Introduction
In their seminal series of experiments, Bargh, Chen, and Burrows demonstrated that
activating a trait construct such as "being old" is sufficient to elicit behavioral
effects in the absence of awareness. Participants who had been unknowingly exposed to
words related to the stereotype of old age walked slower when exiting the laboratory than
those who had not been so exposed, and the effect was claimed to occur without any
awareness of the link between exposure and behavior. This striking finding established
that priming may occur automatically and influence behavior with little or no awareness,
and it subsequently generated considerable further research in social psychology.
Doyen, Klein, Pichon, and Cleeremans sought to replicate Bargh et al.'s experiments for
three reasons. First, the finding had been replicated only twice, with neither replication
being exact. One prior replication required participants to rate the walking speed of a
character on paper and found the expected priming effect, but did not require participants
to actually perform any behavior. A second replication aimed at refining Bargh et al.'s
results by exploring substantial variability among participants in their priming effect;
it managed to replicate the results on walking speed, but (like the original study) can be
questioned based on imprecise timing methods. Second, walking speed in the original study
was measured by a confederate posted in the hallway, a method prone to error and bias, as
manual chronometry requires extra precautions that appear to be absent from the original
study. Third, and more critically, the experimenter who administered the task in Bargh et
al.'s study could very well have been aware of whether the participant was in the prime
condition or not, and could have tuned his or her behavior accordingly. Numerous studies
have indicated that the experimenter's expectations can influence participants' behavior
even in the most controlled experimental environments. In Bargh et al.'s study, only 1 out
of 19 participants was found to be aware of the influence of the primes; this finding
formed the basis for the claim that the effect of stereotype activation on behavior is
unconscious. However, participants can remain unaware of the stimulus, unaware of the
response, or unaware of the fact that the stimulus importantly influenced the response,
and these are not equivalent.
A further motivation was conceptual. The assumption that high-level semantic priming can
occur automatically and outside of conscious awareness conflicts with evidence from
cognitive neuroscience suggesting that top-down attention to the prime and bottom-up
stimulus strength are both necessary to produce large patterns of neural activation in
higher association cortices essential for semantic priming to occur. Additionally, studies
on automatic behavior have suggested that participants tend to mimic the experimenter's
behavior, raising the possibility that experimenters' expectations about primed
participants' behavior may have altered their own behavior, which was then communicated
unconsciously to the participants, so that experimenters' expectations could act as an
amplifier of the priming effect.
The present study reports two experiments aimed at replicating the original findings while
improving on their design. In Experiment 1, we sought to replicate Bargh et al.'s effect
using automated rather than manual chronometry. In Experiment 2, we directly manipulated
the beliefs of the experimenters: half were led to think that participants would walk
slower when primed congruently, and the other half were led to expect the opposite.
Experiment 1
Methods
Participants. 120 undergraduate Belgian French-speaking students (aged 20-34, M =
21.9) took part in Experiment 1. Four experimenters (aged 21-24, M = 22.5) were recruited
for this experiment. The experiment was a mandatory component of a practical course in
psychology and was approved by the Ethics Committee of the University of Brussels.
Procedure and Design. Participants were randomly and blindly assigned to either the
Prime or the No-Prime condition. The design replicated Bargh et al.'s study and took place
in a vacant hallway of the Université Libre de Bruxelles building. Two infrared sensors
were hidden in the hallway, separated by 9.75 m; the recording of the sensors' signal was
started before a participant arrived and stopped upon completion of all tasks. The Prime
version of the scrambled sentences task consisted of 30 sentences, each including a word
related to the concept of old age. The No-Prime version was identical except that each
word related to old age was replaced by a neutral word. Participants were told that they
had to take part in a test of their French linguistic skills. Upon completion of the
scrambled sentences task, participants were thanked and dismissed; they then walked toward
the exit through the hallway, triggering the infrared beams. Before reaching the exit,
participants were called back by the experimenter, who pretended to have forgotten to
administer a final task. The debriefing that followed used a contingency funnel
questionnaire to assess participants' awareness on three levels: awareness of the prime,
awareness of the primed behavior, and awareness of the link between the prime and their
subsequent behavior. One question in the awareness assessment was a 4-AFC task in which
participants chose among four pictures representing social categories (athletic person,
Arabic person, handicapped person, and elderly person), the picture most likely to have
been used as a prime.
Experimenters had no prior knowledge of the original experiment and no prior expectations
toward participants' behavior. Each experimenter randomly tested participants from both
conditions and was instructed to interact with each participant according to a strict
script to minimize potential influence. Questionnaires were enclosed in an envelope that
the participant opened, so as to keep each experimenter blind to the participant's
condition throughout.
Results
Walking speed. The results show no significant difference between the Prime (M =
6.27'' SD = 2.15) and the No-Prime (M = 6.39'' SD = 1.11) groups in the time necessary to
walk along the hallway, F (1, 119)<1, η² = .01.
Awareness of the prime. Analysis of responses to the 4-AFC task shows that the Prime
group chose the picture of the elderly person above chance level, whereas the No-Prime
group was equally likely to choose any of the four pictures, χ2 (1) = 5.43, p = 0.023.
Four participants (6.66%) in the Prime condition reported that the sentences were related
to the stereotype of old persons.
Awareness of the effect. No significant difference was found between the Prime (M =
1.7) and the No-Prime (M = 2.68) groups in the deviation of the slider from the initial
position, t (1, 119)<1, d = 0.022.
Awareness of the link. 96% of participants reported that they could not establish a
link between the scrambled sentences task and their subsequent behavior. No experimenter
reported having entertained any specific expectation about participants' behavior.
Discussion
Despite a sample twice as large as Bargh et al.'s and a more precise measurement method,
we were unable to replicate their automatic priming effect on walking speed. This led us
to assume that crucial factors in the paradigm had remained unidentified, and to design
Experiment 2 to explore the role of experimenter expectancy.
Experiment 2
Methods
Participants. 50 new participants (aged 18-30, M = 21.3) took part in Experiment 2.
Ten experimenters (aged 20-24, M = 21.7) also participated; they had neither prior
expectations toward participants' behavior nor knowledge of the original experiment.
Procedure and Design. We used the exact same design and procedure as in Experiment 1,
with two fundamental modifications. First, experimenters' expectations about primed
participants' behavior were manipulated: half of the experimenters were told that the
primed participant would walk slower (the "Slow" condition); the other half were told that
the primed participant would walk faster (the "Fast" condition). Experimenters'
expectations were shaped through a one-hour briefing and persuasion session prior to the
first participant's session. The first participant each experimenter tested was a
confederate covertly instructed to act in the manner expected by the experimenter;
participants' condition (Prime or No-Prime) was made salient to the experimenter. Second,
experimenters were also asked to measure participants' walking speed using a manual
stopwatch (hereafter "subjective timing") in addition to the infrared sensors ("objective
timing"). The absolute difference between the two, the error level, serves as a measure of
experimenter accuracy. Experimenters followed a strict script and were instructed not to
reveal to the participant the expected result of the prime on subsequent behavior. The
intraclass coefficient for each key dependent variable was always very low and not
significantly different from 0 (p > .30), so observations were treated as independent.
Results
To examine dependencies between observations due to the experimenter, we computed the
intraclass coefficient for each key dependent variable; coefficients were always low and
non-significant (p > .30). We distinguish three components in the timings: subjective
timings, objective timings, and error level.
Subjective timings. For subjective timings, there was a significant interaction
between participants' condition and experimenters' condition, F(3, 46) = 18,82, p<.001, η²
= .295. Primed participants walked significantly slower than No-Prime participants when
tested by an experimenter expecting them to walk slowly (M = 7.25'' SD = .33 vs. M =
6.73'' SD = .32), F(1, 24) = 12,32, p = .002, η² = .339; this replicates Bargh et al.'s
effect. Strikingly, the effect was reversed in the Fast experimenter condition: Primed
participants walked significantly faster than No-Prime participants (M = 5.8'' vs. M =
6.43''), F(1, 24) = 7,55, p = .012, η² = .274.
Objective timings. For the objective timings, we observed a difference between the
Prime (M = 6.95'' SD = .36) and the No-Prime (M = 6.52'' SD = .31) groups in the Slow
experimenter condition, F(1, 24) = 7.07, p = .014, η² = .228, and Bargh et al.'s effect is
thus replicated in this condition. There was no difference between the Prime (M = 6.07''
SD = .57) and the No-Prime (M = 6.01'' SD = .39) groups in the Fast experimenter
condition, F(1, 24) = .231, p = .636, η² = .011. We also found a main effect of
experimenters' condition: participants tested by Slow experimenters (M = 6.75'' SD = .43)
walked significantly slower than those tested by Fast experimenters (M = 6.04'' SD = .47),
F(1, 49) = 30.44, p<.001, η² = .404.
Error level. The reverse effect on walking speed for the Fast experimenter condition
is explained by the absolute error: experimenters in the Fast condition were less accurate
for the No-Prime group (M = 0.73'' SD = .38) than for the Prime group (M = 0.41'' SD =
.24), F(1, 24) = 5.819, p = .023, η² = .183. No significant difference was found between
the Prime (M = 0.3'' SD = .17) and the No-Prime (M = 0.49'' SD = .32) groups in the Slow
condition, F(1, 24) = 3.297, p = .08, η² = .113.
Awareness of the prime. As in Experiment 1, the Prime group chose the picture of the
elderly person above chance level, whereas the No-Prime group was equally likely to choose
any of the four pictures, χ2 (1) = 5.62, p = 0.019.
Awareness of the effect. A significant difference was found between the Prime (M of
the deviation = -7.3 SD = 1.35) and the No-Prime (M of the deviation = 2.01 SD = .87)
groups in the Slow condition, F(1, 24) = 12.43, p = .042, η² = .143; no significant
difference was found in the Fast condition, F(1, 24) = .56, p = .38, η² = .015. For the
most part, primed participants were able to recall the concept related to the scrambled
sentences task.
Discussion
Our findings lead us to reconsider the results of Bargh et al.'s experiments. First, in
Experiment 1, despite the use of a larger sample and an experimental procedure devoid of
the limitations present in the original, we were not able to replicate the automatic
effect of priming on walking speed, leading us to assume that crucial factors in this
paradigm had remained unidentified. Second, in Experiment 2 we were indeed able to obtain
the priming effect on walking speed for both subjective and objective timings, but
crucially this was only possible by manipulating experimenters' expectations such that
they would expect primed participants to walk slower. The results cannot be explained
solely in terms of a pure self-fulfilling prophecy effect, as the primed participants did
not walk faster when tested by an experimenter who believed they would walk faster.
Regarding the subjective timings, we obtained a reverse effect on walking speed, with
participants walking faster. This effect can be explained by the error committed by the
experimenters: most likely as a result of their induced expectations, experimenters in the
Fast condition were prone to committing more errors because they expected the need to
capture a fast event and hence tended to be inaccurate in stopping the watch.
General Discussion
Our findings lead us to reconsider the behavioral priming literature with a critical
eye. Priming alone was not sufficient to promote a priming effect on walking speed
comparable to Bargh et al.'s. We also had to manipulate experimenters' beliefs so that
they would expect the primed subjects to walk slower. Experimenters' expectations seem to
provide a favorable context for the behavioral expression of the prime: experimenters who
expect their participants to walk slower behave differently than those who expect their
participants to walk faster, and those subtle behavioral cues appear to be picked up by
participants. Obviously, this interpretation remains tentative, as we do not know how this
process operates.
The present finding is congruent with recent evidence showing that primed behavior is
sensitive to the context in which it takes place. Our results cannot be explained solely
in terms of a self-fulfilling prophecy, as the primes alone were not sufficient, and
primed participants did not walk faster when tested by an experimenter expecting them to
walk faster. Also relevant is participants' awareness: most primed participants were aware
of the social category they had been primed with, and those who actually walked slower
were in good proportion aware of their diminished walking speed. Whether automatic
behavioral priming can occur without awareness thus remains unclear; participants'
awareness could have led them to exert better conscious control over the latter, thereby
impairing its expression.
In conclusion, although automatic behavioral priming seems well established in the social
cognition literature, it seems important to consider its limitations. In line with our
results it seems that these methods need to be taken as an object of research per se
before using them as a tool. The methods of behavioral priming need to be examined
carefully before behavioral priming can be considered an established phenomenon.
- ¹trait construct–מבנה תכונה
- ²confederate–שותף מחקר סמוי
- ³experimenter expectancy–ציפיית הנסיין
- ⁴semantic priming–הטרמה סמנטית
- ⁵scrambled sentences task–משימת משפטים מעורבבים
- ⁶contingency funnel questionnaire–שאלון משפך
- ⁷4-AFC–4-AFC, בחירה מאולצת מארבע חלופות, four-alternative forced-choice
- ⁸intraclass coefficient–מקדם תוך-מחלקתי
- ⁹self-fulfilling prophecy–נבואה המגשימה את עצמה
- ¹⁰behavioral priming–הטרמה התנהגותית