Affective Priming With Subliminally Presented Pictures

Page created by Tyler May
 
CONTINUE READING
Affective Priming With Subliminally Presented Pictures
         Dirk Hermans and Adriaan Spruyt, University of Leuven, Leuven, Belgium
                         Jan De Houwer, Ghent University, Ghent, Belgium
                        Paul Eelen, University of Leuven, Leuven, Belgium

Abstract Affective priming studies have demonstrated             Eelen, 1994) and forms the basis of an extensive series
that subliminally presented prime words can exert an             of studies (for a recent overview, see Klauer, 1998;
influence on responses towards positive or negative              Klauer & Musch, 2003).
target stimuli. In the present series of experiments, it            In a typical affective priming study, positive and
was investigated whether these findings can be extend-           negative prime stimuli (words or pictures) are present-
ed to pictorial stimuli. Ideographically selected posi-          ed for 200 ms and are followed by a positive or nega-
tive, neutral, and negative picture primes that were             tive target stimulus after an interstimulus interval of 100
sandwich-masked immediately preceded positive or                 ms. This results in a stimulus onset asynchrony of 300
negative target pictures (Experiment 1) or words                 ms (SOA; i.e., the interval between the onset of the
(Experiments 2 & 3). Evaluative categorization                   prime and the onset of the target). Results show that
responses to these target stimuli were significantly             the time needed to evaluate the target stimuli as either
influenced by the valence of the prime. First, it was            “positive” or “negative” is significantly shorter when
demonstrated that high anxious participants were selec-          prime and target share the same valence (positive-posi-
tively slowed when the subliminally presented prime              tive or negative-negative; affectively congruent) as
was negative (Experiments 1 & 2). Second, the affec-             compared to trials on which prime and target are of
tive congruence between primes and targets also exert-           opposite valence (positive-negative or negative-posi-
ed an influence on the responses, but in a direction             tive; affectively incongruent). These data can only be
that is opposite to what is typically observed in affec-         explained if one assumes that the affective valence of
tive priming research. These reverse priming effects             the prime is processed, even though this is not neces-
are situated within a series of recent similar findings,         sary for the task at hand.
and implications for theories of affective priming are              The automatic character of this priming effect is sup-
discussed.                                                       ported by several lines of research, of which a) the
                                                                 subliminal presentation studies and b) the SOA-mani-
                                                                 pulation studies most clearly elucidate this point. With
                                                                 respect to the latter line of research, in a number of
The idea that human organisms evaluate their environ-            studies the interval between the onset of the prime and
ment constantly and unconsciously in terms of “posi-             the onset of the target (SOA) was manipulated (De
tive/pleasant” and “negative/unpleasant” is the                  Houwer, Hermans, & Eelen, 1998; Fazio et al., 1986,
essence of an hypothesis with a long tradition in the            Experiment 2; Hermans et al., 1994, Experiment 1;
history of psychology (Hermans & Eelen, 1997).                   Hermans, Spruyt, & Eelen, 2003). In these studies, for
Nevertheless, it was only in the last two decades that           half of the trials, the interval between the onset of the
experimental evidence started to accumulate to support           prime and the onset of the target was 300 ms, whereas
it. The most compelling and direct support for the               for the other half the SOA was 1,000 ms. If the priming
automatic evaluation hypothesis stems from the seminal           effects observed at SOA 300 are based on consciously
work of Russell Fazio and his colleagues (Fazio,                 controlled processes rather than on automatic process-
Sanbonmatsu, Powell, & Kardes, 1986). These authors              es, one would expect stronger, or at least similar,
employed a modified version of the standard sequential           results if participants are given more time to process
priming paradigm (Neely, 1991) in which the affec-               the prime-target relation, because controlled processes
tive/evaluative relation of the prime-target pairs was           are generally assumed to be more time-consuming than
manipulated. This procedure has become known as the              automatic processes. Nevertheless, although in all three
affective priming paradigm (Hermans, De Houwer, &                studies strong affective congruency effects could be

                               Canadian Journal of Experimental Psychology, 2003, 57:2, 97-114
98                                                                          Hermans, Spruyt, De Houwer, and Eelen

observed at SOA 300, no effects were present at the          affective priming is moderated substantially by SOA,
longer SOA of 1,000 ms. This provides an indirect but        being consistently strong only at a very short SOA
rather strong indication of the automatic nature of the      (67 ms) and decreasing to low levels for SOAs longer
affective priming effect.                                    than 100 ms. With SOAs between 34 and 67 ms and
    Other studies have provided a more fine-grained          masked primes presented for 17 ms, 33 ms, or 50 ms,
analysis of the temporal characteristics of the affective    Draine and Greenwald (1998) were able to demon-
priming effect. Hermans, De Houwer, & Eelen (2001)           strate replicable unconscious affective priming effects.
manipulated the SOA on a within-subjects basis over          Two elements of their study deserve some further con-
five levels (SOA -150, 0, 150, 300, and 450). They were      sideration. First, Draine and Greenwald (1998) were
able to demonstrate affective priming effects for the        able to demonstrate indirect priming effects of the sub-
two shortest, nonnegative SOAs (SOA 0 and SOA 150).          liminally presented stimuli in the absence of direct
For the longer SOAs (e.g., SOA 450), the priming effect      effects (i.e., Experiments 1-3: discriminating masked
disappeared. Similar research has been reported by           prime words from meaningless letter strings;
Klauer and his co-workers (Klauer, Roßnagel, & Musch,        Experiment 4: discriminating positive and negative
1997). Almost simultaneous with, and independent of          masked prime words). This indirect-without-direct
the study by Hermans et al. (2001), they varied the SOA      effect could also be demonstrated for the 17 ms prime
with levels of -100 ms, 0 ms, 100 ms, 200 ms, 600 ms,        duration trials and thereby provides one of the
and 1,200 ms, and obtained an affective priming effect       strongest examples of unconscious semantic process-
for the two shortest nonnegative SOAs (SOA 0 and SOA         ing. Second, the strength of these subliminal effects
100). Based on these studies, Hermans et al. (2001)          (effect sizes ranging from d = 0.34 to d = 1.38) are not
concluded that the activation curve of affective priming     only attributable to the short SOA s, but are likely
has a rather quick onset (SOA 0), with a maximum             increased by the response window procedure and the
around SOA 150, after which the effect rather quickly        sandwich-masking technique. The response window
dissipates. Most probably, an SOA of 300 ms is already       procedure combines the two traditional priming indices
located at the edge of the activation curve. Hence, the      (response speed and accuracy) into one index by
SOA manipulation studies allow for the conclusion that       instructing participants to respond within an interval of
affective priming shows the signature of an automatic,       time that can be described as a response window. It
fast-acting cognitive process.                               has the major benefit of controlling for speed-accuracy
    A second line of research that strongly supports the     trade-off problems by forcing all response latencies to
automatic character of the affective priming effect is a     be relatively similar, thereby avoiding the dilution of
series of studies in which the prime is presented sub-       the priming effect amongst both response latency and
liminally. In this context, Greenwald, Klinger, and Liu      accuracy. The dependent variable with this procedure
(1989) reported three experiments in which primes            is the percentage of correct responses, and the tech-
were dichoptically masked by presenting a random let-        nique typically leads to a large increase in the size of
ter-fragment pattern to the dominant eye, either rapidly     accuracy priming (for an overview, see Klauer &
following the prime (Experiment 1) or presented simul-       Musch, 2003). The sandwich-masking technique sim-
taneously with the prime (Experiments 2 & 3). The            ply entails that the prime is not only followed by a
effectiveness of the masking procedure was demon-            masking stimulus (backward masking), but is also pre-
strated by the participants’ inability to discriminate the   ceded by the same mask. This technique is assumed to
left versus right position of a test series of similarly     allow longer exposure durations of the prime.
masked words. In all three experiments, significant             Combining the response window procedure, the
masked priming effects were obtained: Evaluative deci-       sandwich-masking technique and the use of short SOAs,
sions to evaluatively congruent masked prime-target          other authors were able to successfully replicate the
combinations were significantly faster than those to         subliminal affective priming effects (Abrams &
incongruent combinations. Similar results have been          Greenwald, 2000; Abrams, Klinger, & Greenwald, 2002;
reported by Croizet (1998) and by Otten and Wentura          Klauer, Mierke, & Musch, 2002; Klinger, Burton & Pitts,
(1999).                                                      2000; Musch, 2000, Experiment 5; Otten & Wentura,
    Greenwald, Klinger, and Schuh (1995) compared vis-       1999, Experiment 2). Abrams et al. (2002) demonstrat-
ible and masked priming for evaluative decisions in an       ed that these subliminal priming effects are not based
impressive series of experiments. However, using SOAs        on the automatic activation of a stimulus-response
between 250 and 300 ms, there was only weak evi-             mapping that was acquired during the practice phase,
dence for some amount of priming in the masked con-          but in fact involve the unconscious categorization of
dition. Subsequently, Greenwald, Draine, and Abrams          the prime as either positive or negative (i.e., activation
(1996) report research that suggests that subliminal         of the semantic category). Also, interesting dissocia-
MASKED AFFECTIVE PRIMING                                                                                            99

tions between masked and unmasked affective priming                                Experiment 1
could be established. Musch (2000) demonstrated that
the consistency proportion effect, which refers to the       In this first experiment, the influence of individually
observation that priming effects increase as the propor-     selected and subliminally presented positive, neutral
tion of evaluatively congruent prime-target pairs            and negative prime pictures on the evaluative catego-
increases relative to the proportion of incongruent trials   rization of positive and negative target pictures was
(e.g., Klauer, Roßnagel, & Musch, 1997), can be              investigated. Primes were “sandwich masked,” by
demonstrated at a short SOA for supraliminally present-      employing both a premask as well as a postmask. The
ed stimuli, but disappears when the primes are present-      masking technique was binocular in that primes and
ed subliminally. Greenwald et al. (1996) reported            masks were presented to both eyes. Prime duration
another empirical pattern to differentiate subliminal        was fixed at 20 ms and awareness was assessed follow-
from supraliminal priming. In their research, for supra-     ing the priming task. The set of pictures from which
liminal priming, magnitude of priming was affected by        primes and targets were selected consisted of a series
the relation between prime and target stimuli on the         of complex real-life colour pictures, which has been
just-preceding trial. When the preceding trial was an        successfully employed in supraliminal affective priming
incongruent prime-target pair, supraliminal priming was      studies (e.g., Hermans, De Houwer, & Eelen, 1996;
weaker than when the preceding trial was a congruent         Hermans et al., 1994).
pair. This pattern indicates a form of memory for the
preceding trial’s prime-target configuration. The effect     Method
occurred with visible primes and an SOA of 150 ms. By        Participants. Thirty-five first-year psychology students
contrast, magnitude of subliminal priming was unaffect-      (6 men, 29 women) participated for partial fulfillment
ed by the congruency or incongruency of the preced-          of course requirements. All had normal or corrected-
ing prime-target pair; that is, participants gave no evi-    to-normal vision.
dence of retaining information about the most recent
prime-target configuration.                                  Stimuli and apparatus. A set of 100 colour pictures and
   Based on these data it can be concluded that affec-       100 identical colour slides were used for both primes
tive priming for subliminally presented stimuli is a         and targets throughout the entire experiment. Stimuli
replicable phenomenon. Together with the data of the         were selected in order to obtain a very wide range of
SOA manipulation studies, this provides clear evidence       content as well as affective value (e.g., a mutilated face
for the automatic character of the affective priming         of a young woman, a typewriter on a blue background,
effects and supports the automatic stimulus evaluation       two kittens sitting on a windowsill,...). During the
hypothesis. A characteristic, however, of all studies on     priming procedure, 42 x 64 cm slides were presented,
subliminal affective priming that have been published        at a viewing distance of about 3 m. The masking stim-
until now is that they all have employed word stimuli.       ulus consisted of a colour slide that was composed of
This raises the question of whether these results can be     irregular colour-patches on which a pattern of black
generalized to other types of stimuli. For supraliminal      nonsense figures was superimposed. The masking
affective priming, significant effects have not only been    stimulus was presented 65 x 65 cm in order to mask all
demonstrated for words, but also for stimuli as diverse      primes, of which some were presented in “landscape”
as simple line drawings (Giner-Sorolla, Garcia, & Bargh,     style (64 x 42 cm), whereas others were presented in
1999), complex real life colour pictures (Fazio, Jackson,    “portrait” style (42 x 64 cm).
Dunton, & Williams, 1995; Hermans et al., 1994; Spruyt,          The experiment was run in a dimly lit room. Primes
Hermans, De Houwer, & Eelen, 2002), regular versus           and targets were backprojected on a translucent glass
atonal combinations of tones (Reber, Haerter, &              projection screen, which separated the experimenter’s
Sollberger, 1999), odours (Hermans, Baeyens, & Eelen,        room from the subject’s room. Two random-access
1998), and pictorial stimuli that acquired their evalua-     slide projectors (Kodak Carousel S-RA 2000 and S-RA
tive meaning through a preceding conditioning phase          2500) and a standard slide projector were installed in
(e.g., Hermans, Vansteenwegen, Crombez, Baeyens, &           the former part. All three projectors were equipped
Eelen, 2002; Olson & Fazio, 2001; for an overview, see       with a Compur electronic m3 shutter. Response laten-
Hermans, Baeyens, & Eelen, 2003). In the present             cies were recorded by means of a microphone-activa-
series of studies, it was investigated whether the sub-      ted voice key that stopped a highly accurate Turbo
liminal priming effects can be generalized to real-life      Pascal timer (Bovens & Brysbaert, 1990) upon registra-
colour pictures of positively and negatively valenced        tion of a sound. An IBM compatible XT computer re-
objects, animals, and persons.                               gistered the responses and controlled both slide pre-
                                                             sentation and exposure duration.
100                                                                           Hermans, Spruyt, De Houwer, and Eelen

Procedure. The experiment consisted of three subse-           these figures might be a mere reflection of the charac-
quent phases: the stimulus selection phase, the sublimi-      teristics of our stimulus set, we believe that they are
nal priming procedure, and a postexperimental assess-         also indicative of the fact that there exists a strong posi-
ment of awareness.                                            tive-negative asymmetry in human evaluations (Rozin &
    During the stimulus selection phase participants were     Royzman, 2001; Taylor, 1991) and that there exists a
handed over the 100 colour pictures, and were asked           stronger consistency in ratings of negative stimuli as
to evaluate them on a 21-category scale (-100 = very          compared to positive stimuli (Peeters & Czapinski,
negative/very unpleasant; 0 = neutral; +100 = very pos-       1990). An overall index of consistency makes this
itive/very pleasant). The experimenter stressed that          rather clear: Whereas the stimuli that were selected as a
they should rely on their first, spontaneous reaction         negative prime or target were used for a mean number
towards the picture. To get an idea about the kind of         of 10,7 participants (45%), this was only 5,9 (22%) for
pictures that were included in the set, participants took     the positive stimuli. For the neutral stimuli, this overall
a quick look at the pictures before starting to rate          consistency was also rather low. A total of 35 pictures
them.                                                         were selected as a neutral prime for at least one partici-
    Next, participants were asked to fill out both parts of   pant. The mean number of participants for which
the Spielberger, Gorsuch, Lushene, Vagg, and Jacobs           these 35 pictures were selected was 5,8 (22%). This
(1983) State Trait Anxiety Inventory (STAI; Dutch adap-       lower consistency is possibly the result of a broader
tation by Van der Ploeg, Defares, & Spielberger, 1980).       range of choice, as only 6 stimuli had to be selected for
Meanwhile, out of the participants’ sight, the evaluative     the neutral category (as compared to 15 for the positive
ratings were used to select 18 primes and 18 targets.         and negative categories).
For that purpose the 15 most negatively and the 15                The subliminal priming procedure consisted of a
most positively rated stimuli, together with 6 neutral        series of 18 practice trials and 126 experimental trials.
pictures (rating = 0) were selected. At random, 6 posi-       For the practice trials, primes were randomly assigned
tive, 6 negative and 6 neutral stimuli were designated        to the targets with the restriction that there had to be
as primes, the remaining 9 positively and 9 negatively        equally large sets of affectively congruent (3 Positive-
valenced pictures were chosen as targets. The corre-          Positive, 3 Negative-Negative), affectively incongruent
sponding slides were placed in the proper projector, as       (3 Negative-Positive, 3 Positive-Negative), and control
one random access projector served for the presenta-          prime-target pairs (3 Neutral-Positive, 3 Neutral-
tion of the primes and the second for target presenta-        Negative). The 126 experimental trials consisted of
tion. The masking stimulus was placed in fixed posi-          seven series of 18 trials for which the randomization
tion in the third projector.                                  rule was the same as for the practice trials. The experi-
    Based on the data of 27 of the final 28 participants      mental trials were subsequently subdivided in two
of this experiment (missing data for one participant),        blocks of 63 trials. Within each block the order of the
we checked the consistency with which specific stimuli        trials was completely randomized.
were assigned to the positive, neutral or negative                It was explained that the experiment concerned the
valence category. For the negative stimuli, 38 of the         speed at which people are able to affectively categorize
100 pictures were used as a negative prime or target for      stimulus pictures. It was told that stimuli would be pre-
at least one of the participants. Twelve of these pic-        sented and that the participant’s task consisted in eval-
tures were selected for only a minority of the partici-       uating each of these target stimuli as quickly as possi-
pants (one or two). For the remaining 26 pictures, 3          ble as either “positive” or “negative.” Each trial started
were employed for all participants, 10 for at least 70%       with the presentation of the masking stimulus during
of the participants, and an additional 5 for at least 60%     500 ms (premask). This stimulus was immediately fol-
of the participants. The final 11 pictures were               lowed by the presentation of the prime for a duration
employed for 11 to 33% of our sample. For the posi-           of 20 ms and the masking stimulus (postmask) for 50
tive pictures, the consistency of selected stimuli was        ms. Without delay the target was presented after the
noticeably lower. A total of 68 of the 100 pictures was       postmask, resulting in a stimulus onset asynchrony
used as a positive prime or target for at least one of the    (SOA) of 70 ms. The target remained on the screen
participants. None of these pictures was employed for         until the participant responded by saying “Positive” or
all participants. Twenty-five pictures were selected for      “Negative.” The inter trial interval was always 7 s.
only a minority of the participants (one or two). For             The postexperimental assessment of awareness con-
the remaining 43 pictures, only 8 were used for at least      sisted of three subsequent phases. First, all sponta-
half of the participants, and an additional 7 for at least    neous remarks from the participant concerning possible
40% of the participants. The remainder (35 pictures)          awareness of the primes were recorded during and
were selected for 11 to 33% of our sample. Although           immediately after the experiment. Second, a fixed set
MASKED AFFECTIVE PRIMING                                                                                           101

of questions was asked: 1) “Did you notice anything          primes were considered as “aware” and were excluded
special about this experiment?”, 2) “Did you notice any-     from the analyses. It is important to note that this pro-
thing special about the slide with the colour pattern        cedure did not entail a “forced choice” task to assess
(i.e., the mask)?”, 3) “Did you notice that the slide with   awareness. Given our procedure of assessment, the
the colour pattern was interrupted by another slide for      definition of “unawareness” within the present study is
a short period?”, and in case the participant responded      in line with the idea of “subjective unawareness”
positively to the last question, 4) “Which of those slides   (Cheesman & Merikle, 1985). Within this context,
did you recognize?”. In case the participants did not        Cheesman and Merikle (1986) have argued that the
spontaneously mention the subliminally presented             subjectively defined threshold captures the intuitive
primes and did respond negatively to all five questions,     phenomenological quality of consciousness advocated
the presentations could be regarded as “unconscious”         by Bowers (1984), and that, therefore, this is a satisfac-
following a subjective criterion for conscious percep-       tory way of delineating conscious from unconscious
tion. Nevertheless, it remains possible that some par-       perceptual processing (Öhman, 1999).
ticipants had consciously perceived some of the primes
during the priming procedure, but had forgotten this by      Results
the time of the postexperimental awareness assess-           Awareness assessment. The data of four participants
ment. Hence, it was decided to include an identifica-        had to be excluded from the analyses because of tech-
tion task in this awareness assessment.                      nical problems during the experiment. None of the
    At the start of this identification task, the experi-    remaining 31 participants spontaneously mentioned
menter explained that there indeed had been a stimu-         something related to the presentation of the primes.
lus that was presented for such a short time that con-       Also, none of the participants indicated that they had
scious perception was hampered. The experimenter             noticed anything special about the presentation of the
gave details about the premask/prime/postmask/target         masking stimuli. Most of the participants were actually
sequence and even showed the three slide projectors          surprised that a prime had been presented during the
to demonstrate that there had actually been a series of      presentation of the mask. In response to the identifica-
primes. It was explained that one of the reasons why         tion task, three participants recognized one or more of
the primes might not have been consciously perceived         the primes (the criterion being that the stimulus was at
was that during the priming procedure the participant        least correctly classified as animal, person or object).
was asked to focus all attention to the targets.             One participant correctly recognized one prime, a sec-
Subsequently, a new series of 18 trials was presented,       ond participant correctly recognized two primes, and
during which each of the 18 primes was presented             the third recognized four primes. These three partici-
exactly once. The presentation parameters were the           pants were considered as “aware.” Their data were
same as for the experimental trials. The participant         excluded from the analyses. This left us with 28 partic-
was now asked to focus all attention to the masking          ipants (3 men, 25 women).
stimulus. No response had to be given to the target,
which was now presented for a fixed duration of              Response latencies. The data from trials on which a
600 ms (which equals the mean response latency of            voice key failure occurred or on which an incorrect
previous studies; e.g., Hermans et al., 1994, Experiment     response was given were excluded from the analyses
1). The experimenter explained that 18 different             (2.75%), together with all response latencies shorter
primes would be presented that were also presented           than 250 ms or longer than 1,500 ms (0.17%). The
during the priming phase, and that the participant had       analyses are based on the remaining data (97.1% of all
already seen during the stimulus selection phase. After      observations). The mean reaction times were subjected
each trial the participant was asked what the prime had      to a 2 (Block: trials 1-63 vs. trials 64-126) x 2 (Target:
been. It was stated that it was sufficient to indicate       positive vs. negative) x 3 (Affective Congruence: con-
whether it had been an animal, a person or an object.        gruent vs. control vs. incongruent) ANOVA with repeat-
For each new presentation the experimenter asked to          ed measures on all three variables. As in all following
closely attend to the masking stimulus. If the partici-      experiments, the significance level was set at p < .05
pant indicated that no prime was consciously per-            (two-sided). p-Values will only be reported in the case
ceived, this was written down by the experimenter            of marginally significant (.05 < p < .10) effects.
who instructed to closely watch the following trial.         Greenhouse-Geiser correction of the df was applied
Most participants indicated that they had not seen any-      whenever necessary. Means for the relevant variables
thing or merely a brief flash of light during the presen-    of Experiments 1-3 can be found in Table 1.
tation of the masking stimulus. Participants who were           Against expectation, the ANOVA revealed no main
able to consciously recognize one or more of the             effect of affective congruence, F(2, 54) < 1, MSE = 638,
102                                                                                       Hermans, Spruyt, De Houwer, and Eelen

           Figure 1. Mean response latencies as a function of prime valence (positive/neutral/negative) and state anxiety (high/low)
           (Experiment 1). Confidence intervals following Masson and Loftus (in press).

Mcongruent = 599, Mcontrol = 595, Mincongruent = 599. A                ious (e.g., Eysenck, 1992; Williams, Watts, MacLeod, &
main effect of block, F(1, 27) = 45.13, MSE = 3,114,                   Matthews, 1997), effects like these would not be unex-
reflected a practice effect, Mblock 1 = 618, Mblock 2 =                pected. In essence, this would mean that the observed
577. None of the other main effects or interactions                    effect of affective congruence would actually be based
reached the level of significance.                                     on an effect of prime valence.
    Because of exploratory reasons, two variables were                    To test this hypothesis, a separate ANOVA was con-
added to the design. Based on a median split of both                   ducted for the high state anxious group (N = 14), which
the state and the trait scale of the STAI, participants                now included prime valence as an additional within
were subdivided as low/high state anxious and as                       subjects variable. This 2 (Block) x 3 (Prime: positive
low/high trait anxious. The resulting 2 (State Anxiety:                vs. neutral vs. negative) x 2 (Target: positive vs. nega-
low vs. high) x 2 (Trait Anxiety: low vs. high) x 2                    tive) analysis of variance indeed revealed a main effect
(Block: 1 vs. 2) x 2 (Target: positive vs. negative) x 3               of prime valence, F(2, 26) = 3.38, MSE = 515. Post hoc
(Affective Congruence: congruent vs. control vs. incon-                comparisons (Tukey HSD) revealed that high anxious
gruent) ANOVA with repeated measures on the last                       participants were slower to respond to the target stimu-
three variables revealed a significant interaction involv-             lus if this stimulus was preceded by a negative prime
ing state anxiety and affective congruence, F(2, 48) =                 as compared to a neutral control prime, Mnegative = 598,
3.29, MSE = 611. Inspection of the means learns that                   Mneutral = 587 (see Figure 1). Although the response
high state anxious participants were relatively slower to              latencies were also relatively slowed after a positive
respond to affectively congruent and incongruent trials,               prime, the difference with the neutral controls did not
as compared to control trials, Mcongruent = 596, Mcontrol =            reach significance, Mpositive = 593. A similar main effect
587, Mincongruent = 596, whereas no such pattern was                   was absent when the ANOVA was repeated for the low
present for the low state anxious participants, Mcongruent             state anxious participants (N = 14), F < 1, MSE = 900,
= 602, Mcontrol = 603, Mincongruent = 602.                             Mnegative = 603, Mneutral = 603, Mpositive = 601.
    Although these results seem inconsistent with our
initial predictions, this pattern is fully compatible with             Discussion
an explanation in terms of the affective value of the                  Although no overall affective congruence effect could
stimuli. In fact, affectively congruent and incongruent                be demonstrated, the present data did reveal a signifi-
trials differ from control trials essentially in that the              cant interaction between affective congruence and state
primes in the former types of trials have a strong affec-              anxiety. Further exploration of the data by means of
tive meaning, whereas primes in the control trials are                 simple main effect analyses showed that this interaction
selected because of their neutral affective meaning. In                could be attributed to the fact that high state anxious
the context of research on the automatic processing of                 participants were significantly slowed when the sublim-
emotional stimuli by persons who are high or low anx-                  inally presented prime was of negative valence. This
MASKED AFFECTIVE PRIMING                                                                                                                                     103

        TABLE 1
        Mean Response Latencies (and Standard Errors) for Experiments 1-3, as a Function of Block (1/2), Anxiety (low/high), and
        Affective Congruence (congruent/incongruent/control).

        __________________________________________________________________________________________________________________________________________________
                                                        Block 1                                                            Block 2
                                     _____________________________________________________________________________________________________________________
                                       Low Anxious                High Anxious                           Low Anxious                  High Anxious
                                     _____________________________________________________________________________________________________________________
        Congruence                      M         SE                M         SE                          M         SE                  M         SE
        __________________________________________________________________________________________________________________________________________________
        Experiment 1
           Congruent                   627        21               614       21                         576        16                  577       16
           Incongruent                 625        21               614       21                         580        17                  577       17
           Control                     627        21               600       21                         578        15                  574       15

        Experiment 2
           Congruent                   663        27               701       29                         669        30                  696       33
           Incongruent                 666        29               706       32                         654        30                  687       32
           Control                     673        28               702       31                         660        31                  674       34

        Experiment 3
             Congruent                  750       29               755        29                         719        34                 741        34
             Incongruent                764       33               746        33                         695        27                 727        27
             Control                    743       37               760        37                         748        32                 733        33
        __________________________________________________________________________________________________________________________________________________
        Note. The anxiety variable refers to state anxiety in Experiment 1, and to trait anxiety in Experiments 2 and 3.

observation fits completely within a recent and empiri-                             Kentish, & Bradley, 1993). The group (high anxious vs.
cally founded research tradition on attentional effects in                          low anxious) by stimulus valence (negative vs. neutral)
anxiety disorders (for an overview, see Williams, Watts,                            interaction that is typically observed in these studies is
MacLeod, & Mathews, 1997). Employing research para-                                 identical to the one observed in the present experi-
digms like the emotional Stroop task (e.g., Mathews &                               ment. In both cases the interaction witnesses an inter-
MacLeod, 1985), the visual dot probe task (e.g.,                                    fering influence of the negative valence of the irrele-
MacLeod, Mathews, & Tata,1986), and eye movement                                    vant stimulus (component) on responses towards the
registration (e.g., Hermans, Vansteenwegen, & Eelen,                                relevant stimulus (component). In the present task the
1999), it is now firmly established that high levels of                             relevant stimulus is the target, whereas the colour of
anxiety are associated with a tendency to be distracted                             the word is the relevant stimulus in emotional Stroop
by, as well as to shift attention towards, threat-related                           research. Likewise, the irrelevant stimulus in the prim-
stimuli. These phenomena are also known as selective                                ing procedure is the prime, while this is the semantic
distractibility and attentional bias (Eysenck, 1992) and                            meaning of the word in the emotional Stroop studies.
are based on early phases of automatic information                                      Hence, although the valence of the subliminally pre-
processing. Employing the emotional Stroop task, for                                sented primes did not interact with the valence of the
instance, it has been repeatedly demonstrated that anx-                             target in the present study (i.e., affective priming
ious individuals are relatively slowed in colour-naming                             effect), it did produce a main effect for the high anx-
threat-related words (e.g., for a spider anxious person:                            ious participants, which is similar to the effect that is
spider, cobweb, hairy, crawly) as compared to neutral                               observed in subliminal emotional Stroop studies. This
control stimuli (e.g., table, train, kitchen). This selec-                          by itself provides evidence for the automatic evaluation
tive distractibility or attentional bias has been observed                          of subliminally presented stimuli. Also, although com-
in clinically diagnosed patients with an anxiety disorder                           parable Stroop effects have been demonstrated before,
(see Williams, Mathews, & MacLeod, 1996, for an                                     it is the first time that this is demonstrated for sublimi-
overview) as well as in high trait anxious persons (e.g.,                           nally presented pictorial stimuli. For reasons of clarity
Dawkins & Furnham, 1989), and persons who show                                      it is important to note that the emotional Stroop mecha-
increased levels of state anxiety (e.g., Mathews &                                  nism that is discussed here is different from the Stroop-
MacLeod, 1985). Interestingly, research has shown that                              like account of affective priming that was presented by
this emotional Stroop effect can also be observed when                              Klauer and Musch (2003; see also, De Houwer, 2003).
the negative words are presented subliminally (e.g.,                                    The conclusions that can be drawn from the previ-
MacLeod & Hagan, 1992; MacLeod & Rutherford, 1992;                                  ous study are nevertheless limited. The crucial
Mogg, Bradley, Williams, & Mathews, 1993; Mogg,                                     between-subjects factor (high vs. low anxious) was
104                                                                           Hermans, Spruyt, De Houwer, and Eelen

based on a post-hoc median split on the basis of the          ticipated in the study.
questionnaires that were administered at the end of the       Stimuli and apparatus. Primes were selected from the
experiment. Therefore, in Experiment 2, participants          same set as in Experiment 1. Targets were selected
were explicitly selected on the basis of their ratings on     from a series of 75 nouns, of which one-third was rela-
the State-Trait Anxiety Questionnaire. Based on the           tively positive (e.g., romance, wish, baby), one-third
findings of Experiment 1, and in line with the results of     was relatively neutral (e.g., trumpet, bow, steel), and
emotional Stroop research, we predict that high anx-          one-third was relatively negative (e.g., anxiety, cancer,
ious participants will demonstrate significantly slowed       war). The words were selected from existing norma-
response latencies for trials on which the subliminally       tive studies (Bellezza, Greenwald, & Banaji, 1986;
presented prime was negative as compared to trials on         Brown & Ure, 1969; Toglia & Battig, 1978).
which the prime was neutral. A similar effect is not
predicted for the low anxious group.                          Procedure. The procedure was largely identical to that
                                                              of Experiment 1. One of the differences was related to
                      Experiment 2                            the use of words instead of pictures for the target stim-
                                                              uli. In the stimulus selection phase, a series of 75
In addition to the replication of this differential main      words now also had to be evaluated employing the
effect of prime valence, it was also attempted to             same procedure as described for the pictures in
demonstrate a standard affective priming effect. One          Experiment 1. This rating took place after the rating of
possible explanation for the absence of such a priming        the pictures. The cover story was also slightly different
effect in Experiment 1 might be situated in the use of        from Experiment 1. Because prime pictures were only
pictorial stimuli as targets. Possibly, the affective         presented subliminally during the priming phase, this
impact of these supraliminally presented positive and         part of the experiment was now introduced as a study
negative pictures was too strong for evaluative catego-       on the evaluative processing of words (instead of pic-
rization responses to be influenced by the affective          tures).
impact of the subliminally presented primes. In the              A second difference with the procedure of
subsequent study we therefore employed words                  Experiment 1 was that the number of trials was
instead of pictures as target stimuli.                        reduced from 126 (7 x 18) to 108 (6 x 18). This
   As already mentioned, a second difference with             allowed a more accurate comparison of priming
Experiment 1 was that participants were now selected          effects within the first and second half of the experi-
on the basis of their ratings on an anxiety question-         ment. Because of the uneven number of presentation
naire. More specifically, participants were selected on       series in Experiment 1, the first nine trials of the fourth
the basis of the trait anxiety subscale of the STAI. In       presentation series were counted within the first exper-
addition, the experiment was run about three weeks            iment block, whereas trials 10 to 18 from that series
before the start of their final exams. This was done          were counted within the second experiment block. An
because it can be reasonably assumed that at that             even number of presentation series now allowed to
moment most students also display heightened levels           divide the experiment in two blocks of three presenta-
of state anxiety. Previous studies have shown that dif-       tion series (3 x 18 trials), which were identical with
ferential effects of trait anxiety are strongest when state   respect to the number of trials within each of the differ-
anxiety is relatively high (e.g., MacLeod & Rutherford,       ent prime-target combinations.
1992). This exam-stress procedure has been success-
fully employed in previous studies (MacLeod &                 Results
Mathews, 1988; MacLeod & Rutherford, 1992).                   Awareness assessment. For one participant no data
                                                              could be collected because of technical problems dur-
Method                                                        ing the experiment. None of the remaining 37 partici-
Participants. Participants were selected on the basis of      pants spontaneously mentioned something related to
the STAI-trait scores, which were obtained during a nor-      the presentation of the primes. Also, none of the par-
mative study for the whole group of first-year psychol-       ticipants indicated that they had noticed anything spe-
ogy students (N = 365; see also Hermans, 1994). Due           cial about the presentation of the masking stimuli. In
to gender differences in trait anxiety as indexed by          response to the identification task, four participants rec-
questionnaires like the STAI, 30 female students who          ognized one or more of the primes (the criterion being
had the highest score on this scale, and 30 female stu-       that the stimulus was at least correctly classified as ani-
dents who obtained the lowest score, were invited to          mal, person or object). Three participants correctly
participate in this study. Participation was completely       recognized one prime, a fourth participant correctly
voluntary. A total of 38 students (i.e., 63%) finally par-    recognized two primes. These four participants were
MASKED AFFECTIVE PRIMING                                                                                                         105

        Figure 2. Mean response latencies (and 95% confidence intervals) as a function of affective congruence (congruent/con-
        trol/incongruent) and block (1/2) (Experiment 2). Confidence intervals following Masson and Loftus (in press).

considered as “aware” and their data were excluded                     12.26, MSE = 2,540. High trait anxious participants eval-
from the analyses. The final group consisted of 15                     uated negative targets quicker than positive targets
high trait anxious and 18 low trait anxious participants.              (Tukey HSD; Mpositive = 705, Mnegative = 683). A reverse,
                                                                       but only marginally significant, effect was observed for
Response latencies. The data from trials on which a                    the low trait anxious participants, M positive = 657,
voice key failure occurred or on which an incorrect                    Mnegative = 671.
response was given were excluded from the analysis                        Because of our a priori hypothesis concerning the
(3.45%), together with all response latencies shorter                  differential processing of negatively valenced stimuli by
than 250 ms or longer than 1,500 ms (0.98%). The                       high and low anxious persons, an additional 2 (Trait
analyses are based on the remaining data (95.6% of all                 Anxiety) x 2 (Block) x 2 (Target) x 2 (Prime: neutral vs.
observations). The mean reaction times were subjected                  negative) ANOVA with repeated measures for the last
to a 2 (Trait Anxiety: low vs. high) x 2 (Block: 1 vs. 2)              three variables was conducted. The crucial Trait
x 2 (Target: positive vs. negative) x 3 (Affective                     Anxiety x Prime interaction was significant, F(1, 31) =
Congruence: congruent vs. control vs. incongruent)                     4.23, MSE = 1,318. For the high trait anxious group, a
ANOVA with repeated measures for the last three vari-                  priori contrasts showed slower reaction times after sub-
ables.                                                                 liminally presented negative primes as compared to
   As in Experiment 1, there was no main effect of                     neutral primes, F(1, 31) = 5.75, MSE = 1,318, Mnegative =
affective congruence, F(2, 62) < 1, MSE = 996, Mcongruent              704, Mneutral = 688 (see Figure 3). For the low trait anx-
= 682, Mcontrol = 677, Mincongruent = 678. There was,                  ious participants, no similar difference was found, F <
however, a significant Block x Affective Congruence                    1, Mnegative = 664, Mneutral = 667. It is also important to
interaction, F(2, 62) = 4.33, MSE = 875 (see Figure 2).                note that a similar ANOVA that included positive primes
Tukey HSD a posteriori contrasts revealed that in the                  instead of negative primes did not reveal a comparable
second block of trials response latencies were signifi-                Prime x Trait Anxiety interaction (F < 1; for the high
cantly slower for affectively congruent trials as com-                 anxious participants: Mpositive = 691, Mneutral = 688; for
pared to neutral control trials, Mcongruent = 682, Mcontrol            the low trait anxious group: Mpositive = 662, Mneutral =
= 667. Contrasts involving incongruent trials did not                  667). The effect of prime valence is thus specific for
reach the level of significance, Mincongruent = 671. Also,             negatively valenced primes.
for the first experimental block none of the compar-
isons was significant, Mcongruent = 682, Mcontrol = 687,               Discussion
Mincongruent = 686.                                                    First, the present data replicate the finding of
   Besides the Block x Affective Congruence interac-                   Experiment 1 that high anxious participants are rela-
tion, the analysis revealed a significant interaction                  tively slowed in responding to the target stimuli when
between trait anxiety and target valence, F(1, 31) =                   this target is preceded by a subliminally presented neg-
106                                                                                     Hermans, Spruyt, De Houwer, and Eelen

                 Figure 3. Mean response latencies as a function of prime valence (neutral/negative) and trait anxiety
                 (high/low) (Experiment 2). Confidence intervals following Masson and Loftus (in press).

ative stimulus. This observation can be situated within               iment was that a dichoptical masking procedure was
the broader study of attentional bias and selective dis-              used in Experiment 3, instead of a binocular procedure.
tractibility for negative/threatening stimuli in high anx-            In the dichoptical method, the prime and masking stim-
ious persons. An important aspect of our results is that              ulus are no longer presented to both eyes. Rather, the
the subliminal emotional Stroop effects that have been                masking stimulus is presented to the dominant eye,
demonstrated for word stimuli (e.g., MacLeod &                        whereas the prime is presented to the nondominant
Rutherford, 1992) can be generalized to rather complex                eye. This method has been successfully employed in
pictorial material. In a sense, these data do already                 other subliminal affective priming studies (e.g., Croizet,
support the idea that, at least by some individuals, sub-             1998; Greenwald et al., 1989, 1995). It is assumed that
liminally presented pictorial stimuli are automatically               masking effects employing the dichoptical procedure
evaluated.                                                            are based on binocular rivalry (e.g., Wolfe, 1986), with
   In addition, a significant main effect of affective con-           the dominant eye taking precedence in the perception
gruence could be observed for the second experimen-                   of almost simultaneously presented stimuli, and that it
tal block. Quite unexpectedly, however, the direction                 leads to masking at a more central level, rather than at
of this effect is opposite of what is typically observed              a peripheral level (e.g., at the level of the retina)
in affective priming research, and is unpredicted on the              (Holender, 1986). A pilot study with this method also
basis of all present models of affective priming (e.g.,               revealed that the presentation of the prime could be
De Houwer, Hermans, Rothermund, & Wentura, 2002;                      extended from 20 to 30 ms. Because this study primar-
Klauer & Musch, 2003). Nevertheless, since 1997, simi-                ily aimed at testing whether the reversed priming effect
lar reversed priming effects have been reported more                  could be replicated, selection of participants was no
than once for both subliminal (e.g., Banse, 2001;                     longer based on the scores of anxiety questionnaires.
Wentura, 2001) and supraliminal (e.g., Glaser & Banaji,
1999; Klauer, Rossnagel, & Musch, 1997; Maier, Berner,                Method
& Pekrun, 2003) affective priming studies. But, given                 Participants. A total of 16 first-year psychology stu-
that no such contrast effects had been reported at the                dents (14 women) participated for partial fulfillment of
time Experiment 2 was conducted, the primary aim of                   course requirements. All had normal or corrected-to-
Experiment 3 was to test whether this reversed priming                normal vision.
effect could be replicated.
                                                                      Stimuli and apparatus. The stimulus material was iden-
                      Experiment 3                                    tical to Experiment 2. Because of the dichoptical mask-
                                                                      ing procedure, polarizing filters were screwed on the
The most important difference with the previous exper-                shutters of the slide projectors that were used to pre-
MASKED AFFECTIVE PRIMING                                                                                                 107

                       Figure 4. Mean response latencies as a function of affective congruence (congruent/con-
                       trol/incongruent) and Block (1/2) (Experiment 3). Confidence intervals following Masson
                       and Loftus (in press).

sent the primes and the masking stimulus. Two similar              Response latencies. The data from trials on which a
filters were mounted in the apparatus behind which the             voice key failure occurred or on which an incorrect
participants placed their head. An in-height adjustable            response was given were excluded from the analysis,
chin rest as well as an adaptable head rest assured that           together with all response latencies shorter than 250 ms
the participants’ left and right eye were exactly in front         or longer than 1,500 ms. The analyses are based on
of the two (left and right) polarizing filters. The filters        the remaining data (97.9% of all observations). The
were rotated so that the prime could be presented                  mean reaction times were subjected to a 2 (Trait
exclusively to the participant’s nondominant eye, and              Anxiety: low vs. high) x 2 (Block: 1 vs. 2) x 2 (Target:
the mask exclusively to the dominant eye.                          positive vs. negative) x 3 (Affective Congruence: con-
                                                                   gruent vs. control vs. incongruent) ANOVA with repeat-
Procedure. The procedure was largely identical to that             ed measures on the last three variables. Again, the
of Experiment 2. After the stimulus selection phase,               main effect of affective congruence failed to reach the
participants now filled out both parts of the STAI .               level of significance, F (2, 24) = 1.17, MSE = 2,014,
Subsequently, they were asked to take their place                  M congruent = 741, M control = 746, M incongruent = 733.
behind the dichoptical apparatus and to adjust the                 None of the three a priori contrasts concerning the
height of the chair and chin rest, and to adapt the head           affective congruence variable revealed a significant dif-
rest so that they could sit comfortably and look straight          ference.
through the left and right filter. Next, eye dominance                In line with Experiment 2, however, in the present
was tested following a brief procedure described by                study a Block x Affective Congruence interaction could
Greenwald et al. (1989; p. 37), and the filters of the             again be observed, F(2, 24) = 4.03, MSE = 979 (see
apparatus were oriented accordingly. Primes were                   Figure 4). For the second experimental block, the con-
now presented for 30 ms, in stead of 20 ms.                        trasts revealed faster responses for affectively incongru-
                                                                   ent trials as compared to the neutral control trials, F(1,
Results                                                            12) = 8.92, MSE = 1,349, Mcontrol = 740, Mincongruent =
Awareness assessment. The data of two participants                 711. Also, as compared to the affectively congruent tri-
were excluded from the analyses, because they were                 als, the incongruent trials revealed marginally signifi-
able to correctly identify three primes during the identi-         cant faster response latencies, F(1, 12) = 3.76, p = .076,
fication task. On the basis of their scores on the STAI-           MSE = 1,333, Mcongruent = 730. The congruent trials did
trait questionnaire, the remaining participants (N = 14)           not differ from the control trials, F(1, 12) = 1.48, n.s..
were subdivided in two equally large groups of low                 Also, for the first experimental block, none of the three
and high trait anxious participants.                               comparisons between the levels of the affective con-
108                                                                           Hermans, Spruyt, De Houwer, and Eelen

gruence variable was significant, Fs < 1, Mcongruent =         enced responding to the positive or negative targets.
752, Mcontrol = 752, Mincongruent = 755.                          Unlike Experiment 2, this effect now also interacted
    The interaction between block and affective congru-        with trait anxiety. Although the priming effect was in
ence was mediated by trait anxiety in a triple interac-        the same direction for the high trait anxious partici-
tion, F(2, 24) = 5.80, MSE = 979. Simple interactions at       pants, the facilitated responses for incongruent trials
each of both levels of the trait anxiety variable revealed     were particularly present in the low trait anxious
that the interaction between block and affective con-          group. Similarly, the main effect of prime valence was
gruence was restricted to the low trait anxious partici-       also only present in the low anxious group. Although
pants, F(2, 12) = 11.47, MSE = 815, and was absent for         not incompatible with what is typically observed in
the high trait anxious, F < 1 [means for the relevant          research on affective processing in persons high or low
conditions in Block 2 were Mincongruent = 695, Mcontrol =      in trait anxiety, the slowed responding after negative
748 (low trait anxious), and Mincongruent = 727, Mcontrol =    primes for high anxious participants (see Experiments 1
733 (high trait anxious)].                                     and 2) could not be replicated here.
    To assess the impact of prime valence, the data were
analyzed using a 2 (Trait Anxiety: low vs. high) x 2                              General Discussion
(Block: 1 vs. 2) x 2 (Target: positive vs. negative) x 3
(Prime: positive vs. neutral vs. negative) ANOVA with          Previous affective priming studies demonstrated that
repeated measures on the last three variables. The             the affective meaning of subliminally presented prime
analysis revealed a reliable interaction between block         words can facilitate (inhibit) responses to affectively
and prime valence, F (2, 24) = 4.09, MSE = 621.                congruent (incongruent) target stimuli (e.g., Draine &
Supplementary contrasts indicate that in the second            Greenwald, 1998). The results of the present series of
experimental block responses are faster when the sub-          experiments indicate that this finding can be extended
liminally presented prime was positive as compared to          to subliminally presented pictorial stimulus materials.
when the prime was neutral, F(1, 12) = 5.69, MSE = 993,        The primes that were used in Experiments 1-3 consist-
Mpositive = 720, Mneutral = 740, Mnegative = 721. No similar   ed of colour pictures that varied in content, complexity,
differences were observed for the first 54 experimental        and affective meaning (e.g., a guitar, a corpse with a
trials, Mpositive = 749, Mneutral = 752, Mnegative = 759.      slit throat, a fish with beautiful colours). The conclu-
    As was the case for the previously mentioned Block         sion that the affective meaning of these subliminally
x Affective Congruence interaction, this Block x Prime         presented stimuli was processed by the participants,
interaction also interacted with the trait anxiety vari-       stems from two lines of observations.
able, F(2, 24) = 7.82, MSE = 621. Additional simple                First, the results of Experiments 1 and 2 reveal an
interactions revealed that the significant interaction         emotional Stroop-like effect. High anxious participants
between block and prime was present for the low trait          responded significantly slower to the target stimuli,
anxious participants, F(2, 12) = 10.52, MSE = 646, but         when the prime had a negative valence. For
not for the high trait anxious participants, F(1, 12) =        Experiment 1, this effect was based on a post-hoc
1.01, MSE = 596. The pattern of results for the low trait      analysis of the data. This pattern could, however, be
anxious was similar to the pattern that was revealed by        fully replicated when, in Experiment 2, high and low
the overall Block x Prime interaction.                         anxious participants were invited on the basis of their
                                                               scores on the trait anxiety subscale of the STAI. This
Discussion                                                     effect can be interpreted as an instance of selective dis-
The results of the present experiment replicated the           tractibility for negatively valenced stimulus materials in
data of Experiment 2, in that a reversed priming effect        high anxious participants (Eysenck, 1992) and fits
was again observed. Affectively incongruent trials             seamless within a related line of research on subliminal
resulted in shorter response latencies as compared to          emotional Stroop studies (for an overview, see Williams
neutral control trials. Like in Experiment 2, this con-        et al., 1996, 1997). Although these data clearly demon-
trast effect emerged in the second experimental block.         strate that subliminally presented pictures can be affec-
Although the direction of this priming effect is com-          tively processed, this conclusion should obviously be
pletely opposite to what is predicted by most affective        limited to the group of high anxious participants. It
priming models, the data can nevertheless be interpret-        might nevertheless be quite possible that the primes
ed as direct support for the hypothesis that subliminally      were affectively processed by all participants, but only
presented pictorial stimuli can be automatically evaluat-      resulted in a selective response inhibition for the high
ed. In fact, irrespective of their direction, these prim-      anxious participants. On the basis of the present stud-
ing effects can only be explained if one accepts that          ies, this can, however, not be confirmed.
the affective meaning of the prime has selectively influ-          A second series of observations that leads to the
MASKED AFFECTIVE PRIMING                                                                                              109

conclusion that the affective valence of the prime stim-        priming effects (e.g., Banse, 2001; Glaser & Banaji,
uli was processed relates to the effects of affective con-      1999; Klauer et al., 1997; Maier, Berner, & Pekrun,
gruence that were observed in Experiments 2 and 3.              2003; Wentura, 2001). For instance, Banse (2001;
Quite surprisingly, the direction of these effects was          Experiments 1 & 2) reported standard congruence
opposite to what is typically observed in affective prim-       priming effects for supraliminally presented names and
ing research, and to what could be expected on the              pictures of friends, romantic partners, and disliked per-
basis of most models of affective priming effects.              sons, but observed reverse priming effects for back-
Irrespective of the theoretical implications of these           ward masked primes. Using supraliminally presented
findings, which we come to discuss in a moment, it is           primes and targets, Glaser and Banaji (1999) reported a
important to note that conclusions with respect to the          consecutive series of six studies that demonstrated
affective processing of the primes are, in principle,           reversed priming effects. A factor that was persistently
independent of the precise direction of the affective           associated with the reversed effects in their studies was
congruency effect. The fact that the evaluative                 prime extremity. Whereas standard priming effects
response latencies to positive and negative target stim-        were observed for affectively “moderate” primes, all six
uli are significantly influenced by the affective relation      studies revealed reversed priming for affectively
between the prime and the target, reveals that the              “extreme” stimuli. In analogy to similar contrast effects
affective meaning of the primes must have been                  in controlled judgment research, the authors argued
processed. In contrast to the previously discussed              that these reversed effects reflect an automatic correc-
selective distractibility effect, this effect was not mediat-   tion for the biasing influence of the prime. It is, how-
ed by trait anxiety (Experiment 2). Moreover, in                ever, not easy to see how this “correction-model” could
Experiment 3, where trait anxiety was introduced on a           account for the reverse priming effects that are found
post-hoc basis, the effect of affective congruence was          for subliminally presented primes. Besides the present
particularly present for the low anxious participants.          studies and the experiments reported by Banse (2001),
Taken together with the data of Experiment 2, this indi-        Wentura (2001) also reported a similar contrast effect
cates that the affective influence of the subliminally          for subliminally presented primes.
presented pictures should not be restricted to the high            According to Wentura, the sign of the priming effect
anxious participants.                                           might be dependent upon the instructions that are
   With respect to the generalizability of our data, it         given with respect to the reaction time task. If partici-
has to be noted that there was an overrepresentation of         pants are asked to react as fast as possible (emphasis
female participants in our samples. Given gender dif-           on speed), their responses will be based on any evi-
ferences in trait anxiety and the incomparability of            dence that is available, including the valence of the
scores on the trait anxiety questionnaire, for                  (subliminally presented) prime. Hence a congruent
Experiment 2 it was even decided to only include                prime will facilitate and an incongruent prime will
female participants. To date, however, there exist no           hamper the correct response (i.e., a congruency effect
empirical data that would suggest that gender is a mod-         emerges). In contrast, when participants are asked to
erator of affective priming effects.                            react as accurately as possible, they will employ a strat-
   Another aspect that deserves some attention is relat-        egy in which responses are based solely on the target.
ed to the stimulus selection procedure that was                 In that strategy, the ease of discrimination between
employed in the present experiments. Unlike most                prime and target becomes a crucial variable. For
other studies on subliminal (affective) priming, the            incongruent prime-target pairs, the valence of the stim-
prime stimuli were individually selected before the             uli can be (implicitly) used to discriminate both stimuli.
actual masked presentation phase. This procedure                For congruent trials, it becomes much harder to dis-
might have familiarized our participants with the stimu-        criminate both stimuli, because they share the same
lus materials. It did not, however, hinder the success          affective quality, which is actually the most prominent
of our masking procedure. Future studies could never-           feature of stimuli in the evaluation task. Hence, in case
theless be aimed at investigating the effects of this fea-      the participant bases his responses on an accuracy
ture of our procedure. We will return to this issue             strategy, contrast effects are predicted because it is
when we discuss the reversed character of the priming           time-consuming to disentangle the activated representa-
effects that were obtained in Experiments 2 and 3.              tion of affectively congruent prime and target and to
   Aside from these general conclusions with respect to         arrive at a decision. By manipulating the instructions,
the affective processing of subliminally presented pic-         Wentura induced either an accuracy-based or a speed-
tures, the question remains how the observed contrast-          based strategy. This between-subjects manipulation
effect should be interpreted. The last five years, sever-       indeed resulted in a standard congruency-effect for the
al other researchers have reported reversed affective           “emphasis-on-speed” group, but also in a significant
You can also read