Food Bundling As A Health Nudge - Investigating Consumer Fruit and Vegetable Selection Using Behavioral Economics

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 63

Accepted Manuscript

Food bundling as a health nudge: Investigating consumer fruit and vegetable


selection using behavioral economics

Kathryn A. Carroll, Anya Samek, Lydia Zepeda

PII: S0195-6663(17)30846-2
DOI: 10.1016/j.appet.2017.11.082
Reference: APPET 3679

To appear in: Appetite

Received Date: 14 June 2017


Revised Date: 5 October 2017
Accepted Date: 3 November 2017

Please cite this article as: Carroll K.A., Samek A. & Zepeda L., Food bundling as a health nudge:
Investigating consumer fruit and vegetable selection using behavioral economics, Appetite (2017), doi:
10.1016/j.appet.2017.11.082.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT

1 Food Bundling as a Health Nudge:


2 Investigating Consumer Fruit and Vegetable Selection
3 Using Behavioral Economics
4
5

PT
6 Abstract
7 Displaying bundles of healthy foods at the grocery store is a health nudge that simplifies

RI
8 shopping and may have the potential for increasing fruit and vegetable (F&V) purchasing. To
9 evaluate the impact of food bundling, we conduct an artefactual field experiment with
community participants in a laboratory set up as a grocery store. Dual-self theory suggests that

SC
10
11 food choices may differ depending on whether shoppers are under cognitive load – in our
12 experiment, we exogenously vary whether bundles are displayed (with and without a price
13 discount) and whether shoppers are under cognitive load. Our findings align with prior studies

U
14 that suggest unhealthy options are more likely to be selected when cognitive resources are
15 constrained. When bundles are displayed, we observe increased F&V purchasing. We also
AN
16 observe a significant interaction between cognitive load and price discounting. We find
17 discounted bundles are more effective in the absence of cognitive load, but non-discounted
18 bundles are more effective when shoppers are under cognitive load. Although more research is
M

19 warranted, our findings suggest that when shopping under cognitive load, it is possible that
20 discounts impose additional cognitive strain on the shopping experience. For retailers and
21 policymakers, our results point to the potential power of bundling as a strategy for increasing
D

22 healthy food purchasing.


TE

23

24
25 Keywords: food choice; fruit and vegetable selection; product bundling; cognitive load; dual-self
EP

26 theory; behavioral economics; health concerns; grocery shopping


C
AC
ACCEPTED MANUSCRIPT

1 Food Bundling as a Health Nudge:


2 Investigating Consumer Fruit and Vegetable Selection
3 Using Behavioral Economics
4

5 INTRODUCTION

PT
6 Nearly 70% of Americans are overweight or obese, putting them at risk for a range of health

7 problems including early mortality, diabetes and cardiovascular diseases (Flegal et al., 2012). A

RI
8 major component of this problem is the habitual consumption of large quantities of low-nutrient,

SC
9 high-calorie foods (NIH, 2012). Recent research on increased fruit and vegetable (F&V)

10 consumption has been linked to obesity prevention (He et al., 2004) and the reduced risk of

11

U
cardiovascular disease (Hung et al., 2004; He et al., 2007). For many Americans however, a gap
AN
12 exists between the amount of F&Vs actually consumed and the amount recommended. In the

13 U.S., approximately 23% of adults report consuming vegetables less than 1 time per day, while
M

14 the recommendation is with every meal (CDC, 2013).


D

15 One explanation for low F&V consumption is that consumers often shop for food under
TE

16 cognitive load; that is, while mentally strained (Park et al., 1989). The average number of items

17 in U.S. supermarkets exceeded 39,000 per store in 2015 (FMI, 2015). Grocery shoppers then
EP

18 regularly search through a large number of products before making a purchasing decision, yet

19 routinely shop under time pressures (Aylott and Mitchell, 1998). Grocery shopping is often
C

20 cognitively stressful, especially when required to perform in-store information search activities
AC

21 (Park et al., 1989). The effect of cognitive load on food choice then is particularly relevant when

22 examining consumers’ grocery selections. Prior studies have shown that cognitive load increases

23 exertion of the reasoning system, and can impact preferences and decision making (Hinson et al.,

24 2003; Greene et al., 2008; Benjamin et al., 2013; Deck and Jahedi, 2015). This can lead to a

25 ‘dual-self’: where different decisions are made when cognitively impaired than when not

1
ACCEPTED MANUSCRIPT

26 (Kahneman, 2002; 2011; Fudenberg and Levine, 2006; Mukherjee, 2010).

27 Studies looking specifically at the effect of cognitive load on food choice suggest

28 individuals are more likely to choose unhealthy options when under cognitive load. Shiv and

29 Fedorikhin (1999) found lessening cognitive resources made individuals more likely to select an

PT
30 unhealthy snack option, as did Ward and Mann (2000) and Zimmerman and Shimoga (2014).

RI
31 Similarly, Antonides and Cramer (2013) found that constraining cognitive resources resulted in a

32 stronger endowment effect for hedonic (often unhealthy) versus utilitarian (often healthy) foods.

SC
33 Although statistically insignificant, Deck and Jahedi (2015) also found that more unhealthy

34 snack options were selected under cognitive load. All of these studies however focused solely on

35
U
snack foods, using limited choice sets. While these studies indicate that cognitive load does
AN
36 influence food choice, no study has yet to examine this in the context of grocery shopping

behavior, using an expanded product set and taking into account dietary restrictions.
M

37

38 It is important then to identify ways to alleviate the potential effects of cognitive load on
D

39 grocery shopping behavior. We propose that one option is to simplify the shopping experience
TE

40 through the use of healthy food bundles, which consist of primarily F&V items. Work by Gordon

41 et al. (2014) has discussed healthy product bundling as a potential approach to encourage
EP

42 healthier food purchasing among SNAP recipients. Prior work has also shown that bundles may

43 be preferred over individually priced options (Harris and Blair, 2006) and may increase purchase
C

44 likelihood (Iyengar and Lepper, 2000). While Harris and Blair suggest that bundle preference is
AC

45 motivated by a desire to lessen mental strain, they did not directly manipulate cognitive load nor

46 investigate food products. If bundled choices in a grocery store were to include primarily F&V

47 items, and individuals under cognitive load are more likely to choose bundled options as a means

48 of lessening cognitive effort, then food bundles could be an effective health nudge to improve

2
ACCEPTED MANUSCRIPT

49 food choice. As grocery shopping can be cognitively stressful, exploring the effect of a potential

50 health nudge in the presence of cognitive load could also yield more accurate and meaningful

51 results. No study has yet examined preference for healthy food bundles when individual items

52 are also in the choice set, and none have directly manipulated cognitive load.

PT
53 From a mental accounting perspective, Johnson et al. (1999) suggests that a consumer will

RI
54 exhibit more positive evaluations for a bundle of items, as opposed to the same items

55 individually, in part because of the bundle’s single stated price. Similarly, Sharpe and Staelin

SC
56 (2010) found consumers tended to rate bundled goods as being of an increased value due to the

57 reduction in cognitive effort needed to mentally account for a single price versus several prices.

58
U
By bundling F&V items together, in the long run, it may be possible to change consumers’
AN
59 taste preferences, and thus increase consumer demand for F&V items. Work by Liu et al. (2015)

on ‘vice-virtue’ bundles has already examined preference between bundles with varying
M

60

61 proportions of vice (unhealthy) and virtue (healthy) foods, using a choice set comprised of only
D

62 bundles. We propose here that adding food bundles to an existing mix of grocery items might
TE

63 particularly appeal to consumers wishing to constrain their choice set, and reduce the cognitive

64 overload that comes from comparing and selecting numerous individual products.
EP

65 Retail promotion of healthy food bundles could be an effective and inexpensive display

66 strategy to implement at the store level. Bundles may also reduce consumers’ cognitive load by
C

67 simplifying the shopping experience, and ultimately promote increased purchases of F&Vs. Prior
AC

68 work in behavioral economics has already been used to explore potential health nudges designed

3
ACCEPTED MANUSCRIPT

69 to encourage the consumption of nutritious foods1. Further research is needed to understand

70 whether bundling can serve as a health nudge that helps consumers ‘overcome’ poor food

71 choices induced by cognitive load. Bundled products in the marketplace typically also offer a

72 price discount. However, such discounts may be unnecessary if there is a perceived value from

PT
73 the bundle itself.

RI
74 The objectives of this research are to determine: 1) whether cognitive load influences

75 healthy versus unhealthy grocery selection, 2) if displaying healthy food bundles influences F&V

SC
76 selection, and 3) whether such bundles need to offer a price discount. To answer these questions,

77 we conduct an artefactual field experiment with over 350 community participants in a laboratory

78
U
set up as a grocery store. In the experiment, we exogenously vary whether bundles are displayed
AN
79 (with and without a price discount) and whether shoppers are under cognitive load. We also use a

richer product set than previous studies, which included limited snack item selections. Our
M

80

81 product set more closely mimics the products people may select in a grocery store, including
D

82 three product categories: Fruits and Vegetables, Junk Food/Snacks, and Protein/Dairy/Grains.
TE

83 To the knowledge of the authors, no known work has yet to examine grocery food bundling as a

84 potential behavioral intervention to increase overall sales of F&V items.


EP

1
Previous efforts by the USDA, the U.S. DHHS, and others have included providing advice about healthful
food choices, and requiring nutritional labeling of packaged foods (Welsh et al., 1993). Evidence however is
C

inconclusive on whether such information actually improves food choice (for a few see Barreiro-Hurle et al.,
2010; Roberto et al., 2010; Thorndike et al., 2012; Kiesel and Villas-Boas, 2013). Concepts from behavioral
economics can be used to guide individual decision-making, ultimately improving diet and health (Just, 2006;
AC

Just et al., 2007; Cawley and Ruhm, 2011), and can be used to suggest in-store promotional approaches to
encourage healthier food purchases (Gordon et al., 2014). External cues including food presentation,
expectation of how the food tastes, or even the context of the meal have been examined (Wansink and
Deshpande, 1994; Wansink, 1996; Tuorila et al., 1998; Wansink, 2004). Behavioral economics has also made
inroads in understanding the link between present-biased time preferences and health behaviors (e.g., Bradford
et al., 2014; Courtemanche et al., 2015), and has explored the role that incentives, linked to healthier choices,
can play in food purchases (for an overview see Price and Riis, 2012). Solutions such as increasing or “taxing”
the price of unhealthy foods have also been proposed; however, such solutions may decrease the welfare of
lower-income populations, and changes in food prices would have to be large to have any significant effect
(Kuchler et al., 2005; Duffey et al., 2010).

4
ACCEPTED MANUSCRIPT

85

86 THEORETICAL FRAMEWORK

87 One might assume that individuals who recognize the negative health consequences of eating

PT
88 low-nutrition, high-calorie food would then improve their diet. However, individuals often

89 contradict this assumption. An explanation is that food selection is perhaps a self-control

RI
90 problem, influenced by the amount of cognitive resources available at the time of purchase. One

91 way to lessen the cognitive load of the food selection process may be to offer food bundles as

SC
92 part of the choice set, and thus improve the convenience and cognitive ease of healthy item

U
93 selection. AN
94 We use a dual-system framework first developed by Kahneman (2002, 2011) to explain

95 how cognitive load directly impacts food choice behavior. This dual system is composed of an
M

96 intuitive (impulsive) system and a thoughtful reasoning system. Prior studies have shown that

97 subjecting individuals to cognitive load increases the exertion of the reasoning system, thus
D

98 restricting the system’s ability to regulate decision making (Fudenberg and Levine, 2006;
TE

99 Mukherjee, 2010).

100 In particular, Fudenberg and Levine suggest this ‘dual-self’ influences self-regulation by
EP

101 making it easier to select impulsive choices when subjected to cognitive load. We extend
C

102 Fudenberg and Levine’s dual-self theory by predicting that under cognitive load, individuals may
AC

103 be more likely to select unhealthy foods, and more likely to select a food bundle. We argue that

104 increased bundle selection may occur if selecting a bundle requires less cognitive resources

105 compared to selecting individual items. We test whether food bundling can serve as a healthy

106 nudge designed to lessen cognitive processing, and thus increase the likelihood that healthier

107 food options are selected when cognitively strained.

5
ACCEPTED MANUSCRIPT

108

109 Hypotheses

110 Independent of bundling, cognitive load is expected to have a significantly negative effect

111 on the proportion of healthy (F&V) items selected, and a significantly positive effect on

PT
112 Junk/Snack food items, compared to selections made under no load. We also hypothesize that

RI
113 displaying healthy food bundles will result in greater selection of F&V items, compared to

114 displays without bundles. We anticipate more bundles will be selected by consumers under high

SC
115 cognitive load compared to selections under no load. This would indicate that bundle selection

116 may serve to alleviate the potential effects of cognitive load on food choice. If more bundles are

117
U
selected under high load, displaying healthy food bundles may be able to nudge cognitively
AN
118 stressed consumers towards higher levels of F&V selection.

Whether or not bundles need to be discounted in order to effectively increase F&V


M

119

120 selection is indeterminate. If no significant differences are uncovered between discounted versus
D

121 non-discounted bundle treatments, it could be that consumers perceive value just from the
TE

122 bundles’ one stated price, and that no discount is necessary. The interaction between price

123 discounted bundles and cognitive load is also indeterminate.


EP

124

125 METHODOLOGY
C

126 Experimental Design


AC

127 To explore the influence of cognitive load and product bundling on food choice, we conducted

128 an artefactual field experiment using grocery store shoppers. An artefactual field experiment is

129 similar to a standard laboratory experiment, but different in that it uses participants from the

130 ‘market of interest’ (Harrison and List, 2004; List, 2011); in our case, grocery store shoppers.

6
ACCEPTED MANUSCRIPT

131 This allows us to use a sample that is more representative of the target population. The

132 experiment was conducted at the University of Wisconsin-Madison BRITE Lab during fall 2015

133 through summer 2016.

134 A total of 367 participants were recruited for a single session each lasting approximately 60

PT
135 minutes; on average 9-16 participants were present for any individual session. Participants were

RI
136 recruited through advertisements in local newspapers, online community bulletin boards, at local

137 community centers, public libraries, and at grocery stores within a 20-minute driving radius to

SC
138 the college campus. The advertisements referred to the experiment as a ‘consumer study’ in

139 order to avoid sample selection bias; that is, we did not want only participants with a particular

140
U
interest in food signing up. Participants were prescreened, and excluded from the study if they:
AN
141 1) had not shopped for groceries within the past two months, 2) were under the age of 18, 3)

were full time undergraduate students, 4) had not resided in the U.S. for at least 5 years, and/or
M

142

143 5) had any known food allergies. The prescreen criteria was designed to enroll only adult
D

144 participants who routinely shopped for groceries in the U.S., and who would not be allergic to
TE

145 any of the study food items.

146 The laboratory was set up to resemble a small grocery store, with a display that featured 30
EP

147 different food items, each sized for retail pricing at $1. This allowed for ease of comparison

148 across foods. To ensure that this $1 price closely reflected the marketplace, average retail prices
C

149 for each item were obtained from three different stores in the surrounding area, for the duration
AC

150 of the study period. The average retail price for each of the 30 items ranged between $0.89 and

151 $1.08.

152 To better simulate a store environment and preserve product quality, the grocery display

153 featured store shelving, as well as a commercial display freezer and cooler. Another benefit to

7
ACCEPTED MANUSCRIPT

154 this physical display is that it enabled participants to observe the actual food items. Work by

155 Shiv and Fedorikhin (1999) suggests this is important as food choices may differ depending on

156 whether the item is physically present or symbolically represented, such as on a computer screen.

157 They found that individuals selected an unhealthy snack option less often when a picture was

PT
158 displayed, as opposed to the actual item. Another benefit of participants viewing the grocery

RI
159 display is that it helped further convey that the choice task was non-hypothetical.

160 Individual food items and product bundles were pretested for general appeal with a focus

SC
161 group of 22 consumers prior to conducting the experiment. Pretest participants were asked to rate

162 their preference for the various items and bundles. They were also invited to share their opinions

163
U
concerning the study items. Items and bundles that were poorly rated by the majority of the focus
AN
164 group were excluded from the study. Figure 1 displays the final 30 food items featured in the

study.
M

165

166 Upon arriving at the lab, participants were informed of the study purpose and protocol, and
D

167 provided their signed informed consent to participate. The study protocol was approved by the
TE

168 university’s Institutional Review Board for research on human subjects. Participants received $5

169 in cash for showing up for the study, plus any food items that they selected. They could also
EP

170 receive up to an additional $8, based on their study performance during the experiment.

171 Participants were first read the instructions, and then practiced each type of task featured in
C

172 the study. The practice section served to ensure that participants understood the instructions. For
AC

173 example, the practice food tasks would not allow participants to continue unless they spent their

174 entire budget. The study tasks included a food choice task, memorization tasks, and arithmetic

175 tasks. The memorization tasks were always completed in conjunction with either the food choice

176 task and/or arithmetic tasks, and served to induce cognitive load.

8
ACCEPTED MANUSCRIPT

177 After the practice tasks, participants next completed a food choice task. They were given a

178 $10 budget with which to shop the grocery display. Participants were told to use their entire

179 budget as no change would be given, and were not permitted to use any personal money.

180 Participants were instructed to first walk through the grocery display privately and view each

PT
181 product.

RI
182

U SC
AN
M
D
TE
C EP
AC

9
ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
C EP

183
184 FIGURE 1. Individual Items Featured in the Food Choice Task, by Item Category
AC

185

186 Each participant was randomly shown one of three different versions of the display. The

187 Control display consisted of the 30 different individual food items. The Bundles-No Discount

188 display consisted of Control plus six different preassembled food bundles. The bundles featured

189 primarily F&V items, all of which were also sold individually in the grocery display, and were

10
ACCEPTED MANUSCRIPT

190 priced at “5 items for $5”. The Bundles-Discounted display was a version in which bundles were

191 discounted 20%; “5 items for $4”. The food bundles used in the study can be viewed in Figure 2;

192 all six bundles consist of half or more fruits and/or vegetables, similar to USDA MyPlate

193 recommendations (Post et al., 2011).

PT
194 To simulate a more mentally straining shopping experience, half of participants were

RI
195 randomly assigned to complete the food choice task while simultaneously completing a

196 memorization task to induce cognitive load. Deck and Jahedi (2015) note that the most

SC
197 commonly used technique to simulate cognitive load involves an individual keeping a 6-or-more

198 digit number in their memory, while concurrently completing a separate decision task. Therefore,

199
U
participants were shown and asked to memorize a 7-digit number before starting the food choice
AN
200 task. They were later asked to recall the number after they had made their selections. This

resulted in a between-subjects design of six different treatments for the experiment: 3 Displays X
M

201

202 2 Cognitive Load Conditions.


D

203 Participants were randomly assigned to one of the six treatments upon arrival to the lab;
TE

204 this resulted in 58-66 participants per treatment. After privately viewing the grocery display,

205 participants next made their shopping selections privately on a computer. Participants usually
EP

206 received the food items they selected immediately at the end of their session. However, if not

207 going home right away, they could also arrange to pick up their food items at an alternate
C

208 day/time.
AC

209 After the food choice task, participants were given a five-minute break. They then

210 completed a set of eight arithmetic tasks. These arithmetic tasks were adapted from the

211 manipulation check used by Deck and Jahedi (2015), who suggest that participants successfully

212 under cognitive load should exhibit a drop in arithmetic performance. For these tasks,

11
ACCEPTED MANUSCRIPT

213 participants were asked to multiply m1 x m2, where integer m1 ~ U (13…19) and integer m2 ~ U

214 (5….9), following Deck and

PT
RI
U SC
AN
M
D
TE
C EP
AC

215

12
ACCEPTED MANUSCRIPT

216 FIGURE 2. Preassembled Product Bundles Featured in the Food Choice Task
217

218 Jahedi. The eight arithmetic tasks were evenly split into two sections, with a five-minute break

219 between sections. In one of the sections, the tasks were completed under cognitive load. Half of

PT
220 participants were randomly selected to complete the section under cognitive load first, so that we

221 could later check for order effects.

RI
222 After the arithmetic tasks, participants completed a brief post-experiment questionnaire.

SC
223 The questionnaire asked about eating habits, dietary restrictions, food preferences, shopping

224 behaviors and future shopping plans, and standard demographics. Participants next received the

U
225 food items they had selected earlier in the session. They also received additional payments based
AN
226 on study performance. To incentivize memorization of the 7-digit number, at the end of the

227 session, one of the tasks the participant completed under cognitive load was randomly selected
M

228 for payment. They received an additional $5 if they correctly recalled the 7-digit number; else
D

229 they received an additional $0. In order to incentivize arithmetic performance, at the end of the
TE

230 session one arithmetic task was also randomly selected for payment. Participants received an

231 additional $3 if they answered the math problem correctly; else they received an additional $0.
EP

232 The order of study tasks is presented in Table 1.

233
C

234 Outcome Measures and Econometric Model


AC

235 The 30 food items featured in the study were evenly selected from three categories: Fruit and

236 Vegetable Items, Junk Food/Snack Items, and Protein/Dairy/Grain Items. While our study

237 focuses specifically on increasing the selection of Fruit and Vegetable items, many items in the

238 Protein/Dairy/Grain category were generally also higher in nutritional content. However, items

239 from the Junk Food/Snack category tended to be high in calories but low in nutritional content.

13
ACCEPTED MANUSCRIPT

240 Each category consisted of the same proportion of refrigerated and frozen foods relative to more

PT
RI
U SC
AN
M
D
TE
C EP
AC

241
1
242 Memorization tasks served to induce high cognitive load
2
243 Participants randomized to 1 of 6 treatments upon arrival; treatments can be viewed in Table 1.
3
244 Arithmetic performance serves as a manipulation check for high cognitive load, following Deck and Jahedi (2015).
4
245 Order of tasks randomized within section
5
246 Half of participants randomly selected to complete Section B first; the 7-digit numbers employed were 4319162;
247 8568379; 5862413; 2856979
248

249 TABLE 1. Order of Study Tasks


14
ACCEPTED MANUSCRIPT

250

251 shelf-stable foods.

252 To determine the effect of cognitive load, independent of bundling, differences in the

253 percentage of foods selected from each category are compared for the two control-display

PT
254 treatments (T1: Control, No Load; T4: Control, High Load). To determine the effect of displayed

RI
255 food bundles and discounted bundles on F&V selection, differences in the percentage of foods

256 selected from each category are compared between treatments. For the four treatments that

SC
257 feature bundles, differences between treatments in the percentage selecting one or more bundle

258 are examined. These four treatments are: T2: Bundles-No Discount, No Load; T3: Bundles-

259
U
Discounted, No Load; T5: Bundles-No Discount, High Load; T6: Bundles-Discounted, High
AN
260 Load. Non-parametric Wilcoxon rank-sum 2-sample tests are used for the comparisons. In

addition, the effects of the various treatment components, as well as other explanatory variables,
M

261

262 on the proportion of items selected from each food category are modeled using a multivariate
D

263 fractional regression model.


TE

264 Following Papke and Wooldridge (1996), and Murteira and Ramalho (2014), we use a

265 fractional multinomial logit model fit by quasi-maximum likelihood. Our dependent variable is a
EP

266 vector of proportions such that Y ≡ ( yf, yj, yn ) ′ , the proportion of consumers’ baskets bundles

267 that are allocated to each of the three categories k = f, j, n. These three categories are exhaustive
C

268 and mutually exclusive for purposes of this study. Thus, we are interested in their joint behavior,
AC

269 estimated simultaneously as the three categories are correlated and their selection is inherently

270 bounded between zero and one. This joint behavior is explained by a set of explanatory variables

271 X ≡ (x1, x2,.…xn).

272 Murteira and Ramalho (2014) note that in estimating multivariate fractional response

15
ACCEPTED MANUSCRIPT

273 models, quasi-maximum likelihood estimation based on the Bernoulli distribution often handles

274 boundary observations well. This is particularly useful in this case, as it is plausible that one may

275 select all food items from a single category. Therefore, a final fractional multinomial logit

276 generalized from Papke and Wooldridge (1996) is estimated using Stata 14.1:

PT
277 Bki = X/ βk + ik (2)

RI
278 where Bki = {proportion of items purchased}, and 0 ≤ Bki ≥ 1, with ∑Bki=1. The food category

279 equations then are identically specified, estimated simultaneously, and the omitted category for

SC
280 estimation purposes is n; X/ = { Tik, Qik, Zik}, with i representing each individual consumer, and

U
281 ik is the error term with a zero mean across consumers. Tik consists of treatment component
AN
282 dummy variables HighCognitiveLoad, BundlesDisplayed, and BundlesDiscounted, which are 1

283 if the participant was in a treatment that featured that particular component, 0 otherwise. Also
M

284 included are interactions HighCognitiveLoad*BundlesDiscounted and


D

285 BundlesDisplayed*Female. Qik includes additional explanatory variables for consumer i, which
TE

286 are obtained from the consumer’s post-experiment questionnaire. These include

287 AlreadyPlannedToPurchase, (M=83.9%; SD=36.7%) which is 1 if the participant had already


EP

288 been planning to buy items featured in the study, 0 otherwise, and FollowingSpecialDiet,

289 (M=22.6%; SD=41.9%) which is 1 if the participant indicated following a special dietary
C

290 restriction, 0 otherwise. No significant differences within these explanatory variables were
AC

291 observed between treatments.

292 Lastly, Zik consists of demographic variables. Dummy variables Female (M=63.9%;

293 SD=47.5%), ChildrenUnder18 (M=10.5%; SD=30.5%), and NonCaucasian (M=29.4%;

294 SD=45.6%) are 1 if the participant represents the demographic, 0 otherwise. Age in years (M=31;

295 SD=12) and HouseholdIncome10K in tens of thousands of dollars (M=5.67; SD=4.84), are both

16
ACCEPTED MANUSCRIPT

296 mean centered. No significant differences within participant demographics were observed

297 between treatments. All model variables except Age and HouseholdIncome10K are dummy

298 variables where the value is 0 otherwise. Qik and Zik enable us to identify heterogeneous

299 treatment effects; of particular interest are gender differences, as females are often the primary

PT
300 household grocery shopper (FMI, 2016). Our baseline treatment (T1: Control, No Load) is

RI
301 captured by the intercept term, thus represented by setting model variables HighCognitiveLoad,

302 BundlesDisplayed, BundlesDiscounted, HighCognitiveLoad*BundlesDiscounted, and

SC
303 BundlesDisplayed*Female equal to zero.

304

305 RESULTS
U
AN
306 Treatment Comparisons
307 The average and median percentage of items selected from each food category, by treatment, is
M

308 presented in Table 2. The highest percentage of F&Vs is selected when participants are under no

309 cognitive load and view discounted bundles (T3): 63.51% of these selections contain F&V items.
D

310 In contrast, when participants view discounted bundles under high cognitive load (T6), the
TE

311 lowest percentage of F&Vs (48.19%) and the highest percentage of junk food items (23.3%) are

selected.
EP

312

313 Comparing the three different display versions, overall more junk food items are selected
C

314 when choices are made under cognitive load. This result aligns with prior studies (Shiv and
AC

315 Fedorikhin, 1999; Ward and Mann, 2000; Zimmerman and Shimoga, 2014) examining the effect

316 of cognitive load on food choice, which suggest individuals are more likely to select an

317 unhealthy option when cognitive resources are lessened.

318 Differences between treatments for all three food item categories can be viewed in Table

319 3. Shapiro-Wilk tests for normality are first performed on the percentage of items selected for

17
ACCEPTED MANUSCRIPT

320 each category. Results indicate the rejection of normality for all categories at better than the 1%

321 level; therefore, non-parametric Wilcoxon rank-sum 2-sample tests are performed.

PT
RI
U SC
AN
M
D
TE
EP

322
323 NOTE: This table presents the average and median percentage of grocery items selected by participants from each of
324 the three food item categories. as part of their choice set, by treatment. Standard deviations and interquartile ranges
C

325 in parentheses as noted.


AC

326

327 TABLE 2. Average and Median Percentage Selected, by Item Category and Treatment

328

329 A 9.58% increase (p=0.0398) in Junk Food/Snack Items is observed in the discounted

330 bundle display when participants are under high load (T6) compared to no load (T3). As

18
ACCEPTED MANUSCRIPT

331 previously mentioned, this finding is similar to prior studies that suggest more unhealthy items

332 are selected when under cognitive load. Although not statistically significant, for Junk

333 Food/Snack Items, a 3.05% increase is also observed in the Control display under high load (T4)

334 compared to no load (T1), and a similar 2.91% increase is observed in the non-discounted bundle

PT
335 display under high load (T5) compared to no load (T2).

RI
336

U SC
AN
M
D
TE
C EP
AC

337 1
338 p-values in bold are significant at the 10%*, 5%** and 1%*** level respectively

19
ACCEPTED MANUSCRIPT

339 NOTE: This table presents the average difference in the percentage of grocery items selected by participants from
340 each of the three food item categories, as part of their choice set, using pairwise comparisons between treatments. p-
341 values obtained from non-parametric Wilcoxon rank-sum 2-sample tests. Standard deviations in parentheses.
342

343 TABLE 3. Comparison Statistics for Food Item Category Percentages, by Treatment
344

PT
345 We find that under certain conditions, displaying healthy food bundles is associated with

RI
346 an increase in F&V selection, as earlier hypothesized. We first compare treatment conditions

347 where participants are not subject to high cognitive load. When non-discounted bundles are

SC
348 displayed (T2), a 5.86% increase in F&V items is observed compared to Control (T1), although

349 not statistically significant. Displaying discounted bundles (T3) compared to Control (T1) results

350
U
in a 14.87% increase (p=0.0003) in F&V selection, a 3.06% decrease (p=0.0628) in Junk
AN
351 Food/Snack Items, and an 11.79% decrease (p=0.0007) in Protein/Dairy/Grain Items.

352 Displaying discounted bundles (T3) results in 9.01% more (p=0.0584) F&V items compared to
M

353 when non-discounted bundles are displayed (T2).


D

354 We next compare treatment conditions where participants are subject to high cognitive
TE

355 load. When non-discounted bundles are displayed (T5), a 6.88% increase (p=0.0359) in F&V

356 items is observed compared to Control (T4). This increase in F&V selection appears to come at
EP

357 the expense of Protein/Dairy/Grain Items. A 6.29% decrease (p=0.0560) in Protein/Dairy/Grain

358 items is observed when non-discounted bundles are displayed (T5) compared to Control (T4).
C

359 No significant differences in F&V selection are uncovered when participants discounted bundles
AC

360 are displayed (T6) compared to Control (T4). Interestingly, displaying discounted bundles (T6)

361 results in 8.17% less (p=0.0118) F&V items compared to when non-discounted bundles (T5) are

362 displayed.

363 We compare next treatments between high and no cognitive load conditions. When non-

364 discounted bundles are displayed, subjecting participants to high cognitive load (T5) does not

20
ACCEPTED MANUSCRIPT

365 have a significant effect on F&V selection, compared to selections made in the absence of load

366 (T2). However, we do observe an effect of cognitive load for selections made when discounted

367 bundles are displayed. Under high load (T6), 15.32% less (p=0.0002) F&V items are selected

368 compared to no load (T3). Junk/snack food items account for 9.58% of this shift (p=0.0398).

PT
369 Although only observed when discounted bundles are displayed, this shift towards unhealthier

RI
370 choices while subjected to cognitive load is in line with our earlier hypotheses.

371 When faced with a price discount for bundles, we wonder if subjecting participants to

SC
372 cognitive load could compound the overall cognitive effort of selection. If so, this could lead to

373 overall less healthy choices. This could also possibly negate any cognitive ease associated with

374
U
bundle selection (which we discuss further below). It may be that the amount of cognitive
AN
375 resources available at the time of selection influences whether bundles need offer a price

discount in order to effectively increase F&V selection, although more research in needed in
M

376

377 actual store environments.


D

378 For the displays that feature bundles, the percentage of participants selecting bundles, by
TE

379 treatment, is reported in Table 4.

% Selecting Any
EP

Treatment Bundles
(std dev)
1 Bundle 2 Bundles
C

T2 45.00
36.67 8.33
AC

(Bundles-No Discount, No Load) (50.17)

T3 81.03
25.86 55.17
(Bundles-Discounted, No Load) (39.55)

21
ACCEPTED MANUSCRIPT

T5 66.67
37.88 28.79
(Bundles-No Discount, High Load) (47.50)

T6 56.06
19.70 36.36

PT
(Bundles-Discounted, High Load) (50.01)

380

RI
381 NOTE: This table presents the percentage of participants selecting 1, 2, and any bundles as part of their choice set,
382 by treatment. T1 and T4 are omitted as the treatments did not contain product bundles. Standard deviations in
383 parentheses.

SC
384 TABLE 4. Percentage of Participants Selecting Bundles, by Treatment

385 In the absence of cognitive load, 81.03% of participants select at least one bundle when

386
U
discounted bundles are displayed (T3). The second highest percentage of participants (66.67%)
AN
387 select bundles when under high cognitive load and non-discounted bundles are displayed (T5).
M

388 When non-discounted bundles are displayed, over 20% more participants select multiple bundles

389 when subjected to high cognitive load (T5 compared to T2). This suggests that bundle selection
D

390 may lessen mental strain, as Harris and Blair (2006) have earlier suggested, so long as selecting
TE

391 the bundle itself does not cause additional strain. When discounted bundles are displayed,

392 although more participants selected bundles under no load (T3) compared to high load (T6), in
EP

393 both treatments a greater number of participants select two bundles rather than one, suggesting

394 preference for the bundles’ associated price savings.


C

395 Differences in bundle selection between treatments can be viewed in Table 5. We


AC

396 examine whether subjecting participants to cognitive load influences bundle selection. When

397 non-discounted bundles are displayed, 21.67% more participants (p=0.0147) select bundles

398 when subjected to high cognitive load. However, the opposite is observed when discounted

22
ACCEPTED MANUSCRIPT

399 bundles are displayed: 24.97% less participants (p=0.0031) select a bundle when under high load

400 (T6) compared to no load (T3).

401 It is important to note that when not subjected to cognitive load, 36.03% more

402 participants (p=0.0001) select at least 1 bundle when bundles are discounted (T3) compared to

PT
403 non-discounted (T2). Those looking to use bundles as a nudge to increase retail F&V selection

RI
404 should consider exploring the effect that cognitive resources have on bundle selections made in

405 actual store environments.

U SC
AN
M
D
TE
C EP

406
1
407 p-values in bold are significant at the 10%*, 5%** and 1%*** level respectively
AC

408 NOTE: This table presents the average difference in the percentage of participants who selecting 1 or more bundle
409 as part of their choice set, using pairwise comparisons between treatments. T1 and T4 are omitted as the treatments
410 did not contain product bundles. Standard deviations in parentheses.
411
412
413 TABLE 5. Comparison Statistics for Percentage of Participants
414 Selecting 1 or More Bundle, by Treatment

415

23
ACCEPTED MANUSCRIPT

416 Prior studies such as Park et al. (1989) and Aylott and Mitchell (1998) have noted the

417 mental strains associated with grocery shopping; our results obtained under high cognitive load

418 may give a more accurate picture of how bundles could be successfully implemented at the retail

419 level. Although additional research using actual grocery store shopping is needed, our results

PT
420 suggest that when shopping under cognitive load, food bundles may not need to be discounted to

RI
421 effectively increase F&V selection.

422

SC
423 Marginal Effects
424 Marginal effects are computed from the estimated coefficients of the fractional multinomial

425
U
logit, and presented in Table 6. The results of the estimated fractional multinomial logit model
AN
426 can be viewed in Appendix A.
M
D
TE
C EP
AC

427

24
ACCEPTED MANUSCRIPT

428 NOTE: values in bold are significant at the 10%*, 5%** and 1%*** level respectively
429

430 TABLE 6. Marginal Effects for Determinants of Grocery Selection, by Item Category
431

432 For F&V items, displaying non-discounted product bundles is estimated to increase

PT
433 selection by 4.37%, compared to F&V selection when bundles are not displayed, regardless of

RI
434 cognitive load. However, discounted product bundles are influenced by cognitive load. In the

435 absence of load, displaying discounted bundles is estimated to have an additional 10.16%

SC
436 increase on proportion of F&V items selected. When selections are made under high cognitive

437 load though, displaying discounted product bundles results in a 16.33% decrease in the

438 proportion of F&V items selected.


U
AN
439 For Junk Food/Snack items, we find a possible gender effect of displaying product

440 bundles on item selection. Although displaying bundles increases Junk Food/Snack item
M

441 selection by 7.38%, it actually decreases Junk Food/Snack selection by 9.01% for females.
D

442 Interestingly, only one of the bundles features a Junk Food/Snack item; this particular bundle is
TE

443 ranked third out of the six displayed bundles in terms of selection preference ranking. Those who

444 had already planned to buy items featured in the study, prior to attending the study, select 8.42%
EP

445 less Junk Food/Snack items. Likewise, participants who indicate following a special diet select

446 10.33% less Junk Food/Snack items. For every year older in age beyond the mean age of 29,
C

447 participants select 0.73% less Junk Food/Snack items. Lastly, for every $10,000 increase in
AC

448 household income beyond the mean income of $56,644, participants select on average 0.58%

449 less Junk Food/Snack items.

450

451 Manipulation Check for Cognitive Load

25
ACCEPTED MANUSCRIPT

452 For participants who complete the food choice task under high cognitive load (T4, T5, or T6),

453 recall accuracy for the displayed 7-digit number is assessed. Across treatments, recall accuracy is

454 over 90%, with no significant differences in accuracy observed between treatments. Following

455 Deck and Jahedi (2015), we also assess participants’ arithmetic performance under both high and

PT
456 no cognitive load. If high load is successfully manipulated by recalling a 7-digit number, then

RI
457 participant performance should be significantly worse under high load. On average, participants

458 are 9.79% less accurate (p=<0.001) when performing arithmetic under high cognitive load,

SC
459 compared using non-parametric Wilcoxon signed-rank tests for matched pairs. No significant

460 order effects are observed between the high cognitive load and no cognitive load arithmetic

461 sections.
U
AN
462

CONCLUSIONS
M

463

464 Our findings align with prior studies (Shiv and Fedorikhin, 1999; Ward and Mann, 2000;
D

465 Zimmerman and Shimoga, 2014) that suggest unhealthy options are more likely to be selected
TE

466 when cognitive resources are constrained. In all three versions of the grocery display, higher

467 percentages of junk food items were selected when under cognitive load, although only one of
EP

468 the comparisons is statistically significant. We also observe an interesting significant interaction

469 between cognitive load and displaying discounted bundles. We find the largest percentage of
C

470 F&V items (63.5%) are selected when bundles are discounted, and participants are not subjected
AC

471 to cognitive load. In comparison, when bundles are discounted and participants are subjected to

472 cognitive load, the lowest percentage of F&V items (48.2%) are selected. In the absence of

473 cognitive load, displaying discounted bundles led to healthier choices. Yet under cognitive load,

474 displaying non-discounted bundles results in healthier selections.

26
ACCEPTED MANUSCRIPT

475 As prior research suggests that grocery shoppers often shop under cognitive strain (Park

476 et al., 1989) and under time pressures (Aylott and Mitchell, 1998), it is interesting that we find a

477 price intervention offered through discounted product bundles fails to achieve the desired effect

478 of increased F&V selection for high load consumers. To appeal to mentally strained shoppers,

PT
479 whether or not retailers should display discounted or non-discounted healthy food bundles to

RI
480 increase F&V selection warrants additional research. Our results here indicate that displaying

481 healthy food bundles can lead to healthier selections overall under certain shopping conditions.

SC
482 Overall our results seem to suggest that healthy food bundles may not need to offer a

483 price discount to effectively increase F&V selection. This is perhaps helpful information for

484
U
retailers, who could implement product bundles into existing store displays at little to no cost.
AN
485 We further find evidence that suggests product bundles may be preferred by consumers as a way

to lessen their cognitive strain while grocery shopping. When non-discounted bundles are
M

486

487 displayed, 21.7% more participants select bundles when under high load compared to no load.
D

488 Another interesting result is that while product bundling seems to have a relatively stable
TE

489 effect on F&V selection across cognitive load conditions, price discounts for bundles in this

490 study do not. Comparing no load and high load consumers, when discounted bundles are
EP

491 displayed, high load consumers selected 15.3% less F&V items and 9.6% more junk food items.

492 It may be that the added effect of a price discount, when already operating with limited cognitive
C

493 assets, further depletes one’s cognitive resources, although additional research is needed to test
AC

494 the robustness of this finding in the field. Our findings may be particularly relevant for those

495 shopping on a fixed grocery budget, such as the budget we imposed on our study participants.

496 It is reasonable to assume that lower income consumers may already be under cognitive

497 strain before even reaching the grocery store. While one might initially assume that discounted

27
ACCEPTED MANUSCRIPT

498 food bundles would appeal to such consumers, our results suggest that this might not be the case.

499 This is particularly relevant as prior research reported by the USDA’s Economic Research

500 Service has indicated that low-income households eat on average less F&Vs compared to higher

501 income households (Blisard et al., 2004). Our study was conducted using a participant pool from

PT
502 a Midwestern U.S. city and surrounding (who were not low income). Therefore, it would be

RI
503 interesting to extend this work to understand how these interventions affect low-income

504 shoppers. Healthy food bundles may be an effective nudge to encourage F&V selection among

SC
505 for those who may also be at a higher risk for food related health problems. It is important to

506 note that we do not seek to improve overall health from this study alone. Rather, our goal is to

507
U
provide guidelines about the kinds of health nudges that may improve food selection. We find
AN
508 the use of food bundling as a behavioral nudge may ultimately improve health by facilitating the

selection of healthier foods.


M

509

510 Our research has insights for both academia and for policy and practice. For academics,
D

511 we provide evidence of the role of cognitive load, and its interaction with health nudges, on
TE

512 choice. For policy and practice, our research suggests that product bundles are one way to

513 increase F&V selection by consumers. Cognitively stressed consumers may not necessarily
EP

514 require a price discount in order to purchase bundles that improve the healthfulness of their

515 choices. This study also provides interesting implications for better understanding potential
C

516 marketing techniques designed to increase F&V selection and sales. Such an increase in F&V
AC

517 sales could help combat high levels of obesity and diets of poor nutritional quality. An additional

518 benefit includes the potential for increased profitability among grocery retailers, who are often

519 faced with high perishability and low profit margins for produce items, compared to other food

520 products.

28
ACCEPTED MANUSCRIPT

521 It warrants mentioning that our study is not without limitations. Our design required

522 participants to exhaust their shopping budget during the food choice task, and all of the food

523 items were equally priced. We recognize that in an actual grocery store, consumers often face a

524 variety of prices and can choose whether or not to exhaust their budget. Participants under

PT
525 cognitive load in our study may also have been motivated to complete the choice task quickly,

RI
526 given the incentive for recalling the 7-digit number. It may be that when under cognitive load,

527 selecting non-discounted bundles allowed participants to quickly complete the task, but that

SC
528 discounted bundles instead served as an impediment to task completion. Future research is

529 needed to test the robustness of our findings in an actual grocery store shopping scenario,

530
U
particularly when non-discounted bundles are displayed and decisions are made under cognitive
AN
531 load. Potential study designs in the future could also examine selections made when price

discounted bundles are displayed in grocery stores, and could even test a range of possible
M

532

533 discounts. Another idea for future research would be to look at preference for healthy item price
D

534 discounts separate from bundling, when shopping under cognitive load in an actual grocery store.
TE

535 In conclusion, our results suggest that food bundles may not need to offer a price discount

536 in order to effectively increase F&V selection. Busy consumers may perceive greater value from
EP

537 the effort-saving convenience and cognitive ease associated with bundle selection. Product

538 bundles may also be preferred by consumers as a means through which to lessen the cognitive
C

539 strain of the shopping process. It is important to note that the bundles included in this study were
AC

540 preassembled for the consumer; again, future work is needed to explore the practical application

541 of food bundles in an actual grocery store. Our results indicate that food bundles (consisting

542 primarily of F&V items) could potentially increase retail F&V sales, provided such bundles are

543 preassembled for the shopper.

29
ACCEPTED MANUSCRIPT

REFERENCES
Antonides, G., & Cramer, L. (2013). Impact of limited cognitive capacity and feelings of guilt
and excuse on the endowment effects for hedonic and utilitarian types of foods. Appetite, 68, 51-
55.

Aylott, R., & Mitchell, V. W. (1998). An exploratory study of grocery shopping

PT
stressors. International Journal of Retail & Distribution Management, 26(9), 362 - 373.

Barreiro-Hurlé, J., Gracia, A., & de-Magistris, T. (2010). Does nutrition information on food
products lead to healthier food choices?. Food Policy, 35(3), 221-229.

RI
Benjamin, D. J., Brown, S. A., & Shapiro, J. M. (2013). Who is ‘behavioral’? Cognitive ability
and anomalous preferences. Journal of the European Economic Association, 11(6), 1231-1255.

SC
Blisard, W. N., Stewart, H., & Jolliffe, D. (2004). Low-income households' expenditures on fruits
and vegetables. Washington, DC: US Department of Agriculture, Economic Research Service.

U
Bradford, W. David, Charles Courtemanche, Garth Heutel, Patrick McAlvanah, and Chris Ruhm,
AN
2014. “Elicited Time Preferences and Consumer Behavior: Evidence from Health, Energy, and
Financial Decisions.” NBER working paper:
http://www.nber.org/authors_papers/charles_courtemanche.
M

Cawley, J., & Ruhm, C. J. (2011). The Economics of Risky Health Behaviors. NBER Working
Paper, (w17081).
D

Centers for Disease Control and Prevention (CDC). (2013). “State Indicator Report on Fruits and
Vegetables, 2013.” Atlanta, GA: Centers for Disease Control and Prevention, U.S. Department
TE

of Health and Human Services. Retrieved from: http://www.cdc.gov/nutrition/downloads/state-


indicator-report-fruits-vegetables-2013.pdf

Courtemanche, C., Heutel, G., & McAlvanah, P. (2015). Impatience, incentives and obesity. The
EP

Economic Journal, 125(582), 1-31.

Deck, C., & Jahedi, S. (2015). The effect of cognitive load on economic decision making: A
C

survey and new experiments. European Economic Review, 78, 97-119.


AC

Duffey, K. J., Gordon-Larsen, P., Shikany, J. M., Guilkey, D., Jacobs, D. R., & Popkin, B. M.
(2010). Food price and diet and health outcomes: 20 years of the CARDIA Study. Archives of
internal medicine, 170(5), 420-426.

Flegal KM, Carroll MD, Kit BK, Ogden CL. Prevalence of obesity and trends in the distribution
of body mass index among US adults, 1999–2010. Journal of the American Medical Association.
2012; 307(5):491–97.

Food Marketing Institute (FMI). (2015). “Supermarket Facts- Industry Overview 2015”.
Available online at: http://www.fmi.org/research-resources/supermarket-facts. (Accessed Feb.

30
ACCEPTED MANUSCRIPT

2017).

Food Marketing Institute (FMI). (2016). “U.S. Grocery Shopper Trends 2016”. Food Marketing
Institute Report: Arlington, VA.

Fudenberg, D., & Levine, D. K. (2006). A dual-self model of impulse control. The American
Economic Review, 1449-1476.

PT
Gordon, E., Dawkins-Lyn, N., Hogan-Yarbro, R., Karpyn, A., Shore, K., Weiss, S., & Cash, S.
(2014). “Approaches for Promoting Health Food Purchases by SNAP Participants”. Prepared by

RI
ICF International for the U.S. Department of Agriculture, Food and Nutrition Service, Office of
Policy Support; July 2014.

SC
Greene, J. D., Morelli, S. A., Lowenberg, K., Nystrom, L. E., & Cohen, J. D. (2008). Cognitive
load selectively interferes with utilitarian moral judgment. Cognition, 107(3), 1144-1154.

Harris, J., & Blair, E. A. (2006). Consumer preference for product bundles: The role of reduced

U
search costs. Journal of the Academy of Marketing Science,34(4), 506-513.
AN
Harrison, G. W., & List, J. A. (2004). Field experiments. Journal of Economic Literature, 1009-
1055.
M

He, F. J., Nowson, C. A., Lucas, M., & MacGregor, G. A. (2007). Increased consumption of fruit
and vegetables is related to a reduced risk of coronary heart disease: meta-analysis of cohort
studies. Journal of human hypertension, 21(9), 717-728.
D

He, K., Hu, F. B., Colditz, G. A., Manson, J. E., Willett, W. C., & Liu, S. (2004). Changes in
intake of fruits and vegetables in relation to risk of obesity and weight gain among middle-aged
TE

women. International journal of obesity, 28(12), 1569-1574. doi:10.1038/sj.ijo.0802795

Hinson, J. M., Jameson, T. L., & Whitney, P. (2003). Impulsive decision making and working
EP

memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 29(2), 298.

Hung, H. C., Joshipura, K. J., Jiang, R., Hu, F. B., Hunter, D., Smith-Warner, S. A., Colditz,
G.A, Rosner, B., Spiegelman, D., & Willett, W. C. (2004). Fruit and vegetable intake and risk of
C

major chronic disease. Journal of the National Cancer Institute, 96(21), 1577-1584.
AC

Iyengar, S. S., & Lepper, M. R. (2000). When choice is demotivating: Can one desire too much
of a good thing?. Journal of personality and social psychology,79(6), 995.

Johnson, M. D., Herrmann, A., & Bauer, H. H. (1999). The effects of price bundling on
consumer evaluations of product offerings. International Journal of Research in
Marketing, 16(2), 129-142.

Just, D. R. (2006). Behavioral economics, food assistance, and obesity. Agricultural and
resource economics review, 35(2), 209.

31
ACCEPTED MANUSCRIPT

Just, D., Mancino, L., & Wansink, B. (2007). Could Behavioral Economics Help Improve Diet
Quality for Nutrition Assistance Program Participants?. USDA-ERS Economic Research Report,
(43).

Kahneman, D. (2002). Maps of bounded rationality: A perspective on intuitive judgment and


choice. Nobel prize lecture, 8, 351-401.

PT
Kahneman, D. (2011). Thinking, fast and slow. Macmillan.

RI
Kiesel, K., & Villas-Boas, S. B. (2013). Can information costs affect consumer choice?
Nutritional labels in a supermarket experiment. International Journal of Industrial
Organization, 31(2), 153-163.

SC
Kuchler, F., Abebayehu, T., & M. Harris. (2005) “Taxing Snack Foods: Manipulation Diet
Quality or Financing Information Programs.” Review of Agricultural Economics, Vol. 27, pp. 4-
20.

U
List, J. A. (2011). Why economists should conduct field experiments and 14 tips for pulling one
AN
off. Journal of Economic Perspectives, 25(3), 3-16.
List, J. A., Samek, A.S., & Zhu, T. (2015). Incentives to Eat Healthy: Evidence from a Grocery
Store Field Experiment. CESR-Schaeffer Working Paper, (2015-025).
M

Liu, P. J., Haws, K. L., Lamberton, C., Campbell, T. H., & Fitzsimons, G. J. (2015). Vice-virtue
bundles. Management Science, 61(1), 204-228.
D

Mukherjee, K. (2010). A dual system model of preferences under risk. Psychological


TE

review, 117(1), 243-255.

Murteira, J. M., & Ramalho, J. J. (2014). Regression analysis of multivariate fractional


data. Econometric Reviews, 1-38.
EP

National Institute of Health (2012) “How are Overweight and Obesity Treated?”. Available
online at: http://www.nhlbi.nih.gov/health/health-topics/topics/obe/treatment.html. (July 13,
C

2012).
AC

Papke, L. E., & Wooldridge, J. M. (1996). Econometric methods for fractional response
variables with an application to 401 (k) plan participation rates. Journal of Applied
Econometrics, 11(6), 619-632.

Park, C. W., Iyer, E. S., & Smith, D. C. (1989). The effects of situational factors on in-store
grocery shopping behavior: the role of store environment and time available for
shopping. Journal of Consumer Research, 422-433.

Post, R. C., Haven, J., & Maniscalco, S. (2011). Setting the table with a healthy plate: Make half
your plate fruits and vegetables. Journal of the American Dietetic Association, 111(11), 1644-

32
ACCEPTED MANUSCRIPT

1647.

Price, J., & Riis, J. (2012). Behavioral economics and the psychology of fruit and vegetable
consumption. Journal of Food Studies, 1(1), 1-13.

Roberto, C. A., Larsen, P. D., Agnew, H., Baik, J., & Brownell, K. D. (2010). Evaluating the
impact of menu labeling on food choices and intake. American Journal of Public Health, 100(2),

PT
312.

Sharpe, K. M., & Staelin, R. (2010). Consumption Effects of Bundling: Consumer Perceptions,

RI
Firm Actions, and Public Policy Implications. Journal of Public Policy & Marketing, 29(2), 170-
188.

SC
Shiv, B., & Fedorikhin, A. (1999). Heart and mind in conflict: The interplay of affect and
cognition in consumer decision making. Journal of Consumer Research, 26(3), 278-292.

Thorndike, A. N., Sonnenberg, L., Riis, J., Barraclough, S., & Levy, D. E. (2012). A 2-phase

U
labeling and choice architecture intervention to improve healthy food and beverage
choices. American journal of public health, 102(3), 527-533.
AN
Tuorila, H.M., et al. (1998) “Effect of Expectations and the Definition of Product Category on
the Acceptance of Unfamiliar Foods.” Food Quality and Preference, Vol. 9, pp. 421-430.
M

Wansink, B. (1996) “Can Package Size Accelerate Consumption Volume?” Journal of


Marketing, Vol. 60, pp. 1-14.
D

Wansink, B. (2004) “Environmental Factors That Increase the Food Intake and Consumption
Volume of Unknowing Consumers,” Annual Review of Nutrition, Vol. 24, pp. 455-479.
TE

Wansink, B., and R. Deshpande. (1994) “‘Out of Sight Out of Mind’: The Impact of Household
Stockpiling on Usage Rates.” Marketing Letters, Vol. 5, pp. 91-100.
EP

Ward, A., & Mann, T. (2000). Don't mind if I do: disinhibited eating under cognitive
load. Journal of personality and social psychology, 78(4), 753.
C

Welsh, D., C. Davis, and A. Shaw. (1993) USDA’s Food Guide: Background and Development.
Miscellaneous Publication No. 1514. U.S. Department of Agriculture, Human Nutrition
AC

Information Service.

Zimmerman, F. J., & Shimoga, S. V. (2014). The effects of food advertising and cognitive load
on food choices. BMC public health, 14(1), 1.

33
ACCEPTED MANUSCRIPT

APPENDIX A. Fractional Multinomial Logit Estimates for Determinants of Grocery


Selection, by Item Category

PT
RI
U SC
AN
M
D
TE

1
p-values in bold are significant at the 10%*, 5%** and 1%*** level respectively
NOTE: The share of Protein/Dairy/Grain Items is the excluded category
C EP
AC

A1
ACCEPTED MANUSCRIPT

APPENDICES B – NOT FOR PUBLICATION

TABLE B1. Treatments, Food Choice Task

Cognitive Load

PT
Treatment Grocery Display # of Participants
Condition

Individual Items Only

RI
T1 No Load 59
(Control)

T1 plus “5 for $5” Bundles

SC
T2 No Load 60
(Bundles-No Discount)

T1 plus “5 for $4” Bundles


T3 No Load 58

U
(Bundles-Discounted)
AN
Individual Items Only
T4 High Load 58
(Control)
M

T4 plus “5 for $5” Bundles


T5 High Load 66
(Bundles-NoDiscount)
D

T4 plus “5 for $4” Bundles


T6 High Load 66
TE

(Bundles-Discounted)
Total: 367
C EP
AC

A2
ACCEPTED MANUSCRIPT

TABLE B2. Food Tasks: Recall Accuracy for High Cognitive Load Manipulation, by Treatment
Memorization
Accuracy1 7-Digit
Treatment
% Number
(std dev)

T4 93.10

PT
5186348
(Control, High Load) (25.56)

T5
93.94

RI
(Bundles-No Discount, High Load) 6217457
(24.04)

T6

SC
90.91
(Bundles-Discounted, High Load) 7491248
(28.97)
1
no significant differences observed between treatments using non-parametric Wilcoxon rank-sum 2-

U
sample tests
AN
M
D
TE
C EP
AC

A3
ACCEPTED MANUSCRIPT

TABLE B3. Manipulation Check for High Cognitive Load: Arithmetic Performance Tasks

% Accurate
(std dev)
Treatment1
No Load High Load Signed-rank

PT
Difference3
Section Section2 p-value4

94.46 83.02 11.44


T1 (Control, No Load)

RI
<0.001***
(23.98) (26.34) (20.33)

SC
89.37 81.52 7.85
T2 (Bundles-No Discount, No Load) 0.0038***
(23.17) (25.08) (16.53)

U
92.63 85.28 7.35
T3 (Bundles-Discounted, No Load) 0.0267**
(22.92) (25.43) (20.10)
AN
95.26 83.62 11.64
T4 (Control, High Load) <0.001***
(15.14) (19.62) (14.97)
M

94.17 86.45 7.72


T5 (Bundles-No Discount, High Load) 0.0178**
(21.46) (19.49) (19.16)
D

94.33 81.56 12.77


TE

T6 (Bundles-Discounted, High Load) <0.001***


(14.54) (21.37) (17.05)

93.37 83.58 9.79


EP

Aggregate <0.001***
(21.38) (19.88) (20.49)
1
Within each treatment, half of participants completed the High Load Section prior to the No Load Section. Non-
parametric Wilcoxon rank-sum 2-sample tests were used to test for order effects in accuracy both within and
C

between treatments. No significant order effects were observed. For all participants, each Section consisted of 4
arithmetic problems, for a total of 8 arithmetic tasks.
AC

2
No significant differences in high load accuracy observed between high load treatments, and between high versus
no load treatments, using non-parametric Wilcoxon rank-sum 2-sample tests
3
No significant differences in accuracy difference observed between high load treatments, and between high versus
no load treatments, using non-parametric Wilcoxon rank-sum 2-sample tests
4
p-values obtained from non-parametric Wilcoxon signed-rank tests for matched pairs

NOTE: values in bold significant at the *10%, **5% and ***1% level respectively
NOTE: in the high load arithmetic condition, the 7-digit numbers employed were 4319162; 8568379; 5862413;
2856979

A4
ACCEPTED MANUSCRIPT

FIGURE B1. Layout of Grocery Store Display

PT
RI
U SC
AN
M
D
TE
C EP
AC

A5
ACCEPTED MANUSCRIPT

FIGURE B2. Prescreen Questionnaire for Potential Study Participants


Thank you for your interest in the Consumer Study. Please answer each of the questions below
so that we can determine if you are eligible to participate in this study.

1) In the past two months, how often have you shopped for grocery items?
1-2 times

PT
3 or more times
I have not shopped for grocery times in the past two months

2) Are you 18 years of age or older?

RI
YES
NO

SC
3) Are you currently a full time undergraduate student (enrolled in 12 or more credits)?
YES
NO

4) How long have you been a U.S. resident?


U
AN
Less than 5 years
5 years or more

5) Do you have any known food allergies?


M

YES (please list):____________________


NO
D
TE
C EP
AC

A6
ACCEPTED MANUSCRIPT

FIGURE B3. Instructions Script and Practice Tasks

Welcome to the Study.


You will receive a $5 show-up payment today. You will also receive $10 worth of food items that
you will select. You can earn additional money based on your responses during today's tasks.
Therefore, it is important that you read the directions carefully.

PT
Today's session will last approximately 60 minutes.
If you have questions, please raise your hand and wait for a researcher to come to you. Please

RI
do not talk or communicate with other participants, and no cell phone use.
You will be paid in cash immediately following the end of today's session. You will also receive

SC
your $10 worth of selected food items at the end of today's session.
Click to the next page

U
Some items need to be refrigerated - note that you can choose to pick up your food items at a
AN
later time today or tomorrow if you are not able to take them with you; please notify the
researchers of this when you receive your payment.
Click to the next page
M

You will see 3 different types of tasks in today's study:


D

1. A Food choice task,


TE

2. Memorization tasks, and


3. Math tasks.
EP

Click to the next page


C

1) Food choice task: The first task today will be a food choice task. For the food choice task, you
will be given a set budget of $10 at the start of the task. You should use your entire $10 budget
AC

by selecting food items from those displayed. No change will be given if you do not use the entire
$10. You will receive the food items that you select. Select food items by entering the quantity of
each that you would like to purchase with your $10 budget.
Next practice some food choice tasks.
Click to the next page

A7
ACCEPTED MANUSCRIPT

THIS IS A PRACTICE TASK

PT
RI
U SC
AN
M
D
TE
EP
C
AC

Click to the next page

A8
ACCEPTED MANUSCRIPT

THIS IS A PRACTICE TASK

PT
RI
U SC
AN
M
D
TE
EP
C
AC

A9
ACCEPTED MANUSCRIPT

Click to the next page

2) Memorization tasks: Some tasks today also include a memorization task. You will be asked to
memorize and then recall a 7-digit number. You will be shown the number for 10 seconds before
asked to complete a separate task. You will have 10 seconds to re-type the number immediately

PT
after the task.
Next practice a memorization task. For now, there is no separate task, just the memorization
task.

RI
Click to the next page.

SC
THIS IS A PRACTICE TASK
Wait here

U
Remember the number: 3485192
AN
If you remember the number correctly you will receive $5; else you will receive $0.
M

Next page
D

Enter the number you were asked to memorize: ___________________


TE

Click to the next page.


EP

At the end of the session, 1 memorization task will be randomly selected. The randomly selected
task is the one that will 'count' towards today's payment. You will receive an additional $5 if you
C

recalled the number correctly in the selected task. You receive an additional $0 if you did not
memorize the number correctly.
AC

Click to the next page

3) Math tasks: The last set of tasks will be math tasks. You will complete 8 math problems. You
will have 30 seconds to complete each. At the end of the session, 1 problem will be randomly
selected to count towards today's payment. You will receive $3 if you correctly answered the
selected problem, else you will receive $0 for this task.
Next practice some math tasks.

A10
ACCEPTED MANUSCRIPT

Click to the next page

THIS IS A PRACTICE TASK


If you answer the math problem correctly you will receive $3;

PT
Otherwise you will receive $0
47
+4

RI
Next page

SC
THIS IS A PRACTICE TASK

U
If you answer the math problem correctly you will receive $3;
AN
Otherwise you will receive $0
13
x6
M
D

Next page
TE

You have now completed the Practice Tasks.


EP
C
AC

A11
ACCEPTED MANUSCRIPT

FIGURE B4. Post-experiment Questionnaire Items

1) What is your gender?


Male
Female

PT
Other

2) How many children less than 18 years of age live in your household?

RI
# of children (<18 yrs of age):____________

3) In what year were you born? (Enter only last 2 digits of year):

SC
I was born in....................... 19_____

U
4) Before taxes, what was your total household income level this past year?
Less than $19,000
AN
$19,001 - $36,000
$36,001 - $59,500
M

$59,501 - $96,000
$96,001 or more
D

5) Were any of the food items you selected today items that you were ALREADY planning on
TE

buying soon?
YES
NO
EP

6) I currently follow a.............(select one or more):


vegetarian diet
C

vegan diet
AC

raw foods diet


Paleo diet
low-sodium diet
diabetic (reduced sugar) diet
gluten-free diet
low-carb diet
none of the above

A12
ACCEPTED MANUSCRIPT

7) What is the highest level of education you have completed?


Less than High School Graduate
High School Graduate/GED
Some college, no degree
Associate's degree

PT
Bachelor's degree
Graduate or Professional degree

RI
8) What is your race? (Select 1 or more):

SC
White
Black or African American

U
Asian American
American Indian or Alaska Native
AN
Native Hawaiian or other Pacific Islander
Other (Specify):
M

9) Are you Chicano/a, Hispanic and/or Latino/a?


D

Yes
TE

No
C EP
AC

A13
ACCEPTED MANUSCRIPT

FIGURE 1. Individual Items Featured in the Food Choice Task, by Item Category

Category Individual Food Items

PT
RI
SC
1 Bag Baby Carrots
2 Gala Apples 3 Bananas 2 Navel Oranges 2 Cucumbers (16 oz)

U
Fruit &
Vegetable Items

AN
M
1 Bag Frozen 1 Bag Frozen
Sweet Corn Broccoli Cuts 1 Can Green Beans 1 Can Tomatoes 1 Can Bartlett Pears

D
(12 oz) (12 oz) (14 oz) (14.5 oz) (15 oz)

TE
EP
1 Box Frozen
C

2 Bags Cheez-Its 2 Bags Potato Chips 2 Bags Doritos 6 Oreo Cookies French Fries
AC

(1 oz/bag) (1 oz/bag) (1 oz/bag) (2 oz pack) (4.75 oz)

Junk Food/
Snack Items
ACCEPTED MANUSCRIPT

1 Bag Peanut M&Ms 1 Rice Krispie Treat 1 Bag Skittles 1 Twix 1 Snickers
(1.74 oz) (1.3 oz) (2.17 oz) Ice Cream Bar Ice Cream Bar
(1.93 oz) (1.93 oz)

PT
RI
U SC
1 Container Cheerios 6 English Muffins 1 Bottle 2% Milk 1 Strawberry Yogurt 4 Oatmeal Packets

AN
(1.3 oz) (12 oz bag) (8 oz) (6 oz container) (0.98 oz/packet)

Protein/Dairy/

M
Grain Items

D
TE
1 Box Elbow 1 Can Chunk 1 Can Cream of 1 Can Kidney
Macaroni 1 Box White Rice Light Tuna Mushroom Soup Beans
EP
(16 oz) (16 oz) (5 oz) (10.5 oz) (15.5 oz)
C
AC
ACCEPTED MANUSCRIPT

FIGURE 2. Preassembled Product Bundles Featured in the Food Choice Task

Bundle Bundled Items

PT
RI
Bundle A
(5 F&V Items)

SC
1 Bag Frozen
2 Gala Apples 3 Bananas Sweet Corn 1 Can Green Beans 1 Can Tomatoes
(12 oz) (14 oz) (14.5 oz)

U
AN
M
Bundle B
(5 F&V Items)

D
1 Bag Frozen
2 Navel Oranges 1 Bag Baby Carrots Broccoli Cuts 1 Can Tomatoes 1 Can Bartlett Pears

TE
EP (16 oz) (12 oz) (14.5 oz) (15 oz)

Bundle C
C

(4 F&V Items,
AC

1 Protein/
Dairy/Grain 1 Can Kidney
Item) 3 Bananas 2 Cucumbers 1 Bag Baby Carrots 1 Can Tomatoes Beans
(16 oz) (14.5 oz) (15.5 oz)
ACCEPTED MANUSCRIPT

PT
Bundle D
(4 F&V Items,

RI
1 Junk Food/
Snack Item) 1 Bag Frozen 1 Box Frozen

SC
2 Navel Oranges 3 Bananas Sweet Corn 1 Can Green Beans French Fries
(12 oz) (14 oz) (4.75 oz)

U
AN
Bundle E

M
(3 F&V Items,
2 Protein/
Dairy/Grain

D
1 Bag Frozen 1 Box Elbow
Items) 2 Cucumbers 1 Bag Baby Carrots Broccoli Cuts 1 Box White Rice Macaroni

TE
EP (16 oz) (12 oz) (16 oz) (16 oz)
C

Bundle F
AC

(3 F&V Items,
2 Protein/
1 Container
Dairy/Grain
3 Bananas 2 Navel Oranges 1 Can Bartlett Pears Cheerios 1 Bottle 2% Milk
Items) (15 oz) (1.3 oz) (8 oz)
ACCEPTED MANUSCRIPT

TABLE 1. Order of Study Tasks

Task Sub-Task

1. Obtain informed consent

PT
2. Review session
instructions

RI
3. Practice tasks: i. Food choice practice
ii. 7-Digit number memorization and recall practice1

SC
iii. Arithmetic practice

4. View all products in


grocery display

U
AN
5. Food choice task2

6. 5-Minute break
M

7. Eight arithmetic tasks3: i. Section A (4 tasks)4:


Multiply m1 x m2, where integer m1 ~ U (13…19) and
D

integer m2 ~ U (5….9)
TE

ii. 5-minute break

iii. Section B (4 tasks)5:


Multiply m1 x m2, where integer m1 ~ U (13…19) and
EP

integer m2 ~ U (5….9) concurrently with a


memorization task (7-digit number recall at the end of
each task)
C

8. Post-experiment
AC

questionnaire

i. $5 show-up payment

ii. Receive food choice task items (valued at $10)


9. Receive study
incentives:
iii. Memorization task payment ($0-$5)
1 of 7 tasks randomly selected; receive additional $5 if
correctly recalled 7-digit number, else receive
additional $0
ACCEPTED MANUSCRIPT

iv. Arithmetic Task payment ($0-$3)


1 of 8 tasks randomly selected; receive additional $3 if
correctly answered problem, else receive additional $0

PT
1
Memorization tasks served to induce high cognitive load
2
Participants randomized to 1 of 6 treatments upon arrival; treatments can be viewed in Table 1.
3
Arithmetic performance serves as a manipulation check for high cognitive load, following Deck and

RI
Jahedi (2015).
4
Order of tasks randomized within section
5
Half of participants randomly selected to complete Section B first; the 7-digit numbers employed were
4319162; 8568379; 5862413; 2856979

U SC
AN
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT

TABLE 2. Average and Median Percentage Selected, by Item Category and Treatment

PT
Fruit & Vegetable Junk Food/Snack Protein/Dairy/Grain
Items Items Items

RI
Avg. Median Avg. Median Avg. Median
Treatment

SC
(std dev) (IQR) (std dev) (IQR) (std dev) (IQR)

U
48.64 50.00 16.78 20.00 34.58 30.00
T1 (Control, No Load)

AN
(21.77) (30 to 60) (16.96) (0 to 30) (20.11) (20 to 50)

M
54.50 50.00 16.33 10.00 29.17 30.00
T2 (Bundles-No Discount, No Load)
(27.58) (40 to 70) (25.11) (0 to 25) (22.57) (10 to 40)

D
63.51 66.70 13.72 9.10 22.79 25.00

TE
T3 (Bundles-Discounted, No Load)
(20.34)
EP (45.5 to 75) (20.67) (0 to 18.20) (15.23) (10 to 33.33)

49.48 50.00 19.83 20.00 30.69 20.00


T4 (Control, High Load)
(24.09) (40 to 70) (21.56) (0 to 30) (21.35) (10 to 40)
C

56.36 60.00 19.24 20.00 24.40 20.00


AC

T5 (Bundles-No Discount, High Load)


(27.66) (40 to 80) (26.56) (0 to 30) (21.49) (10 to 40)

48.19 50.00 23.30 18.20 28.51 27.30


T6 (Bundles-Discounted, High Load)
(24.16) (30 to 63.63) (28.27) (0 to 40) (22.10) (10 to 41.70)
ACCEPTED MANUSCRIPT

NOTE: This table presents the average and median percentage of grocery items selected by participants from each of the three food item categories. as part of
their choice set, by treatment. Standard deviations and interquartile ranges in parentheses as noted.

PT
RI
U SC
AN
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT
TABLE 3. Comparison Statistics for Food Item Category Percentages, by Treatment

Baseline Comparison Fruit & Vegetable Junk/Snack Food Protein/Dairy/


Treatment Treatment Items Items Grain Items

Avg. % Avg. % Avg. %


Rank- Rank- Rank-
Difference Difference Difference
sum sum sum

PT
over Baseline over Baseline over Baseline
p-value1 p-value1 p-value1
(std dev) (std dev) (std dev)

RI
T2
5.86 -0.45 -5.41
(Bundles-No Discount, No 0.1928 0.1994 0.1180
(4.14) (0.32) (3.83)

SC
Load)
T3
T1 14.87 -3.06 -11.79
(Bundles-Discounted, No 0.0003*** 0.0628* 0.0007***

U
(Control, No Load) (10.51) (2.16) (8.34)
Load)

AN
T4 0.84 3.05 -3.89
0.9846 0.7187 0.2857
(Control, High Load) (0.59) (2.16) (2.75)

M
T5
6.88 -0.59 -6.29

D
T4 (Bundles-No Discount, 0.0359** 0.4601 0.0560*
(4.86) (0.42) (4.45)
(Control, High Load) High Load)

TE
T6
-1.29 3.47 -2.18
(Bundles-Discounted, High 0.9719 0.9959 0.4084
(0.91) (2.45) (1.54)
EP
Load)
C

T2 T3
9.01 -2.61 -6.38
AC

(Bundles-No (Bundles-Discounted, No 0.0584* 0.6964 0.1662


(6.37) (1.85) (4.51)
Discount, No Load) Load)
T5
1.86 2.91 -4.77
(Bundles-No Discount, 0.4575 0.4383 0.1786
(1.32) (2.06) (3.37)
High Load)
ACCEPTED MANUSCRIPT
T3
T6 15.32 -9.58 -5.72
(Bundles-Discounted, No 0.0002*** 0.0398** 0.2480
(Bundles-Discounted, (10.83) (6.77) (4.05)
Load)
High Load)
T5
8.17 -4.06 -4.11
(Bundles-No Discount, 0.0118** 0.4209 0.3307
(5.78) (2.87) (2.91)
High Load)

PT
1
p-values in bold are significant at the 10%*, 5%** and 1%*** level respectively
NOTE: This table presents the average difference in the percentage of grocery items selected by participants from each of the three food item categories, as part of their choice set, using

RI
pairwise comparisons between treatments. p-values obtained from non-parametric Wilcoxon rank-sum 2-sample tests. Standard deviations in parentheses.

U SC
AN
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT

TABLE 4. Percentage of Participants Selecting Bundles, by Treatment

% Selecting Any
Treatment Bundles
(std dev)
1 Bundle 2 Bundles

T2 45.00
36.67 8.33

PT
(Bundles-No Discount, No Load) (50.17)

RI
T3 81.03
25.86 55.17
(Bundles-Discounted, No Load) (39.55)

SC
T5 66.67
37.88 28.79
(Bundles-No Discount, High Load) (47.50)

U
AN
T6 56.06
19.70 36.36
(Bundles-Discounted, High Load) (50.01)
M

NOTE: This table presents the percentage of participants selecting 1, 2, and any bundles as part of their choice set, by treatment. T1 and T4
are omitted as the treatments did not contain product bundles. Standard deviations in parentheses.
D
TE
C EP
AC
ACCEPTED MANUSCRIPT
TABLE 5. Comparison Statistics for Percentage of Participants Selecting 1 or More Bundle, by Treatment

Avg. %
Difference
Rank-sum
Baseline Treatment Comparison Treatment over
p-value1
Baseline
(std dev)

PT
T3
36.03
(Bundles-Discounted, No Load)
T2 (25.48) 0.0001***

RI
(Bundles-No Discount, No Load)

SC
T5
21.67
(Bundles-No Discount, High Load) 0.0147**
(15.32)

U
AN
T6
T3 -24.97
(Bundles-Discounted, High Load)
(Bundles-Discounted, No Load) (17.66) 0.0031***

M
D
T5 T6
-10.61

TE
(Bundles-No Discount, High Load) (Bundles-Discounted, High Load) 0.2126
(7.50)
EP
1
p-values in bold are significant at the 10%*, 5%** and 1%*** level respectively
NOTE: This table presents the average difference in the percentage of participants who selecting 1 or more bundle as part of their choice set, using pairwise comparisons between
C

treatments. T1 and T4 are omitted as the treatments did not contain product bundles. Standard deviations in parentheses.
AC
ACCEPTED MANUSCRIPT
TABLE 6. Marginal Effects for Determinants of Grocery Selection, by Item Category
Fruit & Vegetable Items Junk Food/Snack Items
Variable Estimated Delta-Method Estimated Delta-Method
Marginal Effect Standard Error Marginal Effect Standard Error
HighCognitiveLoad 0.0281 0.0329 0.0031 0.0289

PT
BundlesDisplayed 0.0437** 0.0509 0.0738** 0.0441

BundlesDiscounted 0.1016** 0.0396 -0.0528 0.0403

RI
HighCognitiveLoad*BundlesDiscounted -0.1633*** 0.0505 0.0644 0.0491

SC
BundlesDisplayed*Female 0.0124 0.0540 -0.0901** 0.0460

U
AlreadyPlannedToPurchase 0.0423 0.0407 -0.0842*** 0.0325

AN
FollowingSpecialDiet 0.0515 0.0311 -0.1033*** 0.0315

Female 0.0196 0.0433 0.0375 0.0354

M
ChildrenUnder18 0.0060 0.0412 -0.0149 0.0377

D
NonCaucasian -0.0080 0.0272 -0.0039 0.0239

TE
Age 0.0042 0.0012 -0.0073*** 0.0015

HouseholdIncome10K 0.0054 0.0029 -0.0058* 0.0027


EP
NOTE: values in bold are significant at the 10%*, 5%** and 1%*** level respectively
C
AC
ACCEPTED MANUSCRIPT
APPENDIX A. Fractional Multinomial Logit Estimates for Determinants of Grocery Selection, by Item Category
Fruit & Vegetable Items Junk Food/Snack Items
Variable Robust Robust
Estimated Estimated
Standard z-statistic Pr > |z|1 Standard z-statistic Pr > |z|1
Coefficient Coefficient
Error Error
HighCognitiveLoad 0.1649 0.1439 1.15 0.252 0.1299 0.2231 0.58 0.560

PT
BundlesDisplayed 0.5030 0.2204 2.28 0.022** 0.8551 0.3357 2.55 0.011**

RI
BundlesDiscounted 0.3670 0.1685 2.18 0.029** -0.1365 0.3111 -0.44 0.661

SC
HighCognitiveLoad*BundlesDiscounted -0.6628 0.2183 -3.04 0.002*** 0.0256 0.3811 0.07 0.947

BundlesDisplayed*Female -0.2535 0.2428 -1.04 0.296 -0.8060 0.3618 -2.23 0.026**

U
AlreadyPlannedToPurchase -0.0696 0.1740 -0.40 0.689 -0.6460 0.2428 -2.66 0.008***

AN
FollowingSpecialDiet -0.0877 0.1350 -0.65 0.516 -0.7938 0.2478 -3.20 0.001***

M
Female 0.2415 0.1926 1.25 0.210 0.4253 0.2749 1.55 0.122
ChildrenUnder18 -0.0204 0.1842 -0.11 0.912 -0.1195 0.2976 -0.40 0.688

D
NonCaucasian -0.0575 0.1273 -0.45 0.652 -0.0652 0.1939 -0.34 0.737
Age -0.0029
TE
0.0046 -0.63 0.529 -0.0536 0.0110 -4.87 <0.001***
EP
HouseholdIncome10K 0.0088 0.0114 0.77 0.439 -0.0355 0.0201 -1.77 0.076*
Intercept 0.3474 0.2273 1.53 0.126 1.1470 0.3713 3.09 0.002***
C

Observations 367 367


AC

Wald chi2(24) 92.18


Prob > chi2 0.0000
1
p-values in bold are significant at the 10%*, 5%** and 1%*** level respectively
NOTE: The share of Protein/Dairy/Grain Items is the excluded category
ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC

You might also like