HUMAN-COMPUTER INTERACTIONRisk Info Based on Food
log Wenjun Sun & Xuan Zhang
Department of ISyE & EEUniversity of Wisconsin–Madison
CS/Psych-770 Human-Computer Interaction
• Risk info plays significant role in decision making, especially for diabetic patients (Mühlhauser & Berger, 2000; Adrian Edwards, 2006; Bernard D Frijling, 2004 )
• Numerous diet and calorie tracking applications & websites http://www.fitday.com/ http://www.my-calorie counter.com/calorie_counter.asp
https://www.supertracker.usda.gov/default.aspx
• No application provide risk relation information in the market through our research.
Motivation & Related work
Mühlhauser, I., & Berger, M. (2000). Evidence‐based patient information in diabetes. Diabetic medicine, 17(12), 823-829.Edwards, Adrian, et al. "Presenting risk information to people with diabetes: evaluating effects and preferences for different formats by a web-based randomised controlled trial." Patient education and counseling 63.3 (2006): 336-349.Frijling, Bernard D., et al. "Perceptions of cardiovascular risk among patients with hypertension or diabetes." Patient education and counseling 52.1 (2004): 47-53.
Research Question
The effects of risk info on decision making of eating habit based on logging and tracking summaries (among diabetic
patients).
Instrument
Smartphone App
Take food pictures-calculate the risk-feedback
Wizard-of-OZ Method
Take food pictures- send to us
-we calculate the risk
-send feedback via email
Video
Risk calculationStep1: Estimate main nutrition compositions through
crowdsourcing. Mechanical turk+link to google doc.
Step2: Calculate the numerical risk value using food healthfulness metric, then convert negative health scale into risk value scale from1.0 to 10.0
Budget: $0.02 per HIT × 10 HIT ×10 participant × 7 days = $ 14.00
Martin, J. M., Beshears, J., Milkman, K. L., Bazerman, M. H., & Sutherland, L. A. (2009). Modeling Expert Opinions on Food Healthfulness: A Nutrition Metric.Journal of the American Dietetic Association, 109(6), 1088-1091
Hypotheses • Condition1: Pictures with descriptive risk info
• Condition2: Picture with numeric risk info
• Condition3: Meal pictures only
Change/Decrease of the risk food taken:
Condition1>Conditon2>Condition3
"A picture is worth a thousand words"
Risk Calculation based on food healthfulness metric
Martin, J. M., Beshears, J., Milkman, K. L., Bazerman, M. H., & Sutherland, L. A. (2009). Modeling Expert Opinions on Food Healthfulness: A Nutrition Metric.Journal of the American Dietetic Association, 109(6), 1088-1091.
Low Risk
7.8
Study design• Variables:
Demographic variable: age, gender Dependent variable: Risk value of the food
• Task
Take pictures of food they eat, get risk feedback via email
• Participant population: convenient sample-- 15 friends5 participants in each group9 males, 6 females
All graduate students Average Age 24.6 (1.85)
Data analysis methods• Calculate the mean risk for three meals each day. Get one risk scale for each day.
• Fit linear mixed model for each group, get slope of risk scale change.
mod3<-lmer(risk~Day+ (Day| ID), data = dsub1)
• Testing using contrast to see if the slope of three groups
POCList = list(c(1,0,-1), c(1,-2,1)), Labels = c("c1group", "c2group")) mod2<- lm(abs(k)~ c1group+c2group+age+sex
Day 1
4.4 7.2 6.8
day1
6.13
Results
• No significant difference between three groups
(b=-0.022, t=-0.414, p=0.6874)
• Compare the change of two intervention groups with thechange of control group
Marginal significant
(b= -0.067, t=-1.890,
p= 0.0853 < 0.1
Holm-Bonferroni adjustment)
Results• Pair Comparison
All adjusted by Holm-Bonferroni adjustment
• No significant difference
Group 1 V.S. Group 2
Group 1 V.S. Group 3
Group 2 V.S. Group 3
p=0.547 p= 0.547 p=0.160
Discussion• Based on the trend of the data, participants who received risk information feedback are more willing to consume low risk food, even though no significant effect was found.
• It is promising to provide risk information as intervention to change users’ behavior.
• Difficulties of risk value estimate.
• Limitation of sample size and time.
QUESTIONS?Wenjun Sun & Xuan Zhang
Department of ISyE&EEUniversity of Wisconsin–Madison
CS/Psych-770 Human-Computer Interaction