Article Review 8, 9 & 10

Each review should be one page, double spaced. The review should be a ½ page summary of the article, and a ½ page application of the material to the real world. Details about each review’s requirements are posted in the assignment instructions section.

JOURNAL OF ACADEMIC RESEARCH IN ECONOMICS

THE EFFECT OF VISUAL ATMOSPHERIC CUES ON

COMFORT BUYING BEHAVIOURS OF CONSUMERS IN

RETAIL STORES

AFEEZ BABATUNDE SIYANBOLA

Department of Fine and Applied Arts, Olabisi Onabanjo University, Nigeria

afeezsegun@yahoo.com

NATHANIEL OLUWASEUN OGUNSEYE

Department of Urban and Regional Planning, Olabisi Onabanjo University, Nigeria

townplannerseun@yahoo.com

Abstract

People routinely engage in shopping for discerning needful, self-gratification, hedonic

reasons and other factors, which are basically driven by the state of mind. This study

investigates the effect of visual atmospheric cues on comfort buying in Shoprite retail stores

in Ikeja, Lagos State, Nigeria. Consequently, this study explored the influence of visual

atmospherics such as colour, typography, images, advertisement and wayfinding signage’s
on the consumers’ behaviours and shopping destinations. Purposive sampling technique was
used in the distribution of a structured questionnaire to three hundred (300) shoppers. Both

descriptive and inferential statistic were employed for data analysis using Statistical Package

for Social Science version 19. The study revealed that consumers often engage in retail

therapy, while visual atmospheric cues mediate an enjoyable shopping experience. The study

also showed that colour usage on advertisement posters greatly influenced consumers’
behaviour as compared to texts and pictures. Further, red and blue colours made significant

contributions to aesthetically pleasing retail environment. Shoppers mostly patronize

clothing and apparel products when engaging in comfort buying. The regression analysis

conducted showed visual atmospheric cues statistically influence shoppers’

behaviours

(F=117.882, p=0.000). The study recommended that the psychological desires of potential

shoppers must be considered in the planning and designing of a retail environment, an in-

depth knowledge of colours moods is essential in the application of colours in a retail

environment, and lighting display in a retail environment should be attractive and sensational.

Keywords: Behaviours, Comfort buying, Retail stores, Shoppers, Visual atmospheric cues

JEL Classification: D10, D1

2

1. INTRODUCTION

Buying is influenced by consumer’s desire to acquire goods that improve
and enhance their daily lives. People routinely engage in shopping for discerning

VOLUME 13 NUMBER 1 MARCH 2021 7

mailto:afeezsegun@yahoo.com

mailto:townplannerseun@yahoo.com

JOURNAL OF ACADEMIC RESEARCH IN ECONOMICS

needful, self-gratification, hedonic reasons and other factors, which are basically

driven by the state of mind. The choice of buying and place do much to enhance

shopper’s perceived personality. Comfort buying simply implies making purchasing
decisions to assuage feelings and enlivens mood. This is also referred to as “retail
therapy” or emotional shopping. Retail therapy is casually defined as shopping to

alleviate negative moods (Kang, 2009; Rick et al., 2014). This term was first

originated from an article published in the Chicago Tribune during the Christmas

Eve of 1986 as expressed in this sentence “we have become a nation measuring out
our lives in shopping bags and nursing our psychic ills through retail therapy”
(Schimich, 1986, p. 1). Lee (2015, p. 70) underscores the prevalence of comfort

buying in the following quotes by some personalities:

“I always say shopping is cheaper than a psychiatrist.” — Tammy

Faye Messner, American singer and television personality

“Win or lose, we go shopping after the election.” — Imelda Marcos,

Previous First Lady of the Philippines

“Everyone needs an occasional dose of retail therapy.” — Susan

Thurston, Tampa Bay Times staff writer

“Whoever said money can’t buy happiness simply didn’t know where
to go shopping.” — Bo Derek, American actress

Comfort buying is relieving and self-gratifying. Personalities are defined and

expressed through choice and places of purchase. Perhaps, most purchasing

decisions are basically influenced by consumer’s mood and state of mind.
Researches have established that people derive more satisfaction spending on

pleasurable experiences (Kang, 2009, p. 21). The sensory feelings and experiences

of a retail environment physical and online stores enrich shopping experience.

Improvement of mood stemmed from imagining consumption, experiencing retail

environments, being well-treated by sales associates, shopping activity, and

purchasing (Kang, 2009, p. 18).

Several studies have shown that most people engage in shopping to improve

their mood. Atalay and Meloy (2011) found that among 69 college participants, 43

(62%) reported having purchased an item to treat themselves in the past one week in

order to repair their mood; in comparison, 19 (28%) were motivated. A study

conducted by TNS Global on behalf of Ebates.com establish that more than half of

the Americans admit to engaging in “retail therapy” (Yarrow, 2013). Loureiro et al.

(2019) explored the effect of consumer-generated media stimuli on emotions and

explained that consumer generated media stimuli are positively related to the

dimensions of emotions. Douce and Janssens (2013) posited that marketing

emphasis has shifted from the product to the creation of consumers’ experiences, and

sensory marketing seems to be integral to stimulating excitement and pleasure. Kim

and Sullivan (2019) emphasized the relevance of emotional branding when

developing marketing strategies for fashion brands in a volatile marketplace.

8 VOLUME 13 NUMBER 1 MARCH 2021

https://www.psychologytoday.com/basics/therapy

https://Ebates.com

JOURNAL OF ACADEMIC RESEARCH IN ECONOMICS

Consequently, the individual`s emotional state influences his comportment

within the environment, framed as “approach–avoidance” response. Positive

emotional response influences sensory stimulation in people in enabling a pleasure-

driven experience while shopping. External impulses leverage an emotional

attachment between the buyer and the shopping environment. Reid (1785) as cited

in Song (2010, p. 3) emphasized the role of external senses in human feelings and

perception:

“The external senses have double provinces that make us feel and

perceive. They furnish us with a variety of sensations, some pleasant,

others painful, and others indifferent; at the same time, they give us a

conception, and an invincible belief of the existence of external objects.

The feeling which goes along with the perception, we call sensation.

The perception and its corresponding sensation are produced at the

same time. In our experience we never find them disjoined. Hence, we

are led to consider them as one thing, to give them one name, and to

confound their different attributes….”

Hong (2016) as cited in Kim and Sullivan (2019, p. 2) noted that purchasing

intentions from television advertising are more likely to result from emotional

responses as advertisement content. Luomala (2002) identified eight types of

therapeutic power stemming from different mood-alleviative consumption activities,

which are distraction, self-indulgence, and activation has links with shopping and

purchasing. Lee (2015) in a study entitled “The emotional shopper: Assessing the

effectiveness of retail therapy” reviewed extant literature on shopping and emotions
in proposing a tripartite approach, which provides a holistic attempt at assessing

retail therapy works based on three perspectives, which are motivational (the goals

and motives that consumers have for shopping); behavioural (the activities

consumers engage in during the shopping process); and emotional (the feelings that

consumers experience while shopping). Nearly all consumers are motivated by

different factors and considerations when making purchasing decision. Consumer

behaviours considered to be either mundane or germane form the basis of

consumption choices. Evaluative consumption preferences are attitudes biased by

elements of rational and perceived irrational preferences. It suffices to note that the

presumed irrational factors influencing choices are rational in the mind of consumer

when it satisfies a particular urge.

According to Stankevich (2017), the consumer decision making process

primarily involves:

i. Problem/need recognition: Consumer recognizes a problem or

need. The need is triggered by internal stimuli and rises to a level high

enough to become a drive like hunger and food. A need can also be

triggered by external stimuli (such as advertisement). For instance,

commercials for a new pair of shoes can stimulate a need for a new pair

of shoes. Advertisements induced urge can also assuage feelings through

cohesive and fascinating atmospheric cues. The shopping experience of

VOLUME 13 NUMBER 1 MARCH 2021 9

JOURNAL OF ACADEMIC RESEARCH IN ECONOMICS

contemporary buyers are made memorable by the ambience of the

shopping environment.

ii. Choice of products: At this stage the shopper makes his/her choice of

product based on the exciting shopping experience stirred by the ambience

of the shopping environment.

iii. Evaluation of alternatives: The comfort buyer is impulsive when making

purchasing decision. Considering viable options are less prioritizing when

the shopper indulges in shopping to improve his mental state.

iv. Post-purchase behaviour: This is the ability the to deliver the purchased

product at buyer’s doorstep in good condition within specified time of

delivery. In ensuring buyer’s satisfaction, relieving buyer’s doubts shortly
after a purchase about whether it was the right decision.

The other sections of this paper is structured as follows. The existing

literature on visual atmospherics and its impact in a retail environment are reviewed

while the research was contextualized in related theories. Then the study area was

briefly presented with the study methodology. The study results were presented and

discussed. This study concluded with implications, recommendations, and future

area of research.

2. LITERATURE REVIEW

2.1. VISUAL ATMOSPHERIC CUES IN STORES

Atmospherics refers to the store’s physical characteristics that project an
image and attract customers (Kotler, 1974; Berman et al., 2007). It is the

psychological feeling a customer gets when entering a retail store. Mehrabian and

Russell’s (1974) as cited in Graa and Dani-elKebir (2012) acknowledged that the

impact of situation on behaviour is mediated by emotional responses in the stimulus

and response model. The model (Figure 1) states that situational conditions initially

generates an emotional (affective, connotative, feeling) reaction, which in turn leads

to a behavioural response. Altinigne and Karaosmanoglu (2017) explored the

importance of website atmospherics with emphasis on visual complexity in online

retailing and concluded that less complex visual atmospherics increases purchasing

intention online. Visual atmospherics has become an important aspect in retail

design. Floor (2006) asserted that consumers enjoy being inspired by a unique range

of experiential shopping environments. Visual atmospherics is basically conceived

and implemented to improve shoppability of a retail store. “Shoppability” of a retail
environment (Heil, 2018; Singh et al., 2014) leverages an in-store experience that is

appealing and immersive. “Shoppability” is classified into two elements which are

shopper engagement and purchase conversion (Burke and Morgan, 2017, p. 52).

Burke and Morgan (2017) benchmarked retail “shoppability” on five dimensions
highlighted in Table 1:

10 VOLUME 13 NUMBER 1 MARCH 2021

JOURNAL OF ACADEMIC RESEARCH IN ECONOMICS

Figure 1. The Mehrabian-Russell model

Source: Graa and Dani-Elkebir (2012)

Table 1: Dimension of Shoppability

Relevance The store has in stock the products that shoppers desire at a

competitive price

Transparency The shopping environment makes it easy for customers to see

and find desired products and limits visual and physical

clutter

Convenience The store reduces shopping time and effort by providing

convenient store access and parking, a quick and easy store

layout, and fast and helpful customer service

Assurance The presentation clearly conveys the unique benefits and

value of each product

Enjoyment The retail experience satisfies the incidental and contextual

needs of shoppers by providing unexpected surprises and a

comfortable environment

Source: Burke and Morgan (2017, p. 55)

Singh et al. (2014) investigated how store atmospherics and layout function

as a predictor of consumer behaviour and store performance, the study revealed that

visual atmospherics are the most significant factors that impacts customer approach

behaviours in a retail environment. Retail store owners and designers focus on

creating an appealing retail environment as a core part of their marketing strategies

to enhance positive customer’s perception of their brand. Sabir (2014) posited that
the elements of store atmospherics include layout design, colour, light, sound, scent

within a store. Turley and Milliman (2000) categorized atmospheric cues into five

components; external cues (architectural style, surrounding stores); general interior

cues (flooring, lighting, colour schemes, music, aisle width, ceiling composition);

layout and design cues (space design and allocation, grouping, traffic flow, racks and

cases); point of purchase and decoration displays (signs, cards, wall decorations,

price displays); and human variables (employee characteristics, uniforms, crowding,

privacy). Lighting impact buyer’s mood either positively or negatively. The
intensity, colour and positioning of lighting display. Paluchová et al. (2016)

researched into the impact of visual atmospheric on consumer behaviours in food

stores and posited that lighting is the most important visual element that elicits

VOLUME 13 NUMBER 1 MARCH 2021 11

JOURNAL OF ACADEMIC RESEARCH IN ECONOMICS

buyers in a food retail store. Consumers are often discouraged from shopping in a

retail store that is not properly illuminated, they prefer strong lighting for safety and

easy identification of the goods they are purchasing (Horská & Berčík, 2014, p. 455).

Kim and Sullivan (2019) in their study noted fashion retailers successfully provide

sensory experiences to consumers in their physical stores. Strang (2015) posited that

Lush cosmetic brand employ sensory marketing in creating sight from round shapes

of visually attractive products, provision of live plants to illustrate their product

ingredients, smell from strong sweet scents and sound from resourceful sale

representatives. However, the previous researches in the area of store atmospherics

have not dwelt elaborately into the graphic visual components, that constitutes the

retail display. The graphic visual components are typography, colours, images,

advertisements, signage’s and others. This study also investigates the products
mostly preferred by shoppers engaging in retail therapy. The specific objectives of

this study are to: ascertain the vulnerability of shoppers to comfort buying; determine

the influence of store ambience in stimulating consumer buying behaviours; evaluate

the impact of colours on comfort buying in a retail store; and determine the line of

products that are mostly preferable to comfort buyers.

3. MATERIALS AND METHODS

3.1. STUDY AREA

The study area is the Shoprite Store within the Ikeja City Mall at Alausa,

Ikeja, Lagos State in southwest Nigeria. Shoprite Store, located in Ikeja, is a

supermarket chain with house-label groceries serving as a meeting point or

recreational spot for families and associates. The Ikeja City Mall accommodates

other activities like the cinema, restaurants, clothing stores, hairdressing/beauty

salon, bars, cafes, banks amongst others.

Figure 2. Map of the Study Area

12 VOLUME 13 NUMBER 1 MARCH 2021

JOURNAL OF ACADEMIC RESEARCH IN ECONOMICS

3.2. METHODOLOGY

3.2.1. RESEARCH DESIGN, TARGET POPULATION AND SURVEY

INSTRUMENT

This study adopted a survey research design. Apparently, the study is

interested in quantitative data. The shoppers at the retail store forms the target

population. In order to obtain primary data from the shoppers, a structured

questionnaire was designed. The questionnaire borders on issues regarding:

Validating people’s susceptibility to comfort shopping? what captures shopper’s
imagination when they go shopping? Which colour shopper find its domination

aesthetically pleasing in a retail shopping environment? Which visual elements

attract shopper attention in advertisement posters of products displayed in a retail

store? and shoppers preferred line of stores whenever you want to enhance your

feelings. Also, questions such as if shopper shop for fun, what enlivens shopper

mood, and does shopping makes you feel happy? were posed to the shoppers.

3.2.2. SAMPLING PROCEDURE AND SAMPLE SIZE

For this study, the shoppers were purposively sampled. This sampling

technique was considered appropriate because there are other activities going on

within the Ikeja City Mall where Shoprite Store is situated. A total of 300 copies

were distributed and retrieved from the shoppers who are willing to participate in the

survey. The study sample (300 shoppers) consisted of 180 females and 120 males.

3.2.3. DATA ANALYSIS

The data collected were collated and analysed using Statistical Package for

Social Sciences (SPSS for Windows, version 19). Data were presented using both

descriptive (frequency and percentage distribution, mean and standard deviation)

and inferential (Multiple regression, ANOVA test, Durbin-Watson Test) statistics.

3.2.4. HYPOTHESIS TESTING

Hypothesis was formulated to determine if relationship exist between store

ambience and consumer buying behaviour. The hypothesis tested is stated as

follows:

Ho: Visual atmospheric cues does not significantly influence consumer

behaviours

H1: Visual atmospheric cues significantly influence consumer behaviours

4. RESULTS AND DISCUSSIONS

4.1. SHOPPERS’ VULNERABILITY TO COMFORT BUYING

Shoppers were asked series of questions of which results of analysis were

presented in Table 2. A majority (95.3%) of the respondents confirmed shopping

VOLUME 13 NUMBER 1 MARCH 2021 13

JOURNAL OF ACADEMIC RESEARCH IN ECONOMICS

make them feel happy, almost half (45.3%) stated they shop to enliven their mood,

larger proportion (68%) do not shop for fun, and a majority (90.0%) opined that retail

store ambience motivates them to make purchase. It can be inferred from the results

that shopping makes shoppers feel happy, but this behaviour does not depend on

whether they want to catch fun. Also, the retail store environment plays a huge role

in motivating shoppers, hence it influences their behaviour. The results in Table 2

confirmed that shoppers engage in comfort buying, which is consistent with findings

of studies (Atalay and Meloy, 2011; Kang, 2009).

Table 2: Shoppers’ Vulnerability to Comfort Buying

Variable Category Freque

ncy Percent

Does shopping make you feel happy?

Yes 286 95.3

No 14 4.7

Total 300 100.0

Do you shop to enliven your mood?

Yes 136 45.3

No 164 54.7

Total 300 100.0

Do you shop for fun?

Yes 96 32.0

No 204 68.0

Total 300 100.0

Does the ambience of the retail store

stimulate you to make purchase?

Yes 270 90.0

No 30 10.0

Total 300 100.0

4.2. FACTORS INFLUENCING STORE AMBIENCE IN

STIMULATING CONSUMERS’ BUYING BEHAVIOURS

The study revealed that visual elements such as texts, pictures and colour

usage on advertisement posters of products displayed in a retail store attract the

attention of shoppers. Over half (50.7%) of the respondents stated colour stimulate

their buying behaviours, 36.0% said pictures and 13.3% claimed texts. Further probe

regarding if colourful signage enhances shoppers’ experience in a retail environment
revealed that majority (86.0%) of the respondents’ shopping experience were

enhanced by colourful directional signage in a retail environment while 14.0%

thought otherwise (Table 3).

Table 3: Influence of visual elements and colourful directional signage

Variable Category Frequency Percent

Pictures 108 36.0

Which of these visual elements Texts 40 13.3
attract your attention in

advertisement posters of products
Colour

usage

152 50.7

displayed in a retail store? Total 300 100.0

Does colourful directional signage Yes 258 86.0

enhance your shopping experience No 42 14.0
in a retail environment? Total 300 100.0

14 VOLUME 13 NUMBER 1 MARCH 2021

JOURNAL OF ACADEMIC RESEARCH IN ECONOMICS

The results presented in Table 4 indicates that most respondents (64%) stated

that advertisement design captures their imagination when they go for shopping,

lighting capture the imagination of 18% of the respondents and 13.7% confirmed

colours. A marginal proportion (4.3%) identified wayfinding signage as factors with

the least influence.

Table 4: Shoppers’ interest during shopping in a retail store

Store Ambience Frequency Percent

Lighting 54 18.0

Advertisement posters 192 64.0

Colours 41 13.7

Wayfinding Signage 13 4.3

Total 300 100.0

4.3. IMPACT OF COLOURS ON COMFORT BUYING IN A RETAIL

STORE

Colours appeal to the viewers and create favourable impact on purchasing

decisions with mean and standard deviation of 3.1467±1.72096 respectively. From

the result in Table 5, Red and Blue were dominant in creating aesthetically pleasing

retail shopping environment as confirmed by 22.3% and 22.0% of the respondents

respectively. Other colours in terms of influence are Orange (18.0%), Purple

(14.3%), Green (13.3%), and Yellow (10.0%).

Table 5. Colour type and contribution to retail shop ambience

Colours Frequency Percent

Blue 66 22.0

Red 67 22.3

Green 40 13.3

Orange 54 18.0

Yellow 30 10.0

Purple 43 14.3

Total 300 100.0

4.4. LINE OF PRODUCTS PREFERABLE TO COMFORT BUYERS

Table 6 shows accordingly the kind of products respondents preferred when

engaging in comfort buying. Clothing and Apparel stores are preferred by 50.7% of

the respondents while Food and Beverages accounted for the preference by 28.3%

of the respondents with mean and standard deviation value of 2.4567±1.06704. This

result may have been influenced by the fact of clothing and food being among the

basic needs of man.

VOLUME 13 NUMBER 1 MARCH 2021 15

JOURNAL OF ACADEMIC RESEARCH IN ECONOMICS

Table 6. Line of Products Preferable

Products
Freque

ncy Percent

Electronics Store 48 16.0

Clothing & Apparel 152 50.7

Phone & Accessories 15 5.0

Food & Beverages 85 28.3

Total 300 100.0

4.5. HYPOTHESIS TESTING

Ho: Visual atmospheric cues does not significantly influence consumer

behaviours

H1: Visual atmospheric cues significantly influence consumer behaviours

The study further conducted a regression analysis to establish the degree of

affinity between dependent variable (do you shop for fun?) and independent

variables (Which of this colour do you find its domination aesthetically pleasing in

a retail shopping environment and Which of these captures your imagination when

you go shopping).

Table 7: Multiple Regression Test Results

Model Summaryb

Std.

Error of

Change Statistics

Durbin-

Watson

R

R Adjusted the Square F Sig. F

Model R Square R Square Estimate Change Change df1 df2 Change

1 .732a .536 .533 .31918 .536 171.882 2 297 .000 .040

ANOVAb

Model Sum of Squares Df Mean Square F Sig.

1 Regression

Residual

Total

35.022

30.258

65.280

2

297

299

17.511

.102

171.882 .000a

a. Predictors: (Constant), Which of this colour do you find its domination aesthetically

pleasing in a retail shopping environment, which of this captures your imagination when you

go shopping

b. Dependent Variable: Do you shop for fun

As shown in Table 7, the R2 value which is a measure of how much of the

variability in the outcome is accounted for by the predictors. From the model, its

value is 0.536, which indicated that 54% of the total variations is accounted for by

the independent variables (predictors). These results, therefore, show that 46% of the

variation is caused by factors other than the predictors

16 VOLUME 13 NUMBER 1 MARCH 2021

JOURNAL OF ACADEMIC RESEARCH IN ECONOMICS

From model summary table (Table 7), Durbin Watson statistics shows 0.040,

which can be approximate to 1 indicate a strong positive autocorrelation among the

variables of the model.

From the ANOVA table (Table 7), which tests the overall regression is a

good fit for data, the F-ratio value (171.882) indicates that the independent variables

statistically predict the dependent variable, F (2, 297) = 117.882, p. (0.000) < 0.5.

The regression model is a good fit of the data. The comparison of the observed

significant value (0.000) with the table value (0.05) clearly shows that the observed

significant value is less than the table significant value. Hence, the alternative

hypothesis (H1) is accepted and the null hypothesis (H0) is rejected. Meaning that

visual atmospheric cues statistically influence the consumer behaviours.

5. CONCLUSION AND RECOMMENDATIONS

This study explored the effects of visual atmospheric cues on comfort buying

behaviours of consumers in a Shoprite retail store in Ikeja, Lagos, southwest Nigeria.

Undoubtedly, this study has made contribution to literature in the fields of arts,

marketing and psychology especially how consumers’ behaviours can be influenced

by visual atmospheric cues. The study established that shoppers were motivated

whenever they engaged in shopping. The consumers’ behaviours are majorly
influenced by the colour on advertisement posters while a majority of the consumers

confirmed colourful directional signage enhances their shopping experience. The

study also established that red and blue colours were both dominant, among other

colours, in their contribution to aesthetically pleasing retail shopping environment.

In this study, it was also discovered that consumers’ preference to patronize the
Clothing and Apparel store is the primary way through which consumers enhance

their feelings. The results of the regression analysis conducted established that visual

atmospheric cues statistically influence shoppers’ behaviours (F=117.882, p=0.000).

We have, therefore learned, through this study that the market success of

retail stores is basically anchored on consumers’ desires, pleasure and satisfaction.

Environmental aesthetics stimulate desires and satisfaction. The retail environment

is vital in influencing the decisions of shoppers either positively or negatively.

Shoppers are enticed and stirred by the visual atmospheric cues in the retail

environment. The visual atmospheric cues elicit sensory experience and stimulate

distinctive robust impressions that permeate sensory organs of people visiting the

environment. Physical convenience of retail environments contributes enormously

to customer perception. However, comfort buying is an emotional behaviour that can

be positively guided by in-store visual atmospheric cues. Hence, experiential

shopping enhances feelings and foster pleasurable attainment of consumption goals.

Based on the study findings, the following recommendations were put

forward.

VOLUME 13 NUMBER 1 MARCH 2021 17

JOURNAL OF ACADEMIC RESEARCH IN ECONOMICS

i. The psychological desires of potential shoppers must be considered in the

planning and designing of a retail environment;

ii. An in-depth knowledge of colours moods is essential in the application of

colours in a retail environment;

iii. The lighting display in a retail environment should be attractive and

sensational;

iv. Advertisements of products available for sale in a retail store should form an

integral aspect of in-store décor; and

v. Functional wayfinding signages are necessary for seamless navigation in a

retail environment.

And finally, future research should focus on the impact of visual cues in

mediating positive consumers’ responses to businesses offering services in area of
healthcare and hospitality industries. Future studies should also focus on

demographic and cultural applicability of emotional visual brand marketing

strategies.

REFERENCES

Atalay, S. & Meloy, M. G. (2011). Retail Therapy: A Strategic Effort to Improve

Mood. Psychology and Marketing, 28, 638-659. DOI: 10.1002/mar.20404.

Altinigne, N., & Karaosmanoglu, E. (2017). The Importance of Web Atmospherics

with Emphasis on Visual Complexity in Online Retailing Based on S-O-R

Paradigm. Journal of Business, Economics and Finance (JBEF), 6(4), 348-354.

DOI: 10.17261/Pressacademia.2017.767

Berman, B., Evans, J. R., & Chatterjee, P. (2007). Retail Management: A Strategic

Approach. Upper Saddle River, NJ.: Prentice-Hall,

Burke, R. & Morgan, N. (2017). Benchmarking Retail Shoppability. Journal of

Shopper Research, 2017, 51-59. Available at:

https://www.researchgate.net/publication/315729863

Douce, L., & Wim, J. (2013). The Presence of a Pleasant Ambient Scent in a Fashion

Store: The Moderating Role of Shopping Motivation and Affect Intensity.

Environment and Behavior, 45(2), 215–238. DOI: 10.1177/0013916511410421

Floor, K. (2006). Branding a Store. Philadelphia, United States of America: Kogan

Page.

Graa, A. & Dani-elKebir, M. (2012). Application of Stimulus and Response Model

to Impulse Buying Behavior of Algerian Consumers. Serbian Journal of

Management, 7(1), 53-64. https://doi.org/10.1177/0092070394224002

Kang, M. (2009). Retail therapy: A Qualitative Investigation and Scale

Development. PhD Dissertation, The University of Minnesota. Available at:

http://hdl.handle.net/11299/52142.

18 VOLUME 13 NUMBER 1 MARCH 2021

https://www.researchgate.net/publication/315729863

https://www.researchgate.net/deref/http%3A%2F%2Fdx.doi.org%2F10.1177%2F0013916511410421?_sg%5B0%5D=H8YfZWoXtCbkhEQ3kd2yB3FEXPweh1-hlKoZdp-5paClK8RgVBtCAqC00pHSDoRnMMXfyZfcfWyW0vmXXa5pfN4Y3w.SdwPVn0kMo_pKuIlyMFbrRUy3E2qFJ43Q2n3-1Swwf23t5_jQZOE49KO80QbscoZaSRKYcG7d06FOhGuM8NMLA

https://doi.org/10.1177/0092070394224002

http://hdl.handle.net/11299/52142

JOURNAL OF ACADEMIC RESEARCH IN ECONOMICS

Kim, Y-K. & Sullivan, P. (2019). Emotional Branding Speaks to Consumers’ Heart:

The Case of Fashion Brands. Fashion and Textile, 62, 1-16.

https://doi.org/10.1186/s40691-018-0164-y

Kotler, P. (1974). Atmospherics as a Marketing Tool. Journal of Retailing, 49(4),

48-64.

Heil, K. (2018). How Retail Shoppability Factors into Unplanned Buying. Palmer

Retail Solutions Blog. https://www.palmerretailsolutions.com/blog/how-retail-

shoppability-factors-into-unplanned-buying on 03/10/2020

Horská, E. & Berčík, J. (2014). The Influence of Light on Consumer Behavior at the

Food Market. Journal of Food Products Marketing, 20(4), 440-492. DOI:

10.1080/10454446.2013.838531

Lee, L. (2015). The Emotional Shopper: Assessing the Effectiveness of Retail

Therapy. Foundations and Trends ® in Marketing, 8(2), 69-145. DOI:

10.1561/1700000035.

Loureiro, S.M.C., Bilro, R. G. & Japutra, A. (2019). The Effect of Consumer-

Generated Media Stimuli on Emotions and Consumer Brand Engagement.

Journal of Product and Brand Management, 29(3), 387-408.

https://doi.org/10.108/JPBM-11-2018-2120

Luomala, H. T. (2002). An Empirical Analysis of the Practices and Therapeutic

Power of Mood-Alleviative Consumption in Finland. Psychology & Marketing,

19(10), 813-836. DOI: 10.1002/mar.10039

Paluchová, J., Berčík, J., & Neomániová, K. (2016). The Atmosphere and its Impact

on Consumer Behaviour in Food Stores. DOI:

http://dx.doi.org/10.15414/isd2016.s10.05, Available at:

http://www.slpk.sk/eldo/2016/dl/9788055215037/files/10/paluchova-et-al

Rick S., Pereira B. D. C. S., & Burson, K. A. (2014). The Benefits of Retail Therapy:

Making Purchase Decisions Reduces Residual Sadness. Journal of Consumer

Psychology, 24(3), 373–380. https://doi.org/10.1016/j.jcps.2013.12.004

Sabir, I. (2014). Impact of Store Atmospherics on Impulse Buying Behavior

Regarding Shopping Goods in Pakistan. Masters Thesis. GRIN Publishing.

ISBN 103668048916

Schmich, M. T. (1986). A Stopwatch on Shopping. Chicago Tribune. December 24.

Available at: https://www.chicagotribune.com/news/ct-xpm-1986-12-24-

8604060073-story.html

Song, J. (2010). Retail Design and Sensory Experience: Design Inquiry of Complex

Reality. Iowa State University, USA. Available at:

http://www.drs2010.umontreal.ca/data/PDF/111

Singh, P., Katiyar N., & Verma, G. (2014). Retail Shoppability: The Impact of Store

Atmospherics & Store Layout on Consumer Buying Patterns. International

Journal of Scientific & Technology Research, 3(8), 15-23.

https://www.ijstr.org/final-print/aug2014/Retail-Shoppability-The-Impact-Of-

Store-Atmospherics-Store-Layout-On-Consumer-Buying-Patterns

VOLUME 13 NUMBER 1 MARCH 2021 19

https://www.researchgate.net/deref/http%3A%2F%2Fdx.doi.org%2F10.1080%2F10454446.2013.838531?_sg%5B0%5D=MramYfVv1YgZIR-oq_0y9JGmZaKzNHf71fdQVCvYwYSbwmusIbxiKdFxJGDI-hcqrleKk_NsYNKV6z0S5igkeM2-GA.cslh54lM6B3n2LN4ES5tcI9EcyxeKD3syU-T2YlqAkOB636sQJlv-j9YW2-7uvnlxg1wSc158Dj1pfKgTR2Nhw

https://www.researchgate.net/deref/http%3A%2F%2Fdx.doi.org%2F10.1002%2Fmar.10039?_sg%5B0%5D=rk7F806jEXIjmktLzpE9je9qo5gKP4rJJPCToC1vZNRXK4Nksn5A4kreMAAQ81J5RTBzPRPIsiGFro-mWBTnnn1jtA.EDpcMkZf-lqqWAzyaXZRKui45h7Qa5MynIN6LQqhP2gGLmXWd5OrvVDKcYCSiJCJ0w4YGVgf9unV4Swdm9iy2A

http://dx.doi.org/10.15414/isd2016.s10.05

http://www.slpk.sk/eldo/2016/dl/9788055215037/files/10/paluchova-et-al

https://doi.org/10.1016/j.jcps.2013.12.004

http://articles.chicagotribune.com/1986-12-24/features/8604060073_1_shopping-shaky-ground-festive-hat

https://en.wikipedia.org/wiki/Chicago_Tribune

https://www.ijstr.org/final-print/aug2014/Retail-Shoppability-The-Impact-Of-Store-Atmospherics-Store-Layout-On-Consumer-Buying-Patterns

https://www.ijstr.org/final-print/aug2014/Retail-Shoppability-The-Impact-Of-Store-Atmospherics-Store-Layout-On-Consumer-Buying-Patterns

http://www.drs2010.umontreal.ca/data/PDF/111

https://www.chicagotribune.com/news/ct-xpm-1986-12-24

https://doi.org/10.108/JPBM-11-2018-2120

https://www.palmerretailsolutions.com/blog/how-retail

https://doi.org/10.1186/s40691-018-0164-y

JOURNAL OF ACADEMIC RESEARCH IN ECONOMICS

Stankevich, A. (2017). Explaining the Consumer Decision Making Process: Critical

Literature Review. Journal of International Business Research and Marketing,

2(6), 7-14. DOI: 10.18775/jibrm.1849-8558.2015.26.3001

Strang, E. (2015). Scents and Sensibility at Lush Cosmetics. Available at:

https://theregister.co.nz/2015/03/26/scents-and-sensibility-lush-cosmetics/

Turley, L. W. & Milliman, R. E. (2000). Atmospheric Effects on Shopping Behavior:

A Review of the Experimental Evidence. Journal of Business Research, 49,

193–211. DOI: https://doi.org/10.1016/S0148-2963(99)00010-7

Yarrow K. (2013). Is Retail Therapy for Real? 5 Ways Shopping Is Actually Good

for You. Available at: https://business.time.com/2013/04/16/is-retail-therapy-

for-real-5-ways-shopping-is-actually-good-for-you/

20 VOLUME 13 NUMBER 1 MARCH 2021

https://doi.org/10.1016/S0148-2963(99)00010-7

https://business.time.com/2013/04/16/is-retail-therapy

https://theregister.co.nz/2015/03/26/scents-and-sensibility-lush-cosmetics

Copyright of Journal of Academic Research in Economics is the property of Journal of
Academic Research in Economics (JARE) and its content may not be copied or emailed to
multiple sites or posted to a listserv without the copyright holder’s express written permission.
However, users may print, download, or email articles for individual use.

527089 ABSXXX10.1177/0002764214527089American Behavioral ScientistZhu and Huberman
research-article2014

Article

To Switch or Not To Switch:
Understanding Social
Influence in Online Choices

American Behavioral Scientist
2014, Vol. 58(10) 1329–1344

© 2014 SAGE Publications
Reprints and permissions:

sagepub.com/journalsPermissions.nav
DOI: 10.1177/0002764214527089

abs.sagepub.com

Haiyi Zhu1 and Bernardo A. Huberman2

Abstract
The authors designed and ran an experiment to measure social influence in online
recommender systems, specifically, how often people’s choices are changed by
others’ recommendations when facing different levels of confirmation and conformity
pressures. In this experiment, participants were first asked to provide their preference
from pairs of items. They were then asked to make second choices about the same
pairs with knowledge of other people’s preferences. The results show that other
people’s opinions significantly sway people’s own choices. The influence is stronger
when people are required to make their second decision sometime later (22.4%)
rather than immediately (14.1%). Moreover, people seem to be most likely to reverse
their choices when facing a moderate, as opposed to large, number of opposing
opinions. Finally, the time people spend making the first decision significantly predicts
whether they will reverse their decisions later on, whereas demographics such as age
and gender do not. These results have implications for consumer behavior research
as well as online marketing strategies.

Keywords
social influence, social media and choices, conformity theory

Introduction

Picture yourself shopping online. You already have an idea about what product you are
looking for. After navigating through the website, you find that particular item as well as
several similar items and other people’s opinions and preferences about them provided
by the recommendation system. Will other people’s preferences reverse your own?

1Carnegie Mellon University, Pittsburgh, PA, USA
2HP Laboratories, Palo Alto, CA, USA

Corresponding Author:
Haiyi Zhu, Carnegie Mellon University, Pittsburgh, PA, USA.
Email: haiyiz@cs.cmu.edu

mailto:haiyiz@cs.cmu.edu

http://crossmark.crossref.org/dialog/?doi=10.1177%2F0002764214527089&domain=pdf&date_stamp=2014-03-28

https://abs.sagepub.com

https://sagepub.com/journalsPermissions.nav

1330 American Behavioral Scientist 58(10)

Notice that in this scenario, there are two contradictory psychological processes at
play. On one hand, when learning of other people’s opinions, people tend to select
those aspects that confirm their own existing ones. Prior literature suggests that once
one has taken a position on an issue, one’s primary purpose becomes defending or
justifying that position (e.g., Nickerson, 1998). From this point of view, if the recom-
mendations of others contradict our own personal opinions, we tend to not take this
information into account and stick to our own choices. But research on social influ-
ence and conformity theory (Cialdini & Goldstein, 2004) suggests that even when not
directly, personally, or publicly chosen as the target of others’ disapproval, individuals
may choose to conform to others and reverse their own opinions in order to restore
their sense of belonging and self-esteem.

To investigate whether online recommendations can sway people’s own opinions, we
designed an online experiment to test how often people’s choices are reversed by others’
preferences when facing different levels of confirmation and conformity pressures. We
used Rankr (Luon, Aperjis, & Huberman, 2012) as the study platform, which provides a
lightweight and efficient way to crowdsource the relative ranking of ideas, photos, or
priorities through a series of pairwise comparisons. In our experiment, participants were
first asked to provide their preferences between pairs of photos. Then, they were asked
to make a second choice about the same pairs with the knowledge of others’ preferences.
To measure the pressure to confirm people’s own opinions, we manipulated the time
between the participants’ two decisions about the same pair of photos. To determine the
effects of social pressure, we manipulated the number of opposing opinions that the
participants saw when making the second decision. Finally, we tested whether other fac-
tors (i.e., age, gender, and decision time) affect the tendency to revert.

Our results show that other people’s opinions significantly sway choices. The influ-
ence is stronger when people are required to make their second decision later (22.4%)
rather than immediately (14.1%) after their first decision. Furthermore, people are
most likely to reverse their choices when facing a moderate number of opposing opin-
ions. Last but not least, the time people spend making the first decision significantly
predicts whether they will reverse their decisions later on, whereas demographics such
as age and gender do not.

The main contribution of this article is that we designed and ran an experiment to
understand the mechanisms of social influence in online recommender systems.
Specifically, we measured the effect of others’ preferences on people’s own choices
under different conditions. The results have implications for consumer behavior
research and online marketing strategies.

Related Work

Confirming existing opinions. Confirmation of existing opinions is a long-recognized
phenomenon (Nickerson, 1998). As Francis Bacon (1939) stated several centuries ago,

The human understanding when it has once adopted an opinion (either as being received
opinion or as being agreeable to itself) draws all things else to support and agree with it.

Zhu and Huberman 1331

Although there be a greater number and weight of instances to be found on the other side,
yet these it either neglects and despises, or else by some distinction sets aside and rejects.
(p. 36)

This phenomenon (often referred to as confirmation bias) can be explained by
Festinger’s (1957) dissonance theory: As soon as individuals adopt a position, they
favor consistent over inconsistent information to avoid dissonance.

A great deal of empirical studies supports this idea (see Nickerson, 1998, for a
review). Many of these studies use a task invented by Wason (1960), in which people
are asked to find the rule that was used to generate specified triplets of numbers. The
experimenter presents a triplet, and the participant hypothesizes the rule that produced
it. The participants then test the hypothesis, by suggesting additional triplets and being
told whether it is consistent with the rule to be discovered. Results show that people
typically test hypothesized rules by producing only triplets that are consistent with the
hypotheses, indicating hypothesis-determined information seeking and interpretation.
Confirmation of existing opinions also contributes to the phenomenon of belief persis-
tence. Ross, Lepper, and Hubbard (1975) showed that once a belief or opinion has
been formed, it can be very resistant to change, even after learning that the data on
which the beliefs or opinions were originally based were fictitious.

Social conformity. In contrast to confirmation theories, social influence experiments
have shown that often people change their own opinion to match others’ responses.
The most famous experiment examining this is Asch’s (1956) line-judgment confor-
mity experiments. In the series of studies, participants were asked to choose which of
a set of three disparate lines matched a standard, either alone or after 1 to 16 confeder-
ates had first given a unanimous incorrect answer. Meta-analysis showed that, on aver-
age, 25% of the participants conformed to the incorrect consensus (Bond & Smith,
1996). Moreover, the conformity rate increased with the number of unanimous major-
ity. More recent, Cosley, Lam, Albert, Konstan, and Riedl (2003) conducted a field
experiment on a movie rating site. They found that by showing manipulated predic-
tions, users tended to rate movies toward the shown prediction. Researchers have also
found that social conformity leads to multiple macro-level phenomena, such as group
consensus (Asch, 1956), inequality and unpredictability in markets (Salganik, Dodds,
& Watts, 2006), unpredicted diffusion of soft technologies (Bendor, Huberman, & Wu,
2009), and undermined group wisdom (Lorenz, Rauhut, Schweitzer, & Helbing, 2011).

Latané (1981) proposed a theory to quantitatively predict how the effect of social
influence will increase as a function of the size of the influencing social source. The
theory states that the relationship between the effect of the social influence (I) and the
size of the influencing social source (N) follows a negative accelerating power func-
tion, I = N t 0 < t < 1 (Latané, 1981). The theory has been empirically supported by s , a meta-analysis of conformity experiments using Asch’s line-judgment task (Bond & Smith, 1996).

There are informational and normative motivations underlying social conformity, the
former based on the desire to form an accurate interpretation of reality and behave

1332 American Behavioral Scientist 58(10)

correctly, and the latter based on the goal of obtaining social approval from others (Cialdini
& Goldstein, 2004). However, the two are interrelated and often difficult to disentangle
theoretically as well as empirically. In addition, both goals act in service of a third under-
lying motive to maintain one’s positive self-concept (Cialdini & Goldstein, 2004).

Both self-confirmation and social conformity are extensive and strong and they
appear under many guises in life, both online and in physical interactions. In what fol-
lows, we consider both processes to understand the users’ reactions to online recom-
mender systems.

Online recommender systems. Compared to traditional sources of recommendations—
peers such as friends and coworkers, experts such as movie critics, and industrial
media such as Consumer Reports—online recommender systems combined personal-
ized recommendations sensitive to people’s interests and independently reporting
other peoples’ opinions and reviews. One popular example of a successful online rec-
ommender system is the Amazon product recommender system (Linden, Smith, &
York, 2003).

Understanding users in online recommender systems. In computer science and the
human–computer interaction (HCI) community, for a long time, most research in rec-
ommender systems has focused on creating accurate and effective algorithms (e.g.,
Breese, Heckerman, & Kadie, 1998). Recently, researchers have realized that recom-
mendations generated by standard accuracy metrics, although generally useful, are not
always the most useful to users (McNee, Riedl, & Konstan, 2006). Researchers started
building new user-centric evaluation metrics (Pu, Chen, & Hu, 2011; Xiao & Ben-
basat, 2007). There are few empirical studies investigating the basic psychological
processes underlying the interaction of users with recommendations, and none of them
addresses both self-confirmation and social conformity. As mentioned above, Cosley
and his colleagues (2003) studied conformity in movie rating sites and showed that
people’s ratings are significantly influenced by other users’ ratings. But they did not
consider the effects of self-confirmation or the effects of different levels of social con-
formity pressures. Schwind, Buder, and Hesse (2011) studied how to overcome users’
confirmation bias by providing preference-inconsistent recommendations. However,
they represented recommendations as search results rather than recommendations
from humans and thus did not investigate the effects of social conformity. Further-
more, their task was more related to logical inference rather than purchase decision
making.

In the area of marketing and customer research, studies about the influence of rec-
ommendations are typically subsumed under personal influence and word-of-mouth
research (Senecal & Nantel, 2004). Past research has shown that word-of-mouth plays
an important role in consumer buying decisions, and use of the Internet brings new
threats and opportunities for marketing (Hennig-Thurau, Gwinner, Walsh, & Gremler,
2004; Senecal & Nantel, 2004; Stauss, 1997). There were several studies specifically
investigating social conformity in product evaluations (Burnkrant & Cousineau, 1975;
Cohen & Golden, 1972; Pincus & Waters, 1977). Although they found substantial

Zhu and Huberman 1333

effects of others’ evaluations on people’s own judgments, the effects were not always
significantly stronger when the social conformity pressures are stronger.1 In Burnkrant
and Cousineau’s (1975) and Cohen and Golden’s (1972) experiments, participants
were exposed to evaluations of coffee with high uniformity or low uniformity. Both
results showed that participants did not exhibit significantly increased adherence to
others’ evaluation in the high uniformity condition (although in Burnkrant and
Cousineau’s experiments, the participants recognized that the difference between high
and low uniformity was significant). On the other hand, in Pincus and Waters’s (1977)
experiments (college students rated the quality of one paper plate while exposed to
simulated quality evaluations of other raters), it was found that conformity effects are
stronger when the evaluations are more uniform.

In summary, although previous research showed that others’ opinions can influence
people’s own decisions, none of that research addresses both the self-confirmation and
social conformity mechanisms that underlie choice among several recommendations.
In addition, regarding the effects of increasing social conformity pressures, experi-
ments using Asch’s line-judgment tasks supported that people are more likely to be
influenced when facing stronger social pressures, whereas the findings of studies
using product evaluation tasks were mixed.

Our experiments address how often people reverse their own opinions when con-
fronted with other people’s preferences, especially when facing different levels of con-
firmation and conformity pressures. The hypothesis is that people are more likely to
reverse their opinions when the reversion causes less self-inconsistency (the confirma-
tion pressure is weaker) or the opposing social opinions are stronger (the conformity
pressure is stronger).

Method

Experimental Design

We conducted a series of online experiments. All participants were asked to go to the
website of Rankr (Luon et al., 2012) to make a series of pairwise comparisons with or
without knowing other people’s preferences (Figure 1). The pictures were collected
from Google Images. We wanted to determine whether people reverse their choices by
seeing others’ preferences.

Basic idea of the experiment. Participants were asked to provide their preferences
between the same pair of items twice. The first time the participant encountered the
pair, he or she made a choice without the knowledge of others’ preferences. The sec-
ond time the participant encountered the same pair, he or she made a choice with the
knowledge of others’ preferences. Social influence was measured according to whether
or not people switched their choices between the first time and second time they
encountered the same pair.

To manipulate the pressure to confirm people’s own opinions, we changed the time
between the two decisions. In the short interval condition, people first compared two

1334 American Behavioral Scientist 58(10)

Figure 1. Example of pairwise comparisons in Rankr.

pictures on their own and they were then immediately asked to make another choice
with available information about others’ preferences. When their memories were fresh,
reversion led to strong inconsistency and dissonance with other people’s choices
(strong confirmation pressure). However, in the long interval condition, participants
compared pairs of items in the beginning of the test followed by several distractor
pairs. This was followed again by the same pairs that the participant had previously
compared, but this time with augmented information about others’ preferences. In this
case, participants’ memories of their previous choices decay, so the pressure to con-
firm their own opinions is less explicit.

To manipulate the social pressure, we changed the number of opposing opinions
that the participants saw when making the second decision. We selected four levels:
the opposing opinions were 2 times, 5 times, 10 times, or 20 times as many as the
number of people who supported their opinions.

In the following section, the details of experimental conditions are discussed.

Conditions. The experimental design was 2 (baby pictures and loveseat pictures) × 3
(short interval, long interval, and control) × 4 (ratio of opposing opinions to supporting
opinions: 2:1, 5:1, 10:1, and 20:1). Participants were recruited from Amazon’s
Mechanical Turk (mTurk) and were randomly assigned into one of six conditions
(baby-short, baby-long, baby-control, loveseat-short, loveseat-long, and loveseat-con-
trol) and made four choices with different levels of conformity pressure.

In the baby condition, people were asked to compare 23 or 24 pairs of baby pictures
by answering the question, “Which baby looks cuter on a baby product label?” Note
that the Caucasian baby pictures in Figure 1 are examples. We displayed baby pictures
from different races in the experiment. In the loveseat condition, the question was,
“Your close friend wants your opinion on a loveseat for their living room. Which one
do you suggest?” People needed to make 23 or 24 choices.

In the short interval condition, people first compared two pictures on their own and
they were then immediately asked to make another choice with available information

Zhu and Huberman 1335

Experimental
pair

Experimental pair displaying
others preferences which are
against people’s previous
choice

Pair

Short interval:

Long interval:

Long interval control:

Pairi, j Pairi, j Pairi, j Pair i, j Pair displaying
others’ preferences

1, 2 2, 1 3, 4 3, 4 5, 6 5, 6 7, 8 7, 8 9,10 9,10 11,12 11,12 13,14 13,14 15,16 15,16 17,18 17,18 19,20 19,20 21,22 21,2223,24 23,24

1, 2 3, 4

Honesty test

Honesty test

Honesty test

5, 6 6, 5 7, 8 9,10 11,1213,14 15,16 17,18 19,20 21,22 23,24 25,26 27,28 8, 7 14,13 29,30 31,32 33,34 2, 1 35,36 16,15

1, 2 3, 4 5, 6 6, 5 7, 8 9,10 11,1213,14 15,16 17,18 19,20 21,22 23,24 25,26 27,28 8, 7 14,13 29,30 31,32 33,34 2, 1 35,36 16,15

Figure 2. Example displaying orders in each condition.

about others’ preferences. Furthermore, we tested whether people would reverse their
first choice under four levels of social pressure: when the number of opposing opin-
ions was 2 times, 5 times, 10 times, and 20 times as many as the number of people who
supported their opinions. The numbers were randomly generated.2 In addition to these
8 experimental pairs, we added 14 noise pairs and an honesty test composed of 2 pairs
(24 pairs in total; see Figure 2 for an example). In this condition, noise pairs also con-
sisted of consecutive pairs (a pair with social information immediately after the pair
without social information). However, others’ opinions were either indifferent or in
favor of the participants’ choices. We created an honesty test to identify participants
who cheated the system and quickly clicked on the same answers. The test consisted
of 2 consecutive pairs with the same items but with the positions of the items
exchanged. Participants needed to make the same choices among these consecutive 2
pairs in order to pass the honesty test. The relative orders of experimental pairs, noise
pairs, and honesty test in the sequence and the items in each pair were randomly
assigned to each participant.

In contrast with the short interval condition, where people were aware that they
reversed their choices, in the long interval condition, we manipulated the order of
display and the item positions so that the reversion was less explicit. People first com-
pared pairs of the items without knowing others’ preferences, and then after 11.5 pairs
later, on average, we showed the participants the same pair (with the positions of items
in the pair exchanged) and others’ opinions. Similarly, with the short interval condi-
tion, we showed 8 experimental pairs to determine whether people reversed their pre-
vious choices with increasing pressures of social influence. In addition, we showed 13
noise pairs (9 without others’ preferences and 4 with others’ preferences) and per-
formed an honesty test (see Figure 2 for an example).

By increasing the time between two choices, we blurred the people’s memories of
their choices so as to exert a subtle confirmation pressure. However, as people pro-
ceeded with the experiment, they were presented with new information to process.
This new information might lead them to think in a different direction and change their
own opinions regardless of social influence. To control for this confounding factor, we

1336 American Behavioral Scientist 58(10)

added a long interval control condition, where the order of the pairs was the same as
with the long interval condition but without showing the influence of others.

Procedures. We conducted our experiment on Amazon’s mTurk (Kittur, Chi, & Suh,
2008). The recruiting messages stated that the objective of the study was to do a survey
to collect people’s opinions. Once mTurk users accepted the task, they were asked to
click the link to Rankr, which randomly directed them to one of the six conditions.
This process was invisible to them.

First, the participants were asked to provide their preferences about 23 or 24 pairs
of babies or loveseats. They were then directed to a simple survey. They were asked to
report their age and gender and answer two 5-Likert scale questions. The questions
were as follows: “Is showing others’ preferences useful to you?” and “How much does
showing others’ preferences influence your response?” After filling out the survey, a
unique confirmation code was generated and displayed on the webpage. Participants
needed to paste the code back to the mTurk task. We matched mTurk users with the
participants of our experiments using the confirmation code, allowing us to pay mTurk
users according to their behaviors. We paid $0.35 for each valid response.

Participants. We collected 600 responses. Of this number, we omitted 37 responses
from 12 people who completed the experiment multiple times; 22 incomplete
responses; 1 response from someone who did not conform to the participation require-
ments (i.e., being at least 18 years old); and 107 responses from those who did not pass
the honesty test. These procedures left 433 valid participants in the sample—about
72% of the original number. According to participant self-reporting, 40% were women;
ages ranged from 18 to 82 years with a median age of 27 years. Geocoding3 the IP
addresses of the participants revealed that 57% were from India, 25% were from the
United States, and the remaining 18% of participants came from more than 34 differ-
ent countries.

The numbers of participants in each condition were as follows: baby-short, 72;
baby-long, 91; baby-control, 49; loveseat-short, 75; loveseat-long, 99; and loveseat-
control, 47.4

People spent a reasonable amount of time on each decision (average = 6.6 seconds;
median = 4.25 seconds).

Among the 433 responses, 243 left comments in the open-ended comments section
at the end of the experiments. Most of them said that they had a good experience when
participating in the survey. (They were typically not aware that they were in an
experiment.)

Measures. The measures are as follows:

1) Reversion: whether people reverse their preferences after knowing others’
opinions

2) Social conformity pressures: the ratio of opposing opinions to supporting
opinions

3) Decision time: the time (in seconds) people spent in making each decision

Zhu and Huberman 1337

0.00%

5.00%

10.00%

15.00%

20.00%

25.00%

30.00%

35.00%

40.00%

Baby Loveseat

Short Interval

Long Interval

Long Interval
Control

Figure 3. Reversion rate by conditions.

4) Demographic information: age and gender
5) Self-reported usefulness of others’ opinions
6) Self-reported level of being influenced

Results

1. Did people reverse their opinions by others’ preferences when facing
different confirmation pressures?

Figure 3 shows the reversion rate as a function of the conditions that we manipulated
in our experiment. First, we found out that content does not matter, that is, although
baby pictures are more emotionally engaging than loveseat pictures, the patterns are
the same. The statistics test also shows that there is no significant difference between
the baby and the loveseat results, t(431) = 1.35, p = .18.

Second, in the short interval condition, the reversion rate was 14.1%, which is
higher than zero [the results of the t test are t(146) = 6.7, p < .001].

Third, the percentage of people who reversed their opinions was as high as
32.5% in the long interval condition, significantly higher than the long interval
control condition (10.1%), which measures the effects of other factors leading to
reversion regardless of the social influence during the long interval. t test shows
that this difference is significant, t(284) = 6.5, p < .001. We can therefore conclude that social influence contributes to approximately 22.4% of the reversion of opin- ions observed.

To summarize the results, in both the long and the short interval conditions, others’
opinions significantly swayed people’s own choices (22.4% and 14.1%5). The effect
size of social influence was larger when the self-confirmation pressure was weaker
(i.e., the time between the two choices is larger).

1338 American Behavioral Scientist 58(10)

Table 1. Linear Regression Predicting the Reversion Percentage.

Predictor Coefficient Standard Error P Value

Conditiona .839 .059 < .001 Ratio of opposing opinions .563 .184 .038 Square ratio of opposing opinions –.142 .049 .045 Intercept –2.42 .151 < .001 Adjusted R2 .96

Note. The squared ratio of opposing opinions has a significant negative value (–0.142, p = .045).
a1 = long interval; 0 = short interval.

2. Were people more likely to reverse their own preferences when
more people were against them?

It is interesting that we saw an increasing and then decreasing trend when the opposing
opinions became exponentially stronger (from 2×, 5×, 10×, to 20×). The condition
with the most uniform opposing opinions (20×) was not more effective in reversing
people’s own opinions than the moderate opposing opinions (5× and 10×). The regres-
sion results are shown in Table 1. Note that the squared ratio of opposing opinions has
a significant negative value (Coef. = −0.142, p < .05), suggesting that the returning effect is statistically significant.

These results might be explained by Brehm’s (1966) finding of psychological
reactance. According to Brehm, if an individual perceives one’s freedom as being
reduced or threatened with reduction, one will become aroused to maintain or
enhance one’s freedom. The motivational state of arousal to reestablish or enhance
one’s freedom is called psychological reactance. Therefore, if the participants per-
ceived the uniform opposing opinions as a threat to their freedom to express their
own opinions, their psychological reactance might be aroused to defend and confirm
their own opinions.

These results can also be explained in terms of Wu and Huberman’s (2010) findings
about online opinion formation. In their work, they used the idea of maximizing the
effect that individuals have on the average rating of items to explain the phenomenon
that later reviews tend to show a big difference from earlier reviews on Amazon.com
and IMDB.com.

We can use the same idea to explain our results. Social influence in product recom-
mendations is not just a one-way process. People are not just passively influenced by
others’ opinions but also want to maximize their effect on other people’s future decision
making (e.g., in our experiments, according to our recruiting messages, participants
would assume that their choices would be recorded in the database and shown to others;
in real life, people like to influence their friends and family). We assume that the influ-
ence of an individual on others can be measured by how much his or her expression will
change the average opinion. Suppose there are X1 supporting opinions and X 2 oppos-
ing opinions and that X 2 > X1 . A person’s choice c (0 indicates confirming his or her

https://IMDB.com

https://Amazon.com

Zhu and Huberman 1339

Table 2. Logistic Regression Predicting the Reversion.

Predictor Coefficient Standard Error P Value

Conditionb 1.26 .152 < .001 Age –.006 6.89e-3 .407 Gender .067 .143 .642 Self-reported usefulness .164 .070 .020 Self-reported influence level .334 .072 < .001 Standardized first decision time .323 .065 < .001 Log likelihood –657.83

b1 = long interval; 0 = short interval.

own choice; 1 indicates conforming to others) can move the average percentage of
opposing opinions from X 2 / (X1 + X 2 ) to (X 2 + c) / (X1 + X 2 +1). So, the influence

X 2 X 2 + con the average opinion is . A simple derivation shows that to −
X + X X + X +11 2 1 2

maximize the influence on average opinion, people need to stick to their own choices
and vote for the minority. Then, their influence gain will be stronger when the differ-
ence between existing majority opinions and minority ones is larger. Therefore, the
motivation to exert influence on other people can play a role in resisting the social
conformity pressure and lead people to confirm their own decisions, especially when
facing uniform opposing opinions.

3. What else predicts the reversion?

We used a logistic regression model to predict the decision-level reversion with the
participants’ age, gender, self-reported usefulness of the recommendation system, self-
reported level of being influenced by the recommendation system, and standardized
first decision time (as shown in Table 2). Note that standardized first decision time =
(time in this decision – this person’s average decision time) / this person’s standard
deviation. So, “first decision time” is an intrapersonal variable.

The results showed that age and gender do not significantly predict reversion (p =
.407, p = .642). Self-reported influence level has a strong prediction power (Coef. =
0.334, p < .001), which is reasonable. The interesting fact is that decision time, a simple behavioral measure, also predicts reversion very well (Coef. = 0.323, p < .001). The longer people spent on the decisions, the more equivalent the two choices were for them. According to Festinger’s (1954) theory, the more equivocal the evidence, the more people rely on social cues. Therefore, the more time people spend on a choice, the more likely they are to reverse this choice and conform to others later on.

Discussion

On one hand, the phenomenon we found (i.e., the returning effects of strong influence
pressure) is quite different from the classic line-judgment conformity studies (Asch,

1340 American Behavioral Scientist 58(10)

1956; Latané, 1981). Notice that these experiments used questions with only one sin-
gle correct answer (Asch, 1956). In contrast, in our experiment we examined the social
influence on people’s subjective preferences among similar items, which might result
in such a different phenomenon. On the other hand, our findings reconcile the mixed
results of studies investigating social conformity in product evaluation tasks (which
are often subjective tasks) (Burnkrant & Cousineau, 1975; Cohen & Golden, 1972;
Pincus & Waters, 1977). Due to the returning effects, strong conformity pressure is not
always more effective in influencing people’s evaluations compared to weak confor-
mity pressure.

In addition, in our experiment we did not look at social influence in scenarios where
the uncertainty level is quite high. For example, in settings such as searching recom-
mendations for restaurants or hotels where people have never been, the level of uncer-
tainty is high and people need to rely on other cues. However, in our experimental
setting, people can confidently make their choice by comparing the pictures of the
settings. Therefore, some caution is needed when trying to generalize our results to
other settings with high uncertainty.

Regarding the tasks used in the experiment, the choice of “which baby looks cuter
on a product label” involves emotions and subjective feelings. Alternatively, similar
choices include the preferences of iTunes music or YouTube’s commercial videos. In
contrast, the other type of question, “which loveseat is better,” is less emotionally
engaging. In this case, people are more likely to consider usability factors, such as
color and perceived comfort, when making these types of choices. Results showed that
the effect of social influence on these two different types of choices is similar and
consistent, which suggests the general applicability of the results.

There might be concern about cultural differences between the global nature of the
participants and the Western-style nature of the tasks. We point out that this intercul-
tural preference difference is inherent in the participant. Therefore, it is assumed to be
consistent throughout the test and should not affect a particular person’s preference
changes and will not confound the results.

During our research, we invented an experimental paradigm to easily measure the
effects of social influence on people’s decision making (i.e., the reversion) and manip-
ulate conditions under which people make choices. This paradigm can be extended to
scenarios beyond those of binary choices, to the effect of recommendations from
friends as opposed to strangers, and to social influence, which could be varied with
different visualizations for the recommendations (e.g., displaying the average ratings
from all the users, or showing the ratio of opposed opinions, or showing only the num-
ber of positive reviews such as “likes”).

Limitations & Future Work

In our experiments, we examined whether people reverse their choices when facing
different ratios of opposing opinions versus supporting opinions (2×, 5×, 10×, and
20×). To further investigate the relationship between the ratio of the opposing opinions
and the tendency to revert, it would be better to include more fine-grained conditions in

Zhu and Huberman 1341

the ratio of opposing opinions. The ideal situation would be a graph with the continuous
opposing versus supporting ratio as the x-axis and the reversion rate as the y-axis.

Also, additional manipulation checks or modification of the design of the experi-
ment would be needed to establish whether processes such as psychological reactance
or the intent to influence others have been operating. For example, we could measure
the degree of perceived freedom in a given task. In addition, it would be revealing to
manipulate the visibility of people’s choices to other participants to see if the intent of
influencing others exists.

Regarding the usage of mTurk as a new source of experimental data, we agree with
Paolacci, Chandler, and Ipeirotis (2010) that “workers in Mechanical Turk exhibit the
classic heuristics and biases and pay attention to directions at least as much as subjects
from traditional sources” (p. 417). In particular, compared to recruiting people from a
college campus, we believe that the use of mTurk has a lower risk of introducing the
Hawthorne Effect (i.e., people alter their behaviors due to the awareness that they are
being observed in experiments), which is itself a form of social influence and might
contaminate our results.

In our experiment, we used several methods, such as an honesty test and IP address
check, to further ensure that we collected earnest responses from mTurk workers. The
average time they spent on the task, the statistically significant results of the experi-
ment, and the comments that participants left all indicate that our results are believ-
able. However, there is still a limitation to our honesty test. On one hand, the
presentation of two consecutive pairs with the positions of items switched was unable
to identify all the users who tried to cheat the system by randomly clicking on the
results, which added noise to our data. On the other hand, the honesty test might also
exclude some earnest responses. Thus, it is possible that immediately after people
made a choice, they regretted it.

Conclusion & Implications

In this article, we presented results of a series of online experiments designed to inves-
tigate whether online recommendations can sway people’s own opinions. These exper-
iments exposed participants making choices to different levels of confirmation and
conformity pressures.

Our results show that people’s own choices are significantly swayed by the per-
ceived opinions of others. The influence is weaker when people have just made their
own choices. In addition, we showed that people are most likely to reverse their
choices when facing a moderate, as opposed to large, number of opposing opinions.
And last but not least, the time people spend making the first decision significantly
predicts whether they will reverse their own later on.

Our results have three implications for consumer behavior research as well as online
marketing strategies. (a) The temporal presentation of the recommendation is impor-
tant; it will be more effective if the recommendation is not provided immediately after
the consumer has made a similar decision. (b) The fact that people can reverse their
choices when presented with a moderate countervailing opinion suggests that rather

1342 American Behavioral Scientist 58(10)

than overwhelming consumers with strident messages about an alternate product or
service, a more gentle reporting of a few people having chosen that product or service
can be more persuasive than stating that thousands have chosen it. (c) Equally impor-
tant is the fact that a simple monitoring of the time spent on a choice is a good indicator
of whether or not that choice can be reversed through social influence. There is enough
information in most websites to capture these decision times and act accordingly.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship,
and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of
this article.

Notes

1. Unlike other conformity experiments such as line-judgment where the pressure of social
conformity is manipulated by increasing the number of “unanimous” majority, experi-
ments about social influence in product evaluation (Burnkrant & Cousineau, 1975; Cohen
& Golden, 1972; Pincus & Waters, 1977) usually manipulate the pressure of social influence
by changing the degree of uniformity of opinions. As discussed in Cohen and Golden (1972),
since it is seldom that no variation exists in the advice or opinions in reality, the latter method
is more likely to stimulate participants’ real reactions. We also use the latter method in our
experiment by manipulating the ratio of opposing opinions versus supporting opinions.

2. We first generated a random integer from 150 to 200 as the total participants. Then, we
generated the number of people holding different opinions according to the ratio. Here are
a few examples: 51 versus 103 (2×), 31 versus 156 (5×), 16 versus 161 (10×), and 9 versus
181 (20×).

3. MaxMind GeoLite (http://dev.maxmind.com/geoip/legacy/geolite/) was used to geocode
the IP addresses and self-reports a 99.5% accuracy rate.

4. Among the 600 responses, originally 20% were assigned for baby-strong; 20% for baby-
weak; 10% for baby-control; 20% for loveseat-strong; 20% for loveseat-weak; and 10%
for loveseat-control. The valid responses in short interval conditions were fewer than the
ones in long interval conditions because the short interval condition had a higher failure
rate in the honesty test. The reason might be that the short interval condition had more
repetitive pairs, fewer new items, and more straightforward patterns, leading to boredom
and casual decisions, which in turn caused failure in the honesty tests.

5. To calibrate the magnitude of our results, we point out that our results are of the same
magnitude as the classic line-judgment experiments. According to a 1996 meta-analysis of
line-judgment experiment consisting of 133 separate experiments and 4,627 participants,
the average conformity rate is 25% (Bond & Smith, 1996).

References

Asch, S. E. (1956). Studies of independence and conformity: I. A minority of one against a
unanimous majority. Psychological Monographs, 70(9), 1-70.

http://dev.maxmind.com/geoip/legacy/geolite/

Zhu and Huberman 1343

Bacon, F. (1939). Novum organum. In E. A. Burtt (Ed.), The English philosophers from Bacon
to Mill (pp. 24-123). New York: Random House. (Original work published 1620)

Bendor, J., Huberman, B. A., & Wu, F. (2009). Management fads, pedagogies, and other soft
technologies. Journal of Economic Behavior & Organization, 72(1), 290-304.

Bond, R., & Smith, P. B. (1996). Culture and conformity: A meta-analysis of studies using
Asch’s (1952b, 1956) line judgment task. Psychological Bulletin, 119(1), 111-137.

Breese, J. S., Heckerman, D., & Kadie, C. (1998). Empirical analysis of predictive algorithms
for collaborative filtering. In Proceedings of the 14th conference on uncertainty in artificial
intelligence. San Francisco, CA: Morgan Kaufmann.

Brehm, J. W. (1966). A theory of psychological reactance. New York, NY: Academic Press.
Burnkrant, R. E., & Cousineau, A. (1975). Informational and normative social influence in

buyer behavior. Journal of Consumer Research, 2(3), 206-215.
Cialdini, R. B., & Goldstein, N. J. (2004). Social influence: Compliance and conformity. Annual

Review of Psychology, 55, 591-621.
Cohen, J. B., & Golden, E. (1972). Informational social influence and product evaluation.

Journal of Applied Psychology, 56(1), 54-59.
Cosley, D., Lam, S. K., Albert, I., Konstan, J. A., & Riedl, J. (2003). Is seeing believing? How

recommender system interfaces affect users’ opinions. In Proceedings of the SIGCHI confer-
ence on human factors in computing systems (CHI ‘03) (pp. 585-592). New York, NY: ACM.

Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7, 117-140.
Festinger, L. (1957). A theory of cognitive dissonance. Stanford, CA: Stanford University Press.
Hennig-Thurau, T., Gwinner, K. P., Walsh, G., & Gremler, D. D. (2004). Electronic word-

of-mouth via consumer-opinion platforms: What motivates consumers to articulate them-
selves on the Internet? Journal of Interactive Marketing, 18(1), 38-52.

Kittur, A., Chi, E. H., & Suh, B. (2008). Crowdsourcing user studies with Mechanical Turk. In
Proceedings of the twenty-sixth annual SIGCHI conference on human factors in computing
systems (CHI ‘08) (pp. 453-456). New York, NY: ACM.

Latané, B. (1981). The psychology of social impact. American Psychologist, 36(4), 343-356.
Linden, G., Smith, B., & York, J. (2003). Amazon.com recommendations: Item-to-item col-

laborative filtering. Internet Computing, 7(1), 76-80.
Lorenz, J., Rauhut, H., Schweitzer, F., & Helbing, D. (2011). How social influence can under-

mine the wisdom of crowd effect. PNAS, 108(22), 9020-9025.
Luon, Y., Aperjis, C., & Huberman, B. A. (2012). Rankr: A mobile system for crowdsourcing

opinions. In Mobile computing, applications, and services (pp. 20-31). Berlin: Springer.
McNee, S. M., Riedl, J., & Konstan, J. A. (2006). Being accurate is not enough: How accuracy

metrics have hurt recommender systems. CHI Extended Abstracts, pp. 1097-1101.
Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review

of General Psychology, 2(2), 175-220.
Paolacci, G., Chandler, J., & Ipeirotis, P. G. (2010). Running experiments on Amazon

Mechanical Turk. Judgment and Decision Making, 5(5), 411-419.
Pincus, S., & Waters, L. K. (1977). Informational social influence and product quality judg-

ments. Journal of Applied Psychology, 62(5), 615-619.
Pu, P., Chen, L., & Hu, R. (2011). A user-centric evaluation framework for recommender sys-

tems. In Proceedings of the fifth ACM conference on recommender systems (RecSys ‘11)
(pp. 157-164). New York, NY: ACM.

Ross, L., Lepper, M. R., & Hubbard, M. (1975). Perseverance in self-perception and social per-
ception: Biased attributional processes in the debriefing paradigm. Journal of Personality
and Social Psychology, 32, 880-892.

https://Amazon.com

1344 American Behavioral Scientist 58(10)

Salganik, M. J., Dodds, P. S., & Watts, D. J. (2006). Experimental study of inequality and
unpredictability in an artificial cultural market. Science, 311(5762), 854-856.

Schwind, C., Buder, J., & Hesse, F. W. (2011). I will do it, but I don’t like it: User reactions to
preference-inconsistent recommendations. In Proceedings of the 2011 annual conference
on human factors in computing systems (CHI ‘11) (pp. 349-352). New York, NY: ACM.

Senecal, S., & Nantel, J. (2004). The influence of online product recommendations on consum-
ers’ online choices. Journal of Retailing, 80(2), 159-169.

Stauss, B. (1997). Global word of mouth. Service bashing on the Internet is a thorny issue.
Marketing Management, 6(3), 28-30.

Wason, P. C. (1960). On the failure to eliminate hypotheses in a conceptual task. The Quarterly
Journal of Experimental Psychology, 12, 129-140.

Wu, F., & Huberman, B. A. (2010). Opinion formation under costly expression. ACM
Transactions on Intelligent Systems and Technology (TIST), 1(1), 5.

Xiao, B., & Benbasat, I. (2007). Ecommerce product recommendation agents: Use, characteris-
tics, and impact. MIS Quarterly, 31(1), 137-209.

Author Biographies

Haiyi Zhu is a PhD candidate in the Human-Computer Interaction Institute at Carnegie Mellon
University. Her research interests include human-computer interaction, social computing, and
online communities. She is specifically interested in conducting large-scale data analysis and
field experiments to understand the underlying principles in online social computing systems.
She received a BA in computer science at Tsinghua University in 2009 and a master’s in human
computer interaction from Carnegie Mellon University in 2012.

Bernardo A. Huberman is a Senior HP Fellow and director of the Social Computing Research
Group at HP Labs, which focuses on methods for harvesting the collective intelligence of groups
of people in order to realize greater value from the interaction between users and information.
Huberman received his PhD in physics from the University of Pennsylvania and is currently a
consulting professor in the Department of Applied Physics at Stanford University. He has been
a visiting professor at the Niels Bohr Institute in Denmark, the University of Paris, and Insead,
the European School of Business in Fontainebleau, France.

Can a Rude Waiter Make Your Food Less Tasty? Social Class Differences in
Thinking Style and Carryover in Consumer Judgments

Jaehoon Lee
Southern Illinois University

Accepted by Amna Kirmani, Editor; Associate Editor, Youjae Yi

Building on the notion that cognitive processes vary across social classes, we predict that social class shapes
thinking style, which in turn affects consumer judgments. In doing so, we employ service failure domains as
a way to understand social class effects. Across four studies, we show that, when faced with a failure incident
occurring in one service dimension (e.g., rude employees), consumers in the low social class, relative to those
in the high social class, carry over to influence their evaluations of the other service dimensions (e.g., food
quality) that are unrelated to the failure incident. We further show that low-class consumers favor a holistic
style of thinking, whereas high-class consumers favor an analytic style of thinking and that these differences
in thinking style account for the carryover effects on evaluations. The pattern of the effects exists when the
service failure is perceived to be severe rather than minor.

Keywords Social class; Thinking style; Carryover; Service failure; Failure severity

Introduction

Social class has long been a topic of interest in mar-
keting. Consumers often identify themselves as
belonging to certain social classes, and those of dif-
ferent social classes are characterized by distinct
lifestyles, preferences, and choices. Much of the
research on social class has examined class differ-
ences in attitudinal and behavioral domains that
include purchase decision-making (Coleman, 1983;
Martineau, 1958; Williams, 2002), financial choices
(Henry, 2005), price judgments (Gaston-Breton &
Raghubir, 2013), preferences for cultural tastes

€(Holt, 1998; Ust€ & Holt, 2010), and spending uner
patterns in emerging economies (Kamakura &
Mazzon, 2013). However, we understand relatively
little about cognitive processes that may underlie
these class differences in consumer contexts.

There is reason to believe that social class shapes
consumers’ cognitive styles. It is well understood
that cognitive processes are malleable and suscepti-
ble to social environments, and become largely auto-
matic and unconscious (Nisbett, Peng, Choi, &
Norenzayan, 2001). More relevant to the present
research, social cognitive perspectives suggest that

Received 26 October 2016; accepted 23 October 201

7

Available online 27 November 2017
This research was supported in part by a grant from Carolan

Research Institute.
Correspondence concerning this article should be addressed to

Jaehoon Lee, Department of Marketing, College of Business,
Southern Illinois University, Rehn Hall 223A, 1025 Lincoln Drive,
Carbondale, IL 62901, USA. Electronic mail may be sent to
jhlee@business.siu.edu.

cognitive processing styles are dependent on the
availability of social and economic resources that
vary across social classes (for a review, see Kraus,
Piff, Mendoza-Denton, Rheinschmidt, & Keltner,
2012). Integrating these perspectives, we propose
that consumers of different social classes exhibit sys-
tematic differences in cognitive styles, which in turn
influence their judgments. Specifically, we examine
class differences in analytic and holistic styles of
thinking. Analytic thinking involves detachment of
individual objects from their context, whereas holis-
tic thinking involves connections between objects
and their context as a whole (for a review, see
Nisbett et al., 2001). We highlight the role of thinking
style in social class effects on consumer judgments.
To demonstrate the effects, we employ service failure
domains that vary in service dimensions.

We provide evidence that for low-class individu-
als, who perceive situations as interconnected and
holistic, their evaluations of the service dimension
where a failure incident occurs carry over to influ-
ence their evaluations of the other dimension that
has no direct link to the failure incident. For exam-
ple, low-class consumers, who experience unpleas-
ant interactions with a restaurant employee, may
extend their negative impression of the employee to
their evaluations of food quality. We also find that
this carryover effect occurs to a lesser degree for

© 2017 Society for Consumer Psychology
All rights reserved. 1057-7408/2018/1532-7663/28(3)/450–46

5

DOI: 10.1002/jcpy.1020

http://orcid.org/0000-0002-3584-4129

http://orcid.org/0000-0002-3584-4129

mailto:jhlee@business.siu.edu

�

�

Social Class Differences in Carryover Effects 451

high-class consumers, who perceive situations as
isolated and analytic. We further document that
thinking style serves as the underlying mechanism
that accounts for the carryover effect.

Social Class and Cognitive Patterns

Social class is a multidimensional concept that
reflects both objective and subjective components of
socioeconomic status (Kraus et al., 2012; Snibbe &
Markus, 2005). Objective social class is rooted in
material resources and assessed with objective indi-
cators such as income, educational attainment, and
occupational prestige (Adler & Snibbe, 2003). In
comparison, subjective social class is rooted in self-
perceptions of social standing relative to others
and assessed with one’s perceived rank in society
or in one’s community (Adler, Epel, Castellazzo, &
Ickovics, 2000). While these objective and subjective
components of social class are related (Johnson &
Krueger, 2006) and often used interchangeably
(Adler & Snibbe, 2003), it is important to note that
subjective class better accounts for psychological
characteristics (Adler et al., 2000), cognitive patterns
(Kraus, Piff, & Keltner, 2009), and judgments
(Greitemeyer & Sagioglou, 2016; Smith & Pettigrew,
2014) than does objective class. Thus, it is reasonable
to predict that subjective class, relative to objective
class, will be a stronger predictor for class differ-
ences in thinking styles in the present research.

Social class influences aspects of the self, according
to models of agency that represent implicit frame-
works of ideas and practices about how to be a nor-
matively good person in the world (Carey & Markus,
2016; Markus & Kitayama, 2003; Snibbe & Markus,
2005; Stephens, Markus, & Townsend, 2007). The
relative abundance of resources and opportunities
among high-class individuals promotes indepen-
dence and encourages them to influence others and
environments (disjoint models of agency). In contrast,
the relative scarcity of resources and opportunities
among low-class individuals promotes interdepen-
dence and encourages them to adapt to others and
environments (conjoint models of agency).

Of particular relevance to the present research is
evidence showing that such resource disparities
across social classes, in turn, shape the way people
perceive and respond to their social environments
within a culture. Low-class individuals, who have rel-
ative resource scarcity and perceived lower rank, feel
constrained while pursuing their goals and interests,
and their actions are often affected by external forces
outside of their own control (Kraus et al., 2012). As a

result, low-class individuals tend to focus their atten-
tion on external contexts and favor contextual expla-
nations of social events over dispositional
explanations (Kraus et al., 2012). For example, Kraus
et al. (2009) demonstrated that low-class individuals
attributed economic inequality to contextual factors
such as the economic structure of society or political
influence over dispositional factors such as hard
work, ability, or money management skills. Further-
more, Kraus, Côte, and Keltner (2010) found that
low-class individuals perceived the emotions of other
individuals more accurately than their high-class
counterparts and that this tendency was attributed to
their focus on features of the external social context.

Conversely, high-class individuals, who have rel-
ative resource abundance and perceived higher
rank, feel free to pursue the goals and interests they
choose for themselves (Johnson & Krueger, 2005;
Lachman & Weaver, 1998) with relatively little con-
cern about their material costs. As a result, high-
class individuals tend to focus on their own internal
states, ignoring contextual influences, and favor
explanations of social events in terms of individual
influences (Kraus et al., 2012). For example, high-
class managers are more likely to blame underper-
formance of an employee on the personalities and
abilities of the employee rather than contextual fac-
tors that hinder performance such as poor job
design (Côte, 2011). Similarly, high-class individu-
als, relative to their low-class counterparts, tend to
think that the realization of their hopes depends
more on themselves and less on external contexts
(Lamm, Schmidt, & Trommsdorff, 1976). In sum,
when explaining social events, people in the low
class tend to focus on contextual factors, while
those in the high class tend to focus on individual
traits. These differences between low and high-class
individuals in cognitive orientations may reflect
holistic and analytic thinking styles, respectively.

Holistic and Analytic Thinking

The main distinction between holistic and analytic
styles of thinking lies in the way people attend to
the environment. Holistic thinkers attend primarily
to interconnections between an object and its con-
text and thus are known to be context-dependent.
Conversely, analytic thinkers attend primarily to an
individual object separate from its context and thus
are known to be context-independent.

Different thinking styles arise from a variety of
factors. One widely recognized factor is a cross-cul-
tural difference. People from Eastern cultures tend

452 LEE

to be holistic, perceiving people, objects, and events
in terms of their inseparable relations, whereas
those from Western cultures tend to be analytic,
perceiving people, objects, and events in terms of
their isolated properties (Ji, Peng, & Nisbett, 2000;
Masuda & Nisbett, 2001; Masuda et al., 2008;
Nisbett et al., 2001). Although holistic and analytic
styles of thinking are predominant in Eastern and
Western cultures, respectively, it has been shown
that thinking styles vary by both individual cogni-
tive differences (Choi, Koo, & Choi, 2007) and situa-
tional factors (Kuhnen€ & Oyserman, 2002;
Miyamoto, Nisbett, & Masuda, 2006; Zhou, He,
Yang, Lao, & Baumeister, 2012) within a culture.
For example, Miyamoto et al. (2006) found that
both Japanese and American participants who were
exposed to the Japanese perceptual environment
attended more to the holistic and contextual infor-
mation than did those who were exposed to the
American perceptual environment. These findings
provide evidence that living in a different environ-
ment within a culture may determine whether peo-
ple favor a holistic or analytic style of thinking.

Importantly, such within-culture differences in
thinking styles have been documented to explain
underlying processes of various responses. For exam-
ple, Baaren, Horgan, Chartrand, and Dijkmans
(2004) found that holistic thinkers engaged in behav-
ioral mimicry more than analytic thinkers because of
holistic thinkers’ reliance on contextual cues. In mar-
keting contexts, Zhu and Meyers-Levy (2009) found
that when a product was viewed on a display table,
holistic thinkers perceived the product and the table
as continuous parts of a larger unit, and as a result,
the concepts elicited by the table’s surface materials
were assimilated with perceptions of the product.
Similarly, Monga and John (2010) found that for
functional brands, holistic thinkers perceived brand
extensions more favorably than analytic thinkers
because of holistic thinkers’ ability to connect a par-
ent brand and its extension. Additionally, Lalwani
and Shavitt (2013) showed that holistic thinkers
viewed price as a signal for quality more than ana-
lytic thinkers because of holistic thinkers’ attention to
interrelations between the elements of a product.

Social Class, Thinking Style, and Service Failure

In the present research, we seek to demonstrate that
consumer judgments differ across social classes and
that these differences are driven by information
processing styles, in particular, holistic and analytic
styles of thinking. As a useful way to understand

class differences in thinking styles in consumer con-
texts, we employ service failure domains that vary
in dimensions of failure. We argue that low-class
individuals, who tend to focus their attention on
contextual factors, may engage in holistic thinking.
If this is correct, it is expected that low-class con-
sumers will carry over their negative evaluations of
the service dimension related to a failure incident
(hereafter, failure-related dimension) to the other
dimension that is unrelated to the failure incident
(hereafter, failure-unrelated dimension). In compar-
ison, high-class individuals, who tend to focus their
attention on focal attributes detached from the
whole context, may engage in analytic thinking. If
so, the carryover effect that occurs in the failure-
unrelated dimension will emerge to a lesser degree
for high-class consumers. Put another way, when
faced with a service failure, low-class individuals
will decrease their evaluations of the failure-unre-
lated dimension greater than their high-class coun-
terparts. When it comes to evaluations of the
failure-related dimension, however, there will be lit-
tle to no difference across social class contexts.

For all these predictions, we identify a key moder-
ating factor. It has been shown that the effects of ser-
vice failure are significantly affected by failure
magnitude. For example, when the magnitude of ser-
vice failure is perceived to be severe versus minor
(e.g., a reserved hotel room is unavailable vs. check-
ing into a hotel room is slightly delayed), consumers
provide more negative evaluations of service encoun-
ters (McCollough, Berry, & Yadav, 2000; Smith,
Bolton, & Wagner, 1999). That is, consumers have
cognitive appraisals of failure severity (Wang, Wu,
Lin, & Wang, 2011) and perceive severe (vs. minor)
failure as being a greater loss of their resources (Smith
et al., 1999). In particular, low-class consumers, who
have cognitive biases and appraise the same situation
as potentially more harmful than their high-class
counterparts (Chen & Matthews, 2001), are likely to
amplify perceptions of loss when the failure becomes
severe versus minor. Accordingly, it is necessary to
consider perceived severity of service failure in our
predictions on cognitive differences across classes,
and we expect that severe service failure will enlarge
the carryover effect among low-class consumers,
relative to their high-class counterparts.

Overview of Studies

In four studies, we test how and why consumer
evaluations of the failure-unrelated dimension sys-
tematically diverge across social class contexts.

�

�

Social Class Differences in Carryover Effects 45

3

Studies 1 and 2 provide evidence that low-class
consumers evaluate the failure-unrelated dimension
lower than high-class consumers when failure
severity is perceived as severe rather than minor.
Studies 3 and 4 examine the underlying mechanism
of the effect obtained in Studies 1 and 2. Specifi-
cally, Study 3 measures analytic-holistic thinking
and shows that low (vs. high) class consumers
favor holistic thinking, which results in a decrease
in evaluations of the failure-unrelated dimension.
Study 4 primes thinking style and shows that prim-
ing low-class consumers with analytic thinking (vs.
baseline) increases evaluations of the failure-unre-
lated dimension, whereas priming those of high
social class with holistic thinking (vs. baseline)
decreases evaluations of the dimension.

In all studies, we measure social class that varies
along a continuum. That is, low class is operational-
ized as a group of people placed at the relatively
low end of the continuum, whereas high-class is
operationalized as a group of people placed at the
relatively high end of the continuum. To demon-
strate class differences in the carryover tendency,
we use two service dimensions: the process dimen-
sion involving how consumers are treated during
service encounters and the outcome dimension
involving what consumers receive from service
encounters.

Study 1

In Study 1, we attempt to show that, under cer-
tain circumstances, exposure to service failure in
one dimension influences evaluations of the other
service dimension that is seemingly unrelated to the
failure. Using a process-failure scenario describing
an inattentive waiter at a restaurant, we examine
how evaluations of the other dimension—the out-
come dimension (e.g., the availability of food)—dif-
fer as a joint function of social class and failure
severity. We predict that low-class individuals will
judge the outcome dimension (failure-unrelated
dimension) less positively than their high-class
counterparts when failure severity is high versus
low. We also predict that there will be no class dif-
ference in judgments of the process dimension (fail-
ure-related dimension).

Method

A total of 216 participants (Mage = 36.21, SD =
12.06) recruited from an online panel via Amazon
Mechanical Turk completed the study in return for
a small payment. The ethnic composition of the

sample was 13.4% African/African American, 6.9%
Asian/Asian American, 69.9% White/Caucasian,
7.4% Hispanic/Hispanic American, and 2.3% other
ethnicities. Upon agreeing to participate, partici-
pants were told that they would participate in short
separate studies. Participants were first asked to
complete a social class scale designed to measure
subjective perceptions of social standing relative to
others (Adler et al., 2000). In this measure, partici-
pants were presented with a drawing of a 10-rung
ladder representing where people stand relative to
others and were asked to indicate their relative
social standing on a 10-point scale (e.g., 1 = bottom
rung, 10 = top rung) with higher numbers associ-
ated with higher social classes (see Appendix S1 for
measures and manipulations of all studies).

Following this task, participants were asked to
read a process-failure scenario adapted from Smith
et al. (1999). Participants were randomly assigned
to either a minor or severe service failure condition.
Participants in the minor failure condition (here-
after, low severity condition) read: “The waiter
brings your beverages and entrees and leaves with-
out asking if you need anything else. He does not
refill your beverages while you are eating.” In com-
parison, participants in the severe failure condition
(hereafter, high severity condition) read: “The
waiter brings your entrees and leaves without ask-
ing if you need anything else. The waiter never
brings your beverages, and he doesn’t stop back to
check on you while you’re eating. He drops off the
bill without asking if you want anything more.” All
participants were then asked to indicate failure
severity on a 7-point scale (1 = not severe at all,
7 = very much severe), which served as a manipula-
tion check. As expected, participants in the high
severity condition (M = 5.13, SD = 1.47) reported
the failure incident to be more severe than those in
the low severity condition (M = 4.26, SD = 1.42; F
(1, 214) = 19.59, p < .001).

For dependent measures, we asked participants
to indicate how much they would agree or disagree
with statements about the waiter’s attitude (failure-
related dimension) and the availability of food (fail-
ure-unrelated dimension), adapted from Chan,
Wan, and Sin (2009), all on 3-item 7-point scales
(1 = strongly disagree, 7 = strongly agree). A sample
item for the process dimension includes “It appears
that the waiter’s attitude is acceptable,” and a sam-
ple item for the outcome dimension includes “It
appears that the availability of food is good
enough.” Responses to the items were averaged to
form composite scores of the process (failure-
related) dimension (a = .71) and the outcome

454 LEE

(failure-unrelated) dimension (a = .87) with higher
numbers reflecting better evaluations.

Finally, we assessed participants’ objective social
class using both educational attainment (1 = did not
finish high school, 2 = high school graduate or some col-
lege, 3 = college graduate, 4 = postgraduate) and annual
household income (1 = under $15,000, 2 = $15,001–
$25,000, 3 = $25,001–$35,000, 4 = $35,001–$50,000,
5 = $50,001–$75,000, 6 = $75,001–$100,000, 7 = Over
$100,000). Participants had a median educational
attainment of college graduate and a median house-
hold income between $35,001 and $50,000.

Results

Evaluations of the failure-unrelated dimension.
We first regressed participants’ evaluations of the
outcome (failure-unrelated) dimension onto subjec-
tive social class (M = 4.69, SD = 1.60), failure sever-
ity (0 = low severity, 1 = high severity), and their
interaction. The results revealed no main effect of
subjective social class (B = �0.17, SE = .17; t(212) =
�1.05, p = .30), but there was a main effect of fail-
ure severity (B = �1.68, SE = .53; t(212) = �3.15,
p < .01). Importantly, there was a two-way interac- tion between subjective class and failure severity (B = 0.23, SE = .11; t(212) = 2.15, p < .05). To decompose this interaction, we conducted a spot- light analysis by creating conditional values for subjective social class that were one standard devia- tion above and below its mean. We used Hayes (2013) PROCESS macro for model 1 with 5,000 bootstrapped samples. As predicted, low-class indi- viduals reported lower evaluations than high-class individuals, and this effect emerged in the high severity condition (B = 0.29, SE = .08; t(212) = 3.66, p < .001), but not in the low severity condition (B = �0.06, SE = .07; t(212) = 0.77, p = .44). The pat- tern of these results is illustrated in the top panel of Figure 1. Our results remained significant when we controlled for objective class indicators—education and income. That is, the analysis calculated with the controls revealed that low-class individuals, rel- ative to their high-class counterparts, judged the outcome dimension less positively in the high severity condition (B = 0.27, SE = .09; t(210) = 3.12, p < .01), but there was no class difference in the low severity condition (B = 0.03, SE = .08; t(210) = 0.43, p = .67). Additionally, to identify the range of the subjective class scale in which the effect of failure severity was significant, we per- formed a floodlight analysis (Spiller, Fitzsimons, Lynch, & McClelland, 2013) using the Johnson– Neyman technique. This analysis revealed that

E
va

lu
at

io
n

s

o
f o

ut
co

m
e

di
m

en
si

on
Evaluations of failure-unrelated dimension

7

6

4.935 4.71
4.75

4

3.79

3

2

1
Low severity

High severity

Low class

High class

Evaluations of failure-related dimension
7

io
n

6

i

m
en

s

5

m
e

d

4ou
tc

o
3 2.67

al
ua

t

io
ns

o
f

2
2.37 2.11

1.99

E
v

1
Low severity High severity

Low class High class

Figure 1. Social class differences in evaluations in response to
process service failure as a function of failure severity (Study 1)

there was a significant negative influence of failure
severity on evaluations among those who scored
lower than 5.60 (BJN = �0.39, SE = .19, t(212) =
�1.97, p = .05) on the 10-point scale of subjective
class, confirming our predictions that those rela-
tively low in social class provided lower evalua-
tions in the high (vs. low) severity failure.

Next, we tested our prediction using objective
class indicators. Both education and income were
positively correlated with subjective social class
(r = .24 for subjective social class and education;
r = .54 for subjective social class and income;
r = .24 for education and income; ps < .001). How- ever, none of these objective measures accounted for class differences in evaluations of the outcome dimension (ps > .38), irrespective of whether we
used each separate measure or averaged them into
one overall measure with education and income
standardized. This finding is consistent with previ-
ous research findings suggesting that subjective
social class shapes people’s judgments (Greitemeyer
& Sagioglou, 2016; Smith & Pettigrew, 2014) and
cognitive patterns (Kraus et al., 2009) above and
beyond objective social class.

Evaluations of the failure-related dimension. Finally,
we regressed participants’ evaluations of the process

Social Class Differences in Carryover Effects 455

(failure-related) dimension onto subjective social
class, failure severity, and their interaction. The only
significant result to emerge was the effect of failure
severity (B = �1.09, SE = .44; t(212) = �2.46,
p = .01), suggesting that both low and high-class par-
ticipants in the high severity condition reported
lower evaluations than those in the low severity con-
dition (see the bottom panel of Figure 1).

Discussion

Study 1 shows the existence of class differences
in evaluations of the failure-unrelated dimension of
service encounters. Using the process-failure con-
text, we find that low-class participants formed less
positive evaluations of the outcome dimension than
high-class participants when the severity of failure
was considered high versus low. These results
remained significant when we controlled for objec-
tive class measures—education and income. More-
over, none of these objective class measures were
associated with class differences in evaluations of
the failure-unrelated dimension. While our findings
provide initial evidence that social class affects the
way people judge service failure, there is the possi-
bility that interpersonal characteristics inherent in
the process dimension produced the effects
observed in Study 1. Accordingly, we address this
issue in Study 2 using an outcome-failure context,
which is noninterpersonal.

Study 2

Method

A total of 207 participants (Mage = 35.70,
SD = 11.22) from an online panel via Amazon
Mechanical Turk completed the study in return for
a small payment. The ethnic composition of the
sample was 5.8% African/African American, 5.3%
Asian/Asian American, 77.3% White/Caucasian,
10.1% Hispanic/Hispanic American, and 1.4% other
ethnicities. The procedure was identical to that of
Study 1. Participants were first presented with a
drawing of a ladder and asked to indicate their
subjective social standing on a 10-point scale. Fol-
lowing this task, participants were asked to read an
outcome-failure scenario adapted from Smith et al.
(1999) describing unfulfilled orders at a restaurant
and were randomly assigned to either a low or
high severity condition. Participants then indicated
their perceived severity of the failure on a 7-point
scale (1 = not severe at all, 7 = very much severe). As
predicted, participants in the high severity

condition (M = 5.08, SD = 1.36) reported the failure
in the scenario to be more severe than did those in
the low severity condition (M = 3.55, SD = 1.77; F
(1, 205) = 48.56, p < .001).

For dependent measures, we asked participants
to indicate how much they agreed or disagreed
with statements about the availability of food (fail-
ure-related dimension) and the waiter’s attitude
(failure-unrelated dimension) using the same scales
reported in Study 1. Responses to the items were
averaged to form composite scores of the outcome
(failure-related) dimension (a = .85) and process
(failure-unrelated) dimension (a = .84). Finally, we
assessed education and income using the same
scales as in Study 1. Participants had a median
educational attainment of college graduate and a
median household income between $35,001 and
$50,000.

Results

Evaluations of the failure-unrelated dimension. We
first regressed participants’ evaluations of process
(failure-unrelated) dimension onto subjective social
class (M = 5.32, SD = 1.64), failure severity (0 = low
severity, 1 = high severity), and their interaction.
There was a marginally significant main effect of
subjective social class (B = �0.29, SE = .17;
t(203) = �1.74, p = .08) and a significant main
effect of failure severity (B = �1.96, SE = .59;
t(203) = �3.32, p < .01). Importantly, a two-way interaction between subjective social class and fail- ure severity was significant (B = 0.25, SE = .11; t(203) = 2.39, p < .05). To probe this interaction, we performed a spotlight analysis using Hayes (2013) PROCESS macro for model 1 with 5,000 boot- strapped samples. The analysis revealed that low- class individuals provided lower evaluations than high-class individuals, and this effect emerged in the high severity condition (B = 0.22, SE = .08; t(203) = 2.86, p < .01) but not in the low severity condition (B = �0.04, SE = .07, t(203) = �0.48, p = .64). The pattern of these results appears in the top panel of Figure 2. Our results remained sig- nificant when we controlled for education and income.

We conducted additional regression analyses
using objective class indicators. Both education and
income were positively correlated with subjective
social class (r = .38 for subjective social class and
education; r = .52 for subjective social class and
income; r = .35 for education and income; ps < .001). However, none of these objective class indicators accounted for class differences in evaluations of the

�

456 LEE

E

va
lu

at
io

ns
o

f o
ut

co
m

e
di

m
en

si
on

E

va
lu

at
io

ns
o

f o
ut

co
m

e
di

m
en

si
on

7
Evaluations of failure-unrelated dimension

6

5
5.20
5.08

4.89

4 4.17

3

2

1
Low severity

Low class

High severity

High class

Evaluations of failure-related dimension
7

6

5

4

2.943
2.72

1.912
1.63

1
Low severity High severity

Low class High class

Figure 2. Social class differences in evaluations in response to
outcome service failure as a function of failure severity (Study 2)

process dimension (ps > .47), irrespective of whether
we used each separate measure or averaged stan-
dardized scores into one overall measure.

Evaluations of the failure-related dimension. Finally,
we regressed participants’ evaluations of the out-
come (failure-related) dimension onto subjective
social class, failure severity, and their interaction.
Consistent with the finding of Study 1, the only sig-
nificant result to emerge was the effect of failure
severity (B = �1.16, SE = .55; t(203) = �2.12, p < .05; see the bottom panel of Figure 2).

Discussion

Using the outcome-failure scenario, Study 2 fully
replicates Study 1, displaying that the observed
effects are extended to noninterpersonal failures.
We provide robust support for class differences in
the carryover effect. Specifically, regardless of
whether the failure is process-related (Study 1) or
outcome-related (Study 2), low-class participants
rated the failure-unrelated dimension significantly
lower than did their high-class counterparts. These
effects emerged when the level of failure severity
was viewed as high rather than low. However,

social class does not pertain to evaluations of the
failure-related dimension.

As noted earlier, we argue that these differential
effects on evaluations of the failure-unrelated
dimension are attributed to class differences in
thinking style. In the next set of studies, we exam-
ine the role of thinking style as a potential mecha-
nism responsible for the carryover effect. In doing
so, we measure (Study 3) and manipulate (Study 4)
thinking style. We predict that low-class individu-
als, relative to their high-class counterparts, will
have a tendency to view situations in a more holis-
tic manner and thus evaluate the failure-unrelated
dimension less favorably in response to a specific
failure incident.

One may speculate that the relative scarcity of
resources among low-class individuals produces
cognitive orientations focused on resource con-
straints (e.g., the unavailability of food), which in
turn decrease evaluations of the resource unavail-
ability. To address this issue, we employ different
dependent measures independent of resource con-
straints: the cleanliness of the hotel rooms (Study 3)
and the quality of the restaurant food (Study 4).

Study 3

In this study, we attempt to generalize our previ-
ous findings and provide evidence for the mediat-
ing role of thinking style in the relationship
between social class and evaluations of the failure-
unrelated dimension. Noting that the effects
observed in the first two studies are most pro-
nounced when failure severity is high, we test our
predictions using the high severity context only. In
addition to testing our predictions, we examine
some alternative explanations in the study. First,
Kraus et al. (2009) find that low-class people experi-
ence a reduced sense of power and thus focus their
attention on contextual information. If so, it is plau-
sible that this reduced sense of power may be asso-
ciated with the class-specific effects observed in our
previous studies. Second, Ruggiero and Marx
(1999) suggest that low-class people, characterized
by fewer resources to reduce psychological stress,
tend to perceive themselves as more discriminated.
Thus, low-class people may decrease their evalua-
tions of the failure-unrelated dimension because of
their feelings of discrimination. Last, we attempt to
rule out the possibility that class differences in eval-
uations of the dimension may be associated with
differences in feelings of compassion (Piff, Kraus,
Cote,^ Cheng, & Keltner, 2010; Stellar, Manzo,
Kraus, & Keltner, 2012).

Social Class Differences in Carryover Effects 457

Method

In a correlational study, 278 participants
(Mage = 37.22, SD = 12.53) from an online panel via
Amazon Mechanical Turk completed the study in
return for a small payment. The ethnic composition
of the sample was 6.5% African/African American,
8.3% Asian/Asian American, 77.0% White/Cau-
casian, 7.2% Hispanic/Hispanic American, and 1%
other ethnicities. Participants were first asked to
read a high-severity process-failure scenario
adapted from Smith et al. (1999) describing an inat-
tentive representative at a hotel. As expected, par-
ticipants perceived the failure in the scenario to be
severe, as compared to a midpoint of 4 (M = 5.77,
SD = 1.06; t(277) = 27.92, p < .001).

After reading the scenario, participants indicated
their agreement with statements about the represen-
tative’s attitude (failure-related dimension) and the
cleanliness of the hotel rooms (failure-unrelated
dimension), all on 3-item 7-point scales (1 = strongly
disagree, 7 = strongly agree). A sample item for the
process dimension (failure-related dimension)
includes “It appears that the representative’s atti-
tude is acceptable,” and a sample item for the out-
come dimension (failure-unrelated dimension)
includes “It appears that the cleanliness of the hotel
rooms is good enough.” The three items for each
scale were averaged to form composite scores of
process (failure-related) dimension (a = .76) and
outcome (failure-unrelated) dimension (a = .76)
with higher numbers reflecting better evaluations.

Next, participants completed a series of mea-
sures, all on 7-point scales (1 = strongly disagree,
7 = strongly agree). To assess participants’ feelings
of discrimination, we instructed participants to
think about how they would feel if they were in
the situation described in the scenario and to indi-
cate their agreement with five items (a = .82). A
sample item is “If I were in the situation, I would
feel discriminated.” Thinking style was assessed
with the 24-item scale of analytic versus holistic
thinking tendency (a = .78) adapted from Choi
et al. (2007). A sample item includes “Everything in
the universe is somehow related to each other.” We
measured participants’ compassion using five items
(a = .92) adapted from Shiota, Keltner, and John
(2006). A sample item includes “I am a very com-
passionate person.” To assess participants’ sense of
power, we used the 8-item scale of power (a = .88)
adapted from Anderson and Galinsky (2006). A
sample item includes “I think I have a great deal of
power.” We then measured participants’ subjective
social class using six items (a = .81) adapted from

Griskevicius, Tybur, Delton, and Robertson (2011).
A sample item is “I feel relatively wealthy these
days.” Finally, we measured participants’ objective
social class using education and income as in Stud-
ies 1 and 2. Participants had a median educational
attainment of college graduation and a median
household income between $35,001 and $50,000.

Results

Evaluations of the failure-unrelated dimension. We
regressed participants’ evaluations of the outcome
(failure-unrelated) dimension onto subjective social
class (M = 3.41, SD = 1.31). As predicted, low-class
participants evaluated the cleanliness of the hotel
rooms (failure-unrelated dimension) significantly
lower than high-class participants (B = 0.20,
SE = .06; t(276) = 3.23, p < .01). This finding remained significant when we controlled for educa- tion and income (B = 0.20, SE = .07; t(274) = 2.95, p < .01). We reran the regression analysis using objective class indicators. While both education and income were positively correlated with subjective social class (r = .24 for subjective social class and education; r = .42 for subjective social class and income; r = .31 for education and income; ps < .001), none of these objective class indicators were associated with class differences in evaluations of the failure-unrelated dimension (ps > .19), irrespec-
tive of whether we used each separate measure or
averaged standardized scores into one overall mea-
sure.

Evaluations of the failure-related dimension. Next,
we regressed participants’ evaluations of the pro-
cess (failure-related) dimension onto subjective
social class. As expected, a regression analysis did
not produce any significant effect (ps > .40), irre-
spective of whether education and income were
entered as control variables or not, replicating our
previous findings that there is no social class differ-
ence in evaluations of the failure-related dimension.

Mediation analysis. We conducted a multiple
mediator analysis using Hayes (2013) model 4 with
5,000 bootstrapped samples. In the multiple media-
tor model, subjective social class was treated as the
predictor variable, evaluations of the failure-unre-
lated dimension as the criterion variable, and feel-
ings of discrimination, analytic-holistic thinking,
compassion, and power as the potential mediators.
Table 1 contains the parameter estimates and confi-
dence intervals for the total and indirect effects of
the four mediator variables on the relationship
between subjective social class and evaluations of
the failure-unrelated dimension. The zero included

458 LEE

Table 1
Indirect Effects of Subjective Social Class on Evaluations of the Cleanliness of the Hotel Rooms Through Feelings of Discrimination, Thinking Style,
Compassion, and Power (study 3)

With no control variables With the control variables (education and income)

95% CI 95% CI

Mediator Parameter estimate SE Lower Upper Parameter estimate SE Lower Upper

Total 0.0462 .0279 �0.0030 0.1079 0.0339 .0255 �0.0094 0.0904
Discrimination �0.0060 .0085 �0.0307 0.0053 �0.0093 .0105 �0.0409 0.0037
Thinking style 0.0204 .0133 0.0016 0.0547 0.0224 .0150 0.0014 0.0615
Compassion 0.0000 .0041 �0.0079 0.0101 0.0000 .0044 �0.0092 0.0094
Power 0.0318 .0225 �0.0057 0.0839 0.0209 .0165 �0.0017 0.0637

in the confidence interval for the total indirect effect
(95% CI = �0.0030 to 0.1079) indicates that the
mediating effect of the combination of the four
mediator variables was not significant. However,
specific indirect effects were examined because sup-
pression effects could obscure the impact of indi-
vidual mediators (MacKinnon, Krull, & Lockwood,
2000; Preacher & Hayes, 2008).

Among the four potential mediators, the only
significant mediator was analytic-holistic thinking
(95% CI = 0.0016 to 0.0547) as demonstrated by
confidence intervals that excluded zero. Specifically,
subjective social class was negatively related to ana-
lytic-holistic thinking (B = �0.06, SE = .03;
t(276) = �2.23, p < .05), which in turn was nega- tively related to evaluations of the failure-unrelated dimension (B = �0.35, SE = .15; t(272) = �2.30, p < .05). When we controlled for objective class indicators (education and income), analytic-holistic thinking remained the only significant mediator (95% CI = 0.0014 to 0.0615). These results suggest that class differences in thinking style account for the carryover effect. However, the alternative

accounts related to discrimination, compassion, and
power were independent of the effect. A summary
of descriptive statistics and correlation matrix for
these variables appears in Table 2.

Discussion

Study 3 replicates the results from Studies 1 and
2 with a different failure scenario and a different
dependent measure, displaying the generalizability
of our findings in other contexts. Importantly, the
multiple mediator model reveals that thinking style
significantly mediated the effect of social class on
evaluations of the failure-unrelated dimension. The
analysis calculated with education and income as
control variables also provides consistent evidence
that the positive association between social class
and evaluations is driven by thinking style. That is,
as compared to high-class participants, low-class
participants favor a holistic style of thinking, which
results in lower evaluations of the failure-unrelated
dimension. However, because of the correlational
nature of these findings, no causal interpretations

Table 2
Correlations Among Subjective SES, Evaluations of the Cleanliness of the Hotel rooms, Discrimination, Analytic-holistic Style of Thinking, Compas-
sion, Power, Education, and Income (Study 3)

M SD 1 2 3 4 5 6 7 8

1. Subjective SES 3.41 1.31 —
2. Evaluations 3.99 1.38 0.19** —
3. Discrimination 4.23 1.46 0.05 �0.12* —
4. Thinking style 4.84 .57 �0.13* �0.17** 0.12* —
5. Compassion 5.33 1.21 �0.01 �0.07 0.21** 0.37** —
6. Power 4.57 1.12 0.29** �0.13* 0.05 0.09 0.17** —
7. Education 2.69 .67 0.24** 0.07 �0.05 �0.09 �0.09 0.16** —
8. Income 4.11 1.80 0.42** 0.07 �0.02 �0.01 0.01 0.35** 0.31** —

*p < .05. **p < .01.

Social Class Differences in Carryover Effects 459

are appropriate. To address this issue,

Study 4

employs an experimental design.

Study 4

In Study 4, we extend our underlying mecha-
nism findings of Study 3 using an experimental
causal chain design (Spencer, Zanna, & Fong, 2005),
in which we manipulate styles of thinking. We have
argued that thinking style is responsible for class
differences in the carryover effect. That is, low-class
individuals favor holistic thinking and thus
decrease evaluations of the failure-unrelated dimen-
sion, as compared to high-class individuals, who
favor analytic thinking. If this reasoning is correct,
priming analytic and holistic thinking for low and
high-class individuals, respectively, should reverse
this pattern. That is, priming low-class individuals
with analytic thinking will increase evaluations of
the failure-unrelated dimension, whereas priming
high-class individuals with holistic thinking will
decrease evaluations of the dimension.

Method

A total of 187 undergraduate students (Mage =
21.76, SD = 3.43) at a large public university partici-
pated in the study for extra course credit. The eth-
nic composition of the sample was 12.3% African/
African American, 19.3% Asian/Asian American,
49.2% White/Caucasian, 9.1% Hispanic/Hispanic
American, and 10.1% other ethnicities. Participants
were told that they would participate in a few sepa-
rate studies and completed all tasks on a computer
in a laboratory. Upon arrival in the lab, participants
were randomly assigned to experimental conditions
in a three-group (thinking style: baseline vs. ana-
lytic vs. holistic) between-subjects design. Partici-
pants assigned to the analytic and holistic thinking
conditions were asked to read a fictitious report
and write a statement that supported the report.
Participants assigned to the baseline condition were
not asked to complete any priming task.

Next, we presented participants with the high-
severity process-failure scenario describing an inat-
tentive waiter at a restaurant as used in Study 1. As
expected, participants perceived the service failure in
the scenario to be severe, as compared to a midpoint
of 4 (M = 5.48, SD = 1.53; t(186) = 13.28, p < .001). To measure evaluations of the failure-unrelated dimension, we asked them to indicate their agree- ment with the quality of food (failure-unrelated dimension) using two items (r = .77) on a 7-point scale (1 = strongly disagree, 7 = strongly agree). The

items are “Food is likely to be delicious” and “Food
is likely to be good.” Note that we used different
items to measure participants’ evaluations of the fail-
ure-unrelated dimension, as compared to those used
in Study 1. Finally, we assessed both subjective class
using the same scale (six items; a = .84) as in Study 3
and objective class using annual household income.
Participants reported a median household income
between $35,001 and $50,000.

Results

Manipulation check. To ascertain the effective-
ness of the manipulation of thinking styles, we per-
formed a pretest using a separate sample of
participants (N = 148) from Amazon Mechanical
Turk. We randomly assigned participants to one of
the three conditions (baseline vs. analytic thinking
vs. holistic thinking) and asked them to indicate
their agreement with each statement at that present
moment using the same 24-item 7-point scale
(a = .79) as in Study 3. A one-way ANOVA
revealed a significant difference among the condi-
tions (F(2, 145) = 9.18, p < .001). As expected, par- ticipants in the analytic thinking condition (M = 4.59, SD = .53) reported a significantly lower score on the scale than those in the holistic thinking (M = 5.06, SD = .57; F(1, 145) = 18.34, p < .001) and baseline (M = 4.84, SD = .52; F(1, 145) = 5.26, p = .023) conditions. Also, there was a significant difference between participants in the holistic think- ing (M = 5.06, SD = .57) and baseline (M = 4.84, SD = .52; F(1, 145) = 4.04, p = .046) conditions.

Evaluations of the failure-unrelated dimension. We
first performed a one-way ANOVA with thinking
style as the experimental factor. There was a signifi-
cant effect of thinking style on participants’ ratings
of food quality (F(2, 183) = 8.59, p < .001). Planned contrasts revealed that participants in the analytic thinking condition (M = 4.12, SD = 1.33) scored higher on ratings of food quality than those in the holistic thinking (M = 3.21, SD = 1.13; F(1, 184) = 17.32, p < .001) and baseline (M = 3.61, SD = 1.19; F(1, 184) = 17.32, p < .02) conditions. The difference between the latter two conditions was marginally significant (F(1, 184) = 3.35, p = .069).

Next, we examined a two-way interaction
between thinking style and subjective class by per-
forming a multiple regression analysis. Our regres-
sion analysis involved a multicategorical variable
with three levels (thinking style: 0 = baseline vs.
1 = analytic thinking vs. 2 = holistic thinking) and
a continuous variable (subjective class). Following
the procedure of Hayes and Montoya (2017), we

460 LEE

transformed thinking style into two dummy vari-
ables—D1 and D2. Thus, the regression model
included subjective class, D1, D2, subjective
class 9 D1, and subjective class 9 D2 as the inde-
pendent variables, and evaluations of food quality
as the dependent variable. To test a two-way inter-
action between subjective class and thinking style,
we conducted a test of significance for the change
in R2 in the regression model using PROCESS
model 1 with 5,000 bootstrapped samples (Hayes
& Montoya, 2017). Our findings revealed that the
two-way interaction was significant (ΔR2 = .036, F
(2, 181) = 3.75, p < .05), suggesting that the effect of subjective class on evaluations of food quality is contingent on thinking style. We further probed the interaction in two separate ways. First, we implemented indicator coding to determine differ- ences in evaluations of food quality between the baseline condition and each of the thinking style conditions for the levels of subjective class that were one standard deviation above and below its mean. As expected, low-class participants reported higher evaluations under the analytic condition, relative to the baseline condition (Manalytic = 4.20 vs. Mbaseline = 3.21; t(181) = 3.21, p < .01), whereas high-class participants reported lower evaluations under the holistic condition, relative to the base- line condition (Mholistic = 3.10 vs. Mbaseline = 4.01; t(181) = �3.15, p < .01). Second, we implemented Helmert coding to identify differences in evalua- tions of food quality between the analytic condi- tion and the holistic condition for the levels of subjective class that were one standard deviation above and below its mean. As expected, low-class participants reported higher evaluations under the analytic condition, relative to the holistic condition (Mholistic = 3.36 vs. Manalytic = 4.20; t(181) = �2.75, p < .01), whereas high-class participants reported lower evaluations under the holistic condition, rel- ative to the analytic condition (Mholistic = 3.10 vs. Manalytic = 4.02; t(181) = �2.95, p < .01). An addi- tional analysis indicated that there was no class difference in either the holistic or analytic thinking condition (ps > .35). Figure 3 displays the pattern
of these results. Our findings remained the same
when we controlled for objective class (income).

Finally, using objective class (income), we reran
the regression analysis involving a multicategorical
variable (thinking style) and a continuous variable
(objective class). While objective class was posi-
tively correlated with subjective class (r = .37;
p < .001), there was no significant interaction between objective class and thinking style (ΔR2 = .015, F(2, 181) = 1.55, p = .22).

E
va

lu
at

io
ns

o
f o

ut
co

m
e

di
m

en
si

on
Evaluations of failure-unrelated dimension

7

6

5
4.20

4.014
4.02 3.36

3 3.103.21

2

1
No prime Analytic prime Holistic prime

Low class High class

Figure 3. Social class differences in evaluations as a function of
thinking style (Study 4)

Discussion

We replicate our previous findings and further
highlight the crucial role of thinking style in class
differences in evaluations of the failure-unrelated
dimension. In doing so, we manipulated thinking
style to strengthen the causal link and generalize our
findings. When making judgments about service
failure, low (high) class participants decreased (in-
creased) evaluations of the failure-unrelated dimen-
sion. Using the experimental causal chain design, we
provide evidence that these class differences were
attributed to differences in thinking style. That is,
priming low-class participants with analytic thinking
increased evaluations of the failure-unrelated dimen-
sion, while priming high-class participants with
holistic thinking decreased evaluations of the dimen-
sion. Furthermore, evaluations of the failure-unre-
lated dimension decreased for participants primed
with holistic thinking and increased for those primed
with analytic thinking, thus indicating that thinking
style is the underlying factor that explains class dif-
ferences in the carryover effect.

General Discussion

We provide converging evidence for our theorizing
that social class affects consumer judgments and
that the underlying factor driving the effect is
thinking style. The pattern of our effects remained
the same regardless of whether service failure was
process-related (interpersonal) or outcome-related
(noninterpersonal) in the domains of both restau-
rant and hotel. Utilizing the two primary dimen-
sions of service encounters–process and outcome–as
a means to explore our predictions, we demon-
strated that low-class individuals responded less

Social Class Differences in Carryover Effects 461

favorably to service failure than their high-class
counterparts, especially when evaluating the fail-
ure-unrelated dimension. However, low-class indi-
viduals did not differ from their high-class
counterparts in evaluations of the failure-related
dimension. Notably, we identified failure severity
as an important moderating factor for the carryover
effect obtained in our studies. That is, the effect
occurs when service failure is considered severe,
but not when it is minor. Supporting this evidence,
previous research shows that when both low and
high-class individuals are exposed to the same
stressful events, low-class individuals have cogni-
tive biases and appraise situations as potentially
more harmful than their high-class counterparts
(Chen & Matthews, 2001), feel more stressed
(Kessler, 1979; Kessler & Cleary, 1980), and increase
more aggression (Greitemeyer & Sagioglou, 2016).
Consequently, such cognitive biases among low-
class individuals may be accentuated most strongly
when the severity is high rather than low.

As one explanation for class differences in evalu-
ations, we demonstrated that individuals in the low
social class were more holistic in thinking style than
those in the high social class, and these differences
in thinking style contributed to the observed effects
on evaluations of the failure-unrelated dimension.
Using a multiple mediator model, we ruled out
alternative accounts related to discrimination, com-
passion, and power. Furthermore, we provide addi-
tional evidence showing an opposite pattern in the
effects of social class on evaluations, such that acti-
vating analytic thinking among low-class individu-
als increased evaluations of the failure-unrelated
dimension, while activating holistic thinking among
high-class individuals decreased evaluations of the
dimension. That is, activating thinking styles elimi-
nated differences in evaluations between social
classes, displaying that thinking style was a driving
force underlying the observed effects.

Our research makes several contributions to the
literature on social class. Although it is widely rec-
ognized that social class, as a useful factor for mar-
ket segmentation, shapes consumer behavior on
diverse consumption contexts (for reviews, see
Henry, 2005; Kamakura & Mazzon, 2013), little is
known about the cognitive processes that might
account for social class effects (for reviews, see
Carey & Markus, 2016; Shavitt, Jiang, & Cho, 2016).
Our research fills this gap. We document the crucial
role of holistic-analytic styles of thinking in con-
sumer judgments across social class contexts. We
indicate that holistic tendencies among low-class
individuals are related to the carryover of their

judgments, while analytic tendencies among high-
class individuals are not linked to it. It is notewor-
thy that these differences between social classes
existed even after we held constant objective class
indicators such as education and income. Moreover,
we reveal that none of these objective class indica-
tors accounted for class differences in the carryover
effect. These findings are consistent with previous
research showing that a subjective sense of social
standing relative to others is a stronger predictor of
social cognitive tendencies than the objective indica-
tors of social class (Greitemeyer & Sagioglou, 2016;
Kraus et al., 2009; Smith & Pettigrew, 2014). Addi-
tionally, a recent Gallup poll reveals that the per-
centage of Americans who identify themselves as
working or lower class has increased to 48% in
2015, as compared to 33% in 2000 (Newport, 2015).
Considering these figures, it becomes evident that
marketers should make more efforts to understand
consumers in the lower half of social class, who
may have previously been neglected in favor of
those higher in social class (Pham, 2016).

Limitations and Future Directions

While the present research focuses on the carry-
over effect showing that low-class consumers tend
to incorporate nondiagnostic information into their
evaluations of a focal attribute, it does not reflect
overall service (brand) evaluations. As in recent
review articles (e.g., Shavitt et al., 2016), it is plausi-
ble that low (vs. high) class consumers, who make
external (vs. internal) attributions for behavior, may
provide their overall evaluations of service less neg-
atively in response to negative publicity of service
(e.g., Monga & John, 2008). For example, viewing
negative reviews about a restaurant may have less
impact on overall evaluations of the restaurant
among low (vs. high) class consumers because of
their tendency to seek out external explanations for
the negative reviews. How class differences in the
carryover tendency influence overall evaluations of
a brand is an empirical question worthy of further
research.

Related to this, some studies find that low (vs.
high) class individuals show higher levels of com-
passion in response to the needs of others (Stellar
et al., 2012) and thus engage in more prosocial
behavior (Piff et al., 2010). Aligning with this per-
spective, one may speculate that low-class con-
sumers are likely to evaluate a service failure less
negatively than their high-class counterparts. How-
ever, recent studies provide contradictory evidence;
high (vs. low) class individuals engage in more

462 LEE

prosocial behavior (Kornd€orfer, Egloff, & Schmukle,
2015) and low (vs. high) class individuals tend to
be more aggressive (Greitemeyer & Sagioglou,
2016). Our findings show that there was no class
difference in evaluations of the failure-related
dimension and levels of compassion. Thus, it is
worthwhile to identify potential moderating factors
for the relationship between social class and com-
passionate behavior.

In addition, our research is limited to negative
service contexts. However, would such a carryover
effect among holistic thinkers exist in positive ser-
vice contexts? Some research offers interesting
insights into this question. For example, in the con-
text of product perceptions, Henderson and Arora
(2010) find that when a product brand is associated
with a social cause, the positive associations with
the brand carry over to other product categories that
share the brand’s name but do not pertain to the
social cause. Similarly, Chernev and Blair (2015) find
that company’s prosocial activities including charita-
ble donations positively alter not just company rep-
utations but also consumer attitudes toward
product performance, even when the activities have
no direct association with company products. More
relevant, within the context of service failure, Bolton
and Mattila (2015) show that company’s prosocial
activities attenuate the negative impact of service
failure on satisfaction and loyalty, irrespective of
whether the failure is process-related or outcome-
related, suggesting that overall positive evaluations
of a company may have a carryover effect on other
specific service dimensions. Thus, it is likely that the
carryover effect observed in our studies may be
extended to positive service contexts.

Our findings also provide avenues for future
research. First, previous research suggests that peo-
ple in the high social class display a disjoint model
of agency in which their good actions are character-
ized as promoting independence, controlling envi-
ronments, and influencing others, while those in
the low social class display a conjoint model of
agency in which their good actions are defined as
promoting interdependence, adjusting the self to
environments, and attending to others (Markus &
Kitayama, 2003; Snibbe & Markus, 2005; Stephens
et al., 2007). Recognizing these differences, Ste-
phens, Hamedani, Markus, Bergsieker, and Eloul
(2009) show that in response to stressful situations
(e.g., Hurricane Katrina), people who are predomi-
nantly grounded in the disjoint model of agency
tend to leave the situations (e.g., find a way to
evacuate), while those who are grounded in the
conjoint model tend to stay in the situations by

having faith and maintaining hope (e.g., adjust the
self to the situations). Similarly, Becker, Kraus, and
Rheinschmidt-Same (2017) show that when faced
with a social disadvantage, low-class individuals
tend to remain politically inactive, as compared to
high-class individuals. In a marketing context, these
findings indicate that in response to service failures,
high-class consumers may take action and leave for
other service providers (e.g., switching behavior),
especially when there is an opportunity to do so,
rather than to remain with their existing service
providers.

Second, previous research shows that a service
experience recalled about a focal company may
influence evaluations of its competing company
depending on modes of recalling the experience
(Bickart & Schwarz, 2001). For example, when the
experience was episodically recalled, the evalua-
tions for the competing company (e.g., Burger King)
were in the same direction as those for the focal
company (e.g., McDonald’s; termed assimilation
effects). However, when it was analytically recalled,
the evaluations for the competing company were in
the opposite direction to those for the focal com-
pany (termed contrast effects). Thus, it seems rea-
sonable to predict that the carryover effect observed
in our studies may differ in judgments of compet-
ing companies among consumers who favor ana-
lytic thinking versus holistic thinking.

Finally, future research may investigate cross-cul-
tural differences in carryover effects across social
classes. Miyamoto and Wilken (2010) find that
American leaders (e.g., those who influence others)
tend to provide analytic explanations, whereas
Japanese leaders tend to have holistic perceptions,
suggesting that high-class individuals may have dif-
ferent thinking styles depending on their culture.
Furthermore, Na, McDonough, Chan, and Park
(2016) show that people from interdependent cul-
tures (relative to independent cultures) are more
sensitive to social contexts in their choices regard-
less of social class and, consequently, the effects of
social class on choices are significantly attenuated
among those from interdependent cultures. Thus,
cultural orientations may moderate the carryover
effect observed in our studies, and the effect may
hold for independent cultures, but not for interde-
pendent cultures.

References

Adler, N. E., Epel, E. S., Castellazzo, G., & Ickovics, J. R.
(2000). Relationship of subjective and objective
social status with psychological and physiological

�

Social Class Differences in Carryover Effects 463

functioning: Preliminary data in healthy, White women.
Health Psychology, 19, 586–592. https://doi.org/10.
1037/0278-6133.19.6.586

Adler, N. E., & Snibbe, A. C. (2003). The role of psychoso-
cial processes in explaining the gradient between
socioeconomic status and health. Current Directions in
Psychological Science, 12, 119–123. https://doi.org/10.
1111/1467-8721.01245

Anderson, C., & Galinsky, A. D. (2006). Power, optimism,
and risk-taking. European Journal of Social Psychology, 36,
511–536. https://doi.org/10.1002/(ISSN)1099-0992

Baaren, R. B. V., Horgan, T. G., Chartrand, T. L., & Dijk-
mans, M. (2004). The forest, the trees, and the chame-
leon: Context dependence and mimicry. Journal of
Personality and Social Psychology, 86, 453–459. https://d
oi.org/10.1037/0022-3514.86.3.453

Becker, J. C., Kraus, M. W., & Rheinschmidt-Same, M.
(2017). Cultural expressions of social class and their
implications for group-related beliefs and behaviors.
Journal of Social Issues, 73, 158–174. https://doi.org/10.
1111/josi.12209

Bickart, B., & Schwarz, N. (2001). Service experiences and
satisfaction judgments: The use of affect and beliefs in
judgment formation. Journal of Consumer Psychology, 11,
29–41. https://doi.org/10.1207/S15327663JCP1101_3

Bolton, L. E., & Mattila, A. S. (2015). How does corporate
social responsibility affect consumer response to service
failure in buyer–seller relationships? Journal of Retailing,
91, 140–153. https://doi.org/10.1016/j.jretai.2014.10.001

Carey, R., & Markus, H. R. (2016). Understanding con-
sumer psychology in working class contexts. Journal of
Consumer Psychology, 26, 568–582. https://doi.org/10.
1016/j.jcps.2016.08.004

Chan, H., Wan, L. C., & Sin, L. Y. M. (2009). The contrast-
ing effects of culture on consumer tolerance: Interper-
sonal face and impersonal fate. Journal of Consumer
Research, 36, 292–304. https://doi.org/10.1086/597329

Chen, E., & Matthews, K. A. (2001). Cognitive appraisal
biases: An approach to understanding the relation
between socioeconomic status and cardiovascular reac-
tivity in children. Annals of Behavioral Medicine, 23, 101–
111. https://doi.org/10.1207/S15324796ABM2302_4

Chernev, A., & Blair, S. (2015). Doing well by doing
good: The benevolent halo of corporate social responsi-
bility. Journal of Consumer Research, 41, 1412–1425.
https://doi.org/10.1086/680089

Choi, I., Koo, M., & Choi, J. A. (2007). Individual differ-
ences in analytic versus holistic thinking. Personality
and Social Psychology Bulletin, 33, 691–705. https://doi.
org/10.1177/0146167206298568

Coleman, R. P. (1983). The continuing significance of
social class to marketing. Journal of Consumer Research,
10, 265–280. https://doi.org/10.1086/208966

Côte, S. (2011). How social class shapes thoughts and
actions in organizations. Research in Organizational
Behavior, 31, 43–71.

Gaston-Breton, C., & Raghubir, P. (2013). Opposing
effects of sociodemographic variables on price

knowledge. Marketing Letters, 24, 29–42. https://doi.
org/10.1007/s11002-012-9201-6

Greitemeyer, T., & Sagioglou, C. (2016). Subjective socioe-
conomic status causes aggression: A test of the theory
of social deprivation. Journal of Personality and Social
Psychology, 111, 178–194. https://doi.org/10.1037/
pspi0000058

Griskevicius, V., Tybur, J. M., Delton, A. W., & Robertson,
T. E. (2011). The influence of mortality and socioeco-
nomic status on risk and delayed rewards: A life history
theory approach. Journal of Personality and Social Psychol-
ogy, 100, 1015–1026. https://doi.org/10.1037/a0022403

Hayes, A. F. (2013). Introduction to mediation, moderation,
and conditional process analysis. New York: Guilford.

Hayes, A. F., & Montoya, A. K. (2017). A tutorial on test-
ing, visualizing, and probing an interaction involving a
multicategorical variable in linear regression analysis.
Communication Methods and Measures, 11, 1–30.
https://doi.org/10.1080/19312458.2016.1271116

Henderson, T., & Arora, N. (2010). Promoting brands
across categories with a social cause: Implementing
effective embedded premium programs. Journal of Mar-
keting, 74, 41–60. https://doi.org/10.1509/jmkg.74.6.41

Henry, P. C. (2005). Social class, market situation, and
consumers’ metaphors of (dis)empowerment. Journal of
Consumer Research, 31, 766–778. https://doi.org/10.
1086/426610

Holt, D. B. (1998). Does cultural capital structure Ameri-
can consumption? Journal of Consumer Research, 25, 1–
25. https://doi.org/10.1086/209523

Ji, L.-J., Peng, K., & Nisbett, R. E. (2000). Culture, control,
and perception of relationships in the environment.
Journal of Personality and Social Psychology, 78, 943–955.
https://doi.org/10.1037/0022-3514.78.5.943

Johnson, W., & Krueger, R. F. (2005). Higher perceived
life control decreases genetic variance in physical
health: Evidence from a national twin study. Journal of
Personality and Social Psychology, 88, 165–173. https://d
oi.org/10.1037/0022-3514.88.1.165

Johnson, W., & Krueger, R. F. (2006). How money buys
happiness: Genetic and environmental processes linking
finances and life satisfaction. Journal of Personality and
Social Psychology, 90, 680–691. https://doi.org/10.1037/
0022-3514.90.4.680

Kamakura, W. A., & Mazzon, J. A. (2013). Socioeconomic
status and consumption in an emerging economy. Inter-
national Journal of Research in Marketing, 30, 4–18.
https://doi.org/10.1016/j.ijresmar.2011.12.001

Kessler, R. C. (1979). Social class and psychological dis-
tress. Journal of Health and Social Behavior, 20, 259–272.
https://doi.org/10.2307/2136450

Kessler, R. C., & Cleary, P. D. (1980). Social class and
psychological distress. American Sociological Review, 45,
463–478. https://doi.org/10.2307/2095178

Kornd€orfer, M., Egloff, B., & Schmukle, S. C. (2015). A
large scale test of the effect of social class on prosocial
behavior. PLoS ONE, 10, e0133193. https://doi.org/10.
1371/journal.pone.0133193

https://doi.org/10.1037/0278-6133.19.6.586

https://doi.org/10.1037/0278-6133.19.6.586

https://doi.org/10.1111/1467-8721.01245

https://doi.org/10.1111/1467-8721.01245

https://doi.org/10.1002/(ISSN)1099-0992

https://doi.org/10.1037/0022-3514.86.3.453

https://doi.org/10.1037/0022-3514.86.3.453

https://doi.org/10.1111/josi.12209

https://doi.org/10.1111/josi.12209

https://doi.org/10.1207/S15327663JCP1101_3

https://doi.org/10.1016/j.jretai.2014.10.001

https://doi.org/10.1016/j.jcps.2016.08.004

https://doi.org/10.1016/j.jcps.2016.08.004

https://doi.org/10.1086/597329

https://doi.org/10.1207/S15324796ABM2302_4

https://doi.org/10.1086/680089

https://doi.org/10.1177/0146167206298568

https://doi.org/10.1177/0146167206298568

https://doi.org/10.1086/208966

https://doi.org/10.1007/s11002-012-9201-6

https://doi.org/10.1007/s11002-012-9201-6

https://doi.org/10.1037/pspi0000058

https://doi.org/10.1037/pspi0000058

https://doi.org/10.1037/a0022403

https://doi.org/10.1080/19312458.2016.1271116

https://doi.org/10.1509/jmkg.74.6.41

https://doi.org/10.1086/426610

https://doi.org/10.1086/426610

https://doi.org/10.1086/209523

https://doi.org/10.1037/0022-3514.78.5.943

https://doi.org/10.1037/0022-3514.88.1.165

https://doi.org/10.1037/0022-3514.88.1.165

https://doi.org/10.1037/0022-3514.90.4.680

https://doi.org/10.1037/0022-3514.90.4.680

https://doi.org/10.1016/j.ijresmar.2011.12.001

https://doi.org/10.2307/2136450

https://doi.org/10.2307/2095178

https://doi.org/10.1371/journal.pone.0133193

https://doi.org/10.1371/journal.pone.0133193

�

�

464 LEE

Kraus, M. W., Côte, S., & Keltner, D. (2010). Social class,
contextualism, and empathic accuracy. Psychological
Science, 21, 1716–1723. https://doi.org/10.1177/
0956797610387613

Kraus, M. W., Piff, P. K., & Keltner, D. (2009). Social
class, sense of control, and social explanation. Journal of
Personality and Social Psychology, 97, 992–1004.
https://doi.org/10.1037/a0016357

Kraus, M. W., Piff, P. K., Mendoza-Denton, R., Rhein-
schmidt, M. L., & Keltner, D. (2012). Social class, solip-
sism, and contextualism: How the rich are different
from the poor. Psychological Review, 119, 546–572.
https://doi.org/10.1037/a0028756

K€uhnen, U., & Oyserman, D. (2002). Thinking about the
self influences thinking in general: Cognitive conse-
quences of salient self-concept. Journal of Experimental
Social Psychology, 38, 492–499. https://doi.org/10.1016/
S0022-1031(02)00011-2

Lachman, M. E., & Weaver, S. L. (1998). The sense of control
as a moderator of social class differences in health and
well-being. Journal of Personality and Social Psychology, 74,
763–773. https://doi.org/10.1037/0022-3514.74.3.763

Lalwani, A. K., & Shavitt, S. (2013). You get what you
pay for? Self-construal influences price-quality judg-
ments. Journal of Consumer Research, 40, 255–267.
https://doi.org/10.1086/670034

Lamm, H., Schmidt, R. W., & Trommsdorff, G. (1976).
Sex and social class as determinants of future orienta-
tion (time perspective) in adolescents. Journal of Person-
ality and Social Psychology, 34, 317–326. https://doi.org/
10.1037/0022-3514.34.3.317

MacKinnon, D. P., Krull, J. L., & Lockwood, C. M. (2000).
Equivalence of the mediation, confounding and sup-
pression effect. Prevention Science, 1, 173–181. https://d
oi.org/10.1023/A:1026595011371

Markus, H. R., & Kitayama, S. (2003). Models of agency:
Sociocultural diversity in the construction of action. In
V. Murphy-Berman, & J. J. Berman (Eds.), Cross-cultural
differences in perspectives on the self, Vol. 49 (pp. 1–57).
Lincoln: University of Nebraska Press.

Martineau, P. (1958). Social classes and spending behav-
ior. Journal of Marketing, 23, 121–130. https://doi.org/
10.2307/1247828

Masuda, T., Ellsworth, P. C., Mesquita, B., Leu, J.,
Tanida, S., & Veerdonk, E. V. D. (2008). Placing the
face in context: Cultural differences in the perception
of facial emotion. Journal of Personality and Social Psy-
chology, 94, 365–381. https://doi.org/10.1037/0022-
3514.94.3.365

Masuda, T., & Nisbett, R. E. (2001). Attending holistically
versus analytically: Comparing the context sensitivity
of Japanese and Americans. Journal of Personality and
Social Psychology, 81, 922–934. https://doi.org/10.1037/
0022-3514.81.5.922

McCollough, M. A., Berry, L. L., & Yadav, M. S. (2000).
An empirical investigation of customer satisfaction after
service failure and recovery. Journal of Service Research,
3, 121–137. https://doi.org/10.1177/109467050032002

Miyamoto, Y., Nisbett, R. E., & Masuda, T. (2006). Cul-
ture and the physical environment: Holistic versus ana-
lytic perceptual affordances. Psychological Science, 17,
113–119. https://doi.org/10.1111/j.1467-9280.2006.
01673.x

Miyamoto, Y., & Wilken, B. (2010). Culturally contingent
situated cognition influencing other people fosters ana-
lytic perception in the United States but not in Japan.
Psychological Science, 21, 1616–1622. https://doi.org/10.
1177/0956797610385954

Monga, A. B., & John, D. R. (2008). When does negative
brand publicity hurt? The moderating influence of ana-
lytic versus holistic thinking. Journal of Consumer Psy-
chology, 18, 320–332. https://doi.org/10.1016/j.jcps.
2008.09.009

Monga, A. B., & John, D. R. (2010). What makes brands
elastic? The influence of brand concept and styles of
thinking on brand extension evaluation. Journal of Mar-
keting, 74, 80–92. https://doi.org/10.1509/jmkg.74.3.80

Na, J., McDonough, I. M., Chan, M. Y., & Park, D. C.
(2016). Social-class differences in consumer choices:
Working-class individuals are more sensitive to choices
of others than middle-class individuals. Personality and
Social Psychology Bulletin, 42, 430–443. https://doi.org/
10.1177/0146167216634043

Newport, F. (2015). Fewer Americans identify as middle class
in recent years. Retrieved from http://www.gallup.c
om/poll/182918/fewer-americans-identify-middle-cla
ss-recent-years.aspx?

Nisbett, R. E., Peng, K., Choi, I., & Norenzayan, A.
(2001). Culture and systems of thought: Holistic versus
analytic cognition. Psychological Review, 108, 291–310.
https://doi.org/10.1037/0033-295X.108.2.291

Pham, M. T. (2016). The forgotten working-class con-
sumer. Journal of Consumer Psychology, 4, 566–567.
https://doi.org/10.1016/j.jcps.2016.09.001

Piff, P. K., Kraus, M. W., Côte, S., Cheng, B. H., & Kelt-
ner, D. (2010). Having less, giving more: The influence
of social class on prosocial behavior. Journal of Personal-
ity and Social Psychology, 99, 771–784. https://doi.org/
10.1037/a0020092

Preacher, K. J., & Hayes, A. F. (2008). Asymptotic and
resampling strategies for assessing and comparing indi-
rect effects in multiple mediator models. Behavior
Research Methods, 40, 879–891. https://doi.org/10.3758/
BRM.40.3.879

Ruggiero, K. M., & Marx, D. M. (1999). Less pain and
more to gain: Why high-status group members blame
their failure on discrimination. Journal of Personality and
Social Psychology, 77, 774–784. https://doi.org/10.1037/
0022-3514.77.4.774

Shavitt, S., Jiang, D., & Cho, H. (2016). Stratification and
segmentation: Social class in consumer behavior. Journal
of Consumer Psychology, 26, 583–593. https://doi.org/
10.1016/j.jcps.2016.08.005

Shiota, M. N., Keltner, D., & John, O. P. (2006). Positive
emotion dispositions differentially associated with Big
Five personality and attachment style. Journal of Positive

https://doi.org/10.1177/0956797610387613

https://doi.org/10.1177/0956797610387613

https://doi.org/10.1037/a0016357

https://doi.org/10.1037/a0028756

https://doi.org/10.1016/S0022-1031(02)00011-2

https://doi.org/10.1016/S0022-1031(02)00011-2

https://doi.org/10.1037/0022-3514.74.3.763

https://doi.org/10.1086/670034

https://doi.org/10.1037/0022-3514.34.3.317

https://doi.org/10.1037/0022-3514.34.3.317

https://doi.org/10.1023/A:1026595011371

https://doi.org/10.1023/A:1026595011371

https://doi.org/10.2307/1247828

https://doi.org/10.2307/1247828

https://doi.org/10.1037/0022-3514.94.3.365

https://doi.org/10.1037/0022-3514.94.3.365

https://doi.org/10.1037/0022-3514.81.5.922

https://doi.org/10.1037/0022-3514.81.5.922

https://doi.org/10.1177/109467050032002

https://doi.org/10.1111/j.1467-9280.2006.01673.x

https://doi.org/10.1111/j.1467-9280.2006.01673.x

https://doi.org/10.1177/0956797610385954

https://doi.org/10.1177/0956797610385954

https://doi.org/10.1016/j.jcps.2008.09.009

https://doi.org/10.1016/j.jcps.2008.09.009

https://doi.org/10.1509/jmkg.74.3.80

https://doi.org/10.1177/0146167216634043

https://doi.org/10.1177/0146167216634043

http://www.gallup.com/poll/182918/fewer-americans-identify-middle-class-recent-years.aspx

http://www.gallup.com/poll/182918/fewer-americans-identify-middle-class-recent-years.aspx

http://www.gallup.com/poll/182918/fewer-americans-identify-middle-class-recent-years.aspx

https://doi.org/10.1037/0033-295X.108.2.291

https://doi.org/10.1016/j.jcps.2016.09.001

https://doi.org/10.1037/a0020092

https://doi.org/10.1037/a0020092

https://doi.org/10.3758/BRM.40.3.879

https://doi.org/10.3758/BRM.40.3.879

https://doi.org/10.1037/0022-3514.77.4.774

https://doi.org/10.1037/0022-3514.77.4.774

https://doi.org/10.1016/j.jcps.2016.08.005

https://doi.org/10.1016/j.jcps.2016.08.005

Social Class Differences in Carryover Effects 465

Psychology, 1, 61–71. https://doi.org/10.1080/
17439760500510833

Smith, A. K., Bolton, R. N., & Wagner, J. (1999). A model
of customer satisfaction with service encounters involv-
ing failure and recovery. Journal of Marketing Research,
36, 356–372. https://doi.org/10.2307/3152082

Smith, H. J., & Pettigrew, T. F. (2014). The subjective inter-
pretation of inequality: A model of the relative depriva-
tion experience. Social and Personality Psychology Compass,
8, 755–765. https://doi.org/10.1111/spc3.12151

Snibbe, A. C., & Markus, H. R. (2005). You can’t always get
what you want: Educational attainment, agency, and
choice. Journal of Personality and Social Psychology, 88,
703–720. https://doi.org/10.1037/0022-3514.88.4.703

Spencer, S. J., Zanna, M. P., & Fong, G. T. (2005). Estab-
lishing a causal chain: Why experiments are often more
effective than mediational analyses in examining psy-
chological processes. Journal of Personality and Social
Psychology, 89, 845–851. https://doi.org/10.1037/0022-
3514.89.6.845

Spiller, S. A., Fitzsimons, G. J., Lynch, J. G., & McClel-
land, G. H. (2013). Spotlights, floodlights, and the
magic number zero: Simple effects tests in moderated
regression. Journal of Marketing Research, 50, 277–288.
https://doi.org/10.1509/jmr.12.0420

Stellar, J. E., Manzo, V. M., Kraus, M. W., & Keltner, D.
(2012). Class and compassion: Socioeconomic factors
predict responses to suffering. Emotion, 12, 449–459.
https://doi.org/10.1037/a0026508

Stephens, N. M., Hamedani, M. G., Markus, H. R., Berg-
sieker, H. B., & Eloul, L. (2009). Why did they “choose”
to stay? Perspectives of Hurricane Katrina observers
and survivors. Psychological Science, 20, 878–886.
https://doi.org/10.1111/j.1467-9280.2009.02386.x

Stephens, N. M., Markus, H. R., & Townsend, S. S. M.
(2007). Choice as an act of meaning: The case of social
class. Journal of Personality and Social Psychology, 93,
814–830. https://doi.org/10.1037/0022-3514.93.5.814

€ uner, T., & Holt, D. B. (2010). Toward a theory of sta-Ust€
tus consumption in less industrialized countries. Journal
of Consumer Research, 37, 37–56.

Wang, Y.-S., Wu, S.-C., Lin, H.-H., & Wang, Y.-Y. (2011).
The relationship of service failure severity, service
recovery justice and perceived switching costs with
customer loyalty in the context of e-tailing. International
Journal of Information Management, 31, 350–359.
https://doi.org/10.1016/j.ijinfomgt.2010.09.001

Williams, T. G. (2002). Social class influences on pur-
chase evaluation criteria. Journal of Consumer Market-
ing, 19, 249–276. https://doi.org/10.1108/07363760
210426067

Zhou, X., He, L., Yang, Q., Lao, J., & Baumeister, R. F.
(2012). Control deprivation and styles of thinking. Jour-
nal of Personality and Social Psychology, 102, 460–478.
https://doi.org/10.1037/a0026316

Zhu, R., & Meyers-Levy, J. (2009). The influence of self-
view on context effects: How display fixtures can affect
product evaluations. Journal of Marketing Research, 46,
37–45. https://doi.org/10.1509/jmkr.46.1.37

Supporting Information

Additional supporting information may be found in
the online version of this article at the publisher’s
website:

Appendix S1. Methodological Details.

https://doi.org/10.1080/17439760500510833

https://doi.org/10.1080/17439760500510833

https://doi.org/10.2307/3152082

https://doi.org/10.1111/spc3.12151

https://doi.org/10.1037/0022-3514.88.4.703

https://doi.org/10.1037/0022-3514.89.6.845

https://doi.org/10.1037/0022-3514.89.6.845

https://doi.org/10.1509/jmr.12.0420

https://doi.org/10.1037/a0026508

https://doi.org/10.1111/j.1467-9280.2009.02386.x

https://doi.org/10.1037/0022-3514.93.5.814

https://doi.org/10.1016/j.ijinfomgt.2010.09.001

https://doi.org/10.1108/07363760210426067

https://doi.org/10.1108/07363760210426067

https://doi.org/10.1037/a0026316

https://doi.org/10.1509/jmkr.46.1.37