ATTITUDES AND SOCIAL COGNITION
What’s “Up” With God? Vertical Space as a Representation of the Divine
Brian P. Meier and David J. Hauser
Michael D. Robinson, Chris Kelland Friesen, and
North Dakota State University
“God” and “Devil” are abstract concepts often linked to vertical metaphors (e.g., “glory to God in the
highest,” “the Devil lives down in hell”). It is unknown, however, whether these metaphors simply aid
communication or implicate a deeper mode of concept representation. In 6 experiments, the authors
examined the extent to which the vertical dimension is used in noncommunication contexts involving
God and the Devil. Experiment 1 established that people have implicit associations between God–Devil
and up– down. Experiment 2 revealed that people encode God-related concepts faster if presented in a
high (vs. low) vertical position. Experiment 3 found that people’s memory for the vertical location of
God- and Devil-like images showed a metaphor-consistent bias (up for God; down for Devil). Experi-
ments 4, 5a, and 5b revealed that people rated strangers as more likely to believe in God when their
images appeared in a high versus low vertical position, and this effect was independent of inferences
related to power and likability. These robust results reveal that vertical perceptions are invoked when
people access divinity-related cognitions.
Keywords: metaphor, embodiment, God, Devil, divinity
The Lord is high over all nations, and his glory is higher than the
The great dragon was hurled down—that ancient serpent called the
The concepts of “God” and “Devil” are well known both within
and outside of Christianity. In the United States, a poll of 889
respondents found that 82% believe in God and 60% believe in the
Devil (Harper, 2005). Many non-Christian religions (e.g., Islam or
Hinduism) have divine figures that are quite similar to the concept
of the Christian God and Devil (Lindsay, 2005). People hold these
beliefs about God and the Devil even though these figures do not
exist in a literal sense (i.e., God and the Devil cannot be perceived
by the five senses). This poses challenges in representation, as it is
known that representational processes often if not typically rely on
perceptual processes (Barsalou, 1999).
Communication about the divine, however, is often done
through metaphors that invoke physical characteristics. A meta-
phor is a figure of speech in which one thing is experienced or
understood in terms of another (Lakoff & Johnson, 1980). For
example, God is referred to as “the light of the world” or as a
“father,” whereas the Devil is referred to as the “prince of dark-
ness” or as a “serpent.” These metaphors are thought to exist
because they allow people to communicate about what they cannot
see, hear, taste, touch, or smell (Lakoff & Johnson, 1999).
Metaphors for the divine consistently employ descriptions of
vertical space in both Christian and non-Christian religions (e.g.,
Haidt & Algoe, 2004; Previc, 2006). For example, God is known
as the “most high” or is considered to reside in the “high Heavens,”
which are “high above earth” (Lattimore, 1996). People are said to
actually gaze upward to indicate the location of Heaven (Favazza,
2004). This figurative pairing (i.e., God with a high vertical
location) is common and long-standing throughout Christian his-
tory. For example, since the 6th century, a typical Catholic Mass
involves the singing or reading of a hymn titled “Glory to God in
the Highest” (Richstatter, 2005).
The opposite end of this same vertical dimension (i.e., down or
low) is utilized in metaphors to describe the Devil or the profane.
God was said to have cast the Devil “down out of Heaven”
(Lattimore, 1996). In turn, the Devil is thought to be a “lowly”
figure often residing in the “underworld” or, in other words, “deep
down in the earth” (Favazza, 2004, p. 91). Although metaphors for
the divine and the profane obviously aid communication, we
suggest that theories of cognition predict that they are more than
simple communication devices, but suggest a deeply perceptual
mode of concept representation.
In many contexts, we learn about what things are like through
our senses. For example, strawberries are red and taste sweet,
raindrops are cool and make us wet, and marbles are round and feel
Brian P. Meier and David J. Hauser, Department of Psychology, Get-
tysburg College; Michael D. Robinson, Chris Kelland Friesen, and Katie
Schjeldahl, Department of Psychology, North Dakota State University.
Correspondence concerning this article should be addressed to Brian P.
Meier, Department of Psychology, Gettysburg College, Gettysburg, PA,
17325. E-mail: firstname.lastname@example.org
Journal of Personality and Social Psychology, 2007, Vol. 93, No. 5, 699 –710
Copyright 2007 by the American Psychological Association 0022-3514/07/$12.00 DOI: 10.1037/0022-3518.104.22.1689
smooth. God and Devil concepts are different in that they cannot
be perceived through the senses. Thus, to have any appreciation for
what God and the Devil are like, it makes sense that people use
metaphors involving physical domains when describing them
(DesCamp & Sweetster, 2005). That is, because we are sensory-
based creatures, we often use sensory-based metaphors to describe
abstract concepts (i.e., concepts that do not have a concrete phys-
This argument converges with views concerning the conceptu-
alization of abstract concepts. Lakoff and Johnson (1980, 1999;
also see Gibbs, 2006) contended that abstract thoughts are possible
because they rely on our capacity for sensory-based metaphor.
They believe that metaphors allow us to communicate with other
individuals and to represent (i.e., to depict or make sense of)
concepts that would otherwise have no link to physical experience.
Lakoff and Johnson (1980, 1999) further contended that human
thought processes are structured on the basis of metaphor; that is,
the manner in which people encode, store, and retrieve information
is thought to be grounded in metaphor. Thus, metaphors may
underlie our most basic representation of concepts, such that
merely thinking about certain concepts cannot be done without
activating relevant perceptual metaphors.
The perspective of Lakoff and Johnson (1980, 1999) is consis-
tent with theories related to the embodied nature of representation.
Proponents of these theories contend that cognition, rather than
being abstract and amodal, is inherently body based (e.g., Bar-
salou, 1999; Gibbs, 2006; Niedenthal, Barsalou, Winkielman, &
Krauth-Gruber, 2005; Wilson, 2002). In this view, off-line cogni-
tion (i.e., cognition that occurs in the absence of environmental
input) is often body based. That is, thinking about a specific
stimulus (e.g., driving a car) in the absence of that stimulus
activates the same perceptual circuits used to interact with the
stimulus as a physical entity (Wilson, 2002).
Barsalou’s (1999) perceptual symbol systems (PSS) model has
been particularly influential in social and cognitive psychology.
Barsalou noted that typical cognitive models of representation
suggest that sensorimotor information is transduced into abstract
and arbitrary “amodal” codes, which are further processed during
subsequent cognitive operations. Although these amodal codes
may be useful in explaining artificial intelligence systems, they
may fail to explain human cognition. In the PSS, sensorimotor
information is acted on directly by higher level cognitive processes
(i.e., without being transduced). Barsalou contended that our per-
ceptual systems capture sensorimotor information during the on-
line processing of a stimulus (e.g., the motor act of grasping when
reaching for a handle). Later, during the off-line conceptualization
of the stimulus (e.g., in the absence of the handle), these same
perceptual systems reenact the sensorimotor state (e.g., grasping).
The PSS (Barsalou, 1999) thus suggests that both abstract and
concrete concepts often involve embodied modes of cognition. A
study by Stanfield and Zwaan (2001) provides an illustration of
embodied representational processes. These researchers asked par-
ticipants to read sentences about objects in particular spatial ori-
entations. After each sentence, participants had to decide whether
a subsequent picture included the mentioned object. Participants
were faster at determining whether an object was in the picture if
the orientation of the object was the same in both the sentence and
the picture. For example, participants were faster at determining
that a vertically drawn pencil was in the preceding sentence if the
sentence read “John put the pencil in the cup” rather than “John put
the pencil in the drawer.”
Experiments by Meier, Robinson, and Clore (2004) illustrate the
manner in which the embodied cognition perspective (e.g., Bar-
salou, 1999) may have broader generality in relation to abstract
concepts without a direct physical basis. Meier et al. noted that
people often use color to describe affect (an abstract concept). For
example, good things are described as white (e.g., “a bright day”),
whereas bad things are described as black (e.g., “a dark time”).
Meier et al. had participants evaluate words as positive or negative
as the font color randomly varied across trials. They found that
words with a negative meaning were evaluated faster and more
accurately when presented in a black (vs. white) font, whereas
words with a positive meaning were evaluated faster and more
accurately when presented in a white (vs. black) font. These results
suggest that evaluations activate perceptual cues for affect (i.e., in
this case brightness variations), which speeded consistent evalua-
tions (e.g., good–white) and delayed inconsistent evaluations (e.g.,
What is common to the metaphor-representation view (Lakoff &
Johnson, 1999) and Barsalou’s (1999) PSS is the idea that even
relatively abstract concepts may build on, and thus activate, per-
ceptual representation processes. In the Lakoff and Johnson (1999)
view, these perceptual representation processes are captured by
metaphor, which reveals the embodied mappings involved. In the
PSS view, the focus is on the neurocognitive mechanisms involved
in accessing concepts rather than their metaphorical nature. None-
theless, the general point is that representational processes appear
embodied in nature (Boroditsky & Ramscar, 2002; Meier et al.,
2004; Stanfield & Zwaan, 2001; Stepper & Strack, 1993; for a
review of distinctions among embodied cognition frameworks, see
Wilson, 2002). It is this general embodiment claim that provides
the necessary theoretical background for our predictions, which
involved representations of divinity and the vertical dimension of
The Vertical Dimension, Affect, and Power
Although our focus on vertical space and divinity is entirely
novel, researchers have considered the role that verticality may
play in the representation of affect and power. One common
metaphor pairs affect with up and down (Lakoff & Johnson, 1999;
Meier & Robinson, 2005). For example, happy people are said to
be “feeling up,” whereas sad people are said to be “feeling down.”
Good things are described as being up (e.g., “thumbs up” for a
good movie), and bad things are described as being down (e.g.,
“thumbs down” for a bad movie). In one experiment, Meier and
Robinson (2004) found that participants evaluated words with a
positive meaning faster if the words were presented in a high
vertical position and evaluated words with a negative meaning
faster if the words were presented in a low vertical position.
Conceptually related results were reported by Crawford, Margo-
lies, Drake, and Murphy (2006), and thus there is general evidence
that affect and verticality are systematically linked in implicit
Schubert (2005) conducted a similar series of experiments based
on the common metaphor that suggests power is “up” and power-
less is “down.” He found a comparable effect with words that had
a powerful (e.g., master) and powerless (e.g., servant) meaning.
MEIER, HAUSER, ROBINSON, FRIESEN, AND SCHJELDAHL
Furthermore, Schubert convincingly demonstrated that the vertical
representation of power is independent of affect. His studies are
consistent with Lakoff and Johnson’s (1999) view that the same
perceptual domain (i.e., vertical position) can be co-opted for
multiple representation-related purposes. That is, vertical position
is used to represent affect and power, despite the fact that we can
often detest powerful figures (e.g., dictators) and love powerless
figures (e.g., children). In our experiments, we set out to determine
whether verticality is also used to represent divinity-related con-
cepts and whether these representations would be at least partially
independent of affect and power, thus furthering the view that the
same perceptual domain can be co-opted for multiple distinct
It is not clear whether conceptions of divinity borrow from the
perceptual domain. On the one hand, because God and the Devil
cannot be directly perceived, there may be a strong tendency to use
perceptual metaphors to understand these figures and the divine–
profane dimension. Simply stated, there might be no way to
conceptualize this domain without relying on perceptual metaphor.
On the other hand, as spiritual beliefs are based on faith quite
independently of the perceptual world, it might be that represen-
tations of the divine and profane are independent of the perceptual
representation processes observed in other domains (as hinted at
by Barsalou, Barbey, Simmons, & Santos, 2005). Along these
lines, faith does not require physical proof (e.g., Favazza, 2004),
and it is therefore uncertain whether embodiment would be in-
volved. These alternative possibilities rendered the current domain
an important one in which to examine possible embodiment, as
well as an important one in understanding divinity-related repre-
The Current Experiments
Investigations of the cognitive processes associated with repre-
senting divinity have been practically nonexistent (McCauley &
Whitehouse, 2005; Seybold, 2004). Therefore, we conducted six
experiments to thoroughly examine the extent to which divinity-
related thoughts invoke perceptions of verticality. The different
experiments focused on diverse cognitive phenomena related to
implicit associations, encoding, memory, and judgment, but con-
vergence was expected.
In Experiment 1, we sought to determine whether people im-
plicitly associate up and down with God and the Devil, respec-
tively, through the use of an Implicit Association Test (IAT;
Greenwald, McGhee, & Schwartz, 1998). It was predicted that
there would be such an implicit association. In Experiment 2, we
asked individuals to categorize God (e.g., Creator) versus Devil
(e.g., Satan) words randomly assigned to low versus high positions
on the computer screen. It was predicted that faster encoding times
would be found when God-related words were presented higher on
the computer screen. In Experiment 3, we asked participants to
recall the vertical location at which God- and Devil-like images
were presented. It was predicted that participants would remember
God-related images as being presented higher than they actually
were. In other words, convergence was expected in paradigms
related to implicit associations, speed of encoding, and memory
retrieval for vertical position.
In Experiments 4, 5a, and 5b, we examine the potential social
judgment–related consequences of the encoding and retrieval pro-
cesses examined in Experiments 1 through 3. In Experiment 4, we
examined whether vertical space might have an effect on partici-
pants’ ratings of a stranger’s belief in God. We asked participants
to rate the extent to which strangers believe in God as their images
randomly appeared on the top or bottom of a screen. In Experi-
ments 5a and 5b, we used this same rating paradigm to determine
whether the vertical representation of divinity is independent of
power and likability inferences. It was predicted that verticality
effects on judgments related to belief in God would be somewhat
independent of verticality effects on judgments related to likability
(Experiment 5a) or power (Experiment 5b).
In Experiment 1, we used the IAT (Greenwald et al., 1998),
which is a useful tool for assessing associations among concepts in
an implicit manner (i.e., without directly asking participants). The
IAT requires participants to categorize words as belonging to one
of four concepts. For example, Greenwald et al. (1998) had people
categorize words that described insects or flowers and words that
had a pleasant or unpleasant meaning. The trials of interest occur
when two category names are mapped onto one response button. In
the present example, Greenwald et al. found that participants were
faster to categorize words when insect and unpleasant were
mapped onto one response button and flower and pleasant were
mapped onto another (rather than the reverse mappings). Although
there is some controversy surrounding the mechanisms responsible
for the IAT effect (e.g., Arkes & Tetlock, 2004), researchers
generally agree that it measures the associative strength between
concepts (Fazio & Olson, 2003). In Experiment 1, we hypothe-
sized that associations would favor God– up and Devil–down
mappings relative to the reversed set of mappings.
Participants were 41 Gettysburg College volunteers (15 men and
26 women) with an average age of 18.76 years (SD ⫽ 1.20 years).
Thirty-eight participants were Caucasian (92.7%).
The IAT required participants to categorize a word that was
randomly selected and presented in the center of a 19-in. (48.26-
cm) computer screen (in green 16-point Arial font on a black
background). We used four words from each of four categories:
God (Almighty, Creator, Deity, and Lord), Devil (Antichrist, De-
mon, Lucifer, and Satan), up (high, top, above, and ascend), and
down (low, bottom, below, and descend). We only used four
God-related and four Devil-related stimuli because there are very
few of these words in the English language. Words representing
the God and Devil categories were similar in number of letters and
word frequency (ts ⬍ 1), as were the words representing the up and
down categories (ts ⬍ 1.34). Thus, associations between God–
Devil and up–down could not be due to word frequency consid-
Participants were told to categorize the words as quickly and as
accurately as possible. Category names appeared on the upper left
and upper right of the screen (in white 16-point Arial font). The
VERTICAL SPACE AS A REPRESENTATION OF THE DIVINE
choice to be made was always between “God” and “Devil” or
between “up” and “down.” Participants were told to press the Q
key if the word belonged to the category on the left and to press the
P key if the word belonged to the category on the right. Incorrect
categorizations were followed by the word INCORRECT in red
font for 1.5 s. Correct categorizations were followed by a 150-ms
blank screen. The word remained on the screen until participants
pressed a key. All of the words were presented with the first letter
The IAT had seven blocks, five of which were practice blocks.
The critical comparison was between Block 4 (“God or up” vs.
“Devil or down”; 48 trials) and Block 7 (“Devil or up” vs. “God
or down”; 48 trials). We randomly assigned participants to one of
two versions of the IAT. The only difference between the two was
the presentation of the combined blocks. In the IAT described
above, God was first paired with up and Devil was first paired with
down. In the other version of the IAT, this pairing was reversed,
thus allowing us to rule out potential block order effects.
Participants completed the Nearness to God scale (Gorsuch &
Smith, 1983) after finishing the IAT. Participants in the remaining
experiments completed this same scale. We included this scale to
measure individual differences in belief in God. Participants an-
swered six questions (e.g., “God is very real to me” and “God
exists in all of us”) using a 4-point scale (1 ⫽ strong disagreement;
4 ⫽ strong agreement). The internal consistency was high (␣
We followed the data reduction procedure outlined by Green-
wald et al. (1998). This procedure involved deletion of the first two
trials, replacement of trials with response times that were below
300 ms and above 3,000 ms with these values, deletion of inac-
curate trials (3.2% of trials), and a log-transformation of the raw
latencies. Analyses were performed on these transformed laten-
cies, but the means are reported in raw milliseconds for ease of
To investigate participants’ implicit associations, we performed
a mixed-model analysis of variance (ANOVA) with three factors.
The two between factors pertained to the counterbalanced order of
the critical “combined” blocks and participants’ score on the belief
in God scale (z scored). The within factor pertained to IAT block
(God ⫹ up/Devil ⫹ down vs. Devil ⫹ up/God ⫹ down). The main
effects of IAT order, F(1, 37) ⫽ 3.71, p ⫽ .062, and belief in God,
F ⬍ 1, the interactions between IAT order and IAT block, F(1,
37) ⫽ 2.58, p ⫽ .117, and IAT order and belief in God, F ⬍ 1, and
the three-way interaction among IAT order, IAT block, and belief
in God, F ⬍ 1, were not significant. The main effect of IAT block
was significant, F(1, 37) ⫽ 216.40, p ⬍ .001,
⫽ .85. Partici
pants were faster at categorizing God-related words when they
were paired with up-related words and Devil-related words when
they were paired with down-related words (M ⫽ 709 ms, SD ⫽ 84
ms) rather than when these pairings were reversed (M ⫽ 976 ms,
SD ⫽ 185 ms). This main effect was qualified by an interaction
between IAT block and belief in God, F(1, 37) ⫽ 6.11, p ⫽ .018,
⫽ .14. We estimated the means for participants high and low in
belief in God (⫾1 SD). Participants high in belief in God had a
larger difference between the two blocks (God ⫹ up/Devil ⫹
down, M ⫽ 700 ms; God ⫹ down/Devil ⫹ up, M ⫽ 1,013 ms) than
participants low in belief in God (God ⫹ up/Devil ⫹ down, M ⫽
717 ms; God ⫹ down/Devil ⫹ up, M ⫽ 940 ms).
An IAT revealed that people implicitly associate God and Devil
concepts with up and down, respectively. Furthermore, people
high in belief in God showed a stronger tendency in this regard.
However, this individual difference variable did not moderate the
findings in any of the remaining experiments so it is not discussed
further. In relation to the main effect of IAT block, we recognize
that Experiment 1 did not involve a manipulation of actual verti-
cality independent of linguistic categories. Experiments 2–5b rec-
tify this concern in relation to more direct manipulations of vertical
In Experiment 2, we sought to examine whether a manipulation
of vertical position would affect the encoding of God- and Devil-
related words. Participants were asked to categorize words as
being related to God or the Devil as the words randomly appeared
near the top or bottom of a computer screen. Our hypothesis was
that participants should be faster to categorize a God-related word
when it is presented at the top (vs. bottom) of the screen, whereas
the reverse should occur for Devil-related words.
Participants were 47 Gettysburg College volunteers (21 men and
26 women) with an average age of 18.83 years (SD ⫽ 0.96 years).
Thirty-nine participants were Caucasian (83.0%).
Participants were asked to categorize words in terms of whether
they better matched a God or a Devil category. We used the same
eight God and Devil words as in Experiment 1. Using a screen
resolution of 1,024 pixels ⫻ 768 pixels, the words appeared
centered at Pixels 20 (top) and 748 (bottom) on a 19-in. (48.26-cm)
computer screen (in white 22-point Arial font on a black back-
ground). Each word was shown seven times in each of the two
vertical positions, and this was true for all participants, resulting in
112 trials (i.e., 28 trials for each cell of the Word Type ⫻ Vertical
Each trial started with a cue (a white circle 40 pixels ⫻ 40 pixels
in size), which was presented in the center of the screen for 300
ms. Participants were told to attend to this cue because it would
signal the start of a trial. The centering cue transitioned to a
300-ms blank screen. Then, a randomly chosen God- or Devil-
related word appeared on the top or bottom of the computer screen.
Participants were instructed to press the Q key on the keyboard for
words related to God and to press the P key on the keyboard for
words related to the Devil. If participants were correct, a 500-ms
blank screen occurred until the next trial began. If participants
were incorrect, the word INCORRECT was shown in red font for
1.5 s before the 500-ms blank screen between trials. The word
remained on the screen until participants pressed a key. Although
MEIER, HAUSER, ROBINSON, FRIESEN, AND SCHJELDAHL
each participant received a different random order of words and
locations, each participant saw the same number of words in the
same number of locations.
We deleted the inaccurate trials (5.5% of trials), log-transformed
the response times for correct trials, and then replaced response
times faster or slower than 2.5 standard deviations from the grand
latency mean with these log cutoff scores (Robinson, 2007). Anal-
yses were conducted on log-transformed data, but means are
reported in raw millisecond values to facilitate comprehension. We
then performed a 2 (word type: God and Devil) ⫻ 2 (vertical
location: up and down) repeated-measures ANOVA.
The main effect of word type was significant, F(1, 46) ⫽ 8.29,
p ⫽ .006,
⫽ .15, indicating that participants were faster to
categorize the God-related words (M ⫽ 782 ms, SD ⫽ 106 ms)
than the Devil-related words (M ⫽ 802 ms, SD ⫽ 127 ms). The
main effect of vertical location was also significant, F(1, 46) ⫽
18.92, p ⬍ .001,
⫽ .29, indicating that participants were faster
to categorize words appearing on the top of the computer screen
(M ⫽ 780 ms, SD ⫽ 121 ms) than words appearing on the bottom
of the computer screen (M ⫽ 805 ms, SD ⫽ 112 ms). As hypoth-
esized, these two main effects were qualified by a significant Word
Type ⫻ Vertical Location interaction, F(1, 46) ⫽ 33.64, p ⬍ .001,
⫽ .42. As shown in the top panel of Figure 1
, participants were
faster to categorize God-related words when they appeared on the
top (M ⫽ 758 ms, SD ⫽ 106 ms) versus the bottom (M ⫽ 807 ms,
SD ⫽ 111 ms) of the computer screen, t(46) ⫽⫺7.42, p ⬍ .001,
d ⫽⫺1.08. However, participants’ speed to categorize Devil-
related words did not differ by vertical location (Top: M ⫽ 801 ms,
SD ⫽ 142 ms; Bottom: M ⫽ 802 ms, SD ⫽ 118 ms; t ⬍ 1).
It is hazardous to interpret the interaction in light of the main
effects that were observed (which have been observed in other
studies as well: Meier, Sellbom, & Wygant, 2007; Schubert, 2005),
neither of which is relevant to our interactive predictions. There-
fore, we used the procedures of Rosnow and Rosenthal (1989) to
examine interactive effects after the removal of the main effects
for word type and vertical location. As shown in the bottom panel
of Figure 1, these residual scores revealed that participants were
faster to categorize the God-related words when presented on the
top of the screen, but faster to categorize the Devil-related words
presented on the bottom of the screen, after removal of the two
Experiment 2 revealed that vertical position biased the en-
coding of God- and Devil-related words. Participants were
faster to identify God-related words when the position of the
word matched rather than mismatched the metaphor-consistent
association. Although simple effects did not establish that
Devil-related words were categorized faster when presented
lower on the computer screen, we follow Rosnow and Rosenthal
(1989) in suggesting that main effects may have obscured the
crossover nature of the interaction. Indeed, after removing main
effects, the interaction was very much crossover in nature,
supporting the God-is-up and Devil-is-down considerations that
led to the interactive hypotheses.
Experiments 1 and 2 required participants to explicitly catego-
rize words as belonging to the concepts of God or the Devil. We
expected, however, that mere exposure to God and Devil stimuli,
regardless of whether these stimuli were explicitly categorized or
consisted of words or pictures, would be enough to invoke vertical
metaphor. We thus modified a paradigm used by Crawford et al.
(2006; see also Crawford & Cacioppo, 2002) to examine memory
for spatial locations. Participants were presented with God-like
images, Devil-like images, and images of neutral objects unrelated
to religion (e.g., a spoon). All such images were randomly as-
signed to one of five vertical positions. Later, during a testing
phase, participants saw a centrally presented image and were asked
to recall the location at which the image had previously appeared.
Using neutral images as the baseline, we predicted that God
images would be remembered as occurring higher than they actu-
ally appeared and that Devil images would be remembered as
occurring lower than they actually appeared.
Participants were 33 Gettysburg College volunteers (16 men and
17 women) with an average age of 18.70 years (SD ⫽ 0.73 years).
Thirty-one participants were Caucasian (93.9%).
Top of Sc reen
Bottom of Screen
Top of Screen
Bottom of Screen
Figure 1. Top panel: Mean categorization speed as a function of word
type and vertical location (Experiment 2). Bottom panel: Residual scores
for the Word Type ⫻ Vertical Location interaction (Experiment 2).
VERTICAL SPACE AS A REPRESENTATION OF THE DIVINE
God- and Devil-like images. We searched for God- and Devil-
like images using the image search function of Google.com. It was
difficult to locate God-like images. We found that simply typing
God into the search engine did not result in many usable images.
Therefore, we chose five images that reflected different versions of
Michelangelo’s painting of God (i.e., the figure on the ceiling of
the Sistine Chapel). Interestingly, Devil-like images were much
easier to find. We chose five Devil-like images that depicted a
figure with scaly skin and horns. An example of each image type
is shown in Figure 2.
We note that American popular culture consistently depicts the
Devil as a humanlike figure with scaly skin and horns and God as
a fatherly figure with a long gray beard (e.g., as shown on televi-
sion shows such as The Simpsons). Therefore, we believed that the
images we chose were reasonably identifiable as God and the
Devil. To validate our assumption, we presented these images to
10 participants not involved in any of the experiments. Participants
were presented with the images and given the following instruc-
tions: “Please tell us what you think is being shown in each
picture” (one response blank was provided). We calculated the
frequency with which participants responded to the images with
“God” or “Devil” descriptors. The results for the God-like images
revealed that 91.84% of responses (45 of 49; one nonresponse)
listed by participants were “God” or a type of God such as “Zeus”
or “Poseidon.” The results for the Devil-like images revealed that
94.00% of responses (47 of 50) listed by participants were “Devil”
or a type of Devil such as “Demon,” “Satan,” or “Lucifer.” These
results reveal that participants had a strong tendency to associate
the images with God and Devil figures, respectively, thus validat-
ing this manipulation.
We also wanted to ensure that our findings could not be due to
a systematic bias in gaze direction such that the God-like images
were gazing up and Devil-like images were gazing down (Friesen
& Kingstone, 1998). We therefore asked 10 new participants to
rate the extent to which the eye gaze for the object in each picture
was looking high or low (1 ⫽ eye gaze is very low;5⫽ eye gaze
is neither low nor high;9⫽ eye gaze is very high). We found that
participants rated the Devil-like images as having a higher eye
gaze (M ⫽ 5.24, SD ⫽ 0.72) than the God-like images (M ⫽ 3.92,
SD ⫽ 0.56), t(9) ⫽ 4.98, p ⫽ .001, d ⫽ 1.57. Although there was
a relation between images and gaze direction, it would work
counter to our suggestion that God-like images are assigned to
higher levels of physical space, precisely because the God-like
images were more likely to be gazing downward rather than
Neutral images. We used 20 images of neutral objects unre-
lated to God or the Devil. These images constituted an important
comparison condition against which we could compare placement
of God and Devil images. The neutral images were taken from the
International Affective Picture System (Lang, Bradley, & Cuth-
bert, 1999). The International Affective Picture System is a col-
lection of images that are normed on a variety of dimensions
including pleasure (using a scale ranging from 1 to 9). The images
used in the current study were rated at the mid-range of the
pleasure scale (M ⫽ 5.01, SD ⫽ 0.17). Example content included
a spoon, a cup, a towel, a leaf, and a lamp. In terms of Lang et al.’s
(1999) image numbers, we used the following images: 5740, 6150,
7000, 7002, 7004, 7009, 7025, 7090, 7100, 7175, 7182, 7184,
7185, 7187, 7207, 7217, 7235, 7491, 7830, and 7950.
Participants were told that researchers were interested in mem-
ory. They were informed that they would see images of common
and uncommon objects (no mention was made of the content) and
that they would receive a memory test after all images were shown.
Furthermore, they were told that to increase their attention and
interest, the images would appear in different locations on the
computer screen, but there was no indication that this aspect of
memory for the pictures would be tested.
Figure 2. Examples of the God- and Devil-like images used in Experi-
ment 3. Top: From Michelangelo’s “Creation of the Sun and Moon”
(1508 –1512). In the public domain. Bottom: Devil (2005). Courtesy of
J&M Costumers, Inc., North Hollywood, CA. Reprinted with permission.
MEIER, HAUSER, ROBINSON, FRIESEN, AND SCHJELDAHL
We presented participants with the images one at a time on a
black background on a 19-in. (48.26-cm) computer screen with a
resolution of 1,024 pixels ⫻ 768 pixels. Each image was presented
at a size of approximately 205 pixels ⫻ 154 pixels and was shown
for 1 s, followed by a 2-s blank screen. Participants saw 30 images
in total (5 God-like images, 5 Devil-like images, and 20 neutral
The images were presented in the horizontal center of the
computer screen, but the vertical location of the images varied
from a high to a low position. Using a height resolution of 768
pixels, Pixel 1 represents the top of the screen, and Pixel 768
represents the bottom of the screen. Images were centered in one
of five vertical locations (Pixels 77, 230, 384, 538, and 691). Each
image was presented in only one vertical position, and images
from each of the three categories were presented equally often at
each of the five positions. That is, there was one God-like and one
Devil-like image shown at each location and there were four
neutral images shown at each location. We chose to present more
neutral images than God- or Devil-like images to mask the theo-
retical interest in memory for divine and profane stimuli as well as
to have a stable baseline condition.
During the memory-testing phase of the experiment, participants
were told that they would now see each image again, one at a time,
in the center of the screen. On the right side of the screen was a
white bar that was 51 pixels in width and spanned from the top to
the bottom of the screen (Pixel 1 to Pixel 768). Participants were
instructed to use the mouse cursor to select a point on the vertical
bar that reflected the vertical position at which the center of the
image had been shown previously. They were asked to recall the
vertical location of all 30 images, which received a new random
ordering during the memory test. After a location choice had been
made for a given image, there was a 100-ms blank screen followed
by the next image.
The neutral images were purposely included as a comparison
standard to eliminate general tendencies to recall stimuli higher
than they actually appeared (Crawford et al., 2006). Accordingly,
we subtracted the remembered location of the religious images
from the remembered location of the neutral images presented at
the same location, with positive (negative) numbers reflecting a
higher (lower) vertical placement relative to the neutral controls.
As predicted, the God images were recalled as appearing higher
than the neutral images (M ⫽ 36.04, SD ⫽ 121.14), whereas the
Devil-like images were recalled as appearing lower than the neu-
tral images (M ⫽⫺21.02, SD ⫽ 113.70), with the difference
between these two means being significant, t(32) ⫽ 2.17, p ⫽ .037,
d ⫽ 0.38.
Experiment 3 used pictures, included a neutral comparison
condition, and examined the novel question of whether memory
would be biased in a manner consistent with the idea that God is
up and the Devil is down. Indeed, this was the case, and the effects
were symmetrical, such that God images were recalled to be higher
than they actually were, whereas Devil images were recalled to be
lower than they actually were, both relative to a neutral image
control condition. Thus, we have presented evidence for God– up
and Devil–down mappings in relation to implicit associations,
encoding processes, and memory processes. The remaining exper-
iments focus on the possible implications of such cognitive data
for social judgment.
Experiments 1 through 3 reveal systematic associations between
higher areas of vertical space and divinity-related inferences. The
metaphor-consistent nature of such results was clear, but the rel-
evance of such findings for social judgment processes is uncertain.
Although humans cannot be divine, they can be more or less
faithful. Thus, in Experiment 4, we used a judgment task in which
participants judged the extent to which strangers depicted in im-
ages believe in God. The vertical location of these images was
systematically varied, such that some appeared higher on the
computer screen and some appeared lower on the computer screen.
The verticality effects obtained in the first three experiments
might be limited to representations of God and the Devil relative
to physical beings, who are after all real people rather than ste-
reotyped divinity-related figures that cannot be directly perceived.
It seemed likely to us, however, that vertical position would bias
judgments related to a stranger’s belief in God, but what direction
should this bias take? Whether basing such predictions on
metaphor-related (e.g., Lakoff & Johnson, 1999) or cognitive
embodiment (e.g., Barsalou, 1999) perspectives, we note that both
perspectives emphasize assimilation-related processes, here in
terms of a high vertical location being associated with a stronger
belief in God. In a recent review, we reinforced the assimilation-
related nature of affective metaphor (Meier & Robinson, 2005).
From a broad theory-informed perspective, then, higher targets
should be judged more faithful.
There are, however, specific reasons to entertain the idea that
images placed lower in vertical space would lead to stronger belief
in God ratings. After all, faithful individuals typically kneel down
to pray. These practices, however, are likely done in a symbolic
manner to suggest that one is flawed in relation to God, and in fact
believers typically assume that they will ascend during the course
of life and in the afterlife (Haidt & Algoe, 2004). Thus, low
vertical position adopted during prayer is in service of higher
levels of vertical space in divinity-related terms. This fact, in
combination with the assimilation-related nature of metaphoric
associations (Lakoff & Johnson, 1999; Meier & Robinson, 2005),
led us to predict that participants would give stronger belief in God
ratings for images appearing in a high vertical location.
Participants were 27 Gettysburg College volunteers (12 men and
15 women) with an average age of 18.23 years (SD ⫽ 0.71 years).
Twenty-four participants were Caucasian (88.9%).
We downloaded 100 grayscale images from the AR Face Da-
tabase (Martinez & Benavente, 1998). These images included 50
men and 50 women photographed from the shoulder to the top of
VERTICAL SPACE AS A REPRESENTATION OF THE DIVINE
the head against a white background. The age range appeared to be
from the late 20s to the early 40s. The models in the photographs
exhibited a neutral facial expression. Two example images are
shown in Figure 3.
Participants were told that we were interested in how people
determine the extent to which a stranger believes in God. They
were informed that they would see an image of a stranger and a
rating scale in the center of the screen, and that their task was to
rate the extent to which they thought the particular person believes
in God. The images were presented on a 19-in. (48.26-cm) com-
puter screen at a size of approximately 250 pixels ⫻ 180 pixels (on
a white background). The rating scale was arranged horizontally
with a range from 1 (no belief in God)to6(strong belief in God)
and was vertically centered for all trials. Participants were told to
use the mouse to place the cursor over a number and to press the
left mouse button to enter their rating for a particular trial.
We randomly varied the location of the images such that they
appeared either near the top (centered at Pixel 115) or bottom
(centered at Pixel 653) of the computer screen. We told partici-
pants that the location of the image would vary from trial to trial
because we found that varying the location kept people focused on
and more interested in the task. After each rating, the image
disappeared, and there was a 1,500-ms delay before the next image
appeared. Participants rated each image once for a total of 100
We sought to determine whether participants rated strangers as
having a stronger belief in God when the strangers’ images ap-
peared higher (vs. lower) on the computer screen. As predicted,
participants rated strangers as more likely to believe in God when
their images appeared near the top of the computer screen (M ⫽
3.61, SD ⫽ 0.34) relative to the bottom of the computer screen
(M ⫽ 3.48, SD ⫽ 0.27), t(26) ⫽ 3.00, p ⫽ .006, d ⫽ 0.58.
Experiment 4 did not present images of God or the Devil, but
rather presented images of real people in the context of an inci-
dental manipulation of verticality. Although participants were in-
structed to essentially ignore vertical position when making their
judgments and there were no divinity-related figures involved, the
results strikingly revealed that mere vertical placement of pictures
influenced social judgment processes such that higher placed in-
dividuals were seen to have stronger in belief in God. Even though
humans are quite different from the divine, this result indicates that
vertical representation transfers from the divine to believers in the
divine (i.e., an assimilation-related bias; Meier & Robinson, 2005).
Experiments 5a and 5b sought to build on these findings while
examining issues related to discriminant validity.
Experiments 5a and 5b
Experiments 1 through 4 provide evidence that divinity-related
concepts invoke perceptions of verticality. These experiments,
however, do not allow us to determine whether the divine invokes
verticality because the divine might be positive (Meier & Robin-
son, 2004), powerful (Schubert, 2005), or both. Before reporting
the new experiments, we report results from several pilot tests in
which naı¨ve participants were asked to rate the valence (1 ⫽ very
negative, 5 ⫽ neutral, and 9 ⫽ very positive) and power (1 ⫽ very
powerless, 5 ⫽ neutral, and 9 ⫽ very powerful) of the stimuli used
in Experiments 1 and 2 (Ns ⫽ 10 for valence and power ratings,
respectively) and Experiment 3 (again, Ns ⫽ 10 for valence and
power ratings, respectively). The two sets of pilot participants
involved different naı¨ve samples.
In relation to the word stimuli used in Experiments 1 and 2,
participants rated the God words as more positive (M ⫽ 6.70,
SD ⫽ 1.19) than the Devil words (M ⫽ 1.98, SD ⫽ 0.87), t(9) ⫽
8.08, p ⬍ .001, d ⫽ 2.55, but there was no difference in power
ratings (God words: M ⫽ 7.55, SD ⫽ 1.26; Devil words: M ⫽
7.65, SD ⫽ 1.30), t(9) ⫽⫺0.23, p ⫽ .823. In relation to the
pictures used in Experiment 3, a similar pattern emerged in that
participants rated the God-like images as more positive (M ⫽ 6.72,
SD ⫽ 1.23) than the Devil-like images (M ⫽ 2.36, SD ⫽ 1.84),
t(9) ⫽ 5.30, p ⬍ .001, d ⫽ 1.68, whereas participants’ power
ratings did not significantly differ by image type (God-like images:
M ⫽ 6.54, SD ⫽ 1.37; Devil-like images: M ⫽ 5.80, SD ⫽ 1.10),
t(9) ⫽ 1.82, p ⫽ .102. These results suggest that our effects are
Figure 3. Two of the 100 images used in Experiment 4. From The AR
Face Database (CVC Technical Report Number 24), by A. M. Martinez
and R. Benavente, 1998, http://rvll.ecn.purdue.edu/⬃aleix/aleix_face_
DB.html. Copyright 1998 by A. M. Martinez and R. Benavente. Reprinted
MEIER, HAUSER, ROBINSON, FRIESEN, AND SCHJELDAHL
probably not due to a power difference between God and Devil
stimuli, but they might be partially due to a valence difference.
In Experiments 5a and 5b, we used the same procedure as in
Experiment 4, but we asked individuals to make multiple ratings.
On some trials, participants rated how much strangers believe in
God, and on other trials, participants rated how likable (Experi-
ment 5a) or powerful (Experiment 5b) strangers might be. Using
this within-participants design, if our effects are due to inferences
related to likability or power, then the manner in which vertical
position influences God-belief ratings should correlate with the
manner in which vertical position influences likability and power
ratings. However, if we are correct, there should be no such
correlations, establishing the independence of the present vertical-
ity effects from those related to likability or power inferences.
Experiment 5a. Participants were 66 North Dakota State Uni-
versity volunteers (16 men and 42 women; 8 participants did not
report demographics) with an average age of 18.23 years (SD ⫽
0.71 years). Fifty-three participants were Caucasian (91.4%).
Experiment 5b. Participants were 55 North Dakota State Uni-
versity volunteers (20 men and 34 women; 1 participant did not
report demographics) with an average age of 20.88 years (SD ⫽
5.88 years). Forty-five participants were Caucasian (83.3%).
We used the same stranger images as in Experiment 4. To
balance the Rating Type ⫻ Vertical Location design, 96 (48 male
and 48 female) of the 100 images used in Experiment 4 were
randomly selected for inclusion here.
Participants were told that we were interested in how people
make inferences about others. Participants were informed that we
wanted them to rate images according to one of two dimensions—
the extent to which the person believes in God and the extent to
which the person is likable (Experiment 5a) or powerful (Experi-
ment 5b). We randomly selected 48 images (24 male and 24
female) to use for the God-belief ratings, and the remaining 48
images were used for the likability ratings of Experiment 5a or the
power ratings of Experiment 5b.
In Experiment 4, we programmed the task such that it randomly
selected an image and randomly placed it in one of two vertical
locations. This, of course, meant that different pictures were as-
signed to different vertical locations rather than balanced in a
systematic way across participants. This factor would be unlikely
to affect the results, but we sought a fully balanced scheme in
Experiments 5a and 5b. Toward this end, we created two computer
programs for each experiment such that the pictures assigned to the
higher vertical location in one program were assigned to lower
vertical location in the other. Thus, it was necessarily the case that
half of the participants in a given experiment saw a particular
individual in a high vertical position, whereas the other half of the
participants saw the same particular image in a low vertical posi-
In Experiment 4, we used a numerical rating scale (1 ⫽ no belief
in God and 6 ⫽ strong belief in God). In Experiments 5a and 5b,
we used a continuous rating scale that allowed for more nuance in
the person inferences involved. We placed a white horizontal bar
on the center of the screen that was 35 pixels in height and 750
pixels in length. The endpoints varied according to the rating type
(left end: “no belief in God,” “not at all agreeable and nice,” or “no
power and prestige”; right end: “strong belief in God,” “very
agreeable and nice,” or “very much power and prestige”). Partic-
ipants were told to simply place the mouse cursor at a point that
reflected their rating and then press the left mouse button. This
continuous rating bar results in a 750-point scale (based on the
750-pixel horizontal rating bar). The remaining procedural details
were the same as in Experiment 4.
We first computed the average image rating for each participant
based on vertical location (top and bottom) and rating type (belief
in God vs. likability). We then performed a 2 (vertical location: up
and down) ⫻ 2 (rating type: God belief and likable) repeated-
measures ANOVA on these mean ratings. The main effect of
rating type was significant because the God-belief ratings were
higher (M ⫽ 425.72, SD ⫽ 67.84) than the likability ratings (M ⫽
400.28, SD ⫽ 71.54), F(1, 65) ⫽ 11.93, p ⫽ .001,
⫽ .16. The
main effect of vertical location was not significant (F ⬍ 1).
The interaction between vertical location and rating type was
significant, F(1, 65) ⫽ 51.26, p ⬍ .001,
⫽ .44. As shown in the
top panel of Figure 4, participants rated strangers as more likely to
believe in God when images appeared high on the computer screen
(M ⫽ 439.96, SD ⫽ 71.39) compared with low on the computer
screen (M ⫽ 412.47, SD ⫽ 71.95), t(65) ⫽ 4.66, p ⬍ .001, d ⫽
0.57. However, participants rated strangers as less likable when
images appeared high on the computer screen (M ⫽ 384.64, SD ⫽
75.64) compared with low on the computer screen (M ⫽ 415.92,
SD ⫽ 73.44), t(65) ⫽⫺6.06, p ⬍ .001, d ⫽⫺0.75. Thus, the
effects of vertical location on the two types of person inferences
were opposite in nature.
To examine the independence of verticality effects on belief in
God and likability judgments, we calculated two difference scores
for each participant. One was specific to belief in God ratings and
subtracted low vertical ratings from high vertical ratings (i.e.,
higher scores ⫽ a higher rating for images on top vs. bottom). The
other difference score was calculated in an equivalent manner, but
was specific to trials involving the likability judgments. We found
that verticality effects on belief in God and likability were inde-
pendent of each other, r(66) ⫽⫺.10, p ⫽ .409.
As in Experiment 5a, we first performed a 2 (vertical location:
up and down) ⫻ 2 (rating type: God beliefs and power) ANOVA
on rating means averaged across the cells of the design. The main
effect of rating type was significant because the God-belief ratings
were higher (M ⫽ 432.14, SD ⫽ 62.69) than the power ratings
(M ⫽ 353.47, SD ⫽ 64.64), F(1, 54) ⫽ 78.51, p ⬍ .001,
The main effect of vertical location was significant because the
VERTICAL SPACE AS A REPRESENTATION OF THE DIVINE
ratings on top were higher (M ⫽ 398.04, SD ⫽ 58.42) than the
ratings on bottom (M ⫽ 386.57, SD ⫽ 56.90), F(1, 54) ⫽ 4.74,
p ⫽ .034,
⫽ .08. As shown in the bottom panel of Figure 4, the
interaction between vertical location and rating type was not
significant (F ⬍ 1).
The nonsignificant interaction might suggest that belief in God
and power ratings are made in a similar manner. That is, because
vertical position affected God and power ratings to the same extent
and in a similar direction, power inferences might be systemati-
cally related to inferences related to belief in God. The pilot test
data reported above rule out this potential confound in relation to
the stimuli used in Experiments 1–3, but admittedly Experiment 5b
used a different paradigm. We therefore calculated two difference
scores reflecting verticality effects on belief in God (high condi-
tion minus low condition) and inferences of power (high condition
minus low condition). We found that verticality effects on the two
types of inferences were independent, as there was no correlation
among these difference scores, r(55) ⫽ .03, p ⫽ .813.
We conducted Experiments 5a and 5b to both replicate and
extend the results from Experiment 4. Replication was found in
that higher vertical images were viewed as higher in belief in God
than were lower vertical images, again showing that inferences
related to divinity rely on vertical representation processes. Exper-
iments 5a and 5b extend Experiment 4 by showing that verticality
effects on inferences related to divinity are independent of verti-
cality effects on inferences related to likability (Experiment 5a) or
power (Experiment 5b). Along with pilot data reported in the
introduction to the current experiments, the nonsignificant corre-
lations establish the independence of these effects, even with fairly
large sample sizes.
Lakoff and Johnson (1980, 1999) contended that abstract con-
cepts are represented in metaphor-consistent terms. In other words,
people do not simply communicate via metaphor, they also think
via metaphor. This sort of argument is consistent with Barsalou’s
(1999) PSS model, which proposes that conceptual representation
processes often rely on perceptual representation processes. The
basic tenet of these theories, then, is that human cognition is body
based, meaning that off-line conceptualization is not accomplished
via disembodied abstract symbols, but through the activation of
sensations and perceptions related to those concepts.
In the current series of experiments, we examined whether
cognitions related to divinity are similarly “embodied” in nature.
In six experiments, we obtained results that are consistent with
embodied theories of cognition generally and with the more spe-
cific idea that representation of the divine versus profane “bor-
rows” from the vertical domain of perception. Verticality effects
occurred on a broad range of dependent measures related to
implicit associations (Experiment 1), encoding tendencies (Exper-
iment 2), memory (Experiment 3), and social judgments (Experi-
ments 4, 5a, and 5b). The diversity of these effects reveals a broad
mapping of divinity and verticality, and we discuss several impli-
cations of these findings below.
Implications Related to Vertical Representations of the
Religious concepts are said to be spiritual in nature and based on
faith. Therefore, these cognitions might not be grounded in per-
ception to any considerable extent. On the other hand, divinity-
related concepts might be represented by verticality because they
are abstract in the strongest sense of this word (i.e., they cannot be
directly perceived). Our data support the second view in that
people’s representations of the divine are closely tied to perceptual
representations of verticality. Although our results have nothing to
say about the actual existence of God, our results nevertheless
support the idea that representations of divinity borrow from quite
mundane body-based perceptual processes.
In fact, it seems that our results are consistent with religious
symbolism and ritual. Preachers, priests, or ministers often speak
to their congregations from an elevated platform known as the
pulpit. This practice intentionally or unintentionally builds on the
sorts of perceptual representational processes examined here, with
a higher vertical position suggesting a closer relationship to God.
Our data are also consistent with the frequent observation, in
anecdotes at least, that higher realms of attention seem to promote
experiences of closeness to God. Along these lines, it has been
reported that military pilots and astronauts tend to have experi-
ences of God when flying high above the earth (Gawron, 2004).
Similarly, Previc (2006) contended that an upward eye gaze is
Top of Screen
Bottom of Screen
Top of Screen
Bottom of Screen
Figure 4. Top panel: Mean rating as a function of rating type and image
location (Experiment 5a). Bottom panel: Mean rating as a function of rating
type and image location (Experiment 5b).
MEIER, HAUSER, ROBINSON, FRIESEN, AND SCHJELDAHL
generally associated with experiences of being closer to God. Such
systematic links seem particularly amenable to conceptions of
divinity emphasizing high regions of vertical space.
Affect, Power, and/or Divinity?
In their compelling analysis of metaphor representation pro-
cesses, Lakoff and Johnson (1999) pointed out that multiple con-
cepts seem to borrow from the same perceptual domain, whether
related to brightness, verticality, or closeness, or some other phys-
ical experience (for a review, see Meier & Robinson, 2005). For
example, temperature is used when describing both anger (e.g., “he
was boiling with anger”) and love (e.g., “they had a steamy
affair”). The reliance of both anger and love on heat-related
metaphors, however, is unlikely to be due to the fact that we feel
anger for those whom we love, or vice versa. Rather, it seems that
both anger and love, somewhat independently, are associated with
emotional experiences that are usefully represented in terms of
Consistent with such independent concept–percept mappings,
Schubert (2005) has shown that the vertical representation of
power is independent of the vertical representation of affect. In this
case, verticality is used to represent both power and affect. The
present results reinforce the independence of such mappings. First,
the God and Devil stimuli used in Experiments 1–3 did not differ
in power, but did differ in valence. Second, Experiment 5a found
that participants rated strangers as more likely to believe in God
when their images appeared on the top versus the bottom, whereas
they rated strangers as less likable when their images appeared on
the top versus the bottom. This is a double dissociation supporting
the independence of vertical metaphors related to belief in God
Although similar analyses of Experiment 5b data revealed that
higher vertical positions were associated with inferences related to
both belief in God and interpersonal power to an equal extent, such
effects were independent of each other. Thus, the general point is
that verticality is co-opted to represent divinity, likability, and
power, but these effects are independent of each other. Analyses of
the sort presented in Experiments 5a and 5b may better establish
the independence of multiple metaphor-related mappings, such as
those related to anger and love both drawing from the temperature
domain (Lakoff & Johnson, 1999).
Possible Asymmetries in Perceptual Representations of
God and the Devil
The design of Experiment 2 was somewhat uniquely suited to
examining symmetry or asymmetry in relation to representations of
the God and the Devil. In point of fact, different analyses resulted in
different conclusions. Without control of main effects, God stimuli
were classified faster when occurring high on the computer screen,
whereas Devil stimuli were not classified faster when occurring low
on the computer screen. However, removing the significant main
effects in the analysis, through the use of residual scores of the sort
favored by Rosnow and Rosenthal (1989), revealed that the interac-
tion was crossover in nature. We recognize that there are controver-
sies in interpreting interactions in light of main effects. Therefore,
although we favor the symmetry idea (i.e., God is high and the Devil
is low), it is useful to examine an asymmetrical perspective for the
results reported in Experiments 1–2.
In point of fact, there are at least anecdotal reasons to favor the
asymmetrical perspective. A typical visit to a museum will reveal
that God is often depicted high in paintings, whereas the Devil is
often depicted in a vertical region of space at level with human
beings. Such paintings reinforce Christian doctrine in suggesting
that for the Devil to exert an influence on living human beings, he
has to be on the “same level” as us. A second perspective relates
to the fact that more Americans believe in God than in the Devil
(Harper, 2005), and it also seems likely that they think more often
about God than about the Devil. Thus, God relative to the Devil
may be a more elaborated concept, and therefore more closely tied
to vertical representation processes. Although our data provide
evidence for symmetrical processes in relation to the God versus
the Devil (Experiments 1–3), we recognize that further research
could be more definitive in this regard.
Individual Differences in God Beliefs
Our hypotheses were based on normative vertical associations,
but of course individuals differ in their beliefs in God (Harper,
2005). It may be of interest whether such individual differences
moderate the effects here. From a knowledge perspective, for
example, it could be proposed that more religious individuals have
more knowledge concerning Christian doctrine and might there-
fore be more influenced by vertical mappings related to divinity.
From an affective perspective, too, one might propose that more
religious individuals are likely to have more positive attitudes
toward God and therefore might be more influenced by vertical
mappings related to divinity.
As mentioned earlier, all experiments administered an individ-
ual difference scale that examined one’s belief in God (e.g., “God
is very real to me”; Gorsuch & Smith, 1983). The moderating
effects involving this individual difference scale were nonsignifi-
cant in five of the six experiments (Fs ⬍ 1.30). Thus, there is no
general pattern of moderation related to individual differences in
belief in God. Such results reinforce the normative conclusions of
our investigation. After all, almost everyone in the United States is
familiar with Christian conceptions of God and is therefore likely
to represent God in similar vertical terms. That is, vertical repre-
sentations of divinity appear quite independent of individual dif-
ferences in belief in God.
An additional potential moderator variable relates to culture.
Most individuals ascribe supernatural power to a God, but there are
exceptions. References to a divine force external to the self are
quite ambiguous in the Buddhist and Hindu religions, which tend
to link divinity to a force within the self (e.g., “karma”) rather than
a deistic entity. Nevertheless, an analysis of such religious–
spiritual traditions suggests that links of divinity to verticality are
found as well. For example, it is common in many world religions,
even non-Christian ones, to build worship temples or monasteries
in a manner that extends high into the sky (Previc, 2006).
Also, cross-cultural analyses reinforce the systematic link of
divinity to verticality (e.g., Haidt & Algoe, 2004). For example,
the concepts of karma and enlightenment involve the achievement
VERTICAL SPACE AS A REPRESENTATION OF THE DIVINE
of a higher position in vertical space. These examples suggest that
multiple religions use vertical space when conceptualizing high
levels of spiritual attainment, even within non-Christian religions.
On the basis of such consistent links of divinity to higher vertical
regions of space, we would expect that the present results would be
obtained in non-Christian religions as well, even those in which a
deity plays a lesser role in conceptions of enlightenment.
Six experiments examined the general hypothesis that represen-
tations of divinity would be linked to higher levels of vertical
space. The experiments found support for such predictions in the
areas of implicit associations (Experiment 1), encoding processes
(Experiment 2), memory processes (Experiment 3), and person
perception judgment processes (Experiments 4 –5b). The conver-
gence of the results is significant in highlighting the manner in
which representations of divinity rely on metaphor-linked pro-
cesses related to vertical perception. In sum, it appears that the
phrase “Glory to God in the Highest” is not merely the title of a
common hymn, but a phrase that reveals an important truth about
our thoughts, memories, and social judgments concerning divinity-
Arkes, H. R., & Tetlock, P. E. (2004). Attributions of implicit prejudice, or
“would Jesse Jackson ‘fail’ the Implicit Association Test?” Psycholog-
ical Inquiry, 15, 257–278.
Barsalou, L. W. (1999). Perceptual symbol systems. Behavioral and Brian
Sciences, 22, 577– 660.
Barsalou, L. W., Barbey, A. K., Simmons, W. K., & Santos, A. (2005).
Embodiment in religious knowledge. Journal of Cognition and Culture,
5, 14 –57.
Boroditsky, L., & Ramscar, M. (2002). The roles of mind and body in
abstract thought. Psychological Science, 13, 185–189.
Crawford, L. E., & Cacioppo, J. T. (2002). Learning where to look for
danger: Integrating affective and spatial information. Psychological Sci-
ence, 13, 449 – 453.
Crawford, L. E., Margolies, S. M., Drake, J. T., & Murphy, M. E. (2006).
Affect biases memory of location: Evidence for the spatial representa-
tion of affect. Cognition and Emotion, 20, 1153–1169.
DesCamp, M. T., & Sweetser, E. E. (2005). Metaphors for God: Why and
how do our choices matter for humans? The application of contemporary
cognitive linguistics research to the debate on God and metaphor.
Pastoral Psychology, 53, 207–238.
Favazza, A. (2004). PsychoBible: Behavior, religion, and the holy book.
Charlottesville, VA: Pitchstone.
Fazio, R. H., & Olson, M. A. (2003). Implicit measures in social cognition:
Their meaning and use. Annual Review of Psychology, 54, 297–327.
Friesen, C. K., & Kingstone, A. (1998). The eyes have it! Reflexive
orienting is triggered by nonpredictive gaze. Psychonomic Bulletin and
Review, 5, 490 – 495.
Gawron, V. (2004). Psychological factors. In F. H. Previc & W. R.
Ercoline (Eds.), Spatial disorientation in aviation (pp. 145–195). Re-
ston, VA: American Institute of Astronautics & Aeronautics.
Gibbs, R. W., Jr., (2006). Embodiment and cognitive science. New York:
Cambridge University Press.
Gorsuch, R. L., & Smith, C. S. (1983). Attributions of responsibility to
God: An interaction of religious beliefs and outcomes. Journal for the
Scientific Study of Religion, 22, 340 –352.
Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring
individual differences in implicit cognition: The Implicit Association
Test. Journal of Personality and Social Psychology, 74, 1464 –1480.
Haidt, J., & Algoe, S. (2004). Moral amplification and the emotions that
attach us to Saints and Demons. In J. Greenberg, S. L. Koole, & T.
Pyszczynski (Eds.), Handbook of experimental existential psychology
(pp. 322–335). New York: Guilford Press.
Harper, J. (2005, December 25). Majority in U.S. believes in God. Wash-
ington Times. Retrieved May 3, 2006, from http://www.washtimes.com
Lakoff, G., & Johnson, M. (1980). Metaphors we live by. Chicago: Uni-
versity of Chicago Press.
Lakoff, G., & Johnson, M. (1999). Philosophy in the flesh: The embodied
mind and its challenges to Western thought. New York: Basic Books.
Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (1999). International affective
picture system (IAPS): Technical manual and affective ratings. Gainesville:
University of Florida, Center for Research in Psychophysiology.
Lattimore, R. (1996). The new testament. New York: North Point Press.
Lindsay, J. (2005). The encyclopedia of religion (2nd ed.). Farmington
Hills, MI: Thomson Gale.
Martinez, A. M., & Benavente, R. (1998). The AR face database (CVC
Tech. Rep. No. 24). Retrieved August 15, 2005, from http://
McCauley, R. N., & Whitehouse, H. (2005). Introduction: New frontiers in the
cognitive science of religion. Journal of Cognition and Culture, 5, 1–13.
Meier, B. P., & Robinson, M. D. (2004). Why the sunny side is up:
Associations between affect and vertical position. Psychological Sci-
ence, 15, 243–247.
Meier, B. P., & Robinson, M. D. (2005). The metaphorical representation
of affect. Metaphor and Symbol, 20, 239 –257.
Meier, B. P., Robinson, M. D., & Clore, G. L. (2004). Why good guys wear
white: Automatic inferences about stimulus valence based on color.
Psychological Science, 15, 82– 87.
Meier, B. P., Sellbom, M., & Wygant, D. B. (2007). Failing to take the
moral high ground: Psychopathy and the vertical representation of
morality. Personality and Individual Differences, 43, 757–767.
Niedenthal, P. M., Barsalou, L. W., Winkielman, P., & Krauth-Gruber, S.
(2005). Embodiment in attitudes, social perception, and emotion. Per-
sonality and Social Psychology Review, 9, 184 –211.
Previc, F. H. (2006). The role of the extrapersonal brain systems in
religious behavior. Consciousness and Cognition, 15, 500 –539.
Richstatter, T. (2005). A walk through the mass: A step-by-step explana-
tion. Catholic Update. Retrieved May 15, 2006, from http://
Robinson, M. D. (2007). Lives lived in milliseconds: Using cognitive
methods in personality research. In R. W. Robins, R. C. Fraley, & R.
Krueger (Eds.), Handbook of research methods in personality psychol-
ogy (pp. 345–359). New York: Guilford Press.
Rosnow, R. L., & Rosenthal, R. (1989). Definition and interpretation of
interaction effects. Psychological Bulletin, 105, 143–146.
Schubert, T. W. (2005). Your highness: Vertical positions as perceptual
symbols of power. Journal of Personality and Social Psychology, 89, 1–21.
Seybold, K. S. (2004). God and the brain: Neuroscience looks at religion.
Journal of Psychology and Christianity, 24, 122–129.
Stanfield, R. A., & Zwaan, R. A. (2001). The effect of implied orientation
derived from verbal context on picture recognition. Psychological Sci-
ence, 12, 153–156.
Stepper, S., & Strack, F. (1993). Proprioceptive determinants of emotional
and nonemotional feelings. Journal of Personality and Social Psychol-
ogy, 64, 211–220.
Wilson, M. (2002). Six views of embodied cognition. Psychonomic Bul-
letin and Review, 9, 625– 636.
Received July 23, 2006
Revision received May 19, 2007
Accepted July 1, 2007 䡲
MEIER, HAUSER, ROBINSON, FRIESEN, AND SCHJELDAHL