|Note: Some of this material is derived from Kihlstrom, J. F., S. Mulvaney, et al., "The Emotional Unconscious", in E. Eich, J. F. Kihlstrom, G. H. Bower, J. P. Forgas and P. M. Niedenthal, Cognition and Emotion (Oxford University Press, 2000).|
What we can call the cognitive unconscious consists, first of all, in automatic cognitive processes that generate conscious percepts, memories, and thoughts; and then percepts, memories, thoughts, and other cognitive states -- states of knowing that are inaccessible to phenomenal awareness but nonetheless influence the person's experience, thought, and action. But cognition isn't all there is to the mind. But cognition is not all there is to mental life. At least since the 19th century, many psychologists and philosophers of mind have classified mental life into three broad faculties, including emotion and motivation as well as cognition.
This idea began
with Christian Wolfe (1679-1754) the German philosopher who
brought the term psychology into common use, and who
classified the mind into the facultas cognoscitova
(knowledge) and the facultas appetiva (desire). Moses
Mendelsohn (1729-1786) added affect (emotion) to the list. The
tripartite classification of mental faculties was consolidated
by the philosopher Immanuel Kant (1724-1804), who wrote in his
Critique of Judgment (1790) that
"there are three absolutely irreducible faculties of the mind, namely, knowledge, feeling, and desire".
By this statement, Kant meant that emotion and motivation each have an existence that is in some sense independent of cognition, and of each other. This position which contrasts with the idea, still popular in psychology, that emotional and motivational states are cognitive constructions -- i.e., that they are beliefs about what one feels. Kant's idea of three independent mental faculties is reflected in what Hilgard (1980) has called "the trilogy of mind": cognition, the mental representation of reality through perception, attention, learning, memory, and thought; emotion, the subjective experience of arousal, pleasure, and displeasure, and their expression in behavior; and motivation, the activation of behavior and its direction toward a goal. All three of these mental states -- knowledge, feeling, and desire; cognition, emotion, and motivation -- enter into the determination of behavior.
We usually think of the cognitive, emotional, and motivational processes that underlie action in terms of conscious mental states. We are aware of what we think, feel, and desire; and we act accordingly. But we have already concluded that cognitive states, such as percepts, memories, and thoughts can affect behavior outside of awareness. If it makes sense to say that cognitive states of perception, memory, and thought can be unconscious, or implicit, it makes sense to ask whether emotional states can be unconscious, or implicit, as well.
We know that there's a cognitive unconscious: is there an emotional (affective) unconscious, and a motivational (conative) unconscious as well? What evidence is there for implicit emotion, and implicit motivation, analogous to implicit cognition?
What might implicit emotion look like? One possibility is that it resembles Sigmund Freud's vision of unconscious mental life. Freud (in the General Introduction to Psycho-Analysis, 1916-1917/1963; and the New Introductory Lectures on Psycho-Analysis, 1933/1964) argued that our conscious mental lives are a product of unconscious primitive affects and drives which he labeled as eros (sex) and thanatos (aggression). So, the classic Freudian defense mechanisms (most of which were actually described by Freud's daughter, Anna) all are designed to render the person unaware of his or her true emotions. In reaction formation, we profess love but really feel hate; in displacement, we declare hatred of one person, when we really hate another person entirely; in intellectualization and rationalization, our behavior is stripped of all of its emotional connections entirely. Freud argued that our true emotions were manifest in our behavior, even if they were not represented in consciousness. The defenses render us unaware of our conflictual feelings and desires, and we're not aware of our defenses either -- otherwise, if you think about it, the defenses wouldn't work. In this way, people are affected by emotional or motivational states of which they are not consciously aware. Psychoanalysis, in Freud's view, permits people to gain insight into their true feelings and desires, and to recognize the unconscious determinants of their conscious experience, thought, and action. But this insight has the character of an inference, rather than an introspection. We are rarely conscious of our true feelings or desires at the time we act on them.
Note: Nobody knows precisely where the iceberg metaphor comes from. It is commonly ascribed to Freud, but in fact he did not use it, at least in print. There's actually a concordance to the Standard Edition of the Complete Psychological Works of Sigmund Freud, just like there is for the Bible, locating and counting every word (except a, and, and the that appears there, and "iceberg" isn't in the list. He or she who identifies the first person to use the iceberg metaphor is going to make a great contribution to the history of psychology.
Suppes and Warren (1975), two mathematical psychologists, have proposed a mathematical model of the kinds of transformations involved in the Freudian defense mechanisms. They begin with a propositional representation of unconscious affect -- of an actor, an action, and an object (x) of the action -- as in the prototypical Freudian emotional self-disclosure:
And, of course, I hate my father, because he is a rival for my mother's affections. And I fear my father, too, because he arouses castration anxiety in me. Now, that's a lot of baggage for a little kid to carry around, and so he must repress his love of his mother, and his hatred and fear of his father, render it all unconscious, find an acceptable substitute, and give that substitute conscious representation. All of this is accomplished by a set of Freudian defense mechanisms.
Suppes and Warren then go on to show that some 44 different
defense mechanisms, including all those included in the
standard list, can be produced by just eight transformations
applied to the actor, the action, or the object, alone or in
combination -- e.g., the transformation of self to other, of
an action into its opposite, or from object x to
-- a glib and vulgar Freudianism, to be sure, but one which nicely illustrates the essential process by which the defense mechanisms are held to operate so as to render the actual emotional and motivational determinants of our behavior inaccessible to phenomenal awareness.
Of course, one does not have to embrace the whole conceptual panoply of classical (or even neofreudian) psychoanalysis -- the division of the mind into id, ego, and superego, the theory of infantile sexuality, the stages of psychosexual development, repression, and whatnot. And one certainly doesn't have to trace all of one's emotional life to primitive sexual and aggressive instincts. The kinds of emotions whose conscious representations are at issue can be represented by the everyday concepts of folk psychology, as reflected in the affect circumplex, Murray's list of needs, and similar ideas. In the present context, what is important in Suppes and Warren's (1975) model is that it suggests two basic ways in which the emotional unconscious can be expressed -- (1) when the original emotion is represented consciously but we are unconscious of the source of that emotion, as in displacement and projection; and (2) where the emotion itself is denied conscious representation, as in reaction formation, intellectualization, and denial.
But really, we want to move beyond glib and vulgar Freudianism -- not just because we don't want to be either glib or vulgar, but because Freudian psychoanalysis is a non-starter. There is no evidence for any specific proposition of psychoanalytic theory. To make things worse, some philosophers of science have argued that Freudian theory is simply untestable -- and thus a poor choice to begin a scientific analysis of the motivational and emotional unconscious. And there's no evidence that psychoanalysis is a particularly effective form of therapy (if it were, that might give us reason to think that the scientific propositions of Freudian theory are correct).
What we want to do, then, is to seek evidence for the
emotional and motivational unconscious in the context of
established psychological theory. So, the relevant
evidence will come in three basic forms:
The first thing to be said is that psychologists have long believed that certain emotional and motivational states were automatically activated by certain stimulus conditions.For example, C.S. Sherrington used the term interoception to characterize the sensory mechanisms involved in homeostatic regulation of the body's internal environment.
Walter B. Cannon made a similar point with regard to the flight or fight response. From his point of view, the physiological arousal that accompanies emotional states of fear and anger occurs automatically in response to a threatening stimulus.
More recently, of course, Bargh's auto-motive model of behavior asserts that goals are automatically activated in response to the appearance of certain environmental stimuli.
for the automatic generation of emotion comes from the
extensive social-psychological literature on the mere
exposure effect (Zajonc, 1968). In his first
experiment, now a classic in social psychology, Zajonc
recruited subjects for an experiment on language learning, and
then taught them how to pronounce "Chinese" ideographs or
"Turkish" words (which weren't actually Chinese or Turkish,
but sort of looked like they were). The items were
presented for varying numbers of trials. Later, the
items were (falsely) identified as adjectives, and the
subjects were asked to guess whether they meant "something
good" or "something bad", on a 0-6 scale.
result was that
the evaluation of the connotative (emotional) meaning of the
item varied, depending on the number of times the item had
been studied. New items, that had not been studied at
all, were given slightly negative ratings, while items that
had been presented 10 or 25 times got slightly positive
ratings. The effect was subsequently replicated using a
wide variety of stimulus materials.
The mere exposure effect (MEE) is, simply that repeated exposure to a stimulus increases likability, even in the absence of substantive contact (such as any serious analysis). The exposure-likability relation, in turn, was thought to be mediated by familiarity -- a sort of priming effect. Familiarity, then doesn't breed contempt -- in fact, we prefer that with which we are familiar. In social psychological theory, the MEE is of interest because it is a purely situational effect -- people seem to prefer whatever they encounter frequently in the environment.
In most studies of
the MEE, the stimuli -- the primes were presented
supraliminally, in clear view, so that the subjects
consciously perceived the items as they were presented.
But in a later study by Kunst-Wilson and Zajonc (1980), the
primes were presented subliminally - -via a tachistoscope, for
only 1 msec, unmasked. Not surprisingly, on a later
test, the subjects did not recognize any of the primes as
having been presented before. If something has not been
consciously perceived, it's not going to be consciously
remembered. But Kunst-Wilson and Zajonc still got a
significant mere exposure effect. As we'll see later,
this effect has been replicated many times.
In fact, a meta-analysis by Bornstein (1989) found that the magnitude of the mere exposure effect was significantly greater with subliminal than with supraliminal stimuli. Apparently, affective judgments are influenced by perceptual fluency, which in turn is enhanced by the priming effects of the subject's initial exposure to the material. When subjects consciously remember the prior exposure, they appear to correct their preference ratings accordingly; but when the initial exposure is subliminal, so that subjects do not consciously perceive (much less consciously remember) it, subjects are unable to engage in discounting, resulting in a stronger effect on preference judgments (Bornstein, 1992; Bornstein & D'Agostino, 1992, 1994; see also Klinger & Greenwald, 1994).
But Zajonc had another explanation, which he expressed in the title of a later paper: "Preferences Need No Inferences". Zajonc doubted that there was anything like this complex cognitive activity going on. He argued that the MEE was simply an automatic response to the prior presentation, whether supraliminal or subliminal. Preferences need no inferences because they are not mediated by cognition -- they're independent of cognition.
Zajonc's paper represented the opening salvo in what I have come to think of as an affective counter-revolution in psychology. The first revolt against behaviorism came from cognitive science and cognitive psychology, where theorists argued that cognitive processes mediated between environmental stimuli and organismal response; and that behavior was mediated by cognitive representations such as perceptions, expectations, goals, and memories. Some cognitive psychologists used the term "cognitive" broadly, to refer to all sorts of mental states. But others took the position that emotional and motivational states were themselves the products of cognitive construction. For example, Schachter and Singer (1962) famously argued that feelings of happiness and anger, and by extension all sorts of other emotional states as well, were mediated by the subject's cognitive appraisal of the situation in which physiological arousal took place. People felt physiologically aroused, but that undifferentiated physiological arousal didn't turn into an emotion until it has been explained; and the nature of the explanation determined the subject's emotional feeling state.
asserted that "preferences need no inferences" -- that there
were at least some emotional states, like pleasure and
displeasure, happiness and sadness, that required no cognitive
mediation, and were elicited directly by appropriate stimuli,
in a more or less reflex-like matter -- which is to say, they
are activated automatically. Zajonc went so far, in
another paper, to argue for "the primacy of affect" -- that
the first response to a stimulus is affective in nature, and
that cognitive appraisal is an afterthought. Zajonc
engaged in an extended debate with Richard Lazarus, a clinical
psychologist at UC Berkeley, who argued for the traditional
position of "the primacy of cognition" -- that, at the very
least, we could employ cognitive processes to regulate our
We can see the influence of this position
in some contemporary approaches to moral reasoning, such as
the theory promoted by Jonathan Haidt, emotion is more
powerful than cognition, and that our intuitive feelings,
generated automatically, drive our reasons. This
position is quite popular now, and was brought into the
popular press by David Brooks, a columnist for the New
York Times. In a 2011 book entitled The Social
Animal, Brooks brought the automaticity juggernaut
together with the affective counter-revolution to argue that
our most important thinking takes place automatically and
unconsciously, and that automatically generated emotions
provide the grounds for rational decision-making.
Setting aside this more or less ideological debate, the subliminal mere exposure effect is of interest as an example of implicit perception. Subjects do not consciously perceive the briefly presented primes, but the subliminal MEE shows clearly that they were perceived, outside of conscious awareness. Put another way, the subliminal MEE is an effect of perception on experience, thought, and action, in the absence of conscious awareness of the percept.
So, one element of the emotional unconscious is that a person may be consciously aware of his or her emotional state, yet unaware of its source in current or past experience. This is the sort of thing the Roman poet and epigrammatist Martial had in mind, when he wrote (Epigrammata, Book 1, No. 32, freely translated by the 17th-century English satirical poet Thomas Brown; see Hayward, 1927):
Non amo te, Sabidi,
nec possum dicere quare:
hoc tantum possum dicere,
non amo te.
I do not love you Dr. Fell,
but why I cannot tell;
But this I know full well,
I do not love you, Dr. Fell.
It is also what Breuer and Freud (1893-1895/1955) had in mind in their Studies on Hysteria:
Hysterics suffer mainly from reminiscences. [But In] the great majority of cases it is not possible to establish the point of origin by a simple interrogation of the patient, however thoroughly it may be carried out... principally because he is genuinely unable to recollect it and often has no suspicion of the causal connection between the precipitating event and the pathological phenomenon.
For Breuer and Freud's unconscious (or repressed) memories, we can substitute implicit ones. Viewed in this way, a conscious emotional state may serve as an index of implicit perception or memory. Even though the emotional states themselves are consciously experienced, these phenomena of implicit perception and memory deserve to be counted in the emotional unconscious.
In a classic demonstration of what we have now come to call spared implicit memory in organic amnesia, Claparede (1911/1951) pricked an unsuspecting Korsakoff's syndrome patient with a pin hidden in his hand -- an event that caused her quite a bit of distress. Claparede subsequently left the room, and returned after the patient had regained her composure. Upon questioning, she failed to recognize Claparede, and had no recollection of the unfortunate incident that had just transpired between them. Nevertheless, she refused to shake his hand. When asked why, she replied, "Sometimes people hide pins in their hands". The story illustrates the phenomenon of source amnesia familiar in studies of memory. But assuming that the prospect of shaking hands made the patient nervous as well, it suggests a dissociation between conscious awareness of an emotional state, which she probably experienced, and conscious recollection of the origins of that state in experience, which she evidently did not.
In a case of hysterical somnambulism reported by Pierre Janet, a French psychologist, "Madame D." suffered a breakdown after some men jokingly deposited her drunken husband on her doorstep and announced that he was dead. Afterward, the woman had no conscious recollection of the event. But whenever she passed by her front door, she froze with terror. Moreover, she complained of dreams in which her husband was brought home dead. Here again, we have an emotional state -- terror, and distress about the dreams -- and emotional behavior -- freezing, as well as the nightmares; but no awareness of why they occur.
A third example comes from a case of phobia for running water reported by Bagby (1928). The patient had no memory of the circumstances under which this intense emotional reaction had been acquired. However, the mystery was solved when the patient was visited by an aunt who said, as an aside, "I have never told". It turned out that the patient, as a child, had gone on a picnic with the aunt; and, as children are wont to do, she had disobeyed instructions, strayed into a nearby creek, and become trapped under a waterfall. The child was rescued by the aunt, who promised to keep her transgression a secret. Apparently, memory for the incident was lost -- perhaps due to a process like repression or dissociation, perhaps merely to childhood amnesia or ordinary forgetting -- but the phobia remained solidly entrenched. In this case, the symptom appears to be an implicit memory for an incident lost to explicit memory.
The evidence for emotion as implicit memory is not limited to anecdote. There are several formal studies that also demonstrate that emotional responses can persist in the absence of corresponding awareness of the past circumstances in which they were acquired.
A study by Johnson, Kim, and Risse (1985) made use of Zajonc's "mere exposure" effect on preferences. They exposed Korsakoff syndrome patients (who are amnesic as a result of damage to diencephalic structures) and controls to unfamiliar Korean melodies. Some melodies were played only once during the study phase, while others were played five or ten times. Later, the subjects were played these melodies, and other Korean melodies that were entirely new, and asked to indicate which they preferred. As expected from the mere exposure effect, both Korsakoff patients and controls preferred old over new melodies, although there was no effect of the number of exposures given to the old tunes. However, the patients showed greatly impaired levels of recognition, compared to the controls. Thus, exposure affected the amnesic patients' preference judgments, and index of emotional response to the melodies, even though the patients showed impaired memory for the exposure trials.
A second study by Johnson et al. provided subjects with more substantive contact with the stimulus materials. The same amnesic and control patients who served in the melodies study were presented with pictures of two male faces, accompanied by fictional biographical information that depicted one individual positively (the "good guy") and the other negatively (the "bad guy"). When asked whom they preferred, control subjects always chose the face that had been paired with the positive information; and they were always able to state that their judgment was made on the basis of the accompanying descriptive information. The amnesic patients also showed a strong (though not unanimous) preference for the "good guy"; however, they were able to recall only a negligible amount of the biographical material presented at the time of study. Again, some aspect of emotional response -- liking of persons instead of melodies -- was altered by information presented during the study phase, in the absence of conscious recollection of this information.
The dissociation between acquired emotional preferences and explicit memory was confirmed by an experimental case study with patient Boswell, who had been rendered densely amnesic following a case of herpes encephalitis (Damasio, Tranel & Damasio, 1989). Despite a profound inability to recognize people, it had been noted that Boswell would go to a particularly generous staff member if he wanted something. In the experiment, Boswell had an extended series of positive, negative and neutral encounters, respectively, with three different confederates. Upon subsequent questioning, Boswell was unable to recall anything about any of the people and never indicated that they were familiar in any way. Nonetheless, when asked on a forced-choice test whom he liked best, and would approach for rewards and favors, he strongly preferred the "good" confederate over the "bad" one, with the neutral confederate falling in between.
These and other studies just described converge on the conclusion that emotional response can serve as an index of implicit memory. That is, subjects can display emotional responses that are attributable to some event in their past history, in the absence of conscious recollection of that event. However, it must be admitted that the evidence in this regard is rather sparse, especially when compared to the vast body of literature on the perceptual and cognitive expressions of implicit memory (e.g., Roediger & McDermott, 1990; Shacter, 1987). More systematic studies are needed, employing both amnesic patients and normal subjects, specifically testing the hypothesis that emotional response, as an expression of implicit memory, can be dissociated from conscious recollection.
There is considerably better evidence for
emotion as an index of implicit perception -- that is
to say, where emotional responses are attributable to some
event in the current environment, in the absence of conscious
perception of that event. Here again, we begin with anecdote,
and proceed to some formal studies. That is, in
fact, how I interpret the work of Zajonc and others on the subliminal
mere exposure effect. The fact that subliminal
exposures affect preferences shows that the subliminal stimuli
were perceived, albeit unconsciously.
In perhaps the most dramatic extension of the Kunst-Wilson and Zajonc (1980) study, Bornstein and his colleagues found that subliminal exposures can affect not only subjects' preferences for people's faces, but also their interpersonal behavior towards those very same people when they actually meet them later (Bornstein, Leone, & Galley, 1987). Subjects who were subliminally exposed to a picture of a confederate during the study phase were more likely to express agreement with that confederate on a judgment task. However, testing of a separate group of subjects indicated that recognition of the "old" confederate was at chance levels, indicating that the faces had not been consciously perceived during the study phase.
Other investigators have found similar sorts of effects, where subliminal or unattended stimuli have effects on judgments and behavior that would be clearly labeled as "emotional". For example, Murphy and Zajonc (1993) found, like Bornstein (1989), that subliminal exposure to emotional faces produced increased liking and preference for Chinese ideographs. In their view, the familiarity produced by subliminal exposure created diffuse positive feelings analogous to (though different in valence from) the clinical concept of free-floating anxiety.
Of course, considerable early evidence for emotion as implicit perception was provided by investigations of perceptual defense, subception, and other aspects of the "New Look" (Bruner & Klein, 1960; see also Erdelyi, 1974; Greenwald, 1992; Kihlstrom et al., 1992a, 1992b). More recently, Bargh and Pietromonaco (1982) found that subjects who had been exposed to hostile words followed by a masking stimulus attributed significantly more negative qualities to a pictured person than subjects who had not received this masked exposure (see also Bargh, Bond, Lombardi, & Tota, 1986). Similarly, Devine (1989) found that unmasked parafoveal presentation of words related to negative stereotypes of African-Americans led to more negative evaluations of a target person whose race was unspecified.
Niedenthal (1990, 1992; Niedenthal, Setterlund, & Jones,1994) and her colleagues have produced yet another emotional adaptation of the basic subliminal priming paradigm. In the study phase of each experiment, briefly presented primes consisting of faces expressing emotions of joy or disgust were rendered subliminal by means of a meta-contrasting presentation of clearly supraliminal cartoon figures. In the test phase, the subjects were asked to discriminate between old, previously presented cartoons and new distractors. These were also preceded by a face prime, which again was rendered subliminal by meta-contrast. On half the test trials, the affect associated with the prime was the same as it had been on the study trials; for the remainder, the prime was drawn from the opposite emotional category. Congruence between the primes generally facilitated recognition of the targets, especially when the prime was negative. A second study, in which emotionally charged faces or scenes primed emotionally neutral women's faces, obtained essentially the same effect. Moreover, the affective valence of the prime influenced the subjects' interpretations of the target's emotional state. Based on research indicating that the perception of emotionally expressive faces induces a similar emotional state in the perceiver (Niedenthal & Showers, 1991), Niedenthal has proposed that a subliminal emotional prime elicits a corresponding emotional state in the perceiver; this state then serves as a cue for both perceptual identification and recognition memory. It is also, therefore, an expression of implicit perception.
Additional evidence that subliminal emotional primes actually elicit conscious feeling states comes from research on subliminal fear conditioning by Ohman and his colleagues (for reviews, see Ohman, 1998). In one line of research (Ohman, Dimberg, & Esteves, 1989), subjects were conditioned to associate an electric shock with presentation of an angry face (an unreinforced happy face served as a control stimulus). In subsequent unreinforced test trials, the angry face was masked by a neutral face. Despite the fact that subjects could not consciously perceive the angry face, they gave conditioned electrodermal responses when it was presented, compared to masked presentation of the happy face. A subsequent pilot study showed that acquisition of a conditioned fear response is possible, even when the conditioned stimulus is masked and therefore not consciously perceptible. In another line of research, Ohman & Soares (1993, 1994) substituted non-masked pictures of snakes, spiders, flowers, and mushrooms as conditioned stimuli. On unreinforced test trials, masked pictures of snakes and spiders elicited conditioned electrodermal fear responses, but masked pictures of flowers and mushrooms did not. The fact that only the snake and spider pictures survived masking was interpreted within the framework of Seligman's (1971) preparedness theory of phobias which argues that, by virtue of our evolutionary history, some stimuli (such as snakes and spiders) automatically give rise to rapid and long-lasting conditioned fear responses. In this case, the assumption of automaticity of the association is supported by the fact that the fear response persists even when the fear stimulus is subliminal, and thus unattended.
On the basis of early studies showing a dissociation between preference judgments and conscious recollection (Moreland & Zajonc, 1977) and conscious perception (Kunst-Wilson and Zajonc, 1980), Zajonc (1980, 1984a, 1984b) argued that emotional processing is independent of, and temporally prior to, cognitive processing (see also Lazarus, 1982, 1984). However, later studies documenting similar dissociations between explicit and implicit expressions of memory (e.g., Schacter, 1987) and perception (e.g., Kihlstrom et al., 1992a) put a different light on the early results. One might just as well conclude, with respect to memory, that dissociations between recall and priming show that memory itself is independent of, and temporally prior to, cognitive processing. It is now clear that, in the early studies and those that followed, some aspect of emotional response is serving as an implicit expression of perception and/or memory. But unconscious cognition is still cognition. Furthermore, if there were an emotional system separate from cognition, it would still need the cognitive capacity to analyze stimuli, link them to prior knowledge, and generate emotional feelings and expressions (Leventhal, 1980, 1984). The fact that such cognitive processes can go on outside of awareness, so that changes in evaluative judgment and other aspects of emotional response can be dissociated from explicit perception and memory, yields one aspect of the emotional unconscious. People can be aware of their emotional states, but unaware of the percepts and memories which evoke these states.
Clinical folklore about post-traumatic stress disorder has revived the notion, originated by Brewer and Freud (1893-1895/1955) that unconscious memories of trauma express themselves implicitly as intrusive feelings (see, e.g., Bass & Davis, 1988; Blume, 1990; Frederickson, 1993; Herman, 1992; Terr, 1994; van der Kolk, McFarlane, & Weisaeth, 1996; for a detailed analysis of the parallels between Freud's theories and later clinical practices, see Bowers & Farvolden, 1996; Crews, 1995; Kihlstrom, 1996b, 1997b, 1998c). For example Frederickson (1992) has distinguished between a conscious recall memory and an unconscious feeling memory:
Feeling memory is the memory of an emotional response to a particular situation. If the situation we are being triggered to remember is a repressed memory, we will have the feelings pertaining to the event without any conscious recall of the event itself. Feeling memory is often experienced as a flood of inexplicable emotion, particularly around abuse issues.... A felt sense that something abusive has happened is a common form of a feeling memory. Some survivors will say, "Yes, I think I was sexually abused, but it's just a gut feeling." These clients are experiencing a feeling memory about being abused, even though at that moment they can recall nothing about their abuse (p. 92).
In some respects, the notion of a "feeling memory" finds support in the literature reviewed in this section, which indicates that emotional responses can, indeed, serve as expressions of implicit memory. However, there is an important difference: the experimental literature we have reviewed provides independent corroboration of the emotion-eliciting event. Implicit memory may be inferred only when such evidence is available, and such information is rarely available in clinical practice. Nevertheless, clinical practitioners may infer a history of prior trauma and abuse from the patient's current emotional symptoms, and then engage in therapeutic practices intended to recover the traumatic memories and restore them to conscious accessibility. Of course, in the absence of objective corroboration of the patient's history, such inferences are tautological, and should be avoided -- not least because the techniques used to recover ostensibly lost memories are highly suggestive, and may lead patients to reconstruct distorted or false memories of trauma and abuse (Kihlstrom, 1996b, 1998c; Lindsay & Read, 1994, 1995; Shobe & Kihlstrom, 1998).
When feelings and desires emerge as expressions of
implicit memory and perception, they are conscious
feelings and desires. Martial may not have known why
he didn't like Sabidi, but he definitely was aware that he
didn't like him. Now, however, we turn to a more
interesting question -- whether feelings and desires themselves
can influence our experience, thought, and action, even though
we're not aware of them.
Here's the "official" list of Murray's needs.
This viewpoint was taken up by Murray's protege, David
McClelland, and McClelland's students, who focused on what
McClelland called the three great social motives:
It's a good idea. Unfortunately, Murray and Morgan never really got around to standardizing their test -- developing a system by which the stories could be scored for various motives, and collecting norms against which an individual's scores could be interpreted. Nevertheless, the TAT became one of the most popular tests used by clinical psychologists for the assessment of personality and the diagnosis of mental illness.
The task of standardization fell to McClelland and his students, who, in turn, focused on those "three great social motives" of achievement, power, and intimacy. They developed new sets of pictures specifically geared toward one or another of these motives, as well as objective reliable schemes for coding the results of the Picture-Story Exercise (PSE) for motive-relevant imagery. And it also fell to McClelland (who by then had succeeded Murray as professor of psychology at Harvard) to offer a theoretical rationale for preferring the TAT and similar methods to the less cumbersome questionnaire technique.
McClelland and his colleagues (1989)
did just that by drawing a distinction between two
expressions of motivation, much as Schacter had done with
memory. All motives drive, direct, and select
McClelland et al. further
proposed that explicit and implicit motives could be
dissociated, much as is the case with explicit and implicit
One early and frequent criticism of the TAT was that motive scores generated by that procedure did not correlate well with motive scores derived from personality inventories such as Jackson's Personality Research Form (PRF), which had been designed specifically to measure individual differences in Murray's needs. And it's true: a number of reviews have indicated that, at least as far as the three great social motives" are concerned, the correlations between TAT (or PSE) and questionnaire measures (like the PRF) are about as close to zero as you can get. To many psychometricians, this implied that the TAT was not a valid technique of personality measurement. But McClelland pointed out that a person's unconscious motives may differ from his conscious ones. Sometimes they go in the same direction, sometimes they don't, in which case, we wouldn't expect a correlation between them.
These early findings have been confirmed in a more recent meta-analysis by Kollner & Schultheiss (2014), which again focused on "the three great social motives" of achievement, affiliation, and power. Again, the correlations between TAT (or PSE) measures of motivation, and questionnaire (or other self-report) measures of the same motive were very low -- about as close to zero as you can get.
Put another way, explicit (conscious) and implicit (unconscious) motives are dissociable, just as explicit and implicit memory are.
That's an interesting idea, and maybe it's true, but there
are some problems. In the first place, the lack of
correlation is prima facie evidence for a
dissociation, but it's really only the first step.
There might be lots of reasons why the TAT doesn't correlate
with the PRF.
Still, it would be a lot better if the two measures didn't differ so much. Consider the lesson from the early studies of implicit memory. Warrington and Weiskrantz found that priming on stem-completion was spared even when free recall was grossly impaired. But word stems provide more cue information than free-recall tests, so this improvement in memory might just be because the stem-completion test was in some sense "easier". That's why Graf and his colleagues compared stem-cued recall to stem-completion. This kept the cue information provided to the subject constant, but varied the instructions so that the stem-cued recall test required conscious recollection while the stem-cued completion test did not. It's a cleaner, more convincing comparison. You'd like to see something like that in the assessment of implicit motivation. (I'll make the same point when we talk about implicit emotion.)
But can people be unaware of their emotional states themselves? The proposition seems to contain an internal contradiction, because emotions must be felt, and feeling is by any ordinary definition a conscious experience (Clore, 1994). But environmental stimuli must be felt, too; yet cognitive psychology and cognitive neuroscience is gradually coming to terms with the possibility that percepts can be unconscious (Kihlstrom et al., 1992a; Greenwald et al., 1996), just as it earlier came to terms with the possibility that memories can be unconscious (Roediger & McDermott, 1990; Schacter, 1987). If there is a cognitive unconscious, in which percepts, memories, and thoughts influence experience, thought, and action outside of phenomenal awareness, then why can't there be an emotional unconscious as well? The answer depends on how we define emotion. If we define emotion as a conscious feeling state, a positive answer is foreclosed, by definition. But if we define emotion differently, the question is at least open to empirical evidence.
In his classic research on experimental neurosis in dogs, Gantt (1937, 1953) observed that separate components of a conditioned fear response could be acquired and extinguished at different rates, and persist for different lengths of time, resulting in an organismic state of schizokinesis reflecting the "disharmony or cleavage in behavioral, somatic, and psychophysiological response systems" (Mineka, 1979, p. 987). The clear implication of Gantt's work is that a multifaceted emotional response, and that these facets can be separated, or dissociated, from each other.
Gantt's (1937, 1953) observations have been confirmed in more recent research on fear conditioning. For example, Mineka (1979) distinguished four quite different response systems which have been used in the study of fear conditioning in nonhuman animals: conditioned emotional responses, increased rate of conditioned avoidance response, passive avoidance, and conditioned heart rate. Further, she showed that these indices of fear could be dissociated from learned avoidance behavior. Animals can behave as if they are afraid, even if they do not appear to manifest fear according to some standard laboratory measures (see also Mineka, 1985, 1992). One interpretation of such findings is that, contrary to Mowrer's (1947) two-process theory, avoidance learning is not motivated by fear. Another is that the subjective experience of fear is only one component of a broader emotional response to fear stimuli. Similar observations have been made in the case of human fears and phobias. For example, in a study of systematic desensitization of snake phobia, Lang and Lazovik (1963) found that some subjects would show substantial changes in avoidance behavior, while still expressing fear of the snake; other subjects would deny fear of the snake, but show elevated cardiovascular activity and persisting avoidance behavior.
Based on observations such as these, Lang (1968, 1971, 1978, 1988; Lang, Rice, & Sternbach, 1972) proposed a multiple-system theory of emotion. According to this theory, every emotional response consists of several components: verbal-cognitive, corresponding to subjective feeling state (e.g., of fear); overt motor, or behavioral, response (e.g., escape or avoidance); and a covert physiological response mediated by the autonomic and skeletal nervous systems (e.g., skin conductance or heart rate). Lang further proposed that these three systems are partially independent, although they also interact with each other in important ways. When all three systems act together, the person experiences intense emotional arousal. Under circumstances of attenuated emotion, however, the correlations among these systems tend to break apart, at the same time as their individual levels of activity are reduced.
Moreover, Lang proposed that the different components of emotion can have different developmental histories. For example, autonomic responses to emotional stimuli may appear early in development, with the behavioral and cognitive responses emerging only later. Or, alternatively, the cognitive component of an emotional state can be acquired first, as for example through the social learning of fear, with the behavioral and physiological components coming on line later, if at all. Lang further proposed that effective psychotherapy for anxiety states and other emotional disorders should be directed at all three components: it cannot be assumed for example, that flooding directed at reducing compulsive behavior will necessarily reduce subjective anxiety and physiological arousal as well.
Rachman and Hodgson (1974; Hodgson & Rachman, 1974; Rachman, 1978, 1981, 1990) picked up Lang's theme and explored the implications of desynchrony among emotional systems, and especially between overt behavior and covert physiology, for the treatment of anxiety disorders. They proposed that different forms of treatment would have different effects on the various components of fear and anxiety: for example, flooding might reduce avoidance behavior but leave autonomic arousal largely intact; on the other hand, spontaneous remission would affect autonomic arousal first, but behavioral avoidance would persist for a longer period of time. Like Lang (1968), Rachman and Hodgson proposed that fear and anxiety should be assessed in terms of all three components, and that treatment should be directed toward that component which was most "abnormal" (see also Norton, DiNardo, Barlow, 1983). One way or another, however, the remaining components would eventually catch up. So long as one component persisted unchanged, in their view, the likelihood of relapse remained high.
The general idea of desynchrony is that an emotional response can be manifest at one level, but not at another (Hugdahl, 1981; Turpin, 1991). Given that they were writing from a tradition of behavior therapy which emphasizes objective measurement, it is perhaps natural, and certainly understandable, that both Lang (1968) and Rachman (1978) placed most emphasis on desynchrony between the behavioral and physiological components of emotion. In the present context, however, we are most interested in cases that represent the emotional analog of the explicit-implicit distinction in memory: where the subjective component of an emotion (conscious feeling state) is absent, while the behavioral and physiological components persist outside of phenomenal awareness. The snake phobic denies fear, but somehow never quite manages to go near the reptile house at the zoo; the agoraphobic claims to be cured, and even ventures outside the house, but blood pressure and heart rate still go way up. If such observations reflected merely denial, or a flight into health, they would not be too interesting. But suppose that the patients' reports accurately reflect the subjective state of affairs -- that they really do not experience the emotions that used to bother them. If the behavioral and autonomic signs of emotion persist unabated, why can't we say that they are displaying unconscious emotion -- or at least an unconscious emotional response?In line with Lang's multiple-component view of emotion, we can distinguish between explicit and implicit emotion as follows:
The idea of dissociable emotion systems is related to the concept of the emotional brain, as it has evolved from Cannon (1929) and Bard (1929) through Papez (1937) and MacLean (1949) to LeDoux (1996). Cannon and Bard found that decorticate animals would still show fear responses, so long as the thalamus and hypothalamus remained intact. These observations led them to propose that the diencephalon, which contains these structures, was the seat of the emotions. According to their view, the diencephalon mediated skeletal and autonomic emotional responses; in contrast, the conscious experience of emotion was mediated by the cortex, which was activated by fibers ascending from the hypothalamus. In such a system, a disconnection (Geschwind, 1965) between the diencephalon and the cortex would impair the subjective experience of emotion, while leaving the behavioral and physiological components intact.
Later research and theory broadened the emotional brain to include the entire limbic system (MacLean, 1949, 1952; Papez, 1937) -- a move which, according to one commentary, "had the appeal of combining behavioral phenomena having no known neurological substrates with anatomical structures having no known function" (Kolb & Wilshaw, 1996, p. 418). For Papez, the afferent messages arriving at the thalamus were transmitted in two separate streams to the sensory cortex (the stream of thought) and to the hypothalamus (the stream of feeling). The hypothalamus, in turn, generated skeletal and autonomic responses to the stimulus, and also transmitted sensory information to the cingulate cortex, which also received inputs from the sensory cortex. When inputs from the hypothalamus were integrated with inputs from the sensory cortex, an emotional feeling state was generated. In such a system, three different disconnections could create a desynchrony between the explicit subjective and implicit emotion: (1) between the thalamus and the sensory cortex; (2) between the hypothalamus and the cingulate cortex; and (3) between the sensory cortex and the cingulate cortex. In any of these cases, the behavioral and physiological responses to an emotional stimulus would run off unimpaired, in the absence of any corresponding subjective feeling state. This is known as Papez' circuit.
MacLean (1949, 1952, 1970, 1990) broadened the limbic system even further, including the amygdala and other structures connecting directly to the hypothalamus, and proposed that a paleomammalian brain mediates the visceral and emotional life of the (mammalian) organism, while the neomammalian brain mediates consciousness, language, and other complex cognitive functions. MacLean also proposed a reptilian brain consisting, essentially, of the brainstem. Thus, along the lines outlined by Papez (1937), a disconnection between the paleomammalian and neomammalian brains could impair explicit emotion, while sparing implicit emotion.
recently, LeDoux (1995, 1996) has proposed a variant on the
Papez/MacLean theory which has the virtue of being more
specific in terms of both anatomy and psychology. Briefly,
LeDoux argues that a particular structure in the limbic
system, the amygdala, mediates a particular emotion, fear
(for a similar analysis, see Damasio, 1994).
Based on his studies of fear conditioning in rats and other
nonhuman animals, but supported by studies of human patients
who have suffered damage to the amygdala and surrounding
brain tissue (e.g., Adolphs, Damasio, Tranel, & Damasio,
1996; LaBar, LeDoux, Spencer, & Phelps, 1995), LeDoux
has proposed that fear stimuli are processed by the
amygdala, which in turn generates appropriate behavioral,
autonomic, and endocrine responses. Cortical arousal,
feedback of somatic and visceral information, and
information about the fear stimulus are then integrated in
working memory to generate the subjective experience of
As with the simpler systems described by Papez and MacLean, a disconnection between the amygdala and the cortex can produce a dissociation between explicit and implicit emotion: the person will respond in a fearful manner without feeling fear or anxiety.
LeDoux's system is especially appealing because it also offers a mechanism by which fear can serve as an implicit expression of memory: if the eliciting stimulus is not represented in working memory, the person will experience fear without being aware of the fear stimulus.
It should be understood that LeDoux's (1995, 1996) analysis of the amygdala applies only to the emotion of fear. Whereas Papez and MacLean implied that all emotions were mediated by a single system (Papez's circuit or the limbic system), LeDoux postulates a number of different systems, each mediating conscious experience, motor behavior, and somatic changes, corresponding to different emotional domains. It is not clear how many such systems there are, but if Ekman and Friesen (1975) are right that some patterns of emotional expression have deep evolutionary roots, a reasonable hypothesis is that there may be at least seven separate systems, corresponding to the "basic emotions" of surprise, happiness, sadness, fear, anger, and disgust. Thus, the range of possible dissociations is not just between explicit and implicit expressions of emotion in general. It may also be possible, at least in principle, to observe in a single patient, desynchrony between explicit and implicit fear, but synchrony between explicit and implicit anger. This is, of course, a nightmare for any prospective researcher of the emotional unconscious.
But can emotions really be unconscious? A 1994 symposium answered this question in the negative (Clore, 1994; Davidson & Ekman, 1994; LeDoux, 1994; Zajonc, 1994). There was general agreement that the cognitive and brain processes underlying emotions could operate outside of conscious awareness and conscious control. In this way, we might not be conscious of the source of our emotions; and, in the absence of source awareness, we might not be aware of precisely which emotion we are experiencing. But as Clore put it, the essence of emotion is feeling, and "emotions that are felt cannot be unconscious by definition" (1994, p. 285). And if, as Clore and Schwarz (Clore,1992; Clore, Schwarz, & Conway, 1994; Schwarz & Clore, 1983, 1988) propose, the function of emotion is to provide information concerning the appraisal and evaluation of the (past, current, or anticipated future) situation, it would certainly be dysfunctional if emotions were unconscious.
This is, of course, an empirical question. So how
would we go about documenting a dissociation between
explicit and implicit memory?
The nightmare is exacerbated by a consideration of the logic of inferring unconscious emotions. Consider the analogy to implicit memory. We know that priming is evidence of implicit memory because we can trace the facilitation in lexical decision, perceptual identification, and the like, to a specific objectively observable event -- the nature of the prime presented to the subject. Furthermore, we can specify objectively the relationship between the prime and the target: same versus different word, same word/same appearance versus same word/different appearance, and so on. Put another way, we can identify an implicit expression of memory because we know what happened to the subject in the past -- what the subject should be remembering.
But by the same logic, in order to identify an implicit expression of emotion, we have to know what emotional state the subject should be experiencing, or which emotional state is being represented, and expressed, outside of conscious awareness. Applying the logic of explicit and implicit memory to the problem of emotion, then, compelling evidence of a dissociation between explicit and implicit emotion makes a number of methodological demands.
First, we need to possess an adequate stimulus for emotion -- that is, a set of stimuli which, under ordinary circumstances, reliably elicits particular emotions in subjects. Unfortunately, the search for such reliable elicitors has not been particularly fruitful. Apparently, just as the experience of pain depends on the subjective meaning of the pain stimulus, the person's emotional response to a situation depends greatly on his or her cognitive appraisal of that situation (e.g., Ellsworth, 1991; Lazarus, 1991; Lazarus & Smith, 1988; Ortony, Clore, Collins, 1988; Smith & Ellsworth, 1985).
Still, some fairly universal elicitors of emotion have been proposed, and these hold promise for desynchrony research. Ekman and Friesen (1975) found, among other relations, that actual or threatened harm elicits fear, while loss of an object to which one was attached induces sadness. Similarly, Scherer, Wallbott, and their colleagues have found, among other relations, that basic pleasures elicited joy, while separation elicited sadness (Scherer, Wallbott, & Summerfield, 1986; Scherer & Wallbott, 1994). To be sure, the relations in question are moderated to some degree by context-specific appraisals and cultural considerations, but there is enough cross-situational consistency that we can have some hope of measuring subjects' subjective, behavioral, and physiological responses to stimuli that, all things being equal, should elicit certain emotions. If we observe diminished subjective awareness coupled with persisting behavioral and physiological responses, we would have evidence of a dissociation between explicit and implicit emotion.
Of course, documenting such a dissociation also requires that we have reliable measures of the subjective, behavioral, and physiological responses to the emotion stimulus; and these are nontrivial problems. Subjective feeling states can be assessed by the usual sorts of self-report measures, but care must be taken to distinguish between the subjects' failure to consciously feel a particular emotion, and their willingness to report what they feel to an experimenter. Implicit emotion is about awareness, not denial. The behavioral component of emotion might be indexed by gross patterns of approach/withdrawal, flight/fight, or activation/inhibition (Gray, 1987), and the physiological component by generalized levels of autonomic arousal (Schacter & Singer, 1962), but again ideally we would like to have implicit measures that are more specifically isomorphic to the lexicon of conscious emotions. On the overt behavioral side, one possibility are facial expressions of the sort documented by Ekman and Friesen (1975), as well as cognate postural and gestural expression. On the level of covert physiology, Levenson (1988, 1992) and his colleagues have been able to document specific patterns of autonomic response accompanying particular indices of emotion (Ekman, Levenson, & Friesen, 1983; Levenson, Carstensen, Friesen, & Ekman, 1991; Levenson, Ekman, & Friesen, 1990; Levenson, Ekman, Heider, & Friesen, 1988). Similarly, Davidson (1993) has suggested that particular patterns of cortical activation may also differentiate certain basic emotions.
The ideal structure of an experimental
demonstration of implicit emotion is now clear. To begin
with, we assume that a particular emotional state is a
hypothetical construct defined by the logic of converging
operations (Garner, Hake, & Eriksen, 1956; see also
Campbell & Fiske, 1956; Kihlstrom, 1984; Stoyva &
Kamiya, 1968). These operations include specification of an
eliciting stimulus and the measurement of subjective
experience, overt behavior, and physiological response.
When all operations agree, we can be fairly confident that the person is in an emotional state such as fear or happiness. Agreement among the operations is sufficient to establish the presence of an emotional state. However, none of these operations is necessary for this purpose, and under some circumstances some combination of two or three operations would suffice. For example, a subject might report feeling no fear in response to a real or imagined threat. At the same time, he or she should continue to manifest facial, postural, and gestural expressions of fear, as well as autonomic and cortical signs of fear. Under these circumstances, the diagnosis of implicit fear might well be irresistible. As we grow more confident with respect to classification of emotional stimuli, and multivariate measurement of emotional responses, we will be in a better position both to evaluate the multiple-systems theory of emotion and to search for evidence of desynchronies between emotional systems -- especially the particular pattern(s) of desynchrony characteristic of a dissociation between explicit and implicit emotion.
The more a study of implicit emotion looks like a study of implicit memory,
the more convincing that study will be.
Moving from theory (whether psychological or neuroscientific) to experimentation, what empirical evidence suggests that explicit and implicit emotion can be dissociated?
Much, perhaps most, of the evidence bearing on the concept of implicit emotion comes from recent social-psychological work on attitudes, stereotypes, and prejudice. In social psychology, attitudes have a central affective component: they are dispositions to favor or oppose certain objects, such as individuals, groups of people, or social policies, and the dimensions of favorable-unfavorable, support-oppose, pro-anti naturally map onto affective dimensions of pleasure-pain or approach-avoidance. As Thurstone put it, "attitude is the affect for or against a psychological object" (1931, p. 261). Like emotions, attitudes are generally thought of as conscious mental dispositions: people are assumed to be aware that they are opposed to nuclear power plants, or favor a women's right to choose. Similarly, people are generally believed to be aware of the stereotyped beliefs that they hold about social outgroups, and of the prejudiced behavior that they display towards members of such groups. And for that reason, attitudes and stereotypes are generally measured by asking subjects to reflect and report on their beliefs or behavior. However, Greenwald and Banaji (1995) proposed an extension of the explicit-implicit distinction into the domain of attitudes. Briefly, they suggest that people possess positive and negative implicit attitudes about themselves and other people, which affect ongoing social behavior outside of conscious awareness.
Following the general form of the explicit-implicit distinction applied to memory, perception, learning, and thought in the cognitive domain, Greenwald and Banaji distinguished between conscious and unconscious expressions of an attitude:
Similarly, Blair and Banaji (1996) conducted a series of experiments in which subjects were asked to classify first names as male or female. Prior to the presentation of each target, the subjects were primed with a word representing a gender-stereotypical or gender-neutral activity, object, or profession. In general, Blair and Banaji (1996) found a gender-specific priming effect: judgments were faster when the gender connotations of the prime were congruent with the gender category of the name. This means that gender stereotypes influenced their subjects' classification behavior.
In the area of racial stereotypes, Gaertner and McLaughlin (1983) employed a conventional lexical-decision task with positive and negative words related to stereotypes of Blacks and whites, and the words "black" or "white" serving as the primes. There was a priming effect when positive targets were primed by "white" rather than "black", but no priming was found for the negative targets, and this was so regardless of the subjects' scores on a self-report measure of racial prejudice. thus, the effect of attitudes on lexical decision was independent of conscious prejudice.
Similarly, Dovidio Evans, and Tyler (1986) employed a task in which subjects were presented with positive and negative trait labels, and asked whether the characteristic could ever be true of black or white individuals. While the judgments themselves did not differ according to race (even the most rabid racist will admit that there are some lazy whites and smart blacks), subjects were faster to endorse positive traits for whites, and to endorse negative traits for blacks. Thus, even though conscious attitudes did not discriminate between racial groups, response latencies did.
These studies, and others like them (e.g., Devine, 1989), seem to reveal the implicit influence of sexist or racist attitudes on behavior. However, at present, interpretation of these results is somewhat unclear. In the first place, the logic of the research is that stereotype-specific priming indicates that subjects actually hold the stereotype in question -- that, for example, the subjects in Blair and Banaji's (1996) experiment really (if unconsciously) believe that males are athletic and arrogant while females are caring and dependent. However, it is also possible that these priming effects reflect the subjects' abstract knowledge of stereotypical beliefs held by members of society at large, though they themselves personally reject them -- both consciously and unconsciously. Thus, a subject may know that people in general believe that ballet is for females and the gym is for males, without him- or herself sharing that belief. Even so, this knowledge may affect his or her performance on various experimental tasks, leading to the incorrect attribution of the stereotypical beliefs to the subject.
Moreover, most studies of implicit attitudes lack a comparative assessment of explicit attitudes. Implicit measures of attitudes may be useful additions to the methodological armamentarium of the social psychologist, but in the present context their interest value rests on demonstrations of dissociations between explicit and implicit expressions of emotion. Accordingly, it is important for research to show that implicit measures reveal different attitudes than those revealed explicitly. Just as the amnesic patient shows priming while failing to remember, and the repressive subject shows autonomic arousal while denying distress, we want to see subjects displaying attitudes or prejudices which they deny having, and acting on stereotypes which they deny holding.
Wittenbrink, Judd, and Park (1997) performed a formal comparison of explicit and implicit racial attitudes. Their subjects, all of whom were white, completed a variety of traditional questionnaire measures of self-reported racial attitudes. They also performed a lexical-decision task in which trait terms drawn from racial stereotypes of whites and blacks were primed with the words black, white, or table. Analysis of response latencies found, as would be anticipated from the studies described above, a race-specific priming effect: white speeded lexical judgments of positive traits, while black speeded judgments of negative traits. However, the magnitude of race-specific priming was correlated with scores on the questionnaire measures of racial prejudice. In this study, then, implicit attitudes about race were not dissociated from explicit ones. Such a finding does not undermine the use of implicit measures in research on attitudes and prejudice (Dovidio & Fazio, 1992), but a clear demonstration of a dissociation is critical if we are to accept implicit attitudes as evidence of an emotional unconscious whose contents are different from those which are accessible to phenomenal awareness.
Beginning in 1998, Greenwald, Banaji, and their colleagues
have introduced the Implicit Attitude Test (IAT),
which is expressly designed to measure implicit
attitudes. The IAT consists of a series of
dichotomous judgments, which we can illustrate with a
contrived "Swedish-Finnish IAT" that might be used to detect
prejudice of Swedes against Finns (or vice-versa).
Just to make it perfectly clear:
Link to demonstrations of the IAT on the "Project Implicit" website. As of October 2015, the online IAT had been completed more than 17 million times.
For example, in
one early study, Greenwald, Banaji, and their colleagues
looked at white subjects implicit attitudes toward
blacks. Subjects responded faster when stereotypically
"White" names shared a response key with "Positive", than
when stereotypically "Black" names shared a response key
with "Positive", thus implying that they associated White
with good, and Black with bad.
Greenwald et al. also measured their subjects' explicit racial attitudes with a standard technique known as the attitude thermometer, which is basically a numerical rating scale with one pole labeled "positive" and the other pole labeled "negative". The correlation (r) between IAT and the attitude thermometer varied from .07 to .30, depending on the sample.
In another study, they looked at
attitudes among certain Asian ethnicities. Koreans
responded faster when "Korean" names shared a key with
"Positive", and 'Japanese" names shared a key with
"Negative". Japanese subjects did precisely the
opposite, responding faster when "Japanese" and "Positive"
shared a key than when 'Korean" and "positive" did.
Thus implying that Koreans associated Japanese with bad,
while Japanese made the same associations with Koreans.
Greenwald et al. also measured these
subjects' explicit ethnic attitudes with a standard
technique known as the attitude thermometer, which
is basically a numerical rating scale with one pole labeled
"positive" and the other pole labeled "negative". The
correlation (r) between IAT and the attitude
thermometer varied from -.04 to .64, depending on the
By now a huge literature has developed
in which the IAT has been used to measure almost every
attitude under the sun. Nosek (2007) summarized this
literature with a graph showing the average
explicit-implicit correlation, across a wide variety of
attitude objects. These correlations varied widely,
but the median explicit-implicit correlation was r =
.48. Note, however, that this mean correlation is
substantially higher than the corresponding correlations
between measures of explicit and implicit motives, as
reported by investigators such as Kollner and Schultheis
Based on findings such as these, Banaji and Greenwald have concluded that the IAT reveals an attitudinal "blindspot" -- that we carry around in our heads a set of "hidden biases", developed over a lifetime of social learning, that affect our social interactions -- especially with racial and ethnic minorities and other outgroups -- outside our conscious awareness and control.
Greenwald, Banaji, and their colleagues have claimed that the explicit-implicit correlations obtained between the IAT and the attitude thermometer and other self-report measures are relatively low, and this suggests that unconscious attitudes can, indeed, be dissociated from conscious ones. But there are some problems with this argument (for early critiques, see Kihlstrom, 2003; Fiedler et al., European Review of Social Psychology, 2006; Blanton et al., J. Exp. Soc. Psych., 2006; and an exchange between ).
Why does this last point matter? If
we're interested in implicit attitudes, then counterbalancing is
just part of proper methodology. The problem is that we're not just interested in attitudes in
general. From the beginning, the IAT has been touted
as an instrument for the assessment of an individual's
attitudes. That's why individuals who complete the IAT online are given their
scores (and told how much they're unconsciously racist, or
whatever). That's why there are training programs, for
police officers, teachers, and even office workers, to help them
overcome whatever biases they may, individually, harbor deep in
their unconscious minds.
So how does the IAT stack up as a psychometric
instrument? All psychological tests, such as the Wechsler
Adult Intelligence Scale or the Minnesota Multiphasic
Personality Inventory, have to have certain psychmetric properties.
Let's see how the IAT stacks up against these standards (for
further discussion of these psychometric standards, see the
lectures on "Trait
Research" in the lecture
supplements for my course on
"Personality"; for a shorter
version, see the lectures on "Personality
and Social Interaction" in the lecture
supplements for my course on "General Psychology".
As if 2017, there had been no formal assessments of the utility of the IAT. In fact, psychological tests are rarely accompanied by assessments of utility -- which is one reason why "projective" tests like the Rorschach have lasted as long as they have. But we can get some sense of the utility of the IAT from the literature on its validity, much of which compares the validity of the IAT to the validity of alternative assessments of attitudes.
With respect to validity, the problem for any assessment of attitude is determining exactly what the person's attitude is. It is one thing for a Swede to actively dislike Finns, but it is another thing entirely for a Swede to like Finns well enough, but like Swedes better. The IAT cannot distinguish between these two quite different attitudinal positions. All it does is make an inference of relative attitude from relative reaction times.
In psychometric terms, there are four different forms of validity.
Now with all this background, let's focus on the IAT.
Ordinarily, the simplest approach would be to determine how well the IAT correlates with an obvious measure of attitudes, such as a simple self-report or ratings on an attitude thermometer. But that won't work for the IAT, precisely because there is supposed to be a dissociation between explicit and implicit attitudes. In a sense, a low correlation between IAT and attitude thermometer scores is evidence for the validity of the IAT as a measure of implicit, unconscious attitudes. If the correlations were high, the IAT might be a measure of attitudes, but it wouldn't be a measure of implicit attitudes that are different from their explicit counterparts. And we already know those correlations are low -- that's the evidence for a dissociation between explicit and implicit attitudes. So we can't take the usual tack, which would be to correlate the IAT with self-report measures of attitude. However, Lane et al. (2007) actually cite the correlations between IAT and explicit measures as evidence of the validity of the IAT. They noted that, across 17 studies, the median correlation between IAT and some explicit measure of the same attitude was r = .22 -- which, while admittedly low (as indeed it should be!), is still statistically significant. Thus, the IAT has its cake and eats it too: it's correlation is high enough to support its validity, but not so high as to compromise its status as a measure of implicit, rather than explicit, attitudes.
We can get some assessment of convergent validity by
looking at the correlation of the IAT with other measures of implicit
attitudes. As noted earlier, priming can also be used to
measure implicit attitudes. If presentation of a word
relevant to the African-American stereotype prompts white subjects
to think of "black", as in Devine's (1989) early study, then we
might -- might -- be justified in concluding that a subject who
scores high on the IAT really does hold negative attitudes towards
African-Americans. Lane et al. (2007) surveyed a number of
such studies, mostly involving priming, and concluded that "the
IAT was uncorrelated with the other implicit measures" (p.
Unfortunately, the Bosson study is not an outlier. The
general thrust of this literature is that the IAT shouldn't
correlate with explicit measures, and it doesn't. But its
correlations with other implicit measures are even lower,
and the other implicit measures don't correlate much with each
other. Lane et al. suggest that this outcome may be due to
attribute this failure to the IAT's low levels of internal
consistency, as measured by Cronbach's alpha. While it's
true that correcting for measurement error will raise these
correlations somewhat, it's not going to put them anywhere near
the much higher values routinely observed with explicit
measures. So, it's hard to say that the IAT has the degree
of convergent validity that we'd like to see.
Lane et al. also discuss the discriminant validity of the IAT. They conclude that the IAT is correlated with explicit measures of the same attitude, but not with explicit measures of different attitudes. So, for example, the race IAT might correlate with scores on an attitude thermometer measuring racial bias, but not with an attitude thermometer measuring gender bias. Lane et al. conclude that this is the case, so that the IAT possesses discriminant validity. However, implicit measures of X are not supposed to correlate with explicit measures of X. That is the whole point of claiming a dissociation. So, this sort of evidence of discriminant validity cuts both ways. The more discriminant validity the IAT shows, with respect to explicit attitude measures, the less validity it can have as a measure of implicit attitudes.
Another approach to validity involves criterion groups. That is, individuals known to have a particular characteristic (e.g., a trait or attitude) should score higher on tests of that characteristic, compared to those known not to have this characteristic. This is, essentially, how the MMPI, CPI, and many other classic personality inventories were developed. For the MMPI "schizotypy" scale, for example, patients diagnosed with schizophrenia were asked to answer a large number of personality-type questions (my favorite is "Sometimes my thoughts turn into butterflies"). Items that the patients endorsed more frequently than a control group then were collected into a "schizotypy" scale measuring tendencies toward schizophrenia. In this way, so the developers argued, external validity was built into the scale from the beginning. Lane et al. (2007), summarize a few studies which seem to suggest that the IAT possesses this kind of criterion validity. In a study by Greenwald et al. (1998), for example, Japanese- and Korean-Americans, social groups known for their strong ingroup favoritism (note to Japanese and Korean readers: if you object to this characterization, take it up with the authors of the paper; I'm just citing it to illustrate this approach to validity).
Last, we turn to
construct validity. How well does the IAT
predict some construct-relevant external criterion, such as
a Swede's willingness to hire a Finn, or let him marry his
daughter? Greenwald et al. (JPSP 2009)
performed a comprehensive meta-analysis of more than 100
papers, including almost 15,000 subjects
comparing the ability of self-report measures
of explicit attitudes, and IAT measures of implicit
attitudes, to predict attitude-relevant behavior. Because of the obvious social importance of the
problem, these studies focused on stereotyping and
prejudice directed at racial minorities and ethnic
example, how well do explicit and
implicit measures of racial attitudes predict
nonverbal behavior of whites towards blacks?
How well do they predict competition between Japanese and
Koreans in the Prisoner's Dilemma?
Such findings strengthen the case that the IAT is actually a measure of attitudes -- precisely because it is able to predict attitude-relevant behavior. But none of them support the inference that the IAT is a measure of unconscious attitudes -- as opposed to attitudes of which the subjects are consciously aware, but reluctant to disclose to other people.
But these promising findings are tempered by a further
meta-analysis of studies of racial attitudes performed by Oswald
et al. (JPSP 2013). Although Greenwald et al. (2009)
focused on racial and ethnic prejudice (after all, these attitudes
are much more societally relevant than attitudes towards
vegetables or insects), the depend variables in their
meta-analysis covered an extremely wide range , from overt
interpersonal behavior (e.g., cooperation vs. competition in the
Prisoner's Dilemma game) to brain activity (e.g., fMRI). To
make a long story short, they added a number of studies published
since 2006 (the end-date for Greenwald et al.'s analysis), and
classified the studies which Greenwald et al. had lumped together
into various categories (for details, read the paper).
These effects are significant, because of the Ns in the studies involved, but they're small even by the standards of social-science research: implicit attitudes, as measured by the IAT, explained only about 2-3% of the variance in racial/ethnic stereotyping/prejudice.
Why were the effects observed by Oswald et al smaller, in
general, then those observed by Greenwald et al. (2009)?
Oswald et al. concluded that the IAT is a "flawed instrument for
the assessment of attitudes, including implicit
attitudes. And, to boot, that theories of the role of
implicit bias in prejudice and discrimination, insofar as they
were based on research using the IAT, were also, perforce,
flawed. They write:
"The initial excitement over IAT effects gave rise to a hope tht the IAT would prove to be a window on unconscious sources of discriminatory behavior..... This closer look at the IAT criterion studies in the domains of ethnic and racial discrimination revealed, however, that the IAT provides little insight into who will discriminate against whom, and provides no more insight than explicit measures of bias. The IAT is an innovative contribution to the multidecade quest for subtle indicators of prejudice, but the results of the present meta-analysis indicate that social psychology's long search for an unobtrusive measure of prejudice that reliably predicts discrimination must continue..." (p. 188).
Not surprisingly, Greenwald et al. (JPSP 2015) responded
with a full-throated defense of the IAT. They began by
noting procedural differences between the 2009 and 2015
meta-analyses. They also argued, correctly, that small
effects, when multiplied, over many individuals or occasions, can
have substantial, societally important, aggregate effects.
Oswald et al. (JPSP, 2015) responded to Greenwald et al.
(2015) by restating that
"all of the meta-analyses converge on the conclusion that, across diverse methods of coding and analyzing the data, IAT scores are not good predictors of ethnic or racial discrimination, and explain, at most, small fractions of the variance in discriminatory behavior in controlled laboratory settings. The thought experiments presented by Greenwald et al. go well beyond the lab to claim systematic IAT effects in noisy real-world settings, but these hypothetical exercises depend crucially on untested and, arguably, untenable assumptions" (p. 562)
But wait! There's more, in the form of yet another
meta-analysis, by Forscher et al. currently (March 2017) under
review but already widely circulated, which largely supports the
conclusions of Oswald et al. (2013, 2015). The interesting
feature of this analysis is that two of its seven co-authors are
closely associated with Greenwald and Banaji's Project Implicit
the IAT Corp -- one of them its current Director of research, the
other a co-author on several of the pioneering studies of the
IAT. They employed an innovative meta-analytic technique to
aggregate the results of more than 500 studies investigating
various ways ot changing implicit bias (this account is based on
an earlier draft, from 2016, summarizing more than 400
studies. . This is an important problem, because
implicit bias, if it is truly unconscious, means that it is
operates either outside conscious awareness or voluntary control
or both. Of course, if you're unaware of a bias, you can
hardly control it. So most of these interventions use the
IAT or a similar procedure (like priming) to make people aware of
their implicit biases; then perform some sort of intervention --
e.g., to strengthen positive associations or weaken negative ones;
and finally test the effectiveness of the intervention -- on
post-test implicit attitude scores scores, explicit attitudes, and
discriminatory or prejudicial behavior.
The general conclusion of this analysis is that, on the whole,
these sorts of interventions produce small but significant changes
in implicit bias, corresponding to an effect size of d
< .30. Although the effects were weak, they were,
interestingly, stronger than the effects of the same procedures on
explicit bias, or on discriminatory or prejudicial
behavior. Put another way, changes in performance on the IAT
were not associated (very much) with changes in either explicit
attitudes or attitude-related behavior. The implication,
consistent with the validity studies summarized in the Greenwald
et al. (2009) and Oswald et al. (2013) meta-analyses, is that the
IAT has very low levels of construct validity.. Some
interventions were more effective than others, but for our
purposes we are mostly interested in validity issues -- the
ability of the IAT to predict actual discriminatory and
It is possible to argue, as some members of Project Implicit
have, that these sorts of problem are less relevant when the IAT
is used in laboratory research on prejudice and stereotyping
involving dozens or hundreds of subjects. But that's not the
only way the IAT is being used. In fact, since its
introduction in 1998, up at leas through the publication of Blindspot,
including many appearances on national TV and other media, the IAT
has been presented as an instrument for revealing the extent of an
individual's unconscious biases.
The IAT Controversy
With respect to the unconscious status of the attitudes
ostensibly assessed by the IAT, one implication of the
Greenwald/Banaji argument is that the IAT picks up on attitudes
that influence behavior automatically, outside our voluntary
control. Which brings to mind Jacoby's Process-Dissociation
Procedure, which enables us to separate the influence of automatic
and controlled processes on task performance.
Inspired by Jacoby's work, Sherman and his colleagues at UC Davis have developed the QUAD Model of automatic bias in stereotyping and prejudice.
A study by Jennifer Beer et al. (2008; Beer took her PhD from UC Berkeley) illustrates the logic of the QUAD Model, as applied to a Black-White version of the IAT, in which subjects must classify faces as black or white, and words as positive or negative (much like the paper-and-pencil version of the IAT shown above).
By means of a subtraction logic not unlike Jacoby's PDP, the QUAD
model permits us to estimate the value of each of the parameters
underlying performance on the IAT.
Interestingly, in two studies of the Black-White IAT appear to
suggest that automatic activation plays a relatively small role in
performance on the IAT. In both studies, the subjects were
white college students. Both studies computed separate
parameter estimates for the automatic association (AC) of "black"
with "unpleasant" and "white" with "pleasant".
Another aspect of construct validity has to do with group differences. Koreans appear prejudiced against Japanese, and Japanese against Koreans, and that's what we'd expect if the IAT really measured prejudicial attitudes. But, for example, there is little evidence concerning other ingroup-outgroup differences in IAT performance. The problem is encapsulated in the title of a famous critique of the IAT, entitled "Would Jesse Jackson fail the IAT?". If, for example, African-American subjects also "favor" whites when they take the IAT, it would be hard to characterize a Black-White IAT as a measure of prejudice against African-Americans.
The invocation of Jesse Jackson stems from a remark the civil-rights advocate once made, to the effect that when he sees a group of black urban youth coming toward him on the street, even he feels a desire to cross to the other side to avoid them.
The biggest problem, however, is that correlation between explicit and implicit prejudice, which Nosek reports at a median r of .48. That's not a perfect correlation of 1.00, but it's also not a zero correlation. In fact, it's a big correlation by the standards of social-science research -- and, in fact, it's about as big as it can get, given the test-retest reliability of the IAT. If explicit and implicit attitudes were truly dissociable, we'd expect the explicit-implicit correlation to be a lot lower than it is. The fact that explicit-implicit correlations are typically positive, and in many cases quite substantial, suggests, to the contrary, that people's implicit attitudes are pretty much the same as their explicit attitudes. They're not dissociated, they're associated.
This last problem is confounded by the fact that improvements in the scoring procedure for the IAT have actually led to increases in the correlation between explicit and implicit attitudes. But if the IAT were really measuring unconscious attitudes, we'd expect psychometric improvements to decrease the correlation -- to strengthen the evidence that explicit and implicit attitudes are truly dissociable.
The fact that explicit and implicit attitudes are significantly correlated, and that the correlation increases with psychometric improvements, suggests instead that the IAT may be an unobtrusive measure of attitudes that are consciously accessible, but which subjects are simply reluctant to disclose -- something on the order of a lie-detector. But an unobtrusive measure of a conscious attitude shouldn't be confused with a measure of an unconscious attitude.
But even that isn't entirely clear, because of the technical problems with the IAT described earlier -- issues of target familiarity, task difficulty, distinguishing between unfavorable attitudes and those that are simply less favorable. For this reason, I think it's premature for its promoters to promote the IAT as a measure of any kind of attitude, conscious or unconscious.
Frankly, the IAT brings us full circle, back into Freudian territory -- though without the lurid claims about primitive sexual and aggressive motives. Freud was quite content to tell people what their problems were -- that, for example, they loved their mothers and hated and feared their fathers. And when people would say it wasn't true, he would explain to them the concept of repression. And when they continued to resist, he'd tell them that their resistance only indicated that he was right. In much the same way, it's a little disturbing to find the promoters of the IAT using it to tell people that they're prejudiced, only they don't know it. Because it isn't necessarily so.
This is the problem of what William James and John Dewey called the psychologist's fallacy -- the idea that, first, every event has a psychological explanation; and, second, that the psychologist's explanation is the right one. Freud thought that he knew better than his patients what their feelings and desires were. The "IAT Corporation" (yes, there really is one, offering the IAT to government and corporate personnel and human-relations departments concerned about workforce diversity) claims to know better than you do whether your prejudiced against African-Americans, or Hispanics, or Japanese, or Koreans.
At this point it's important to be reminded of what William James wrote about the unconscious mind. It's critical that assessments of unconscious motivation and emotion, no less than unconscious cognition, be based on the very best evidence. Otherwise, unconscious mental life will become the "tumbling-ground for whimsies" that James warned it could be.
This page last modified