Explaining experimental changes in consumer behaviour in realistic settings using observational data Adriaan Kole, René de Wijk, Daniëlla Stijnen, Anna Maaskant Wageningen UR: Food & Bio based Research Consumer Sciences & Intelligent Systems group
Restaurant of the Future A normal university cafetaria? Look inside Diners being observed
A multifunctional research facility Restaurant for 200 persons Grand café Research kitchen Sensory laboratory Mood rooms Mind room 45 video cameras 7 video analysis workstations 3 km of cabling
Restaurant of the Future in the press
Laboratories for sensory and physiological research Measure physiological and emotional responses Study effects of food odor, taste and texture
The Restaurant; designed to study the effects of changing environments on consumer behaviour Light Odor Temperature Buffet lay out Price Assortment. Etc,
Observational research Observe food selection and consumption Study effects of environmental and social variables
The behavioral approach: Food choice in Restaurant 300 registered consumers: Age (yrs) 40.88 % Males 51% BMI 23.76 NeophobiaScore 26.56 Education High
Some results We know the nutritional content of the products, we know who is buying what: Clusters of consumers based on nutritional intake Consistency of repeated lunch selections Lunch composition; the effect of soup on the other lunch selections
Cluster analysis Three clusters of consumers 1 2 3 Lunch energy from carbohydrates (%) 31 42 53 Lunch energy from protein (%) 16 19 19 Lunch energy from fat (%) 52 37 24 Relatively healthy food choice carbohydrates : fat : protein 50%:25%:25%
Cluster analysis Three clusters of consumers 1 2 3 Lunch energy from carbohydrates (%) 31 42 53 Lunch energy from protein (%) 16 19 19 Lunch energy from fat (%) 52 37 24 Need more carbohydrates and less fat
Cluster analysis Three clusters of consumers 1 2 3 Lunch energy from carbohydrates (%) 31 42 53 Lunch energy from protein (%) 16 19 19 Lunch energy from fat (%) 52 37 24 Needs MUCH more carbohydrates and MUCH less fat
Cluster analysis Three lusters of consumers 1 2 3 Lunch energy from carbohydrates (%) 31 42 53 Lunch energy from protein (%) 16 19 19 Lunch energy from fat (%) 52 37 24 Basis for health intervention: how to move consumers from cluster 1 to 3.
Consistency calculated for: Nutrients: fat, carbohydrates & proteins Energy Price Weight Number of lunch items Location of lunch items Type of lunch items
100 90 80 Consistency of repeat purchase (%) Snack Bread Soup Salad Sandwich filling Drinks Sandwich Desserts Hot meals Butter Fruit Type of food/drink 70 Rate of repeat purchase (%) Some lunch items attract fewer, but loyal consumers Consistency Rate 60 50 40 30 20 10 0
while other lunch items attract more, but less loyal consumers Snack Bread Soup Salad Sandwich filling Drinks Sandwich Desserts Hot meals Butter Fruit 100 90 80 70 60 50 40 30 20 10 0 Consistency of repeat purchase (%) Rate of repeat purchase (%) Consistency Rate Type of food/drink
Most consumers are consistent with regard to their choice of buffets and 50 Number of consumers (% of total) 45 40 35 30 25 20 15 10 5 0 50 51-60 61-70 71-80 81-90 91-99 100 Consistency of repeat purchases (%)
Na with regard to lunch price, weight & number of items Price Weight No.products Consistency buffet Consistency food/drink Energy in lunch Proteins Fat Saturated fat Trans fat Unsaturated fat Multiple unsaturatedf Fat Carbohydrates Mono/disaccharides Dietary fiber 1.2 1 0.8 0.6 0.4 0.2 0 Coefficient of variation (st.dev/mean)
Na Consumers are not consistent with regard to the nutritional composition of their lunches. Price Weight No.products Consistency buffet Consistency food/drink Energy in lunch Proteins Fat Saturated fat Trans fat Unsaturated fat Multiple unsaturatedf Fat Carbohydrates Mono/disaccharides 1.2 1 0.8 0.6 0.4 0.2 0 Coefficient of variation (st.dev/mean) Dietary fiber
Summary Consumers pay relatively little attention to their nutritional needs in terms of energy and nutritional composition. They rely more on factors such as portion size and habitual routes along buffets.
Some results: consumption patterns Lunch composition; the effect of soup on the other lunch selections. Hypothesis: Soup is supposed to be relatively satiating. Hence, we expect fewer calories on a soup tray compared to a non-soup tray.
Some background information Soup and non-soup consumers are similar with regard to their personal characteristics. Soup and non-soup lunches are similar with regard to energy from macro-nutrients.
What else is in the lunch? (Percent energy from product categories) No Soup Soup Sandwich fillings 6.3 2.5 Sandwiches 6.8 3.9 Butter 1.9 1.1 Bread 19.0 10.6 Dessert 9.9 3.7 Drinks 23.8 23.9 Fruit 0.8 0.4 Green Salad 4.6 1.6 Meal Salad 4.6 2.9 Snack 14.8 11.6 Soup 0.0 34.1 Hot meal 7.1 3.2 Non-soup lunches contain more bread, sandwich fillings and snacks
What else is in the lunch?: Total energy Lunch Without soup With soup 1867 kjoules 1677 kjoules Over 200 lunches, the difference adds up to 10000 calories, or approximately less 3 LB bodyweight if everything else is constant(1 LB bodyweight = appr. 3500 cals).
Conclusions of the soup case Results support the satiating properties of soup. However, if soup eaters consume fewer calories, why are they not thinner... Or are they compensating on other meals?
Routine buffets: Faster visits result in more purchases % visits resulting in purchase Time of visit w.o.purchase (s) Bread 86 5,4 Fruits & Juices 91 2,6 Sandwich Fillings 75 4,0 Salads 51 5,0 Soups 72 9,1 r= -0.8, sig. Sandwiches 37 7,4 Desserts 65 3,7 Snacks 62 11,5
Routine buffets: % visits resulting in purchase Time of visit w.o.purchase (s) Bread 86 5,4 Interventions Fruits & Juices should 91 focus on less 2,6 habitual foods which are thoroughly inspected first. Sandwich Fillings 75 4,0 Salads 51 5,0 Soups 72 9,1 Sandwiches 37 7,4 Desserts 65 3,7 Snacks 62 11,5
Tracing walking patterns Challenge Observe the walking patterns of visitors to learn their attention patterns during choosing behaviour Solution LED-based tracking solution Camera s can observe individuals, even when camera s do not overlap Preliminary result Individual tracks showing browsing behaviour and choosing times Aggregated tracks showing hot zones
Entrance Juices Juices Sandwich Fillings The Standard Route Salads Sandwiches Bread Soups Bread Coffee & Tea Bread Soups Lemonades Coffee Cash & Tea register Snacks Desserts Sandwiches Desserts Cash register Lemonades
Tracking results: a possible basis for interventions Food choice behaviour is not random Food choice behaviour may be partly related to proximity of buffets Combinations of buffets suggest equivalence of choices (e.g., snacks vs sandwiches)
Summary 2 1. Restaurant well-suited for longitudinal studies on: 1. (variations in) consumer preferences/food choice 2. Repeat purchasing of specific products 3. Effects of behavioral and/or environmental interventions with regard to food choice
Interventions: the Effect of product information on routing 1.00 Relative time during menu selection (0-1.0) 0.80 0.60 0.40 0.20 Cola Cucumber DrinbBuffet FrenchFries FriedOnion Ketchup Lettuce Mayo Onion ReadHealthy ReadNew ReadWelfare-friendly SelectedChicken Tomato Vegetables Water 0.00 New Healthy Welfare- Friendly Selected chicken product
Interventions: the Effect of product information on routing 1.00 Relative time during menu selection (0-1.0) 0.80 0.60 0.40 0.20 Cola Cucumber DrinbBuffet FrenchFries FriedOnion Ketchup Lettuce Mayo Onion ReadHealthy ReadNew ReadWelfare-friendly SelectedChicken Tomato Vegetables Water 0.00 New Healthy Welfare- Friendly Order of food choice varies with the type of chicken product. Selected chicken product Different consumers, different strategies?
Intervention: effect ambient aromas on choice 100 90 P=0.02 % of visitors 80 70 60 50 40 P=0.05 P=0.03 CitrusAroma VanillaAroma 30 20 Citrus odour: more combination meals chosen Vanilla odour: more often fish/meat with staple foods Bron: FBR/CICS
Intervention: Effect of ads on food choice 60.0 50.0 % of visitors 40.0 30.0 20.0 SaladPhoto StirrFryPhoto 10.0 0.0 NoDessert FruitDessert VanillaDessert Photos of salads decrease demand for desserts. Bron: FBR/CICS
Intervention: calorie labeling 600 Reference period Information period 550 500 450 Average lunch energy (kcals) 400 About 30% of the consumers eats less calories Bron: FBR/CICS
... And 70% increases calorie intake Recuctions: toppings, bread, meals, cold snacks Increase: salads, warm snacks Bron: FBR/CICS
Interventions: CO2 labeling
Effect of new technology on salad sales 70 Humidifier increases attention for the salad buffet Percentage approach per day 60 50 40 30 20 10 0 60 59 54 46 Experimental condition 59 No humidifier Half humidifier Full humidifier 58 Humidifier decreases salad sales (food technology phobia) Percentage choice per approach 56 54 52 50 48 46 44 50 51 No humidifier Half humidifier Full humidifier Experimental condition Bron: FBR/CICS
Effect of product on food choice
Eating gesture analysis Challenge: observe the eating gestures to learn about eating habits and emotions (liking/disliking) Solution Formalised framework describing which eating gestures exist Computer vision software to detect occurring gestures Preliminary result Individual eating gestures and eating patterns can automatically be detected
Smaller bites result in smaller meals P < 0.05 P < 0.001 P < 0.05 500 LS HS Intake (g) 400 300 200 100 small large free 30% reduction in grams eaten when taking small bites compared to large bites Bron: Dieuwerke Bolhuis. WUR/HV
Eating faster is eating more, esp. with smaller bite sizes Intake (g) 600 500 400 a a b c Short (20 s/100g) Long (60 s/100g) 300 200 LB 6.7 bites/100g HB 20 bites/100g Lab: bite sizes can be reduced by using stronger flavours Bron: Dieuwerke Bolhuis. WUR/HV
Conclusions and discussion Effects can be described empirically (this is what we observe that consumers do): observational measures We can influence what people do and observe.
Conclusions and discussion Usually psychological theory explain effects based on how people are and how their brain works. If we only observe we don t know these things How can we bridge the gap between lab and real life.
Conclusions and discussion Or do we not need to, since only the effects matter??
Thank you for your attention! Contact: adriaan.kole@wur.nl