 2011). We have specified many different boundary conditions and
moderators (which increases the falsifiability of our account), the processing steps necessary to
produce changes in evaluation, and important details about how inferential reasoning might
occur. Yet, there are still unconstrained factors. Most importantly, we have not specified (all) the
specific inferences (e.g., I
) that might be involved in evaluative stimulus-action effects. Our
reason for this is that there is much variation in the specific inferences that individuals make (e.g.,
because this depends on the availability of information in a person’s information network which
may strongly differ between people). Indeed, in a recent study, we asked participants to indicate
why they would rate an approached stimulus as more positive than an avoided stimulus. We
found marked variability in the specific responses obtained. Many of the provided reasons were
not highly elaborated (e.g., ‘avoiding a stimulus means that something is wrong with it’)
suggesting that participants often use heuristic rules (e.g., an availability heuristic) to infer
stimulus evaluation on the basis of approach-avoidance (AA) training (and instructions).
Because of the complexity and uncertainty with regard to the inferences people might
contact, propositional or inferential models are sometimes thought to be inferior to more simple
and/or domain specific models (e.g., associative models which assume that evaluative stimulus-
action effects occur on the basis of simple links that are formed as the result of repeated pairings).
Note, however, that parsimony should not come at the cost of explanatory or influence value. Our
model can be considered useful because it has high heuristic and predictive value (i.e., it allows
us to explain existing findings, predict novel findings, and influence human behavior in novel
ways). Note also that although individual association formation accounts might be considered
simple (e.g., Phills et al., 2011), these accounts cannot explain many of the relevant findings in
this literature (see above). As a class, however, association formation models have a high degree
of flexibility, which allows proponents of these models to always make post-hoc adaptations to
explain obtained results. However, different findings require different adaptations that are often
logically inconsistent. In sum, our model leads to a whole set of a priori predictions for which it
is difficult to see how a similar set of predictions could be made on the basis of any individual
association formation model or subclass of logically consistent association formation models. It
is also possible that a dual-process framework incorporating inferential and associative processes
could explain evaluative stimulus-action effects. However, an account that can explain these
effects on the basis of one coherent processing mechanism (i.e., inferential reasoning) should be
preferred over an account that additionally postulates the existence of an entirely different second
mechanism (e.g., association formation) for reasons of parsimony.
It is also important to note that by providing a predictive processing framework for our
account we are directing it away from accounts that require a strong ‘homunculus’ factor. We
provide an explanation of how evaluative stimulus-action effects might occur in a more or less
automatic or uncontrolled manner and under which circumstances these effects should arise. In
doing so, we clarify that effects that occur under specific automaticity conditions do not
necessarily require associative explanations. Inferential accounts can explain all aspects of
evaluative stimulus-action effects - not only those that are highly controlled (e.g., effects
resulting from demand compliance). Moreover, we also clarify that inferential reasoning can
operate on the basis of (neurologically plausible) mental mechanisms that have specific
characteristics. This also opens up the possibility of eventually specifying a more elaborate
mathematical model of these complex effects by implementing Bayesian probability calculus
(Friston, 2003).
In providing this inferential framework for evaluative stimulus-action effects, we have
also opened up new avenues for the construal of novel accounts of other (evaluative) learning
phenomena (e.g., EC) that might operate on the basis of similar processes. These accounts
distinguish themselves from (current) propositional accounts in that they provide an explanation
based on the generation of inferences (a subset of propositions) and that they provide an elaborate
explanation of how this mechanism might occur. Of course, it is important to appreciate that our
account provides only an initial framework (applied to one specific type of effects). Because we
do not want to extend our model too far beyond the data, we think that further specification
should come on the basis of empirical findings.
Concluding Remarks
In the current paper we have described a model of evaluative stimulus-action effects that
draws on inferential rather than simple associative processes. We have specified the process steps
that underlie these effects as well as how the processes underlying these steps might work and
under what conditions they operate. We hope that this new framework can help improve our
understanding of evaluative stimulus-action effects (and other evaluative learning phenomena)
and further improve the utility of (action-based) procedures in clinical domains.
Ajzen, I., & Fishbein, M. (2005). The influence of attitudes on behavior. In D. Albarracín, B. T.
Johnson, & M. P. Zanna (Eds.), The handbook of attitudes and attitude change (pp. 173-
221). Mahwah, NJ: Erlbaum.
Baeyens, F., Eelen, P., Crombez, G., & Van den Bergh, O. (1992). Human evaluative
conditioning: Acquisition trials, presentation schedule, evaluative style and contingency
awareness. Behaviour Research and Therapy, 30, 133–142. doi: 10.1016/0005-7967
Banks, A.P. (2013). The influence of activation level on belief bias in relational reasoning.
Cognitive Science, 37, 544-577. doi:10.1111/cogs.12017
Bastos, A. M., Usrey, W. M., Adams, R. A., Mangun, G. R., Fries, P. & Friston, K. J. (2012).
Canonical microcircuits for predictive coding. Neuron, 76, 695-711.
Bayes, T. (1958). Studies in the history of probability and statistics: IX. Thomas Bayes' (1763)
essay “Towards solving a problem in the doctrine of chances” in modernized notation.
Biometrika, 45, 296–315
Becker, E. S., Ferentzi, H., Ferrari, G., Möbius, M., Brugman, S., Custers, J., et al. (2016).
Always approach the bright side of life: A general positivity training reduces stress
reactions in vulnerable individuals. Behaviour Research and Therapy.
Becker, D., 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.
Bem, D. J. (1972). Self-perception theory. In L. Berkowitz (Ed.), Advances in experimental
social psychology, Vol. 6. (pp. 1–62). New York: Academic Press.
Benartzi, S., Beshears, J., Milkman, K. L., Sunstein, C. R., Thaler, R. H., Shankar, M., Tucker-
Ray, W., W.J. Congdon & Galing, S. (2017). Should Governments Invest More in
Nudging? Psychological Science. Advance online publication.
Boddez Y., De Houwer J., & Beckers T. (2017). The inferential reasoning theory of causal
learning: Towards a multi-process propositional account. In: Waldmann M. (Eds.), The
Oxford Handbook of Causal Reasoning. Oxford: Oxford University Press.
Brighton, H., Gigerenzer, G. (2012). Homo heuristicus: less-is-more effects in adaptive
cognition. Malaysian Journal of Medical Science, 19, 6–16.
Burt, C. (1911). Experimental tests of higher mental processes and their relation to general
intelligence. Journal of Experimental Pedagogy, 1,93-112
Butz, M. V. (2016). Toward a Unified Sub-symbolic Computational Theory of Cognition.
Frontiers in Psychology, 7, 925.
Cacioppo, J. T., Priester, J. R., & Berntson, G. G. (1993). Rudimentary determinants of attitudes
II: Arm flexion and extension have differential effects on attitudes. Journal of Personality
and Social Psychology, 65, 5–17. doi:10.1037/0022-3514.65.1.5
Centerbar, D.B., Schnall, S., Clore, G.L., & Garvin, E.D. (2008). Affective incoherence: When
affective concepts and embodied reactions clash. Journal of Personality and Social
Psychology, 94, 560-578. doi: 10.1037/0022-3514.94.4.560
Chetverikov, A. & Kristjánsson, Á. (2016). On the joys of perceiving: Affect as feedback for
perceptual predictions. Acta Psychologica, 169, 1-10.
Cisek, P. & Pastor-Bernier, A. (2014) On the challenges and mechanisms of embodied decisions.
Philosophical Transactions of the Royal Society, 369: 20130479.
Cleeremans, A. (2014). Connecting conscious and unconscious processing. Cognitive Science,
38, 1286-1315.
Cristea, I. A., Kok, R. N., & Cuijpers, P. (2015). Efficacy of cognitive bias modification
interventions in anxiety and depression: meta-analysis. The British Journal of
Psychiatry, 206, 7-16; doi: 10.1192/bjp.bp.114.146761
Cunningham, W. A., Zelazo, P. D., Packer, D. J., & Van Bavel, J. J. (2007). The iterative
reprocessing model: A multilevel framework for attitudes and evaluation. Social
Cognition, 25(5), 736-760. doi: 10.1521/soco.2007.25.5.736
De Houwer, J. (2009). The propositional approach to associative learning as an alternative for
association formation models. Learning & Behavior, 37, 1-20. doi: 10.3758/LB.37.1.1
De Houwer, J. (2014a). A Propositional Model of Implicit Evaluation. Social and Personality
Psychology Compass, 8, 342-353. doi:10.1111/spc3.12111
De Houwer, J. (2014b). Why a propositional single-process model of associative learning
deserves to be defended. In J. W. Sherman, B. Gawronski, & Y. Trope (Eds.), Dual
processes in social psychology (pp. 530-541). NY: Guilford.
De Houwer, J., Gawronski, B., & Barnes-Holmes, D. (2013). A functional- cognitive framework
for attitude research. European Review of Social Psychology, 24, 252–287.
Echabe, A. E. (2013). Relationship Between Implicit and Explicit Measures of Attitudes: The
Impact of Application Conditions. Europe’s Journal of Psychology, 9, 231−245.
Eder, A. B., & Klauer, K. C. (2009). A common-coding account of the bidirectional evaluation–
behavior link. Journal of Experimental Psychology: General, 138, 218–235. doi:
Evans, J. S. B. T. (2010). Intuition and reasoning: A dual process perspective. Psychological
Inquiry, 21, 313–326
Feldman, H. & Friston, K. J. (2010). Attention, uncertainty, and free-energy. Frontiers in Human
Neuroscience, 4. doi:.3389/fnhum.2010.00215.
Festinger, L. (1957). A Theory of Cognitive Dissonance. Stanford, CA: Stanford University Press.
Festinger, L., & Carlsmith, J. M. (1959). Cognitive consequences of forced compliance. The
Journal of Abnormal and Social Psychology, 58, 203–210.
Fletcher, P. C., & Frith, C. D. (2009). Perceiving is believing: a Bayesian approach to explaining
the positive symptoms of schizophrenia. Nature Reviews Neuroscience. 10(1):48-58.
Friston, K. (2003). Learning and inference in the brain. Neural Networks, 16, 1325-1352
Friston, K. (2008). Hierarchical models in the brain. PLoS Computational Biology, 4 (11),
Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews
Neuroscience, 11:127–38.
Gawronski, B., & Bodenhausen, G. V. (2006). Associative and propositional processes in
evaluation: An integrative review of implicit and explicit attitude change. Psychological
Bulletin, 132, 692-731. doi: 10.1037/0033-2909.132.5.692
Gawronski, B., & Bodenhausen, G. V. (2015). Theory and explanation in social psychology.
New York: Guilford Press.
Gawronski, B., Bodenhausen, G. V., & Becker, A. P. (2007). I like it, because I like myself:
Associative self-anchoring and post-decisional change of implicit evaluations. Journal of
Experimental Social Psychology, 43, 221–232. doi: 10.1016/j.jesp.2006.04.001
Gentner, D. (2016). Language as cognitive tool kit: How language supports relational thought.
American Psychologist, 71, 650-657.
Gentner, D., & Smith, L. A. (2013). Analogical learning and reasoning. In D. Reisberg (Ed.), The
Oxford handbook of Cognitive Psychology (pp. 668-681). New York, NY: Oxford
University Press.
George D, & Hawkins J. (2009). Towards a Mathematical Theory of Cortical Micro-
circuits, PLoS Computational Biology, 10, e1000532. doi:10.1371/journal.pcbi.1000532
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, 1464–80. doi: 10.1037/0022-3514.74.6.1464
Hahn, U., & Chater, N. (1998). Similarity and rules. Distinct? Exhaustive? Empirically
distinguishable? Cognition, 65, 197-230.
Helmholtz, H. (1962). Handbuch der physiologischen optic. J. P. C. Southall (Ed.), Vol. 3. New
York: Dover.
Hofmann, W., De Houwer, J., Perugini, M., Baeyens, F., & Crombez, G. (2010). Evaluative
conditioning in humans: a meta-analysis. Psychological Bulletin, 136, 390–421. doi:
10.1037 /a0018916
Hommel B., Müsseler J., Aschersleben G., & Prinz W. (2001). The Theory of Event Coding
(TEC): a framework for perception and action planning. Behav. Brain Sci. 24, 849–937.
Hsee, C., Tu, Y., Lu, Z., & Ruan, B. (2014). Approaching aversion: Negative reactions toward
approaching stimuli. Journal of Personality and Social Psychology, 106, 699–712.
Hughes, S., Barnes-Holmes, D., & De Houwer, J. (2011). The dominance of associative
theorizing in implicit attitude research: propositional and behavioral alternatives.
Psychological Record, 61, 465–496.
Hughes, S. Van Dessel, P., Smith, C. T., & De Houwer, J. (2018). A comparative analysis of
different evaluative learning procedures. Manuscript in preparation.
Hummel, J. E. (2010). Symbolic Versus Associative Learning. Cognitive Science, 34(6), 958-
965. doi:10.1111/j.1551-6709.2010.01096.x.
Jones, A., Hardman, C. A., Lawrence, N., & Field, M. (2017). Cognitive training as a potential
treatment for overweight and obesity: A critical review of the evidence. Appetite.
Advance online publication. doi: 10.1016/j.appet.2017.05.032.
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, 13, 989-996. doi: 10.1037/a0033164
Kahneman, D., Slovic, P. & Tversky, A. (1982). Judgment under uncertainty: Heuristics and
biases. Cambridge University Press.
Kawakami, K., 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.
Kawakami, K., Steele, J. R., Cifa, C., Phills, C. E., & Dovidio, J. F. (2008). Approaching math
increases math = me and math = pleasant. Journal of Experimental Social Psychology, 44,
818–825. doi:10.1016/j.jesp.2007.07.009
Kruglanski, A. W., & Gigerenzer, G. (2011). Intuitive and deliberative judgements are based on
common principles. Psychological Review, 118, 97–109.
Krypotos, A.-M., Arnaudova, I., Effting, M., Kindt, M., & Beckers, T. (2015). Effects of
Approach-Avoidance Training on the Extinction and Return of Fear Responses. PLOS
ONE, 10(7), e0131581. doi:10.1371/journal.pone.0131581
Lagnado, D. A., Waldmann, M. R., Hagmayer, Y., & Sloman, S. A. (2007). Beyond covariation:
Cues to causal structure. In A. Gopnik & L. Schulz (Eds.), Causal learning: Psychology,
philosophy, and computation (pp. 154–172). Oxford University Press.
Laham, S. M., Kashima, Y., Dix, J., Wheeler, M., & Levis, B. (2014). Elaborated contextual
framing is necessary for action-based attitude acquisition. Cognition & Emotion, 28,
1119-1126. doi:10.1080/02699931.2013.867833
Laird, J. D. (1974). Self-attribution of emotion: The effects of expressive behavior on the quality
of emotional experience. Journal of Personality and Social Psychology. 29, 475–
486. doi:10.1037/h0036125.
Lovibond, P.F. (2003). Causal beliefs and conditioned responses: Retrospective revaluation
induced by experience and by instruction. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 29, 97-106.
Mann, T. C., & Ferguson, M. J. (2015). Can we undo our first impressions? The role of
reinterpretation in reversing implicit evaluations. Journal of Personality and Social
Psychology, 108, 823-849.
Meiser, T. (2011). Much pain, little gain? Paradigm-specific models and methods in experimental
psychology. Perspectives on Psychological Science, 6(2), 183-191.
Melnikoff, D.E., & Bargh, J.A. (2018). The Mythical Number Two. Trends in Cognitive
Sciences, 22, 280-293. doi: 10.1016/j.tics.2018.02.001.
Mertens, G., Van Dessel, P., & De Houwer, J. (2018). The contextual malleability of approach-
avoidance training effects: Approaching or avoiding fear conditioned stimuli modulates
effects of approach-avoidance training. Cognition & Emotion, 32, 341-349
Metzinger, T., & Wiese, W. (2017) Philosophy and predictive processing. Frankfurt am Main:
MIND Group.
Mitchell, C. J., De Houwer, J., & Lovibond, P. F. (2009). The propositional nature of human
associative learning, Behavioral and Brain Sciences, 32, 183-198. doi:
Moors, A. (2014). Examining the mapping problem in dual process models. In J. Sherman, B.
Gawronski & Y. Trope (Eds.), Dual process theories of the social mind (pp. 20-34). New
York: Guilford Press.
Moors, A. (2016). Automaticity: Componential, causal, and mechanistic explanations. Annual
Review of Psychology, 67.
Moors, A., Boddez, Y., & De Houwer, J. (2017). The power of goal-directed processes in the
causation of emotional and other actions. Emotion Review. 9(4), 310-318.
Olson, J. M., & Stone, J. (2005). The influence of behavior on attitudes. In D. Albarracín, B. T.
Johnson, & M. P. Zanna (Eds.), Handbook of attitudes and attitude change (pp. 223-271).
Mahwah, NJ: Erlbaum.
Penny, W. (2012). Bayesian models of Brain and Behaviour. Biomathematics, 785791,
Pezzulo, G., Rigoli, F. & Friston, K. (2015). Active inference, homeostatic regulation and
adaptive behavioural control. Progress in Neurobiology, 134, 17–35.
Phills, C. E., Kawakami, K., Tabi, E., Nadolny, D., & Inzlicht, M. (2011). Mind the gap:
Increasing associations between the self and Blacks with approach behaviors. Journal of
Personality and Social Psychology, 100, 197–210. doi: 10.1037/a0022159
Proulx, M. J. (2014). The perception of shape from shading in a new light. PeerJ, 2:e363.
Ranganath, K. A., Smith, C. T., & Nosek, B. A. (2008). Distinguishing automatic and controlled
components of attitudes from direct and indirect measurement methods. Journal of
Experimental Social Psychology, 44, 386–396.
Remue, J., Hughes, S., De Houwer, J., & De Raedt, R. (2014). To be or want to be: Disentangling
the role of actual versus ideal self in implicit self-esteem. PloS ONE, 9, e108837
Sanborn A. N., & Chater, N. (2016) Bayesian brains without probabilities. Trends in Cognitive
Sciences, 20(12):883–893.
Seth, A. K. (2013). Interoceptive inference, emotion, and the embodied self. Trends in Cognitive
Sciences, 17, 565-73. doi: 10.1016/j.tics.2013.09.007.
Shanks, D. R. (2007). Associationism and cognition: human contingency learning at 25.
Quarterly Journal of Experimental Psychology, 60, 291–309. doi:
Snelleman, M., Schoenmakers, T. M., & de Mheen, D. V. (2015). Attentional Bias and
Approach/Avoidance Tendencies Do Not Predict Relapse or Time to Relapse in Alcohol
Dependency. Alcoholism-Clinical and Experimental Research, 9, 1734-1739. doi:
Spruyt, A., De Houwer, J., Tibboel, H., Verschuere, B., Crombez, G., Verbanck, P., Hanak, C.,
Brevers, D., & Noël, X. (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
Strack, F., & Deutsch, R. (2004). Reflective and impulsive determinants of social behavior.
Personality and Social Psychology Review, 8, 220-247. doi:
Taylor, C. T., & Amir, N. (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
Vandenbosch, K., & De Houwer, J. (2011). Failures to induce implicit evaluations by means of
approach–avoid training. Cognition & Emotion, 25, 1311–1330. doi:
Van Dessel, P., De Houwer, J., & Gast, A. (2016). Approach-Avoidance Training effects are
Moderated by Awareness of Stimulus-Action Contingencies. Personality and Social
Psychology Bulletin, 42, 81-93. doi:10.1177/0146167215615335
Van Dessel, P., De Houwer, J., Gast, A., & Smith, C. T. (2015). Instruction-Based Approach–
Avoidance Effects: Changing Stimulus Evaluation via Mere Instruction to Approach or
Avoid Stimuli. Experimental Psychology, 62, 161-169. doi:10.1027/1618-3169/a000282
Van Dessel, P. De Houwer, J., Gast, A., Roets, A., & Smith, C. T. (2018). Comparing effects of
AA instructions and AA training on racial prejudice. Manuscript in preparation.
Van Dessel, P., De Houwer, J., Gast, A., Smith, C. T., & De Schryver, M. (2016). Instructing
Implicit Processes: When Instructions to Approach or Avoid Influence Implicit but not
Explicit Evaluation. Journal of Experimental Social Psychology, 63, 1-9.
Van Dessel, P., De Houwer, J., Roets, A., & Gast, A. (2016). Failures to Change Stimulus
Evaluations by means of Subliminal Approach and Avoidance Training. Journal of
Personality and Social Psychology, 110, e1-e15. doi:10.1037/pspa0000039
Van Dessel, P., De Houwer, J., & Smith, C. T. (2018). Relational Information Moderates
Approach-Avoidance Instruction Effects on Implicit Evaluation. Acta Psychologica, 184,
137-143. doi:10.1016/j.actpsy.2017.03.016.
Van Dessel, P., Eder, A. B., & Hughes, S. (in press). Mechanisms Underlying Approach-
Avoidance Training Effects on Stimulus Evaluation. Journal of Experimental Psychology:
Learning, Memory, and Cognition.
Van Dessel, P., Hughes, S., & De Houwer, J. (in press). Consequence-Based Approach-
Avoidance Training: A New and Improved Method for Changing Unwanted Behavior.
Psychological Science.
Van Dessel, P., Hughes, S., De Houwer, J., & Smith, C. T. (2018). Testing an Inferential Model
of Approach-Avoidance Training Effects on Stimulus Evaluation. Manuscript in
Van Dessel, P., Smith, C.T., De Houwer, J., Hussey, I., Hughes, S., De Neve, J. & Perugini, M.
(2018). The Role of Awareness in Effects of Different Learning Pathways on Evaluations
of novel stimuli. Manuscript in preparation.
Wagner, A. R. (1981). SOP: A model of automatic memory processing in animal behavior. In:
N.E. Spear & R.R. Miller (Eds.), Information processing in animals: Memory
mechanisms (pp. 5–47). Hillsdale, NJ: Erlbaum.
Wells, G. L., & Petty, R. E. (1980). The effects of overt head movements on persuasion:
Compatibility and incompatibility of responses. Basic and Applied Social Psychology, 1,
White, G. L., Fishbein, S., & Rutstein, J. (1981). Passionate love and misattribution of arousal.
Journal of Personality and Social Psychology, 41, 56–62.
Wiers, R. W., Eberl, C., Rinck, M., Becker, E. S., & Lindenmeyer, J. (2011). Retraining
automatic action tendencies changes alcoholic patients’ approach bias for alcohol and
improves treatment outcome. Psychological Science, 22, 490–497. doi:
Wittekind, C. E., Feist, A., Schneider, B. C., Moritz, S., & Fritzsche, A. (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:
Woud, M. L., Becker, E. S., & Rinck, M. (2011). Induction of implicit evaluation biases by
approach-avoidance training: A commentary on Vandenbosch and De Houwer. Cognition
and Emotion, 25, 1331-1338. doi: 10.1080/02699931.2011.628300
Zogmaister, C, Perugini, M., Richetin, J (2016). Motivation modulates the effect of approach on
implicit preferences. Cognition & Emotion, 30, 890-911.
Pieter Van Dessel is supported by a Postdoctoral fellowship of the Scientific Research
Foundation, Flanders (FWO-Vlaanderen). Jan De Houwer is supported by Methusalem Grant
BOF16/MET_V/002 of Ghent University. We thank Andreas Eder, Yannick Boddez, Baptist
Liefooghe, and Senne Braem for their comments on an earlier version of this manuscript.
Figure 1. Schematic of the basic steps involved in the effects of actions on stimulus evaluation
according to the inferential account of evaluative stimulus-action effects.
... Commonly, AAMTs present stimuli cueing either functional or dysfunctional behavior on a computer screen (eg, alcoholic vs nonalcoholic beverages for individuals with alcohol use disorder) and require participants to use pull and push movements with a joystick to move dysfunctional stimuli away from themselves and functional stimuli toward themselves [32]. Arguably, the repeated approach and avoidance of training stimuli initiates an inferential process that results in a modification of the subjective stimulus evaluation [33]. By pushing a stimulus away, the valence associated with this avoidance reaction (ie, "This is something I have to avoid") is assumed to become associated with said stimulus. ...
... As the conscious display of emotions can be assumed to carry a valence much stronger than that of a small wrist or hand motion [50], it can be hypothesized that stress-causing beliefs can be effectively modified by AAMTs using positive emotions to pull functional beliefs toward oneself and negative emotions to push dysfunctional beliefs away from oneself. Moreover, it is of note that emotions embody specific meanings and facilitate certain action tendencies [51][52][53] that might further increase the efficacy of the intervention by providing additional information in the inferential process aimed to modify stimulus evaluations [33]. ...
... First, to move stimuli in eAAMT-SP, the training requires participants not to perform the typically used swiping motion but to enact emotional expressions, which are arguably associated with a greater valence than swiping or joystick motions. Similar to the inferential mechanisms proposed by Van Dessel et al [33], the eAAMT-SP harnesses the transformative power of emotions as the greater valence of the emotional training reaction is assumed to be transferred to the stimuli, thus modifying the evaluation of these stimuli more strongly than can be expected in a training using swiping motions. Second, our training used functional and dysfunctional stress-related beliefs as training stimuli, thus targeting potentially stress-inducing belief systems directly. ...
Background A key vulnerability factor in mental health problems is chronic stress. There is a need for easy-to-disseminate and effective interventions to advance the prevention of stress-related illnesses. App-based stress management trainings can fulfill this need. As subjectively experienced stress may be influenced by dysfunctional beliefs, modifying their evaluations might reduce subjective stress. Approach-avoidance modification trainings (AAMT) can be used to modify stimulus evaluations and are promising candidates for a mobile stress intervention. As the standard training reactions of the AAMT (swiping and joystick motion) have little valence, emotions could be incorporated as approach and avoidance reactions to enhance the effectiveness of AAMTs. Objective We aimed to evaluate the feasibility of a mobile emotion-enhanced AAMT that engages users to display sadness to move stress-enhancing beliefs away and display positive emotions to move stress-reducing beliefs toward themselves (emotion-based AAMT using sadness and positive emotions [eAAMT-SP]). We explored the clinical efficacy of this novel intervention. Methods We allocated 30 adult individuals with elevated stress randomly to 1 of 3 conditions (eAAMT-SP, a swipe control condition, and an inactive control condition). We evaluated the feasibility of the intervention (technical problems, adherence, usability, and acceptability). To explore the clinical efficacy of the intervention, we compared pretest-posttest differences in perceived stress (primary clinical outcome) and 3 secondary clinical outcomes (agreement with and perceived helpfulness of dysfunctional beliefs, emotion regulation, and depressive symptoms) among the conditions. Results The predetermined benchmarks of 50% for intervention completion and 75% for feasibility of the study design (completion of the study design) were met, whereas the cutoff for technical feasibility of the study design (95% of trials without technical errors) was not met. Effect sizes for usability and acceptability were in favor of the eAAMT-SP condition (compared with the swipe control condition; intelligibility of the instructions: g=−0.86, distancing from dysfunctional beliefs: g=0.22, and approaching functional beliefs: g=0.55). Regarding clinical efficacy, the pretest-posttest effect sizes for changes in perceived stress were g=0.80 for the comparison between the eAAMT-SP and inactive control conditions and g=0.76 for the comparison between the eAAMT-SP and swipe control conditions. Effect sizes for the secondary clinical outcomes indicated greater pretest-posttest changes in the eAAMT-SP condition than in the inactive control condition and comparable changes in the swipe control condition. Conclusions The findings regarding the feasibility of the intervention were satisfactory except for the technical feasibility of the intervention, which should be improved. The effect sizes for the clinical outcomes provide preliminary evidence for the therapeutic potential of the intervention. The findings suggest that extending the AAMT paradigm through the use of emotions may increase its efficacy. Future research should evaluate the eAAMT-SP in sufficiently powered randomized controlled trials. Trial Registration German Clinical Trials Registry DRKS00023007;
... That is, avoiding a stimulus leads to a more negative and approaching one to a more positive evaluation. According to van Dessel and colleagues' inferential account [35], these findings can be explained by a (nonconscious) inferential process where the evaluation of a certain action is transferred to the stimulus it relates to. Hence, if a certain action is "coded" as positive, the stimulus this action refers to will be evaluated as positive, which in turn will facilitate approach behavior towards this stimulus in the future (e.g., "I find myself approaching this stimulus repeatedly, so it must be positive (and shall therefore be approached in the future)"). ...
... As outlined above, AAMTs have been investigated as promising treatments for various forms of psychopathology, with, albeit, much room for further improving their efficacy in particular when used as standalone interventions. Most prominently, based on the inferential account [35], it can be argued that the finger or wrist movements performed in previous AAMTs do not represent valence with the necessary clarity/intensity. Thus, the efficacy of AAMTs might be increased by utilizing responses that clearly represent salient desired or undesired states. ...
... For this purpose, we will run a pilot study with the following aims: (a) evaluate the feasibility of the intervention and study design and (b) explore its efficacy in influencing clinical outcomes when compared to an inactive control condition and a swipe-based AAMT (swipe control condition). Additional exploratory questions are whether the specific emotions anxiety, anger, sadness, and disgust differ with respect to their effectiveness in eAAMTs and whether deviating from the common 50:50 ratio of approach vs. avoidance responses (e.g., [35,47]) in favor of a 1:4 bias towards approach responses would lead to stronger effects on the reduction of perceived stress. ...
Full-text available
Background Stress levels and thus the risk of developing related physical and mental health conditions are rising worldwide. Dysfunctional beliefs contribute to the development of stress. Potentially, such beliefs can be modified with approach-avoidance modification trainings (AAMT). As previous research indicates that effects of AAMTs are small, there is a need for innovative ways of increasing the efficacy of these interventions. For this purpose, we aim to evaluate the feasibility of the intervention and study design and explore the efficacy of an innovative emotion-based AAMT version (eAAMT) that uses the display of emotions to move stress-inducing beliefs away from and draw stress-reducing beliefs towards oneself. Methods We will conduct a parallel randomized controlled pilot study at the Friedrich-Alexander-Universität Erlangen-Nürnberg, Germany. Individuals with elevated stress levels will be randomized to one of eight study conditions (n = 10 per condition) — one of six variants of the eAAMT, an active control intervention (swipe-based AAMT), or an inactive control condition. Participants in the intervention groups will engage in four sessions of 20–30 min (e)AAMT training on consecutive days. Participants in the inactive control condition will complete the assessments via an online tool. Non-blinded assessments will be taken directly before and after the training and 1 week after training completion. The primary outcome will be perceived stress. Secondary outcomes will be dysfunctional beliefs, symptoms of depression, emotion regulation skills, and physiological stress measures. We will compute effect sizes and conduct mixed ANOVAs to explore differences in change in outcomes between the eAAMT and control conditions. Discussion The study will provide valuable information to improve the intervention and study design. Moreover, if shown to be effective, the approach can be used as an automated smartphone-based intervention. Future research needs to identify target groups benefitting from this intervention utilized either as stand-alone treatment or an add-on intervention that is combined with other evidence-based treatments. Trial registration The trial has been registered in the German Clinical Trials Register (Deutsches Register Klinischer Studien; DRKS00023007; September 7, 2020).
... That is, because they typically sample for stimuli that have positive consequences, it may be reasonable to infer most of the time that if one samples something often, it is more likely to be positive. Such a distinction mirrors work in the literature on approach-avoidance (AA) that first theorized AA effects on evaluation may be driven by a mere association between approach-positivity and avoidance-negativity (e.g., Cacioppo et al., 2003), whereas more recent work suggests it is people's inferences about their AA behavior that is critical for its impact on evaluation (e.g., Cornielle & Stahl, 2019;Van Dessel et al., 2019). Following a similar logic in the current context, this would suggest that rather than sampling behavior always predicting a positive evaluative shift, if people were in contexts in which this inference does not apply (e.g., if someone is sampling for stimuli that have negative consequences), increased sampling should not predict a positive evaluative shift. ...
... Subsequent work questioned the extent to which basic associative processes underlie these effects by highlighting the role of people's interpretation in this process (e.g., Cornielle & Stahl, 2019;Van Dessel et al., 2019). For instance, studies have demonstrated approach-avoidance effects could be eliminated if an action was not viewed as approach/avoidance behavior, or that the same action could have opposite effects depending on if it was framed as approach or avoidance (De Houwer et al., 2001;Hütter & Genschow, 2020;Markman & Brendl, 2005). ...
... However, while much of the past research demonstrating the malleability of approachavoidance effects has centered on people's interpretation of whether an action is approach, avoidance, or neither, recent theorizing has also argued that the valences of what approach and avoidance denote per se may also be malleable to interpretation (Mertens et al., 2018;Van Dessel et al., 2019). Such a possibility mirrors the findings in the current work, which manipulated sampling goals to change people's understanding of what their sampling behavior denotes. ...
Full-text available
People often have some degree of choice over the stimuli they sample and learn more about. These sampling decisions can play an important role in evaluative learning, with recent work showing that sampling a stimulus more frequently predicts a positive shift in its evaluation (Hütter et al., 2022). The current work suggests sampling does not merely have a direct effect of positivity on evaluations, but instead is malleable to people’s interpretations of their sampling behavior’s meaning. Across five experiments, participants sampled faces to interact with across a series of trials. On each trial, the sampled face was paired with a positive or negative image, and we manipulated participants’ sampling goals. Participants given a goal to sample for positivity showed a positive evaluative shift toward the faces they sampled more frequently, regardless of whether it was consistently paired with positive or negative images. Participants given a goal to sample in a balanced way tended to show a similar but weaker effect on evaluative shift. Finally, this effect was eliminated (or reversed) among participants given a goal to sample for negativity. Complementary shifts in evaluation were also observed for faces participants chose not to sample. Thus, these results highlight the role of people’s interpretations of what their sampling behavior denotes: in contexts in which sampling should not necessarily predict liking (i.e., when one’s goal is to sample for negativity), sampling a stimulus more (vs. less) often does not create positive evaluative shifts.
... However, several studies did not find an analogous mediation effect (see e.g., Dickson et al., 2016;Machulska et al., 2016;Taylor & Amir, 2012;Wiers et al., 2011). Furthermore, according to an alternative account of AAT effectsthe propositional inference account-a behavioural training is not even necessary, because knowledge about relations between stimuli and AA-related behaviours ("I approach stimulus X and avoid stimulus Y") and inferences about the evaluation of stimuli based on this relational knowledge ("I like X more than Y because I have repeatedly approached it") can be formed without behavioural training (Van Dessel et al., 2019). In fact, studies demonstrated that the mere instruction to approach or avoid a stimulus is sufficient to produce changes in explicit and implicit evaluations of that stimulus (Van Dessel et al., 2015. ...
... With respect to theoretical accounts of AAT effects, results are in line with explanations that deemphasize the role of a repeated pairings, or association formation, between stimuli and particular and highlight the role of relational knowledge acquisition during AAT phases. According to the inferential account (Van Dessel et al., 2019), AAT tasks serve the purpose to transmit knowledge about relations between specific stimuli and actions ("I approached Person X"), and the outcomes that is generated by the action ("I usually approach a person that I like"), which is used for inferential reasoning about the liking or attractiveness of training stimuli ("I like this person because I approached her"). This relational knowledge can be acquired even without behavioural training, for example, via verbal instruction (Van Dessel et al., 2015). ...
Full-text available
Approach-avoidance training (AAT) procedures were developed with the prospect that they can modify action impulses to approach or avoid specific stimuli. Research suggested that the outcome of AAT procedures is mediated by training-induced changes in implicit response tendencies. This study investigated whether AAT procedures affect implicit response tendencies because of a training of goal-related responses or due to a training of motoric actions effecting approach and avoidance. Participants in three internet-based experiments (total n = 514) were trained to approach and avoid two fictitious social groups by steering a symbolic representation of the self towards and away from group members. They alternated between the training task and a flanker-like test task that probed for training-induced changes in response tendencies consistent with the trained action or with the trained AA goal. Results demonstrated a transfer of relations between the stimuli and AA goals from training to test tasks. In contrast, relations to the motoric acts subserving these goals had no effect on implicit response tendencies. It is concluded that a relation to approach- and avoidance related goals, and not to the motoric action, were established with the AAT procedure. Implications for associative and inferential accounts of AAT effects are discussed.
... This does not require Person A to voluntarily and effortfully make this inference. Indeed, propositional processes can be automatic (De Houwer, 2018;Van Dessel et al., 2019). ...
... From a propositional perspective, increased identification may both strengthen the vicarious approach-avoidance effect, and modulate the consequences of agency. Indeed, propositional evaluative processes are thought to strongly depend on momentary goals (De Houwer et al., 2021;Van Dessel et al., 2019). The increased identification with the model, by guiding participants to process evaluatively relevant information through the eyes of the model, may therefore have two effects. ...
Social learning plays a prominent role in shaping individual preferences. The vicarious approach-avoidance effect consists of developing a preference for attitudinal objects that have been approached over objects that have been avoided by another person (model). In two experiments (N = 448 participants), we explored how the vicarious approach-avoidance effect is affected by agency (model's voluntary choice) and identification with the model. The results consistently revealed vicarious approach-avoidance effects in preference, as indicated by the semantic differential and the Implicit Association Test. Agency increased the size of the preference assessed through the semantic differential but did not significantly impact preference in the Implicit Association Test. Identification with the model had no significant impact on the vicarious approach-avoidance effect. Theoretical and practical implications of the results are discussed.
... An alternative reasoning-but yielding to consistent predictions-can also be considered with the single-process propositional model (De Houwer, 2009, 2014Van Dessel et al., 2019). According to De Houwer (2014), the endorsement and then the activation of-sometimes opposite-propositions (e.g., "I like tobacco", "I should avoid tobacco") could drive approach/avoidance tendencies toward the product. ...
Full-text available
The relationship between heaviness of use and the approach bias (i.e., stronger approach than avoidance tendencies) toward tobacco remains ambiguous at both theoretical and empirical levels. Indeed, some models of addition would formulate opposite predictions (i.e., positive vs. negative relationship) and, as it turns out, current evidence is mixed. In three studies, we investigated this relationship among smokers (relying on a continuous measure of heaviness) and compared approach/avoidance tendencies of light smokers and non-smokers (relying on group comparison). To measure approach/avoidance tendencies, we used the Visual Approach/Avoidance by the Self Task (VAAST) that visually simulates whole body movements. This task was used as irrelevant-feature version (i.e., instructions about another dimension). Heaviness of use was assessed continuously with daily cigarette use. Data were analyzed in two Integrative Data Analyses (IDAs; a kind of meta-analysis considering jointly the raw data of the three studies), thus taking into account both significant and non-significant effects (total N = 173). In our first integrative analysis (Studies 1-3), we observed an increase in the approach bias toward tobacco as a function of heaviness of use, as well as an avoidance bias among light smokers. In our second integrative analysis (Studies 2 and 3), we found that light smokers have a stronger avoidance bias than non-smokers. While the positive relationship between heaviness of use and approach tendencies toward tobacco is consistent with most addiction models, our finding on light smokers’ avoidance bias stands in sharp contrast. These findings, however, can be incorporated into general motivational models or single-process propositional models that consider the role of goal-oriented or propositional processes, respectively.
... IBR research highlights the powerful role of causal inferences and expectancies in how people interpret their environment, including the perception of their self. In doing so, it connects to a long tradition of research on placebo effects (De Houwer, 2018a;Kirsch, 2018), hypnosis (Lynn et al., 2019), therapeutic compliance (Kanter et al., 2002(Kanter et al., , 2004, instruction-based learning (Kang et al., 2022), and contemporary research on predictive processing (Chancel et al., 2022;Clark, 2016;Martin & Pacherie, 2019;Van Dessel, Hughes, et al., 2019) IBR research may help identify cases where nondeceptive placebo treatments are effective. In nondeceptive placebo studies, patients receive verbal information about placebo effects and they are explicitly told that they will undergo a placebo treatment (for a recent discussion, see Colloca & Howick, 2018). ...
Full-text available
A variety of psychological effects have been recently replicated in studies where participants merely received information describing experimental tasks, while participants experienced these tasks in studies where these effects were originally established. We argue that these successful instruction-based replication studies raise challenging questions for contemporary psychological research: (1) What does psychological science tell us about effects beyond common knowledge? (2) Does performing the experienced version of the task add to the effect, how much so, and why? (3) Should the effect be considered an experimental demand artifact? Throughout the article, we discuss methodological challenges and solutions associated with these questions. We conclude that instruction-based replication studies offer opportunities for theoretical, methodological, and empirical development in psychological science.
... Notably, basic cognitive research on ApBM in healthy volunteers has yielded results that better fit with a single-process inferential perspective than with a dual-process perspective (6). For example, training effects require conscious awareness and can sometimes be generated by instruction only, rather than requiring repeated training. ...
Full-text available
Objective Despite their potential in improving health behaviors, such as physical activity (PA), the effectiveness of interventions targeting automatic precursors remains contrasted. We examined the effects of a single session of ABC training – a personalized consequence-based approach-avoidance training – on PA, relative to an active control condition and a control condition. Methods Middle-aged US participants (N = 360, 53 % of women) either completed an ABC training (being instructed to approach PA to obtain self-relevant consequences), an approach-avoidance training (approaching PA in 90 % of trials), or a control training (approaching PA in 50 % of trials). Participants selected antecedents (e.g., “When I have little time”) in which personalized choices between PA and sedentary alternatives were likely to occur. In the ABC training only, after approaching PA, self-relevant consequences were displayed (e.g., increase in the health status of participant’s avatar). Primary outcome was self-reported PA seven days after the intervention. Secondary outcomes included choices for PA (vs sedentary) alternatives in a hypothetical free-choice task, intention, automatic and explicit attitudes toward PA. Results No significant effect of the ABC intervention on PA was observed, so as on intention and explicit attitudes. However, the ABC intervention was associated with higher odds of choosing PA alternatives in the free-choice task and with more positive automatic attitudes toward PA. Conclusions While the ABC training was not effective at improving PA, its effects on choices and automatic attitudes suggest that this intervention may still have potential. Future studies with intensive trainings and device-based measures of PA remains needed.
Full-text available
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 = 1547) 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-hour delay and unhealthy snacking in a taste test. Our results contrast with dominant (association-formation) accounts of approach-avoidance training effects 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.
Full-text available
2007 by Alison Gopnik and Laura Schulz. All rights reserved. Causal induction has two components: learning about the structure of causal models and learning about causal strength and other quantitative parameters. This chapter argues for several interconnected theses. First, people represent causal knowledge qualitatively, in terms of causal structure; quantitative knowledge is derivative. Second, people use a variety of cues to infer causal structure aside from statistical data (e.g. temporal order, intervention, coherence with prior knowledge). Third, once a structural model is hypothesized, subsequent statistical data are used to confirm, refute, or elaborate the model. Fourth, people are limited in the number and complexity of causal models that they can hold in mind to test, but they can separately learn and then integrate simple models, and revise models by adding and removing single links. Finally, current computational models of learning need further development before they can be applied to human learning.
Full-text available
Over the past decade an increasing number of studies across a range of domains have shown that the repeated performance of approach and avoidance (AA) actions in response to a stimulus leads to changes in the evaluation of that stimulus. The dominant (motivational-systems) account in this area claims that these effects are caused by a rewiring of mental associations between stimulus representations and AA systems that evolved to regulate distances to positive and negative stimuli. In contrast, two recently forwarded alternative accounts postulate that AA effects are caused by inferences about the valence of actions and stimuli (inferential account) or a transfer of valenced action codes to stimulus representations (common-coding account). Across four experiments we set out to test these three competing accounts against one another. Experiments 1-3 illustrate that changes in stimulus evaluations can occur when people perform valenced actions that bear no relation to a distance regulation, such as moving a manikin upwards or downwards. The observed evaluative effects were dependent on the evaluative implication of the instructed movement goal rather than whether the action implied a movement towards or away from the stimuli. These results could not be explained with a rewiring of associations to motivational systems. Experiment 4 showed that changes in stimulus evaluations occurred after participants passively observed approach-avoidance movements, supporting an explanation in terms of cognitive inferences.
Full-text available
Governments are increasingly adopting behavioral science techniques for changing individual behavior in pursuit of policy objectives. The types of “nudge” interventions that governments are now adopting alter people’s decisions without coercion or significant changes to economic incentives. We calculated ratios of impact to cost for nudge interventions and for traditional policy tools, such as tax incentives and other financial inducements, and we found that nudge interventions often compare favorably with traditional interventions. We conclude that nudging is a valuable approach that should be used more often in conjunction with traditional policies, but more calculations are needed to determine the relative effectiveness of nudging.
Full-text available
The aim of this review is to critically evaluate the effectiveness and candidate mechanisms of action of psychological interventions which aim to either (a) improve the capacity for self-regulatory, reflective processes or (b) reduce the impact of automatic appetitive processes, in an attempt to influence food intake and associated weight-gain. Our aim was to examine three important issues regarding each type of intervention: i) whether the intervention influenced behaviour in the laboratory, ii) whether the intervention influenced behaviour and/or body mass index in the real world, and iii) whether the proposed mechanism of action was supported by evidence. We systematically searched three commonly used databases and identified 32 articles which were relevant to at least one of these issues. The majority of studies attempted to manipulate food intake in the laboratory using associative learning paradigms, in normal-weight female participants. Most of the laboratory studies demonstrated the predicted effects of interventions on behaviour in the laboratory, but studies that attempted to translate these interventions outside of the laboratory yielded more mixed findings. The hypothesised mechanisms of action received inconsistent support across studies. We identified several limitations which may complicate interpretation of findings in this area, including heterogeneity of study methods, small sample sizes, and absence of adequate control groups. We provide recommendations for future studies that aim to develop and evaluate these promising interventions for the reduction of overweight and obesity.
Full-text available
Previous research demonstrated that instructions to approach one stimulus and avoid another stimulus can result in a spontaneous or implicit preference for the former stimulus. In the current study, we tested whether the effect of approach-avoidance instructions on implicit evaluation depends on the relational information embedded in these instructions. Participants received instructions that they would move towards a certain non-existing word and move away from another non-existing word (self-agent instructions) or that one non-existing word would move towards them and the other non-existing word would move away from them (stimulus-agent instructions). Results showed that self-agent instructions produced stronger effects than stimulus-agent instructions on implicit evaluations of the non-existing words. These findings support the idea that propositional processes play an important role in effects of approach-avoidance instructions on implicit evaluation and in implicit evaluation in general.
Full-text available
Previous research showed that the repeated approaching of one stimulus and avoiding of another stimulus typically leads to more positive evaluations of the former stimuli. In the current study, we examined whether approach and avoidance training (AAT) effects on evaluations of neutral stimuli can be modulated by introducing a regularity between the approach-avoidance actions and a positive or negative (feared) stimulus. In an AAT task, participants repeatedly approached one neutral non-word and avoided another neutral non-word. Half of the participants also approached a negative fear-conditioned stimulus (CS+) and avoided a conditioned safe stimulus (CS-). The other half of the participants avoided the CS+ and approached the CS-. Whereas participants in the avoid CS+ condition exhibited a typical AAT effect, participants in the approach CS+ condition exhibited a reversed AAT effect (i.e., they evaluated the approached neutral non-word as more negative than the avoided non-word). These findings provide evidence for the malleability of the AAT effect when strongly valenced stimuli are approached or avoided. We discuss the practical and theoretical implications of our findings.
Full-text available
The extreme version of the Whorfian hypothesis—that the language we learn determines how we view the world—has been soundly rejected by linguists and psychologists alike. However, more moderate versions of the idea that language may influence thought have garnered recent empirical support. This article defends 1 such view. I propose that language serves as a cognitive tool kit that allows us to represent and reason in ways that would be impossible without such a symbol system. I present evidence that learning and using relational language can foster relational reasoning—a core capacity of higher order cognition. In essence, language makes one smarter.
It is often said that there are two types of psychological processes: one that is intentional, controllable, conscious, and inefficient, and another that is unintentional, uncontrollable, unconscious, and efficient. Yet, there have been persistent and increasing objections to this widely influential dual-process typology. Critics point out that the 'two types' framework lacks empirical support, contradicts well-established findings, and is internally incoherent. Moreover, the untested and untenable assumption that psychological phenomena can be partitioned into two types, we argue, has the consequence of systematically thwarting scientific progress. It is time that we as a field come to terms with these issues. In short, the dual-process typology is a convenient and seductive myth, and we think cognitive science can do better.