Visualization designers usually start with good common-sense ideas about human perception, attention and cognition. Often these are now formalized into predictive theories and even computational models. As technology advances speed up, it is easy to assume that we know all about how end users work, forgetting everyday variations in age, expertise level, real-world knowledge and task goals. This talk will consider the adaptability of human visual processing to such factors, and some practical implications for work with geospatial visualizations.
1. This talk had four main areas: the lessons we've been learning from visual attention experiments in recent
years (upper left); our growing understanding of the geospatial knowledge that people bring to bear about an
experienced environment (lower left); specific issues around users and tasks, specifically concerning
expertise and age (lower right), and some tentative research agenda items that arise from all this (upper
right).
2. So, the basic idea was to
try to see some new hints
or insights about the
person holding the mobile
device, or selecting
among those layers of GI
data.
3. Part 1: human visual attention, as
applied to geographic
visualisations. (Most of my own
work on this particular area has
been with aerial imagery, often
combined with an outline map
overlay.)
4. There's been a lot of
interest in visual salience
maps over the past 15
years, but they seem to
predict only the first few
hundred ms after
someone first sees an
image, without expecting
what would be in it, and
without having a
particular task to do with
it. Clearly, that's not the
scenario of most of our
users, most of the time -
and they also have more
time to find what they
need.
5. The backlash against
visual salience has
perhaps gone too far
the other way in some
respects: De Graef's
statement is one I'd
very much disagree
with. Except when
watching movies or
TV, or walking through
a strange town, most
of the scenes we see
are pretty much what
we expect to see next:
we don't have to
explore them for
meaning at all (and
may not even pay
much attention to
them). It's not like
we're going "Hey look,
a street! I'm suddenly
in a forest! Oh, there's
a map on my GIS
screen!"
However, the three
stages suggested by
DeGraef as what we
do when we are faced
with an unexpected
image or scene are
quite useful: we
probably do apply top-
down 'schemas' from
our past knowledge, in
deciding where to look
and what to pull out of
the scene.
6. It's tempting, then, to
start seeing human
visual attention as
entirely top-down
(expectation)-driven.
However, visual salience
(in other words, good
design - highlighting the
important stuff with
colour, contrast etc. as
we've all been taught to
do) does help even in
expert scenarios, as
these graphs from one
study show. If the item
that people had to spot
change in, or recall later,
was close to a cluster of
more visually salient
items, even expert
performance improved
(this was
photogrammetrists who
do such change
detection as a major part
of their job, versus
relatively novice viewers
of aerial imagery).
7. At this point, I showed a brief video of an expert
change-detection photogrammetrist, looking
over some British suburban aerial imagery to
spot changes (comparing both building outlines
and street features). I showed that although he
chose to scan systematically up and down the
streets, a lot of the work he was doing used his
peripheral rather than foveal vision: this is an
important feature of attention that eyetracking
studies can't capture. Also, a common problem
with dynamic attention tasks such as this is that
people aren't completely reliable in recalling
where they've already looked (hence this user's
attempt at a systematic strategy - but this falls
down when there is multiple panning among
screenfuls, as here). Thus things can be
missed, even by expert users. Still, visually
salient items can help to efficiently direct the
eye around a visualisation.
8. And so to part 2, the spatial
cognition that may lie behind/
above our basic visual processing.
9. Of course, our visualisations are
'geo' - that's what makes them
so interesting… so what do we
know about human
understanding of (experienced)
environmental space?
10. In 2014, the Nobel prize for medicine and
physiology was shared between scientists
who had discovered place cells in the rat
hippocampus - which 'learn' to fire only when
the rat reaches a particular location within a
well-learned maze - and grid cells, which fire
at a regular series of locations across the
space, as if forming some sort of coordinate
system. Left out of the Nobel laureate, but
potentially equally important, were 'head
direction' cells, which only fire when the rat is
looking in a certain direction.
11. How the international
media reported the
story… but how come
we're actually not very
GPS-like? What else is
going on? First, it's
important to remember
that place and grid cells
have that amazing
accuracy only in very,
very over-learned
spaces. Prior to (and as
well as) that, we need
other sources of help.
12. Paul Dudchenko's book
is a useful summary of
some of this. In
particular, and unlike a
GPS, there's a larger
focus in human
navigation on visual
landmarks, and on the
route topology (not
always accompanied
by accurate metric
knowledge).
13. There seems to be a
particularly strong brain
pathway that simply links
what you see right here,
with where to go next.
There are even
'conjunction' cells that link
head direction with place -
i.e., that fire at a particular
spot *only* if looking in a
specific direction there. So,
until our place cells are
fully tuned for a given
environment, this may be
one key source of help.
14. Rather nicely, this ties in with what we've seen for many years in behavioural studies of wayfinding and
spatial cognition. With a relatively unfamiliar environment, people tend to be quite inaccurate, and non-
transitive, in distance and direction estimates - but much stronger on identifying route choices from vistas at
decision points. Thus our spatial knowledge seems to be not a 'map' but a mishmash of visual, topological
and metric information - what Tversky (1993) called a 'cognitive collage', but Meilinger (2008) theorised
more predictively as a 'network of reference frames' theory.
15. OK, so we could crudely think of our spatial knowledge as only loosely
metric (preserving distance and direction), but more strongly topological.
Reminds us not of topographic 'maps', but of...
16. Topological maps deliberately (unlike us?) lose some of the metric accuracy of the real space. If you look at this
version of it, the dotted lines show stations which are actually only 5-10 mins' walk from each other.
17. But look what happens when someone tries to make the map closer to the real spatial distances and directions among stations (though
this version still had to distort to leave room for the names). It immediately looks more complex to us, and this hints a reason why our
brains rely on topological simplifications of space: the principle of 'cognitive economy'. It's a lot more efficient! Thus, good visualisations
of a space that users have/will also experience (which, of course, isn't always the case in geovisualisation) need to reflect the key
aspects we use to orientate: landmarks, and connections.
18.
19. Part 3 of the presentation
considered some human factors
issues which are easily
overlooked in our user studies
and design processes.
20. Users may, or may not, know the actual space being depicted, as discussed above. Even if they don't, they may have (perhaps *should* have) some
expertise in its geography: the types of features it contains, and the ways they do and don't relate together. Outsourcing GI data creation or
visualisation to cheaper workers in another country, for instance, has been known to lead to expensive disasters where the cultural knowledge of the
geography was lost. On the other hand, expertise in the type of visualisation may often be overplayed, though sometimes important. In my own work
I've sometimes *not* seen any expert-novice differences with a specialist type of mapping, just because people adapt to visual representations
surprisingly quickly. Finally, while academics have liked to muse over what's "special about spatial", the USIS work of the early 1990s (Davies &
Medyckcyj-Scott 1994, 1996, in IJGIS) showed that often users are more cognitively overloaded by their complex GIS or other task and system issues,
than by the map.
21. Age is an aspect of individual differences
which has often been relegated to studies of
the elderly. But work such as that by Tim
Salthouse at Virginia has shown more
recently that actually, many of the effects
shown here (especially cognitive abilities)
tend to start falling off in our 20s, and in the
past this was masked by test-retest practice
effects - i.e., you'd done the test before (or
similar tasks) so you seemed to do just as
well as before, even though age was actually
making you slower. Thus we should be
designing, and user testing, NOT just with
20-year old undergraduate students.
22. A few more, slightly random,
thoughts based on my
experience of studies in this
area. To my knowledge, few
geovisualisation-using tasks in
the real world need to be done
in just a few seconds, yet a lot
of what we've done to establish
rules of visual attention,
perception and cognition was
done by comparing reaction
times in milliseconds (as with
the visual salience work above).
Screen size and zoom level, on
the other hand, are an under-
researched issue: there's a bit
of evidence that people working
constantly at a zoomed-in level
won't build such an integrated
overview of the space
(unsurprisingly). The user
interface is often more of a
problem for users than the map
is, if poorly matched to the task,
and often the task that users
have to do with
geovisualisations is a rather
mundane one of matching
geometry, colour etc. to a paper
plan: these tasks aren't actually
'geographic' at all in their
requirements. HCI has many
useful tools for task and user
modelling (see any HCI
textbook): I'd recommend
considering these for
geovisualisation design too.
23.
24. Finally, just a short
'wishlist' of research I'd
like to see, that would help
us to see where the issues
I've raised here do, and
perhaps don't so much,
matter for particular types
of visualisation and
application.