![[Previous Chapter]](/icons/button-back.gif){kind=icon}
![[Table of Contents]](/icons/button-contents.gif){kind=icon}
![[Next Chapter]](/icons/button-next.gif){kind=icon}
The measure of the representativeness of virtual environments was measured using three perception tasks. It was also measured in the form of self-reported evaluations from the participants. The results are analyzed and discussed on a task by task basis first, and then they are discussed as a whole.
Size Estimates - Analyzing the Data.
The room size estimating task required participants to estimate the three dimensions, length, width and height, of three different spaces in the museum. Because of the size differences in the three rooms, the results were normalized as percentages of actual room dimensions (1 is a perfect estimate of distance, and .9 is 90% short of the actual length). In this way, it was possible to statistically combine the results of the differently shaped rooms. It was also a convenient way to convey under and overestimates relative to the actual dimensions.
The terms "length" and "width" were used for convenience during the experiment. Participants each had their own definitions for what constituted the "length" or the "width" of the spaces. Some judged that the shorter dimension was the width. Others defined width to be the dimension of a space perpendicular to the entrance. The reversibility in the definition of these dimensions suggests that they can be treated as one relatively shorter or longer dimension. I therefore combined all the horizontal estimates. This doubling of data strengthened the statistical value of the results.
There are two important factors which suggest that the height dimension, on the other hand, should not be combined with the other spatial dimensions of width and length. The first is procedural. To estimate a room's horizontal dimensions, participants generally imagined the number of body lengths required to span the space. This was a difficult task. Subjects had little information as to the size of their bodies laying on the floor because they were standing (or because their bodies were not in the model). The estimate for the vertical dimension was simpler because their standing posture "clued them in" to a large portion of the distance.
The second difference is that, in many instances, participants said that the nature of building construction limited the range of possible vertical dimension. Whereas the widths and lengths of rooms vary, room heights often have standard dimensions because building materials often come in standard dimensions. This limited the variability in the estimates. Of the 144 horizontal estimates (2 dimensions x 3 rooms x 24 participants), only one estimate was exact, representing less than 1% of the total number of guesses. Of the 72 height estimates (1 dimension x 3 rooms x 24 participants), 17 were perfect, about 24%. As a result, the size task results had to be divided into "horizontal" and "vertical" dimension estimates.
Size Estimates - General Observation.
Fig. 7.1 - Actual area of rooms as a function of estimated area.
The combined means across all conditions suggest that participants usually
underestimated the dimensions of the rooms. These underestimates were related
to the size of the rooms. The underestimating increased as the size of spaces
increased. The six actual horizontal dimensions of the three rooms in the
museum are 17.5, 18.5, 23, 28, 47 and 48 feet, represented as numbering from 1
to 6 respectively (Fig. 7.1).
The height estimates behave quite differently. Participants' results were all
close to the actual height and there is no apparent relationship to the actual
size. Two of the room heights were 14 feet and one was 20 feet, represented as
numbering from 1 to 3 respectively (Fig. 7.2).
Fig. 7.2 - Actual height of rooms as a function of estimated height.
These general observations confirm the importance of distinguishing between
vertical and horizontal distance estimates. The height overestimates also
indicate that a particular architectural design can have the effect of making
spaces feel taller, if desired.
A multi-comparison one factor ANOVA for horizontal distance estimates
shows a significant main effect due to display conditions with p<.0001 at
the 95% confidence interval. The underestimates for dimensions in all of the
simulation conditions are significantly different from the Real condition.
Results for the Real display condition were quite accurate (Fig. 7.3).
Fig. 7.3 - Average horizontal distance estimates of spaces in the four test
conditions.
Among the simulation conditions, the underestimated values in the Tracked
condition were significantly smaller than the other two display conditions,
according to the Fisher PLSD test. Furthermore, the more conservative Scheffe
F-test suggests that the Tracked condition was also significantly different
from the Real condition (Table C.1 in Appendix C: Statistical Tables).
Fig. 7.4 - Average height estimates of spaces in the four test conditions.
A multi-comparison one-factor ANOVA for height estimates suggests there is a
main effect between display types, with p<.0003 at 95% confidence interval.
The means for the simulation conditions are all smaller than the Mean for the
Real condition, as was the case for the Horizontal dimension (Fig. 7.4).
Both the less conservative Fisher PLSD test and the more conservative Scheffe
F-test indicate that the Tracked condition results are significantly smaller
than the Monitor and the Real condition. Both tests for individual differences
also show that, in the two Eyephone conditions, Fixed and Tracked, the
estimates were significantly smaller than in the Real condition (Table C.2 in
Appendix C: Statistical Tables).
In general, there was little variation in the self-reports for ease and
confidence in doing the size estimation task (Fig. 7.5). Furthermore, people's
standards for what constituted "easy" and "difficult" varied. For these
reasons, the significance of this data is diminished.
Fig. 7.5 - "Level of confidence" in the size task.
A one factor ANOVA for Ease has a main effect across display conditions with
p<0.003 at the 95% confidence interval. According to the Fisher PLSD test,
participants in the monitor and the fixed conditions felt their tasks were
easier than participants in the Real condition (Table C.3 in Appendix C:
Statistical Tables).
Fig. 7.6 - "Ease of estimating sizes" (Means).
Similar results were expressed for the confidence participants had in their
estimates. There was a main effect for confidence with p<0.02 at the 95%
confidence interval. Participants in the Fixed condition expressed more
confidence in their values than those in the Monitor and Real conditions,
according to the Fischer PLSD test (Fig. 7.6 and Table C.4 in Appendix C:
Statistic Tables).
All of the horizontal dimension estimates in the simulation conditions
were significantly smaller than those in the Real condition. There are several
possible explanations.
The model. Since the dimensions were underestimated in all of the
simulation conditions, the results could suggest there was something inherent
to the scale of the model which made the spaces appear smaller. However, this
is not likely because the scale of model in the present system was calculated
very accurately.
Movement. Another explanation is related to the difference in the way
people moved through the spaces. As discussed earlier, the act of physically
walking is a very important cue for getting a sense of rate of movement and
distance. Participants in the real condition physically paced across the
spaces. Participants in the simulation conditions had no kinesthetic feedback
for their movement. Their rate of movement could only be perceived visually.
There is little reason to believe however that simulated movement should
necessarily make distances appear to be shorter. Furthermore, the participants
were stationary while doing the distance estimates.
Field of View. With regard to the field of view, these findings concord
with other studies which have shown that the perception of size and distances
diminish as the field of view narrows (Dolezal 1982, Alfano 1990). This is
better know as the size-constancy problem. The moon illusion is a well know
example. The moon appears to be larger when it is close to the horizon than
when it is directly above. Yet the moon does not really diminish in size. The
field of view for both eyes was constrained to 90deg. in all the simulation
conditions. More specifically, the peripheral vision was deprived of about
45deg. per eye. Although the 90deg. field of view is more generous than those
in the Dolezal or the Alfano studies (explored the 60deg. to 12deg. field of
view), it certainly seems to follow that trend.
Previous experience. Participants in the Real condition knew the space
significantly better than the other participants (Please refer to Appendix B:
Data Figures - Participant Profile). This might explain some of the differences
between the simulation conditions and the Real condition.
Other factors. There are of course numerous other differences between
the Real museum and the simulated one. People walking through the real museum
could hear their footsteps. They could feel and smell the qualities of the
air, and they probably perceived many other aspects to the space which did not
exist in the simulation condition, and yet which played a role influenced their
perceptions (see Canter 1977).
The Tracked condition estimates were also significantly different from both the
Fixed and the Monitor simulation conditions. The head-tracking device
differentiated both the Fixed and the Monitor condition from the Tracked
condition. In the Tracked condition, participants turned their heads to change
directions, whereas in the other two conditions, participants rotated the
Spaceball.
Search pattern. The results suggest that source of difference between
the non-tracked and the tracked conditions lies in the difference in the
searching pattern. In the non-tracked conditions, participants wishing to see
another portion of the space must rotate the model with the Spaceball. In this
scenario, the participants trying to estimate the dimensions of a room are
basically panning the images in front of their field of view until the edge of
the room appears. This is a rather "passive" way to search the spaces.
The process of searching the spaces is more "active" in the Tracked condition.
Users turn their heads (and bodies) in search of the room edges, much as they
would in a real space. They are potentially better able to anticipate where the
corners should be precisely because the position of their body is mapped
one-to-one to the model. In that process, the gaze is engaged. While turning
their eyes in the direction of the anticipated corner of the room, they end up
seeing through the edges of the eyephones, precisely where the distortion of
the optics is greatest.
Tracked condition setup. Another explanation might be related to the
way participants interacted with the model. The space ball was fixed to the
table in all three conditions. In the case of the Fixed or Monitor conditions,
this was a convenience. However, it was much less appropriate to use in the
Tracked condition. It was difficult for participants to use, especially when
turning full circles. This might have affected the process by which they
explored the spaces. In any event, it made the task awkward.
Gender. A note might be said about the gender distribution across the
display conditions. At the same time, results suggest women underestimated
room sizes more than men. There were many more women in the Tracked condition
than in the other three conditions. Of the 24 subjects, 5 were women, and
three of those experienced the Tracked condition. While their subject group
was very small, it might explain at least in part why the scores were so much
lower than the Fixed condition (Table C.5 in Appendix C: Statistics Tables).
As indicated earlier, the height estimates were greatly affected by the
known standardization of building materials. While this improved people's
estimates, it did so evenly across display conditions. All of the estimates in
the simulation conditions were smaller than the Real condition estimates, as
was the case for horizontal estimates. The explanations for the differences
between the vertical dimension estimates in the Real and all three simulation
conditions is probably the same as it was for the horizontal dimension.
When looking at individual differences between the display conditions, it was
found that both stereoscopic conditions were significantly different from the
Real condition. This further supports the conclusions about the underestimated
horizontal results, which suggests that the limited field of view has the
effect of diminishing people's perception of distances.
There is also a study which indicates that estimates of size and distance are
more accurate in conditions where the peripheral vision has at least
some visual stimulus, even if the stimulus has no direct relationship with
the information in the fovea (Hagen 1978). Hagen conducted an estimating task
comparing, among other things, the perception of size and distance in a real
setting as viewed through a truncated window, with the perception of size and
distance when seeing a slide view of the same scene. The field of view is
identical in both instances. While not statistically significant, subjects did
tend to underestimate distances and sizes more in the "truncated" view of the
real scene than in the view of the slide. This suggests that the information
in the periphery, in this case, the room where the projection was taking place,
the seats and the walls, while entirely unrelated to the projected image,
nonetheless served as a kind of context by which to better judge the sizes of
objects in the fovea field of view.
Their study is similar to the Museum study. The Monitor condition is
equivalent to the "slide" condition in the Hagen study, and the Eyephone
conditions are equivalent to the "truncated view" of the real scene. Both the
monocular and binocular conditions display the model 90deg. around the fovea.
In the Eyephone conditions, the participant has no visual information in the
periphery. They only see the black rubber of the device. In the Monitor
condition, the participant can see the rest of the laboratory in their field of
view. It serves as a context, or a frame of reference, much as the projection
room did in the other study. The means between the Monitor and the Eyephone
conditions do show that, like the Hagen study, the Monitor condition values
were less underestimated.
Fig. 7.7 - Schematic representation of perceived size in a virtual and real
space. Perceived volume in Tracked condition (gray) and the Actual volume
(black)
The significant underestimating which is perceived in virtual spaces is an
important finding. If the values for the means of the three dimensions are
multiplied, they would describe a volume of the main gallery of the museum to
be 60% that of the actual size (Fig. 7.7).
The results for Ease and Confidence are not very reliable because they were
self-reported values. They nonetheless indicate that it was easier to estimate
the size of the volumes when viewing the simulations rather than when being in
the real space. This is interesting because it would support the purpose of
this representation. By being free of distractions and unnecessary details,
representations make it easier for people to judge and evaluate specific
spatial attributes, in this case, the dimensions of the spaces.
In the angle task, participants were asked to point to or face in the
direction of three objects they had seen previously during their tour. Walls
obstructed the target objects from view. As a result, the participant's task
was to remember where the object was, and then to estimate its location.
In the Real condition, participants pointed in the selected direction with
their hand. In all of the simulation conditions, rather than pointing, they
were asked to center their view in the direction of the target.
The estimated angles were subtracted from the actual angle of the target. The
data is therefore converted into "angle in error from actual target direction".
Since the size of the target was quite large, and because the technique for
expressing a desired direction was not extremely precise, it is assumed that
differences under 5deg. are insignificant.
A first look at the data suggests that the values from the three target
tasks need to be normalized before they can be averaged because they have a
bias. Participants consistently overestimated the angle to the target in the
same direction (Fig. 7.8).
Fig. 7.8 - Bias in the angle task.
In the first two target conditions, the target is to the right of the closest
normal (in geometric terms, the normal is the line which describes a
perpendicular angle between the observer and a facing wall). In those
conditions, participants across all four conditions estimated the target to be
even farther to the right. In the third target condition, the position of the
target is the mirror invert of the first two target tasks; it is to the left of
the nearest normal to the facing wall. Subsequently, estimates were more to
the left than the target. Biases are common to the pointing task. Generally,
people are biased by the direction from which they have been traveling (Okabe
1986, Lindberg 1980).
A scattergram for the data points by target shows the two trends for the
distribution of error (Fig. 7.9). Most of the estimates for the first and
second angle are to the right (positive), and most of the estimates for the
third angle are to the left (negative angles).
Fig. 7.9 - Distribution of angle data before it is normalized.
The data can be normalized by multiplying the values from the third angle by
(-1). The scattergram illustrates this (Fig. 7.10). Positive values become
overestimates in the direction of the target, relative to the closest normal,
and negative values become underestimates.
Fig. 7.10 - Distribution of angle data after it is normalized.
After the data is normalized, a one factor ANOVA suggests there is a
significant difference between the three target estimates across all four
conditions, p<0.03 at the 95% confidence interval. The Fisher PLSD test for
individual differences shows that estimates for the third target were much
better than the other two (Table C.6 in Appendix D: Statistics Tables). This
might indicate that participants' cognitive map of the museum improved as their
visit progressed. In any case, a 2 factor ANOVA for display and target versus
the normalized angle shows no main effect for the interaction, p<0.65 at the
95% confidence interval (Table C.7 in Appendix D: Statistics Tables).
Fig. 7.11 - Distribution of normalized angle data.
A multi-comparison one factor ANOVA for the normalized angles shows no main
effect for display condition, p < 0.17 at the 95% confidence interval (Table
C.8 in Appendix C: Statistics Tables). However, when taking a look at the
distribution of the data, certain trends become apparent. The diagram below
shows the distribution of error by display type. While it is not clear that
the means are different, the variance in the distribution of the
data is on the other hand quite different. This trend was not signaled in the
ANOVA statistical test because this test is sensitive to differences in
means, rather than differences in variance. While the mean is an
important measure, the success of the orientation task is just as much a
function of the variance of the estimates. In the Real condition, the
tightness of the distribution reflects a certain consistency across all the
subjects, even though the angle scores were on average 9.2deg. off their target
(Fig. 7.11).
In the simulation conditions, the distribution has a much greater variance.
This could reflect a greater uncertainty in participants' task of estimating
the direction to the targets. A post hoc ANOVA on the variance of the
angles from the mean of each display condition suggest that there is a main
effect at the 90% confidence interval, with p<0.07. The diagram is another
representation of the means and standard deviations, by display condition (Fig.
7.12 and Table C.9 in Appendix C: Statistics Tables).
Fig. 7.12 - Means and standard deviations of angle data.
The Orientation Task - Self-reported Ease and Confidence.
Participants did not find the orientation task to be significantly
easier in any one display condition. A one factor ANOVA shows no main effect,
with p < 0.11 at the 95% confidence interval (Table C.10 in Appendix C:
Statistics Tables).
There was no relationship between confidence participants had in their angle
estimates and their display condition, p<0.49 at the 95% confidence interval
(Table C.11 in Appendix C: Statistics Tables).
There was a general trend across all display conditions to overestimate
the angle to the target by an average of 8deg.. This is common in orientation
tasks and in cognitive maps in general. In similar studies, it has been shown
that people's angle estimates can be biased by the path of their tour. Okabe
has found that, "When a trail was winding and there were no landmarks along the
trail, direction judgment was biased toward the direction of the vanishing
point of a traversed trail."(Okabe 1986). Kevin Lynch was a pioneer in the use
of sketches to study cognitive maps of places. He has many examples of how
people systematically straighten curved lines (Lynch 1960, Canter 1977). But
while this seems to be clearly a case of bias, the reason for the bias is
beyond the scope of this study.
While there was no significant difference in angle estimates between the four
display conditions, subsequent analysis confirmed to some extent (at 90%
significance) that the variance in the estimates in the simulation conditions
were significantly greater than in the Real condition. These results begin to
reveal how difficult it is to situate oneself in the computer generated
environments.
There are several reasons to believe that the little differences in variance
mask big differences in people's sense of orientation. The first is due to the
simplicity of the test space. The museum has a symmetrical plan. Two pairs of
rooms are identical in size which are laid out symmetrically about the plan.
The three non paired spaces are distinctly different spaces. The building
footprint is the rectangle. All of these attributes of "rectalinearity"
greatly simplify the orientation task.
Another factor which greatly simplified the orientation task is that the main
gallery and the entrance hallway were visited twice. Studies have shown that
"repeatedly viewing a spatial event increases the accuracy of the observer's
spatial representation" (Allen 1978, Passini 1984). Both the simplicity of the
test site and the repeated viewing of the spaces raise the expectancy in the
accuracy of the estimates. Participants should be about as good as they will
ever be. This means that little errors in angle estimates are really quite
significant.
Normally, when evaluating the results of the orientation tasks, it is assumed
that the chance expectancy is 90deg. because errors can range from 0deg. to
180deg.(Presson 1987). But because of the simplicity of this task, the chance
expectancy is quite a bit lower. The combination of the rectangularity and
symmetry of the space limited the possible range of errors from 0deg. to
45deg., and therefore reduced the chance expectancy to 22.5deg.. In the context
of this range, the simulation results were approaching randomness.
There are several possible explanations for the apparent randomness, or at
least the lack of consistency in the estimates of the simulation conditions,
when compared to the results of the Real condition:
Field of view. One of the differences between the Real condition and
the three simulation conditions is size of the field of view. As discussed
earlier (Chapter 4. Hypothesis), it has been shown that "the overlap of
peripheral and fovea information is necessary for veridical perception to
occur, ... and that restricting the field of view will interfere with both
perception and visuomotor performance " (Alfano 1990). Although these studies
concern a field of view truncated to 60deg., rather than the 90deg. in the
eyephones, it remains nonetheless a plausible explanation for the imprecision
in the estimates.
One of the other effects of having a limited field of view is that participants
could not back up enough in the space to see as much of it as participants
could in the Real condition. So while participants in the simulation
conditions might have known quite well where the target was, their view of the
space was too limited to express the direction accurately. It is very similar
to the difficulties encountered when one tries to hang up a poster by oneself.
Although one knows what looks straight and what does not, it is nonetheless
difficult to place the poster because one does not see enough of the rest of
the room in one's field of view.
As a result, it is unclear whether the limited field of view hampered people's
ability to build a cognitive map, or whether it just made the task of
identifying their selected direction difficult.
Movement. The other important difference between the simulation
conditions and the Real condition is the way participants move through the
spaces. Participants in the Real condition physically walked around the space,
whereas in the simulation conditions, they experienced the illusion of movement
without the kinesthetic feedback. This was probably a considerable
disadvantage for making accurate estimates of direction because the distance
traveled could only be interpreted from the visual modality.
There is also an important factor which might have tainted the results for the
Tracked condition. The Spaceball was fixed to the table in all simulation
conditions (Appendix D: Interface Hardware). While this was appropriate for
the non-tracked conditions, it served as an important indication of
(dis)orientation in the Tracked condition. Since the orientation of the
virtual space is mapped onto that of the laboratory space, participants had a
constant kinesthetic cue as to the direction they were facing. The Spaceball
would have been stripped of its role as an "anchor" to the participant's sense
of orientation if they had been obliged to carry it with them. This would most
probably have lead to different results.
The qualitative questionnaire required participants to rate the
qualities of the main gallery space, using an adjective checklist. There were
13 bi-polar adjectives to choose from (see Appendix A: Questionnaire). The
purpose of this task was to look for resemblance in the overall spatial
description of the gallery, rather than to study specific differences in
attributes of the space.
Fig. 7.13 - Descriptive task averages by question number.
There appears to be a remarkable similarity in the results across display
conditions (Fig. 7.13).
Because the adjectives are non-parametric, each question is measured
separately across the four display conditions. 13 one-factor ANOVAs were run,
one for each question. There were only significant differences between display
conditions in question 2 and 3. Question 2 had a p<.02 and question 3 had a
p<.007 at 95% confidence (Table C.12 and Table 13 in Appendix C: Statistic
Tables).
In order to properly compare these non-parametric results as a whole, a simple
regression was computed (Table 7.1). The graph below summarizes the
findings.
Monitor Fixed Tracked
Fixed .74
Tracked .83 .84
Real .88 .79 .91
Regression values (r2)
There exist no absolute values which describe the qualities of interior spaces.
Instead, the "objective" attributes are based on the average of many people's
perception of the space. If this is the case then, the goodness of the
simulations can be measured as a function of the distance of these scores from
the Mean of the Real condition scores for each question. A post hoc
analysis suggests, although not at a statistically significant level, the
Tracked condition results were on average much closer to the Real condition
than the other two simulation conditions (Fig. 7.14).
Fig. 7.14 - Average difference in variation from the mean of the Real condition
(based on the 1 to 7 semantic differential scale).
The results from the adjective checklist are all relatively similar.
However, the comparison between the difference of the values in the simulation
conditions and those from the control group (Real condition), indicate a
greater concordance between the Real condition and the Tracked condition than
with the other two simulation conditions. This would seem to confirm that
people are better able to judge the feel of models in virtual environments than
in walkthroughs. In a sense, these conclusions are to be expected because it
is much easier to judge the feel of a space when one is present in it
than when one looks at it. Of all the simulation conditions, the
Tracked condition the most effective at creating the illusion of presence.
The concept of having a "sense of presence" in a computer model comes "hand in
hand" with the invention of virtual interfaces. Magazines about virtual
interfaces are even entitled "Presence". In fact, the sense of presence is so
much greater in virtual environments than when viewing walkthroughs, that it is
surprising to not find a greater difference.
Perhaps the lack of greater difference between the Tracked condition and the
other two conditions is due to the simplicity of the test space. As a museum
space, the main gallery was purposefully neutral and free of unnecessary
details. There was a limited number of attributes to describe it.
Furthermore, the list of attributes in the questionnaire was much shorter than
its author intended (see Chapter 3). The combination of these two factors
might explain the relative similitude in participants' perception of the space
across the various display conditions.
In addition to the results above, participants were asked to perform
other less easily quantifiable tasks. The purpose of these tasks was to
explore and probe for other unknown possible differences between the display
conditions. These include a sketch task, questions about participants' most
and least favored spaces, as well as open ended questions about the interface
they experienced.
After their visit, participants were asked to draw the plan of the
gallery. Their plans were rated as a function of their perception of (1) the
location of rooms relative to each other, (2) the path of visit and (3) their
ability to rank the spaces by size from smallest to largest. This less formal
task was designed to capture information which might not have been apparent
through the size and angle tasks.
The maps are loosely rated as a function of the number of errors made. The
scale is the following; 0 errors = 100%, 1 error = 50% and 2 or more errors =
0%. For example, if a participant forgot to draw one of the rooms, their
sketch would have a score of 50%.
In general, the maps were all quite accurate. This is to be expected,
given the simplicity of the test site.
1. Relative location of rooms: Participants in the Real and Tracked conditions
had a perfect sense for the relative location of spaces. Some participants in
the Monitor and Fixed conditions had a much less clear cognitive map of the
gallery spaces (Fig. 7.15).
Fig. 7.15 - Accuracy in recalling the relative location of spaces.
2. Path of Visit: Participants in the Real condition had a perfect
recollection of their movements through the gallery. The simulation
participants had more difficulties doing this task (Fig. 7.16).
Fig. 7.16 - Accuracy in recalling path of visit.
3. Relative Sizes Volumes: Participants had problems distinguishing the size
of the two small paired rooms, A & C and E & F (for a plan view, see
Fig. 6.3). Some participants did not even recognize that the small rooms were
identical in size. The task was as difficult for three of the four conditions.
Participants in the Fixed condition fared much worse (Fig. 7.17).
Fig. 7.17 - Accuracy in ordering spaces by volume size.
Overall, results from the sketch task suggest that participants in all four
conditions had a rather accurate cognitive map of the museum. However, the
average accuracy of the three sketch tasks indicate that the participants in
the Tracked condition might have been better able to build an accurate
cognitive map of the spaces than those in the other two simulation conditions
(Fig. 7.15, 7.16 and 7.17). If that were the case, it would further emphasize
the fact that the pointing task was not effective in measuring the accuracy of
people's cognitive maps.
The second portion of this task consisted in comparing general
subjective views about the gallery as a whole. Participants were asked to rate
which space was (1) easiest to size-up, (2) the most pleasant and (3) the least
pleasant. They were asked to explain their answers. What quickly becomes
apparent when looking at the data is the tremendous variability due to the
differences in the participants. What makes a space pleasant for one person
only partly applies to the next. As a result, the effect of the environment is
difficult to distinguish.
1. Easiest space to size-up: Most participants agreed that either the first
small rooms "A" and "C" were the easiest to size up or the main gallery space
"D" (Fig. 7.18). The small ones were easy because participants could see more
of the walls without distortion, and their human scale helped them judge
distances. Those who felt the larger room was easier usually said it was
because of the scale figures.
Some suggested that spaces which were visited more than once became familiar
and therefore easier to size up. Both "D" and "C" fit this category (for a
plan view, see Fig. 6.3).
Fig. 7.18 - Easiest rooms to "size-up".
In the Monitor condition, the task of sizing up spaces was judged easiest in
rooms which required no panning, which included scale figures, or which were
small.
Participants in the Fixed condition felt that, as in the Monitor condition,
distortions due to panning of spaces made the task difficult. They also
preferred spaces with a scale figure. What is new about this group is the
effect of adaptation. They, more than any in any other condition, expressed a
need to adapt to the simulation model. They found it easier to size up rooms
which they had visited multiple times, or after they had felt more comfortable
with the experience.
In the Tracked condition, participants found small spaces to be easiest to
size-up.
Like the other conditions, Real condition participants felt that the size of
the room affected the ease of the sizing task. But unlike the simulation
conditions, participants did not feel as great a need for scale figures.
Instead many expressed they could use themselves as scale figures.
2. Most pleasant space: 70% to 80% of the participants across all conditions
preferred the large main gallery space. People who liked cozy spaces preferred
room "A". Those who liked more open spaces preferred the main gallery space
"D" (for a plan view, see Fig. 6.3).
3 The least pleasant space: There was no general agreement as to which space
was most unpleasant (diagram F). Real condition participants did not like the
end space "G" much. The participants in the simulation condition did not get
exactly the same introduction to the hallway "B" and certainly did not get to
appreciate the detail it actually contains, and it was therefore often
mentioned as an unpleasant space. But most participants agreed that spaces
where unpleasant either because they were movement spaces rather than pause
spaces, or because of the proportions of the room.
(For a more complete decomposition of the results for the sketch task, please
refer to Appendix B: Data Figures - Sketches)
In an attempt to further explore participants' cognitive maps, they were
asked to judge whether or not they got lost during the tour, the number of
times this happened, and how that affected their sense of well being. None of
these questions delivered statistical findings (refer to Appendix B: Data
Figures - General Experience Evaluation).
These are self-reported open-ended questions about the simulation
interfaces. The purpose of these questions was to (1) look for the main
factors which participants felt influenced their ability to do the tasks and
(2) to identify and prioritize the aspects of virtual interface technologies
which require most urgent improvements.
Participants listed 3 things about the experiment that made their task
difficult, and 3 things which would make their task easier. The answers were
clustered after the study was conducted and divided into 4 main categories:
Procedural, Model, Interaction and Input Device.
Architects overall felt that both the input device and the rendering of the
model were the most important factors in helping or hindering their tasks
during the experiment (Fig. 7.20). They felt the tasks would have been easier
if there had been more cues about the scale of the space, such as scale figures
and texture mapping. Many found the Spaceball difficult to use. Many
participants also complained about the low resolution of the eyephones and
their limited field of view (a complete breakdown of these evaluations is
located in Appendix B: Data Figures - Participant evaluation of interface).
Size Estimates - The Effect of Display in the Horizontal Dimension.
Size Estimates - The Effect of Display in the Vertical Dimension.
Size Estimates - Self-reported Ease and Confidence - General
Observations.
Size Estimate - Horizontal Dimension - Discussion.
Size Estimate - Vertical Dimension - Discussion.
The Orientation Task - Analyzing the Data.
The Orientation Task - General Observation.
The Orientation Task - The Effect of Display
The Orientation Task - Discussion.
Descriptive Questionnaire - Analyzing the Data.
Descriptive Questionnaire - General Observation.
Descriptive Questionnaire - The Effect of Display.
Table 7.1 - Descriptive task regressions results.
The greatest correlation is between the Tracked condition and the
Real condition. The Monitor and the Real condition also had a high
correlation. The Fixed condition was perhaps the only one to be somewhat
different from the others, although a r2=.79 is a rather high
correlation.
Descriptive Questionnaire - Discussion.
Additional Questionnaire Results.
Sketches Task.
Sketches Task - Cognition Map Accuracy.
Sketches Task - Subjective Evaluations.
General Experience with Interface - Wayfinding and Presence.
Interface Evaluation - General Observation.
