Conceptual Overviews - Scatter Image Plots
A scatter image plot is a scatterplot with point markers which are images
representing each specific data point (defined by the scatterplot {X,
Y} coordinates). The basic idea of using images in a scatterplot is to
represent individual points as particular graphical objects, so that the
points with similar characteristics (e.g., same gender in a Height vs.
weight scatterplot, or individuals in a certain age bracket in a weight
vs. blood pressure scatterplot) can be easily identified. This pictorial
version of the scatterplot can reveal a great deal of information about
other aspects (in addition to the X and Y variables) of the problem; there
may be clustering of individuals falling into the same age brackets, or
of individuals of the same gender. These are only a few of the numerous
situations where using images for the points can significantly add to
the information that a simple scatterplot would not be able to depict.
In general, in this type of plot the points on the scatterplot are represented
by images rather than the point markers. Different image files (Statistica
supports bitmap, metafile, PNG, and JPG formats) can be attached to different
data points, and a weight variable can be used to scale specific image
markers accordingly. Unlike Scatter
icon plots, the images in a scatter image plot are not icons that
are defined by specific variables; they are rather user-selected graphics
files.
Scatter icon plots vs. Scatter image
plots. Both Scatter icon
plot and Scatter
image plot are basically scatterplots and for that reason show relations
between two variables; the two coordinates (in 2D plots) that determine
the location of each point (represented as particular graphical objects)
correspond to its specific values on the two variables. However,
the two types of plots differ in the sense that the individuals in a scatter
icon plot are represented by icons (star, polygon, Chernoff face, etc.)
that are defined by a set of variables, while in the scatter image plot
the individual points are represented by images that are nothing more
than pictures (symbolizing a common characteristic). Thus, the scatter
image plots cannot provide information about the within variability of
a set of variables as the scatter icon plots can do, where the variables
of interest are used to define the graphical object (icon).
Creating images for Scatter Image Plots.
The images that can be used in these types of plots
are graphical objects that can be stored in image files which can be attached
to different data points (defined by the scatterplot X and Y coordinates).
Statistica supports files in bitmap, metafile, PNG, and JPG formats. Although,
the choice of images is not an involved task, yet an appropriate choice
of images can help in depicting any hidden aspects (to be uncovered by
assigning images to points) much more clearly. Unlike scatter icon plot,
one has a choice to use as many images in a scatter image plot as desired
(even a different image can be used for each point). In a given situation,
it may be useful to use different sizes for the same image so as to depict
some important feature (e.g., suppose your data points represent animals;
you could use images of different sizes to denote differences in the animals'
weight). A weight variable can be used to automatically scale specific
image markers according to certain feature. For example, if gender is
used as the weight variable, then all images representing males will be
of the same size, but different from the common size of images representing
females in the scatterplot. Statistica provides options also for manually
controlling image sizes.
Applications. Two-dimensional
scatterplots are used to visualize relationship between two variables
(e.g. height and weight). There are many situations where a scatter image
plot can provide information that will remain hidden if a simple two-dimensional
scatterplot is used instead. For example, the scatterplot can indicate
the lack of homogeneity in the sample by forming distinctive clouds of
points in the graph, but the information as to whether this clustering
can be attributed to certain feature (e.g., gender in the height vs. weight
scatterplot) will remain hidden unless the points corresponding to males
and females are represented by different images in the scatter plot, i.e.,
a scatter image plot is created.
The fitting of functions to scatter points helps to identify and summarize
the patterns of relations between variables. However, in a given situation,
it may be more useful to fit different functions to different sets of
observations, rather than one common function. A scatter image plot can
provide guidance with respect the different functions that should be fit
to different groups of cases.
In other situations, the scatter image plot may suggest the use of piecewise
linear regression, if two different trends are observed for points represented
by two different images that were used on the basis of a cut off value
(e.g., 400o
C) of Temperature.
Finally, scatter image plots can be used to create interesting and appealing
presentations of data. By replacing the standard point markers with interesting
relevant pictures, perhaps scaled so as to highlight an important trend,
major conclusions can be "packaged" in engaging ways.
Scatter Image Plots in Statistica.
Statistica can create both 2D and 3D Scatter image plots. It provides
a several options to select images of the different forms and sizes. A
variety of image file types (bitmap, metafile, PNG, and JPG) are supported
that can be attached to individual data points. Scatter image plots in
Statistica can be created from the Graphs menu. Select Scatter image plot
from 2D Graphs
or 3D XYZ Graphs
to display the Scatter
image plot dialog box or 3D
Scatter image plot dialog box, respectively, in which appropriate
selections can be made to create the desired scatter image plot.