More recently, a number of investigators have taken the position that an affect is simply another form of information so that what the affect is about
may be unconscious, but the emotion itself is a conscious experience [9
]. Lambie and Marcel [10
] also place great emphasis on the phenomenology of emotion experience, taking the view that what is nonconscious are the processes and representations that underlie the actual conscious experience of the emotion. Interestingly, Zajonc’s [11
] position, although differing from Lambie and Marcel’s in a number of significant ways, is similar in that what is unconscious causes the conscious emotion experience, but is not itself to be thought of as an emotion. This position finds some support in split brain patients who demonstrate the appropriate affect associated with a stimulus presented to the right hemisphere (e.g., the word “laugh” elicits laughter) but are unable to identify its cause. Breuer and Freud [13
] described a similar condition in patients with intact brains but who suffered from hysterical dissociation. This kind of disconnection between expressed and experienced affect without awareness of its cause has also been described in normal hypnotized subjects who enact hypnotic suggestions. In all these instances, however, the instigating stimuli were supraliminal, but were unavailable either by reason of brain damage (split brain patients), psychopathology (hysterical dissociation or repression), changed states of consciousness (hypnosis), leaving open the question as to what happens when the stimuli are entirely unconscious, the subject of interest in the studies to be reported.
Among neuroscientists, Panksepp argues in favor of the conscious nature of affects on quite other grounds. For Panksepp, affects must be qualitatively distinguished from cognitive processes that are capable of becoming unconscious because affects function in an entirely different manner. Cognitive processes provide information about the outer and inner world, but affects provide the internal value codes (e.g., good/bad) necessary to sustain ongoing behavior and could not perform that role if they were unconscious [14
]. According to Panksepp, affects are generated deep in the brainstem and are intrinsically conscious, while cognitive processes are more closely associated with the cortex and can become unconscious. Neuropeptidergic neuromodulaters regulate distinct affect states, while cognitive processes depend more on informationally resolved glutamatergic transmission. Consciousness rather than being restricted just to higher mental functions is assumed also to be present in the deeper, older parts of the brain [16
]. On the other hand, Damasio [18
] and Berridge and Winkielman [19
], among others, although granting all the qualitative differences cited by Panksepp, have concluded that these deeper brain structures subserving “core” affects, are largely implicit and unconscious, and that consciousness emerges when they interact with higher cortical functions. On the basis of SCR, EMG, and EEG evoked potential evidence elicited by subliminal valence stimuli, Shevrin [20
] has also taken the position that affects can be unconscious as well as conscious.
The first and second author of this paper are thus on opposite sides of the issue of whether affects can be unconscious. It was this difference emerging from intensive discussions that led to a productive adversarial collaboration resulting in the experiments to be reported. The study was conducted in the Ormond and Hazel Hunt Laboratory at the University of Michigan Medical Center under the first author’s direction. The third and fourth authors are colleagues and collaborators of the first author. They participated in setting the problem, designing the experiments, analyzing the data and contributing to writing the paper. They share the first author’s position on unconscious affect.
As we next turn to examining the relevant literature we will measure success by these two criteria: (1) meeting exacting subliminal conditions, (2) absence of a conscious experience similar to the presumed unconscious cause.
1.2. Problems in Subliminal Methodology: Differences between Objective and Subjective Thresholds
There are two fundamental ways to operationalize whether unconscious effects exist when employing the method of subliminal activation: (1) Objective threshold approaches, and (2) Subjective threshold approaches. In the former, more conservative approach, stimulus conditions are arranged such that participants cannot discriminate the relevant stimuli above chance in direct forced-choice tasks. In the latter approach, stimulus conditions are arranged such that participants simply deny being aware of the stimuli, even though they can nonetheless discriminate the stimuli above chance. Accordingly, objective threshold conditions are more stringent than subjective threshold conditions [5
Each approach has potential advantages and disadvantages. As Reingold and Merikle [22
] suggested, objective threshold conditions may inadvertently eliminate not only conscious but unconscious influences as well, thus throwing out the unconscious baby with the conscious bathwater. However, reviews of the literature (see [23
]) suggest that robust effects are readily obtainable under objective threshold conditions—particularly under objective detection
threshold (ODT) conditions, widely agreed to be the most stringent conditions of all for establishing that a stimulus is unconscious [24
]. At the ODT, participants cannot discriminate the sheer presence (vs.
absence) of a stimulus at all.
On the other hand, as Macmillan and Creelman [25
] have noted, basing their critique on signal detection theory, subjective threshold approaches may simply index very faint conscious perceptions that are below participants’ response criterion—that is, so weak that participants possess insufficient confidence to say that they really saw the stimuli. This is a serious problem, which most investigators attempt to surmount by demonstrating convergent qualitative differences between effects obtained under subjective threshold versus
fully conscious conditions. Even then, however, such qualitative differences may simply differentiate effects produced by weak versus
strong conscious stimuli [26
], rather than demonstrate truly unconscious effects.
Nevertheless, subjective threshold methods may capture an important class of genuinely unconscious phenomena—namely, reflectively unconscious events that are phenomenally conscious [28
]. The cocktail party experience is a commonplace example of phenomenal consciousness occurring in the absence of reflective consciousness. One hears and yet does not know one has heard a remark made in a nearby exchange; a moment later when attention is directed to what one has just heard, the content of the remark enters reflective consciousness. At that point one is able to recall that in fact one was aware of the remark at that earlier time, but because attention was directed elsewhere the response criteria for consciousness had not been met. In contrast, objective threshold methods index phenomenally unconscious events. Accordingly, a simple conscious/unconscious dichotomy is insufficient. Rather, a tripartite model emerges: (1) Completely unconscious (i.e.
, both phenomenally and reflectively unconscious); (2) Reflectively unconscious but phenomenally conscious; and (3) Completely conscious (i.e.
, both phenomenally and reflectively conscious). To ensure that we are dealing with phenomenally unconscious affect it thus becomes critical that the stimuli involved should be at the objective detection threshold.
We have selected two studies that come closest to meeting the two criteria, but each for different reasons fall short: Bernat, Bunce, and Shevrin [29
] and Winkielman, Berridge, and Willbarger [30
]. In the Bernat et al.
study in which highly pleasant or unpleasant words according to the Osgood valence criteria [31
] were presented, subliminal exposure conditions were at the objective detection threshold (1 ms with luminance at 5 foot/lamberts). Detection d′
was not significantly different from zero. The main finding was significantly greater positive ERP amplitude across a number of components, particularly strong on the left side for negative
valence stimuli. Supraliminally the same positive amplitude difference was found, but the effects were largely bilateral. It was possible to deliver a reliable stimulus at 1 ms, not attainable with computer screens, because an electronically triggered tachistoscope was used that made such brief flashes possible.
Although this result would seem to support the existence of unconscious affect at the most stringent threshold conditions, it could be argued by those who favor the affects-must-be-conscious view that it remained possible for a change in conscious
mood to have occurred consonant with the valence of the subliminal stimuli, even though participants might have remained unaware of its cause. Essentially, this is the position of Zajonc [12
] and Ohman, Flykt, and Lundqvist [21
], who have published evidence for unconscious emotional stimuli causing
conscious affect changes, while denying that such stimuli cause unconscious affects. Thus the main limitation of the Bernat et al.
] study is that no effort was made to measure conscious mood changes.
This limitation was not present in two experiments reported by Winkielman, Berridge, and Wilbarger [30
] and the experimental method summarized in Berridge and Winkielman [19
]. Conscious mood was measured following the exposure of angry, happy, and neutral faces in a computerized forward and backward masking procedure. The participant’s task was to identify the gender of a supraliminal neutral face, used as the backward mask, presented immediately following the subliminal stimulus. Participants varied in their degree of thirst, the second factor in the experiment. The main result of interest was an interaction between subliminal stimulus valence and amount of a drink consumed. Following the happy face, more drink was consumed than following the angry face by those participants who rated themselves as very thirsty at the start of the experiment. The second result of interest was the absence of any rated effect on conscious mood in response to the two classes of emotional stimuli. Citing evidence from other studies that expressive faces are known to cause affect experiences, the authors inferred from their two findings that unconscious affect experience caused changes in drinking consonant with stimulus valence, while causing no concomitant change in conscious affect experience. Therefore, they were prepared to make the strong claim that affects can be unconscious.
Of critical importance is the claim made by Winkielman et al.
] that on a test of perceptibility performance did not significantly exceed chance. The test only briefly described in the paper was the same procedure as in Winkielman et al.
]. In that report, participants were asked to choose which of two faces had been the one just presented and masked (angry or happy). One of the choices exactly matched the just-presented face; the other was another face differing either in gender or ethnicity, but had the same
facial expression—that is, the response choices were always both happy or both angry. Results were at chance when these comparisons were made. At first glance the results would appear to have met objective threshold conditions for an identification task rather than a detection task.
However, a close examination of the subliminal conditions in the two experiments reported raise serious doubts concerning how unconscious the stimuli actually were with respect to the relevant subliminal dimension facial expression
. Neither gender nor ethnicity was at issue. The crucial difference between the faces—whether the expression was angry or happy—was held constant
; only gender or ethnicity were varied. Accordingly, it is not clear that they achieved the objective identification threshold with respect to the relevant stimulus dimension
, facial expression
. Indeed, when Murphy and Zajonc [33
] had participants make this
, where the response choices varied in expression), using stimulus exposure conditions likely as or more stringent than Winkielman et al.
], they found that performance was clearly well above chance. With this in mind, had Winkielman et al.
] tested perceptibility for the relevant facial expression dimension they would not have attained the objective but rather subjective threshold conditions. Accordingly, although such effects may safely be regarded as reflectively unconscious, there is little reason to regard them as phenomenally unconscious. If so, their interpretation changes crucially—rather than being completely out of awareness, the subliminal faces were phenomenally conscious but weak, such that participants denied their existence. Further, although still noteworthy, if reflectively unconscious effects were what the authors intended to demonstrate, these effects are clearly less pertinent to the fundamental question of whether emotions can be completely (phenomenally) unconscious. Lastly, the absence of a correlation between perceptibility and conscious mood reported by Winkielman et al.
] is irrelevant because they were not correlating the truly relevant dimension, perceptibility of the difference in facial expression. Also, since there is no clear theory of how affective consciousness may operate, it would have been useful if there had been a psychological measure of thirst or appetite for the drink prior to the drinking test (i.e.
, the most relevant experienced affect for the behavioral outcome measure was not monitored).
Although in the Bernat et al.
] study stimuli were presented at the objective detection threshold and produced positive results, it failed to incorporate a measure of conscious mood. The Winkielman et al.
] and the Berridge and Winkielman [19
] study, on the other hand, did measure a large variety of conscious mood variables, but failed to establish that the subliminal stimuli were phenomenally unconscious. The need is for a study that both presents emotional stimuli at the objective detection threshold and
measures conscious mood changes. The study to be reported below undertook to do both. Briefly, in two experiments, two rigorously subliminal blocks of either pleasant or unpleasant words were presented, followed by two blocks of the other valence. During these blocks, participants made dummy bright vs.
dim ratings on each trial (in actuality, this was not manipulated). Participants rated their conscious mood (on valence, arousal, and potency scales) after each individual block.