This article was downloaded by: [152.3.102.254] On: 30 July 2019, At: 10:59Publisher: Institute for Operations Research and the Management Sciences (INFORMS)INFORMS is located in Maryland, USA

Management Science

Publication details, including instructions for authors and subscription information:http://pubsonline.informs.org

The Primacy of “What” over “How Much”: How Type andQuantity Shape Healthiness Perceptions of Food PortionsPeggy J. Liu, Kelly L. Haws, Karen Scherr, Joseph P. Redden, James R. Bettman, Gavan J.Fitzsimons

To cite this article:Peggy J. Liu, Kelly L. Haws, Karen Scherr, Joseph P. Redden, James R. Bettman, Gavan J. Fitzsimons (2019) The Primacy of“What” over “How Much”: How Type and Quantity Shape Healthiness Perceptions of Food Portions. Management Science65(7):3353-3381. https://doi.org/10.1287/mnsc.2018.3098

Full terms and conditions of use: https://pubsonline.informs.org/page/terms-and-conditions

This article may be used only for the purposes of research, teaching, and/or private study. Commercial useor systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisherapproval, unless otherwise noted. For more information, contact permissions@informs.org.

The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitnessfor a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, orinclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, orsupport of claims made of that product, publication, or service.

Copyright © 2018, INFORMS

Please scroll down for article—it is on subsequent pages

INFORMS is the largest professional society in the world for professionals in the fields of operations research, managementscience, and analytics.For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org

MANAGEMENT SCIENCEVol. 65, No. 7, July 2019, pp. 3353–3381

http://pubsonline.informs.org/journal/mnsc/ ISSN 0025-1909 (print), ISSN 1526-5501 (online)

The Primacy of “What” over “How Much”: How Type and QuantityShape Healthiness Perceptions of Food PortionsPeggy J. Liu,a Kelly L. Haws,b Karen Scherr,c Joseph P. Redden,d James R. Bettman,e Gavan J. Fitzsimonse

aKatz Graduate School of Business, University of Pittsburgh, Pittsburgh, Pennsylvania 15260; bOwen Graduate School of Management,Vanderbilt University, Nashville, Tennessee 37203; c School of Medicine, Duke University, Durham, North Carolina 27708; dCarlson Schoolof Management, University of Minnesota, Minneapolis, Minnesota 55455; e Fuqua School of Business, Duke University, Durham,North Carolina 27708Contact: peggy.liu@pitt.edu, http://orcid.org/0000-0002-7049-4929 (PJL); kelly.haws@owen.vanderbilt.edu (KLH); karen.scherr@duke.edu,

http://orcid.org/0000-0002-1972-6472 (KS); jredden@umn.edu (JPR); jrb12@duke.edu (JRB); gavan@duke.edu (GJF)

Received: January 22, 2017Revised: November 6, 2017Accepted: March 19, 2018Published Online in Articles in Advance:December 7, 2018

https://doi.org/10.1287/mnsc.2018.3098

Copyright: © 2018 INFORMS

Abstract. Healthy eating goals influence many consumer choices, such that evaluating thehealthiness of food portions is important. Given that both the type and quantity of foodjointly contribute to weight and overall health, evaluations of a food portion’s healthinessought to consider both type and quantity. However, existing literature tends to examinefood type and food quantity separately. Across seven studies, we show that consumerstreat type as a primary dimension and quantity as a secondary dimension, such thata change in type (versus quantity) has a greater impact on perceived healthiness or healthgoal impact, even when holding objective impact constant in terms of calories. We alsoexamine whether one reason this effect occurs is because most consumers consider type (acategorical attribute) before quantity (a continuous attribute). We conclude by discussingextensions of these ideas to other perceptual assessments involving both type and quantity(e.g., price perceptions).

History: Accepted by Yuval Rottenstreich, decision analysis.Funding: Funding was provided by University of Pittsburgh, Vanderbilt University, Duke University,and University of Minnesota for the respective authors affiliated with these institutions.

Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2018.3098.

Keywords: health goal • goal means • quantity • portion sizes • healthiness perceptions • calories • attribute evaluation

1. IntroductionImagine a consumer with a healthy eating goal who isdeciding whether to consume a particular portion ofchocolate candies. Like most consumers with healthyeating goals, this consumer’s main health goal consistsof losing or maintaining weight (International FoodInformation Council Foundation 2012), such that caloriesare typically considered a major objective indicator ofhealth goal impact (Cochran and Tesser 1996, Chandonand Wansink 2007a, Huang et al. 2012, Campbell andWarren 2015). Given the importance of calories as amajor aspect of healthiness for most consumers’ healthgoals, there are two main aspects of the food con-sumed that ought to combine to jointly determine howa given food portion will affect the consumer’s healthgoals: the type of food and the quantity of the food.However, do a consumer’s judgments of the healthinessof a food portion fully factor in both type and quantity, orare they driven more by one dimension than the other?

In this research, we examine the effects of varyingfood types (e.g., chocolates versus almonds versuscrackers) and varying food quantities (e.g., 1/2 servingversus 1 serving versus 2 servings) on healthiness

perceptions. In comparison, the existing food and healthliterature generally either (1) separately examines typeor quantity; or (2) does not distinguish between typeand quantity, instead treating them as somewhat in-terchangeable routes to healthier consumption in termsof decreasing calories. We instead introduce an explicitcomparison between pursuing health via changing thetypeversus quantity of a food.Ourmainproposition is thatfor healthiness perceptions, food type is treated as a “pri-mary dimension” (i.e., a highly salient dimension thatdominates judgments), whereas food quantity is treated asa “secondary dimension” (i.e., a dimension that does notaffect judgments much unless made salient, and even thenaffects judgments less than a primary dimension).We demonstrate the secondary nature of quantity

and the spontaneous tendency to underweight quan-tity through multiple lines of evidence. First, we showthat when consumers are asked to judge the healthinessof a consumed food portion, their judgments are highlysensitive to food type but not to food quantity. Namely,most consumers either do not adjust their perceptionsto quantity at all, or when the salience of quantity isincreased, adjust them but only to a small extent. Second,

3353

although our main focus is on healthiness perceptions,we also examine consumers’ caloric perceptions andshow that they track less than 1:1 with size perceptions,again suggesting an underweighting of quantity incaloric perceptions. In fact, we show that the secondarynature of quantity in healthiness assessments holds evenwhen consumers are provided with caloric information.Further, we show process by moderation evidence thatthe predicted secondary nature of quantity in healthinessassessments is driven by the (majority segment of) con-sumers who adopt a healthiness evaluation strategyof focusing first on food type and then subsequentlyadjusting for food quantity (and is mitigated among theminority segment of consumers who instead adopt anevaluation strategy of focusing first on food quantity andthen subsequently adjusting for food type). Finally, weexamine choice implications with consumers instructedto adopt a weight loss or management health goalchoosing between portions of two calorically dense butdifferentially healthy foods (chocolates versus almonds).We find that consumers choose the healthier food type(almonds) even when the size of the almond portionmeans that the caloric content of the almond portion farexceeds the caloric content of the chocolate candiesportion. This finding suggests that the secondary natureof quantity is also reflected in consumers’ food choices.

We encapsulate our findings through a descriptivetheory that food type is a salient primary dimension ofhealthiness, whereas food quantity is a less noticedsecondary dimension (see e.g., Dryer 2006 on the valueof descriptive theories). Further, we also present ex-planatory reasoning underlying this descriptive theoryby drawing from work on categorical versus continuousattributes and heuristic versus systematic processing.This theory fully accounts for our main empirical find-ings, and it also makes several broader contributions.We add to the goals literature by showing that whena change in type or quantity results in the same caloricchange (or even when the change in quantity results ina greater caloric change), the type changes are per-ceived by consumers to affect healthiness to a greaterextent. This stands in contrast to a common tendency totreat calories as a main aspect of healthiness, given thatweight loss or management is a common health criterionfor many Americans (e.g., Kuo et al. 2009). Further, ourresearch also contributes to the food decision-making lit-erature, which has mainly used two different choice par-adigms (either choice among different food types or choiceamong different food quantities) to examine the healthi-ness of food choices. We show that the choice paradigmused affects consumers’ perceptions of the healthinessdifferences between options, suggesting that conclusionsfrom one choice paradigm may not always translate di-rectly to contexts involving the other choice paradigm.

The remainder of this paper is structured as fol-lows. First, we review the extant food literature (which

generally examines type and quantity separately) andthe goals literature (which often treats type and quantitysimilarly). Second, we review literature relevant to cat-egorical versus continuous attribute processing andheuristic versus systematic processing, which leads toour main proposition about the secondary nature ofquantity. Third, we present seven studies testing thisproposition in different ways. Finally, we conclude bydiscussing theoretical and managerial contributions andfuture directions, including howour primary–secondaryaccountmay extend to other contexts in which both typeand quantity ought to jointly influence attribute judg-ments yet type may still exhibit primacy.

2. Conceptual Development2.1. Food Type and Food QuantityA recent survey of American adults found that 54% hada goal to lose weight, and 25% had a goal to maintaintheir weight (International Food Information CouncilFoundation 2012). With weight loss or weight man-agement as one’s main health goal, calories can beconsidered a key objective indicator of health impact(Cochran and Tesser 1996, Chandon and Wansink2007a, Huang et al. 2012, Campbell and Warren 2015).However, particularly when health is defined in termsof decreasing and managing caloric intake, consumerscan pursue health via food consumption in two mainways: changing what they eat, or how much they eat. Inother words, consumers can pursue health by switchingto healthier food types or to smaller quantities of thesame less healthy food types (or, they could use a com-bination of these approaches, butwe focus on comparingthem). These two broad approaches to healthier con-sumption are frequently used and are well studied butare nearly exclusively examined in isolation.1

The first approach (via type) involves the standardchoice between a more versus less healthy food type(e.g., fruit versus cake) (Dhar and Simonson 1999, Shivand Fedorikhin 1999, Kivetz and Zheng 2006, Laran2010, Gal and Liu 2011, Liu et al. 2015). Here, the se-lection of the less healthy option in a choice set indicatesamore indulgent and less self-controlled behavior.Muchresearch has been produced using this standard choiceparadigm. For instance, consumers making decisionsaffectively (versus cognitively) are more likely to choosea less healthy (versus healthier) food type (Shiv andFedorikhin 1999). Further, when consumers feel thatthey deserve to reward themselves for previous effortfulacts, the choice share of chocolate cake (versus fruit salad)increases (Kivetz and Zheng 2006).The second approach (via quantity) involves

choosing from different-sized portions, typically ofunhealthy foods (e.g., different sizes of fries or soda)(Sharpe et al. 2008, Dubois et al. 2012, Wansink 2012,Haws and Winterich 2013, Cornil and Chandon2016). Understanding how people perceive, judge,

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food Portions3354 Management Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS

and respond to portion size is of high practical im-portance. Many public policy advocates and academicresearchers have focused on how large (and growing)portion sizes are a key factor contributing to obesity(Young and Nestle 2002, Nielsen and Popkin 2003,Raynor 2014, Wansink and Chandon 2014, Dallas et al.2015). Because consumers often eat most of the foodthey select for themselves or are served (Collins 2006,Schwartz et al. 2012), larger portion sizes typicallylead to increased consumption (Wansink 1996,Wansinkand Park 2001, Rolls et al. 2002). Most portion size re-search has focused on the effect of researcher-providedlarger portions on increased consumption, attributingthis effect to difficulties with monitoring consumptionquantity (Rolls et al. 2002,Wansink et al. 2005), behavioralnorms (Miller et al. 2015), or biases in portion sizeestimation (Raghubir and Krishna 1999, Wansink andVan Ittersum 2003, Chandon and Ordabayeva 2009,Ordabayeva and Chandon 2016).

In sum, the extant literature generally examines foodtype or food quantity separately and has not system-atically examined whether type and quantity means areperceived differently in terms of subjective health impact.The present research examines how these two dimensionsof food portions (“what/type” and “how much/quan-tity”) differentially affect perceptions of how healthy foodportions are. Our main proposition is that food type willbe a “primarydimension” that is quite salient,with a strongimpact on perceived healthiness and subjective health goalimpact, whereas food quantity will be a “secondary di-mension” that is attended to less by most consumers andthus has a much weaker impact on these perceptions.

2.2. Primary vs. Secondary DimensionsOur primary–secondary account draws on literature onthe order of processing information (Hogarth 2001,Kahneman and Frederick 2002): we suggest that foodtype (a categorical attribute) is primary because categor-ical attributes tend to be processed first. By contrast, foodquantity (a continuous attribute) is secondary becausecontinuous attributes tend to be processed afterward.

Prior food decision-making research suggests thatpeople quickly and automatically categorize food typesas either inherently healthy or unhealthy, pointingtoward food type as a primary dimension. Consumersevaluating food products often display highly cate-gorical thinking (Rozin et al. 1996, Oakes 2005, Oakesand Slotterback 2005, Chernev and Gal 2010), begin-ning to categorize foods as good or bad starting ata very young age, perhaps in part owing to frequentexposure to messages from their parents and schoolteachers that they should eat certain foods and avoidothers (Nguyen 2007). Importantly, categorizations arequite powerful in guiding judgments and deci-sion making (Peeters 2002, Fox et al. 2005, Chernevand Gal 2010). Once an item makes it into a particular

“category,” that categorization will have an undue in-fluence on subsequent decisions. For example, Chernevand Gal (2010) found that the tendency to categorizefoods as unhealthy or healthy led people to estimate thecaloric content of a combination of the two types (e.g.,cheeseburger and side salad) using an averaging ap-proach that allows the healthy food to cancel out theunhealthy food (rather than the appropriate additiveapproach), and Oakes (2004) found that people per-ceived the addition of a negatively categorized ingre-dient (e.g., caramel coating) to invalidate the positivenutrients of a food it covered (e.g., an apple). Further,Oakes (2005) found that perceived healthiness is moreinfluenced by food type than by information on caloriccontent, and Chandon andWansink (2007a) found thatperceived caloric content is influenced more by thebrand of fast-food restaurant than by the actual caloriccontent. Other research in the food context that is con-sistent with strong categorical thinking is the notion of“dose insensitivity”—the mistaken belief held by a mi-nority of individuals (approximately 20%) that a smallamount of salt or fat in the diet is worse than none (Rozinet al. 1996). Collectively, these findings all show thatconsumers have a spontaneous tendency to discretelycategorize a food as either healthy or unhealthy, and theyoffer hints at the tendency for quantity to be somewhatneglected (Rozin et al. 1996, Chernev and Gal 2010).In the present research, we offer a more systematic

examination of when and to what extent food quantityfactors in and also offer an underlying explanation.Again, we posit that food quantity will serve as a sec-ondary dimension. Our underlying explanation for thisprediction draws both on the literature on categoricalthinking in the food domain (referenced above) andalso from general cognitive processing literature. Thisliterature distinguishes between heuristic (system 1)and systematic (system 2) processing (Kahneman andFrederick 2002). Because system 1 processing acts moreon “prototypes” (Kahneman and Frederick 2002, p. 51),and a categorical attribute is inherently more based onprototypes, such work suggests that food type may becovered by quicker system 1 processing. Indeed, asHogarth (2001) notes, categorization based on salientfeatures—which we would suggest food type is—isbased on a quick, intuitive process. Indeed, to offer anexample2 from animal taxonomy, natural classification ofanimals is likely based more on the (categorical) attributeof feature presence (e.g., presence or absence of feathersfor birds) than the (continuous) attribute of animal size(e.g., hummingbirds and ostriches are both birds).Given that quantity is a continuous attribute, whereas

type is a categorical attribute, we thus propose that typewill be primary in healthiness perceptions and quantitywill be secondary. Additionally, some evidence in thefood domain is consistent with this secondary nature ofquantity: research has found that consumers often

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food PortionsManagement Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS 3355

fail to monitor consumption quantities (Wansink2004). In our research, we mainly examine subjectiveperceptions of healthiness3 but also examine caloricperception measures in one study because calories areoften considered the objective indicator of health impactwhen it comes to a weight loss or management goal. Ourprimary–secondary account makes predictions abouthow subjective health perceptions are affected, yet thenotion that quantity is secondary in health perceptionsalso suggests that caloric perceptions may underweightquantity. In sum, our main proposition is that foodtype is primary and food quantity is secondary in per-ceptions of how healthy food portions are.

Our conceptualization of quantity as a secondarydimension in healthiness judgments also informs theprocess-related prediction that (1) increasing the salienceof quantity will increase the effect of quantity differenceson healthiness judgments of food portions, but that(2) quantity will still remain secondary. That is, evenwhen quantity is made salient, quantity will still besecondary to type (albeit to a lesser extent). Additionally,it informs the underlying process prediction that mostconsumers will prefer to acquire food type informationbefore food quantity dimension information whenmaking healthiness judgments of food portions. Inthat sense, this process hypothesis also correspondsto the anchoring-and-insufficient-adjustment modelused to explain the effect of numeric anchors (Epley andGilovich 2006), whereby food type is analogous to theinitial anchor and insufficient adjustment ismade for foodquantity. Finally, it supports the prediction that we shouldobservemoderation by individual consumer differences inevaluation strategy. Specifically, although we predict thatthemajority of consumerswill choose to process food typefirst, a minority of consumers may choose to process foodquantity first; our effects should be larger among thoseconsumers who choose a type-then-quantity evaluationstrategy over a quantity-then-type evaluation strategy.

3. Overview of Studies andAnalytical Approaches

3.1. Overview of StudiesWe test our primary–secondary account in seven studies(see Table 1 for a summary). Study 1 provides evidencefor our primary–secondary account by showing thatconsumers’ healthiness evaluations of food portions re-flect food type differences but are largely insensitive tofood quantity differences, even though consumers areable to perceive the quantity differences.

Study 2 then further tests our primary–secondaryaccount by adding emphasis to the actual consumptionor eating of (nearly) the entire food portion. Further,study 2 examines the additional outcomes of caloricperceptions and caloric estimation.

Studies 3a to 3c then test predictions stemming fromour primary–secondary account by examining three

theoretically motivated ways of increasing the salienceof quantity at the time of making healthiness judg-ments. Our account posits that quantity is a secondarydimension (not a nondimension); hence increasing thesalience of quantity should lead quantity to be incorpo-rated into healthiness judgments. Study 3a increases thesalience of quantity byprompting a joint evaluationmodeamong participants (Hsee 1996, Gonzalez-Vallejo andMoran 2001). Study 3b increases the salience of quantityby drawing from expectancy-disconfirmation theory(Bettman 1979, Helgeson and Beatty 1987) to createa condition in which participants evaluate a portionsize much larger than would be expected for con-sumption on one occasion. In doing so, study 3b testswhether a latitude of acceptance exists where quantitiesthat could reasonably be consumed in a single sitting donot factor into healthiness assessments, but that outsidethis latitude consumers become considerably more sen-sitive. Study 3c then increases the salience of quantity bygauging a specific aspect of health very closely tied toquantity: perceived weight impact over time.4

Study 4 then further tests our primary–secondaryaccount of healthiness perceptions by examininghealthiness perceptions when caloric informationabout food portions is explicitly provided. Importantly,Study 4 also includes additional measures to examineour underlying attribute processing order explanation,testing our prediction that healthiness perception ef-fects will be moderated by individual differences inchoice of evaluation strategy (i.e., type-then-quantityversus quantity-then-type). By testing for moderationby individual differences in choice evaluation strategy,we are able to address alternative accounts for thefindingsbased on interpretation of what healthiness means.Finally, to further test our primary–secondary ac-

count, study 5 examines consumers’ choices rather thanperceptions of health impact. Consumers were givena weight loss or management health goal and instructedto choose between two calorically dense food portions,with one portion composed of a healthier food type andthe other composed of a less healthy food type.We positthat once a food is perceived as being a healthier foodtype, it is also perceived to be a better choice that is lessthreatening to one’s health largely regardless of therespective portion sizes. Thus, the healthier food typewill be chosen more often than its caloric contentwould justify, potentially having a negative impact onefforts to lose or maintain weight through reducedcaloric consumption.Next, we present aspects of decisions regarding data

collection and statistical analyses that are commonacross studies, and we then present the individualstudies. All sample sizes were decided in advance of anydata analysis (Simmons et al. 2011). Note that in allstudies we use the term “portion size” in reference toany quantity of food presented and “serving size” with

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food Portions3356 Management Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS

Tab

le1.

Summaryof

Stud

ies

Stud

y

Stim

uli

Mainfind

ing

Food

type

(s)

Portionsizesa

Portion

depiction

form

at“H

ealth

y”“U

nhealth

y”

1Ba

bycarrots,

almon

ds,

Whe

atTh

ins

Minic

hocolate

chip

cook

ies,

Che

etos,p

lain

M&Ms

1/2,

1,2

Before

and

afterph

otos

inavisu

alfood

diary

Basicev

iden

ceforthesecondarynature

offood

quan

tity

inhea

lthinessevaluations:

Acrossarang

eof

food

quan

titiesthat

couldbe

consum

edin

atypicalsitting,

consum

erslargelyfocu

sedon

food

type

differen

cesan

ddidno

tad

just

healthiness

percep

tions

inresp

onse

tofood

quan

titydifferen

ces.Fu

rthe

r,pe

ople

perceive

dsize

differen

cesbe

tweenfood

portions:the

second

aryna

ture

offood

quan

tityis

not

driven

byinab

ility

tope

rceive

size

differen

ces.

2Ba

bycarrots,

almon

ds,

Whe

atTh

ins

Minic

hocolate

chip

cook

ies,

Che

etos,p

lain

M&Ms

1/2,

1,2

Before

and

afterph

otos

inavisu

alfood

diary

Examiningother

assessmen

tsof

hea

lthim

pact,includingcaloricperception

san

dcaloricestimates:S

tudy

2exam

ined

severald

ifferentde

pend

entmeasuresga

uging

percep

tions

ofhe

alth

impa

ctto

furthertest

ourprim

ary–

second

aryaccoun

t.Stud

y2

used

ahealthinesspe

rcep

tionmeasure

focusedclearlyon

eatin

g(e.g.,“W

aseatin

gthis

snack...”)an

dshow

edthat

thismeasure

also

reflectedthesecond

aryna

ture

ofqu

antity.

Stud

y2also

includ

edcaloricpe

rcep

tionmeasuresan

dshow

edthat

caloricpe

rcep

tions

also

seem

edto

unde

r-incorporatequ

antitydiffe

rences.

3aAlm

onds

PlainM&Ms

1,2

Before

and

afterph

otos

inavisu

alfood

diary

Increa

singsalien

ceof

food

quan

tity:W

henfood

quan

titywas

mad

esalient,con

sumers

somew

hatincorpo

ratedqu

antityinto

theirh

ealth

inesspe

rcep

tions,but

toalesser

extent

than

wou

ldbe

expe

cted

basedon

thecaloricdiffe

rences

(studies

3aan

d3b

)or

size

percep

tiondiffe

rences

(study

3c)betw

eenqu

antities.Th

esestud

iesprov

ideeviden

cethat

quan

tityisasecond

arydimension

,not

ano

ndim

ension

.•Stud

y3a

utilizesasimilarhealthinessmeasure

asstud

y1,

increasing

thesalienceof

quan

tityby

man

ipulatingevalua

tionmod

e(sep

arateversus

joint)(H

see1996).

•Stud

y3b

utilizesasimilarhealthinessmeasure

asstud

y1,

increasing

thesalienceof

quan

tityusingexpe

ctan

cydiscon

firm

ation,by

includ

ingacond

ition

inwhich

participan

tsevalua

tedafood

quan

titymuchlargerthan

wou

ldbe

expe

cted

tobe

consum

edinatypical

sitting

(Bettm

an1979,H

elgesonan

dBe

atty

1987).

•Stud

y3c

utilizesameasure

ofweigh

timpa

ctov

ertim

e,which

increasesthesalienceof

quan

tityviathesing

ular

focuson

weigh

tim

pact

over

time.

3bAlm

onds

PlainM&Ms

1,2,

8Be

fore

and

afterph

otos

inavisu

alfood

diary

3cBa

bycarrots,

almon

ds,

Whe

atTh

ins

Minic

hocolate

chip

cook

ies,

Che

etos,p

lain

M&Ms

1/2,

1,2

Before

and

afterph

otos

inavisu

alfood

diary

4Alm

onds

Pean

utM&Ms

1/4cu

p(peanu

tM&Ms);1

/4or

1/2cu

p(alm

onds

)

Verba

lExten

dingto

contextswithcaloricinform

ationan

dex

aminingtheproposed

underlyingprocess:W

efurthe

rtested

ourprim

ary–

second

aryaccoun

tinacontext

with

caloricinform

ation.

Wealso

used

measu

resad

aptedfrom

Kride

ret

al.(20

01)to

furthe

rexam

inetheun

derlying

attributeprocessing

orde

rexplan

ationforou

rprim

ary–

second

aryaccoun

t.5

Alm

onds

PlainM&Ms

1/2(plain

M&Ms);1

or2

(alm

onds

)

Photos

ofafood

portion

Choice-based

eviden

ceforthesecondarynature

offood

quan

tity:C

onsumers

instructed

tochoo

sewith

ahe

alth

goal

consistin

gof

weigh

tloss

orman

agem

ent

gene

rally

chosealmon

dsov

erM&Ms,rega

rdless

ofqu

antity(and

thus

calories)of

almon

ds,sho

wingthat

choicesalso

reflectthesecond

aryna

ture

ofqu

antity.

a Inman

ufacturer-su

ggestedservingsizes,un

less

othe

rwisesp

ecified

.Given

that

themostcommon

health

goal

isto

lose

weigh

tor

maintainweigh

t,larger

portions

shou

ldob

jectivelybe

considered

less

healthy—

espe

cially

fortheman

ycalorically

densefood

sthat

weinclud

edacross

ourstud

ies(nearlyallfood

sweexam

inein

this

research

,other

than

baby

carrots).

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food PortionsManagement Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS 3357

respect to externally recommended serving size quan-tities (typically based on the manufacturer’s guidelines).

3.2. Null-Hypothesis Significance Testing andEffect Sizes

In some of the studies (e.g., study 1), traditional null-hypothesis testing suggests that despite noticeabledifferences in portion sizes, people sometimes do notaccount for these easily perceivable differences inquantity in their healthiness assessments. In otherwords, there seems to be a potential null effect ofquantity (complete quantity insensitivity). In suchcases, we additionally report Bayesian analyses, whichcan provide valid statistical support for a null hy-pothesis that food quantity receives no weight injudgment (see next section). We also report effect sizesthroughout alongside p-values, given that effect sizemeasures do not vary according to sample size (al-though effect size measures are less stable at smallersample sizes). For analysis of variance (ANOVA)F-tests, we report partial η2 as an effect size measure;typical benchmarks refer to 0.01 as a small effect size,0.06 as a medium effect size, and 0.14 as a large effectsize (Cohen 1988). For t-tests, we report Cohen’s d as aneffect size measure; typical benchmarks refer to 0.2 asa small effect size, 0.5 as amedium effect size, and 0.8 asa large effect size (Cohen 1988). Of note, effect sizemeasures for within-subject factors (e.g., the foodquantity factor for studies 1, 2, and 3c; the food quantityfactor in the joint evaluation condition in study 3a)generally overestimate the “true” effect size (Dunlapet al. 1996, Olejnik and Algina 2003, Maxwell andDelaney 2004, Lakens 2013). If anything, however,inflated effect sizes for the effect of food quantity rep-resent a more conservative test of our main proposi-tion that food quantity is secondary to food type.Finally, for χ2 tests, we report Cramer’s V as an effectsize measure; typical benchmarks refer to 0.1 as a smalleffect size, 0.3 as a medium effect size, and 0.5 as a largeeffect size (Cohen 1988).

3.3. Bayesian StatisticsTo test whether a null effect of food portion quantity onhealthiness assessments occurs, we report Bayes factors(BF10 and BF01; calculated using JASP version 0.7.5.6and 0.8.2 and the default priors) for our main analyseswhen a null effect seems possible on the basis of tra-ditional null-hypothesis significance testing (Kassand Raftery 1995, Jeffreys 1998, Rouder et al. 2009,Wagenmakers et al. 2018). BF10 represents a ratio of thepredictive performance of a model predicting the al-ternative hypothesis to a model predicting the nullhypothesis, and BF01 represents the inverse (Jeffreys1998). In other words, BF01 = 1/BF10. Thus, BF10 > 1represents evidence in favor of the alternative model(i.e., that there are differences between groups), whereas

BF01 > 1 represents evidence in favor of the null model(i.e., that there are not differences between groups).Accordingly, if there is quantity insensitivity, we ex-pect to find BF01 > 1 for a model with portion sizecondition as the independent variable; relatedly, un-der conditions for which we expect quantity insensi-tivity to be eliminated, we expect to find BF10 > 1.Although there are known issues with using verbal la-bels with respect to Bayes factor sizes, an interpretationprovided by Wagenmakers et al. (2018) is that BF from1 to 3 provides anecdotal or marginal evidence for thenull over the alternative (in the case of BF01) or the al-ternative over the null (in the case of BF10), BF from 3 to 10provides moderate evidence, BF from 10 to 30 providesstrong evidence, BF from 30 to 100 provides very strongevidence, and BF > 100 provides extreme evidence. Insum, using Bayes factors allows us to establish the degreeof support for a null effect of quantity (i.e., completequantity insensitivity).

4. Study 1: Secondary Nature of FoodQuantity Relative to Food Type

Study 1 examines healthiness perceptions of variousfood portions (1/2, 1, and 2 times the manufacturer’sserving size) and various food types using a pro-totypical healthiness inquiry used to evaluate differentoptions (Wilcox et al. 2009, Irmak et al. 2011, Liu et al.2015). There were two important aspects of study 1’sstimuli and methods. First, study 1 utilized a beforeand after visual “food diary” approach to present eachsnack episode. By asking participants to evaluate avisual food diary involving before and after photosto indicate the snack consumed, it was made clear toparticipants that we were asking them to evaluate thehealthiness of everything eaten in total on this par-ticular snack episode. Second, study 1 also measuredportion size perceptions to examine whether partici-pants are able to perceive the quantities as beingdifferent. We also assessed size perceptions becausesome prior work suggests that consumers have dif-ficulty with accurate size perception (Chandon andWansink 2006, Chandon andWansink 2007b, Chandonand Ordabayeva 2009, Van Ittersum and Wansink 2012,Ordabayeva and Chandon 2016). Thus, an alternativeexplanation for the secondary nature of quantity could bean inability to perceive size differences. Our primary–secondary account does not rely on failure to perceivesize differences as the reason for the secondary nature ofquantity, and thuswe attempt to establish that consumersindeed can perceive size differences between the foodportions used in this study.

4.1. Study 1 MethodA total of 185 participants recruited from AmazonMechanical Turk (MTurk) (48% female; Mage = 35.3years) participated. Participants were randomly

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food Portions3358 Management Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS

assigned to view either a set of three relatively healthysnacks (baby carrots, almonds, andWheat Thins) or a setof three relatively unhealthy snacks (mini chocolate chipcookies, Cheetos, and plain M&Ms). Food quantity wasvaried at three levels for each snack (small, medium, andlarge were equivalent to 1/2, 1, and 2 times the man-ufacturer’s recommended serving size for each food,respectively). Food quantity was a within-subject factor,with each participant being exposed to each of the threesizes balanced across the set of three snacks evaluated.We created ordered conditions such that participantswere randomly presented one of the six (i.e., 3 ways toassign three snacks across three sizes) possible combi-nations of pictures for their randomly assigned healthyor unhealthy condition. The three foods (and associatedquantities) were then presented in a random order. Forinstance, a participant might rate amedium portion ofbaby carrots, a large portion of almonds, and thena small portion of Wheat Thins. Another participantmight rate a small portion ofmini chocolate chip cookies,a large portion of Cheetos, and then a medium portion ofplain M&Ms.

We utilized a procedure carefully designed to makeclear the approximate quantity consumed by pre-senting a set of pictures for each of the three foods thatwas said to be from someone’s visual food diary. Spe-cifically, all participants read:

Please imagine that someone you know has been tryingto improve their diet and they have been tracking theirconsumption of food using a visual diary, in which theyrecord everything they eat using pictures.

First, they take a picture of their plate/bowl/containerat the beginning of their snack and then again when theyare finished.

You will be shown, in no particular order, snacks thatthey consumed on 3 different recent days. Please look ateach entry in their visual food diary and answer thequestions that follow.

Participants then saw a food diary entry with twopictures (“start” view and “end” view), which includedplausible dates and similar time stamps to enhance therealism of the visual food diary.We held the quantity offood on the ending plate to a small and constant quan-tity across size conditions of the same food. As such, foreach of the six food stimuli, the ending quantity was setto be as close to 10% of the small portion as possible. SeeFigure 1 for one diary entry (Figure A.1 in the appendixcontains the full set of diary entries).

Under each food diary entry (i.e., a before and anafter picture; Figure 1), participants rated their per-ceptions of the healthiness of that snack episode onthree items, adapted from Irmak et al. (2011): “Pleaseindicate how healthy you believe that this snack was”(1 = not at all healthy, 9 = very healthy), “Please indicatehow nutritious you believe that this snack was” (1 = not

at all nutritious, 9 = very nutritious), and “How wellwould this snack fit within this person’s overall diet?”(1 = not at all, 9 = very much so). After answering,participants were then shown the next visual before–after food diary entry.After all healthiness assessments were made, partic-

ipants were shown the same three before–after diaryentries a second time and asked about perceptionsof the size of the portion to ensure that participantsindeed had the ability to potentially perceive these dif-ferences (these size perceptions were also used sub-sequently as a reference for studies 2 and 3c). Specifically,following each of the same three food diary entries frombefore, participants indicated size perceptions on a9-point scale anchored by “very small” and “very large.”This study and all others concluded with demographicinformation.

4.2. Study 1 ResultsWe first examined perceptions of the size of each snackto ensure that they were perceived as different. Foreach of the six food types, a one-way ANOVA of sizecondition on size perceptions revealed the expectedsignificant trend, in that each successive sizewas viewedas larger (see means in Table 2a). Thus, any subsequenteffects do not merely reflect an inability to perceive thesize differences.To account for the design of our study, we performed

a mixed-model analysis on our key dependent variableof healthiness perceptions, which was a composite ofour three measures (α = 0.98; see means in Table 2b).The model included food quantity as a within-subjectfactor, food type as a between-subjects factor, and theirinteraction. There was a main effect of food type[F(1, 183) = 1119.98, p < 0.001, ηp

2 = 0.86], no effect offood quantity [F(2, 366) = 0.89, p = 0.410, ηp

2 = 0.005], andno interaction [F(2, 366) = 0.27, p = 0.762, ηp

2 = 0.001]. Toconfirm these results and provide more support for thehypothesis that food quantity receives no weight injudgment, we also conducted a mixed Bayesian ANOVAwith food quantity as a within-subject factor, food typeas a between-subjects factor, and their interaction. Thisrevealed extreme evidence for food type having aneffect (BF10 > 100), strong evidence for food quantityhaving a null effect (BF01 = 21.28; BF10 = 0.047), andstrong evidence for the main effects model over theinteraction model by a Bayes factor of 20.83.To test our specific predictions, we performed

planned contrasts. As noted earlier, there was a maineffect of food type with higher healthiness evaluationsfor the healthier foods [M= 7.00 versusM=1.93;F(1, 183) =1119.98, p < 0.001, ηp

2 = 0.86]. Moreover, providing evi-dence of the secondary nature of quantity, there were notany significant pairwise least significant difference (LSD)results between the small (Msmall = 4.59) and mediumportion (Mmedium = 4.45; p = 0.394, d = 0.05), small and

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food PortionsManagement Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS 3359

large portion (Mlarge = 4.39; p = 0.166, d = 0.07), or themedium and large portion sizes (p = 0.695, d = 0.02).

Thus, despite viewing three different snacks at threedifferent quantities (and being able to perceive thosequantity differences), participants did not perceivedifferences in healthiness on the basis of food quantity.Rather, healthiness perceptions seemed to be drivensolely by type of food.

4.3. Study 1 DiscussionStudy 1 used a visual before and after food diary ap-proach and provided evidence that people’s natural re-sponse when assessing a snacking episode’s healthinessusing a prototypical healthiness measure is to largelyignore quantity information (i.e., to treat quantity as asecondary dimension). Rather, healthiness perceptionswere driven entirely by food type. This finding held forboth relatively healthier food types (carrots, almonds,and Wheat Thins) as well as less healthy food types(M&Ms, cookies, andCheetos). Thisfinding that quantityinsensitivity occurred for both food types addressesa prescriptive norms alternative account that per-haps consumers are correctly gauging that consuminglarger portions of the healthier food types—carrots inparticular—is similarly healthy as (or even more healthythan) smaller portions. It is much harder to make thatcase for the less healthy food types. Additionally,quantity insensitivity occurred even though the before

andafter fooddiary setupwasdesigned to focuspeople onthe overall healthiness of the snack portions being eatenand the change in quantity was easily apparent in thebefore and after photos. It seems that people do notfind it natural to incorporate quantity into judgmentsof overall healthiness. Finally, study 1 found differencesin subsequent size perceptions, providing evidenceagainst the alternative account that an underlying visualperceptual bias in viewing the pictures drove the health-iness perception effects. Participants could easily perceivethe food quantity differences, but they still failed tospontaneously consider these differences when judginghealthiness.

5. Study 2: Changing the Measure ofHealth Impact

Study 1 used a healthiness perceptions measure that iscommonly used in prior research (i.e., variants on “howhealthy is this snack”) (Wilcox et al. 2009, Irmak et al.2011, Liu et al. 2015), combined with before and afterpictures of the snack to facilitate potentially noticingquantity. Study 2 used several considerably differentmeasures of health impact: (1) a different measure ofhealthiness that is focused clearly on eating (i.e., variantson the question “was eating this snack . . .”), (2) a caloricperception measure, and (3) a caloric estimation measure.The purpose of using a different healthiness measurephrased to focus clearly on eatingwas to examinewhetherour primary–secondary account generalizes to a differentphrasing other than variants of “how healthy is thissnack.”Although “howhealthy is this snack” is a commonmeasure of healthiness perceptions and also mapsonto how consumers typically evaluate the healthi-ness of different options as they make food decisions,one possibility is that the secondary nature of quantityis specific to this kind of healthiness assessment.5 Wepredicted that using a different healthiness assess-ment focusing clearly on eating might increase thesalience of quantity (because eating refers to foodintake, which emphasizes volume more), such thatquantity insensitivity would be somewhat mitigated.

Figure 1. (Color online) Study 1 Before and After FoodDiary Entry Example

Table 2a. Study 1 Mean Size Perceptions for Snack, Depending on Type and Quantity

Food type

Food quantity

Omnibus test for differencesSize 1 (1/2 serving) Size 2 (1 serving) Size 3 (2 servings)

“Healthy”Almonds 2.97 4.39 6.30 F = 31.33, p < 0.001, ηp

2 = 0.410Carrots 3.06 4.87 6.97 F = 40.37, p < 0.001, ηp

2 = 0.473Wheat Thins 2.91 5.50 6.03 F = 33.49, p < 0.001, ηp

2 = 0.427“Unhealthy”Cookies 2.87 4.30 5.67 F = 20.80, p < 0.001, ηp

2 = 0.318M&Ms 2.93 5.70 6.53 F = 33.77, p < 0.001, ηp

2 = 0.431Cheetos 2.80 3.78 5.50 F = 21.54, p < 0.001, ηp2 = 0.326

Note. These size perceptions are also used as a comparison reference for studies 2 and 3c because the same stimuli are used in studies 1, 2, and 3c.

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food Portions3360 Management Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS

Importantly, however, we still predict that quantitywill have a relatively small impact even when usingthis different measure of healthiness.

The purpose of including the caloric perception andcaloric estimation measures was to extend our un-derstanding of the role of food quantity in consumers’caloric perceptions. Although our primary–secondaryaccount is with respect to healthiness evaluations, notcaloric perceptions, the notion that quantity is sec-ondary in healthiness perceptions is suggestive thatcaloric perceptions may underweight quantity. Thus,we examined caloric perceptions and caloric estimatesby exploring to what extent these measures incorporateperceived food quantity. By definition, caloric per-ceptions should map 1:1 with size perceptions, but anunderweighting of food quantity would suggest lessthan a 1:1 mapping with size perceptions.

5.1. Study 2 MethodA total of 180 participants recruited from MTurk(44% female;Mage = 35.2 years) participated. The samedesign, visual before and after food diary procedure,and stimuli were used as in study 1, and a similar samplesize was also used. The primary difference was in themeasures collected.

First, for each snack episode (as depicted by a fooddiary entry consisting of a before photo and an afterphoto), participants responded to four items explic-itly phrased to focus on eating the snack: “Was eatingthis snack” (1 = not at all healthy, 9 = very healthy),“Was eating this snack” (1 = not at all nutritious, 9 =very nutritious), “Was eating this snack” (1 = a badthing, 9 = a good thing), and “Did eating this snack fitwithin this person’s overall diet?” (1 = not at all, 9 =very much so).

We then measured perceived caloric content. Partici-pants were shown the same three diary entries a secondtime and asked to respond to two questions regardingthe perceived caloric content of the food portion. Spe-cifically, “How many calories do you think this snackhas?” (1 = very few calories, 9 = a lot of calories) and “Howmany calories do you estimate are in this snack?”

(numeric free-response). We did not assess size per-ceptions in study 2 to keep the survey length man-ageable; instead, because we used the same stimuli asin study 1, we refer to the size perceptions data instudy 1 as a reference (as a size pretest in many ways).

5.2. Study 2 Results5.2.1. Healthiness Perceptions. We performed amixed-model analysis on the dependent variable ofhealthiness perceptions, which was a composite ofthe fourmeasures (α = 0.97; see Table 3a for themeans).The model included food quantity as a within-subjectfactor, food type as a between-subjects factor, andtheir interaction. There was a main effect of food type[F(1, 178) = 616.40, p < 0.001, ηp

2 = 0.78], a main effectof food quantity [F(2, 356) = 4.31, p = 0.014, ηp2 = 0.024],and no interaction [F(2, 356) = 0.15, p = 0.860, ηp2 = 0.001].We also conducted a mixed Bayesian ANOVA withfood quantity as a within-subject factor, food type asa between-subjects factor, and their interaction. Thisrevealed extreme evidence for food type having an effect(BF10> 100), anecdotal evidence for food quantity havingan effect (BF10 = 1.20), and strong evidence for the maineffects model over the interaction model by a Bayesfactor of 23.00. We then performed planned contrasts.As noted earlier, therewas amain effect of food typewithhigher healthiness evaluations for the healthier foods[M = 7.21 versusM = 2.64; F(1, 178) = 616.40, p < 0.001,ηp2 = 0.78]. Providing evidence of the secondary nature

of quantity, LSD contrasts indicated that there weresome statistically significant but small differences asa function of food quantity: there was a significant dif-ference between the small (Msmall = 5.11) and mediumportion (Mmedium = 4.81; p= 0.020, d= 0.11) and betweenthe small and large portion (Mlarge = 4.75; p = 0.007,d = 0.13) but not between themediumand large portions(p = 0.660, d = 0.02).6

5.2.2. Caloric Perceptions. A mixed-model analysison caloric perceptions (see Table 3b formeans) revealedamain effect of food type [F(1, 178) = 26.59, p < 0.001, ηp2 =0.13], a main effect of food quantity [F(2, 356) = 24.39,

Table 2b. Study 1 Mean Healthiness Perceptions for Snack, Depending on Type and Quantity

Food type

Food quantity

Omnibus test for quantity effectSize 1 (1/2 serving) Size 2 (1 serving) Size 3 (2 servings)

“Healthy”Almonds 7.47 7.86 7.31 F = 1.60, p = 0.21, ηp

2 = 0.034Carrots 7.96 7.80 8.14 F = 0.37, p = 0.69, ηp

2 = 0.008Wheat Thins 5.88 5.04 5.45 F = 1.77, p = 0.18, ηp

2 = 0.038“Unhealthy”Cookies 2.35 2.14 1.92 F = 0.86, p = 0.43, ηp

2 = 0.019M&Ms 1.82 1.53 1.67 F = 0.66, p = 0.52, ηp

2 = 0.015Cheetos 2.02 2.14 1.74 F = 0.69, p = 0.51, ηp

2 = 0.015

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food PortionsManagement Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS 3361

p < 0.001, ηp2 = 0.127], and no interaction [F(2, 356) = 0.80,

p = 0.450, ηp2 = 0.004]. In terms of the main effects, caloric

perceptions were lower for the healthier food type thanthe less healthy food type (M = 4.23 versus M = 5.39;p < 0.001, d=0.77). For portion size, therewas a significantdifference between the small (Msmall = 4.16) and mediumportions (Mmedium = 4.85; p < 0.001, d = 0.33), themedium and large portions (Mlarge = 5.42; p = 0.001, d =0.26), and the small and large portions (p < 0.001,d = 0.59).

Of note, although we cannot say whether type orquantity is factored in to a greater extent for caloricperceptions (nor is that the focus of our primary–secondary theory, which is centered on healthinessperceptions), we can examine how caloric perceptionsmap onto size perceptions (i.e., compare Table 2afrom study 1 with Table 3b in study 2; see Figure 2,which helps to visualize this profile mapping). If sizeperceptions were completely incorporated into caloricperceptions on a 1:1 basis, then the caloric perceptionsshould map 1:1 onto the size perceptions data col-lected in study 1, because both are on similar 1 to 9scales. We performed a profile analysis to test thisproposition. Specifically, we tested whether the sizeperceptions and caloric perceptions lines in Figure 2are nonparallel via a profile analysis to gauge anyinteraction between actual size and type of measure.

Because there was no three-way interaction with typeof food (healthy, unhealthy) [F(2, 722) = 1.76, p = 0.172,ηp2 = 0.005], we collapsed across food type to conduct

a profile analysis with food quantity as a within-subject factor (small, medium, large) and type ofdependent measure as a between-subjects factor (sizeperceptions, caloric perceptions8). This analysis revealedno main effect of type of measure on ratings [F(1, 363) =1.66, p= 0.198, ηp

2 = 0.005], but supporting our propositionthat size perceptions are not incorporated on a 1:1 basisby caloric perceptions, there was a significant interaction[F(2, 726) = 36.70, p < 0.001, ηp2 = 0.092] such that theprofile of size perceptions was significantly differentthan the profile of caloric perceptions. Overall, the profileanalysis is consistent with quantity being a secondarydimension.

5.2.3. Caloric Estimates. Finally, we also analyzedcaloric estimates (see Table 3c for means). A mixed-model analysis on caloric estimates with food quantityas a within-subject factor, food type as a between-subjects factor, and their interaction indicated a maineffect of food type [F(1, 178) = 17.68, p < 0.001, ηp2 =0.09], a main effect of food quantity [F(2, 356) = 23.79,p < 0.001, ηp

2 = 0.12], and an interaction [F(2, 356) = 7.33,p = 0.001, ηp

2 = 0.04]. We did not predict this interactiona priori, but given the interaction, we examined the

Table 3a. Study 2 Mean Healthiness Perceptions for Food Portions, Depending on Type and Quantity

Food type

Food quantity

Omnibus test for quantity effectSize 1 (1/2 serving) Size 2 (1 serving) Size 3 (2 servings)

“Healthy”Almonds 8.07 7.84 7.07 F = 5.09, p = 0.008, ηp

2 = 0.105Carrots 8.06 8.22 8.04 F = 1.64, p = 0.200, ηp

2 = 0.036Wheat Thins 5.93 5.69 6.00 F = 0.24, p = 0.789, ηp

2 = 0.005“Unhealthy”Cookies 3.53 2.68 2.38 F = 3.79, p = 0.027, ηp2 = 0.080M&Ms 2.35 2.14 2.62 F = 1.02, p = 0.366, ηp

2 = 0.023Cheetos 2.92 2.85 2.38 F = 1.22, p = 0.301, ηp

2 = 0.027

Table 3b. Study 2 Mean Caloric Perceptions for Food Portions, Depending on Type and Quantity

Food type

Food quantity

Omnibus test for quantity effectSize 1 (1/2 serving) Size 2 (1 serving) Size 3 (2 servings)

“Healthy”Almonds 3.93 4.97 5.97 F = 7.16, p = 0.001, ηp

2 = 0.141Carrots 2.71 2.76 2.80 F = 0.02, p = 0.982, ηp

2 < 0.001Wheat Thins 4.48 4.97 5.43 F = 1.88, p = 0.159, ηp2 = 0.041

“Unhealthy”Cookies 4.23 5.11 6.63 F = 18.10, p < 0.001, ηp

2 = 0.294M&Ms 4.97 5.72 5.93 F = 2.05, p = 0.135, ηp

2 = 0.045Cheetos 4.66 5.52 5.77 F = 2.19, p = 0.118, ηp2 = 0.048

Note. Caloric perceptions seem to be less sensitive to the food quantity differences than are the size perceptions (see Table 2a from study 1),suggesting an underweighting of size perceptions for caloric perceptions.

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food Portions3362 Management Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS

effect of food quantity on caloric estimates sepa-rately for the healthier food types and the less healthyfood types. For the healthier food types there was nosignificant difference between caloric estimates for thesmall (Msmall = 126.47) andmediumportions [Mmedium=150.11; t(89) = 1.34, p = 0.185, d = 0.18] or between themedium and large portions [Mlarge = 158.73; t(89) =0.54, p = 0.591, d = 0.07], but there was a significantdifference between the small and large portions [t(89) =2.02, p = 0.046, d = 0.28]. For the less healthy food types,there was a significant difference between caloric esti-mates for all portion sizes: small (Msmall = 157.81) versusmedium [Mmedium = 205.02; t(89) = 4.87, p < 0.001, d =0.39], medium versus large [Mlarge = 266.46; t(89) = 4.53,p< 0.001, d = 0.44], and small versus large [t(89) = 8.83, p<0.001, d = 0.77]. The interaction seems to suggest aheightened sensitivity to quantity when assessing caloriesfor unhealthy food types. Interestingly, we observe that

the numeric caloric estimates generally exhibit much lessthan a doubling relationship from 1/2 serving to 1serving and from 1 serving to 2 servings, suggestingan underweighting of quantity in caloric estimates. Ofcourse, this could be due in part to not perceiving adoubling in size relationship, but the profile analysiscomparing caloric perceptions with size perceptionssuggests that this does not explain the entire numericcaloric estimate results.

5.3. Study 2 DiscussionStudy 2 used several measures of health impact thatdiffered from those examined in study 1. First, usinga healthiness perception measure focusing clearly oneating (which we expected could make quantity moresalient), we found that quantity insensitivity wassomewhat attenuated but that quantity was still sec-ondary to food type. Second, we also examined caloric

Figure 2. Perceived Size (Study 1), Caloric Content (Study 2), and Healthiness–Eating (Study 2), as a Function of Actual Size(i.e., Serving Size)

1

2

3

4

5

6

7

8

9

1/2 serving 1 serving 2 servings

Rat

ings

Actual Serving Sizes

Size for healthy type (S1) Size for unhealthy type (S1)Calories for healthy type (S2) Calories for unhealthy type (S2)Healthiness, emph. eating for healthy type (S2) Healthiness, emph. eating for unhealthy type (S2)

Note. Error bars denote standard errors of the mean.

Table 3c. Study 2 Mean Caloric Estimate for Food Portions, Depending on Type and Quantity

Food type

Food quantity

Omnibus test for quantity effectSize 1 (1/2 serving) → Size 2 (1 serving) → Size 3 (2 servings)

“Healthy”Almonds 154.61 ×1.19 183.91 ×1.05 193.00 F = 0.46, p = 0.632, ηp

2 = 0.011Carrots 75.58 ×1.10 83.45 ×1.09 91.00 F = 0.26, p = 0.774, ηp

2 = 0.006Wheat thins 151.94 ×1.18 179.48 ×1.07 192.20 F = 1.26, p = 0.288, ηp

2 = 0.028“Unhealthy”Cookies 156.40 ×1.21 189.67 ×1.61 306.10 F = 10.06, p < 0.001, ηp

2 = 0.188M&Ms 155.16 ×1.42 220.96 ×1.24 275.07 F = 5.90, p = 0.004, ηp

2 = 0.119Cheetos 162.10 ×1.26 204.97 ×1.06 218.20 F = 1.51, p = 0.227, ηp

2 = 0.034

Notes. When examining numeric caloric estimates (i.e., a ratio variable), the size perceptions from Table 2a in study 1 are not readily comparable(because size perceptions were an interval variable). We do observe that the numeric caloric estimates generally exhibit much less than a ×2relationship from1/2 serving to 1 serving and from1 serving to 2 servings, particularly for the healthier food types, indicating anunderweighting ofsize perceptions for caloric perceptions. Although this could potentially be in part because people do not perceive a ×2 relationship in size, theconsiderable size differences in Table 2a from study 1 suggest that size perceptions may not fully account for these caloric estimates.

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food PortionsManagement Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS 3363

perceptions and caloric estimates. Although ourprimary–secondary account mainly applies to health-iness evaluations, the notion that quantity is secondaryin healthiness perceptions suggests that caloric judg-ments also may underweight quantity. Thus, we alsoincluded caloric judgments in this study, and we foundthat the caloric perceptions and caloric estimates bothseemed to underweight food quantity (and that thecaloric perceptions did not seem to factor in size per-ceptions on a 1:1 basis). We also found an interactionbetween food type and food quantity in caloric esti-mates, such that caloric estimates for unhealthy foodtypes were more sensitive to quantity than caloric es-timates for healthier food types. Future research mayfollow up on this finding to determine whether it con-sistently occurs and why.

6. Studies 3a–3c: Increasing the Salienceof Food Quantity

In studies 3a–3c, we used three different ways of in-creasing the salience of the quantity dimension to testwhether consumers adjust their healthiness percep-tions in response to food quantity differences and, if so,to what extent.

In studies 3a and 3b, we return to utilizing healthi-ness measures like those in study 1, because they areboth a prevalent way of gauging healthiness in theliterature (Wilcox et al. 2009, Irmak et al. 2011, Liu et al.2015) and also a main way in which consumers thinkabout the healthiness of their food choices. In doing so,we tested for further evidence that quantity is a sec-ondary dimension (i.e., factored in when its salienceis increased but still to a lesser extent than type isfactored in). We also tested a practically relevant in-tervention (study 3a) and whether a latitude of accep-tance exists for incorporating quantity (study 3b). Insum, using two different ways of increasing the salienceof the quantity dimension and holding constant thehealthiness dependentmeasure for all participantswithina given study, studies 3a and 3b tested whether con-sumers adjusted their healthiness perceptions in responseto food quantity differences and, if so, to what extent.

Then, in study 3c, we increased the salience of quantityby utilizing a very specific evaluation focused only onweight impact, without mentioning the word “health”to participants. Specifically, we gauged perceptions ofweight impact from consuming a given foodportion onceper day. Because weight impact perceptions are likelystrongly linked to caloric perceptions (which show someincreased quantity salience, per study 2), and because theweight impact effects of consuming given food quantitiespresumably accumulate over time, we expected thatquantity sensitivity would emerge in weight impactperceptions. However, we also expected that, likecaloric perceptions in study 2, even perceptions of weightimpact over time might under-incorporate quantity

differences relative to consumers’ size perceptions, pro-viding additional evidence of underweighting of quantity.

6.1. Study 3a: Increasing Salience of Quantity viaJoint (vs. Separate) Evaluation Mode

In study 3a, we utilize a choice paradigm comparingseparate versus joint evaluation to test our predictionbased on past work that quantity insensitivity would beevident in a separate evaluation context (as in study 1) butreduced in a joint evaluation context (Hsee 1996).

6.1.1. Study 3a Method. A total of 257 participantsrecruited from MTurk (39% female, two did not pro-vide gender information; Mage = 33.6 years) partici-pated. The procedure involved the visual diary methodused in studies 1 and 2, with the modification thatparticipants were told that the person using the diaryhas been trying to “lose weight” (versus “improve theirdiet” in the prior studies). This modification was madebecause in a weight loss context, managing (and inparticular decreasing) caloric intake is commonly ac-cepted as the health goal, such that larger portions ofcalorically dense food would clearly be consideredcounterproductive to a health goal. Participants wererandomly assigned to one of two food types (a healthyfood type: almonds; or an unhealthy food type: M&Ms),as well as one of the following three evaluation con-ditions: (1) separate evaluation of a small portion(one serving, according to the manufacturer’s label);(2) separate evaluation of a large portion (two servings);or (3) joint evaluation of both the small and the largeportions. See Figure A.2 in the appendix for stimuli.Under the before and after food diary entry pictures,

participants rated their perceptions of the healthiness ofthat snack episode using the item “Please indicate howhealthy you believe that this snack was” (1 = not at allhealthy, 9 = very healthy). In the joint evaluation condition,participants responded to the healthiness measure forthe first snack (always the smaller portion) and thenfor the second snack, while having both sets of snackpictures in view. The cover story in the joint evaluationcondition was that each snack episode was from a dif-ferent person that they were evaluating.After healthiness assessments were made, partici-

pants were shown the diary entry pictures a secondtime and asked about perceptions of the size of theportion to ensure that participants were able to po-tentially perceive these size differences in both theseparate and joint evaluation conditions. Specifically,following each of the same pictures from before, par-ticipants indicated their portion size perceptions on a9-point scale anchored by “very small” and “very large.”

6.1.2. Study 3a Results. We first examined perceptionsof the size of each snack to check that they were per-ceived as different. Indeed, for both food types and in

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food Portions3364 Management Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS

both evaluation modes, a t-test (independent-samplesin the case of the separate evaluation conditions, paired-samples in the case of the joint evaluation condition) ofsize condition on size perceptions revealed the expectedsignificant result: the larger size was viewed as largerby participants (see means in Table 4a). Thus, any sub-sequent effects do not merely reflect an inability toperceive the size differences.

To first look for evidence of quantity insensitivity inour separate evaluation conditions,we performed a 2× 2ANOVA on health perceptions using food type andfood quantity as factors (see means in Table 4b). Asexpected, there was a large main effect of food type,with higher healthiness evaluations for the almonds [M =7.57 versus M = 1.64; F(1, 169) = 958.47, p < 0.001, ηp

2 =0.850].More importantly, as predicted, therewas nomaineffect of size on healthiness evaluations [F(1, 169) = 1.51,p = 0.220, ηp2 = 0.01], replicating the quantity insensi-tivity observed in study 1. Specifically, there wasno significant difference between the small and largeportion of the almonds [Msmall = 7.73 versusMlarge = 7.42;t(85) = 1.07, p = 0.290, d = 0.23] or for theM&Ms [Msmall =1.72 versusMlarge = 1.56, t(84) = 0.65, p = 0.517, d = 0.14].The interaction was also not significant [F(1, 169) = 0.15,p = 0.704, ηp

2 = 0.001]. To confirm these results andprovide more support for the null effect of portion sizecondition, we also conducted a Bayesian 2 × 2 ANOVA.This revealed extreme evidence for food type having aneffect (BF10> 100), moderate evidence for portion sizecondition having a null effect (BF01= 5.29; BF10 = 0.189),and moderate evidence for the main effects model overthe interaction model by a Bayes factor of 3.92.

We then examined healthiness perceptions in thejoint evaluation conditions, which increased the sa-lience of food quantity (see means in Table 4b). A 2 × 2mixed ANOVA using food type as a between-subjects

factor and food quantity as a within-subject factorrevealed a very large main effect of food type [F(1, 82) =158.58, p < 0.001, ηp2 = 0.66] and the focal predictedrelatively small main effect of food quantity [F(1, 82) =13.37, p< 0.001, ηp2 = 0.14]. For both the almonds [Msmall =7.27 versus Mlarge = 6.66; t(40) = 2.50, p = 0.017, d = 0.36]and the M&Ms [Msmall = 2.58 versusMlarge = 2.19; t(42) =2.96, p = 0.005, d = 0.21], we found that indeed, thejoint evaluation context eliminated quantity insensitivity.There was no interaction [F(1, 82) = 0.61, p = 0.428,ηp2 = 0.01]. A Bayesian 2 × 2 mixed ANOVA confirmed

these results, showing extreme evidence for food typehaving an effect (BF10 > 100), very strong evidence forfood quantity having an effect (BF10 = 51.77), andmoderate evidence for the main effects model overthe interaction model by a Bayes factor of 3.06.Finally, we also analyzed whether the differences

between the findings in the joint and separate evalu-ation conditions were significant for each food type.Following the special t-statistic procedures used byHsee (1996) to compare a between-subjects compari-son with a within-subject comparison, analyses indi-cated that there was a significant difference betweenthe joint and separate evaluations for both the almonds[t(125) = 2.09, p = 0.039] and the M&Ms [t(126) = 2.85,p = 0.005].

6.1.3. Study 3a Discussion. Overall, these results showthat increasing the salience of quantity (via joint evalu-ation mode) mitigates quantity insensitivity in healthi-ness evaluations. This finding that quantity indeed can befactored into healthiness assessments when made quitesalient is important because the literature on categori-cal thinking hints that people simply do not factor inquantity. In particular, our finding is consistent with ourproposed account that quantity is a secondary dimension,

Table 4a. Study 3a Mean Size Perceptions for Snack, Depending on Type, Quantity, and Evaluation Mode

Food type Evaluation mode

Food quantity

Test for differencesSize 1 (1 serving) Size 2 (2 servings)

“Healthy” (Almonds) Separate 4.00 5.56 t = 4.09, p < 0.001, d = 0.88Joint 4.41 6.71 t = 7.89, p < 0.001, d = 1.53

“Unhealthy” (M&Ms) Separate 4.79 6.26 t = 3.51, p = 0.001, d = 0.76Joint 5.37 7.07 t = 7.40, p < 0.001, d = 0.71

Table 4b. Study 3a Mean Healthiness Perceptions for Snack, Depending on Type, Quantity, and Evaluation Mode

Food type Evaluation mode

Food quantity

Test for quantity effectSize 1 (1 serving) Size 2 (2 servings)

“Healthy” (Almonds) Separate 7.73 7.42 t = 1.07, p = 0.290, d = 0.23Joint 7.27 6.66 t = 2.50, p = 0.017, d = 0.36

“Unhealthy” (M&Ms) Separate 1.72 1.56 t = 0.65, p = 0.517, d = 0.14Joint 2.58 2.19 t = 2.96, p = 0.005, d = 0.21

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food PortionsManagement Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS 3365

not a nondimension. Of importance, in further support ofour prediction that quantity is a secondary dimension,food quantity continued to play a small role in explainingthe healthiness assessments, even in joint evaluationmode (whereas food type played a big role). Finally, wenote that on the practical side, this study provides apractically implementable intervention for increasingsensitivity to quantities.

6.2. Study 3b: Increasing Salience of Quantity viaExpectancy Disconfirmation

In study 3b, we further test our primary–secondaryaccount by increasing the salience of quantity viadrawing from expectancy disconfirmation theory. Weexamined both healthy (almonds) and unhealthy foods(plain M&Ms) at three different quantities (1, 2, and8 servings, according to the manufacturer). The 1- and2-serving quantities were included as representative oftypical quantities consumed in a single sitting (consistentwith the prior studies),whereas the 8-serving portionwasincluded as a very large portion size, not representativeof typical portion sizes for a single sitting. Based on ex-pectancy disconfirmation (Bettman 1979, Helgeson andBeatty 1987), we suggest that for typical food portions,quantity is not particularly salient but should becomemore salient for an obviously large portion size (e.g., an8-serving portion). In a sense, there may be a latitude ofacceptance where quantities that could be reasonablyconsumed in a single typical sitting do not factor inmuchto healthiness assessments, but that outside this latitudeconsumers become considerably more sensitive.

6.2.1. Study 3b Method. In a before and after visualdiary setup similar to that of studies 1 and 2, 218 un-dergraduate participants (50% female) receiving coursecredit for their participation were randomly assigned toview one of six conditions representing a 2 (food type:a healthy food type [almonds] or an unhealthy foodtype [M&Ms]) × 3 (food quantity: low quantity salience[1 or 2 servings] or high quantity salience [8 servings])between-subjects design; see Figure A.2 in the appendixfor stimuli. Following the same three-item healthinessperceptions measures (α = 0.97) as used in study 1, weagain included portion size assessments (9-point scaleanchored by very small and very large).

6.2.2. Study 3b Results. A 2 × 3 ANOVA revealeda significant interaction between food type and foodquantity [F(2, 212) = 4.59, p = 0.011, ηp2 = 0.04] as wellas main effects for both food type [F(1, 212) = 1210.69,p < 0.001, ηp

2 = 0.85] and food quantity [F(2, 212) = 11.21,p < 0.001, ηp

2 = 0.10] on healthiness perceptions. We alsoconducted a Bayesian 2 × 3 ANOVA, which confirmedthe interaction results and the main effect of food type,but not food quantity: the Bayesian analysis revealedextreme evidence for food type having an effect

(BF10 > 100), moderate evidence for food quantity havinga null effect (BF01 = 4.48; BF10 = 0.223), and marginalevidence for the interaction model over the main modelby a Bayes factor of 2.78.Follow-up analyses (in the form of LSD contrasts,

conducted separately for each food type, because vari-ance was higher for the almonds than the M&Ms) re-vealed the expected insensitivity between healthinessevaluations for typical 1- and 2-serving quantities for bothfood types, consistent with the prior studies (M = 7.07versusM = 7.06, p = 0.978, d = 0.01 for almonds;M = 1.64versusM = 1.65, p = 0.944, d = 0.01 forM&Ms). In the caseof the almonds, the 8-servings portion size (M= 5.86)wasperceived to be significantly less healthy than either oftwo smaller sizes (p’s < 0.001, d’s > 0.80), indicatingthat people do become quantity sensitive once the por-tion size is very large. This same comparison was inthe expected direction, but marginally significant, forthe M&Ms, with a mean healthiness rating of 1.38 in the8-servings portion size condition (p = 0.063, d = 0.44, andp = 0.053, d = 0.51 for comparisonswith 1 and 2 servings,respectively). However, we note that the very lowhealthiness ratings for even the smaller portions ofM&Ms may have left little room for further decreaseswith the very large portion (see Figure 3). As such, thelarger decreases in healthiness perceptions for the al-monds compared with the M&Ms led to the significantinteraction.Finally, analysis of the size perception measure

confirmed again that participants indeed perceived thevarious portion sizes as different, including the 1-servingversus 2-serving portions. For the almonds, serving sizeperceptions ranged from 4.17 to 5.67 to 7.56; for theM&Ms, size perceptions ranged from 4.89 to 6.35 to8.08 for the 1-, 2-, and 8-serving portion sizes, respec-tively. The overall effect of sizewas significant [F(2, 212) =77.81, p < 0.001, ηp

2 = 0.42], as were all paired com-parisons (p’s < 0.001, d’s > 0.98 for almonds; p’s < 0.001,d’s > 0.86 for M&Ms).

6.2.3. Study 3b Discussion. In study 3b we found thatvery large portion sizes mitigate the quantity insensi-tivity observed in study 1. In combination with study3a, two different ways of increasing salience of quantitywith similar healthiness assessments as used in study 1led quantity to be factored into healthiness assessmentsto a small extent (food type consistently played a largerole). Thus, food quantity is not a nondimension. In-stead, these findings provide additional support for ourkey proposition that the type of food serves as a primary(most salient) dimension in healthiness perceptions, andthe quantity of the food serves as a secondary dimensionthat does not immediately impact healthiness perceptionsbut can affect perceptions if made salient. Of note,quantity still remains secondary even when the salienceof quantity is increased in these different ways.

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food Portions3366 Management Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS

6.3. Study 3c: Increasing Salience of Quantity viaa Weight Impact Over Time Measure

In study 3c, we increased the salience of quantity bychanging the measure of healthiness evaluation. Ratherthan examine general healthiness evaluations of a sin-gle snacking occasion, we focused very specifically onperceived weight impact, without mentioning theword “health” to participants. Presumably, weight im-pact is most closely linked to caloric perceptions, whichshould heighten quantity salience because, objectivelyspeaking, quantity should be incorporated on a 1:1 basisinto caloric perceptions. Further, because a single eatingepisode of any kind is unlikely to have considerable“impact,”we included a once-per-day qualifier in study3c, consistent with Oakes (2005), to further allow forany potential quantity effects to manifest, which, ifanything, would magnify any effect of quantity differ-ences on weight impact. In sum, examining this mea-sure of perceived weight impact over time thus let usfurther test our primary–secondary account, in a senseby examining the furthermost bounds of our accountwhen using a measure highly in favor of detectingquantity salience.

Further, we tested our process account that type isprimary over quantity in part because type is processedbefore quantity. Participants completed a supplementalmeasure at the end of study 3c about the order in whichthey would prefer to acquire these pieces of information(food type, food quantity) when evaluating the health-iness of a snack portion. We predicted that the majorityof participants would prefer to acquire food type first.

6.3.1. Study 3c Method. A total of 180 participantsrecruited from MTurk (49% female; Mage = 35.9 years)participated. The same design, visual before and afterfood diary procedure, and stimuli were used as instudies 1 and 2, and a similar sample size was alsoused. The difference was in the measures collected.

First, for each snack episode (shown with a beforeand after food diary photo set), we asked “What willbe the impact on this person’s weight from consumingthe pictured food, if eaten once per day?” (1 = lose a lot,9 = gain a lot).

Then, to begin to test our underlying process, weincluded several supplemental measures at the end ofstudy 3c. Specifically, “When evaluating the healthi-ness of a snack portion, you can acquire informationabout the type of food first or the amount of food first.Which piece of information do you prefer to acquirefirst to make your healthiness evaluation?” ( food type,food quantity) (the order of mentioning type of food oramount of food was randomized both in the questionand in the answer options). Additionally, to address thepossibility that either healthy or unhealthy food typeswere viewed as more satiating, participants wereshown the diary entry pictures a second time and asked,“After eating this portion of food, how long do youbelieve this person will wait before eating again?” (sliderfrom 0 = very little time to 100 = a long time).

6.3.2. Study 3c Results.6.3.2.1. Weight Impact. Given the conceptual connec-tion between caloric perceptions and weight impact,we followed a similar analysis approach as for caloricperceptions in study 2. Specifically, we first performeda mixed-model analysis on weight impact (see Table 5for the means). The model included food quantity asa within-subject factor, food type as a between-subjectsfactor, and their interaction. There was a main effectof food type [F(1, 178) = 96.54, p < 0.001, ηp2 = 0.35],a main effect of food quantity [F(2, 356) = 17.66, p < 0.001,ηp2 = 0.0909], and no interaction [F(2, 356) = 1.08, p = 0.339,

ηp2 = 0.006].In terms of the main effects, weight impact per-

ceptions were lower (i.e., indicating lower expectedweight gain) for the healthy than for the less healthyfood type (M = 4.15 versusM = 5.82; p < 0.001, d = 1.46).For portion size, there was a significant differencebetween the small (Msmall = 4.63) and medium por-tions (Mmedium = 4.94; p = 0.023, d = 0.18), the mediumand large portions (Mlarge = 5.42; p = 0.001, d = 0.27),and the small and large portions (p < 0.001, d = 0.45).As with caloric perceptions in study 2, although we

cannot say whether type or quantity is factored in toa greater extent for weight impact perceptions in thisstudy, we can examine how weight impact perceptionsmap onto size perceptions (compare Table 2a fromstudy 1 with Table 5 in study 3c; see Figure 4, whichhelps to visualize this profile mapping). Although themapping from size perceptions to weight impact is lessstraightforward than for mapping from size perceptionsto caloric impact, we can examine whether size per-ceptions seem to be incorporated into weight impact ona 1:1 basis by comparing weight impact ratings withthe size perceptions from study 1 (both on 1 to 9 scales).Specifically, a profile analysis tested whether weightimpact perceptions are less sensitive to food quantitydifferences than the size perceptions are, suggestingunderweighting of quantity even when it comes to

Figure 3. Study 3b Mean Healthiness Perceptions of Snack,Depending on Type and Quantity

123456789

"Healthy" (Almonds) "Unhealthy" (M&Ms)

Perc

eive

d H

ealth

ines

s

1 serving 2 servings 8 servings

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food PortionsManagement Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS 3367

weight impact perceptions. We thus tested whether thesize perceptions andweight impact perceptions lines inFigure 4 are nonparallel via a profile analysis in whichwe examined whether there is a significant interactionbetween actual size and type of measure. Becausethere was no three-way interaction with type of food(healthy, unhealthy) [F(2, 722) = 1.53, p = 0.217, ηp2 =0.004], we collapsed across food type to conduct aprofile analysis with food quantity as a within-subjectfactor (small, medium, large) and type of dependentmeasure as a between-subjects factor (size perceptions,weight impact perceptions10). This analysis revealeda main effect of type of measure on ratings [F(1, 363) =7.22, p = 0.008, ηp

2 = 0.020], but more importantly,supporting the notion that size perceptions are notincorporated on a 1:1 basis by weight impact per-ceptions, there was a significant interaction [F(2, 726) =77.17, p < 0.001, ηp2 = 0.175]. Specifically, the profile ofsize perceptions was significantly steeper than theprofile of weight impact perceptions. Overall then, thisprofile analysis is also consistent with quantity beinga secondary dimension.

6.3.2.2. Process Underlying Food Type PrioritizationAccount. We then examined the supplemental mea-sures administered at the end of the study. As predicted,regarding preferences about attribute information ac-quisition order, the majority of participants (82.8% [n =149] versus 17.2% [n = 31]; exact binomial test: p < 0.001)preferred to acquire information about food type beforefood quantity, consistent with our account that foodtype (a categorical attribute) will come before foodquantity (a continuous attribute) for combined eval-uative judgments.11

We also examined satiety perceptions. A mixed-model analysis on satiety perceptions with food typeas a between-subjects factor, food quantity as a within-subject factor, and their interaction revealed amain effectof food quantity [Mhealthy = 44.14 versus Munhealthy =35.13;F(1, 178) = 12.35, p= 0.001, ηp2 = 0.065], amain effectof food quantity [Msmall = 32.77 versus Mmedium =40.40 versus Mlarge = 45.58; F(2, 356) = 30.48, p < 0.001,ηp2 = 0.146], and no interaction [F(2, 356) = 0.02, p = 0.985,

ηp2 < 0.001]. The means for each food (collapsing across

portion size) were as follows: almonds (M = 46.18),

Table 5. Study 3c Mean Weight Impact Perceptions for Food Portions, Depending on Type and Quantity

Food type

Food quantity

Omnibus test for quantity effectSize 1 (1/2 serving) Size 2 (1 serving) Size 3 (2 servings)

“Healthy”Almonds 3.72 3.95 5.07 F = 6.76, p = 0.002, ηp

2 = 0.136Carrots 3.55 3.31 3.24 F = 0.27, p = 0.767, ηp

2 = 0.006Wheat Thins 4.30 4.91 5.33 F = 4.69, p = 0.012, ηp

2 = 0.098“Unhealthy”Cookies 5.42 6.00 6.17 F = 2.20, p = 0.117, ηp2 = 0.048M&Ms 5.71 6.14 6.43 F = 1.90, p = 0.156, ηp

2 = 0.041Cheetos 4.97 5.53 6.08 F = 4.79, p = 0.011, ηp

2 = 0.098

Note. Weight impact perceptions seem to be less sensitive to the food quantity differences than are the size perceptions (see Table 2a fromstudy 1), suggesting an underweighting of size perceptions for weight impact perceptions.

Figure 4. Perceived Size (Study 1) and Weight Impact (Study 3c), as a Function of Actual Size (i.e., Serving Size)

1

2

3

4

5

6

7

8

9

1/2 serving 1 serving 2 servings

Rat

ings

Actual Serving Sizes

Size for healthy type (S1) Size for unhealthy type (S1)Weight impact for healthy type (S3c) Weight impact for unhealthy type (S3c)

Note. Error bars denote standard errors of the mean.

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food Portions3368 Management Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS

carrots (M = 42.39), Wheat Thins (M = 43.84), cookies(M = 38.42), M&Ms (M = 33.36), and Cheetos (M =33.60). These analyses suggest that consumers have thebelief that healthy food types are more satiating thanunhealthy food types but also have a similarly, if not justas strong belief, that larger food quantities are moresatiating than smaller food quantities. Thus, perceivedsatiety differences do not readily explain our accountthat food type is prioritized over food quantity inhealthiness evaluations (because satiety seems simi-larly affected by both). However, they might point toone underlying contributing reason why food type isemphasized in healthiness evaluations (i.e., food type isbelieved to considerably affect satiety).

6.3.3. Study 3c Discussion. In study 3c, we found thata weight impact over time measure somewhat atten-uates quantity insensitivity, as expected. Althoughshowing this increased sensitivity to quantity, this weightimpact measure—like the caloric perceptions measureutilized in study 2—also seemed to underweight foodquantity differences when compared against consumers’size perceptions, further supporting our account of thesecondary nature of quantity even when utilizing ameasure seemingly highly in favor of fully detectingquantity differences.12

Finally, supplemental measures collected at the end ofstudy 3c showed thatmost consumers report preferring toacquire information about food type before food quantitywhen evaluating the healthiness of a food portion, pro-viding some initial support for our underlying informa-tionprocessingorder explanation for a primary–secondaryaccount. We examine this underlying processing orderexplanation in greater depth in the next study.

7. Study 4: Extending to a Context withCaloric Information

Study 4 further tests our primary–secondary accountof healthiness perceptions in a context in which par-ticipants were provided with caloric information. Prac-tically, consumers often have access to caloric informa-tion (e.g., from aNutrition Facts Label), and theoretically,examining this context let us further test our primary–secondary account of healthiness perceptions. Addi-tionally, rather than using a visual food diary with beforeand after photos to depict a snack episode, participantswere instead provided with verbal information for bothfood quantity and food type corresponding with a snackepisode.Our account is that food typewill continue toplaya greater role in health perceptions than food quantity.

In study 4, participants evaluated the healthinessof changing from “1/2 cup of peanut M&Ms (400 cal-ories)” to one of three randomly assigned alternatives:(1) “1/2 cup of almonds (400 calories),” (2) “1/4 cupof peanut M&Ms (200 calories),” or (3) “1/4 cup ofalmonds (200 calories).”

Our primary–secondary account proposes that quan-tity differences are secondary to food type differencesin consumers’ healthiness evaluations. Accordingly, ourprimary–secondary account predicts that consumers willview consuming “1/2 cup of almonds (400 calories)” or“1/4 cup of almonds (200 calories)”—which both involvechanging food type from peanut M&Ms to almonds—ashealthier than consuming “1/4 cup of peanut M&Ms(200 calories)”—which involves a quantity change from1/2 cup to 1/4 cup. Of note, prior work by Oakes (2005)indicates that perceived healthiness is more influencedby food type than by caloric content: most applicable tostudy 4, this prior work would thus indicate that con-suming 200 calories of almonds is healthier than 200calories of peanutM&Ms (or that consuming 400 caloriesof almonds is healthier than consuming 400 caloriesof peanut M&Ms). Extending beyond Oakes (2005)though, we additionally predict that consuming“1/2 cup of almonds (400 calories)” will be viewedas a bigger health improvement over “1/2 cup ofpeanut M&Ms (400 calories)” than the improve-ment from consuming “1/4 cup of peanut M&Ms(200 calories).” This test is a highly conservative testof our primary–secondary account, because the explicitcaloric content of the almond portion doubles that of thepeanut M&Ms portion.Finally, study 4 also further tests our underlying

process more formally. To do so, study 4 includes self-report process measures of (1) whether food type orfood quantity factored more into one’s healthiness eval-uations, and (2) the order of processing the food typeand food quantity attributes. These measures wereadapted from Krider et al. (2001), who proposed aprimary–secondary account of area assessments forgeometric figures. Our primary–secondary accountoffers twomain sets of predictions with respect to thesemeasures. First, we predict that the majority of partic-ipants will report that (1) both dimensions are factoredinto their healthiness evaluations, but that food quantityis factored in to a lesser extent, and that (2) they chosea type-then-quantity evaluation strategy.Second, not all consumers may show this tendency;

thus, we test for moderation of our effects using theseprocess measures and predict that our effects will belarger (1) among those consumers who naturally factor infood type more into their healthiness evaluations, and(2) among those consumers who choose a type-then-quantity evaluation strategy over a quantity-then-typeevaluation strategy. Importantly, process by moder-ation based on choice of evaluation strategy (i.e., foodtype first or food quantity first) explicitly argues againstany artifact or interpretation account for healthiness.For example, if consumers simply interpret healthinessto refer to things like vitamin content for which toomuchquantity is not bad within bounds, then that would noteasily explain why our effects would be larger among

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food PortionsManagement Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS 3369

those consumers adopting the type-then-quantity eval-uation strategy.

7.1. MethodA total of 161 participants recruited from MTurk(50% female;Mage = 34.3 years) participated. Participantswere randomly assigned to one of three conditions(healthiness route: healthier food type, healthier foodquantity, or healthier food type and healthier foodquantity).

All participants first read, “Imagine that someoneyou know typically has a 1/2 cup of peanut M&Ms(400 calories) as a daily snack.” Participants then readone of three sentences, depending on random assign-ment to healthiness route conditions. In the healthierfood type condition, participants read, “However, onthis occasion, they have been trying to improve theirdiet, so they instead switch to a 1/2 cup of almonds(400 calories).” In the healthier food quantity condition,participants read, “However, on this occasion, they havebeen trying to improve their diet, so they instead switchto a 1/4 cup of peanut M&Ms (200 calories).” In thehealthier food type and healthier food quantity condi-tion, participants read, “However, on this occasion, theyhave been trying to improve their diet, so they insteadswitch to a 1/4 cup of almonds (200 calories).”Note thatthis information was closely based on actual calorieinformation because peanut M&Ms and almonds havesimilar caloric density.

Participants then answered two questions to tap intoperceptions of how healthy the consumer’s behaviorwas on this occasion: “Howhealthywas this consumer’sbehavior on this occasion?” (1 = not at all healthy, 9 = veryhealthy) and “How much health goal progress has thisconsumermade from this occasion?” (1 = very little healthgoal progress, 9 = a lot of health goal progress).

Participants then completed two measures adaptedfrom study 1 of Krider et al. (2001). To gauge whetherparticipants explicitly recognize they are using aprimary–secondary account with greater weight onfood type, they were first asked, “When you wereevaluating the healthiness of eating [depending on con-dition: 1/2 cup of almonds, 1/4 cup of peanut M&Ms,1/4 cup of almonds], which did you consider more, thetype of food [depending on condition: almonds, peanutM&Ms] or the quantity of food [depending on condition:1/2 cup, 1/4 cup]? Allocate 100 points between foodtype and food quantity to reflect the extent towhich eachdimension played a role in your healthiness evaluation.”Second, as a further test of our underlying type-then-quantity information processing order account, partici-pants were then asked: “When evaluating the healthinessof eating [depending on condition: 1/2 cup of almonds,1/4 cup of peanut M&Ms, 1/4 cup of almonds], whichbetter captures what you did?” (choice options in ran-domized order: I basically looked at the type of food but made

some adjustments for the quantity, I basically looked at thequantity of food butmade some adjustments for the type of food).

7.2. Results7.2.1. Perceptions of Healthiness. As predicted, a3-group (healthiness route: healthier food type, healthierfood quantity, simultaneously healthier food type andfood quantity) one-way ANOVA on healthinessperceptions (α = 0.80) was significant [F(2, 158) = 12.15,p < 0.001, ηp2 = 0.133]. Follow-up LSD contrasts testedour primary–secondary account, which predicts thatconsumers will view consuming “1/2 cup of almonds(400 calories)” or “1/4 cup of almonds (200 calories)”—which both involve changing food type from peanutM&Ms to almonds—as healthier than consuming“1/4 cup of peanut M&Ms (200 calories)”—which in-volves a quantity change from 1/2 cup to 1/4 cup. Indeed,as a highly conservative test of our primary–secondaryaccount, participants perceived consuming a healthierfood type to be healthier than consuming a healthierfood quantity, despite there being a caloric decrease inthe healthier food quantity condition but no caloric de-crease in the healthier food type condition (Mtype = 5.93versusMquantity = 4.86, p = 0.003, d = 0.62). Although lessconservative (and conceptually replicating Oakes 2005),participants also perceived consuming a simultaneouslyhealthier food type and food quantity (Mtype & quantity =6.53) to be highly significantly healthier than consuminga healthier food quantity of the same caloric content(p < 0.001, d = 0.97). Finally, although not a central testfor our account, we found a marginally significant dif-ference between the healthier food type and simulta-neously healthier food type and food quantity conditions(p = 0.073, d = 0.33).

7.2.2. Process Measure: Point Allocation to Food Typevs. Food Quantity. First, we tested our prediction thatthe majority of consumers would naturally allocategreater focus to food type. Although there was a mar-ginally significant effect of healthiness route on pointallocation to food type [F(2, 158) = 2.44, p = 0.090, ηp2 =0.030], more importantly, three one-sample (two-tailed)t-tests showed that point allocation to food type (versusfood quantity, out of 100 points) was greater than 50 ineach condition, consistent with our primary–secondaryaccount for type–quantity [healthier food type:M = 69.52,t(51) = 5.13, p < 0.001; healthier food quantity:M = 60.31,t(47) = 2.56, p = 0.014; simultaneously healthier food typeand food quantity: M = 70.66, t(60) = 7.07, p < 0.001].Second, we tested our process by moderation pre-

diction that the effect of healthiness route on healthinessperceptions would emerge to a greater extent among con-sumers who naturally allocate greater focus to food type.With the caveat that point allocations are left-skewed aspredicted, we conducted a 3 (healthiness route: healthierfood type, healthier food quantity, simultaneously

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food Portions3370 Management Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS

healthier food type and food quantity) × continuouspoint allocation to food type (mean-centered) ANOVAon healthiness evaluations, which revealed a signifi-cant main effect of healthiness route [F(2, 155) = 16.49,p < 0.001], no main effect of point allocation [F(1, 155) =1.89, p = 0.172], and a significant interaction [F(2, 155) =16.54, p < 0.001]. See Figure 5, which shows that thepredicted effect of healthiness route emerges as point al-location to type increases. Indeed, a Johnson-Neyman testfor a multicategorical independent variable (Hayes andMontoya 2017) showed that the effect of healthiness routewas significant for point allocations greater than 51.64(which 69.6% of participants had).13

7.2.3. ProcessMeasure: Choice of Attribute ProcessingOrder. We also examined participants’ choice of evalu-ation strategy, which we predicted would show similarresults but more directly tests our underlying attributeprocessing order explanation. A χ2 test indicated nodifference between conditions [χ2(2) = 2.76, p = 0.252,Cramer’s V = 0.13], so we collapsed across all threeconditions to conduct an exact binomial test, whichshowed that participantswere significantlymore likely toindicate that they followed a “focus on food type first”than a “focus on food quantity first” healthiness evalu-ation approach (70.8%versus 29.2%, p< 0.001), consistentwith our proposed underlying information processingorder explanation for food type serving as a primarydimension. Additionally, with the caveat that only 47 of161 participants selected a “focus on food quantity food”approach, we also conducted a 3 (healthiness route:healthier food type, healthier food quantity, simul-taneously healthier food type and food quantity) × 2(evaluation approach: food type first, food quantityfirst) ANOVAon healthiness evaluations, which revealeda significant main effect of healthiness route [F(2, 155) =6.70, p = 0.002], no main effect of evaluation approach[F(1, 155) = 0.52, p = 0.473], and a significant interac-tion [F(2, 155) = 8.80, p < 0.001]. See Figure 6, which

shows that our predicted key effects emerge among themajority of participants who report a strategy of fo-cusing first on type and adjusting for quantity. A one-way ANOVA on these participants who focus first ontype and adjust for quantity (n = 114) was significant[Mtype = 6.35 versusMquantity = 4.32 versusMtype & quantity =6.68;F(2, 111) = 19.46, p< 0.001,ηp2 = 0.260]. Follow-upLSDtests showed that participants perceived the health-via-quantity route as significantly less healthy thaneither the health-via-type path or the health-via-type-and-quantity path (both p’s < 0.001, d’s = 1.32 and 1.39),and the latter twopathswere not viewed as significantlydifferent (p = 0.377, d = 0.19).14

7.3. DiscussionStudy 4 provided consumers with caloric informationregarding (different) food portions. As noted earlier,the goals literature often refers to calories as a mainobjective measure of health goal impact, given theprevalence of weight loss and weight managementas a primary health goal (Cochran and Tesser 1996,Chandon and Wansink 2007a, Huang et al. 2012,Campbell and Warren 2015). However, we found thatconsumers evaluate the healthiness of a given caloriedifferently in a manner consistent with our primary–secondary account of food type and food quantity.Moreover, using process measures adapted from

Krider et al. (2001), we also found process evidencesupporting this primary–secondary account. First, wefound support for our prediction that the majority ofpeople indeedweight food type over food quantity andalso that the majority of people report choosing anevaluation strategy in which they focus first on foodtype and then adjust subsequently for food quantity.Second, we found moderation evidence that our effectsemerge more strongly among this majority of peoplewho weight food type over food quantity and whofocus first on food type and then adjust subsequentlyfor food quantity. Importantly, the finding that theeffects of healthiness route on healthiness percep-tions emerged among consumers who chose thetype-then-quantity processing order (but not amongconsumerswho chose the quantity-then-type processingorder) provides support for our underlying process-ing order explanation while also arguing against al-ternative accounts based on particular interpretationsof healthiness (e.g., that healthiness is about vitamincontent).

8. Study 5: Choice Evidence of theSecondary Nature of Quantity

In the final study, we go beyond healthiness per-ceptions to examine choice, testing whether the sec-ondary nature of quantity is evidenced in choicebehavior in potentially negative ways. Specifically, wefocus on a choice between portions of two calorically

Figure 5. Study 4 Healthiness Perceptions of DifferentRoutes to Healthiness as a Function of Participants’Allocation to Food Type vs. Food Quantity

Note. Because only 16 people (i.e., < 10% of participants) had scoresfrom 0 to 29, the figure begins at 30.

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food PortionsManagement Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS 3371

dense foods, a healthier food (almonds) and a lesshealthy food (M&Ms), varying the portion size of thealmonds. We predicted that the majority of consumerswould select the healthier food (almonds) over the lesshealthy food (M&Ms), with little effect of varying theportion size of the almonds. Although not central toour theory, we also explored potential downstreamcaloric consumption consequences for later in the day(reported in the online appendix).

8.1. MethodParticipants from a paid university participant pooltook part in this study, which had two parts completedon two sequential days. One-hundred ten participants(72% female,Mage = 23.3 years) completed part 1 for $5each, and of these, 98 (74% female, Mage = 23.1 years)returned the next day to complete part 2 for $10 each. Part1 occurred in the afternoons, and part 2 occurred in themornings the next day (two-part methodology adaptedfrom study 2 of Liu et al. 2015). Both parts took place atcomputer stations, with dividers keeping participantsfrom viewing each other’s consumption.

All participants were told that they would listen toan audio program and have a snack while listening tothe program. To support the cover story, participantschose between two audio programs (actually the sameprogram to facilitate the same listening experience, butlabeled with different names). Participants then madethe focal choice between a healthy- versus unhealthy-by-type snack portion. Specifically, because our aimwas to examine choice implications of having a healthgoal focus revolving around weight control, all par-ticipants were instructed, “We would also like you tochoose a snack. For choosing this snack, we would likeyou to choose like you are actively trying to pursuea health goal consisting of losing or managing yourweight.” They were then asked to “Please select thesnack bowl you want to eat today while you listen tothe audio program” and shown pictures of two snack

bowls. For the unhealthy snack bowl, the picture depicteda portion size of chocolate M&M candies equivalentto 1/2 the suggested serving size according to themanufacturer (21 g) to ensure that this was not a largeamount of candy that would never be chosen. For thehealthy snack bowl, the picture depicted varying portionsizes of almonds (1 serving or 2 servings). Thus, allparticipants were randomly assigned to one of twochoice sets: (1) 1/2 serving M&Ms versus 1 servingalmonds or (2) 1/2 serving M&Ms versus 2 servingsalmonds. See FigureA.3 in the appendix for pictures andcalorie information.Research assistants then served participants their

snack bowls, and participants listened to the approx-imately 10-minute audio program while having theirsnack. Theywere then asked to set aside the snack bowlbefore completing a survey about the audio program andsnack.15

The next day, participants returned and were in-structed to complete a dietary recall of everything theyate and drank the previous day, using a three-stepmultiple-pass recall method in which participantsprovided detailed information about what foods andbeverages they consumedand inwhat amounts (Guentheret al. 1997, Scott et al. 2007). A portion size estimationguide, local restaurant menus, and a sheet with addi-tional detailed food prompts (Scott et al. 2007) wereprovided to help participants recall their consumptionmore accurately. These recalls were subsequently codedusing online calorie databases and restaurant and productwebsites to determine caloric intake for the rest of theday after their part 1 snack session.16 Finally, partici-pants completed a final sur-vey including their demo-graphic information.

8.2. ResultsOur main outcome was choice shares. To test whetherthe portion size of almonds affected choice shares,we used a logistic regression with portion size condition

Figure 6. Study 4 Healthiness Perceptions of Different Routes to Healthiness as a Function of Participants’ Choice ofEvaluation Strategy

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food Portions3372 Management Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS

as a categorical predictor and choice of almonds orM&Ms as the outcome variable. The majority ofparticipants (72.7%) indicated that they would selectalmonds over M&Ms, providing evidence of foodtype as the primary attribute that dominates (exactbinomial test: p < 0.001).

Further, consistent with quantity as a secondarydimension, there was no significant overall effect ofportion size condition on choice of almonds overM&Ms [choice set with 1 serving almonds: 75.9% chosealmonds; versus choice set with 2 servings almonds:69.6% chose almonds; Wald χ2 (1) = 0.55, p = 0.460].Cramer’s V for the effect of portion size condition onchoice was 0.071. To confirm these results and providemore support for the hypothesis that relative foodquantity does not affect choice, we also conducteda Bayesian test, which revealed moderate evidence forchoice set condition having a null effect (BF01 = 3.67;BF10 = 0.273).

These choices meant that the average calories selectedsignificantly increased as the almond portion size in-creased, from 148 to 257 calories for the 1- and 2-portionsize conditions, respectively [t(108) = 7.60, p < 0.001].Thus, quantity insensitivity is further evidenced inchoices, because people given an explicit goal of makinghealthy choices for weight loss or management chosethe healthy-by-type option regardless of the relativeportion sizes (and thus caloric content) of the moreand less healthy-by-type options.

8.3. DiscussionStudy 5 provides further evidence of the secondarynature of quantity, showing that it manifests in foodchoices such that those given an explicit health goalcentered on weight loss or management chose a health-ier food type snack over a less healthy food type snack,regardless of their relative portion sizes. Further, thedisregard of portion size had consequences for thecaloric content of the snacks individuals chose. Althoughit is possible that greater consumption of almonds couldstill lead to a healthy overall diet, assuming this replacedother consumption later in the day, this remains a ques-tion for future work.

9. General DiscussionOverall, seven studies conducted across a wide ran-ge of contexts provided support for our primary–secondary account of food type and quantity as thebasis for healthiness perceptions of food portions (seeTable 1 for summary and Table A.1 in the appendixfor a summary of dependent variables and processmeasures). First, study 1 provided basic evidence forour primary–secondary account, showing that with acommon healthiness perceptions measure utilized inthe literature, food type strongly affects healthinessperceptions, whereas food quantity does not. Study 2

then examined different health impact measures.Most importantly, study 2 showed that the primary–secondary account is clearly evidenced with an alter-native healthiness perception measure centered on theact of eating (which should further allow any quantityeffect to manifest yet not be overly biased towardeither type or quantity). Additionally, although lesscentral to testing our account but valuable for expandingunderstanding to different health impact measures,we also examined caloric perceptions and estimates.These measures still seemed less sensitive to foodquantity differences than consumers’ size perceptionsare, further suggesting an underweighting of quantityconsistent with our account of quantity as a secondarydimension.Studies 3a and 3b then returned to the healthiness

perception measure used in study 1 to show that twodifferent ways of increasing salience of quantity withoutchanging the healthiness measure led quantity to befactored into healthiness assessments to a small extent(food type consistently played a large role). Thus, thesestudies showed that food quantity is not a non-dimension; rather they provide additional support forour key proposition that the type of food serves as aprimary (most salient) dimension in healthiness per-ceptions and that the quantity of the food serves as asecondary dimension that does not immediately affecthealthiness perceptions but can affect perceptions ifmade salient. Study 3c then used a third way of in-creasing the salience of quantity, by changing the healthimpact measure to one very specifically focused onweight impact over time, without any mention of theword “health.” Thismeasure ofweight impact over timedid show increased salience to quantity, as expected;however, like caloric perceptions, weight impactremained less sensitive to food quantity differencesthan consumers’ size perceptions, again suggestingsome underweighting of quantity consistent withour account of quantity as a secondary dimension.Study 4 then further tested our primary–secondary

account of healthiness perceptions by examininghealthiness perceptions when caloric informationabout food portions is explicitly provided, showingthat consumers still seem more sensitive to changesin food type than food quantity. Further, study 4provided evidence of our underlying attribute pro-cessing order explanation, including showing processby moderation.Finally, study 5 provided choice-based evidence of

our primary–secondary account, because consumersinstructed to choose with a health goal consisting ofweight loss or management chose a healthier typeof snack over a less healthy type of snack, regardlessof the quantity (and thus calories), showing that choicesalso reflect the secondary nature of quantity.

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food PortionsManagement Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS 3373

9.1. Theoretical and Practical ContributionsOur research offers two main contributions. The firstcontribution is to the health goals literature, in whichcalories are a commonly used measure of healthgoal impact (Cochran and Tesser 1996, Chandon andWansink 2007a, Huang et al. 2012, Campbell andWarren 2015). We show that subjective health im-pact reflects the primacy of food type over quantity,even when calorie information is provided and heldconstant. Indeed, althoughwhatwe consume and howmuchwe consume are understood to jointly influencecaloric intake (Chandon and Wansink 2012), prior re-search has not examined how systematically changingthe type versus quantity of consumption differs inperceived goal impact. Rather, these two broad classesof goal means are often treated as interchangeableways of decreasing one’s caloric intake. In the healthdomain, this tendency is reflected by studies that sep-arately examine what we consume and how much weconsume (Shiv and Fedorikhin 1999, Khan and Dhar2006, Sharpe et al. 2008, Laran 2010, Dubois et al. 2012,Wansink 2012, Haws and Winterich 2013, Cornil andChandon 2016). Given the importance of perceptionsof health goal progress as a construct, future goalsresearch may build on our differentiation betweenchanging food type versus quantity to examine anarray of questions. For instance, future work couldtest whether decreasing calories via changing type(versus quantity) is more motivating (Etkin and Ratner2012) or leads to more long-term goal commitment(Duckworth et al. 2007, Woolley and Fishbach 2016)and, if so, whether perceptions of health goal progressmediate such effects.

Relatedly, our second major contribution is to thefood decision-making literature, because two choiceparadigms for gauging the healthiness of consumers’food choices predominate: choice between food types(Dhar and Simonson 1999, Shiv and Fedorikhin 1999,Kivetz and Zheng 2006, Laran 2010, Gal and Liu 2011,Liu et al. 2015) or choice between food quantities(Sharpe et al. 2008, Dubois et al. 2012, Wansink 2012,Haws and Winterich 2013, Cornil and Chandon 2016).Our research shows that consumers perceive choicebetween food types to involve greater healthinessdifferences than choice between food quantities, despiteboth paradigms being used to gauge the healthinessof consumers’ choices.

Our research also offers some practical implicationsfor consumer health. First, the secondary nature ofquantity and the occurrence of quantity insensitivitymay be problematic in several ways. Although con-suming a larger portion of some healthier food types(e.g., carrots) may be just as good, if not better, thanconsuming a smaller portion of that food, it is hard tomake that claim for many junk foods (e.g., candies) orcalorically dense “healthy” foods (e.g., granola, nuts). In

our research, however, peoples’ healthiness evalua-tions were often quantity insensitive or seemed tounderweight perceived quantity differences for bothfood typeswithin a considerable range (1/2 to 2 servingsin our studies), suggesting a latitude of acceptance. Thesefindings suggest that an individual who chooses toconsume a larger portion of junk food would not codethat as a less healthy snack episode, provided thatthe portion is plausibly within the range of typicallyconsumed portion sizes for a single sitting. Further,study 4 indicates that providing caloric informationstill would not lead consumers to factor in quantityto the same extent that they factor in food type. How-ever, one potential intervention may be offered bystudy 3a, which suggests that facilitating joint eval-uation of different food quantities may heightensensitivity.Second, as study 5 shows, individuals with health

goals consisting of weight loss or management maychoose a portion of a healthy but calorically densefood type well past the point at which the caloriccontent exceeds a small portion of an unhealthy foodtype. Of course, this choice is not necessarily bad ifthe health-seeking person adjusts later consumption(e.g., from feeling fuller from the large portion or fromcorrectly estimating caloric content). However, peopleoften have difficulty adjusting later consumption epi-sodes to compensate for prior eating unless directed toconsider daily consumption guidelines (Roberto et al.2010), and people also seek ways to justify indulgentconsumption (Mukhopadhyay and Johar 2009, Hubertset al. 2014).Third, the primacy of type seems related to an

avoidance mindset. Research has documented backfireeffects of dietary approaches based on completeavoidance of unhealthy-by-type foods (McFarlane et al.1999, Urbszat et al. 2002, Liu et al. 2015, David andHaws 2016). Accordingly, retraining people to eval-uate healthiness on the basis of both type and quan-tity may lead to more sustainable and healthierconsumption patterns, less likely to result in a “whatthe hell effect” backfire pattern (Cochran and Tesser1996) when small portions of unhealthy foods areconsumed. This remains a question for future work,which could randomly assign people to a healthi-ness strategy (either choosing healthier food typesor smaller food quantities) and measure downstreamconsequences (rather than examine downstream con-sumption after consumers self-select a particularstrategy).Finally, our research also offers some practical im-

plications for understanding which health goal means(choosing a healthier food type versus choosing asmaller portion of an unhealthy food type) consumerswill assess as more effective when their main goal is tomake healthier food decisions. Our results show that

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food Portions3374 Management Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS

choosing a healthier type is viewed as more effective,which is at odds with recent work showing the key roleand malleability of consumption quantities and norms(Wansink 2004). Future work could explore ways tohelp consumers appreciate the importance of foodquantity (e.g., always creating their own portions toforce a quantity decision versus mindlessly adoptingprepackaged portions). Our findings also have impli-cations in a social context, based onwhether consumerswant to communicate more or less goal progress toothers. For instance, obese consumers are often stig-matized for unhealthy behavior (Puhl and Heuer 2010)and thus may want to communicate greater health goalprogress in their food choices. Our research indicatesthat these consumers may benefit from pursuing healthvia healthier food types while avoiding even smallportions of indulgent foods owing to the stronger impliedsignaling. Finally, from a food industry perspective, ourfindings indicate that when targeting health-concernedindividuals, offering smaller portions of unhealthyfoods (e.g., M&Ms) may not necessarily make themseem sufficiently impactful on health compared withlarger portions of healthy but calorically dense foods(e.g., almonds), even when caloric information is pro-vided. Perhaps consumers may be more likely tochoose small portions of unhealthy foods if they areinformed that such portions will not necessarily nega-tively affect their health goals and if it is emphasized thatsmall portions of such foods may be more rewarding inother ways (e.g., taste) that can make sticking to a targetcaloric goal easier.

9.2. Potential Future Extension to Other DomainsInvolving Type and Quantity

Although our research focused on an attribute anddomain with considerable importance to consumers,organizations, and policy makers, namely healthinessassessments of food portions, future research may ex-tend our primary–secondary account to other attributesand domains. Our theory is that when both type (i.e.,“what”) and quantity (i.e., “how much”) should, ob-jectively speaking, jointly affect an attribute’s value (inour case, healthiness assessments), quantity will tend tobe neglected in assessments of “portions” within a con-siderable range unless otherwise made salient and, evenwhen made salient, assessments are still likely to reflectthe secondary nature of quantity.

One potential extension is to the financial domain, inwhich—like the health domain—there exists a commonnumeric indicator of total financial impact or savingsgoal impact (i.e., dollars) (Cheema and Bagchi 2011,Huang et al. 2012, Campbell and Warren 2015). In this

domain, total spending depends jointly on “what” ispurchased (e.g., what tier of product brands) and“how much” is purchased (e.g., the size or quantityof products). Our theory predicts that evaluations offinancial impact will be more influenced by the type ofthe purchased products than the quantity of products.For example, luxury goodsmay havemore of a judgedfinancial impact than nonluxury goods, even if thegoods are equivalent in price. Indeed, analogous tothe food domain, in which people display highly cat-egorical thinking for the type of food, the brandsliterature suggests that consumers also display highlycategorical thinking for the type of product. Brandedproducts are categorized quickly by consumers (ThomaandWilliams 2013) and readily categorized as “higher-end” or “lower-end” (Fang and Mishra 2002). Con-sumers also begin at an early age to categorize brandedproducts (Ross and Harradine 2004), and even whenbrands do not meaningfully differ, consumers still try tomeaningfully differentiate and categorize them (Carpenteret al. 1994). Our theory then predicts that perceived fi-nancial impact will be greater for an equivalent mone-tary savings obtained via a change in product brandtier versus a quantity change.

9.3. ConclusionIn closing, we provide substantial evidence that al-though both the types and quantities of foods eatenjointly contribute to weight and overall health, con-sumers treat type as a primary dimension and quantityas a secondary dimension. Accordingly, a food’s type(versus quantity) has greater effect on perceived healthimpact, across multiple study designs and stimuli,and even when controlling for caloric content orproviding caloric information. Given that healthinessperceptions are an important input to food choice, thetendency towardneglecting or underweighting quantitymay have negative effects on those pursuing healthgoals, especially weight loss goals. We hope this re-search will lead individual decisionmakers, researchers,food industry representatives, and public policy makersto become aware of consumers’ tendencies, with theultimate aim of improving decisionmaking andwelfare.Additionally, we offer suggestions for future researchto extend our primary–secondary account to othercontexts in which both “what” and “howmuch” oughtto (yet may not) jointly affect assessments.

AcknowledgmentsThe authors thank their respective institutions for supportingthis research. The authors especially thank the research as-sistants and lab manager at the Fuqua School of Businessbehavioral lab for assistance with study 5.

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food PortionsManagement Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS 3375

Appendix. Study Stimuli

Notes. Because study 3c and the online appendix studywere conducted in 2017, the date stamps on these visual food diary entries were changedto 2017 (from 2015) for these studies. All other aspects of the food diaries were the same.

Figure A.1. (Color online) Studies 1, 2, 3c, and Online Appendix Study: Visual Food Diary Entries.

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food Portions3376 Management Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS

Figure A.2. (Color online) Study 3b (and 3a): Visual Food Diary Entries

Note. The 1- and 2-serving stimuli shown here were also used in study 3a, either individually in the separate evaluation conditions (i.e., just the1-serving or just the 2-serving size) or jointly in the joint evaluation condition.

Figure A.3. (Color online) Study 5: Portion Sizes of M&Ms andAlmonds and Calorie Information: Snack Bowl Choice Options

Notes. Participants did not see the serving size or calorie information in study 5. Calorie counts from www.fatsecret.com.

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food PortionsManagement Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS 3377

Endnotes1The only exceptions that we are aware of are research that examineschoice among “vice-virtue bundles” that simultaneously vary bothtype and quantity within a single product (Liu et al. 2015, Haws andLiu 2016a) and research that examines the effect of pricing and calorielabeling on choice from a menu with choice options varying in bothtype and quantity (Haws and Liu 2016b). In contrast to this priorwork, the present research systematically examines both of these twogeneral strategies for pursuing a healthy food consumption goal,showing that they differ in their impact on subjective healthiness (orhealth goal impact), an important construct in the goals literature.

2We thank an anonymous reviewer for this natural classificationexample.3Of course, healthiness is a complex assessment with multipleaspects, and although calories often emerge a major standard,healthiness is a subjective assessment. As such, our researchexamines healthiness perceptions as a primary focus (not caloricperceptions or estimates). However, because calories are com-monly treated as an objective indicator of health impact in thegoals literature, and because of the widespread availability andprominence of caloric information (e.g., on Nutrition Facts La-bels, on restaurant menus), we also examine calories as a sec-ondary focus.

Table A.1. Dependent Variables and Process Measures Used Across Studies

Study Dependent variables and process measures

1 • Participants viewed before and after visual food diary entry logs and answered three items adapted from Irmak et al. (2011) foreach before and after entry:◦ Please indicate how healthy you believe that this snack was (1 = not at all healthy, 9 = very healthy)◦ Please indicate how nutritious you believe that this snack was (1 = not at all nutritious, 9 = very nutritious)

• How well would this snack fit within this person’s overall diet? (1 = not at all, 9 = very much so)

2 • Participants viewed before and after visual food diary entry logs and answered four items for each before and after entry:◦ Was eating this snack (1 = not at all healthy, 9 = very healthy)◦ Was eating this snack (1 = not at all nutritious, 9 = very nutritious)◦ Was eating this snack (1 = a bad thing, 9 = a good thing)◦ Did eating this snack fit within this person’s overall diet? (1 = not at all, 9 = very much so)

• Participants also answered two questions regarding perceptions of caloric content:◦ How many calories do you think this snack has? (1 = very few calories, 9 = a lot of calories)◦ How many calories do you estimate are in this snack? (numeric free-response)

3a • Participants viewed before and after visual food diary entry logs and answered one item:• Please indicate how healthy you believe that this snack was (1 = not at all healthy, 9 = very healthy)

3b • Participants viewed before and after visual food diary entry logs and answered three items adapted from Irmak et al. (2011):◦ Please indicate how healthy you believe that this snack was (1 = not at all healthy, 9 = very healthy)◦ Please indicate how nutritious you believe that this snack was (1 = not at all nutritious, 9 = very nutritious)◦ How well would this snack fit within this person’s overall diet? (1 = not at all, 9 = very much so)

3c • Participants viewed before and after visual food diary entry logs and answered one item for each before and after entry:◦What will be the impact on this person’s weight from consuming the pictured food, if eaten once per day? (1 = lose a lot, 9 =gain a lot)

• Participants then answered one item for each before and after entry to gauge satiety perceptions:◦After eating this portion of food, how long do you believe this personwill wait before eating again? (Slider bar from 1 to 100)

• Participants then answered a question gauging our type-then-quantity information processing ordering explanation:◦When evaluating the healthiness of a snack portion, you can acquire information about the (order randomized: “type of foodor the amount of food first” or “amount of food or the type of food first”). Which piece of information do you prefer toacquire first to make your healthiness evaluation? (order randomized: food type, food quantity)

4 • Participants viewed verbal food portion information and caloric information (e.g., 1/4 cup of peanut M&Ms [200 calories]) andanswered two items:◦ How healthy was this consumer’s behavior on this occasion? (1 = not at all healthy, 9 = very healthy)◦ How much health goal progress has this consumer made from this occasion? (1 = very little health goal progress, 9 = a lot ofhealth goal progress)

• Participants completed an item gauging whether people self-report a primary–secondary account, adapted from Krider et al.(2001):◦Allocate 100 points between food type and food quantity to reflect the extent to which each dimension played a role in yourhealthiness evaluation

• Participants completed an item gauging our type-then-quantity information processing ordering explanation underlyinga primary–secondary account, adapted from Krider et al. (2001):◦When evaluating the healthiness of eating [particular food portion], which better captures what you did? (I basically looked atthe type of food but made some adjustments for the quantity, I basically looked at the quantity of food but made some adjustments for thetype of food)

5 • Participants viewed pictures of two different snack portions side-by-side and chose between them given a health goal consistingof weight loss or weight management.

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food Portions3378 Management Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS

4We thank an anonymous reviewer and the editor for suggesting thata weight impact measure should seemingly make quantity moresalient.5Note that an alternative explanation is that perhaps consumers arenever sensitive to quantity when utilizing a measure such as “howhealthy is this snack” (in study 1). However, later, we address thisalternative account in studies 3a and 3b, in which we hold constantthe healthiness measure and instead use other ways to increase thesalience of quantity, to show that the same healthiness measureshows some quantity sensitivity under particular conditions. We alsoaddress this alternative account in study 4, by holding the healthimpact measure constant and testing for moderation by individualdifferences in choice of evaluation approach.6We also performed LSD contrast comparisons for each of the lesshealthy food types (cookies: small versus medium [p = 0.062, d = 0.45],medium versus large [p = 0.434, d = 0.22], small versus large [p =0.009, d = 0.70]; M&Ms: small versus medium [p = 0.517, d = 0.19],medium versus large [p = 0.158, d = 0.34], small versus large [p = 0.431,d = 0.20]; Cheetos: small versus medium [p = 0.843, d = 0.05], mediumversus large [p = 0.213, d = 0.23], small versus large [p = 0.156, d = 0.39])and healthier food types (almonds: small versus medium [p = 0.490,d = 0.19], medium versus large [p = 0.018, d = 0.64], small versus large[p = 0.003, d = 0.70]; carrots: small versus medium [p = 0.110, d = 0.43],medium versus large [p = 0.129, d = 0.36], small versus large [p = 0.945,d= 0.02];Wheat Thins: small versusmedium [p= 0.611, d= 0.14],mediumversus large [p = 0.511, d = 0.15], small versus large [p = 0.876, d = 0.04]).7As noted earlier, the within-subject effect sizes for portion sizethroughout the paper ought to be interpreted with caution becauseit is likely an overestimated effect size (Lakens 2013).8We recognize that the size perceptions and caloric perceptions werecaptured in separate studies and participants were not randomlyassigned to one or the other. However, because they are captured onsimilar 1–9 scales, a profile analysis is utilized to comparewhether theprofiles are parallel. Additionally, if we Z-score standardize both sizeperceptions and caloric perceptions (i.e., such that the overall mean ofall size perceptions, collapsed across food type and food quantity, isone with a standard deviation of zero, and the overall mean of allcaloric perceptions, collapsed across food type and food quantity, isalso one with a standard deviation of zero) and conduct a similarmixed model analysis, we still find a key significant interaction[F(2, 726) = 37.26, p < 0.001, ηp2 = 0.093] between type of measure andfood quantity, suggesting that size perceptions are incorporated onless than a 1:1 basis by caloric perceptions.9Again, as noted earlier, thewithin-subject effect sizes for portion sizeought to be interpreted with caution because it is likely an over-estimated effect size (Lakens 2013).10Again,we recognize that thesewere captured in separate studies, andparticipants were not randomly assigned to one or the other. However,because bothmeasureswere captured on 1–9 scales, a profile analysis isutilized to compare whether they are parallel. Additionally, if weZ-score standardize both size perceptions and weight impact andconduct a similar mixed model analysis, the interaction between typeof measure and food quantity remains significant [F(2, 726) = 54.93,p < 0.001, ηp

2 = 0.131], showing that size perceptions are incorporatedat less than a 1:1 basis by weight impact perceptions.11Given that few participants (31 of 180) indicated that they wouldacquire information about food quantity before food type, we did notconduct a formal analysis of whether acquisition order preferencemoderated the effects of food type and food quantity differences onweight impact evaluations. However, the two consumer segments’means for perceived weight impact based on food type (collapsingacross food quantity) and food quantity (collapsing across food type)are available in the online appendix. An examination of the means issuggestive, consistent with our theory, that consumers who indicatethat they would acquire information about food type first seem to

show a larger food type effect and a smaller food quantity effect thanconsumers who indicate that they would acquire information aboutfood quantity first.12 Study 3c purposely utilized a measure of weight impact that madequantitymore salientwithout explicitlymentioning theword “quantity”in the evaluation measure, which we believed served as an ideal furthertest of the bounds of our primary–secondary account. However, onecould even consider health impact measures that explicitly referencethe quantity of food, in a sense representing the most extreme form ofquerying health impact in away that prioritizes quantity. Thus, althoughnot central to our paper, we conducted an additional study to examinewhether the secondary nature of quantity emerges even under queryingconditions most biased to make quantity focal (see online appendix forfull details on this additional study).13The Johnson-Neyman test also revealed another significant regionless than 22.74. However, because only 12 participants out of 161 hadscores lower than 22.74, this region should be interpretedwith greatercaution, and we thus focus on the other region greater than 51.64,which captures the majority of participants.14With the caveat again that relatively few participants report astrategy of focusing first on quantity and then adjusting for type, wenote that the one-way ANOVA for this much smaller segment wasnonsignificant [Mtype = 4.90 versusMquantity = 5.78 versusMtype & quantity =6.04; F(2, 44) = 1.89, p = 0.162, ηp

2 = 0.079], and the means followeda different pattern.15Although our primary focus in this studywas on having a real choice,we did have research assistants weigh remaining snacks after eachsession. Unfortunately, an administrative mistake led to approximatelyhalf of the participants being served serving sizes larger than what wasdepicted in the photos they selected from. For these reasons, and others,such as a relatively short amount of time for consumption, we do notdiscuss consumption quantities further in the paper. Please refer to theonline appendix for further information on snack consumption.16Because the main aim of study 5 was to examine snack choice, thepost-snack caloric consumption results are presented in the onlineappendix. Overall, there was no main effect of participants’ self-selected snacks on calories consumed the rest of the day after thestudy, but as discussed in the General Discussion (Section 9), futureworkmay test the effect of randomly assigning participants to a givensnack (or more generally, either a health strategy emphasizing foodtypes or a strategy emphasizing food quantities) and then measuringpost-snack caloric consumption.

ReferencesBettman JR (1979) Information Processing Theory of Consumer Choice

(Addison-Wesley, Reading, MA).Campbell MC, Warren C (2015) The progress bias in goal pursuit:

When one step forward seems larger than one step back.J. Consumer Res. 41(5):1316–1331.

Carpenter GS, Glazer R, Nakamoto K (1994) Meaningful brands frommeaningless differentiation: The dependence on irrelevant at-tributes. J. Marketing Res. 31(3):339–350.

Chandon P, Ordabayeva N (2009) Supersize in one dimension,downsize in three dimensions: Effects of spatial dimensionalityon size perceptions and preferences. J. Marketing Res. 46(6):739–753.

Chandon P, Wansink B (2006) How biased household inventoryestimates distort shopping and storage decisions. J. Marketing70(4):118–135.

Chandon P, Wansink B (2007a) The biasing health halos of fast-foodrestaurant health claims: Lower calorie estimates and higher side-dish consumption intentions. J. Consumer Res. 34(3):301–314.

Chandon P, Wansink B (2007b) Is obesity caused by calorie un-derestimation? A psychophysical model of meal size estimation.J. Marketing Res. 44(1):84–99.

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food PortionsManagement Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS 3379

Chandon P, Wansink B (2012) Does food marketing need to make usfat? A review and solutions. Nutrition Rev. 70(10):571–593.

Cheema A, Bagchi R (2011) The effect of goal visualization on goalpursuit: Implications for consumers and managers. J. Marketing75(2):109–123.

Chernev A, Gal D (2010) Categorization effects in value judgments:Averaging bias in evaluating combinations of vices and virtues.J. Marketing Res. 47(4):738–747.

Cochran W, Tesser A (1996) The “What the Hell” Effect: Some Effects ofGoal Proximity and Goal Framing on Performance (Lawrence Erl-baum Associates, Hillsdale, NJ).

Cohen J (1988) Statistical Power Analysis for the Behavioral Sciences(Routledge Academic, New York).

Collins K (2006)New Survey on Portion Size: Americans Still Cleaning Plates(American Institute for Cancer Research, Washington, DC).

Cornil Y, Chandon P (2016) Pleasure as a substitute for size: Howmultisensory imagery can make people happier with smallerfood portions. J. Marketing Res. 53(5):847–864.

Dallas SK, Liu PJ, Ubel PA (2015) Potential problems with increasingserving sizes on the nutrition facts label. Appetite 95(1):577–584.

David ME, Haws KL (2016) Saying “no” to cake or “yes” to kale:Approach and avoidance strategies in pursuit of health goals.Psych. Marketing 33(8):588–594.

Dhar R, Simonson I (1999) Making complementary choices in con-sumption episodes: Highlighting vs. balancing. J. Marketing Res.36(1):29–44.

Dryer MS (2006) Descriptive Theories, Explanatory Theories, and BasicLinguistic Theory (Mouton de Gruyter, Berlin).

Dubois D, Rucker DD, Galinsky AD (2012) Super size me: Productsize as a signal of status. J. Consumer Res. 38(6):1047–1062.

Duckworth AL, Peterson C, Matthews MD, Kelly DR (2007) Grit:Perseverance and passion for long-term goals. J. Personality Soc.Psych. 92(6):1087–1101.

DunlapWP, Cortina JM, Vaslow JB, BurkeMJ (1996)Meta-analysis ofexperiments with matched groups or repeated measures de-signs. Psych. Methods 1(2):170–177.

Epley N, Gilovich T (2006) The anchoring-and-adjustment heuristic:Why the adjustments are insufficient. Psych. Sci. 17(4):311–318.

Etkin J, Ratner RK (2012) Goal pursuit, now and later: Temporalcompatibility of different versus similar means. J. Consumer Res.39(5):1085–1099.

Fang X, Mishra S (2002) The Effect of Brand Alliance Portfolio on thePerceived Quality of an Unknown Brand (Association for Con-sumer Research, Valdosta, GA).

Fox CR, Ratner RK, Lieb DS (2005) How subjective grouping ofoptions influences choice and allocation: Diversification bias andthe phenomenon of partition dependence. J. Experiment. Psych.Gen. 134(4):538–551.

Gal D, LiuW (2011) Grapes of wrath: The angry effects of self-control.J. Consumer Res. 38(3):445–458.

Gonzalez-Vallejo C, Moran E (2001) The evaluability hypothesisrevisited: Joint and separate evaluation preference reversal asa function of attribute importance.Organ. Behav. Human DecisionProcess 86(2):216–233.

Guenther P, DeMaio T, Ingwersen L, BerlinM (1997) Themultiple-passapproach for the 24-h recall in the continuing survey of food intakesby individuals, 1994-1996. Amer. J. Clin. Nutrition 65(4):S1316.

Haws KL, Liu PJ (2016a) Combining food type (s) and food quantitychoice in a new food choice paradigm based on vice-virtuebundles. Appetite 103(1):441–449.

Haws KL, Liu PJ (2016b) Half-size me? How calorie and price in-formation influence ordering on restaurant menuswith both halfand full entrée portion sizes. Appetite 97(1):127–137.

Haws KL, Winterich K (2013) When value trumps health in a su-persized world. J. Marketing 77(3):48–64.

Hayes AF, Montoya AK (2017) A tutorial on testing, visualizing,and probing an interaction involving a multicategorical

variable in linear regression analysis. Commun. Methods Meas.11(1):1–30.

Helgeson JG, Beatty SE (1987) Price expectation and price recall error:An empirical study. J. Consumer Res. 14(3):379–386.

Hogarth RM (2001) Educating Intuition (University of Chicago Press,Chicago).

Hsee CK (1996) The evaluability hypothesis: An explanation forpreference reversals between joint and separate evaluations ofalternatives.Organ. Behav. Human Decision Process 67(3):247–257.

Huang S-c, Zhang Y, Broniarczyk SM (2012) So near and yet so far:The mental representation of goal progress. J. Personality Soc.Psych. 103(2):225–241.

Huberts JCDW, Evers C, De Ridder DT (2014) “Because I am worthit”: A theoretical framework and empirical review of a justifi-cation-based account of self-regulation failure. Personality Soc.Psych. Rev. 18(2):119–138.

International Food Information Council Foundation (2012) 2012 Food&Health Survey: Consumer Attitudes Toward Food Safety, Nutrition,& Health (International Food Information Council Founda-tion, Washington, DC).

Irmak C, Vallen B, Robinson SR (2011) The impact of product nameon dieters’ and nondieters’ food evaluations and consumption.J. Consumer Res. 38(2):390–405.

Jeffreys H (1998) Theory of Probability (Oxford University Press, Oxford,UK).

Kahneman D, Frederick S (2002) Representativeness Revisited: AttributeSubstitution in Intuitive Judgment (Cambridge University Press,New York).

Kass RE, Raftery AE (1995) Bayes factors. J. Amer. Statist. Assoc.90(430):773–795.

Khan U, Dhar R (2006) Licensing effect in consumer choice.J. Marketing Res. 43(2):259–266.

Kivetz R, Zheng Y (2006) Determinants of justification and self-control. J. Experiment. Psych: General 135(4):572–587.

Krider RE, Raghubir P, Krishna A (2001) Pizzas: Π or square? Psycho-physical biases in area comparisons. Marketing Sci. 20(4):405–425.

Kuo T, Jarosz CJ, Simon P, Fielding JE (2009) Menu labeling asa potential strategy for combating the obesity epidemic: A healthimpact assessment. Amer. J. Public Health 99(9):1680–1686.

Lakens D (2013) Calculating and reporting effect sizes to facilitatecumulative science: A practical primer for t-tests and ANOVAs.Frontiers Psych. 4:1–12.

Laran J (2010) Goal management in sequential choices: Consumerchoices for others are more indulgent than personal choices.J. Consumer Res. 37(2):304–314.

Liu PJ, Haws KL, Lamberton C, Campbell TH, Fitzsimons GJ (2015)Vice-virtue bundles. Management Sci. 61(1):204–228.

Maxwell SE, Delaney HD (2004) Designing Experiments and AnalyzingData: A Model Comparison Perspective (Psychology Press, NewYork).

McFarlane T, Polivy J, McCabe RE (1999) Help, not harm: Psycho-logical foundation for a nondieting approach toward health.J. Soc. Issues 55(2):261–276.

Miller N, Reicks M, Redden JP, Mann T, Mykerezi E, Vickers Z (2015)Increasing portion sizes of fruits and vegetables in an elementaryschool lunch program can increase fruit and vegetable con-sumption. Appetite 91(1):426–430.

Mukhopadhyay A, Johar GV (2009) Indulgence as self-reward forprior shopping restraint: A justification-based mechanism.J. Consumer Psych. 19(3):334–345.

Nguyen SP (2007) An apple a day keeps the doctor away: Children’sevaluative categories of food. Appetite 48(1):114–118.

Nielsen SJ, Popkin BM (2003) Patterns and trends in food portionsizes, 1977-1998. J. Amer. Medical Assoc. 289(4):450–453.

Oakes ME (2004) Good foods gone bad: ‘Infamous’ nutrients di-minish perceived vitamin and mineral content of foods. Appetite42(3):273–278.

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food Portions3380 Management Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS

Oakes ME (2005) Stereotypical thinking about foods and perceivedcapacity to promote weight gain. Appetite 44(3):317–324.

OakesME, Slotterback CS (2005) Too good to be true: Dose insensitivityand stereotypical thinking of foods’ capacity to promote weightgain. Food Quality Prefer. 16(8):675–681.

Olejnik S, Algina J (2003) Generalized eta and omega squared sta-tistics: Measures of effect size for some common research de-signs. Psych. Methods 8(4):434–447.

OrdabayevaN,Chandon P (2016) In the eye of the beholder: Visual biasesin package and portion size perceptions. Appetite 103(1):450–457.

Peeters G (2002) From good and bad to can and must: Subjectivenecessity of acts associatedwith positively and negatively valuedstimuli. Eur. J. Soc. Psych. 32(1):125–136.

Puhl RM, Heuer CA (2010) Obesity stigma: Important considerationsfor public health. Amer. J. Public Health 100(6):1019–1028.

Raghubir P, Krishna A (1999) Vital dimensions in volume perception:Can the eye fool the stomach? J. Marketing Res. 36(3):313–326.

Raynor H (2014) What to do about portion sizes? Roundtable discussionat the forefronts in portion size conference. Internat. J. Obesity38(1):S34–S36.

Roberto CA, Larsen PD, Agnew H, Baik J, Brownell KD (2010)Evaluating the impact of menu labeling on food choices andintake. Amer. J. Public Health 100(2):312–318.

Rolls BJ, Morris EL, Roe LS (2002) Portion size of food affects energyintake in normal-weight and overweight men andwomen.Amer.J. Clin. Nutrition 76(6):1207–1213.

Ross J, Harradine R (2004) I’m not wearing that! Branding and youngchildren. J. Fashion Marketing Management 8(1):11–26.

Rouder JN, Speckman PL, Sun D, Morey RD, Iverson G (2009)Bayesian t tests for accepting and rejecting the null hypothesis.Psych. Bull. Rev. 16(2):225–237.

Rozin P, Ashmore M, Markwith M (1996) Lay American conceptionsof nutrition: Dose insensitivity, categorical thinking, contagion,and the monotonic mind. Health Psych. 15(6):438–447.

Schwartz J, Riis J, Elbel B, Ariely D (2012) Inviting consumers todownsize fast-food portions significantly reduces calorie con-sumption. Health Affairs 31(2):399–407.

Scott AR, Reed DB, Kubena KS, McIntosh WA (2007) Evaluation ofa group administered 24-hour recall method for dietary as-sessment. J. Extension 45(1):1RIB3.

Sharpe KM, Staelin R, Huber J (2008) Using extremeness aversion tofight obesity: Policy implications of context dependent demand.J. Consumer Res. 35(3):406–422.

Shiv B, Fedorikhin A (1999) Heart and mind in conflict: The interplayof affect and cognition in consumer decisionmaking. J. ConsumerRes. 26(3):278–292.

Simmons JP, Nelson LD, Simonsohn U (2011) False-positive psy-chology: Undisclosed flexibility in data collection and analysisallows presenting anything as significant. Psych. Sci. 22(11):1359–1366.

Thoma V, Williams A (2013) The devil you know: The effect of brandrecognition and product ratings on consumer choice. JudgmentDecision Making 8(1):34–44.

Urbszat D, Herman CP, Polivy J (2002) Eat, drink, and be merry, fortomorrow we diet: Effects of anticipated deprivation on foodintake in restrained and unrestrained eaters. J. Abnorm. Psych.111(2):396–401.

Van Ittersum K, Wansink B (2012) Plate size and color suggestibility:The Delboeuf illusion’s bias on serving and eating behavior.J. Consumer Res. 39(2):215–228.

Wagenmakers E-J, Love J, Marsman M, Jamil T, Ly A, Verhagen J,Selker R, et al. (2018) Bayesian inference for psychology. PartII: Example applications with JASP. Psych. Bull. Rev. 25(1):58–76.

Wansink B (1996) Can package size accelerate usage volume?J. Marketing 60(3):1–14.

Wansink B (2004) Environmental factors that increase the food intakeand consumption volume of unknowing consumers. Annu. Rev.Nutrition 24(1):455–479.

Wansink B (2012) Package size, portion size, serving size . . . Marketsize: The unconventional case for half-size servings. MarketingSci. 31(1):54–57.

Wansink B, Chandon P (2014) Slim by design: Redirecting the ac-cidental drivers of mindless overeating. J. Consumer Psych. 24(3):413–431.

Wansink B, Park S (2001) At the movies: How external cues andperceived taste impact consumption volume. Food Quality Prefer.12(1):69–74.

Wansink B, Van Ittersum K (2003) Bottoms up! The influence ofelongation on pouring and consumption volume. J. ConsumerRes. 30(3):455–463.

Wansink B, Painter JE, North J (2005) Bottomless bowls: Whyvisual cues of portion size may influence intake. Obesity Res.13(1):93–100.

Wilcox K, Vallen B, Block L, Fitzsimons GJ (2009) Vicarious goal ful-fillment: When the mere presence of a healthy option leads to anironically indulgent decision. J. Consumer Res. 36(3):380–393.

Woolley K, Fishbach A (2016) For the fun of it: Harnessing immediaterewards to increase persistence in long-term goals. J. ConsumerRes. 42(6):952–966.

Young LR, Nestle M (2002) The contribution of expanding portionsizes to the US obesity epidemic. Amer. J. Public Health 92(2):246–249.

Liu et al.: How Type and Quantity Shape Healthiness Perceptions of Food PortionsManagement Science, 2019, vol. 65, no. 7, pp. 3353–3381, © 2018 INFORMS 3381