Grocery Shopping for America: External vs. Internal Threats to … · 2018-12-12 · GROCERY...
Transcript of Grocery Shopping for America: External vs. Internal Threats to … · 2018-12-12 · GROCERY...
Running Head: GROCERY SHOPPING FOR AMERICA 1
Grocery Shopping for America:
External vs. Internal Threats to National Identity
Sonal Pandya∗, Luca Cian†, & Raj Venkatesan†
October 29, 2018
∗Corresponding Author. Department of Politics, University of Virginia, P.O. Box 400787,Charlottesville, VA 22904-4787; Phone: 434-243-1573; Fax: 434-243-3359; [email protected]†Darden School of Business, University of Virginia
GROCERY SHOPPING FOR AMERICA 2
Abstract
Nationalist political strategies capitalize on the psychology of external threats
to justify harshness towards outgroups. We hypothesize that while external threats
strengthen national identification, harshness towards outgroups that degrades in-
group’s constituent values (internal threat) weakens national identification. We
test the causal effects of US war casualties (external threat) and Abu Ghraib
torture scandal (internal threat) on national identification using weekly sales of
American-sounding supermarket brands, a behavioral proxy for national identifi-
cation. In our sample spanning over 8,000 brands and 1,100 supermarkets, the
market share of American-sounding brands increased in stores following the death
of a solider from the same county. These same brands’ national market shares
declined during Abu Ghraib. A July 2018 lab experiment reveals that Chinese im-
port competition (external threat) strengthens Americans’ national identification
but backlash against refugee family border separations (internal threat) weakens
identification. Our findings suggest that nationalist political strategies can backfire
if pushed too far.
GROCERY SHOPPING FOR AMERICA 3
The rise of nationalism, evidenced by Brexit, the election of Donald Trump, white
supremacist rallies, and extremist political parties, suggests political strategies that build
on the psychological foundations of national identity. Such strategies emphasize exter-
nal threats—war, trade, immigration—to justify harsh treatment of the nation’s puta-
tive enemies. We compare the effects of external threats—threats to the nation from
outgroups—and internal threats—threats from within the nation—on Americans’ na-
tional identification.
Social identity theory (SIT) posits that individual identity derives partially from
group affiliations (Tajfel & Turner, 1979) via categorization (e.g., “I am American.”)
and self-enhancement (e.g., “I am proud to be American.”) (Hogg, 2006). SIT pos-
tulates a fundamental need for positive self-esteem, which is partially based on inter-
group comparisons and dynamics (Sedikides, 1993; Sedikides & Strube, 1997). As such,
external threats to the group may strengthen individual identification with the group
(Branscombe, Ellemers, Spears, & Doosje, 1999; Davies, Steele, & Markus, 2008), re-
flected in the adoption of prototypic group behaviors (Akerlof & Kranton, 2000). When
considering internal threats, instead, one common reaction consists of the rejection of
threatening ingroup member(s) (Elsbach & Bhattacharya, 2001), the easiest cognitive
strategy to preserve a positive self-view (Marques, Paez, & Abrams, 1998).
Do the dynamics of social identity give nationalist politicians carte blanche? We
propose a distinct type of internal threat, actions of ingroup members that degrade the
group’s constituent values. These system-wide threats render the common strategy of
rejecting threatening ingroup members less effective. We hypothesize that the same self-
enhancement mechanism that strengthens national identification in response to external
threats acts as a circuit breaker, weakening national identification in response to na-
tionalist excesses. Weaker identification is the easiest cognitive strategy to preserve a
positive self-view in these scenarios.
We test our hypotheses using war-related threats the US faced in 2004. By observing
GROCERY SHOPPING FOR AMERICA 4
the differential effects of external and internal threats, we can establish their opposite
effects on national identification. War casualties are an external threat: US soldiers
(ingroup members) die while defending against an outgroup. By the end of 2004 nearly
1500 American soldiers had died in Iraq and Afghanistan. The Abu Ghraib scandal
was an internal threat; US soldiers tortured members of the outgroup. In late April
2004, images of American soldiers torturing Iraqi prisoners emerged and were seen by
76% of Americans within two weeks (Pew Research Center, 2004). Abu Ghraib was
widely condemned as an affront to American ideals; a majority of Americans reported
being upset or angry about the incident (Greenberg, 2005).1 The scandal undermined
American democratic ideals, one of the main justifications for the Iraq War (Greenberg,
2005). For example, US Defense Secretary Rumsfeld described the incident as “un-
American” and “inconsistent with the values of our nation.”2
We expect that war casualties strengthened Americans’ national identification, re-
flecting well-established SIT mechanisms. By contrast, we argue that Abu Ghraib weak-
ened national identification.
Using Branscombe et al.’s (1999) formulation of social identity threats, war casual-
ties and Abu Ghraib are best interpreted as external and internal threats to national
identity, respectively. In early 2004, 70% of Americans described the Iraq War as going
“very/fairly well” (Pew Research Center, 2004), suggesting casualties were not a compe-
tence threat. Nor do the events reflect realistic vs. symbolic threats (Stephan, Ybarra, &
Morrison, 2009). War casualties embodied both the realistic and symbolic threats under-
lying war motives: prevention of terrorism and weapons proliferation, democratization,
human rights.3
Compelling empirical tests of our claims require comparing simultaneous internal
and external threats within a single design. Extant research relies on laboratory experi-
1See Supplemental Materials for in-depth discussion of public response to the Abu Ghraib scandal.2Hearing, Congressional Armed Services Committees, May 7, 2004.3US Congressional Joint Resolutions 107-40-September 18, 2001 and 107-243-October 22, 2002.
GROCERY SHOPPING FOR AMERICA 5
ments, which may lack external validity, or self-reported surveys, often subject to social
desirability bias and demand effects (Barbera, Jost, Nagler, Tucker, & Bonneau, 2015).
Real-time threats are difficult to be reproduced in the lab and fictionalized threats may
be unrealistic. Recall of past threats is vulnerable to ex post rationalization (Aronczyk,
2013; Brewer, 1984).
We measure shifts in Americans’ national identification using weekly supermarket
scanner data by taking American-sounding supermarket brands as symbols of American
identity. This approach capitalizes on tight links between social identity and super-
market purchases (Escalas & Bettman, 2005). Consumers prefer brands consistent with
their most salient social identities (Khan, Misra, & Singh, 2013), and brands that bol-
ster their self-worth (Shachar, Erdem, Cutright, & Fitzsimons, 2011). Compensation for
own/other’s ethical failings shapes consumption behavior (Mari, Andrighetto, Gabbia-
dini, Durante, & Volpato, 2010; Zhong, Liljenquist, & Cain, 2009).
Mass consumption is a behavioral proxy for psychological constructs. By testing the
observable implications of our claims for consumption, our empirical tests have a high
degree of external validity and overcome limitations of extant research designs (Webb,
Campbell, Schwartz, & Sechrest, 1966). Supermarket purchasing is a frequent, con-
sistent, and nearly universal behavior in the United States. The average household
purchases groceries weekly (Kahn & Schmittlein, 1989). We observe nearly real-time
responses to threats and can evaluate temporal shifts in the strength of national identifi-
cation. Supermarket scanner data also provide detailed contextual information including
price and selection. As compared to opinion surveys, purchases are less subject to social
desirability bias.
In Study 1, we analyze weekly sales of American-sounding brands for a representative
sample of over 1,100 stores in 50 geographic markets. Our data span over 8,000 brands
across 30 product categories. We measure brands’ perceived American origin using
surveys. For a given store, weekly casualty exposure is the count of war casualties in a
GROCERY SHOPPING FOR AMERICA 6
week whose hometowns are in the same county as the store. Internet search patterns
capture weekly variation in Abu Ghraib exposure.
Our estimates represent the causal effects of threats on the strength of national
identification. War casualties and Abu Ghraib were both exogenous shocks that influ-
enced market share only though their effects of consumers’ national identification. For
each store-week in 2004, we model the change in the weekly market share of American-
sounding brands in 2004 as compared to the same store-week in 2001. By analyzing
differences, we hold constant time invariant store characteristics including the ex ante
demand for American-sounding brands, customer demographics, and seasonal fluctua-
tions. The timing of a store’s exposure to local causalities is quasi-random, which holds
constant all national-level factors that may influence national identification such as politi-
cians’ rhetoric or war performance. We control for zip code-level military enlistment to
account for cross-sectional variation in potential casualty exposure. Abu Ghraib was
a common shock across all stores. We control for weekly zip code-level housing prices
in case time-varying economic conditions incidentally coincided with casualties or Abu
Ghraib.
We find that in 2004, the market share of American-sounding brands increased in
stores in the weeks they had a local casualty, reflecting stronger national identification.
During the Abu Ghraib scandal, by contrast, the average market share of American-
sounding brands declined nationwide, evidence of weaker national identification.
The Abu Ghraib-induced decline is robust to restricting the sample to stores that
experienced casualties before the scandal and measuring exposure as cumulative local
casualties since 2002. Indeed, we show that higher cumulative casualties produce a
larger rise in American-sounding brands’ market share. This finding suggests that, over
time, cumulative external threats magnify shifts in national identification. Our research
uniquely lends itself to assessing cumulative effects. Stores with ex ante stronger as-
sociations to national identity saw larger changes in market share in response to both
GROCERY SHOPPING FOR AMERICA 7
threats.
Study 2 confirms the hypothesized psychological processes with an experiment. Our
July 2018 study drew treatments from real-time nationalist policies: Chinese trade com-
petition (external threat) and public outrage over family separation at the US-Mexico
border (internal threat). Consistent with our hypotheses, external (internal) threats
strengthened (weakened) national identification, measured as personal identification with
American identity, pride in that identity, and America’s superiority.
Study 1: Observational Evidence
Method
Measurement. Our analysis requires measurement of three concepts: the perceived
American origin of brands; weekly supermarket purchases; and the exposure to local war
casualties and the Abu Ghraib scandal.
Perceived Brand Nationality. We measure perceived brand nationality based on
product brand names because names are a highly salient, readily available cue (Usunier
& Shaner, 2002). For American consumers, brand names based on foreign languages
frequently evoke associations with a foreign country through distinctive letter combina-
tions and special characters, such as umlauts and accent marks that do not occur in
English. By contrast, brands that incorporate geographic locations in the US or Ameri-
can cultural symbols imply American-made products. Survey and experimental evidence
shows consumers frequently misidentify the national origin of products because they in-
fer nationality from marketing cues, rather than searching for official country of origin
labels (Balabanis & Diamantopoulos, 2011; Samiee, Shimp, & Sharma, 2005). Brand
nationality is a cue that operates outside of consumers’ conscious awareness in a man-
ner analogous to social stereotypes (Liu & Johnson, 2005; Martin, Lee, & Lacey, 2011).
GROCERY SHOPPING FOR AMERICA 8
Consumers draw inferences based on prior associations between the implied country and
the product.
We administered surveys to assess the perceived nationality of brands via an online
US-based subject pool. 1203 participants received a randomly selected brand name and
its product category and asked “What nationality does this brand most make you think
of?” Ten possible responses included eight nationalities (American, Chinese, English,
French, German, Italian, Japanese, Spanish), “none,” and “other.” Participants were
paid for each brand evaluation and allowed a maximum of 20 evaluations to minimize
respondent fatigue. Each of the over 8,000 brands in our data had seven independent
evaluations.
Using these data we calculated AmericanScoreb , which ranges 0-7 reflecting the
number of respondents who deemed brand b to be American. Table 1 provides examples
of brands at each variable value. Brands with AmericanScoreb = 7 exhibit strong Amer-
ican nationality cues including geographic references and historical figures. Coca-Cola
is an example of a high profile brand perceived as American though without explicit
American branding cues. Lower-scoring brands have distinctively foreign elements in-
cluding words in other languages and foreign geographic references (see Table S1 in the
Supplemental Material for the distribution of American Score across product categories).
American-sounding brands are typically objectively American in as much as they are
brands owned by US-based firms and/or are trademarks owned by Americans.
Pre Test. In a pre test we verified that American-sounding brands indeed symbolize
American national identity:4 The brands that score high (low) on our American Score
should be strong (weak) symbols of America.
To measure American symbolism, we adapted a three-item scale from Steenkamp,
Batra, and Alden (2003; To me, this brand is a symbol of America; I associate this
brand with things that are American; To me, this brand represents American values.
4Preregistered at AsPredicted.org.
GROCERY SHOPPING FOR AMERICA 9
Table 1.Brand Examples Across American Score ValuesAmericanScoreb Brand Example (Product Category)
7Sam Adams Boston Lager (beer)Coca Cola (carb. beverages)
6Land O’ Lakes (margarine/butter)Phillies (hot dogs)
5Olde Cape Cod (spaghetti sauce)Swanson American Recipes (frozen dinners)
4New England (ketchup/mustard)Dad’s Root Beer (carb. beverages)
3Maple Leaf (hot dogs)Van De Kamps (frozen dinners)
2Life in Provence Aioli (mayonnaise)Jubilee (ketchup/mustard)
1Royal Scot (margarine/butter)World Trend (toothbrushes)
0Konig Ludwig Weiss (beer)Anna Mario’s (spaghetti sauce)
The rating options ranged from 1 to 7, with 1 meaning “I completely disagree with this
statement,” and 7 meaning “I completely agree with this statement.”). As such, we
expected a correlation between our American Score and the American Symbolism scales.
Given the massive number of brands (8,644) already evaluated with the American
Score scale, we decided to sample only a selection of them (40 brands) to test the corre-
lation between the American Score and the American Symbolism scales.
Brand Selection. We randomly selected five different product categories (beer, laun-
dry detergent, toothpaste, spaghetti sauce, and frozen dinner). For each one of these
categories, we needed to include brands belonging to different levels of the American
Score scale. As such, we randomly selected two brands with an American Score of zero,
three, five, and seven. Summarizing, the selection of these 40 brands was the result
of two randomly selected brands for each of four levels of the American Score in five
different brand categories.
Study design. We assessed that evaluating 40 brands in a single session could generate
survey fatigue. As such, we assigned participants to evaluate only 20 brands (instead of
GROCERY SHOPPING FOR AMERICA 10
40), in a random order. Each respondent evaluated a brand that scored zero, three, five,
and seven on each of the five different brand categories. This generated a mix-model
design, in which we had 20 brands, within-subjects (four American Score levels five brand
categories), and two randomly generated brand lists, between-subjects. To exemplify,
Respondent 1 evaluated 20 brands (one brand for four different American Score levels
five brand categories—what we called “List A”). Respondent 2 evaluated the other 20
brands (one brand for four different American Score levels five brand categories—what
we called “List B”). Respondent 3 evaluated List A, Respondent 4 evaluated List B,
and so on. Table 2 shows the 40 brands used and identifies to which of the two lists it
belonged.
Respondent sample. 400 US-based participants from an online pool participated in
this study in exchange for money.
Analysis. We analyzed the Cronbach’s alpha of the three items we used to measure
American symbolism. Given that the alpha was high (.96), we decided to average these
three items in an index (“American Symbolism scale”). We then performed a mix-model
ANOVA with the two Lists (A and B) as the between-subjects independent variable,
and the 20 brands evaluated by each person as the within-subjects dependent variable.
The means of each brand are reported in Table 2.
Table 2.American Symbolism means for each of the 40 brands tested
Category List Brand NameAmerican
Score
American
Symbolism
Scale (mean)
American
Symbolism
Scale (95% CI)
beer List A Guinness 0 2.82 2.57, 3.06
beer List B Tequiza 0 2.25 2.07, 2.43
beer List A Sierra Blanca 3 2.70 2.50, 2.91
beer List B Molson 3 2.99 2.78, 3.20
beer List A Red Wolf 5 3.81 3.59, 4.03
beer List BBrooklyn
Brewery5 5.09 4.87, 5.31
GROCERY SHOPPING FOR AMERICA 11
beer List A Budweiser 7 5.81 5.63, 5.99
beer List BGreat Lakes
Brewing7 5.18 4.98, 5.38
laundry
detergentList A Blanca Nieves 0 2.41 2.21, 2.60
laundry
detergentList B Paloma 0 2.81 2.62, 2.99
laundry
detergentList A Ariel 3 3.18 2.97, 3.39
laundry
detergentList B Citra Suds 3 3.11 2.91, 3.31
laundry
detergentList A Ajax 5 4.69 4.46, 4.91
laundry
detergentList B Method 5 4.08 3.86, 4.30
laundry
detergentList A
Arm &
Hammer7 5.56 5.39, 5.73
laundry
detergentList B Tide 7 5.52 5.33, 5.70
toothpaste List A Elgydium 0 2.49 2.29, 2.69
toothpaste List B Dabur 0 2.17 2.00, 2.34
toothpaste List A Butler 3 3.40 3.18, 3.62
toothpaste List B Shane 3 3.28 3.06, 3.50
toothpaste List A Mentadent 5 3.65 3.43, 3.87
toothpaste List B Choice 5 3.73 3.51, 3.95
toothpaste List A Colgate 7 5.38 5.20, 5.56
toothpaste List B Crest 7 5.44 5.26, 5.62
spaghetti
sauceList A Cucina Antica 0 2.41 2.21, 2.62
spaghetti
sauceList B
Anna
Mario’s0 2.94 2.75, 3.14
spaghetti
sauceList A Prego 3 3.41 3.19, 3.63
spaghetti
sauceList B Roland 3 3.32 3.12, 3.52
GROCERY SHOPPING FOR AMERICA 12
spaghetti
sauceList A
Sonoma
Gourmet5 4.34 4.12, 4.57
spaghetti
sauceList B Ragu 5 4.51 4.29, 4.74
spaghetti
sauceList A
California
Seasonings7 4.77 4.55, 4.99
spaghetti
sauceList B
Uncle
Dave’s7 4.43 4.21, 4.65
frozen
dinnerList A Ajinomoto 0 2.27 2.07, 2.46
frozen
dinnerList B
Gallina
Blanca0 2.31 2.13, 2.49
frozen
dinnerList A
Michelina’s
Signature3 3.65 3.43, 3.87
frozen
dinnerList B
Bobby
Salazars3 3.01 2.82, 3.19
frozen
dinnerList A
Healthy
Choice5 4.85 4.64, 5.06
frozen
dinnerList B
Seeds of
Change5 3.62 3.41, 3.84
frozen
dinnerList A
Boston
Market7 5.45 5.26, 5.64
frozen
dinnerList B Uncle Ben’s 7 5.20 5.00, 5.41
This mix-model ANOVA revealed a non-significant effect of the list (F(1, 398) = 1.30,
p = .254, ηp2 = .003 ) and significant main effect of the brands (different brand names
were evaluated differently on the American Symbolism scale; F(19, 7,562) = 348.15, p <
.001, ηp2 = .467). The interaction between the two factors was significant (F(19, 7,562)
= 19.63, p < .001, ηp2 = .047). A significant interaction was unexpected but simply
indicated that, within the same brand level, sometime the means were higher in List A
and some other times were higher in List B.
Given that there was no general main effect of the of the Lists, we pooled the data
GROCERY SHOPPING FOR AMERICA 13
and grouped the 20 brands of List A and 20 brands of List B together. We then ran
a correlation between American Score (0, 3, 5, 7) and American Symbolism scale. The
two scales were strongly and positively correlated, r(7,998) = .56, p < .001 (note that
7,998 degrees of freedom, n = 8,000, equals to 20 brands ∗ 400 people). Table 3 shows
how each subsequent level of the American Score corresponded to a higher rating on the
American Symbolism scale.
Table 3.Average American Symbolism Means for Each Level of AmericanScore
American ScoreAmerican Symbolism
Scale (mean)American Symbolism
Scale (95% CI for mean)0 2.49 2.43, 2.553 3.20 3.14, 3.275 4.24 4.17, 4.317 5.27 5.21, 5.34
Main Study. We measure consumer behavior using weekly supermarket sales data
supplied by Information Resources Inc. (IRI), a leading source of US supermarket scan-
ner data (Bronnenberg, Kruger, & Mela, 2008). These data cover a representative sample
of 1,145 supermarkets across 50 IRI-designated geographic markets.5
The 135 supermarket chains represented in the data collectively accounted for about
80% of US supermarket sales in 2004. During the sample approximately 70% of American
grocery purchases were in supermarkets.
We construct our store-level measure of consumer response using weekly unit sales for
8,644 brands across 30 product categories: beer, blades, carbonated beverages, cigarettes,
coffee, cold cereal, deodorant, diapers, facial tissue, frozen dinners, frozen pizza, house-
hold cleaners, hot dogs, laundry detergent, butter, mayonnaise, milk, mustard/ketchup,
paper towel, peanut butter, photo, razors, salty snacks, shampoo, soup, spaghetti sauce,
sugar substitutes, toilet tissue, toothbrush, yogurt. Major supermarket chains stock ma-
5See Figure S2 in the Supplemental Material for map.
GROCERY SHOPPING FOR AMERICA 14
ture brands and maintain a relatively stable portfolio of brands within each store. We
aggregate data across multiple stock keeping unit (SKU) codes of a single brand-product
category (e.g., six-pack of Coke, two-liter bottle of Coke) but not across distinct but
related brands (e.g., Coke and Diet Coke).
For each product category-store-week in our dataset, we model the change in market-
share growth rate between 2001 and 2004. Our outcome of interest is indexed by:
i: 8 American Score levels,
j: 1,154 supermarkets,
k: 30 product categories, and
t: 52 weeks.
A brand’s weekly store market share is the number of brand product units sold as a
percentage of all units in the product category sold in that store-week. For example, if
brand b in product category k (e.g., yogurt) had a 50% market share in a given store j for
week t, the brand accounted for half of all units of yogurt sold in that store in that week.
Measuring market share, as opposed to the total number of units sold, allows us to scale
that store’s sales of a brand relative to overall demand for that product category in that
store-week. Changes in market share also capture shifts in demand for brands distinct
from changes in demand for a particular product category. For each category-store-
week we calculate the average market share across brands at each of the eight levels of
AmericanScorei. This aggregation reflects our interest in change across AmericanScorei
levels rather than individual brands and reduces the sample to a computationally feasible
size. As compared to sampling a subset of stores, this approach minimizes computational
burden, maintains generalizability and utilizes variation in casualties across all stores.
For every American Score level-product category-store-week in our sample, we calcu-
late the change in market share between 2004 and 2001 (Share2004ijkt -Share2001
ijkt ). 2001 is
the first year for which scanner data are available. Measuring change in demand within
each store allows us to hold constant all time-invariant baseline characteristics of the
GROCERY SHOPPING FOR AMERICA 15
store’s customer base that influence sales, including ex ante customer preferences. If we
were to observe sales only in 2004, we could not differentiate between a change in demand
and preexisting low demand. We choose 2001, the first year for which scanner data are
available, as a baseline because it precedes almost all war casualties.6 For each store, we
retain only brands that were sold in all weeks of 2004 and 2001 so our results are not
biased by attrition and entry. We also hold constant seasonal fluctuations by comparing
2004 and 2001 shares in the same week. For example, sales of American-sounding brands
may rise around July 4th or Memorial Day.
External and Internal Threats to National Identity: War Casualties and
Abu Ghraib. From the perspective of given supermarket, a “local” casualty is the
death of a deployed US soldier whose hometown is in the same US county as the store.7
Casualty data are from US Defense Department press releases as compiled by the Asso-
ciated Press. We matched each casualty’s self-reported hometown to its corresponding
county and summed county-week casualties. Hometown is distinct from service unit.
For example, Fort Hood, Texas had 504 Iraq War casualties representing 427 unique US
hometowns. We measure casualties at the county level because it is the most conser-
vative measure of exposure we can accurately construct and is consistent with existing
studies of local casualty responses (Kriner & Shen, 2010). Figure 1 summarizes weekly
national casualty counts (denoted on left vertical axis) in 2004. In most weeks the US
had no more than twenty five casualties total.
Consumers are more likely to be aware of local casualties. Proximity increases the
likelihood of exposure to information about the casualty through local media and social
networks and a personal connection. Local casualties shift war support among non-
consumers of news media, suggesting they obtain information about casualties through
other conduits (Althaus, Bramlett, & Gimpel, 2012). Even in the absence of a tangible
6If 9/11 and/or the 11 US war casualties in 2001 increased sales of American-sounding brands, thiswould bias against our expected finding for 2004.
7See Table S2 and Figure S3 in the Supplemental Material for a distribution across stores.
GROCERY SHOPPING FOR AMERICA 16
Fig. 1. Weekly Trends: US War Casualties vs. Abu Ghraib
connection, consumers and casualties share identity rooted in a common place of origin.
Self-reported personal connection to someone injured or killed in Iraq War increases the
likelihood that Iraq War drives choice of political candidate (Gartner, 2009) and presi-
dential approval (Gartner, 2008). The relatively high percentage of survey respondents
who report knowing a casualty, implausible given the number of troops deployed, sug-
gest strong perceived connections to casualties (Gartner, 2009; Kriner & Shen, 2010).
Experimental evidence shows stronger opposition to war in response to casualties from
respondent’s own state independent of local news framing (Kriner & Shen, 2012). Our
focus on local casualties holds constant all national-level factors that may influence re-
sponses to external threats including current state of the two wars and priming by elected
official and/or national media.
We measure weekly variation in Americans’ exposure to Abu Ghraib based on the
volume of web searches for “Abu Ghraib.” Online search trends capture the revealed
salience of the event for the mass public. Data are from Google Trends, which is based
GROCERY SHOPPING FOR AMERICA 17
on US Google searches to produce a normalized score that can be compared across
weeks. The Abu Ghraib scandal came to light in the last week of April 2004, the
week scored as 100 on the right vertical axis of Figure 1. Intense search activity lasted
three to four weeks. The figure suggests that Abu Ghraib did not influence casualties
counts in subsequent weeks. See Abu Ghraib in the Supplemental Material for a detailed
discussion.
Empirical Model. We estimate a ordinary least squares model:
Share2004ijkt − Share2001ijkt = β0 + β1(AbuGt ∗ AmScoreijk) (1)
+ β2(Casjt ∗ AmScoreijk)
+ β3(HomePrice2004jt ∗ AmScoreijk)
+ β4(Enlistments2004j ∗ AmScoreijk)
+ β5AbuGt
+ β6Casjt
+ β7AmScoreijk
+ β8HomePrice2004jt
+ β9Enlistments2004j
+ β10(Price2004ijkt − Price2001ijkt )
+ β11(NumSkus2004ijkt −NumSkus2001ijkt ) + eijkt
The coefficients of interest are β1 and β2. They measure to what extent Abu Ghraib
and local casualties respectively shifted the market share of relatively more American-
sounding brands.
GROCERY SHOPPING FOR AMERICA 18
Control Variables. In order to interpret our findings as causal, we assume that
local casualties and Abu Ghraib influenced brand purchases only by shifting consumers’
identification with American national identity. Neither war casualties nor Abu Ghraib
produced explicit calls for Americans to change their consumption, or systematically
changed brand characteristics or availability. We control for store-level price and stock
changes in the same week, the only real-time responses available to retailers. Controlling
for military enlistment, a store’s exposure to casualties was quasi-random. For each
store, we control for zip code-level military enlistment to control for the non-random
component of local casualty exposure. In case unobserved economic shocks correlated
with response to local casualties and/or Abu Ghraib, we also control for weekly average
home prices by store zip code.
The model includes Enlistmentj, the sum of military enlistments for 2001-2003 in the
same zip code as store j. Enlistment data covers all military branches and are assigned
to zip codes based on enlistees’ home address. These are Department of Defense data
obtained via a Freedom of Information Act request. Conditional on military enlistment, a
store’s exposure to war casualty is quasi-random. The timing of local casualties is quasi-
random but enlistment is not because the US has an all-volunteer military. In 2004 all
zip codes in our data had at least one enlistment. To the extent that communities with
higher enlistment are systematically different, we can draw sound inferences about only
those communities.
The model also includes HomePricejt, average home price in store j’s zip code in
week t.8 These data are from zillow.com. The variable accounts for zip code-week wealth
shocks that may influence the propensity to respond to one or both threats. Local wealth
shocks in a given week are unlikely to correlate with weekly exposure to local casualties.
Prices of American-sounding brands are not systematically different such that wealth
shocks would influence consumer purchases independent of casualties.
8Results are unchanged if variable calculated as the 2004-2001 difference or if the interaction of marketshare in 2001 and American Score is included to predict market share in 2004.
GROCERY SHOPPING FOR AMERICA 19
As is standard in empirical marketing analyses, we control for two time-varying brand-
store characteristics that influence fluctuations in market share (Ataman, Van Heerde,
& Mela, 2010). ∆Price2004 -2001ijkt controls for exogenous price changes and the effect
of promotional, time-limited price discounts. Non-pricing responses, such as advertising,
were less likely because they require longer lead times to implement. Price promo-
tions are retailers’ fastest response to negative demand shocks. Retailers’ contracts with
manufacturers forbid changes to products’ shelf space allocation and location, so no
retailer-driven change in product supply or location is possible in the short run.
We also control for weekly changes in the average number of varieties across all the
brands within each American score level that is stocked by a store in a product category.
All else equal, consumers are more likely to purchase brands belonging to an American
score level if a store stocks more varieties. ∆NumVariants2004 -2001ijkt is the change in
the average number of SKUs across all brands with AmericanScorei for product category
k in store j from 2001 in week t.
Our controls for prices and number of product varieties stocked are relatively stable
across weeks, as is characteristic of sales in well-established grocery retailers.
Empirical Findings
Table 4 summarizes estimates of our baseline models.9 In Model 1, the interaction of
Abu Ghraib and American Score is negative and statistically significant (β1 = -4e-06, p
< .01), confirming our expectation that American consumers reduced their purchases of
American-sounding brands in the weeks that Abu Ghraib was most salient. This result
is unchanged when we add local casualties and its interaction with American Score
(Model 2). In Model 2, the interaction between casualty in store j in week t (Casjt) and
AmericanScorei in store j in category k (AmScoreijk) is positive and significant (β2 =
2.17e-04, p < .01). This result demonstrates strengthened national identity in response
9See Table S3 in the Supplemental Material for summary statistics.
GROCERY SHOPPING FOR AMERICA 20
to external threats.
Table 4.External vs. Internal Threats: Baseline ResultsVariable Model 1 Model 2
Intercept0.001***(.00012)
9.9e-04***(1.2e-04)
AbuGTt ∗ AmScoreijk-4e-06***(1e-06)
-4e-06***(1e-06)
Casjt ∗ AmScoreijk2.37e-04***(5.7e-05)
AbuGTt1e-06(3e-06)
2e-06(3e-06)
AmScoreijk-5.24e-04***(2.6e-05)
-5.2e-04***(2.6e-05)
Casjt-6.5e-04**(2.6e-04)
HomePricejt2.7e-05***(3e-06)
2.7e-05***(3e-06)
Enlistmentsjt-1.74e-03***(2.8e-04)
-1.68e-03***(2.8e-04)
HomePricejt ∗ AmScoreijk-5e-06***(1e-06)
-5e-06***(1e-06)
Enlistmentsjt ∗ AmScoreijk2.88e-04***(6.2e-05)
2.67e-04***(6.2e-05)
(Price2004ijkt − Price2001ijkt )-5.28e-04***(1e-05)
-5.27e-04***(1e-05)
(NumSkus2004ijkt −NumSku2001ijkt )0.0085***(9e-06)
0.0085***(9e-06)
R-square 12.7% 12.7%n 6,344,222 6,344,222
***p<0.01, **p<0.05, *p<0.10
The average change in market share between 2004 and 2001 across all brands and
stores in any week was -.08%. The estimated model predicts that in counties with a single
casualty, the change in market share for brands with an American Score = 2 increases by
65% and the change in market share for brands with an American Score = 5 increases by
154%. The model also predicts that the share of American Score = 2 brands decreases
further by 91% and the share of American Score = 5 brands decreases further by 227%
during Abu Ghraib.
GROCERY SHOPPING FOR AMERICA 21
These findings are robust to zip code-level controls for weekly average housing prices
and military enlistment. Shares of American-sounding brands decreased more in stores
located in zip codes with higher consumer wealth in 2004 (β3 = -5e-06, p < .01) and
increased more in stores with greater enlistments or likelihood of exposure to casualties
(β4 = 2.67e-04, p < .01). The effect of internal and external threats are hence also robust
to differences in consumer wealth in 2004 (as measured by the average home prices in a
zip code in 2004) and a zip code’s likelihood of exposure to casualties. Controls for price
and number of varieties perform as expected.
Table 5 summarizes estimates based on alternative measures of casualty exposure.
Model 1 restricts the sample to stores that experienced their first local casualty prior
to May 2004 (i.e., prior to Abu Ghraib). Focusing on this subsample of stores is a
more challenging test because all stores were previously exposed to external threats
that strengthen national identification. Our baseline results holds: sales of American-
sounding brands decline during Abu Ghraib but rise in response to local casualties.
Model 1 further shows the decline during Abu Ghraib happened even in stores that
were exposed to casualties beforehand (i.e., exposed to external threats that strengthen
national identity.) The effect size of the external threat is also higher among stores that
experienced an internal threat prior to the external threat.
Table 5, Model 2 is based on cumulative total casualties from the beginning of 2002
through week t. Responses to internal and external threats may vary based on the
frequency of external threats. CumulativeCasjt*AmScoreijk is positive and statistically
significant. Finding is robust to using log(CumulativeCasjt). See Tables S4 and S5 in
the Supplemental Material.
We use Models 3 and 4 to compare the marginal effect of a single casualty and cu-
mulative casualties on AmericanScorei = 4 and AmericanScorei = 7 brands. In Figure
2 for each of the two AmericanScorei levels, we plot the change in market share be-
tween 2004 and 2001 at three values of cumulative casualties: one standard deviation
GROCERY SHOPPING FOR AMERICA 22
Table 5.External vs. Internal Threats: Dimensions of Local CasualtyExposureVariable Model 1 Model 2
Intercept1.35e-03***(1.68e-04)
2.42e-03***(1.80e-04)
AbuGTt ∗ AmScoreijk-5.13e-06***(1.28e-06)
-4.61e-06***(1.27e-06)
Castj ∗ AmScoreijk3.21e-04***(8.23e-05)
CumulativeCasjt ∗ AmScoreijk4.02e-05***(3.56e-06)
AbuGTt-1.26e-06(5.84e-06)
-3.80e-06(5.80e-06)
AmScoreijk-5.20e-04***(2.52e-05)
-6.69e-04***(2.90e-05)
Casjt-1.22e-03***(3.75e-04)
CumulativeCasjt-3.20e-06(1.68e-05)
HomePricejt1.03e-05***(2.57e-06)
-1.37e-05***(2.92e-06)
Enlistmentsjt-1.21e-04(2.51e-04)
-1.36e-03***(2.61e-04)
(Price2004ijkt − Price2001ijkt )-7.54e-04***(2.10e-05)
-7.37e-04***(2.10e-05)
(NumSkus2004ijkt −NumSku2001ijkt )9.17e-03***(1.85e-05)
9.17e-03***(1.85e-05)
R-square 13.6% 13.6%n 1,605,300 1,605,300
***p<0.01, **p<0.05, *p<0.10
below the mean (“low”), mean (“medium”), and one standard deviation above the mean
(“high”). Comparing weekly casualties across the two AmericanScorei levels demon-
strates larger market share growth for more American-sounding brands in response to
local casualties in the same week. Comparing cumulative casualties, market share growth
of AmericanScorei = 7 brands is marked higher for stores with above average cumula-
tive casualties. This finding suggests that repeated exposure to external threat magnifies
the embrace of national identity.
10Based on Table 5, Column 2 Estimates
GROCERY SHOPPING FOR AMERICA 23
Fig. 2. Cumulative Casualties Generate Sharper Increase in High AmericanScore Market Share10
Finally, we assess how ex ante variation across stores in the strength of national
identification shapes responses to external and internal threats. We split the stores in
our sample into two groups: stores whose average market share of AmericanScorei >
3 brands in 2001 (prior to war) was above the sample median, and stores whose value
was below the median. We estimated equation 1 separately for each group. Figure 3
plots regression coefficients for war casualties (external threat) and Abu Ghraib (internal
threat) for each group (See Table S6 and S7 in the Supplemental Material for underlying
regression results.) For the above median sample—stores with stronger national identifi-
cation in 2001—effect sizes are smaller for both types of threats as compared to stores in
the below median sample. These pattern confirm existing findings that high identifiers
may be generally more immune to threat, and are therefore better able to rationalize
(internal or external) threats precisely because of the strength of their group identity
(Branscombe et al., 1999; Ellemers & Haaker, 1995).
GROCERY SHOPPING FOR AMERICA 24
Fig. 3. Response to External and Internal Threats by Ex Ante National Identification
Study 2: Experimental Evidence
Study 1 establishes the causal effects of external and internal threats on national iden-
tification with a high degree of external validity. Study 2, a preregistered experiment,
provides cross-methodological evidence to establish internal validity. The experiment,
conducted in July 2018, capitalizes on real time threats to American national identity:
Chinese trade competition (external threat) and public opposition to refugee family sep-
arations at the US-Mexico border (internal threat). A July 2018 Gallup poll found that
62% of Americans believe Chinese trade policies are unfair to the US.11 At the same time,
two-thirds of Americans opposed family separation and the policy was widely condemned
by leaders across the political spectrum.12 74% of Americans reported being bothered
by images of children separated from their parents.13 Our reliance on real and on-going
threats addresses the challenge of generating realistic threats in an experimental context
and avoids the risk of ex post rationalization in survey-based studies.
11https://news.gallup.com/poll/236843/americans-say-china-trade-unfair-trade-canada-fair.aspx12https://fivethirtyeight.com/features/separating-families-at-the-border-is-really-unpopular/13https://www.washingtonpost.com/politics/most-americans-oppose-key-elements-of-trump-
immigration-policy/2018/07/05/36124360-7e3d-11e8-b0ef-fffcabeff946 story.html
GROCERY SHOPPING FOR AMERICA 25
Method
Subjects. 352 Amazon MTurk participants based in the US (Mdnage bracket = 30-
49 years old, SDage bracket = .754; 48% female) completed this study in exchange for
monetary compensation.
Procedure. We employed a between-subjects design. In the internal threat condi-
tion, participants read an article about abuse of refugee children separated from their
parents at the US border. In the other condition (external threat), participants read an
article about the US imposition of import tariffs on Chinese goods that accused China
of protectionist practices that harm American businesses.14 Both issues were on-going
threats during the experiment.
After reading the article about the (external/internal) threat, participants were told,
“We will come back on this article later. In the meanwhile, please answer to the following
unrelated questions”, to lower the demand effect. Then, to assess participants’ national
identity, we used the National Attachment Scale proposed by Huddy and Khatib (2007).
The scale is composed of three sub-constructs, namely: National Identity, National Pride,
and Nationalism evaluated on a ten-point scale (for more details, see Study 2 Scales in
the Supplemental Material available online). Finally, participants reported their gender,
age, ethnicity, income, political affiliation, and rural/urban living.
Results
We conducted a series of one-way ANOVAs on both the scale as a whole and on each
of its sub-constructs. The first ANOVA on the National Attachment Scale (α = .943),
controlling for gender, age, ethnicity, income, political affiliation, and rural/urban living,
revealed a significant main effect of threat (F(1, 344) = 9.75, p = .002, ηp2 = .03). As
predicted, participants had a lower (higher) identification with national identity in the
14For more details, see Study 2 manipulations in the Supplemental Material available online
GROCERY SHOPPING FOR AMERICA 26
presence of an internal (external) threat (Minternal = 6.46, SDinternal = 2.10, CI [6.15,
6.71]) vs. Mexternal = 7.02, SDexternal = 1.77; CI [6.78, 7.33]). With respect to the
covariates, age (F(1, 344) = 9.12, p = .003, ηp2 = .03) and income (F(1, 344) = 16.90,
p < .001, ηp2 = .05) had a significant effect in the model. Ethnicity (F(1, 344) = .74, p
= .391, ηp2 < .00), rural/urban living (F(1, 344) = 1.67, p = .197, ηp2 < .00), political
orientation (F(1, 344) = .99, p = .320, ηp2 < .00), and gender F(1, 344) = 2.65, p =
.105, ηp2 = .01) had no significant effect. Without the covariates, the significant effect
of our manipulation still holds.
Additional Calculations. We also ran an ANOVA for each of the sub-constructs
of the scale. For ease of reading, we condensed the results in the tables below.
Table 6.Dependent variable: National Identity (α = .892)
An ANOVA on National Identity, controlling for gender, age, ethnicity, income, political affiliation,
and rural/urban living, revealed a significant main effect of threat (F(1, 344) = 6.99, p = .009,
ηp2 = .02). As predicted, participants had a lower (higher) identification with national identity
in presence of an internal (external) threat (Minternal = 6.88, SDinternal = 2.41, CI [6.52, 7.25])
vs. Mexternal = 7.41, SDexternal = 2.01; CI [7.12, 7.70]). Covariate measures are reported below.
Without including covariates, the significant main effect of our manipulation still holds.
Covariate df F p-value ηp2
gender F(1, 344) 0.04 .843 .00
age F(1, 344) 11.21 .001 .03
ethnicity F(1, 344) 0.14 .704 .00
income F(1, 344) 19.30 .000 .05
rural/urban living F(1, 344) 3.39 .067 .01
political orientation F(1, 344) 0.66 .417 .00
GROCERY SHOPPING FOR AMERICA 27
Table 7.Dependent variable: National Pride (α = .912)
An ANOVA on National Pride, controlling for gender, age, ethnicity, income, political affiliation,
and rural/urban living, revealed a significant main effect of threat (F(1, 344) = 15.05, p < .001,
ηp2 = .42). As predicted, participants had a lower (higher) identification with national identity
in presence of an internal (external) threat (Minternal = 6.02, SDinternal = 2.04, CI [5.71, 6.33])
vs. Mexternal = 6.71, SDexternal = 1.81; CI [6.45, 6.98]). Covariate measures are reported below.
Without including covariates, the significant main effect of our manipulation still holds.
Covariate df F p-value ηp2
gender F(1, 344) 7.41 .007 .02
age F(1, 344) 12.08 .001 .03
ethnicity F(1, 344) 2.16 .143 .01
income F(1, 344) 13.44 .000 .04
rural/urban living F(1, 344) 1.46 .228 .00
political orientation F(1, 344) 0.62 .431 .00
Table 8.Dependent variable: Nationalism (r = .80)
An ANOVA on Nationalism, controlling for gender, age, ethnicity, income, political affiliation,
and rural/urban living, revealed a significant main effect of threat (F(1, 344) = 4.30, p = .039,
ηp2 = .01). As predicted, participants had a lower (higher) identification with national identity
in presence of an internal (external) threat (Minternal = 6.49, SDinternal = 2.46, CI [6.12, 6.86])
vs. Mexternal = 6.95, SDexternal = 2.17; CI [6.63, 7.26]). Covariate measures are reported below.
Without including covariates, our manipulation has a marginally significant effect on Nationalism
F(1, 350) = 3.47, p = .063, ηp2 = .01).
Covariate df F p-value ηp2
gender F(1, 344) 2.61 .107 .01
age F(1, 344) 2.24 .135 .01
ethnicity F(1, 344) 0.32 .569 .00
income F(1, 344) 9.32 .002 .03
rural/urban living F(1, 344) 0.24 .628 .00
political orientation F(1, 344) 1.11 .293 .00
GROCERY SHOPPING FOR AMERICA 28
General Discussion
External and internal threats to social groups produce opposite shifts in national identi-
fication. Study 1 distinguishes between threats external—the death of US soldiers—and
internal threats—torture committed by US soldiers. Study 1 showed war casualties
strengthened Americans’ national identification, corresponding to greater market share
of American-sounding brands. The Abu Ghraib scandal reduced these same brands’
market share, indicative of weaker national identification. Study 2 leveraged real-time
threats to verify that external and internal threats produce opposite shifts in identifica-
tion.
Our work highlights a distinctive type of internal threat that weakens group iden-
tification. Similar threats can emerge for other dimensions of social identity. Consider
Catholic clergy sex abuse. As the magnitude of abuse and complicity becomes clearer,
Catholics must reconcile their need for self-enhancement with their religious identity.
Revelations about systematic sexual harassment and biases pose an analogous internal
threat across a broad spectrum of industries.
We note several possible extensions. More nuanced study of threats can establish
which types of internal threats weaken identification, the tipping point for weaker identi-
fication (Marques et al., 1998), and circumstances when internal criticism can be accepted
(Hornsey, Oppes, & Svensson, 2002). Also, the point at which internal threats weaken
national identification instead of scapegoating (Marques et al., 1998) may depend on
the exact context and the nature of the threat. Although not examined here, this is an
interesting avenue for future work.
The emotional underpinnings of threat responses—shame, guilt, rage—likely produce
distinctive responses and need further investigation. The boundary conditions of self-
enhancement’s effect merit more study. Individual-level moderators may also shape
threat responses. Future research can also explore which factors amplify the moderating
GROCERY SHOPPING FOR AMERICA 29
effect of internal threats across social identities.
Our empirical contribution is the use of real-time, high frequency, fine-grained be-
havioral proxies for psychological phenomena. Brand choice is a shift in identification
observable in an unrelated domain. Existing work on threat and national identity rely
on past events (e.g., Simon, Pantaleo, & Mummendey, 1995), which may differ from
real-time responses and risk ex post rationalization. Unobtrusive measures are more
challenging tests of individual-level psychological processes. Causal inference based on
observational data can better assess change over time such as cumulative effects of threat.
Our research highlights psychological foundations of democratic accountability. In
establishing that threat responses drive real-time mass behavior, we establish a tighter
link between individual responses to social threats and political behavior like protest
and voting. Our internal threat findings suggest that cognitive responses to internal
threats may guard against nationalist excesses. Consumption taps into a wide range
of politically and socially relevant identities. We hope that this work will spur further
exploration of this topic.
GROCERY SHOPPING FOR AMERICA 30
References
Akerlof, G. A., & Kranton, R. E. (2000). Economics and Identity. Quarterly Journal of
Economics , 115 (3), 715-753.
Althaus, S. L., Bramlett, B. H., & Gimpel, J. G. (2012). When War Hits Home: The
Geography of Military Losses and Support for War in Time and Space. Journal of
Conflict Resolution, 56 (3), 382-412.
Aronczyk, M. (2013). Branding the nation: The global business of national identity.
Oxford, England: Oxford University Press.
Ataman, M. B., Van Heerde, H. J., & Mela, C. F. (2010). The Long-Term Effect of
Marketing Strategy on Brand Sales. Journal of Marketing Research, 47 (5), 866-
882.
Balabanis, G., & Diamantopoulos, A. (2011). Gains and Losses from the Misperception of
Brand Origin: The Role of Brand Strength and Country-of-Origin Image. Journal
of International Marketing , 19 (2), 95-116.
Barbera, P., Jost, J. T., Nagler, J., Tucker, J. A., & Bonneau, R. (2015). Tweeting From
Left to Right: Is Online Political Communication More Than an Echo Chamber?
Psychological Science, 26 (10), 1531-1542.
Branscombe, N. R., Ellemers, N., Spears, R., & Doosje, B. (1999). The Context and
Content of Social Identity Threat. In Social identity: Context, commitment, content
(p. 35-58). Oxford, England: Blackwell Publishers Inc.
Brewer, J. D. (1984). Looking Back at Fascism: A Phenomenological Analysis of BUF
Membership. The Sociological Review , 32 (4), 742-760.
Bronnenberg, B. J., Kruger, M. W., & Mela, C. F. (2008). Database Paper: The IRI
Marketing Data Set. Marketing Science, 27 (4), 745-748.
Davies, P. G., Steele, C. M., & Markus, H. R. (2008). A nation challenged: The impact
of foreign threat on America’s tolerance for diversity. Journal of Personality and
GROCERY SHOPPING FOR AMERICA 31
Social Psychology , 95 (2), 308-318.
Ellemers, N., & Haaker, M. (1995). Feedback and self-esteem: Individual and group level
manipulations and responses. University of Amsterdam.
Elsbach, K. D., & Bhattacharya, C. B. (2001). Defining Who You Are By What You’re
Not: Organizational Disidentification and The National Rifle Association. Orga-
nization Science, 12 (4), 393-413.
Escalas, J. E., & Bettman, J. R. (2005). Self-Construal, Reference Groups, and Brand
Meaning. Journal of Consumer Research, 32 (3), 378-389.
Gartner, S. S. (2008). The Multiple Effects of Casualties on Public Support for War:
An Experimental Approach. American Political Science Review , 102 (1), 95-106.
Gartner, S. S. (2009). Evaluating Claims of Social Connection to International Conflict
Casualties. International Interactions , 35 (3), 352-364.
Greenberg, K. J. (Ed.). (2005). The Torture Debate in America. Cambridge University
Press.
Hogg, M. A. (2006). Social Identity Theory. In P. J. Burke (Ed.), Contemporary Social
Psychological Theories (p. 111-136). Stanford, CA: Stanford University Press.
Hornsey, M. J., Oppes, T., & Svensson, A. (2002). “It’s OK If We Say It, but You
Can’t”: Responses to intergroup and intragroup criticism. European Journal of
Social Psychology , 32 (3), 293-307.
Huddy, L., & Khatib, N. (2007). American Patriotism, National Identity, and Political
Involvement. American Journal of Political Science, 51 (1), 63-77.
Kahn, B. E., & Schmittlein, D. C. (1989). Shopping trip behavior: An empirical inves-
tigation. Marketing Letters , 1 (1), 55-69.
Khan, R., Misra, K., & Singh, V. (2013). Ideology and Brand Consumption. Psycholog-
ical Science, 24 (3), 326-333.
Kriner, D. L., & Shen, F. X. (2010). The Casualty Gap: The Causes and Consequences
of American Wartime Inequalities. Oxford, England: Oxford University Press.
GROCERY SHOPPING FOR AMERICA 32
Kriner, D. L., & Shen, F. X. (2012). How Citizens Respond to Combat Casualties: The
Differential Impact of Local Casualties on Support for the War in Afghanistan.
Public Opinion Quarterly , 76 (4), 761-770.
Liu, S. S., & Johnson, K. F. (2005). The Automatic Country-of-Origin Effects on Brand
Judgments. Journal of Advertising , 34 (1), 87-97.
Mari, S., Andrighetto, L., Gabbiadini, A., Durante, F., & Volpato, C. (2010). The
Shadow of the Italian Colonial Experience: The Impact of Collective Emotions on
Intentions to Help the Victims’ Descendants. International Journal of Conflict and
Violence, 4 (1), 17.
Marques, J. M., Paez, D., & Abrams, D. (1998). Social identity and intragroup differ-
entiation as subjective social control. In S. Worchel, J. F. Morales, D. Paez, &
J.-C. Deschamps (Eds.), Social identity: International perspectives (p. 124-141).
London: Sage Publications.
Martin, B. A., Lee, M. S. W., & Lacey, C. (2011). Countering negative country of origin
effects using imagery processing. Journal of Consumer Behaviour , 10 (2), 80-92.
Pew Research Center. (2004, May). 76% Have Seen Prison Pictures; Bush Ap-
proval Slips (Tech. Rep.). Washington, DC: Pew Research Center. Retrieved
from http://assets.pewresearch.org/wp-content/uploads/sites/5/legacy
-pdf/213.pdf
Samiee, S., Shimp, T. A., & Sharma, S. (2005). Brand origin recognition accuracy: Its
antecedents and consumers’ cognitive limitations. Journal of International Busi-
ness Studies , 36 (4), 379-397.
Sedikides, C. (1993). Assessment, enhancement, and verification determinants of the
self-evaluation process. Journal of Personality and Social Psychology , 65 (2), 317-
338.
Sedikides, C., & Strube, M. J. (1997). Self-evaluation: To thine own self be good, to
thine own self be sure, to thine own self be true, and to thine own self be better.
GROCERY SHOPPING FOR AMERICA 33
In Advances in Experimental Social Psychology (Vol. 29, p. 209-269).
Shachar, R., Erdem, T., Cutright, K. M., & Fitzsimons, G. J. (2011). Brands: The
Opiate of the Nonreligious Masses? Marketing Science, 30 (1), 92-110.
Simon, B., Pantaleo, G., & Mummendey, A. (1995). Unique Individual or Interchange-
able Group Member? The Accentuation of Intragroup Differences Versus Similar-
ities as an Indicator of the Individual Self Versus the Collective Self. Journal of
Personality and Social Psychology , 69 (1), 106-119.
Steenkamp, J.-B. E., Batra, R., & Alden, D. L. (2003). How perceived brand globalness
creates brand value. Journal of International Business Studies , 34 (1), 53-65.
Stephan, W. G., Ybarra, O., & Morrison, K. R. (2009). Intergroup Threat Theory.
In T. D. Nelson (Ed.), Handbook of Prejudice, Stereotyping, and Discrimination
(p. 43-59). New York, NY: Psychology Press, Taylor & Francis Group.
Tajfel, H., & Turner, J. C. (1979). An integrative theory of intergroup conflict. In
W. G. Austin & S. Worchel (Eds.), The Social Psychology of Intergroup Relations
(p. 33-47). Monterey, CA: Brooks/Cole Publishing Company.
Usunier, J.-C., & Shaner, J. (2002). Using linguistics for creating better international
brand names. Journal of Marketing Communications , 8 (4), 211-228.
Webb, E. J., Campbell, D. T., Schwartz, R. D., & Sechrest, L. (1966). Unobtrusive
measures: Nonreactive research in the social sciences. Oxford, England: Rand
Mcnally.
Zhong, C.-B., Liljenquist, K., & Cain, D. M. (2009). Moral self-regulation: Licensing
and compensation. In D. De Cremer (Ed.), Psychological Perspectives on Ethi-
cal Behavior and Decision Making (p. 75-89). Charlotte, NC: Information Age
Publishing.