Abstract
Gaslighting is a form of psychological manipulation that, over time, causes a victim to doubt their sense of reality, often leading to a loss of agency and emotional and mental instability. Currently, mechanistic explanations for gaslighting are rooted in unfalsifiable psychodynamic theory. We propose a theoretical framework that draws upon prediction error minimization, symbolic interactionism, attachment theory, self-verification theory, and shared reality theory to illustrate the cognitive mechanisms that allow gaslighting to occur. We hypothesize that gaslighting depends on normative social-cognitive mechanisms operating in atypical social situations. Our model assumes that (close) relationships fulfill important epistemic needs—close others shape and verify our self-views and our experience of the world. This privileged position of close others is what gives gaslighters the epistemic leverage required for gaslighting to be effective. We then apply our theoretical framework to the cycle of gaslighting and conclude by distinguishing gaslighting from other related phenomena.
Public Abstract
Gaslighting is a type of emotional abuse where someone manipulates another person into doubting their own sense of reality. Psychology lacks clear scientific explanations for how this abuse makes people feel like they’re losing touch with what’s real. In this report, we look at research from brain science and social psychology to explain what might be going on inside the minds of people who experience gaslighting. Our explanation focuses on how people learn from their experiences, and we also include ideas about how relationships and social situations can shape behavior. The goal is to offer a scientific explanation of gaslighting.
Keywords
“If I could only get inside that brain of yours and understand what makes you do these crazy, twisted things.”
“Gregory, are you trying to tell me I’m insane?”
“That’s what I’m trying not to tell myself.”
“But that’s what you think, isn’t it? That’s what you’ve been hinting and suggesting for months now, ever since. . .”
“Hum, since what?”
“Since the day I lost your broach.”
Paula then remembers that her hardship actually began with the discovery of a letter that could incriminate Gregory. Cunningly, Gregory first endorses Paula’s memory about the letter, telling her, “Yes, you’re right, that was the beginning. . .” but quickly turns the tables on her, denying the existence of the letter and claiming he was “staggered” by her hallucinations. Gregory reminds Paula that her mother was institutionalized for madness, describing her mother’s insanity in graphic detail; he concludes by saying, “now perhaps you will understand the other things about yourself. . .why I cannot let you meet people. . .” He thereby seals Paula’s fate: with no third-party witnesses, Gregory’s version of events becomes Paula’s reality. This scene from the classic film Gaslight demonstrates the essence of gaslighting; it is a form of psychological manipulation that causes targets to doubt their sense of reality, leading to emotional and mental instability and, perhaps most importantly, a loss of agency—that is, their sense of autonomy, efficacy, power, and mastery (e.g., Abramson, 2014; Calef & Weinshel, 1981; Klein et al., 2023).
In addition to forming the plot for the multi-Oscar-winning film, gaslighting has generated much theoretical interest since the 1940s. Originally, discussion was largely constrained to psychiatric and psychodynamic spheres but has more recently gained traction in psychology, philosophy, and sociology. While much of the discourse on gaslighting has occurred in the academic milieu, in the mid-2010s, the term began to frequently appear in lay dialogue. Indeed, the surge of interest in the phenomenon is evidenced by a 1,740% increase in web searches for the term in 2022, which led Merriam-Webster to choose “gaslighting” as its word of the year in 2022 (Merriam-Webster, n.d.).
How gaslighting is conceptualized has changed over the course of its intellectual history. Today, interpersonal gaslighting is often viewed as occurring when a perpetrator tries to convince a target that the target is not of sound mind. However, at different points in time and in different disciplines, what is considered essential to gaslighting has varied. In this paper, we begin by presenting a brief history of the topic of gaslighting. Our purpose here is to give readers a sense of how the term has evolved over the years and to gain a clearer understanding of the essence of the phenomenon (for a more in-depth historical review, see Klein, Wood, and Bartz, preprint). We then propose a theoretical framework that draws upon theories of learning to shed light on the cognitive and neurocomputational mechanisms that enable gaslighting to occur. 1 We hypothesize that the effects of gaslighting depend on normative social-cognitive mechanisms operating in atypical social situations, which are explainable in terms of the prediction error minimization framework. Finally, we conclude with a discussion of outstanding questions and directions for future work in this area.
A Brief History of Gaslighting
Early Psychodynamic Perspectives
Barton and Whitehead introduced the term gaslighting to the peer-reviewed literature in 1969. They viewed gaslighting as a conscious manipulative attempt to have a mentally well individual institutionalized on false pretenses for financial, emotional, or personal gain. Over the years, several case studies were published involving manipulative individuals who tried to have their family members committed to asylums on false pretenses to make tangible gains. These case studies were largely positioned as cautionary tales to alert healthcare professionals to possible cases in which family members attempt to have relatives institutionalized in psychiatric hospitals on false pretenses. These articles warn that seemingly concerned relatives may have nefarious motivations when bringing their family to the attention of psychiatrists.
By the 1980s, the appearance of case reports describing manipulative attempts to have targets institutionalized on false pretenses had largely ceased to appear, possibly because of widespread deinstitutionalization. At this time, psychodynamic researchers began to redefine gaslighting, causing a shift in the academic usage of the term. They viewed gaslighting as a conscious or unconscious attempt to convince another person that they were insane or otherwise neurotic. Psychodynamic researchers abandoned the requirement of convincing third parties that the target was insane. Instead, they highlighted attempts to convince the target that the target themself is mentally unwell. In these cases, the gaslighters’ motivations were often less tangible, and more emotional in nature. For example, Calef and Weinshel (1981) describe a case where a client gaslights their own psychotherapist in order to avoid doing the psychodynamic work required to address their neurosis. This literature appeals to the psychodynamic process of “projective identification” in which people project aspects of their own psychology that make themselves uncomfortable onto the other person. Both Dorpat (1996) and Calef and Weinshel (1981) argue that gaslighting is either a type of, or closely related phenomenon to, projective identification. Finally, in contrast to earlier theorists, the psychodynamic researchers viewed gaslighting as much more common, even ubiquitous (Calef & Weinshel, 1981; Dorpat, 1996).
Non-Academic Approaches and Stern’s Book, The Gaslight Effect
Following this period of psychodynamic research, gaslighting remained a relatively stagnant topic for almost 20 years, although the concept received attention and thus developed outside of the academy. The most significant work published during this time is the self-help book The Gaslight Effect (2007), written by Dr. Robin Stern, a psychodynamic psychotherapist. Based on her personal and clinical experience, the book is meant to help readers identify when they are experiencing gaslighting so that they can escape and recover from this kind of abuse. In contrast to earlier theorists that primarily focused on the gaslighter, Stern (2007) views gaslighting as a “tango” (e.g., p. 32), in which both the gaslighter and target are active participants. Her book includes a list of questions readers may ask themselves to determine whether they are being gaslighted. This list is frequently used in modern psychological studies as a measure of gaslighting, including such questions as, “You think twice before bringing up certain seemingly innocent topics of conversation” and “You have the sense that you used to be a very different person—more confident, more fun-loving, more relaxed.” The Gaslight Effect remains one of the most cited texts on gaslighting today.
Interdisciplinary Approaches to Study Gaslighting
Throughout the 2010s and 2020s, there has been growing interdisciplinary interest in gaslighting. Psychologists, philosophers, sociologists, and medical professionals have begun to weigh in on the topic (e.g., Abramson, 2014; Ahern, 2018; Fielding-Singh & Dmowska, 2022; Kukreja & Pandey, 2023; Riggs & Bartholomaeus, 2018; Spear, 2019, 2020; Sweet, 2019). Recently psychologists, sociologists, and other researchers have studied gaslighting in the context of romantic relationships (e.g., Graves & Samp, 2021; Klein et al., 2023; Sweet, 2019) but also in other close relationships, including parent–child relationships (e.g., Li & Samp, 2023; Riggs & Bartholomaeus, 2018). Gaslighting perpetrated by medical doctors against their patients has also been discussed, with a particular focus on the experience of female patients (e.g., Fielding-Singh & Dmowska, 2022; Sebring, 2021). Medical professionals are themselves identified as the target of gaslighting by their own professional institution (e.g., Ahern, 2018), as are academics (Rodrigues et al., 2021). These latter two examples fall under the domain of “institutional gaslighting” (i.e., when gaslighting occurs at an organizational level). The role of gender and race in gaslighting has also been discussed by sociologists and philosophers (e.g., Davis & Ernst, 2019; Roberts & Andrews, 2013; Sweet, 2019). Additionally, philosophers have proposed various distinct types of gaslighting as well as mechanisms by which gaslighting and recovery from gaslighting may operate (e.g., McKinnon, 2017; Sodoma, 2022; Spear, 2019).
In sum, the initial application of the term “gaslighting” was very narrow, but over the past several decades, its application has broadened substantially. Nonetheless, the historical and contemporary gaslighting literature has begun to converge on several key features of gaslighting, which we discuss next.
Core and Associated Features of Gaslighting
In our view, there are two core features of gaslighting: (1) an attempt by the perpetrator to convince a target that they (the target) are epistemically incompetent—that is, incapable of either grasping some (or all) aspect(s) of reality, potentially including some aspect of themselves (e.g., Klein et al., 2023; Spear, 2019) and (2) the target’s epistemic trust in the perpetrator (see Figure 1). Regarding the latter, the target needs to believe that the perpetrator is epistemically trustworthy, as this is what allows the perpetrator to develop the epistemic leverage needed to manipulate the target. Essentially, it is the target’s belief that the gaslighter is an accurate interpreter of reality that enables the target to accept and internalize the gaslighter’s view of the target (as epistemically incompetent); once this view is internalized by the target, the gaslighter may further control the target.
Core and associated features of gaslighting.
As noted, and as discussed in detail below, epistemic trust is a normative feature of typical close relationships—people, in general, are inclined to bestow epistemic trust on close relationship partners. That said, there are factors that can influence how much epistemic trust people bestow on their partners. These factors include structural features of the relationship, such as (1) the intensity of the emotional bond (e.g., Klein et al., 2023; Sarkis, 2018; Stern, 2007), and (2) differential power positions in the relationship, including financial and/or structural dependency (Klein et al., 2023; Stern, 2007). Certain target characteristics may also influence the degree of epistemic trust; these characteristics mainly center around (three) emotional dependency—for example, low self-esteem, interpersonal insecurity, as well as a history of abuse may incline targets to bestow greater epistemic trust in their close relationship partners (e.g., Abramson, 2014; Gass & Nichols, 1988; Li & Samp, 2023), and finally the internalization of harmful stereotypes (e.g., Graves & Samp, 2021; see Figure 1). We briefly note these structural and personal characteristics here, as these factors could put targets at greater risk of being gaslit and, as such, are associated features of the construct (we return to a more thorough discussion of risk factors in the Open Questions and Future Directions).
Finally, although not necessarily a core feature per se, we resonate with the early usage of the term (and the aforementioned psychodynamic literature) and believe that in most cases the manipulation of the target’s behavior is for the gaslighter’s benefit. The benefit may be for material gains (e.g., money, property; Barton & Whitehead, 1969; Bashford & Leschziner, 2015), or immaterial gains, for example, allowing the gaslighter to control or dominate the target (e.g., Calef & Weinshel, 1981; Sweet, 2019), or to avoid accountability or responsibility for their actions (Klein et al., 2023). A related point—which is still debated in the literature—is whether gaslighting is perpetrated consciously (e.g., Dumitrascu et al., 2015) or whether it can be unconscious (e.g., Abramson, 2014; Dickson et al., 2023; Spear, 2020). Although this discussion is a bit beyond the scope of this paper as it has to do with the gaslighter’s state of mind, we do not believe that the gaslighter must explicitly be aware that they are engaging in these behaviors. For example, gaslighting has been used to describe socio-cultural systems of oppression where individuals occupying privileged epistemic positions internalize stereotypes about marginalized others and subsequently devalue, ignore, and doubt marginalized others’ testimonies, leading those others to feel epistemically incompetent (Zembylas, 2024). Here, the biases at play are often unconscious and unintentional but nonetheless can benefit the gaslighter.
The Epistemic Function of Close Relationships: A Preamble to Our Theoretical Model
A fundamental assumption of our theoretical model is that, in addition to satisfying fundamental needs related to attachment and belonging, (close) relationships also fulfill important epistemic needs. Before describing our theoretical model, we briefly review the literature supporting the epistemic function of close relationships, as this literature is key to understanding the target’s susceptibility to gaslighting. These theories fall into two broad categories: theories about the role of (close) others on how we understand (1) ourselves and (2) the world/experience in general.
Close Others and Self-Understanding
The notion that others, and especially close others, influence self-understanding dates to over 100 years ago with the proposal of the “looking glass self” (Cooley, 1902) and the sociological theory of symbolic interactionism (Mead, 1934). These revolutionary ideas emphasized that the self is developed and understood in the social realm, that others’ appraisals shape our understanding of ourselves, and that individuals are created through social interactions. The notion that people internalize their relationships with significant others, which then influences their sense of self and subsequent interpersonal experiences, is also at the heart of attachment theory (Bowlby, 1969; Mikulincer & Shaver, 2003). According to attachment theory, early interactions with caregivers are used to generate mental models for close relationships that reflect how individuals expect close others to treat them (e.g., Can I rely on this person? Will this person be responsive to my needs?) and how they believe they deserve to be treated (e.g., Do I deserve to be loved? Do I deserve to be supported?). Attachment theory is one of the most prolific theories in psychology to date, and research indicates that attachment models play a key role in self-understanding as well as perception, cognition, and behavior more broadly.
The role of close others in self-understanding is the basis of Andersen and Chen’s (2002) theory of the relational self. Because of our fundamental need for attachment and belonging, they argue that the self and personality are largely shaped by our experiences with close others. According to Andersen and Chen (2002), the self, essentially, is an associative network of relationships, consisting of linkages between representations of significant others and representations of the version of the self we are when we are with each significant other. When a significant other is activated (either by their actual physical presence or reminders), that activates the affect, goals, and behaviors associated with that relationship, biasing the self “to be” who we are in that relationship. A detailed discussion of this theory is beyond the scope of this paper; our point here is to illustrate how the self is shaped by close others. In fact, according to Andersen and Chen (2002), the self is inherently relational—there is no self that is devoid of our relational experiences.
In addition to shaping our self-views, close others play a key role in verifying and reinforcing we who we think we are over time. In particular, research on self-verification indicates that we seek correspondence between how we see ourselves and how others see us. This is evidenced by the observation that people prefer to interact with others who see them as they see themselves, rather than others who see them differently than they see themselves (Swann & Read, 1981). In fact, this need for correspondence between self-views and others-views is so strong that people even prefer to interact with others who see them in a negative light if that is how they see themselves (i.e., low self-esteem). This phenomenon is true for new relationship partners as well as for established close relationship partners. It is thought that such verification gives us a sense of coherence, control, and predictability (Swann, 1990; Swann & Buhrmester, 2012). This theory underscores how fragile the self-concept can be (am I who I think I am?), and that people critically rely on others for continual verification of who they think they are.
(Close) Others and Information Processing About the World
The role of others in our understanding of the world is central to the foundations of social psychology—in particular, research on conformity and social influence 2 more broadly highlight the powerful role of our social conspecifics on information processing. Early research by Asch (1955) demonstrated the powerful influence of others on our perception and behavior by showing that students at elite universities would endorse obviously incorrect beliefs (about the relative length of various lines) when under the pressure of group consensus. Some participants even insisted that their perceptions matched the incorrect beliefs that they had endorsed (Asch, 1955). These foundational social psychology studies are examples of what Cialdini (2001) calls informational social influence. Now, over a century of research shows that humans are highly attuned to the attitudes of others (Latané, 1981), and look to others for information about what is good, right, proper, valued, and beautiful (Heine, 2010). Importantly, social influence is not just about outward obedience. For example, functional imaging research shows that conformity is accompanied by corresponding changes in the subjective computation (i.e., neural encoding) of value, suggesting internal endorsement and acceptance of outwardly imposed beliefs and evaluations (e.g., Zaki et al., 2011).
In addition to telling us what to believe, research on shared reality shows that our social conspecifics play a key role in our experience of the world more generally. According to this line of research, humans are fundamentally motivated to create a sense of shared reality with other people—that is, a perceived commonality of inner states such as feelings, beliefs, preferences, and values. In the same way that others verify our sometimes-tenuous self-concept, experiences of shared reality are important because they help us to understand what is “real” (Etcherhoff et al., 2009; Etcherhoff & Schmalbach, 2018). When we experience shared reality with another person—when we feel that we are “on the same page”—that fundamentally alters our experience, making the experience more intense (Boothby et al., 2014), and more real (Boothby et al., 2017).
Recently, it has been proposed that shared reality is a distinct type of social influence that simultaneously satisfies both epistemic and relational motives (Echterhoff & Higgins, 2025). Experiences of shared reality can occur in many kinds of relationships (with friends, acquaintances, and even with strangers), they are especially powerful in romantic relationships because shared reality acts as a means of verifying one’s beliefs about the world while also facilitating feelings of interpersonal closeness, thereby satisfying both epistemic and relational needs (Rossignac-Milon et al., 2021). Indeed, as Rossignac-Milon et al. (2021) noted, disruptions to this joint sense-making can explain, in part, why losing a romantic relationship can be one of the most distressing life experiences. In essence, others and especially close others ground our experience; without others, we are like an unanchored boat adrift at sea.
Finally, social baseline theory (Beckes & Sbarra, 2022; Beckes & Coan, 2011) also emphasizes the critical role of others in our experience. According to this theory, humans have evolved to be fitted to a social ecology, and thus our physiological and cognitive systems are at a baseline (i.e., natural state) when we are around others. This social ecology is important because it reduces the metabolic cost of perception and action through the reduction of vigilance, as it is assumed that processes of avoiding harm and seeking opportunity are distributed amongst the group. Social baseline theorists cite a wide variety of empirical studies that demonstrate how social others impact information processing, which generally point toward cognition being more effortless and less vigilant when people are in their social ecology (Beckes & Sbarra, 2022; Beckes & Coan, 2011). Thus, others not only serve to ground our experience of ourselves and the world but they also help us “carry the load” by reducing the cognitive effort required to navigate life. As discussed in detail later, this theory is of particular relevance to gaslighting because it specifically argues that one effect of being in the social ecology is that people relax their defenses and become less vigilant (because they can “share the load”). While this phenomenon is thought to be adaptive (resource conservation), it is a vulnerability when one is in a relationship with a gaslighter.
In sum, these theories all share the view that others and especially close others have a privileged epistemic position: because we rely on close others for our sense of ourselves and of reality, we need to trust them. This trusting relationship facilitates learning, we adopt beliefs (about the world and about ourselves) from those that we trust and share reality with. The privileged position close others occupy in our epistemic abilities is a normative feature of interpersonal relationships, but gaslighters abuse this privilege for their own gain. Specifically, gaslighters develop epistemic leverage by using this need for trust, as well as other aspects of the relationship (e.g., dependency, intimacy, authority, etc.), to (further) manipulate the target. Once a bond has been formed between the gaslighter and target, the target will develop an expectation that the gaslighter is trustworthy, generous, and kind; the target will then give more weight to the gaslighter’s views and beliefs compared to the views of psychologically distant others when determining their own views and beliefs. At its core, gaslighting is an abuse of this epistemically privileged position that close relationship partners possess.
A Theoretical Framework for Studying the Phenomenon of Gaslighting
We now turn to our theoretical model, which aims to understand the cognitive mechanisms that allow gaslighting to occur; again, our focus is on the target’s experience. Our model is based on theories of learning and, specifically, the prediction error minimization (PEM) framework. We believe this framework is ideal for illustrating the phenomena of gaslighting as gaslighting is, essentially, a learning process by which the target comes to view themselves, via the gaslighter’s repeated actions, as epistemically incompetent. Before describing our theoretical model, we briefly overview of PEM theories and review relevant terminology.
Overview of PEM Theories
The origins of PEM have been situated as far back as the 1860s, when Herman Von Helmholtz described perception as a type of inference (Clark, 2013). Since then, components of PEM have developed in an interdisciplinary fashion in computer and cognitive science (e.g., Buckner & Garson, 2019; McClelland et al., 1986; Rumelhart et al., 1986). The PEM framework has also been used to model visual processes such as binocular rivalry (Hohwy et al., 2008) and end-stopping (Rao & Ballard, 1999), as well as other perceptual processes such as interoception (Allen et al, 2019) and audition (Schröger et al., 2015). Outside of perception, PEM has been used to model emotion (Barrett, 2006), attention and learning (e.g., Holland & Schiffino, 2016), classical conditioning (Rescorla & Wagner, 1972), and even schizophrenic hallucinations (Fletcher & Frith, 2009; Sterzer et al., 2018). The most extreme PEM theorists view it as a potential unifying theory of cognition, perception, and action. While PEM has contributed to our understanding of a wide variety of phenomena, it remains to be seen if it might be applied more widely to social-cognitive processes (De Bruin & Michael, 2017, 2021).
Next, we provide a brief explanation and example of how PEM works. (A complete overview of PEM and its empirical basis is outside of the scope of this paper, but there are several reviews that outline various theories within PEM and ground them in empirical studies for readers who would like more detail, e.g., Bastos et al., 2012; Clark, 2013; Friston, 2010; Karnick & Brick, 2024; Smith et al., 2021; Spratling, 2017; Teufel & Fletcher, 2020.) The PEM framework posits that the body simulates an integrated multimodal representation that combines the external world along with the body’s internal state. This representation is a top-down prediction based on prior sensory experience and is iteratively compared to incoming sensory information (i.e., information that influences behavior from the bottom-up). In other words, the mind anticipates what will be happening in the body and the world and compares that prediction to input from the various sensory receptors. Through each comparison cycle, prediction errors (i.e., mismatches between expected and experienced bodily states) can occur. A prediction error can be thought of as the mind making an incorrect guess about what will be occurring in the body or the world. For example, as illustrated in Figure 2, imagine you typically place your morning coffee on a coaster on the right side of your desk. One day while cleaning your desk you unwittingly move the coaster to the other side of the desk; the next morning, you sit down at your desk and place your coffee on the coaster; the coffee is thus in an unusual spot (Figure 2(1)). After reviewing your to-do list, you reach for the coffee without looking, but you grasp air; this bottom-up bodily experience (grasping air) generates a prediction error (i.e., your expectation of where the coffee cup is located has been violated; Figure 2(2)).
Prediction error and resolution by active and perceptual inference.
The task of the brain/mind is to resolve this prediction error; there are two ways it can be resolved. The first is to move the coffee and coaster back to the expected spot; this is an example of active inference (Figure 2(3a)). 3 This action ensures that the actual state of the world corresponds to the predicted (i.e., modeled) state of the world. In other words, when the mind experiences that the state of the body or the world is “not quite right,” the mind can resolve this issue by changing the state of the body or the world to bring it in line with how the mind thinks it ought to be. The second way to reduce prediction errors is by perceptual inference, that is, adjusting priors (i.e., updating the internal simulation to match incoming sensory data; Clark, 2013). For example, as depicted in Figure 2(3b), updating your mental model of the location of the coffee cup (from the right side of the desk to the left). There are two additional points worth noting about PEM. First, prediction errors can vary in magnitude—that is, the size of the difference between the expected and actual input (using the example above, how far the cup is from the expected position). Second, prediction errors can vary in their degree of influence on inferences—that is, signal strength, or weight, which reflects the degree of confidence in the prediction error signal. Returning to the coffee cup example, if one were in a dim room, one might have less confidence in one’s predictions about the location of the coffee cup and, in turn, less confidence in the prediction error signal. Because of variation in signal weight, small deviations from expectations can still be influential if the signal is weighted strongly and, conversely, large deviations from expectation can have little influence if they are not weighted strongly. This concept of signal weight will be important when we apply PEM to gaslighting.
We believe that PEM has epistemological features that make it well-suited to explain gaslighting, a uniquely epistemic form of abuse. The influence of top-down priors on bottom-up sensory processing can illuminate how gaslighters capitalize on their leverage to influence their target’s thoughts, beliefs, and behavior. Gaslighting involves the induction of a particular type of perceptual inference in the target, namely, that the target is epistemically incompetent; this belief (prior) then leads the target to down-weigh their other priors (i.e., give them less signal weight) and, in this way, gives the gaslighter greater epistemic leverage over the target and the target’s experience of reality. Over the course of repeated experiences of being gaslit, targets come to question whether they can accurately model reality and, consequently, begin to reduce their confidence in their other beliefs about the world and about themselves; this down-weighting of other priors gives the gaslighter leverage as the down-weighing allows prediction errors caused by the gaslighter to exert more influence (increased strength/weight) over the target, which fuels the cycle of gaslighting. We next map out how gaslighting can be explained within the PEM framework. As noted, we believe gaslighting arises as a function of typical social-cognitive mechanisms operating in atypical social situations (i.e., when the close other has malevolent intentions). We thus begin by explaining how the relevant, typical social-cognitive mechanisms operate in close relationships.
Of note, while gaslighting has been identified in multiple contexts, we focus on close, romantic relationships as they are most likely to possess the features that allow gaslighting to be effective. In particular, close, romantic relationships are marked by a high degree of interdependence (Proulx et al., 2007), which, as noted, should promote epistemic trust. Moreover, close, romantic relationships are particularly potent for learning given the amount of time people spend together; notably, this time together is typically associated with decreased time spent with others (who could challenge the would-be gaslighter). Finally, much, but not all, of the existing literature on gaslighting has focused on romantic relationships. That said, we do not believe romantic relationships are unique (at least in the context of gaslighting); rather they are extreme exemplars of interdependence. As such, our model should apply to gaslighting in any interpersonal relationship that has a high degree of epistemic trust (see Discussion).
Setting the Stage: Relationship Initiation and Establishing Higher-Order Priors
A key step in the development of romantic and other close relationships is the experience of epistemic trust, which allows for the intimacy that characterizes such relationships, as well as other processes like the unconditional provision of support, etc. (e.g., Holmes, 1981, 1991). The presence of epistemic trust is an enabling constraint (i.e., a requirement) for gaslighting to occur. That said, the development of epistemic trust is not a part of gaslighting, but rather a precondition.
Within the PEM framework, the development of romantic (and other close) relationships involves weighing various forms of evidence to establish a high-order prior (i.e., belief) that one’s new partner is a trustworthy and appropriate relationship partner. Thus, each affiliative act a new partner makes at this stage contributes to the predictive cognitive representation of the new partner’s trustworthiness. In PEM terms, at the outset of a relationship someone may have a weak, neutral-to-slightly-positive view (i.e., prior) of their new partner. The new partner’s behaviors (e.g., punctuality, generosity, kindness) will be compared against the lower-level expectations that constitute this prior. When the behavior is better than expected, the violation will cause an appetitive prediction error, which typically results in an updated view of the new partner as more trustworthy, loving, etc. than expected. Conversely, when the behavior is worse than expected, these violations cause an aversive prediction error, which typically results in an updated view of the new partner as less trustworthy, etc. than expected. In PEM terms, this is how typical new relationships (including those that degrade into gaslighting) unfold: people build up priors about the trustworthiness of new relationship partners.
The Powerful Influence of Close Others on Prediction Errors
Before turning to the gaslighting cycle, we pause to reflect again upon the epistemic features of close relationships and, specifically, the notion that we turn to others to make the world more understandable and predictable (Rossignac-Milon & Higgins, 2018; Rossignac-Milon et al., 2021), as this has direct implications for the processing of prediction errors. As noted, several influential theories suggest that when we form a bond with another person, we develop epistemic trust in that person; we depend on them and relax our own personal responsibility in understanding our experiences. By definition, to have epistemic trust in another person means that you will “take their word for it,” in other words, you will learn from them. In PEM terms, this reduction of responsibility would be reflected in the reduced weighting of prediction errors coming from sources other than the trusted other and thus, greater reliance on priors (this idea is also echoed in the social baseline theory argument that people are less vigilant when they are in their social ecology; Beckes & Sbarra, 2022; Beckes & Coan, 2011). At the same time, because of epistemic trust, people should weigh prediction errors stemming from the trusted other more heavily than prediction errors generated by non-close others or the environment. Thus, prediction error signals stemming from trusted others should have greater signal strength, and, consequently, be more influential in learning. This typical feature of close relationships facilitates the epistemic leverage that gaslighters use to manipulate targets.
Of note, while the influence of close others on our beliefs about ourselves and the world is bolstered in a general sense (given more weight), there is a special case in which close others’ influence is actually diminished—namely, when it comes to our beliefs about that close other. When one’s romantic partner acts in a manner that is inconsistent with a prior expectation, the resulting prediction error will have a weaker influence on belief updating than an equivalently large (i.e., equivalent deviation from expectation) prediction error generated by a stranger (Park et al., 2021). For example, when confronted with an expectancy violation (e.g., witnessing someone stealing money from a tip jar), a target would be less likely to update their beliefs about a romantic partner than they would be to update their beliefs about, say, a co-worker. This feature of close relationships may have particularly powerful implications in gaslighting contexts. Once a strong prior expectation of the gaslighter’s value and trustworthiness is established, this prior will be particularly resistant to change in the face of contradictory evidence. This feature of close relationships (the influence of close others on prediction error processing) sheds light on why targets do not come to quickly view the gaslighter as untrustworthy even when the gaslighter displays (increasingly) untrustworthy behavior. See Table 1 for a summary of the consequences that close others have on the weighting of different sources of prediction errors.
Consequences of Close Others on the Weighting of Different Sources of Prediction Errors.
Note. Prediction errors stemming from non-close others regarding the environment, targets, or close other will be down-weighted, resulting in greater reliance on priors. In contrast, predictions errors stemming from close others regarding the environment and/or target will give greater weight, this corresponds to a sense of dependence on or epistemic trust in the close other. When close others generate prediction errors about their own character or behavior, these prediction errors are down-weighted, maintaining preexisting perceptions about close others and protecting the relationship.
The Gaslighting Cycle I: Inaccurate Resolution of Prediction Errors
In the above section, we offer a mechanistic account of how the stage is set for gaslighting; we now turn to the actual cycle of gaslighting. Once epistemic trust is established the gaslighter takes advantage of their privileged epistemic position to manipulate the target’s sense of reality. This manipulation occurs through (1) generating salient prediction errors, followed by (2) offering explanations for these prediction errors that suggest, either implicitly or explicitly, that the target is epistemically incompetent (e.g., Dorpat, 1996). For example, a gaslighter may cause confusion by manipulating the target’s environment by hiding objects (e.g., “where are my keys? I swore I put them right here. Am I going crazy?”). In PEM terms, these gaslighting behaviors (hiding keys) generate prediction errors, which need to be resolved. In gaslighting relationships, they tend to be resolved in ways that favor the gaslighter and erode the target’s belief in their own epistemic abilities, thereby perpetuating the cycle of gaslighting (see Figure 3(1). Returning to the misplaced keys example, the most accurate way for a target to resolve this prediction error would be to change their positive view of the gaslighter to a negative one (e.g., “I am certain I left my keys here. There is no other explanation for their absence; the perpetrator must have taken them”). However, the gaslighter offers an alternative explanation; namely, that the target is epistemically incompetent (e.g., “you’re a little crazy right now”) and at fault for the misplaced keys (e.g., “it [your craziness] is making you so forgetful!” If the target adopts this auxiliary explanation, this would be an example of perceptual inference). 4
Prediction-error minimization framework of gaslighting.
We chose a simple example to illustrate our point; most cases of gaslighting do not involve physically manipulating the victim’s environment, but rather manipulating their social and emotional environment (e.g., Klein et al., 2023). For example, in response to one’s partner setting a boundary, a gaslighter might begin to express intense sadness, which, in turn, causes the target to question whether they were right to explicitly state this boundary in the first place (see Klein et al., 2023). Essentially, the gaslighter moves the conversation away from the partner’s boundaries to the gaslighter’s emotions. As another example, the gaslighter tries to prevent the target from going out on a planned night on the town with friends by reminding the target that they are too irrational to budget properly (see Figure 3(2a)). Here, the gaslighter generates a prediction error (“it’s a bad idea to go out”) and then influences the target to respond to this prediction error by updating their beliefs about their own cognitive and epistemic abilities, rather than their beliefs about the gaslighter. These events cause the target’s self-perception to change, as well as their behavior to conform to the gaslighter’s desires (see Figure 3(2d) and (e)). While the above examples differ from the misplaced keys example, the same underlying psychological mechanism holds. Namely, gaslighters create a situation that violates the target’s expectations, and then offer auxiliary explanations that center on the target’s epistemic incompetence. Of note, in gaslighting relationships, these deviations from expectations are typically aversive—that is, the actual input is (1) more negatively valanced than the expected input and (2) stressful, two factors that create more potent opportunities for learning (Diederen & Fletcher, 2021; Robinson et al., 2013).
Why would the target adopt the gaslighter’s alternative explanation? Why does the partner (gaslighter) hold a more privileged epistemic position than the target themselves? First, it is important to keep in mind that, as noted, close relationships fulfill fundamental needs related to belonging and security and, for these reasons, people are highly motivated to see their close relationships and close relationship partners in the most positive light, even if that involves some liberties with the truth (Murray et al., 1996; Murray & Holmes, 2017). As Park et al. (2021) noted, people are especially likely to maintain priors when prediction errors are about close others. Here, it is worth noting that this need to see the relationship and partner in a positive light is even more intense for those being gaslit because they are often socially isolated and, consequently, rely solely on the gaslighter to meet their attachment/relational needs.
The second reason why targets are biased to adopt the gaslighter’s explanation is because these events typically start out as relatively minor, “one-offs,” and grow in intensity (and frequency) over time. So, when the target is confronted with a singular event (lost keys, accusations of financial irresponsibility, “you didn’t see what you think you saw”), accepting the alternative hypothesis that they’ve made some form of misjudgment in the moment is arguably easier—and less psychologically threatening—than updating the view of the trusted other as untrustworthy or, even worse, malevolent. Indeed, people are especially likely to accept auxiliary hypotheses for prediction errors when they allow us to maintain our priors about close others (Kim et al., 2020). As an aside, even if the target were to update their view of the gaslighter as untrustworthy, that would also involve accepting some epistemic incompetence as they would have to admit that they misjudged the gaslighter’s trustworthiness. In fact, this captures the “double-bind” that Dorpat (1996) discusses as a process relevant to gaslighting.
Finally, a related point to emphasize is that the gaslighting process unfolds overtime. Over repeated events, the gaslighter creates prediction errors and then offers explanations for these prediction errors—explanations that center on the target’s epistemic incompetence. The gaslighter weakens, bit by bit, the target’s prior that they are sane. As we turn to next, the slow but steady creep is a necessary condition as that is what creates the cognitive conditions (in PEM terms) that allow for successful gaslighting and the erosion of agency in the target.
The Gaslighting Cycle II: Erosion of Agency
A key proposition of our theoretical model is that repeated experiences of gaslighting lead to an erosion of agency—that is, the target’s sense of power, autonomy, efficacy, and mastery (Bakan, 1966; Wiggins, 1991). From the PEM perspective, agency depends on having finely tuned predictive models of the world (i.e., priors; see Farb et al., 2015). This erosion of agency comes about via the gaslighter’s repeated accusation of epistemic incompetence (and auxiliary explanations), which makes targets come to question their grasp on reality. As the target’s faith in their own epistemic abilities is diminished (by adopting the kinds of auxiliary hypotheses described above), the target becomes increasingly dependent on the gaslighter for their sense of / grasp on reality and, thus, even more motivated to view the gaslighter as trustworthy (see Figure 3(1)). Gaslighting essentially creates a dilemma wherein the target’s need to maintain their sense of individual agency and their faith in their own ability to accurately perceive and respond to reality is pitted against their need for attachment, belonging, and the epistemic certainty and meaning they derive from their close relationships. Normative cognitive mechanisms, as well as motivated relationship maintenance processes, ultimately influence this dilemma to be resolved in a manner that challenges the target’s epistemic competence. The gaslighter essentially creates an environment that leads targets to question their rationality. Over time, when faced with repeated gaslighter-generated prediction errors, the target’s priors as well as subsequent prediction errors become less precise (akin to reaching for one’s coffee cup in a dark room, as per the earlier example); that is, targets come to question both their priors (“am I sane?”) and their predictions errors (“maybe I did not leave my keys here?”), leading to a dissociated state wherein they are increasingly stripped of their agency. By degrading the target’s epistemic self-trust, the gaslighter is positioned as an epistemic authority, which increases the gaslighter’s influence on the target’s cognitive processing (Echterhoff et al., 2009; Knausenberger et al., 2019). 5
Another feature of gaslighting relationships that contributes to the target’s erosion of agency is that gaslighters often isolate their targets from family, friends, and colleagues (Klein et al., 2023; Sarkis, 2018; Sodoma, 2022). This social isolation further augments the target’s propensity to resolve prediction errors by questioning their own epistemic competence. This is because socially isolated (vs. connected) individuals have fewer opportunities to reaffirm their epistemic competence through interactions with (non-manipulative) others. Moreover, socially isolated (vs. connected) individuals have fewer opportunities to interact with others who explicitly call the trustworthiness of the gaslighter into question, thus challenging the target’s strongly held belief that the gaslighter is trustworthy. Indeed, affirming targets through affective empathy is a proposed “antidote” to gaslighting (Sodoma, 2022). Finally, socially isolated individuals have fewer opportunities to meet epistemic and belongingness needs through interactions with individuals other than their gaslighter, which would both reduce the gaslighter’s leverage and attenuate the target’s motivation to view the gaslighter positively.
Insight and Recovery: Accurate Resolution of Prediction Error
We now turn to the question of recovery. While the cycle of gaslighting could, in theory, go on indefinitely, recent evidence suggests that many targets do escape and recover from gaslighting relationships (Klein et al., 2023). Recovery is not an essential component of gaslighting, but we discuss recovery here as it can also be explained in terms of the resolution of prediction errors and, thus, bolsters our use of the PEM framework in understanding gaslighting. To the extent that gaslighting relies on the inaccurate resolution of prediction errors, recovery from gaslighting should involve the accurate resolution of prediction errors, ultimately resulting in updating the target’s beliefs about their epistemic competence and the gaslighter’s trustworthiness. There are two parts to recovery: (1) insight and (2) activities that reaffirm the self as an epistemic agent.
Insight can arise from a singular catalytic event (e.g., the gaslighter says the target will never succeed in their career that they are incompetent and their coworkers will figure this out some day; but the target gets a big promotion), or from the mounting of PEs that undermine what the gaslighter is imposing on the target (an accumulation of positive job appraisals). In PEM terms, insight involves decreasing epistemic trust in gaslighters (decreasing signal weight of gaslighter/not seeing gaslighter as a reliable source) and augmenting the prior that they (the target) are epistemically competent. We believe that insight is very rare without some outsider intervention—that is, another (trustworthy) individual who points out the abuse and who can meet (at least some of) the target’s epistemic and relational needs. In PEM terms, the new person becomes the epistemically trusted other, who can then strengthen the target’s prior about their epistemic competence.
Following insight, targets often engage in a number of activities that promote recovery. Of note, many of the activities associated with recovery are described as re-embodying (Klein et al., 2023), and the PEM framework offers a mechanistic explanation for why. Within the PEM framework, selfhood arises primarily from predictions of self-generated sensory signals (Farb et al., 2015). Thus, activities that help targets re-establish a sense of self are also those that require heightened interoceptive awareness. This heightened interoceptive awareness allows for self-generated interoceptive prediction error signals to exert more influence on higher order representations of the self as an embodied agent in the world. By engaging in activities that require paying attention to interoceptive signals, such as mindfulness, yoga, physical activity, art, and music, targets can begin to reintegrate various levels of their hierarchical self-representation, thereby re-establishing self-concept clarity, self-esteem, and well-being. Similarly, activities such as journaling and creative expression cause recovering targets to focus on emotionally laden signals, which is itself a form of interoception (Klein et al., 2023). Lastly, time spent with close others, especially those engaging in validating affective empathy, who are responding to targets in a non-manipulative manner, will help retrain recovering targets’ self-world model to be more accurately attuned to the world via the same normative mechanisms that enabled the gaslighting to occur in the first place (Hailes, 2022; Klein et al., 2023; Sodoma, 2022). Finally, time spent with others allows targets to practice engaging in healthy relationships, thereby re-establishing adaptive and accurate priors for modeling the self, close others, and the world.
Discussion and Analysis of Our Theoretical Model
In sum, drawing upon the PEM framework, we offer a mechanistic account of the cognitive processes underlying the phenomenon of gaslighting. Our objective is to offer a mechanistic model of gaslighting that explains why close others are able to make a person question their own experience of themselves and the world. First, the target’s strong prior that the gaslighter is trustworthy is created, which results in the gaslighter’s epistemic leverage. Second, with the gaslighter having secured a position to manipulate the target’s environment, the gaslighter creates prediction errors and offers false explanations for these prediction errors that center directly or indirectly on the target’s epistemic incompetence. The target is biased toward resolving these prediction errors by accepting the gaslighter’s explanations, rather than questioning the trustworthiness of the gaslighter. Third, over time and after repeated gaslighting, the target increasingly comes to doubt their epistemic competence and, consequently, comes to question their priors as well as the prediction errors they experience; this, in turn, makes them even more epistemically dependent on the gaslighter. We have focused primarily on the erosion of the target’s psychological sense of agency, but within the PEM framework, there is a continuity between psychological and bodily agency (also see Krol et al., 2020). In line with this, evidence suggests that gaslighting targets also experience distortions in their sense of being an embodied self (Klein et al., 2023).
Why the PEM Framework?
We believe gaslighting is, essentially, a form of learning. We have drawn upon the PEM framework to make our case, as PEM theories represent the cutting edge of theories on learning, cognition, and perception (e.g., Barron et al., 2020; Clark, 2013; Corlett et al., 2022; Diederen & Fletcher, 2021; Hohwy, 2020; Subramanian et al., 2022; Zhang et al., 2020). Indeed, learning theories have been useful theoretical frameworks for studying abuse in the past. For example, the theory of learned helplessness has historically been used as a framework for understanding the psychological impact of physical abuse in close relationships (Walker, 1979). The ability for PEM theories to generate formal falsifiable hypotheses has been utilized to successfully explain perceptual processes such as vision and interoception (De Bruin & Michael, 2017, 2021). As noted, there is very little empirical work on the phenomena of gaslighting, though decades of theoretical work indicate that gaslighting is distinct from other related phenomena. Our proposal not only meets explicit calls to apply PEM to social-cognitive processes (De Bruin & Michael, 2021) but also generates specific testable hypotheses that will enable researchers to quantify the phenomenon of gaslighting and to understand the psychological mechanisms that enable gaslighting to occur. Moreover, the combination of the aforementioned social-cognition research programs with the neurocomputational theory of PEM has the potential to generate a variety of future research directions and testable hypotheses that are relevant not only to gaslighting but also to psychology more broadly. That said, we have not committed to any particular formalization of PEM. Many different precise formalizations of the general PEM framework have been proposed, and some may be better suited to our model than others.
PEM Versus Other Relationship Science Explanations: A Different Level of Analysis
The psychological aspects of gaslighting are consistent with many theories in social psychology and relationship science. Gaslighting could be explained in terms of Harold Kelley’s classic covariation model (1973) by framing gaslighting as a process through which the gaslighter manipulatively influences the attributions of the target. For example, the gaslighter induces intense emotions in the target and then attempts to convince the target that their feelings or responses are “crazy” as they are different from how others would act (i.e., low consensus) and furthermore that such “crazy” thoughts and reactions are how the target consistently behaviors across a wide variety of situations (i.e., high consistency, low distinctiveness). Thus, the target would be influenced to attribute their unpleasant emotions to themselves rather than the gaslighter.
As we acknowledged at the outset motivations to maintain our relationships that have been identified by relationship scientists likely cause targets to continue to view gaslighters in a positive light, even after the gaslighter has become abusive. Indeed, for any set of behaviors as complex, dynamic, and temporally extended as gaslighting one could come up with a list of dozens of relevant psychological processes. We are not doubting the veracity or utilities of these and other relationship science theories. We acknowledge the possibility that these psychological phenomena may be at play in gaslighting, but we are not attempting to explain the psychological phenomena of gaslighting in terms of other psychological phenomena. Here, we are concerned with proposing a different level of analysis, namely a computational mechanism. The principles of PEM could also potentially provide a mechanistic account for attribution, motivated cognition, and many other psychological phenomena without undermining the veracity of or replacing any of these observed effects. The utility of PEM is to bridge the gap between psychological phenomena and brain function by providing a computational framework that can (1) be implemented by the brain and (2) explain complex psychological experiences and behavior.
Testing the Proposed Model
Our model can be verbalized in terms of the consequences of close others on the weighting of different sources of prediction error (see Table 1). First, in general, (most) prediction errors stemming from close others should have a higher weight than prediction errors from other sources. That is, individuals should be more likely to incorporate information coming from close others versus non-close others when updating most prior expectations. For example, imagine someone approaches you and says, “aliens have landed on earth!” The degree to which this statement causes you to update your view of reality will be greater if the person is a close other rather than an acquaintance or stranger. Second, the weight of close other generated prediction errors about the close other’s own character or behavior should be decreased. For example, imagine you saw your close other receive the wrong change from a cashier (e.g., a $20 when they should have received $5) and they put the extra money in their pocket without saying anything to the cashier. In this situation, you would be more likely to attribute your close other’s behavior to external sources (e.g., mistake, lack of attention) than to update your expectations about their character. Conversely, people should be more suspicious of an acquaintance who pocketed the change, as we have less experiences on which to base our view of the acquaintance. This is because our prior expectations about our close other’s character and behavior will be more precise because those predictions are based on a larger sample of behaviors (and because of motivated biases to see our partners in a positive light). Thus, our prior expectations about close others will be more change resistant compared to prior expectations about an acquaintance or stranger because our expectancies are based on more data points.
There are several ways researchers could test these propositions. The least resource-intensive way would be to find existing data sets that are consistent with the propositions and assess how well existing PEM models explain the data. Additionally, lab-based experiments could be set up to assess the behaviors implied by all three propositions. Of note, these experiments could be adapted for electroencephalogram (EEG) methods to determine whether predicted behaviors are linked to neural signatures of PEM (e.g., N400) in a theory-consistent manner (i.e., are neural signatures of prediction error upweighted and down-weighted consistently in the proposed conditions).
In addition to evaluating the presence of these normative mechanisms, it will also be important to study whether the psychological consequences of gaslighting depend on these mechanisms. That is a slightly more difficult task, as it would require manipulating accusations of epistemic incompetence by close others and observing the consequences of those accusations, which would require working with a sensitive population (i.e., survivors of abuse). Carefully conducted observational studies could test for differences in the strength of prior expectations in gaslighting survivors vs. the general population. Additionally, interventions geared at recovery that make use of the proposed neurocognitive mechanisms could be developed, that, if successful, would validate our proposition. For example, cognitive tasks aimed at reestablishing survivor’s view of themselves as epistemically trustworthy could be designed. Again, EEG could be used to test whether the influence of top-down predictions in general has been eroded in victims of gaslighting (compared to controls). Given the caution that is required to study such a population, this would likely be best after the basic research has been conducted on the general population, in order to help develop interventions for survivors of gaslighting.
Limitations
The proposed model has limitations. First, as described in the previous section, while individual components of our model could be verified with correlational research, or tested experimentally, it may not be feasible to wholistically verify the entire model aside from using computational simulations. Second, we have not committed to any particular formalization of PEM. Many different precise formalizations of the general PEM framework have been proposed, and some may be better suited to our model than others. Third, gaslighting is, obviously, a multi-faceted construct; our model is focused entirely on the psychological unraveling of the target and does not make predictions about how the gaslighter will respond to different situations or how the gaslighting process impacts the psychology and motivations of the gaslighter. Finally, our model does not incorporate risk factors for gaslighting victimization (or perpetration), although we do discuss such risk factors below.
Open Questions and Future Directions in the Study of Gaslighting
Setting aside the specific tenants of our theoretical model, there are several broader questions that warrant investigation. We end with a discussion of what we believe to be the most pertinent. We begin by discussing question that are directly related to our theoretical model and then address broader questions for this emerging field of inquiry.
Risk Factors for Gaslighting Victimization
Our model assumes that gaslighting depends on normative cognitive mechanisms, and thus, anyone can be vulnerable. However, one can image that there could be some degree of individual variation in vulnerability and here we discuss a few such candidates. We preface this section by noting that this review is concerned with the target’s experience and the mechanisms that allow being gaslit to come about. Thus, we focus primarily on target-related factors (interested readers are referred to footnote 6 for a discussion of factors related to gaslighting perpetration).
In our view, any characteristic that makes people more dependent on their partner should make them more vulnerable to gaslighting victimization. Attachment anxiety is an obvious candidate. Anxiously attached individuals have a strong desire for closeness but chronic concerns about abandonment (Mikulincer & Shaver, 2016); because of this, anxiously attached individuals are more likely to subordinate their own needs over their partner’s needs. Moreover, attachment anxiety has been linked to low agency (Bartz et al., 2004; Bartz et al., 2015). These factors (high dependency, low agency) should make these individuals more vulnerable to gaslighting victimization. For the same reason, low self-esteem (e.g., Abele & Hauke, 2018) and rejection sensitivity (e.g., Meehan et al., 2018) may also be a risk factor for gaslighting victimization, to the extent that these factors augment dependency and/or are associated with diminished agency. Low self-concept clarity—that is, the extent to which the self-concept is clear, consistent, and stable—may also be a vulnerability factor for gaslighting victimization—less because of their interpersonal insecurity and more because they rely on others to verify their sense of self (e.g., Emery et al., 2018). Intriguingly, recent research suggests that those low in self-concept clarity are more vulnerable to the rubber hand illusion, suggesting pre-existing weak priors about the self (Krol et al., 2020). Of course, certain mental health conditions may be risk factors for gaslighting victimization—especially those related to Cluster B personality disorders (e.g., borderline)—given the interpersonal insecurities that characterize these disorders. That said, while popular science blogs and books often suggest that gaslighting perpetration is related to Cluster B personality disorders, to our knowledge, these claims have not been verified in peer-reviewed studies (e.g., Sarkis, 2018).
Importantly, situational and structural factors that augment dependency may also be risk factors. As noted at the outset, differential power positions in the relationship arising from financial and/or structural dependency can be risk factors that augment the gaslighter’s epistemic leverage. Indeed, the role of power in gaslighting has generated much interest. Theoretical perspectives on gaslighting have focused primarily on power differentials arising from social inequities (e.g., gender, race) or differing social/role status (e.g., psychiatrists and patients). While the role that power plays in gaslighting has been studied, this work has only looked at one type of power, namely, relationship dependence power (Graves & Samp, 2021). Finally, as noted earlier, social isolation is a risk factor for gaslighting severity because socially isolated individuals have fewer opportunities to (1) reaffirm their epistemic competence through interactions with (non-manipulative) others, (2) interact with others who explicitly call the trustworthiness of the gaslighter into question, and (3) meet their epistemic and belongingness needs through interactions with individuals other than their gaslighter.
While the proposed model does not presently differentiate between profiles of those who may be at greater or lesser risk of being gaslit, once these individual differences are established the model’s parameters could be modified to apply to such cases. For example, say there is a robustly identified trait-like feature that puts individuals at risk of being gaslighted (for example, a disorder or personality trait). This susceptibility could be explained in terms of one of, or some combination of the following: (1) The stability or strength of these individual’s prior expectations in general, (2) the particular makeup of these individual’s relationship-specific prior expectations, (3) how these individuals weight prediction errors in general, and (4) the strength of the effect of close other’s on the weighting of predictions errors among these individuals. These are just four examples of parameters in the present model that could explain individual variation in susceptibility (or resistance) to gaslighting.
Gaslighting in Non-Romantic Relationships
In the discussion of our theoretical model, we focused exclusively on gaslighting in close, romantic relationships. An open question is whether and how our model would apply to other gaslighting contexts, for example, gaslighting in parent–child relationships, friendships, the workplace, medicine, and institutional betrayal. We believe that many aspects of our model would apply to these other situations. Epistemic trust is a core feature of many of these alternative relationship contexts—indeed parents and other authority figures (doctors, religious leaders) are particularly relevant as these individuals are thought to “know best,” which could establish a set of priors that facilitate gaslighting. Essentially, we believe that the same principles should apply so long as there is a relationship of epistemic trust between the target and gaslighter. That said, there may be important moderator variables. For example, one could imagine weaker effects in less intimate relationships if those relationships lack the degree of epistemic trust found in more intimate relationships. Relatedly, more intimate relationships typically have a more powerful impact on sense of self (Aron & Aron, 1997; Rusbult et al., 2009). Thus, the notion of epistemic trust and epistemic leverage presented here may be a useful framework for understanding how gaslighting occurs between individuals who have less frequent contact and less familiarity than romantic partners. For example, when a gaslighter is acting on behalf of an organization that the victim has epistemic trust in (e.g., a university department, a hospital, the government, etc.) the gaslighter may leverage the victim’s trust in the institution that they (the gaslighter) represent. In these cases, gaslighting in the absence of intimacy is explainable. Future studies may also investigate how individuals that represent larger organization may effectively borrow the epistemic leverage of their organizations, and how this may enable gaslighting.
What neurobiological mechanisms underlie gaslighting?
Cognitive neuroscientists have linked the computational principles of PEM to specific biological implementations in the brain. Prediction error signal and minimization is most strongly linked to dopamine. While a full discussion of the role of dopamine in PEM theories is outside the scope of this paper, dopamine is believed to signal reward prediction errors (i.e., dopamine release is associated with unexpected rewards), or encode signal weight (i.e., subjective probability) of outcomes more broadly (for a more in-depth discussion, see Steinberg et al., 2013; Friston et al., 2014).
While dopamine is the most widely discussed neurotransmitter in the PEM framework, there is evidence for modulatory roles of other neurotransmitters as well. Most relevant to this discussion is oxytocin, a neurohormone that plays a major role in attachment bonds (e.g., Carter, 1998; Walum & Young, 2018). Recent research showed that oxytocin (vs. placebo) administration significantly attenuated prediction error signals (Crucianelli et al., 2019). 7 This finding, though preliminary, suggests that oxytocin may not only play a role in the formation of close relationships but also in fulfilling the epistemic function of these relationships via its attenuating effect on prediction error signals (also see Quattrocki & Friston, 2014).
Intriguingly, recent research also points to the role of oxytocin in people’s tendency to resolve prediction errors via the gaslighter’s explanations. Within PEM frameworks, interoceptive signals are viewed as forming the basis of emotionally laden intuitions (e.g., Critchley & Garfinkel, 2017; Farb et al., 2015). As noted above, preliminary evidence suggests that oxytocin attenuates interoceptive prediction error signals (e.g., Crucianelli et al., 2019; Quattrocki & Friston, 2014). This oxytocin-induced attenuation of interoceptive prediction error signals should then blunt the influence of “gut-feelings,” or emotional information on learning (i.e., updating of priors); in this way, opening the potential for other sources of information (e.g., the testimonies of close others, including gaslighters) to exert more influence on the updating of higher-order mental representations. Oxytocin’s influence on prediction errors may also facilitate the continuation of gaslighting. Research indicates that oxytocin administration can specifically blunt learning about the trustworthiness of social others, thereby serving to maintain pre-existing impressions of others (whether positive or negative; Ide et al., 2018). Thus, oxytocin may simultaneously (1) reduce reliance on one’s own emotional responses to the world, in favor of information generated by others, while also (2) reducing the ability to evaluate the trustworthiness of social others as a source of information. This research is a good illustration of the generalizability and applicability of the PEM framework for advancing our understanding of a psychological phenomenon and, specifically, the interplay between the behavioral, psychological, and neurological levels of analysis.
In addition to oxytocin, the stress hormone cortisol may play an important role in gaslighting given the stressful nature of gaslighting, and the impact of stress and, specifically, cortisol on learning. For example, both acute stress and post-traumatic stress disorder have been shown to reduce prediction error signals, thereby inhibiting learning (Carvalheiro et al., 2021; Ross et al., 2018). Thus, in early stages of gaslighting, acute stress during gaslighting interactions may bias targets away from learning negative information about their gaslighters while learning positive information about the gaslighter during non-stressful interactions is left unaffected. In later stages of gaslighting, chronic stress may also contribute to the maintenance of gaslighting behavior by preventing the development of new patterns of behavior.
Conclusions
In sum, the phenomenon of gaslighting has intrigued scholars for decades, yet there is little by way of a mechanistic account that can explain the phenomenon. We believe gaslighting arises as a function of typical social-cognitive mechanisms operating in atypical social situations, that is, the tendency to prioritize the testimony of close others combined with an epistemically manipulative close other. In most healthy (adult) relationships, the influence of close others on beliefs is (1) reciprocal, (2) primarily implicit, and (3) non-manipulative. Close other’s corrective feedback is typically delivered in good faith. In such cases, this process likely results in beliefs that are more accurate, or at least adaptive within one’s social context. Furthermore, the reciprocity of this process in healthy relationships promotes both partners finding a middle ground. Gaslighting, on the other hand, is characterized by a unidirectional and manipulative use of this social-cognitive mechanism. Overtime, gaslighting leads to the erosion of agency and fuels a cycle of dependency, which the gaslighter leverages.
We drew upon the PEM framework to model this phenomenon. We propose that in PEM terms, gaslighting involves the induction of a prior in the target that they themselves are not a reliable epistemic agent via repeated violations of explanations coupled with manipulative explanations from the gaslighter. We hope that our proposal can not only provide a starting point of a specific falsifiable theory of the mechanisms that underlie gaslighting, but, so long as much of theory is demonstrated to be accurate, it will also contribute to the emerging general theory of mind and behavior that is PEM.
Positionality Statement, Citation Statement, and Constraints on Generality Statement
The concept of gaslighting has its origins in a Western cultural context, as do all three authors of the present work. These origins have undoubtedly impacted the broader gaslighting literature and the author’s perspective on the topic. That said, while early work on gaslighting was fairly homogenous in terms of subjects, authors, and disciplinary perspective (primarily psychodynamic theory), more recent work is considerably more interdisciplinary and culturally diverse. For example, we cite communications theorists studying gaslighting among LGBTQ+ individuals, research on the personality correlates of gaslighting among an Australian sample, conducted by an international team of researchers, philosophical investigations of how trans-individuals experience gaslighting, clinical case studies investigating how parents gaslight their transgender children, critical theory analyses of the gaslighting of African American teachers, and several other studies conduct by diverse scholars and/or on diverse populations. We draw upon both the early work and more recent literature in this review. More specific statements regarding positionality, citations, and generality follow below.
The authors of this manuscript are diverse in gender identity, national origin, academic rank, and sub-disciplinary expertise. The first author is a Jewish Canadian male, enrolled in a PhD program at the time of writing, his academic background is in psychology, cognitive science, and philosophy; the second author is a White, American cis-gender female and Associate Professor in the Teaching-Stream in a Department of Psychology at a Canadian University, with considerable expertise in the psychology and neurobiology of learning and memory; and the third authors is a Caucasian female and Full Professor in the Department of Psychology at a Canadian University, with considerable expertise in the psychology and neurobiology of attachment, prosocial behavior and social cognition. The first author has received significant training and mentorship from the other two authors. The contents of this manuscript are informed by the authors’ subject matter expertise, broader scientific and scholarly expertise, and lived experience.
Regarding the cited material, the academic backgrounds of the cited scholars cited are diverse. The present work draws heavily on behavioral neuroscience, social psychology, cognitive psychology, philosophy, sociology, psychiatry, and critical theory. There is also an element of temporal diversity in the cited works, with the oldest works being over 50 years old at the time of writing. Additionally, there is significant national and gender diversity with regard to the authors cited.
The cited work draws on populations beyond the typical WEIRD populations of psychology. While many cited empirical studies draw on convenience samples (e.g., university students), even these studies draw upon samples from various countries (e.g., USA, Italy, Canada, Germany). Of note, the quantitative studies cited sample individuals from similar G7 countries but also samples from India, China, and Australia. Participants in qualitative studies and case reports varied greatly in terms of gender identity, sexual orientation, national origin, age, social-economic status, and even time period. Thus, we believe the present work attains a high degree of generalizability by integrating across these diverse literatures.
Footnotes
Acknowledgements
The authors would like to acknowledge Amy Gregory, Dr. Ian Gold, Dr. Kristina Tchalova, Dr. Melanie Dirks, Andrew Burcar, Molly Feffer, Melissa Shemirani, Nikita Guarascio, Cendrine Blais, and Thera Phillips for their helpful comments and assistance. This research was funded in part by a grant from the Natural Sciences and Engineering Council of Canada (to J.A.B.)
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: We acknowledge the support of the Fonds de Recherche du Québec - Société et Culture (FRQSC) Team Grant (#305471; DOI attribué : https://doi.org/10.69777/305471) awarded to J.A.B., an internal Social Science and Humanities Development grant funded by McGill’s SSHRC Institutional Grant (File No.: 633-2015-0042) awarded to J.A.B, and a Fonds de Recherche du Québec - Société et Culture (FRQSC) Doctoral Fellowship awarded to W. K. (https://doi.org/10.69777/365189).
