Portfolio
Ping Du
Ph.D. Candidate
Iowa State University
Email: pdu@iastate.edu
Du, Ping - Portfolio
Research Project 1: Eye-tracking data predict feature importance and size change saliency
Corresponding Publication: Du, P. and MacDonald, E., 2014, "Eye-Tracking Data Predict Importance of Product Features and Saliency of Size
Change," Journal of Mechanical Design, 136(8), 081005.
Features, or visible product attributes, are indispensable product components that influence customer evaluations of functionality,
usability, symbolic impressions and other qualities. Two basic components of features are visual appearance and size. This work tests
whether or not eye-tracking data can (1) predict the relative importances between features, with respect to their visual design, in
overall customer preference; and (2) differentiate noticeable and unnoticeable feature size changes.
Stimuli: Images of cars and electric bicycles were prepared in Photoshop by merging different feature designs into base images. Sizes of
some features were also changed to form several levels of size changes.
The side mirror in the right car is 20% larger
The handlebar in the right electric bicycle is 10%
than that in the left one.
larger than that in the left one.
Figure 1. Sample feature design variations and size changes
Method: A computer-based
experiment with two general
conditions that differ in stimuli
presentation (Side-by-Side vs.
Sequential) was created. Survey
questions in the experiment were
regarding product preferences
(Sections I and II), product
differentiation (Section III), feature
importance rating (Section IV), and
demographics (Section V).
Participants’ eye movements were
tracked by a screen-based eyetracker throughout the experiment.
Figure 2. Experiment flow (Side-by-Side condition)
Du, Ping - Portfolio
Research Project 1: Eye-tracking data predict feature importance and size change saliency
Analysis: (1) Areas of Interest (AOIs), as shown in Fig. 3, were
manually created for each stimulus to obtain the eye-tracking
Fixation Time
data for particular product features, using the Attention Tool
Fixation Count
software. (2) Both R and JMP were used to process the eyeFirst-Located Time
tracking data and survey responses, and to conduct statistic
analysis (e.g. correlation analysis, ANOVA test, linear regression
and logistic regression).
Figure 3. Sample Areas of Interest for the car
Results: Based on 70+ participants’ inputs, the results demonstrate that feature importance is significantly correlated with a variety of
eye-tracking data. Results also show that there are significant differences in fixation time and count for noticeable vs. unnoticeable size
changes. Statistical models of gaze data can predict feature importance and saliency of size change.
600
400
200
0
0
1
2
Noticeable size change
Side-by-Side Condition
3
4
5
6
7
Importance Ratings
(1=Not important at all, 7=Very important)
Figure 4. Sample results showing the correlation between feature
importance and eye-tracking data
Average Fixation Time on
a feature pair with size change
(ms)
Average Fixation TIme (ms)
Sequential Condition
2000
*
Unnoticeable size change
***
1500
1000
(Note:
‘*’ p<0.05,
‘***’ p<0.0001)
500
0
Sequential
Seq
Condition
Side-by-Side
SBS
Condition
Figure 5. Sample results showing the difference between the noticeable and
unnoticeable size changes in the eye-tracking data
Implications for Design and Design Research: This study suggests that feature importance can be identified at the individual level in
only three questions, without directly asking about feature importance. This could (a) significantly reduce people’s mental burden
associated with current methods such as discrete choice analysis and complex rating schemes and (b) remove context effects caused
by drawing attention to the purpose of the experiment (ascertaining feature importance), and instead let people evaluate products
naturally. The demonstrated potential possibility of using the eye-tracking data to identify whether or not someone notices a change in
the size of a product feature can be used in a variety of ways, such as determining when manufacturing imperfections in the form of
geometrical variations are noticeable.
Du, Ping - Portfolio
Research Project 2: Products' shared visual features do not cancel in customer decisions
Corresponding Publication: Du, P. and MacDonald, E., 2015, "Products' Shared Visual Features Do Not Cancel in Consumer Decisions", Journal of
Mechanical Design, 137(7), 071408.
Customers’ product purchase decisions typically involve comparing competing products’ visual features and functional attributes.
Companies strive for “product differentiation”, which makes customers’ product comparisons fruitful but also sometimes challenging.
Psychologists that study decision-making have created models of choice such as the cancellation-and-focus (C&F) model. C&F explains
and predicts how people decide between choice alternatives with both shared and unique attributes: the shared attributes are
“cancelled” (ignored) while the unique ones have greater weight in decisions. However, this behavior has only been tested with text
descriptions of choice alternatives. To be useful to designers, this study tests the C&F model with product visuals.
Stimuli: Two types of structured product pairs: Unique-good and Unique-bad pairs as
shown in Fig. 6 were formed. The “good” (harmonious with the product styling, though not
necessarily beautiful) and “bad” (ugly and/or mismatched with the overall product styling)
feature designs were selected and modified from web images using our design expertise.
The good and bad designs were verified with participants in a pilot study using a
combination of card sorting and rating approaches.
Unique-good Pair
Unique design
Shared design
Good design
Unique-bad Pair
Bad design
Figure 6. Sample unique-good and unique-bad pairs
Method: A computer-based experiment was conducted. It had six conditions defined by:
the representation mode (text-only, image-only, and image-with-text) and presentation
(sequentially, or side-by-side) of choice alternatives. In the experiment, participants
evaluated product stimuli and answered survey questions as shown in Fig. 7. Eye
movements of the participants were tracked during the process.
Pair 1 – Car X
Page Break
Page Break
Page Break
Pair 1 – Car Y
Page Break
1. Please compare Car Y to Car X and indicate your
preference using the scale below:
Strongly
Slightly
Slightly
Strongly
prefer
prefer
prefer
prefer
Car X
Car X
Car Y
Car Y
2. Please evaluate your decision according to the following
instructions.
(1) Please think about the car you prefer in this pair and rate
your satisfaction with the decision using the scale below:
Very
Slightly
Slightly
unsatisfied
unsatisfied
satisfied
(2) Please rate for Car X using the scale below:
Very
satisfied
Slightly
Very
Slightly
good
bad
bad
(3) Please rate for Car Y using the scale below:
Very
good
Very
bad
Slightly
bad
Slightly
good
Very
good
Figure 7. Illustration of the experiment process
(Image & Sequential condition)
Du, Ping - Portfolio
Research Project 2: Products' shared visual features do not cancel in customer decisions
Results: Based on 100+ participants’ inputs, we identify
the inability of the C&F model to predict preference or
post-preference evaluation trends in unique-good and
unique-bad pairs when the choice alternatives include
images. Results indicated that the shared visual feature
designs between alternatives may reinforce good or
bad impressions that are consistent with the valence of
these designs, even though they attract less gaze
attention than the unique ones. Using the eye-tracking
data, the study confirms in five out of the six conditions
that differences between choice alternatives attract
more gaze attention than commonalities.
Figure 8. Sample eye-tracking heat map showing the focus on the unique attributes
Implications for Design and Design Research: This study highlights the importance of shared visual features in design, an already
intuitively-important concept in fields such as industrial design. Whether designing features to be shared with product predecessors,
shared with products in the same product line, or shared with competing products, designers must study what reinforcements they
may create through shared features. Designers should consider and test attitudes and preference for potentially shared features in
addition to considering production costs and ease of mass-customization. Otherwise, they risk damaging customers’ overall
impressions of a newly-designed product or an entire brand portfolio with the presence of inappropriately shared features.
Additionally, product differentiation remains an important target in design, as unique features are confirmed to attract extra gaze
attention.
Du, Ping - Portfolio
Research Project 3: Investigate effects of visual product design cues on customer decisions (Ongoing)
Visual design of a product carries all kinds of information about the product itself. It leaves aesthetic impressions, indicates functionality,
usability and social significance of the product, and helps the product to be categorized. This study investigates if products can use visual
design cues on themselves to help communicate certain properties of the products to customers. The study also considers potential
differences between market segments in their product evaluation strategies and decisions.
Stimuli: Different designs of two case products (electric bicycle and electric heater) and a practice product (hand mixer) were modeled
and rendered in Solidworks based on available designs on the Web as well as my own thoughts. Those designs were following certain
rules in order to serve the study purposes.
Figure 9. Sample designs modeled in Solidworks
Example visual cues
selected to communicate
that the electric heater is
environmental friendly
Figure 10. Example product visual cues
Method: A computer-based experiment was created. It includes tasks like implicitly learning of cues, preference evaluation, rating
products’ environmental friendliness, completing a comprehension test, and providing demographical information. Participants’ eye
movements are tracked during the experiment.