Chapter 2 has very simple thesis.
And it comes from
Jock Mackinlay who is one of
the I think the trailblazers in
modern computational visualization.
And he came up with this idea
that he called Effectiveness.
And he basically says,
look, how can we tell if one
visualization is better than another one?
And he just came up with this heuristic,
but I think it's very effective.
And it relies on our understanding
of human perception.
So you can see he says a visualization,
it,
it's just more effective than another one,
if the information it's conveyed is more
readily perceived than the information
in the other visualization.
So in other words, in, if you want to
learn how to make visualizations that can
communicate powerfully, it's critical
to understand how the human mind
perceives shape, color, line,
space, motion, and interaction.
And these elements together, shape how we
actually perceive what's on the display.
So in this section,
I'm going to quickly talk about
some of the important qualities of human
perception and see how they affect
the way that we should structure
data when we present it visually.
So the first is just a quick walk through
of Gestalt's ecology, which is basically
a school of psychology that came out of
Germany, and it really describes something
very interesting, which is that the mind
looks at objects in their entirety,
before or in parallel with
perception of individual parts.
And this means that,
you know, there's this very
intriguing statement at the end here.
The whole is other than
the sum of its parts.
So, let's talk about what this means.
Look at this first concept here.
The idea is called similarity.
And what it means here,
is that the mind takes shapes
that have similar semantic relationship.
Shape, and it brings them together.
So what you see here is this eagle shape,
but
around the eagle are really effectively
just a series of triangles.
But our mind reads that those triangles
create a kind of halo or circle behind
the form of the eagle, if you look
at the shapes, just by themselves.
So don't think of them as a halo.
What they are is just
a series of triangles, but
the mind reads them differently.
So what's important to understand is how
the mind reads these symbols together.
I'll give you another example here.
The principle is called Anomaly.
An anomaly, very simply, tells the story
that if there's continuity, and
a break in the continuity,
that that shape,
that is no longer contiguous
is going to stand out more.
So what's important in human perception,
is putting shapes together and
identifying shapes that stand
out from that togetherness.
This has several interesting consequences.
The first,
is this principle called Continuity.
And what this means, is that we
the human mind will look at the cross,
the swoop that goes across that H and
read that as a continuous path.
And while it's just a shape,
it's, it's a negative space.
What we really read that to mean,
is that the leaf has blown through the H.
So there's meaning in the con,
the connection and
continuation of shapes that
are next to one another.
Another important idea is
something called Closure.
And what this means, is that the mind
makes shapes contiguous that
are not formed entirely.
And so if you look at the top of this
panda bear shape, what you'll see is
there's actually no top of the bear drawn,
but your mind draws it in for you.
So you can see how clearly the human mind
is really interesting at forming patterns.
And these patterns are not
necessarily visible.
So it's critical that we understand
these patterns when we're
putting together things that
are more data oriented.
So this other concept is called Proximity.
And Proximity means that the mind believes
that things that are closer together,
have more meaning.
In other words, the dots,
the squares in this particular form,
are less related than the forms
in this particular shape.
So when we see dots that
are closer together,
we interpret that to mean similar.
So it's really important
that we understand that,
because you'll see that as you create
visualization, it's very often
easy to accidentally put things together
that don't have the same semantic meaning.
So we need to be very careful of
this because when we see shapes,
we infer meaning.
Another example of proximity here,
is the way that
we just look at this distribution
of shapes that are the human form.
And because of the way that
they're approximal to one another,
we interpret this as movement and
as a formation of people.
But if you tried to describe
this in computational terms,
it's really just the same image,
just in different orientations and
locations relative to one another.
But the human mind reads this as motion.
A cloud.
So let's talk about how
the way that these features in
the human perception affect the way
that we should design visualizations.
So I'll ask a question.
Very simple,
look at the following shapes and
say which of the, which of them is bigger.
So it's pretty clear that the there's
a tall and short rectangle.
Now here's this question.
How much bigger is the tall rectangle?
Well, the human mind can make this
assessment relative much better,
than it can this assessment.
So interestingly, if we were to say,
how much taller is the tall
rectangle than the short rectangle,
it's far easier for the human mind to see
that actually it's three times as tall.
Make the same determination
of this circle.
It's actually significantly harder for
us to make that same determination.
But interestingly,
they are exactly the same heights.
Let's talk about color for a moment.
Let's ask the question, which of
the following squares is brighter?
Give everyone a minute,
make a determination.
Okay.
So the answer is the one on the left
is slightly brighter, but what
you'll see is many people will actually
not have been able to tell the difference.
Let's look at another one.
Look at this pair of images and
say which is brighter.
Actually, what you can see is
the one on the right is brighter,
if we describe them in RGB space.
And this amount of difference is actually
perceivable by a large number of people.
So when we look at the way
the human mind understands color,
what we have to understand is that
even though there are differences
that we might be able to label
in RGB space, the human mind
is only capable of perceiving what we
call the just noticeable difference.
In other words,
there're some steps in color space,
that are not perceptible
to the human mind.
So what we have to realize,
is that if we don't allow for
enough space between individual color
values in a scientific visualization,
a human mind won't be able to
perceive key differences in the data.
This is a common, common,
common problem you see in scientific
visualization, that uses continuous color.
It overlooks this very
important fact of perception,
which is that if you don't allow for
humans to perceive the differences,
your data will effectively
become subsampled.
Because people won't be able to see
the difference between a value and
another value that are close
enough in color space,
even though their differences
are clear in numerical space.
So let's think about,
what is this mean, if we're going to
use how the human mind perceives color,
and bring this into a visualization.
well, actually color and shape, et cetera.
So, what we find important is
this idea going back to McKinley.
The one of the early
visualization thinkers,
is he put forth these heuristics.
The first being something called
the principal of consistency.
And that means that
the properties of the image
should match the properties of the data.
In other words if something's big,
the data should make it look big.
If it's small, the data should look,
make it look small.
And then the second idea is the importance
of the principle of importance ordering.
And this is something
that we often overlook.
The idea is that take the most
important variables, and
encode them in the most effective way.
And our understanding of human perception,
is such that we perceive differences of
magnitude differently when they're
encoded using different encoding methods.
And so
if you can see that position is much more
clear of a difference to
the human mind than color.
So what this means is that when you
are making a visualization, the shape,
excuse me, the,
the variables that are the most important,
are often best to map to position and
length rather than color,
because it's hard for the human mind
to perceive differences in color.
And this is something that you see so
very often in scientific visualization.
That, someone will make an example where
they're trying to show difference in
volume, where the same numerical
difference would be much
more clearly demonstrated by differences
in length, or in angle, or in slope.
So you have to choose the dimensions
that you have available, and
think about the saliency that they
have when people are perceiving what
they're looking at, and
think about it not you as the scientist,
the creator of the visualization,
but the human who is not you,
who doesn't know the data,
will look at something and
not be able to perceive differences
that are visible to you.
So let's try and
quantify this a little bit specifically.
This'll be something that will
probably have occurred in a number of
the other lectures.
If we take data and
map them to different types.
So the first is nominal.
These are often category labels.
So if you have data that describe fruits,
the categories would be things
like apples and oranges.
There might be numerical data, but
it's of an ordinal nature, or numerical or
non-numerical data that's
of an ordinal nature.
So the example I put here
is the quality of meat,
where the example would be that you
have A, and double A, and triple A.
And what they mean is,
that they are relative to one another,
but magnitude is not important.
Double A is probably
not twice as good as A.
And then lastly, qualitative, which,
excuse me quantitative, which is
numerical and often continuous values.
That's something like
a measure like length.
So let's think about,
if we have data that have certain types,
that if we want to map those
particular values to the way
that the human mind perceives,
you'll see that quantitative values,
ordinal values and
nominal values all map best to position.
But you'll also see that if you,
the next best value for nominal is hue.
So that means color.
So if you want to label categories,
the, one of the best ways
is differences in color, but I think we'll
show in a moment, not continuous color.
Whereas if you look at
hue under quantitative,
it's actually about two thirds
of the way down the column.
So if you want to show a number, and
it has magnitude, and it's a quantitative
number, and you're picking hue as the
mapping, you should probably think again
about that, because it's not going to
be easy for the human mind to perceive.
Let's pick a more specific example.
color.
So we've talked previously
about this idea, that for,
that hue is one of the important ways
that we can encode a nominal value.
But if you look here,
what you can see is that the human mind,
based on what we described earlier, and
our ability to perceive differences,
means that on a,
a color scale like black to white,
that the human mind is at best able
to perceive seven differences.
And so, if you look at the continuous
values, you can see that co, hue can
encode continuous values, but
it's far less good than its ability
to include ordinal values, which you
can use with different magnitude,
with hue of a single
monotonically changing function.
Another thing that I think you can use,
is that hue is normally
perceived as unordered.
And so, this means that hue is a really
good way to label nominal variables.
And so, when you have things that are,
have no ordering, picking color is
really valuable, because color,
hue does not have an order.
Whereas if you look at color value,
it does have an order.
And so,
it effectively is not nearly as good for
nominal values,
than the unordered color values.
And we could show how
these are applied if we
look at different kinds of
the way that color is used.
So if we look at for example in this
illustration from an anatomy textbook,
how different anatomical features
are colored using different values.
So here, red might mean artery and
blue might mean vein and
yellow might mean capillary.
But they're effectively nominal values.
So colors being used nominally.
You can look similarly at this
presidential election map from the from
NBC, and you can see here
that we've also used color in
a way where red is mapped to
the Republicans and blue to the Democrats.
So here,
we've just simply taken two colors and
made a very easily distinguishable map.
But if we look at quantitative uses
of color, and this is where the most
often you most often see mistakes in Earth
science data visualizations by
quantitative applications of color.
So here, what you have is an example
of Global Average Temperature and
it's rendered from blue to red.
Now there're a number of reasons based
on what we just learned about color,
that described why this
is not the best mapping.
So first of all, the,
there's a, it's very hard for
the human mind to perceive more
than seven seven different colors.
So it's really hard for a hum, for
a person to look along the equator and
tell me the different average
temperature between Panama and Senegal.
And they're probably not the same,
but in red, they're relatively close.
So it's hard for
the eye to distinguish values like that.
Another example here is this
fluid dynamics simulation,
where velocity is encoded as color.
And what you see here is that these hues,
there is no natural ordering.
If you look at the yellow in the middle,
it's really hard to tell where
that lies in color space.
And different,
our ability to perceive light
is emphasizes different
kinds of scalar values.
So if you look at the way that color
can be applied in an ordinal fashion,
it's very useful to pick
a a single saturation,
excuse me, to vary luminance and
saturation together, and you get these
very nicely marked color scales.
Where what you're able to see very clearly
are the demarcations between each of
the different values.
And you can do the same for
different types of color,
where if color is diverging, in other
words, if you want to emphasize the mid
point of a color range, you can
pick a neutral color in the middle.
And when you apply this to a map,
you'll see that the midpoint
that use saturated colors on
the different end points, that they'll be
distinct from the midpoint of the map.
So just at a high level review,
if you're going to be thinking about
color in data visualization,
you should use only a few colors and
that, try to be able to make
them as distinct as possible.
Strive for harmony,
be very cautious of cultural conventions.
Be cautious about bad
interactions between colors.
And one of the best ways to determine
if your visualization is going to be
effective, is before adding color,
see if without it you can distinguish the
different variables in black and white.
We'll ahead into the next chapter.