Content uploaded by Russell E Jackson
Author content
All content in this area was uploaded by Russell E Jackson on Jul 08, 2014
Content may be subject to copyright.
Current research sometimes generalizes findings from
largely horizontal distance-perception studies to all dis-
tance perception without considering that height per-
ception may result from processes not shared with other
distance perception (Avraamides, Loomis, Klatzky, &
Golledge, 2004; Foley, Ribeiro-Filho, & Da Silva, 2004;
Wu, Ooi, & He, 2004). Does height perception result from
all of the same processes as does other distance percep-
tion, or does it result from limited specialized processing?
The former may appear more parsimonious, but we should
answer empirical questions such as this experimentally.
Theorized mechanisms for specialized height percep-
tion include differential eye movements (Wundt, 1874, as
cited in Winslow, 1933), differentially excitatory retinal
meridians (Avery & Day, 1969), and discord among reti-
nal, visual, and gravitational orientations (Higashiyama,
1996). Peripheral mechanisms such as these may explain
how we perceive heights, but they fail to explain why we
might perceive heights differently from other distances.
Although height perception is surely accomplished by
some perceptual mechanism, only a causal theory can
suggest why height perception might differ from other
distance perception.
Two current causal theories of specialized height per-
ception are foreshortening of receding horizontals and
gravity theory. Segall, Campbell, and Herskovitz (1966)
maintain that foreshortening of receding horizontals in-
creases perceived vertical length in order to represent
distance accurately, because vertical retinal surfaces can
represent environmentally horizontal receding surfaces.
Gravity theory (see Howard & Templeton, 1966, p. 37) and
other effort-based theories suggest that perceived distance
corresponds with anticipated navigation effort. Gravity
theory suggests that vertical surfaces should appear longer
than horizontal surfaces because heights require greater
energy to navigate than do horizontal surfaces.
Evolved Navigation Theory
Disparity between the modern environment and the envi-
ronments in which humans evolved provides an important
tool for behavior researchers (Jackson, 2004). In this article,
we propose a broad evolutionary approach to investigating
perceptual and navigational systems and then test a subset
of predictions as they apply to height perception. Evolved
navigation theory (ENT) suggests that physical and psycho-
logical navigation processes reflect natural selection from
navigational costs reliably present over evolutionary time.
ENT applied to height perception focuses on the major cost
inherent in vertical or angled navigation: falling.
Falling is more common and costly on vertical surfaces
than on horizontal ones. One might thus expect vertical
surfaces to be processed differently than horizontal sur-
faces, in a way that incorporates this cost asymmetry when
making navigational decisions. An effective mechanism
that we know would decrease navigation of a costly surface
would be to make the surface appear longer, because, intui-
tively and empirically, organisms pursue the nearer of oth-
erwise equivalent navigational goals (Somervill & Somer-
vill, 1977). ENT applied to height perception suggests that
specialized perceptual mechanisms could subvert prefer-
ence for nearer goals by making vertical surfaces appear
longer than horizontal ones. This height overestimation
would facilitate horizontal navigation choices, when avail-
able, helping to avoid falling costs. If height perception in
human ancestors contained genetic variance in accuracy,
then some degree of overestimation likely incurred the
least net cost, over time, because falling inflicts potentially
severe costs.
353 Copyright 2007 Psychonomic Society, Inc.
Evolved navigation theory and the descent illusion
RUSSELL E. JACKSON AND LAWRENCE K. CORMACK
University of Texas, Austin, Texas
Researchers often assume that height perception results from all of the same mechanisms as does other
distance perception (Avraamides, Loomis, Klatzky, & Golledge, 2004; Foley, Ribeiro-Filho, & Da Silva, 2004;
Wu, Ooi, & He, 2004). Evolved navigation theory (ENT) proposes that natural selection has differentiated some
psychological processes, including height perception, in response to the navigational outcome of falling. We
tested predictions from three theories in two experiments. Only ENT predicted greater height perceived from
the top than from the bottom of a vertical surface (because descent results in falls more often than does ascent).
Participants across experiments perceived an average of 32% greater vertical distance when viewing from the
top than when viewing from the bottom. We discuss selected implications and suggest ENT for uniting isolated
findings, including the vertical–horizontal illusion.
Perception & Psychophysics
2007, 69 (3), 353-362
R. E. Jackson, russelljackson@mail.utexas.edu
354 JACKSON AND CORMACK
Falling injuries likely influenced reproductive fitness
in all environments in which humans evolved. The most
effective way to avoid falling costs would be to avoid com-
pletely those surfaces associated with falling. However, if
height navigation did not always result in falls, and if it
reliably held important resources, then height perception
might instead weight navigational decisions by falling
cost likelihood. What important resources might height
navigation have offered reliably over evolutionary time?
Heights offer advantages similar to those suggested for
human bipedalism (Conroy, 1997, p. 227)—namely, in-
creased visibility and access to targets such as animal prey,
plant resources, water, or other humans. Moreover, access
to elevated surfaces, and the increased visibility that they
afford, helps in forming large and detailed cognitive maps.
Second, geologic heights provide escape and defense
from predators and other humans—a major cause of mor-
tality over evolutionary time (Keeley, 1996). Ancestral
humans navigated geologic heights such as hills, steppes,
boulders, and cliffs within normal human ranges and in
migrations leading to modern dispersal. Habitable heights
offer the above benefits at night, when humans sleep and
have reduced ability to see approaching threats. Habita-
tion of heights has occurred in places as diverse as the
Acropolis in Athens, Swayambunath in Nepal, and the
Cliff Palace of Mesa Verde in Colorado.
Third, humans require large amounts of fresh water,
which is often associated with steep surfaces (because
flowing water erodes its substrate and water flows to the
lowest of available surfaces). It is not uncommon that ac-
cessing sources of fresh water, and the food and materials
located therein, requires vertical navigation. Access to the
ocean and its associated food and materials also often re-
quires vertical navigation.
Finally, trees offer access to arboreal prey, eggs, fruits,
nuts, shelter, and material items such as wood and leaves.
Arboreal species who share recent common ancestors
with humans overestimate vertical in relation to horizontal
distance (Dominguez, 1954).
Initial natural selection leading to the hypothesized
height overestimation may have occurred phylogenetically
prior to distinct humans—as suggested by the pres-
ence of vertical overestimation in nonhuman primates
(Dominguez, 1954). The factors outlined above suggest
at least maintenance selection for height overestima-
tion in ancestral and modern humans. Prehuman ances-
tors surely interacted with the above-mentioned heights
(and others) that just as likely selected for the onset of
height overestimation—which selection could then main-
tain in humans, for the same reasons. Ultimately, height
overestimation could persist in any derived lineage, even
without current benefits, either because of the absence of
sufficient genetic variation or because height overestima-
tion inflicted insignificant cost. The root of ENT is that
any species navigating surface orientations that reliably
affect reproductive fitness might face selection for per-
ceptual tendencies such as height overestimation. ENT as-
serts that humans either are, or are descended from, such
organisms.
One ENT Prediction: Falling Cost Asymmetry
Across Position on Vertical Surface
There is an inherent asymmetry in vertical navigation
not present in horizontal navigation that distinguishes an
ENT-derived approach from other theories. Ascending
from the bottom and descending from the top of a height
pose different costs. Ascending expends more energy
than descending (Bassett et al., 1997; Minetti, Moia, Roi,
Susta, & Ferretti, 2002; Teh & Aziz, 2002), but descend-
ing results in falls more often than ascending. Svanstrom
(1974) found that falls on stairs occurred 76% of the time
during descent. Tinetti, Speechley, and Ginter (1988)
found that stepping down resulted in injurious falls among
the elderly four times as often as stepping up (p. 1704).
Haslam and Bentley (1999) found that postal workers fell
while descending steps or inclined drives 15 times as often
as when ascending (p. 39). Cohen and Lin (1991) found
that ladder accidents occurred during descent nearly twice
as often as during ascent (p. 31).
Body mechanics of descent increase its falling risk.
Posture during descent requires leading with less dexter-
ous feet, and it places the eyes posterior to the direction of
travel, thereby reducing visibility. Whereas those ascend-
ing can keep their bodies close to the vertical surface, those
descending must place their bodies out from the vertical
surface in order to be able to see where they are going.
This position results in poorer selection of holds, which
(nonprehensile) feet cannot grip, while simultaneously
positioning one’s body farther from the surface, making it
more difficult to retain grip than during ascent.
Descending inhibits control of velocity and direction
of travel. Ascending necessitates muscle contractions that
pull one’s body weight to a specific point, but descend-
ing necessitates relaxation of muscles to allow partially
controlled drops in the direction of gravity. This produces
poorer control of velocity and poorer selection of hand
and foot placement than does ascending.
Lack of experience on a height when initiating verti-
cal movement also increases costs of falling more during
descent than ascent. The beginning of an ascent or descent
predicts success in traversing the surface. When ascend-
ing, one can predict ascent safety by testing handholds
while holding oneself off the ground with one’s feet, or
by being able to fall a few inches relatively safely. When
descending, however, one cannot predict safety by testing
a surface, because the likeliest cost of failure is to fall the
entire vertical length. The inability to practice descending
forces greater certainty of success to guide the decision to
descend than is the case with the decision to ascend.
Predictions From Three Theories
All of the following differential height perception theories
argue against a viewpoint, which we title unitary distance
perception, in which all distances are perceived via the same
processes with similar outputs. We tested these theories by
comparing participants’ estimates of the height of a surface
against its true vertical length. Specifically, we compared
participants’ estimates from the top of a vertical surface with
height estimates from the bottom of the same surface.
EVOLVED NAVIGATION THEORY AND THE DESCENT ILLUSION 355
Foreshortening of receding horizontals (Segall, Camp-
bell, & Herskovits, 1966) predicts that we adjust percep-
tion of foreshortened surfaces in order to give exocen-
trically veridical distance estimates. Foreshortening of
receding horizontals specifically predicts equal height
estimation from both the top and the bottom of a vertical
surface, as long as the actual distances are equal and are
equally foreshortened.
Gravity theory (see Howard & Templeton, 1966, p. 37)
predicts overall height estimates exceeding the true
height, because vertical surface navigation is more ener-
getically expensive than other distance navigation. Grav-
ity theory specifically predicts greater distance estimates
from the bottom of a vertical surface than from the top,
because ascending is more energetically expensive than
descending.
Evolved navigation theory–derived predictions predict
overall height estimates exceeding true height, because
vertical navigation is associated with falling risk. These
predictions specifically predict a descent illusion, in
which height is overestimated more from the top of a ver-
tical surface than from the bottom, because descent results
in the costs of falling more than does ascent.
We tested predictions from each theory in two experi-
ments, in which we used different methods and settings.
EXPERIMENT 1
Participants estimated an outdoor vertical surface
length while positioned at its top and at its bottom, by ap-
proximating an equal horizontal length.
Method
Sixty-six participants from an introductory psychology course
who reported normal (20:20) or corrected-to-normal vision met a re-
search assistant (RA) in a campus office, completed a consent form,
and answered a few screening questions. The questions screened for
normal vision, body height, and fear of heights. The participants
then proceeded to the outdoor testing site with the RA.
The participants estimated the height of a 14.39-m vertical sur-
face from both its top and its bottom, as illustrated in Figure 1.
The RA randomly determined participant starting-position order
(top first or bottom first) and then positioned participants near the
same lateral point on the vertical surface for both estimates. Using a
plainly worded script, the RA instructed the participants to estimate
the distance from the top railing to the ground directly beneath it.
The participants estimated vertical length by adjusting the distance
from the RA to the vertical surface with hand signals; the RA then
measured the distance back to the vertical surface with a wheeled
measure.
In order to study this phenomenon in an ecologically valid set-
ting, with rich stimuli, we modeled these procedures after previ-
ous outdoor distance-estimation research (Chapanis & Mankin,
1967; Dixon & Proffitt, 2002; Higashiyama, 1996; Higashiyama
& Ueyama, 1988; Yang, Dixon, & Proffitt, 1999). Understandably,
there exists some imprecision in the task and measurement under
such conditions, but an important aspect of these techniques lies in
the invariance between estimation procedures at the top and bottom
positions. Predictions tested here focus on top- and bottom-estimate
differences, and we used equivalent procedures between positions.
We were not aware of any by-products of our methods that would
artifactually alter distance judgments. We also implemented these
techniques as precisely as possible via high selectivity in RA train-
ing and research site choice.
Author R.E.J. trained all RAs, who were not allowed to collect
data until they made measurements within 1 in. (2.54 cm) of mea-
surements by the author and other trained RAs. The measurement
procedures were simple and specific, allowing accurate measure-
ment with little room for interpretation. All of the RAs were blind to
the experimental predictions and were observed unknowingly over
the duration of the experiment to ensure that they followed all of the
procedures precisely. No deviance from the designed techniques
was observed.
The testing environment consisted of a parking garage selected
for ecological validity to cues important to the theories under in-
vestigation. The height was great enough to inflict falling costs, and
both participant positions were not impassably obstructed, such as
with a window or screen. The participants could turn and lean freely
and take as much time, and make as many adjustments, as they de-
sired. Use of an elevator minimized arousal differences between
estimates, and both top and bottom areas consisted of similar asphalt
and cement surfaces.
Results
Figure 2 shows a scatterplot of participant height es-
timates in meters. Each data point in the figure plots the
top estimate versus the bottom estimate for 1 participant.
Arrows represent mean height estimate from the top
(26.54 m, SD 6.83 m) and the bottom (20.61 m, SD
4.65 m). Top and bottom estimates correlated moderately
highly [r(66) .609, p .001]. The dashed lines in each
dimension represent actual height: 14.39 m. Mean esti-
mates exceeding the actual distance suggest height over-
estimation from both positions. Indeed, every participant
overestimated height from the top. Not surprisingly, a
manipulation check against true height indicates height
overestimation from both the top [t(65) 14.462, p
.001] and the bottom [t(65) 10.879, p .001].
The solid line in Figure 2 represents the slope of unity
between the top and bottom estimates. Most data points
in this plot occur above the slope of unity, which illus-
trates the finding of most interest: Participants overesti-
mated height more from the top than from the bottom of
the vertical surface. In this case, 58 of the 66 participants
overestimated the height more from the top position than
from the bottom position. The binomial probability of
this many (or more) data points falling above the slope of
unity, if perceived height was actually the same, is roughly
9.0 10
=11
.
Figure 1. The Experiment 1 participant position during top
(outline) and bottom (solid) estimates. Dotted icons represent the
same participant during each estimate. Dashed lines represent
the horizontal path in estimating the vertical height at each posi-
tion. Figure not drawn to scale.
356 JACKSON AND CORMACK
Figure 3 illustrates mean estimates by position, with
error bars showing 95% confidence intervals. The dashed
line in Figure 3 represents the true height. The mean height
estimate from the bottom was 20.61 1.12 m, whereas the
mean height estimate from the top was 26.54 1.65 m.
The mean estimate was 29% greater at the top than at the
bottom, with a difference of 5.93 m. A 5.93-m position
difference between estimates of a 14.39-m surface sug-
gests a strong effect of position, with mean estimates at
the top equaling 184% of the actual distance.
Not surprisingly, a paired samples t test indicated a
significant difference between participants’ estimates of
length from the top and from the bottom [t(65) 8.865,
p .001]. Participants perceived much greater vertical
distance from the top of the height than from the bottom.
These data support the primary ENT prediction and are
inconsistent with predictions from unitary height percep-
tion and the two other differential height theories.
An independent samples t test indicated that order of
testing (i.e., beginning at the top or at the bottom) failed to
influence top- and bottom-estimate differences [t(64)
1.019, p .312]. Significance tests of Pearson product–
moment correlations failed to indicate that participant
body height correlated with top- and bottom-estimate
differences in men [r(33) =.099, p .584] or women
[r(33) .043, p .813]. Average estimate difference
(5.93 m) exceeded even the greatest body height (2.08 m)
by more than a factor of two. Neither order of testing nor
body height changed participants’ perception of greater
height from the top than from the bottom.
We asked participants to rate their fear of heights in re-
sponse to the question “Do you fear heights?” on a 5-point
Likert scale, with response options of not at all, slightly,
moderately, very much, and intense, possibly irrational,
fear. We were concerned about the chance that screen-
ing for participant height fear might lead participants to
consider heights as more important during the experiment
than they otherwise would have. However, we were unable
to avoid asking about fear of heights, because we wanted
to ensure that no acrophobic participants (characterized
by responding with intense, possibly irrational, fear) par-
ticipated and, as a result, experienced undue stress during
our experiments. The extent to which screening for fear
of heights influenced our methods appears minimal, be-
cause the participants were uninformed about the actual
distances and the existence or position of upcoming es-
timates. We employed a double-blind method concerning
our predictions, and testing order failed to have an impact
on difference scores.
Discussion
Participants perceived greater height when standing
at the top of a vertical surface than when standing at the
bottom. These data are consistent with ENT-derived pre-
dictions, exclusive of predictions from the other causal
height-perception theories. ENT suggests that costs of
falling present in height navigation produced adaptive
processes that we propose resulted in overestimation of
specific geographical distances associated with falling: a
descent illusion.
Although these data are consistent with ENT, we de-
cided to expand this investigation with a second ex-
periment. The first reason was simply to generalize; we
wanted to make sure that nothing specific about the task
or the testing site was responsible for the pattern of ob-
served results.
Figure 2. The Experiment 1 participant distance estimates, by
position, in meters. Arrows indicate mean estimates from the top
and the bottom. Dashed lines indicate actual vertical distance.
The solid line indicates the slope of unity between the top and
bottom estimates.
50
40
30
20
10
Top Estimate (m)
10 20 30
Bottom Estimate (m)
40 50
Figure 3. The Experiment 1 mean participant distance estimate,
by position, in meters. Bars represent 95% confidence intervals
about the means. The dashed line indicates actual vertical distance.
30
20
15
25
10
Estimated Distance (m)
Participant Position
Bottom Top
EVOLVED NAVIGATION THEORY AND THE DESCENT ILLUSION 357
Second, the monocular depth cue of familiar size might
have mitigated the effect in Experiment 1. Since partic-
ipants adjusted the distance from the same RA in both
positions, familiar size could have biased estimates from
both positions to be more similar than they would have
been without a familiar size cue.
Finally, the distance from the participants’ eyes to the
plane of the vertical surface differed between the top
and bottom positions in Experiment 1. The participants
stood out from the surface by about 60 cm when making
estimates from the bottom, because the vertical surface
extended all the way to the ground. However, when view-
ing from the top, the participants’ eyes were positioned
10–20 cm beyond the plane of the vertical surface. We
therefore wanted to collect data in a setting in which the
participants’ eyes could be roughly the same distance from
the vertical surface plane from both the top and bottom
positions.
EXPERIMENT 2
In Experiment 1, participants estimated a fixed verti-
cal interval via a variable horizontal interval. In Experi-
ment 2, the participants estimated a fixed horizontal inter-
val via a variable vertical interval. We also used a small,
unfamiliar ball to denote distance, which presumably pro-
vided less of a familiar size cue than was used in Experi-
ment 1. Further, the study site in Experiment 2 included
an overhang that allowed participant positioning directly
beneath the vertical surface, more closely matching the
viewing geometry from the top position (see Figure 4).
This experiment allowed us to use an environment and
methods different from those used in Experiment 1, for
greater generalizability.
Predictions From Three Theories
The phenomenon of interest is identical between the
two experiments, but predictions of which distances par-
ticipants will perceive as greater are inverted from Ex-
periment 1, because Experiment 2 participants matched a
fixed horizontal with a variable vertical interval:
Foreshortening of receding horizontals predicts, as be-
fore, overall accurate and equal height estimates from the
top and from the bottom.
Gravity theory predicts greater perceived height from
the bottom than from the top, because ascending is more
energetically expensive than descending. The participants
should therefore match the fixed horizontal interval with
a smaller physical vertical interval at the bottom than at
the top.
Evolved navigation theory–derived predictions predict
the same descent illusion of greater perceived height from
the top than from the bottom, because descent results in
falls more than does ascent. The participants should there-
fore match the fixed horizontal interval with a larger ver-
tical interval at the bottom than at the top.
Method
One hundred forty-four participants from an introductory psy-
chology course who reported normal (20:20) or corrected-to-normal
vision met an RA in a campus office and completed the consent
form and screening questions used in Experiment 1. The partici-
pants then proceeded to the outdoor testing site with the RA, who
was trained in the same method used in Experiment 1.
For the testing site, we selected an outdoor campus staircase, for
the reasons of ecological validity and experimental control stated in
Experiment 1.
The participants estimated the length of a 7.20-m horizontal
distance from both the top and the bottom of the 15.90-m vertical
surface illustrated in Figure 4. The RAs did not tell the participants
that the horizontal distance was equal in both positions. The RA ran-
domly determined starting-position order (top first or bottom first).
The RA then positioned participants at the same lateral point on the
vertical surface for top and bottom estimates and verbally instructed
the participants from a plainly worded script.
The participants’ task was to estimate the horizontal distance be-
tween the vertical surface and a point 7.20 m away in both positions
(denoted by a white line or a brick wall). The participants estimated
the top horizontal distance by telling an RA standing next to them to
lower or raise a black, white-speckled 4.5-cm-diameter rubber ball
suspended from clear monofilament fishing line until the distance
to the ball appeared equal to the horizontal length. The participants
changed the distance to the ball while at the bottom position by
showing an “up” or “down” sign to an RA at the top of the height.
The starting position for the ball was at participant chest height in
both positions. The RAs allowed the participants as much time and
as many adjustments as they desired during estimation, then spooled
in the fishing line and took an elevator to the next position. The RAs
measured the participants’ estimates after the testing session.
The RAs initially instructed the participants to estimate the dis-
tance between the ball and the ground for the bottom estimate and
the distance between the ball and a chest-high railing for the top esti-
mate. However, 3 of the first 15 participants verbally commented on
their strong preference to use the distance from their eyes to the ball
in both positions, so the RAs allowed those and all remaining partici-
pants to do so by describing this estimation. This participant prefer-
ence was likely caused by the overhanging structure under which the
Figure 4. The Experiment 2 participant position during top
(outline) and bottom (solid) estimates. Dotted icons represent the
same participant during each estimate. Dashed lines represent
the vertical path in estimating the horizontal distance for each
position. The barred lines indicate the horizontal distance to esti-
mate for each position. Figure not drawn to scale.
358 JACKSON AND CORMACK
participants stood for the bottom estimate. Positioning participants
directly beneath the vertical surface necessitated a nearly 180º head
rotation, whereas participants in Experiment 1 regularly rotated
their heads only roughly 90º.
We gathered additional data (the distance from the eyes to the
surface of estimation) to reflect participants’ desired estimates. We
compiled this distance for all participants in both positions. Because
the instructions initially suggested estimating from the ground or the
railing, we will first report the results of participants’ estimates as if
they had used the ground when at the bottom and the railing when
at the top, and then report participants’ estimates from their eyes, in
both positions.
Results
Estimates from the railing and from the ground
.
The mean estimate from the bottom position was 8.68 m,
SD 1.31 m, and from the top position it was 5.01 m,
SD 1.30 m. Although these results seem to confirm
ENT predictions, the fact that (most) participants esti-
mated the distance from their eyes, rather than from the
ground or from the railing, would exaggerate the effect
predicted by ENT. Thus, to be conservative and in order to
reflect participants’ intended estimates accurately, we cor-
rected the data as follows. For estimates from the bottom,
we subtracted a standard distance from the eyes to the top
of the head (15.24 cm) from each individual participant’s
height and subtracted the resulting distance from estimates
from the bottom. For estimates from the top, we added the
average distance from the top railing to the participants’
eyes while leaning over the railing (30.48 cm). These dis-
tances reproduced the distance from the eyes to the ball for
each participant and replicated what (most) participants
set as perceptually equal to the fixed horizontal interval.
Estimates from the eyes
. Figure 5 shows a scatterplot
of the data from Experiment 2, plotted in the same way
as the data are plotted in Figure 2. Arrows represent the
mean estimate of horizontal from the top (5.31 m, SD
1.30 m) and from the bottom (7.12 m, SD 1.28 m). As
in Experiment 1, the top and bottom estimates correlated
[r(144) .222, p .01]. Dashed lines in each dimension
represent the actual horizontal distance: 7.20 m. Mean es-
timates less than the actual distance suggest height over-
estimation. A manipulation check against true horizontal
distance indicates statistically significant height overes-
timation from the top [t(143) =17.411, p .001] but
not from the bottom (unlike in Experiment 1) [t(143)
=0.328, p .744].
The solid line in Figure 5 represents the slope of unity
between the top and bottom estimates. The ENT predic-
tion suggested a descent illusion of greater height overes-
timation from the top than from the bottom of the vertical
surface. Figure 5 would support this prediction if the ma-
jority of points fell below the slope of unity. In fact, 129
of the 144 points do fall below the slope of unity. An effect
this large or larger would only occur with a binomial prob-
ability of 4.3 10
=24
if participants actually perceived
height equally from the top and the bottom.
Figure 6 shows mean estimates by position, with error
bars showing 95% confidence intervals about the means.
The dashed line in Figure 6 represents true horizontal
distance. The mean height estimate from the bottom was
7.12 0.21 m, whereas the mean height estimate from
the top was 5.31 0.21 m. The mean estimate was 34%
greater at the bottom than at the top, with a difference of
1.81 m. A 1.81-m position difference between estimates
of a 7.20-m surface suggests a strong effect of position,
with the actual distance constituting 136% of the mean
estimate at the top.
Not surprisingly, a paired samples t test indicated a
significant difference between participants’ estimates of
length from the top and from the bottom [t(143) 13.465,
p .001]. Participants perceived much greater vertical
distance from the top of the height than from the bottom,
consistent with the descent illusion predicted by ENT.
An independent samples t test suggested that order of
testing (i.e., beginning at the top or at the bottom) failed to
influence top and bottom estimate differences [t(142)
=.620, p .536]. Significance tests of Pearson product–
moment correlations failed to suggest that participant body
height correlated with top- and bottom-estimate differences
in men [r(77) .044, p .707] and women [r(66) =.006,
p .959] (1 participant failed to report sex). Neither order
of testing nor body height changed participants’ percep-
tions of greater height at the top than at the bottom.
Discussion
These data appear consistent with the descent illusion
predicted by ENT, in which participants perceive greater
distance from the top of a height than from the bottom.
Additionally, these data suggest that participants overes-
timate height distance in comparison with horizontal dis-
tance when at the top of a height. ENT suggests that costs
of falling present in height navigation could have pro-
duced an adaptive overestimation of specific geographi-
Figure 5. The Experiment 2 participant distance estimates, by
position, in meters. Arrows indicate the mean estimates from the
top and bottom. Dashed lines indicate the actual vertical distance.
The solid line indicates the slope of unity between top and bottom
estimates.
Top Estimate (m)
Bottom Estimate (m)
12
10
8
6
4
2
0
2 4 6 8 10 12 14
EVOLVED NAVIGATION THEORY AND THE DESCENT ILLUSION 359
cal distances associated with falling—a descent illusion,
in this scenario.
Lack of height overestimation from the bottom position
may have resulted artifactually from the testing site. We re-
quired participants to stand directly underneath the vertical
surface, in order to better equate viewing conditions across
positions. However, many participants wanted to move
back, in order to have a more natural view of the vertical
surface; this would have made their view more similar to
that provided in Experiment 1. Mark, Jiang, Steinbach, and
Paasche (1999) found a similar effect of viewing restrictions
hampering distance estimates. Looking directly up, across a
gap, may have reduced ecological validity of this particular
method, because such an estimate would likely rarely occur
in an ecologically valid setting—people would usually have
additional views of a surface from farther away.
Data analyses suggested that restricting participants
to estimate across the gap affected judgments. If we sub-
tract the distance from the top of participants’ heads to
the bottom of the vertical surface at the bottom estimate
(M 6.23 m, SD 1.23), a t test against the true distance
(7.20 m) now suggests a significant height overestimation
[t(143) =9.437, p .001], and a paired samples t test
still suggests a significant difference between top and bot-
tom estimates [t(143) =4.650, p .001]. These data, in
conjunction with Experiment 1, may also reflect the find-
ing that vertical extents are overestimated in proportion to
their perceived size (Chapanis & Mankin, 1967; Dixon &
Proffitt, 2002; Yang et al., 1999).
We allowed participants to use their eyes as the end
reference point when making estimates, instead of the
chest-high railing at the top and the ground at the bottom.
However, the initial instructions likely influenced some
participants. If we were able to subtract the estimates of
individuals who used the instructed reference points to any
degree, the mean top estimate would decrease and the mean
bottom estimate would increase. This would broaden the
disparity between top and bottom estimates and increase
the effect of position observed in Experiment 2 data.
GENERAL DISCUSSION
We demonstrated a descent illusion in which participants
overestimate vertical height much more when viewing it
from the top than from the bottom. J. Stefanucci (personal
communication, August 14, 2005) found similar results in
recent work using different methods. Our participants in
Experiment 1 perceived the vertical surface from the top
as 1.84 times its true height, on average. The participants
in Experiment 2 perceived the vertical surface from the
top as 1.36 times its true height, on average. Expressed
differently, the perceived height from the top averaged
32% greater than the perceived height from the bottom
(29% and 34% in Experiments 1 and 2, respectively). Not
only the average effects are large; we found some overesti-
mation in 100% (66 of 66) of participants in Experiment 1
and in 97% (139 of 144) of participants in Experiment 2.
These results illustrate a key ENT-derived prediction: that
perceived surface height relates to potential navigation
cost. We posit from ENT that the inherently greater risk
of falling associated with descending than with ascending
a vertical surface accounts for the asymmetrical descent
illusion seen in these distance estimates.
We had participants estimate via distance matching.
The key prediction compared estimates across the top and
bottom positions; therefore, invariance across these two
estimates is the most important quality of the procedures.
Because our procedures did not differ across position, it
is highly unlikely that anything related to the procedures
per se could have produced the observed descent illusion.
We know of no procedure that would produce the magni-
tude of effects found in these data by chance, and similar
distance-matching procedures have been used widely in
similar distance estimates (Chapanis & Mankin, 1967;
Dixon & Proffitt, 2002; Higashiyama, 1996; Higashiyama
& Ueyama, 1988; Yang et al., 1999). Further, Da Silva and
Dos Santos (1984) have shown that a variety of distance
estimate techniques produce very similar results.
An obvious question raised by these experiments is why
the size of the effect, expressed as a percentage of the ac-
tual distance estimated, varied across the two experiments
for estimates from both locations. Can we assume that a
perceptual distortion such as the overestimation of heights
would behave like a meridional magnification, so that ob-
jects would appear expanded along the vertical axis? This
would result in an effect similar to wearing cylindrical (as-
tigmatic) lenses. The data are inconsistent with this simple
idea, however.
When we compare Experiment 1 with Experiment 2,
the distance to estimate decreased from 14.39 m to
7.20 m. This corresponded to overestimation from the top
decreasing from a factor of 1.84 to 1.36 (i.e., from an 84%
Figure 6. The Experiment 2 mean participant distance es-
timate, by position, in meters. Bars represent 95% confidence
intervals about the means. The dashed line indicates actual hori-
zontal distance.
Estimated Distance (m)
Participant Position
7.5
6.5
6.0
7.0
5.0
5.5
Bottom Top
360 JACKSON AND CORMACK
to a 36% overestimate), and overestimation from the bot-
tom decreasing from 1.43 to 1.01 (i.e., from a 43% to a 1%
overestimate). Although an analogy to magnification is
seemingly parsimonious, no a priori reason suggests that
it is the form that an adaptation to falling costs would take.
Whereas meridional magnification is trivial to implement
in an optical system, there is unlikely to be a simple neural
analogue to a cylindrical lens in a system as distributed as
the human visual system.
Furthermore, data suggest that vertical extents are as-
sociated with greater overestimation on the basis of per-
ceived size, not meridional magnification. Yang et al.
(1999) found that participants overestimated a vertical
line (of PVC tubing) in natural scenes to a greater percent-
age when viewing the real stimuli outdoors (12% overesti-
mation) than when viewing photographs (2% overestima-
tion) or drawings (3% overestimation) of the same objects.
The authors suggest that this outcome results from greater
perceived object size when in the environment than when
viewing pictures or line drawings. Yang et al. subsequently
ran additional experiments in which they confirmed that
overestimation (expressed as a percentage) increased with
the perceived object size. Chapanis and Mankin (1967)
also demonstrated this effect. These studies suggest that
participants overestimate vertical distances relative to
perceived object size. Here and elsewhere (Jackson &
Cormack, 2006a, 2006b, 2006c, 2006d), we suggest that
perceived size may determine overestimation only insofar
as size approximates likely falling costs.
Differences between our two experiments are consis-
tent with greater vertical extents inducing greater percep-
tual distortion as a proxy for falling cost. Distance esti-
mate differences between top and bottom positions were
roughly the same (about 40%) across both experiments.
Figure 7 illustrates that overestimation from the top, ex-
pressed as a relative change from the estimate from the
bottom, is nearly constant for both experiments. This fig-
ure plots (et eb)/eb for each experiment, in which et and
eb are the estimates from the top and bottom, respectively.
The error bars show 95% confidence intervals about the
means. As the figure shows, the additional overestimation
as a result of viewing from the top rather than the bottom
is substantial (roughly 30%), but not substantially differ-
ent across experiments. This is consistent with the idea
that the amount of overestimation, in general, varies with
size (i.e., falling cost) of the vertical interval estimated,
but the degree of overestimation due to relative viewing
position is roughly constant.
The Vertical–Horizontal Illusion
Ultimately, ENT suggests that the descent illusion or
other response to falling costs could occur outside of situ-
ations with falling costs—as long as doing so imposed
fewer costs than failure to overestimate when in the pres-
ence of ecologically valid falling costs. This illustrates
how ENT may explain the vertical–horizontal illusion
(VHI). The classical VHI (see Figure 8) is the tendency
to overestimate vertical lines in comparison with equal
horizontal lines. The VHI is interesting because we still do
not understand why it occurs, even though it is one of the
oldest psychological phenomena to have been investigated
by modern science (Fick, 1851, cited in Finger & Spelt,
1947; Oppel, 1854, cited in Hicks & Rivers, 1906), having
been researched by the founder of psychological science
(Wundt, cited in Winslow, 1933), among others.
ENT suggests that the classical VHI may be a by-
product of mechanisms that evolved to weight naviga-
tional choices by the falling costs of vertical navigation
(Jackson, 2005). All humans descended from species
that suffered falling costs, such as ancestors in common
with monkeys. Consistent with ENT, new- and old-world
monkeys exhibit the VHI (Dominguez, 1954). ENT also
suggests cross-cultural VHI prevalence, because vertical
surfaces pose falling costs to all humans. Consistent with
ENT, all human populations yet studied display the VHI;
this includes populations in environments with minimal
receding horizontals, such as the Todas of southern India
and the Papuans of New Guinea at the turn of the 20th
century (Rivers, 1905), as well as jungle-dwelling Peru-
vians (Bolton, Michelson, Wilde, & Bolton, 1975). ENT
predicts VHI existence across humans as a species, but
ENT also predicts VHI differences within and between
Figure 7. Percent overestimation from the top, in relation to
the estimate from the bottom, for each experiment. Plots (et
eb)/eb, in which et and eb are the estimates from the top and bot-
tom, respectively. Bars represent 95% confidence intervals about
the means.
% Overestimate
Experiment
50
30
20
40
0
10
12
Figure 8. Three common VHI stimuli: upside-down T, L, and
separated lines.
EVOLVED NAVIGATION THEORY AND THE DESCENT ILLUSION 361
human populations, resulting from variable falling experi-
ence and exposure to fall-inducing environments.
Further, ENT suggests that the VHI may increase as
viewing conditions become less impoverished and more
similar to the conditions in which these adaptations
evolved. As mentioned above, Yang et al. (1999) found
that participants overestimated a vertical line in natural
scenes to a greater percentage when viewing the real
stimuli outdoors (12% overestimation) than when view-
ing photographs (2% overestimation) or drawings (3%
overestimation) of the same objects. Both Chapanis and
Mankin (1967) and the above authors showed that the per-
centage of magnitude of VHI increased with the perceived
object size. Further, Dixon and Proffitt (2002) demon-
strated greater VHI when participants viewed an image
in a virtual 3-D environment and large, flat virtual movie
screen than when they viewed the same image on a virtual
computer screen—even though the stimuli subtended the
same visual angle across conditions.
Of course, the relation between the classical 2-D VHI
and adaptations for 3-D route navigation are thus far spec-
ulative. We are currently investigating this potential link
with predictions derived from ENT (Jackson & Cormack,
2006b).
Descent Illusion Mechanism
These experiments identified a novel perceptual phe-
nomenon, which we term the descent illusion, by using a
causal theory without detailing the underlying mechanism.
Obviously, some cue or combination of cues produces the
descent illusion via some perceptual mechanism, but we
are unaware of any obvious cue that could have produced
such a large discrepancy between estimates from the top
and bottom positions. We are thus far agnostic as to what
specific mechanism produces the descent illusion.
1
The
contributions of causal theories such as ENT primarily
lie in uniting isolated findings and generating novel, test-
able hypotheses. These data demonstrate the utility of a
causal approach, because researchers now know to look
for a mechanism where they previously did not know one
existed.
We are currently attempting to determine the cue or
combination of cues necessary to produce the descent il-
lusion across both real-world and virtual reality settings
(Jackson & Cormack, 2006a). Given the availability of
multiple cues about one’s position in space, we find it
likely that selection operated on several cues simultane-
ously. That said, obvious candidates we are exploring in-
clude total visual extent of environment, vestibular cues to
head orientation, and high-level situational awareness.
Conclusions
People overestimate the height of a vertical surface, and
we have shown that they do so to a dramatically greater
extent while positioned at its top. This descent illusion
increased as the actual height, and thus the likely falling
cost, increased across experiments. Nevertheless, overes-
timation from the top approximated a relatively constant
percentage of the height estimated from the bottom. The
height perceived from the top averaged 32% greater than
the height perceived from the bottom, and was as much as
84% greater than the true height. We predicted the descent
illusion by applying a broad, causal theory—ENT—to
height perception. ENT focuses on adaptations that arose
in response to navigation costs over evolutionary time.
ENT applied to height perception directed our investiga-
tions to the navigation cost asymmetry of falling, which is
much greater for descent than for ascent. This investiga-
tion speaks against a view of unitary distance perception
and suggests some specialization for both height percep-
tion and within different height-perception scenarios.
AUTHOR NOTE
We thank Emily Bromberg, Joanna Casas, Ashley DePierri, David
Dubin, Nancy Egbert, and Ryan Monteiro for collecting data, and Clarke
Burnham and two anonymous reviewers for comments. Correspondence
concerning this article should be addressed to R. E. Jackson, Depart-
ment of Psychology, California State University, 333 S. Twin Oaks Val-
ley Rd., San Marcos, CA 92096-0001 (e-mail: russelljackson@mail
.utexas.edu).
REFERENCES
Avery, G. C., & Day, R. H. (1969). Basis of the horizontal–vertical illu-
sion. Journal of Experimental Psychology, 81, 376-380.
Avraamides, M. N., Loomis, J. M., Klatzky, R. L., & Golledge,
R. G. (2004). Functional equivalence of spatial representations de-
rived from vision and language: Evidence from allocentric judgments.
Journal of Experimental Psychology: Learning, Memory, & Cogni-
tion, 30, 801-814.
Bassett, D. R., Vachon, J. A., Kirkland, A. O., Howley, E. T., Dun-
can, G. E., & Johnson, K. R. (1997). Energy cost of stair climbing
and descending on the college alumnus questionnaire. Medicine &
Science in Sports & Exercise, 29, 1250-1254.
Bolton, R., Michelson, C., Wilde, J., & Bolton, C. (1975). The
heights of illusion: On the relationship between altitude and percep-
tion. Ethos, 3, 403-424.
Chapanis, A., & Mankin, D. A. (1967). The vertical–horizontal illu-
sion in a visually-rich environment. Perception & Psychophysics, 2,
249-255.
Cohen, H. H., & Lin, L. (1991). A scenario analysis of ladder fall ac-
cidents. Journal of Safety Research, 22, 31-39.
Conroy, G. C. (1997). Reconstructing human origins. New York:
Norton.
Da Silva, J. A., & Dos Santos, R. A. (1984). The effects of instructions
on scales for perceived egocentric distance in a large open field. Bul-
letin of the Psychonomic Society, 22, 189-192.
Dixon, M. W., & Proffitt, D. R. (2002). Overestimation of heights in
virtual reality is influenced more by perceived distal size than by the
2-D versus 3-D dimensionality of the display. Perception, 31,
103-112.
Dominguez, K. E. (1954). A study of visual illusions in the monkey.
Journal of Genetic Psychology, 85, 105-127.
Finger, F. W., & Spelt, D. K. (1947). The illustration of the horizontal–
vertical illusion. Journal of Experimental Psychology, 37, 243-250.
Foley, J. M., Ribeiro-Filho, N. P., & Da Silva, J. A. (2004). Visual
perception of extent and the geometry of visual space. Vision Re-
search, 44, 147-156.
Gibson, J. J. (1979). The ecological approach to visual perception. Bos-
ton: Houghton Mifflin.
Haslam, R. A., & Bentley, T. A. (1999). Follow-up investigations of
slip, trip and fall accidents among postal delivery workers. Safety Sci-
ence, 32, 33-47.
Hicks, G. D., & Rivers, W. H. R. (1906). The illusion of compared hori-
zontal and vertical lines. British Journal of Psychology, 2, 243-260.
Higashiyama, A. (1996). Horizontal and vertical distance perception:
The discorded-orientation theory. Perception & Psychophysics, 58,
259-270.
362 JACKSON AND CORMACK
Higashiyama, A., & Ueyama, E. (1988). The perception of vertical and
horizontal distances in outdoor settings. Perception & Psychophysics,
44, 151-156.
Howard, I. P., & Templeton, W. B. (1966). Human spatial orientation.
New York: Wiley.
Jackson, R. E. (2004). Evolutionary psychology. In W. E. Craighead &
C. B. Nemeroff (Eds.), The concise Corsini encyclopedia of psychol-
ogy and behavioral science (3rd ed., pp. 341-343). New York: Wiley.
Jackson, R. E. (2005). Falling towards a theory of the vertical–horizontal
illusion. In H. Heft & K. L. Marsh (Eds.), Studies in perception and
action VIII (pp. 241-244). Mahwah, NJ: Erlbaum.
Jackson, R. E., & Cormack, L. K. (2006a). Descent illusion mecha-
nism. Manuscript in preparation.
Jackson, R. E., & Cormack, L. K. (2006b). Evolved navigation theory
and acrophobia. Manuscript in preparation.
Jackson, R. E., & Cormack, L. K. (2006c). Evolved navigation theory
and the environmental vertical illusion. Manuscript submitted for
publication.
Jackson, R. E., & Cormack, L. K. (2006d). Evolved navigation the-
ory and the origin of the vertical–horizontal illusion. Manuscript in
preparation.
Keeley, L. H. (1996). War before civilization. New York: Oxford Uni-
versity Press.
Mark, L. S., Jiang, Y., Steinbach, K., & Paasche, J. (1999). The im-
pact of visual exploration on judgments of whether a gap is crossable.
Journal of Experimental Psychology: Human Perception & Perfor-
mance, 25, 287-295.
Minetti, A. E., Moia, C., Roi, G. S., Susta, D., & Ferretti, G. (2002).
Energy cost of walking and running at extreme uphill and downhill
slopes. Journal of Applied Physiology, 93, 1039-1046.
Rivers, W. H. R. (1905). Observations on the senses of the Todas. Brit-
ish Journal of Psychology, 1, 321-396.
Segall, M. H., Campbell, D. T., & Herskovits, M. J. (1966). The in-
fluence of culture on visual perception. Indianapolis: Bobbs-Merrill.
Somervill, J. W., & Somervill, C. Z. (1977). Descent behavior of
rats and chicks in a cliff and non-cliff situation. Perceptual & Motor
Skills, 45, 391-397.
Svanstrom, L. (1974). Falls on stairs: An epidemiological accident
study. Scandinavian Journal of Social Medicine, 2, 113-120.
Teh, K. C., & Aziz, A. R. (2002). Heart rate, oxygen uptake, and energy
cost of ascending and descending the stairs. Medicine & Science in
Sports & Exercise, 34, 695-699.
Tinetti, M. E., Speechley, M., & Ginter, S. F. (1988). Risk factors
for falls among elderly persons living in the community. New England
Journal of Medicine, 319, 1701-1707.
Winslow, C. N. (1933). Visual illusions in the chick. Archives of Psy-
chology, 153, 83.
Wu, B., Ooi, T. L., & He, Z. J. (2004). Perceiving distance accurately
by a directional process of integrating ground information. Nature,
428, 73-77.
Yang, T. L., Dixon, M. W., & Proffitt, D. R. (1999). Seeing big things:
Overestimation of heights is greater for real objects than for objects in
pictures. Perception, 28, 445-467.
NOTE
1. Our approach, wherein the perception of a surface is perforce influ-
enced by the cost of navigating that surface, might bring to mind Gibson’s
(1979) notion of affordance. We do not use the term because it carries
theoretical implications that we do not necessarily want to convey.
(Manuscript received September 8, 2005;
revision accepted for publication March 15, 2006.)
