64 TEACHERs COLLEGE COLUmbIA UNIvERsITy
A multi-stage Attention Hypothesis of Drawing Ability
rendering that contributes to successful object rec-
ognition—exactly along the lines of psychological
theories of object identification (e.g., Biederman,
1987).
In revisiting the question of what underlying
skills and strategies are associated with better draw-
ing ability, we suggest that one answer is simply
more astute selection of what information from
a model should be emphasized in a drawing. As
argued by Gombrich (1960) and Kozbelt and See-
ley (2007), artists, as a result of intensive training,
develop knowledge about what visual information
will best capture the illusion of the 3-D form of an
object in a 2-D depiction. In this view, when trained
artists view a model to draw, their visual attention
is biased to focus on essential visual features that
will most accurately represent the specific appear-
ance of the model, while ignoring information that
does not contribute to recognition. Our evidence
(Ostrofsky, Kozbelt, & Seidel, in press) hints that
vertices may be an example of such information,
which is preferentially selected and emphasized in
depiction. Future research may investigate other
visual cues that are emphasized differently between
skilled and unskilled drawers – a potentially fruitful
direction of inquiry.
Enhancement of processing selected
information and suppression of
processing non-selected information
The results from our limited line tracing task
suggest that what visual information is selected to
be included and excluded in an observational draw-
ing will influence how accurately the drawing rec-
ognizably represents the model stimulus. However,
the subsequent visual processing of selected versus
non-selected information can also potentially influ-
ence drawing accuracy. It has long been known that
when a visual target is selected to be processed,
attentional resources are dedicated to preferentially
processing that stimulus, versus stimuli that have
not been selected (Kastner & Ungerleider, 2000).
However, if the perceptual resources that are dedi-
cated to processing the selected target stimulus do
not exceed the capacity of our perceptual systems,
additional resources will be allocated to the pro-
cessing of non-selected, ignored information (Lavie,
et al. 2004). Under some circumstances, the addi-
tional processing of non-selected visual information
can potentially cause errors in perceiving and acting
on the visual information that has been selected for
processing (Franz, 2003).
Perhaps nothing demonstrates this point bet-
ter than visual illusions. Many visual illusions arise
from perceptual judgments about a target stimulus
attribute being distorted by task-irrelevant, non-
selected contextual cues. Take the observation of
size constancy, for instance (Murray, Boyaci & Ker-
sten, 2006). Size constancy is observed when two
objects of identical size and appearance are presented
simultaneously in conjunction with a contextual
background composed of depth cues such as linear
perspective, texture gradients, and shading. The two
objects are typically presented as being at different
perceived distances from the observer. If individuals
are asked to compare the optical size of these objects,
most individuals tend to judge the “farther” object
as larger in size than the “closer” object, presum-
ably because the visual system makes the assump-
tion that objects perceived to be farther away are
larger in objective size than their appearance indi-
cates. Since this misperception disappears when the
objects are presented at equated perceived distances,
this clearly demonstrates that the processing of the
selected visual targets (the size of the two objects)
is influenced by the processing of the non-selected,
task irrelevant contextual depth cues found in the
background. This also makes the more general pointFigure 3. Displays the average frequencies in which
artists and non-artists traced T-, L-, Fork and Ar-
row Junctions in their reproductions of the elephant
photographic model. A star above a comparison in-
dicates that the difference between artists and non-
artists is statistically reliable at the α = .05 level.