mobile ad effectiveness: hyper-contextual targeting with ... · mobile ad effectiveness:...

Articles in Advance, pp. 1–17ISSN 0732-2399 (print) � ISSN 1526-548X (online) http://dx.doi.org/10.1287/mksc.2015.0905

© 2015 INFORMS

Mobile Ad Effectiveness:Hyper-Contextual Targeting with Crowdedness

Michelle AndrewsGoizueta Business School, Emory University, Atlanta, Georgia 30322; and Fox School of Business, Temple University,

Philadelphia, Pennsylvania 19122, [email protected]

Xueming LuoFox School of Business, Temple University, Philadelphia, Pennsylvania 19122, [email protected]

Zheng FangSichuan University, Sichuan 610064 China, [email protected]

Anindya GhoseStern School of Business, New York University, New York, New York 10012, [email protected]

This research examines the effects of hyper-contextual targeting with physical crowdedness on consumerresponses to mobile ads. It relies on rich field data from one of the world’s largest telecom providers who

can gauge physical crowdedness in real-time in terms of the number of active mobile users in subway trains.The telecom provider randomly sent targeted mobile ads to individual users, measured purchase rates, andsurveyed purchasers and nonpurchasers. Based on a sample of 14,972 mobile phone users, the results suggestthat, counterintuitively, commuters in crowded subway trains are about twice as likely to respond to a mobileoffer by making a purchase vis-à-vis those in noncrowded trains. On average, the purchase rates measured2.1% with fewer than two people per square meter, and increased to 4.3% with five people per square meter,after controlling for peak and off-peak times, weekdays and weekends, mobile use behaviors, and randomlysending mobile ads to users. The effects are robust to exploiting sudden variations in crowdedness induced byunanticipated train delays underground and street closures aboveground. Follow-up surveys provide insightsinto the causal mechanism driving this result. A plausible explanation is mobile immersion: As increased crowd-ing invades one’s physical space, people adaptively turn inwards and become more susceptible to mobile ads.Because crowding is often associated with negative emotions such as anxiety and risk-avoidance, the findingsreveal an intriguing, positive aspect of crowding: Mobile ads can be a welcome relief in a crowded subwayenvironment. The findings have economic significance because people living in cities commute 48 minutes eachway on average, and global mobile ad spending is projected to exceed $100 billion. Marketers may consider thecrowdedness of a consumer’s environment as a new way to boost the effectiveness of hyper-contextual mobileadvertising.

Data, as supplemental material, are available at http://dx.doi.org/10.1287/mksc.2015.0905.

Keywords : mobile advertising; hyper-contextual targeting; crowdedness; field study; new technologyHistory : Received November 25, 2013; accepted December 11, 2014; Preyas Desai served as the editor-in-chief

and K. Sudhir served as associate editor for this article. Published online in Articles in Advance.

1. IntroductionAs mobile advertising rapidly evolves, marketers areusing contexts such as location to target consumers.Examples of such contextual targeting include geo-fencing (i.e., sending mobile coupons to people withinthe virtual perimeter of a store) and Bluetooth-basedbeacons (i.e., sending deals to devices within venues)(Luo et al. 2014). Interestingly, hyper-contextual tar-geting may even allow marketers to engage withother environmental contexts such as weather todynamically influence behavior (Marshall 2014).

This research investigates how another hyper-context, physical crowdedness, may affect consumerresponse to mobile ads. We define physical crowded-ness as the degree of population density or number

of people per unit area (Stokols 1972). The uniqueenvironment of subway trains is used to test theeffects of physical crowdedness because of the nat-ural setting of trains with different levels of crowddensity. In the city of this study, the undergroundsubway is equipped with mobile signal receivers andallows passengers to use their mobiles throughouttheir commute. Cellular technology can record in real-time the number of users in each mobile-equippedsubway train. Traditionally, crowding has been diffi-cult to measure in conventional retail environments.Today, mobile technologies allow researchers to moreprecisely determine people’s spatial proximity to oneanother and to gauge crowd density as the number ofpassengers per square meter.

1

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

mailto:[email protected]




http://dx.doi.org/10.1287/mksc.2015.0905

Andrews et al.: Hyper-Contextual Targeting with Crowdedness2 Marketing Science, Articles in Advance, pp. 1–17, © 2015 INFORMS

A large mobile telecom company provided fielddata for our study. The experiment design and theunique attributes of this data can causally identifythe effects of crowding. The company sent targetedmobile ads via short message services (SMS) to con-sumers in subway trains who could purchase themobile ads by responding to the SMS. The companyrandomly sampled commuters from the targeted sub-way population using an instant computational andrandomization procedure. Because crowding is natu-rally confounded with the time of day and the day ofthe week, the field data can isolate the effect of crowd-edness on mobile purchases by controlling for self-selection and user heterogeneity across (i) peak versusnon-peak hours, and (ii) weekdays versus weekends.To further address endogenous selection concerns andrule out alternative explanations, our identificationstrategy exploits sudden variations in crowdednessthat are induced by unanticipated train delays under-ground and street closures aboveground. Based on14,972 mobile phone users, the results show that com-muters in crowded subway trains are about twiceas likely to respond by making a purchase vis-à-visthose in noncrowded trains. On average, the purchaserates measured 2.1% with fewer than two people persquare meter, and increased to 4.3% with five peopleper square meter. However, the effect of crowdednesson response to mobile ads is neither simple nor lin-ear, but rather nonlinear and positive after a lowerthreshold.

Why would people welcome targeted ads on theirmobile phones in crowded trains, more so than innoncrowded trains? Behavioral constraint theory pro-vides a possible answer (Sommer 2009, Stokols 1972).Specifically, crowding in subway trains can invadepeople’s physical space and restrict their behaviors,giving rise to adaptive strategies. Commuters adap-tively turn inwards, i.e., focus less on the physicalcrowding but more on their personal mobile phones.This mobile immersion may thus lead to greater sus-ceptibility to mobile ads in crowded trains.

This research offers several contributions. Our find-ing of the positive effect of crowding is counterintui-tive given prior studies that largely note negativeeffects. Specifically, in crowded retail stores, peopletend to become more anxious, reduce their shoppingtime, and adopt risk-avoidance behaviors (Harrellet al. 1980, Maeng et al. 2013). Yet, in the con-text of subway trains, we find that crowding mayactually increase purchase likelihood. In this sense,we contribute to the literature on crowding andretail environments by revealing a positive aspect ofcrowding: Mobile messages can be a welcome reliefin a crowded subway environment. Theoretically, thisfinding is intriguing because crowding may reduceoutside options and lead people to adaptively focus

their attention inwards. This new insight into howpeople may behave in crowded environments sug-gests that crowding makes audiences for ads moreattentive and susceptible.

We also extend the mobile ad literature by advanc-ing a new way to think about mobile targeting.Crowding makes people turn inwards which, intoday’s age, is manifested by immersion in theirmobile phones. Spending more time browsing on thephone when idle in subway trains increases the prob-ability of increased susceptibility to mobile ads. Thissusceptibility is crucial to the literature since extantwork notes that there can be “noise” and distrac-tions when people construe mobile ads, regardlessof whether such ads are displayed on mobile opti-mized sites (Bart et al. 2014) or are inserted withinapps (Ghose and Han 2014). Paradoxically, we findthat when surrounded by others, people may beless influenced by outside noise but rather inwardlymore attentive to their private mobile devices andresponsive to mobile ad signals. Also, the effect ofcrowding may not be simply about the pure pres-ence of others, because we do not find a linear effect.Rather, the effect is nonlinear with a lower thresh-old, confirming that crowding may pose physical con-straints that make people turn inward and engagein higher mobile immersion. That is, the effect ofcrowding kicks in after a certain threshold of crowd-ing, whereby people can no longer move around topreserve personal space. Prior studies on consumers’use of mobile phones and responsiveness to mobileads have examined contextual factors such as users’geographic mobility (Ghose and Han 2011), location(Molitor et al. 2014), location and time at whichthey receive a promotion (Luo et al. 2014), weather(Molitor et al. 2013), and product characteristics (Bartet al. 2014). We extend these studies by advancing anovel nonlinear view of mobile ad response in thecontextual environment of subway crowding.

Our findings also have meaningful managerialimplications. Mobile marketers are excited about har-nessing a better understanding of the context thatdrives consumer response to mobile ads. With mobilead spending projected to near $100 billion by 2018(eMarketer 2014), marketers have a vested interestin reaching consumers when and where they aremost receptive to mobile ads. Our study suggests thatfor managers, crowding represents a new setting forgauging the effectiveness of context-sensitive mobilemessages. Public transit commutes provide a uniquemarketing environment and thus a window of oppor-tunity for marketers to target mobile users. In mostcities people spend a considerable amount of timecommuting, averaging 48 minutes each way, accord-ing to Census Bureau reports (McKenzie and Rapino2011). This commuting time in the subway can be aboon for marketers to deliver ads to mobile users at

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Andrews et al.: Hyper-Contextual Targeting with CrowdednessMarketing Science, Articles in Advance, pp. 1–17, © 2015 INFORMS 3

the right time and location. As more cities facilitatemobile use in underground subways (Flegenheimer2013), marketing opportunities to target consumersbased on ambient factors such as crowding willabound.

2. Background and Hypothesis2.1. Literature on Mobile Ad ResponseOur work builds on two streams of research. First, themobile ad literature reveals several contexts that influ-ence consumers’ likelihood of responding to mobilepromotions. In terms of location, Ghose et al. (2013a)find that consumers are more likely to click on linksto stores close to them and on higher-ranked linkson their mobile screens. Echoing this, Molitor et al.(2014) found that mobile coupon redemption ratesincreased the closer consumers were to a store and thehigher the offer was displayed on the screen, condi-tional on the discount. Luo et al. (2014) show how thelocation of consumers relative to a promoted venueand the time at which they receive a promotion affectstheir mobile purchase likelihood. Researchers havealso demonstrated how real-time mobile promotionstransmitted within grocery stores can increase shop-pers’ travel distance and, consequently, boost theirunplanned spending (Hui et al. 2013). In addition,scholars have found that for mobile display adver-tising, products higher on involvement and utilitar-ian dimensions generate more purchase intentions(Bart et al. 2014). Studies have also found that theinterdependence between PC and mobile use cre-ates positive synergistic effects that impact consumerresponsiveness to mobile advertising (Ghose et al.2013b). Market competition and environmental fac-tors may also explain why consumers make mobilepurchases (Fong et al. 2015, Molitor et al. 2013).

2.2. Literature on CrowdednessOur study is also related to the literature on crowd-edness. In the sociology and psychology literature,studies have linked crowding to physical and men-tal diseases and juvenile delinquency (Schmitt 1966).Crowding can increase stress (Collette and Webb1976), frustration (Sherrod 1974), and hostility (Griffittand Veitch 1971). Individuals may consider them-selves more anonymous in crowds, which can reducetheir social interactions and fuel antisocial behavior(Zimbardo 1969). In more crowded areas, for exam-ple, crime rates were found to spike due to the higherlikelihood that criminals could avoid detection (Jarrelland Howsen 1990). Scholars have demonstrated thatin crowded settings, people perceive less control overtheir situations (Aiello et al. 1977, Hui and Bateson1991). In consumer behavior research, crowding hasbeen found to induce avoidance behaviors and reduce

shopping times (Harrell et al. 1980). Other studieshave shown that crowding can threaten consumers’sense of uniqueness, which may lead them to pur-chase more distinctive products to restore their per-ceived individuality (Xu et al. 2012). Researchersalso found that crowding can boost purchase varietydue to consumers’ desire to assert freedom throughchoice (Levav and Zhu 2009). Relatedly, Maeng andcolleagues (2013) note that crowded environmentscan spur shoppers to become more risk-averse andfavor safety-oriented options such as visiting a phar-macy rather than a convenience store. Following theseresearch streams, we suggest that the effectiveness ofmobile targeting may also depend on another context,i.e., how crowded a consumer’s environment is.

2.3. The Effect of Crowdedness onMobile Ad Response

Mobile ads may elicit higher response rates whenpeople are in crowded subway trains vis-à-vis non-crowded trains. Crowding invades personal spaceand restricts behavior; people can adapt by turninginwards and becoming more responsive to mobileads.1 More specifically, crowding arises from thephysical proximity of others in confined subwaytrains, where “riders are often squeezed into veryclose proximity” (Milgram and Sabini 1978, p. 32).As more commuters occupy the train, individuals’personal and physical spaces are invaded. This spa-tial limitation poses a behavioral constraint that canlead people to experience a reduction in the out-side options of things to do. For example, peoplewill be less prone to look around because crowdingcan boost the likelihood of catching unwanted gazes(Aiello et al. 1977, Evans and Wener 2007). Behav-ioral constraint theory (Stokols 1972) suggests thatpeople use adaption strategies in a crowd to alle-viate perceived behavioral restrictions: People mayadapt to crowded situations by turning inwards tofilter out inputs from social and physical surround-ings (Milgram 1970). Similarly, scholars have docu-mented that when consumers perceive a threat totheir freedom due to spatial confinement, they mayreact by engaging in actions designed to reassert theirfreedom (Brehm 1966, Wicklund 1974). For exam-ple, consumers may use their purchasing decisions toexert control over their environment (Levav and Zhu2009). In today’s age, people can adapt by turninginwards and paying more attention to their mobiledevices. That is, people can immerse themselves intheir private mobiles “as a means of escape, a way bywhich a person can avoid unwanted encounters” andgain a sense of control over her space and privacy(Sommer 2009, p. 1227). Crowding can lead people

1 We acknowledge an anonymous reviewer for this point.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


to adaptively respond by escaping into the “tech-nologically mediated private realm” of their mobiles(Bull 2005, p. 354). This mobile immersion allows pas-sengers to psychologically withdraw from the phys-ical crowd and can boost their attention to mobileads (Maeng and Tanner 2013), thus likely leading tohigher purchase responses in more crowded trains.

However, this effect of crowding on purchase like-lihood may be neither simple nor linear, but rathernonlinear with a lower threshold. In the closed envi-ronment of subway trains, individuals may initiallyadapt to crowdedness by moving around to preservepersonal space. Once this becomes less feasible asthe subway train fills up, they may resort to turninginwards via mobile immersion after a certain thresh-old level of crowding.2 If so, we test the following:

Hypothesis. Ceteris paribus, physical crowdedness en-genders higher consumer response to mobile ads. Yet theeffect of crowdedness is nonlinear and positive after a lowerthreshold.

3. Field Data, Identification,and Results

An ideal way to test the effects of crowdednesswould be to conduct a randomized field experimentin which the level of crowdedness is varied. Yet this isextremely difficult because it would require randomlyassigning commuters to more versus less crowdedsubway trains. In other words, it is quite hard tomanipulate the level of crowdedness in a field setting.Thus, our study relies on field data in which users arerandomly targeted with mobile coupons as a functionof the level of crowdedness.

In our field data, a total of 10,690 mobile userswho rode the subway were randomly selected tobe sent a promotion through a text message, andtheir responses were recorded. The SMS advertised amissed call alert service, which, if purchased, wouldnotify mobile users of the time and number of thecalls they missed. The study took place in Asia inSeptember 2013, where cell phone plans are piecemealrather than all-inclusive. As such, mobile users must

2 In theory, an upper-boundary may exist: Too dense a crowd wouldnegatively affect mobile ad response because congestion wouldrestrict the ability to even use a mobile phone. For example, inTokyo, subway commutes can be packed to capacity with up to 11passengers/m2, and in Hong Kong, transit authorities are consid-ering removing seats from subway trains to give commuters “moreroom to interact with their devices” (Agence France-Presse 2014).In the city of our study, however, the highest level of crowding weobserve is 5 people/m2, so we cannot test for an upper threshold.Also, as our research aims to make a causal inference of crowd-edness on mobile ad sales, we recognize that consumer responseto mobile ads while in a crowded environment may be multiplydetermined.

purchase a missed call alert service separately to benotified of missed calls. The SMS read “Missed a calland want to know from whom? Subscribe to [Wireless Ser-vice Provider’s] missed call alert service and be notifiedby SMS of the calls you missed! Only ¥9 for 3 months!Get ¥3 off if you reply ‘Y’ to this SMS within the next20 minutes!” Because the average subway commute is30 minutes, the reply time was restricted to 20 min-utes to ensure that most commuters would respondto the SMS while still in transit. The provider offereda ¥3 rebate to incentivize mobile users to respond. Forthose who responded, the cost was charged to theirphone bills. Thus, we measure response to mobileads as the purchase rate (1 = purchase, 0 = no pur-chase), also labeled as mobile ad effectiveness, i.e.,our dependent variable. Of the 10,690 mobile userswho received an SMS, 334 replied and purchased thepromoted service, corresponding to a 3.22% responserate in our data. These response rates seem low, butare fairly high compared with the 0.6% response ratefor mobile coupons in Asia (eMarketer 2012) and the1.65% response rate for the effectiveness of location-based mobile coupons (Molitor et al. 2014).

Crowdedness, our key independent variable, ismeasured as the number of subway passengers persquare meter. We did this in a step-by-step fashion.First, in our study, the subway consisted of articulatedtrains modeled after accordion-style buses where theabsence of doors between cars enables passengers totravel the length of the subway train without restric-tion. Each subway train is composed of six articulatedcars. We measured the volume of mobile users in eachsubway train as the number of mobile users who areautomatically connected to the subway-tunnel cellularlines.3 Second, we calculated passenger volume. Themobile phone penetration rate in major Asian citiesis very high. In the large city we studied, the tele-com provider serves 70% of the population. We thusderived passenger volume by dividing the mobileuser volume by 0.7. Third, we calculate crowdednessby dividing this passenger volume by the total areaof each subway train. Each subway car is 19 meterslong and 2.6 meters wide, totaling 296.4 m2 (6 cars ×

19 m length × 206 m width).4 We measure crowded-ness at the time the SMS mobile ad is sent. Table 1

3 The telecom provider can identify all mobile users’ phone num-bers, including those of phones that are off (due to telecommuni-cation company chip-tracking technology). Thus, both phones thatare on and those that are off are recognized and recorded. Accord-ing to the provider, over 99.9% of phones are on during the day.4 Because crowdedness differs from city to city and culture to cul-ture (e.g., cities with a high population density such as HongKong are different from those with a low population density suchas Auckland), tolerance for crowdedness may also vary. BecauseAsians are more tolerant of crowdedness given their large popula-tion, the effects in our field study may be more conservative.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


Table 1 Level of Crowdedness in Subway Trains

Weekday sample Weekend sample

Time of day Train Crowdedness Crowdedness

7:30–8:30 1 4.18 1.802 3.93 1.87

10:00–12:00 3 2.99 3.074 2.93 3.185 2.91 3.20

14:00–16:00 6 2.01 3.097 1.91 3.42

17:30–18:30 8 5.36 4.679 4.57 4.54

21:00–22:00 10 0.96 1.5111 0.94 1.6212 0.91 1.6413 0.89 1.6514 0.83 1.72

Note. Crowdedness is the number of passengers per square meter.

presents the measured crowdedness per time cycleper train. As shown in Table 1 for the 14:00–16:00time cycle, in train 7 the crowdedness (shown in bold)is 1.91 passengers/m2 on a weekday but is muchhigher at 3.42 on a weekend, as expected. As depictedin Figure 1, mobile purchase likelihood increases asa function of crowdedness. This pattern gives initialmodel-free evidence for the effect of crowdedness onconsumer response to mobile ads. Next, we presentmodel-based evidence.

3.1. Identification and Model-Based ResultsBecause crowding is naturally confounded with thetime of day and day of week, the validity of resultsbased on field data is often threatened by alter-native explanations. For greater confidence in theresults, our identification strategies are multifacetedto: (1) address selection on observable variables thatinclude peak versus nonpeak crowdedness hours,weekdays versus weekends, mobile use behavior,and randomly sending mobile ads, and (2) exploit

Figure 1 (Color online) Crowdedness and Purchase Rates

4.1

3.9

3.7

3.5

3.3

Purc

hase

rat

e (%

)

3.1

2.9

2.7

2.50.91 1.96

Crowdedness as passengers/m2

3.05 4.02 4.97

sudden variations in crowdedness induced by streetclosures aboveground and unanticipated train delaysbelowground in the subway system.

3.1.1. Identification with Peak-Hour, Weekend,and Mobile Usage Behavior. Self-selection concernsarise when people self-select into more or lesscrowded trains depending on factors such as the timeof day and whether it is a weekday or a weekend. Forinstance, during morning rush hours, there is likely tobe more crowding with a greater proportion of work-ing professionals, whereas at other times, seniors,children, and homemakers may be more likely touse the subway. If this is the case, the observedeffects may reflect systematic differences in behaviorbetween people who travel during rush hours ver-sus other times, rather than the effect of crowdednessper se. That is, purchases may be driven by variationsin commuter type rather than crowdedness. Thus, toreduce the concern of self-selection we address thisthreat in several ways.

First, crowdedness is more likely to occur duringpeak hours of travel. To isolate the effects of crowd-edness from peak hours, we have field data acrosspeak and nonpeak hours of crowdedness. This isbecause the SMS was sent from morning to evening.For each day of our study, five different times wereselected to represent the five cycles people may expe-rience during an average day. Each time cycle rep-resents a peak or nonpeak hour of crowdedness andthereby a potentially different level of crowdedness.The first time cycle was from 7:30–8:30 and representsthe morning rush hour. The second was from 10:00–12:00, representing the lull after the morning rushhour traffic. The third was from 14:00–16:00, the after-noon lull before the evening rush hour traffic. Thefourth was from 17:30–18:30, the evening rush hour.The fifth was from 21:00–22:00, the after-dinner traf-fic. Within each time cycle, the provider sent SMS totwo or three trains, for a total of 14 trains a day. Thesefive time cycles nested in the 14 subway trains helpcontrol for the systematic differences between varioustypes of subway commuters, i.e., business commutersat rush hours versus nonbusiness commuters at non-rush hours.

In addition, crowding may differ across weekdaysand weekends. On weekdays, work schedules mightforce business travelers to self-select into crowdedtrains during rush hours, while nonbusiness travel-ers (e.g., retirees and homemakers) might self-selectinto noncrowded trains during lull hours. Also, onweekends, leisure schedules allow people to self-select into more or less crowded trains. Thus, to iso-late the effects of crowdedness from different types ofcommuters across weekdays and weekends, we havefield data from two different days, i.e., a weekday(Wednesday) and a weekend (Saturday). This helps

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


control for the systematic differences between week-day and weekend crowdedness in our model.

Our model estimates the purchase likelihood ofeach mobile user as PurchaseLikelihoodi. We model theunobserved likelihood of making a mobile purchaseas a logit function of crowdedness in subway trains.Following Agarwal et al. (2011, p. 1063) and Guadagniand Little (1983), we assume an i.i.d. extreme valuedistribution of the error term in the logit model:

PurchaseLikelihoodi =exp4Ui5

1 + exp4Ui5

Ui = �+�× crowdednessi +� × weekendi

+ �× peak houri + � ×Xi + �i11

(1)

where Ui denotes the utility of a mobile purchase,and Xi is a vector of control variables with mobileuse behaviors. Because weekends and weekdays areexpected to have different crowdedness patterns,we control for these effects with a weekend dummy(1 = weekend, 0 = weekday). We further identify theeffects of crowdedness by controlling for time of dayeffects. Because peak hours and nonpeak hours areexpected to have different crowdedness patterns, weinclude a peak hour dummy (1 = peak hours during7:30–8:30 and 17:30–18:30, 0 = otherwise). Also, weconducted additional analyses with more time dum-mies of time-specific effects, beyond the one weekenddummy and peak hour dummy, and found robusteffects of crowding on purchase rates of the SMSmobile ads. Also, �i is comprised of the idiosyncraticerror terms. Of key interest here is the effect (�)of crowdedness on mobile purchase likelihood, aftercontrolling for mobile user behaviors, peak hours,and day effects. Our model controls for mobile usehabits based on mobile users’ individual monthly bills(average revenue per user (ARPU)), call minutes used(monthly minutes of usage (MOU)), SMS sent andreceived, and data use (general packet radio service(GPRS)).5 Appendix A presents a histogram of ARPU.

5 Government regulations prohibit the wireless provider fromrevealing customers’ private information such as income. However,income is not relevant in our context because the promoted prod-uct costs about 50 cents per month. ARPU is the revenue that onecustomer’s cellular device generated. Users with a higher ARPUmight be more likely to purchase the missed call alert promotionbecause they might connect with more people. Individual MOUis how much voice time a user spent on her mobile. MOU canhelp control for customer heterogeneity, as business travelers mayhave a higher MOU. Also, SMS is the amount of monthly text mes-sages sent and received, and can also help tease out the effects ofconsumer age as younger generations are more prone to SMS use.GPRS is a measure of the individual monthly volume of data usedwith the wireless service provider. GPRS is useful in controlling fortraveler habits in using the mobile Web and downloading mobilecontent. In additional analyses, we also tried to use factor analy-sis with Principal Component Analysis via VARIMAX rotation and

Also, although we do not have random treatmentof crowdedness, the wireless company randomlyselected subjects from the targeted subway popula-tion to randomize away user heterogeneity. The tar-geted population had not previously subscribed to amissed call alert plan, nor received a similar SMS fromthe wireless company. This also helps rule out poten-tial carryover effects of prior marketing campaigns.For example, if passengers have already received asimilar mobile promotion, promotion repetition orfatigue may drive the results. For each user fromthis targeted subway population, a random numberwas assigned via SAS software’s random number gen-erator and RANUNI function, which returns a ran-dom value from a uniform distribution (Deng andGraz 2002). The random numbers are then sorted insequence from which a sample is extracted. Integratedin the wireless provider’s IT system, this algorithmcan compute and randomize users instantly to accu-rately gauge the level of crowdedness while sendingthe SMS in real time. Because the wireless companymaintains the purchase records of all of its clients,it knows immediately whether a given mobile userin the subway train had previously subscribed to thepromoted service. Such dynamic but instant compu-tation and randomization are difficult to execute if notdone by the wireless company, thereby representing aunique feature of our field data.6

Table 2, Panel A presents the logit model results.Various models test the effects of crowding withand without the peak hour and weekend covariates,in addition to the mobile use behaviors. As shownin Table 2, the effects of crowdedness on consumerresponses to mobile ads (i.e., purchase likelihood) arestatistically significant (all p < 0005), after controllingfor the effects of observable user heterogeneity dueto peak versus nonpeak hours, weekday or week-end variables, and mobile use behaviors. These effectsare consistently significant across the models withor without the weekday versus weekend effects andpeak versus nonpeak hour effects.

obtained one underlying common factor of ARPU, MOU, SMS, andGPRS. The results are robust to this factor analysis approach tomobile use behavior.6 The field data further control for several other factors. First, werule out station effects by using only one station in the subwaysystem. Compared with stations in suburban areas, those in down-town areas are likely to be more crowded. Also, some subway sta-tions have food stands and newspaper stands, whereas others donot. To control for these differences, the SMS was pushed fromonly the fourth stop along the subway line. The fourth station wasselected to ensure that enough passengers had boarded the trainsince the first station, but that they also still had a sufficient distanceto travel to stations further along the line. Also, we rule out theeffect of the subway direction in which passengers travel becausethe wireless company sent the SMS to trains travelling in a singledirection towards the city center.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


Table 2 Effect of Crowdedness on Consumer Response to Mobile Ads

Panel A: Effect of crowdedness

Crowdedness 00239∗∗ 00213∗∗ 00178∗∗ 00114∗∗

4000675 4000755 4000715 4000425Day effects (weekend dummy) No No Yes YesTime of day effects No Yes No Yes

(peak hour dummy)

Ln(ARPU) 00415∗∗ 00411∗∗ 00303∗∗ 00305∗∗

4001815 4001735 4001245 4001285Ln(MOU) −00045 −00045 −00043 −00044

4000725 4000665 4000705 4000695Ln(SMS) 00004 00004 00003 00006

4000765 4000755 4000735 4000735Ln(GPRS) −00002 −00002 −00001 −00001

4000285 4000275 4000255 4000245

Observations 10,690 10,690 10,690 10,690

Panel B: Lower threshold with under 2 passengers/m2

Crowdedness −000844002705

Day effects (weekend dummy) Yes YesTime of day effects (peak hour dummy) Yes YesLn(ARPU) 00354 00352

4002675 4002655Ln(MOU) −00235 −00231

4001425 4001425Ln(SMS) 00083 00074

4001465 4001485Ln(GPRS) −00028 −00026

4000525 4000535

Observations 2,886 2,886

Notes. ARPU, Average revenue per user; MOU, minutes of use; SMS, num-ber of texts sent and received per user; GPRS, data use with the wirelessprovider. Note also that Panel B restricts the observations to under twopassengers/m2.

∗p < 0010; ∗∗p < 0005.

Furthermore, we test the threshold level for theeffects of crowding. Because fewer than two passen-gers per square meter is not truly representative ofcrowded environments, we should not see effects ofcrowdedness on mobile purchase likelihood for sucha low threshold of crowdedness. We report our resultsin Table 2, Panel B. As expected, we find no signifi-cant effect of crowdedness on mobile purchase whencrowdedness is less than two passengers per squaremeter. This suggests a lower boundary threshold ofcrowdedness effects. Put differently, for crowdednessto have an effect, there must be at least two or morepassengers per square meter.

Therefore, these results support our overall hypoth-esis: Physical crowdedness engenders higher con-sumer response to mobile ads. Still the effect ofcrowdedness is nonlinear and positive after a lowerthreshold.

We also assess the economic significance of crowd-edness on mobile ad purchases. As shown in Figure 1,

compared with the baseline of a lower level of crowd-edness (1.96 passengers/m2), when crowdedness dou-bles (4.02 passengers/m2), the estimated marginalmean likelihood of mobile purchases increases by16.0% (=4000319 − 0002755/000275). In addition, whencrowdedness spikes even more (4.97 passengers/m2),the likelihood of mobile purchases jumps 46.9%(=4000404 − 000275)/0.0275). On average, purchaserates measured 2.1% with fewer than two people persquare meter, and increased to 4.3% with five peo-ple per square meter. As such, commuters in crowdedsubway trains may welcome targeted ads on theirmobile phones and are about twice as likely to respondby making a purchase vis-à-vis those in noncrowdedtrains.

Note that in our setting, subways were not packedto capacity (such as some commutes in Tokyo, withup to 11 passengers/m25. In that case, it is logical tospeculate an upper boundary effect of crowdedness:Too dense a crowd would negatively affect mobilepurchases because congestion would restrict the abil-ity to use a mobile phone. However, we checked thesquared and cubed terms of crowdedness, but failedto find either of them significant (p > 0010).

Even after accounting for observables, some unob-servables may drive both crowdedness and pur-chases. That is, because commuters may self-selectinto different trains, the crowd may be endogenouslyinduced, and some unobserved variable may driveboth crowdedness and purchases. In other words, thecrowd may be endogenous and, if so, may confoundthe results.7 We address this potential endogeneitythreat by exploiting sudden variations in crowded-ness induced not only by unanticipated street closuresaboveground but also by train delays belowground.

3.1.2. Identification to Address Endogeneity viaUnanticipated Street Closures. We exploit suddenvariations in crowdedness due to a traffic halt above-ground. Via street closures, the local government tem-porarily halted vehicular traffic for an hour on aweekday. Because the traffic intervention was en-forced to provide a high-security police escort to gov-ernment personnel, passengers were not forewarned.Thus, the temporary traffic intervention changes thelevel of crowdedness within the subway trains, creat-ing a sudden spike in crowdedness. During this traf-fic halt intervention, the wireless provider randomly

7 We acknowledge the associate editor and two anonymous review-ers for this insight. One common strategy to address endogeneitythreats is to use instrument variables (Shriver et al. 2013, Sonnieret al. 2011, Stephen and Toubia 2010). However, a valid instrumentvariable (correlated with crowding but not correlated with purchaserates) is hard to obtain in our natural setting of crowding in sub-way trains. Another common strategy is Propensity Score Matching(PSM) as discussed later in §3.2.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


sent SMS to passengers on subway trains. It promoteda different product category with the same purchaseprice for greater generalizability of the findings. Thepromoted service enabled users to stream videos ontheir mobile devices. The SMS read “To watch thenewest videos on your mobile anytime, anywhere, for only¥3 per month, reply KTV3 now and get ¥3 off of yournext month’s bill!” For mobile users who respondedwith the provided mobile short code, the cost wascharged to their phone bill. Crowdedness was mea-sured in the same manner and mobile users who hadnot previously subscribed to this video service wereselected. Thus, we can pool the street closure sam-ple and add the interaction between the street clo-sure and crowdedness. A significant interaction herewould be a stronger test of the impact of crowdednessbecause this impact is driven by the sudden varia-tions in crowdedness induced by the street closures.To that end, we developed the following model:

Ui = �+�× crowdednessi +�× street closurei+�× crowdednessi × street closurei +� × weekdayi

+ �× peak houri + � × locationi + � ×Xi + �i21 (2)

where street closure is a dummy variable (coded as = 1for users in the street closure sample and = 0 if oth-erwise). Because the street closures affected two spe-cific subway stations, we added a location variable inmodel 2.

We report the results in Table 3, Panel A. Inmodel 1 we entered the mobile covariates to controlfor user heterogeneity in terms of observable mobileuse behaviors. In model 2 we entered the street clo-sure variable to control for its main effects. In model 3we entered the crowdedness variable, and in model 4we entered the interaction between crowdedness andthe street closure. As shown in Table 3, model 2, thestreet closure dummy was negative and insignificant,which makes sense because the street closure itselfshould not lead to a higher purchase likelihood. Asshown in model 4 of Table 3, the interaction betweencrowdedness and the street closure was positive andsignificant. This confirms that variations in mobilepurchases are indeed significantly driven by suddenvariations in crowdedness induced by the street clo-sure shock aboveground.

We also verified that the average crowdedness inthis street closure during the same affected hour(14:00–15:00 during the business week on Thursday)was 3.257 passengers/m2. As expected, this is indeedsubstantially higher than the counterpart (2.01 pas-sengers/m2) not in this street closure but in the samehour. In addition, there may be a concern that thestreet closure intervention could bring a systemati-cally different type of commuter population (one thatnormally uses surface transportation modes) into the

Table 3 Evidence with Sudden Variations in Crowdedness

Panel A: Sudden street closures effects

Crowdedness× 00492∗∗

Sudden street closures 4001875Crowdedness 00126∗∗ 00114∗∗

4000415 4000425Sudden street closures −00120 −00142 −10887

4001175 4001775 4100575Ln(ARPU) 00301∗∗ 00308∗∗ 00308∗∗ 00306∗∗

4001185 4001195 4001195 4001195Ln(MOU) −00043 −00043 −00044 −00044

4000655 4000655 4000655 4000655Ln(SMS) 00014 00014 00015 00013

4000695 4000695 4000695 4000695Ln(GPRS) −00001 −00001 −00001 −00001

4000245 4000235 4000235 4000235

Day effects (weekend dummy) Yes Yes Yes YesTime effects Yes Yes Yes Yes

(peak hour dummy)

Observations 11,960 11,960 11,960 11,960

Panel B: Sudden train delays effects

Crowdedness× 00669∗∗∗

Sudden train delays 4001485Crowdedness 00306∗∗∗ 00282∗∗∗

4001085 4001195Sudden train delays 00142 00130 00162

4001615 4001845 4001745Ln(ARPU) −000003 −000003 −000003 −000003

40000165 40000165 40000165 40000165Ln(MOU) 000001 000001 000001 000001

40000015 40000015 40000015 40000015Ln(SMS) 000001 000001 000001 000001

40000035 40000035 40000035 40000035Ln(GPRS) −000000 −000000 −000000 −000000

40000005 40000005 40000005 40000005

Day effects (weekend dummy) Yes Yes Yes YesTime effects Yes Yes Yes Yes

(peak hour dummy)

Observations 13,702 13,702 13,702 13,702

Notes. ARPU, Average revenue per user; MOU, minutes of use; SMS, num-ber of texts sent and received per user; GPRS, data use with the wirelessprovider.

∗p < 0010; ∗∗p < 0005.

subway system, which may confound our results. Wechecked and confirmed that the observable character-istics of people in the street closure sample are statis-tically the same as their counterparts not in the streetclosure sample (see Table 4). Thus, this alleviates theconcern that street closures bring a different popula-tion into the subway and further supports the effectof crowding on mobile purchases.

Still, we cannot rule out the possibility that thisstreet closure aboveground may involve a differenttype of commuter population belowground. Thus,we need sudden variations in crowdedness thatinvolve the same subway commuter population. This

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


Table 4 Summary Statistics of Mobile Users’ Wireless UsageBehavior

Regular Street closures Train delayspassengers passengers passengers

(N = 101690) (N = 11270) (N = 31012)

Mean S.D. Mean S.D. Mean S.D.

ARPU 590424 410144 590551 410488 590118 400147MOU 5670024 7310090 5610063 7360778 5500804 7190550SMS 3110913 2220189 3120346 2240979 3150522 2280972GPRS 27,856.6 8,867.1 28,445.8 8,944.3 27,336.8 8,506.7

Notes. F –tests of ARPU, MOU, SMS, and GPRS show no statistical differ-ences (all p > 0020) across the three samples, and t-tests of them also showno statistical differences (all p > 0020) across any combination of two sam-ples. ARPU, MOU, SMS, and GPRS comprise key indicators of wireless usebehavior. ARPU is the revenue that one customer’s cellular device generated.MOU is how much voice time a user spent on her mobile. SMS is the amountof monthly text messages sent and received. GPRS is a measure of the indi-vidual monthly volume of data used with the wireless service provider.

is achieved by exploiting unanticipated train delaysin the subway system.

3.1.3. Identification to Address Endogeneity viaUnanticipated Train Delays. On a Friday in the cityof our study, the subway trains were delayed for sev-eral minutes because a passenger’s backpack becameaccidentally stuck in a train door. As a result, thetrains suddenly became more crowded. Because thetrain delays were unexpected a priori, this naturalexperiment with sudden variations in crowdednessfurther addresses the potential endogenous selec-tion bias via the same population of commuters inthe subway system. During the unanticipated traindelays, the wireless provider randomly sent mobileads to users to stream videos on their mobile devices.Crowdedness was measured in the same manner.Thus, we can add the interaction between unantici-pated train delays and crowdedness. Significance ofthis interaction would be a stronger test of the impactof crowdedness because this impact is driven by thesudden variations in crowdedness induced by unan-ticipated train delays with the same population ofsubway passengers. Thus we developed the followingmodel:

Ui = �+�× crowdednessi +�× train delayi

+�× crowdednessi × train delayi +� × weekdayi

+ �× peak houri + � ×Xi + �i31 (3)

where train delay is a dummy variable (coded as = 1for users in the delayed trains, and = 0 if otherwise).Table 3, Panel B presents the results. In model 1 weentered the mobile covariates to control for user het-erogeneity. In model 2 we entered the unanticipateddelay shock variable to control for its main effects.

In model 3 we entered the crowdedness variable,and in model 4 we entered the interaction betweencrowdedness and the unanticipated train delay. Asshown in Table 3, Panel B, model 2, the unanticipatedtrain delay shock, itself, was insignificant. This makessense because the unanticipated delay, per se, shouldnot lead to a higher purchase likelihood. As shownin models 3 and 4, crowdedness significantly boostsmobile ad purchase likelihood (p < 0005). Most impor-tant, the interaction between crowdedness and theunanticipated train delay was positive and significant.This confirms that variations in mobile purchases areindeed significantly driven by the sudden variationsin crowdedness that are exogenously induced by theunanticipated train delays.8 We also determined thatthe average crowdedness with train delays around10:00 during the business week on Friday was 3.63passengers/m2. As expected, this is indeed higherthan the corresponding level (2.91 passengers/m25without train delays in the same hour. We also verifiedthat the observable characteristics of commuters in thetrain delay sample are statistically the same as theircounterparts not in this sample (see Table 4). This con-firms that the same subway population was involvedin the train delays. Therefore, these results providemore empirical evidence for the effects of crowdingby exploiting an exogenous shock of unanticipateddelayed trains in the subway system.

3.2. Robustness Checks withPropensity Score Matching

We adopt PSM to determine consumer responses tocrowding after matching passengers via observablecovariates. PSM tests whether consumer responsesare driven by crowding rather than user heterogene-ity; after matching covariates via PSM, users are notheterogeneous but rather homogenous. With PSM,the field data mimics randomized field experimentsand becomes a quasi-experiment design with pseudo-treatment and pseudo-control groups (Huang et al.2012). Then, the pseudo-treatment is exogenous andconfounds are randomized away, so that the effectson mobile ad purchases are attributed to crowdedness(Rosenbaum and Rubin 1983, Rubin 2006).9

8 Note that unanticipated delays can lead passengers to use theirphones more (e.g., to inform others about the delays), which cansystematically change the purchase rates. However, these resultsprovide more evidence that, together with the other analyses, thereis a causal effect of crowdedness on purchase rates. We thank ananonymous reviewer for this point.9 We acknowledge the associate editor and an anonymous reviewerfor this suggestion. Huang et al. (2012, p. 133) note that “PSM andinstrumental variables are two common techniques to correct forselection bias” and if instruments are hard to find, PSM is a viablesolution. In essence, conditional on covariates, PSM helps ensurethat the assignment of commuters to pseudo-treatment or control

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


PSM involves two stages. The first is to derive thepropensity score to match individuals, and the secondis to test the effects of crowding conditional on thematched homogenous individuals. In the first stage,we mirrored randomization by matching the proba-bility of passengers who would experience the sud-den variations of crowding with PSM. The propensityscore is the probability that passengers would entertrains with sudden variations of crowding (pseudo-treatment group) versus otherwise (pseudo-controlgroup), given their observed covariates of mobile usebehavior. We derive propensity scores using a logis-tic regression in which the dependent variable = 1if a commuter enters trains with sudden variationsin crowding, and = 0 if otherwise. We then matchthe propensity scores, and as a result can comparecommuters who are virtually similar to each other inthe pseudo-treatment group with those in the controlgroup. Matching these two groups allows us to isolateand test the pseudo-treatment effects of crowdednessin the second stage, after accounting for the effects dueto observable covariates.

Table 5 reports the two-stage results of the PSM.Panel A summarizes the results using the suddenvariations in crowding from street closures, whereasPanel B does so for sudden variations from traindelays. In Panel A, we have 782 matched passen-gers (391 matches between the pseudo-treatment andcontrol groups) from the passengers who rode thesame train in the same affected hour. In Panel B, wehave 2,270 matched passengers (1,135 matches). InTable 5 first-stage results in Panels A and B suggestthat many linear and squared terms of mobile usebehaviors are significant in determining the propen-sity score.10 As shown in Figure 2, Panels A and B, thepropensity score distributions between the pseudo-treatment and control groups are quite different in thehistogram plots before PSM using nearest neighbor

crowding is independent of crowdedness outcomes (Rosenbaumand Rubin 1983). When the propensity scores of users in the twotypes of crowdedness are identical, users are equally likely toexperience pseudo-treatment of crowding, since the values of theconfounding covariates indicate an equal chance. We also ana-lyzed the data with same-train-same-time commuters but withoutPSM and found consistent support for the significant effects ofcrowding.10 The PSM literature suggests that use of the squared terms in thefirst stage is common. For example, Huang et al. (2012, p. 134)used population squared, income per capital squared, Herfindahlindex squared, distance in nearest Walmart store squared, andtotal population within two miles squared. Entering squared termsmakes sense when they improve the model fitness in generatingthe propensity score, i.e., generating a better propensity score formatching observables. Nevertheless, we ran the PSM model with-out such squared terms and found consistent evidence for theeffects of crowding, albeit with worse first stage model fitness ingenerating the propensity score.

Table 5 Robustness Checks with Propensity Score Matching

Estimate Pr > ChiSq

Panel A: Street closures PSM

First-stage PSM parameterLn(ARPU) −10387 00015Ln(MOU) 10642 00002Ln(GPRS) 00303 00000Ln2(ARPU) 00458 00000Ln2(MOU) −00131 00001Ln2(SMS) −00038 00018Ln2(GPRS) −00045 00000

Second-stage PSM parameterCrowdedness×Sudden street closures 00318∗∗∗

4000755Crowdedness 00131∗∗ 00125∗∗

4000525 4000475Sudden street closures −00082 −00095

4000695 4000765

Observations 782 782

Panel B: Train delays PSM

First-stage PSM parameterIntercept 109.6 <000001Ln(ARPU) 309236 <000001Ln(MOU) −001596 005231Ln(SMS) 002555 005131Ln(GPRS) −1705559 <000001Ln2(ARPU) −004188 <000001Ln2(MOU) −0000349 008699Ln2(SMS) −000193 005941Ln2(GPRS) 006442 <000001

Second-stage PSM parameterCrowdedness×Sudden train delays 00519∗∗∗

4001315Crowdedness 00281∗∗∗ 00236∗∗∗

4001115 4001315Sudden train delays 10124 10074

4001905 4001855

Observations 2,270 2,270

Notes. In the first-stage of PSM, the predicted likelihood in this logit modelis the propensity score P 4X5. After matching P 4X5 with nearest neighborscores, the sample t-tests of ARPU, MOU, SMS, and GPRS show no statisti-cal differences (p > 0010) between the pseudo-treatment and pseudo-controlgroups. ARPU, Average revenue per user; MOU, minutes of use; SMS, num-ber of texts sent and received per user; GPRS, data use with the wirelessprovider.

∗p < 0010; ∗∗p < 0005; ∗∗∗p < 0001.

matching. However, after matching, the distributionsare virtually identical between the two groups. Thus,PSM helps to ensure that the commuters are virtuallysimilar to each other, suggesting that we have estab-lished user homogeneity to the extent possible.

In the second stage of PSM, we test our results ofcrowdedness effects conditional on matched propen-sity scores. The second-stage results in Table 5, Pan-els A and B confirm that crowdedness significantlyaffected mobile purchase likelihood after PSM. Of

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


Figure 2 (Color online) Propensity Score Distribution

particular importance, conditional on user homogene-ity via PSM, the interaction between crowdednessand street closures (Panel A) and unanticipated traindelays (Panel B) were both statistically significant(p < 0005). As such, this confirms that after usingPSM to derive virtually similar passengers, consumerresponse to mobile ads are indeed driven by crowd-edness, and that the lifts in mobile ad response are

robust to sudden variations in crowdedness inducedby both street closures aboveground and train delaysbelowground.

3.3. More Evidence with Survey DataFurther evidence with survey data provides deeperinsights into the effects of crowding. With its customerservice call-center, the wireless provider surveyed300 passengers. Among these, 180 had received and

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


Figure 2 (Color online) Continued

Panel B1: Train delays before PSM

Histograms of propensity scores by delay group

Panel B2: Train delays after PSM

Histograms of propensity scores by delay group

0

10

20

30

40

50

60Pe

rcen

t

0

20.017.515.012.510.07.55.02.5

020.017.515.012.510.07.55.02.5

0

10

20

30

40

50

60

Perc

ent

N 3.035Mean 0.73Median 0.98Mode 1.00

NormalKernel normal

N 1.135Mean 0.65Median 0.66Mode .

NormalKernel normal

N 2.382Mean 0.34Median 0.34Mode .

NormalKernel normal

0

–0.16 –0.08 0 0.08 0.16 0.24 0.32 0.40 0.48 0.56 0.64 0.72 0.96 1.04 1.12 1.20 1.280.80 0.88

1

Perc

ent

Perc

ent

01

Estimated probability

0.18 0.22 0.26 0.30 0.34 0.38 0.42 0.46 0.50 0.54 0.58 0.62 0.74 0.78 0.860.82 0.90 0.94 0.98 1.02 1.060.66 0.70

pscore

purchased the promoted service, and 120 had receivedbut had not purchased it. The surveys can help iden-tify possible reasons that crowdedness affects pur-chase likelihood, after matching the purchase recordsand measured crowdedness from the field data.11

Appendix B shows the survey instrument adopted by

11 A caveat of surveys is that correlation, rather than causality, isinvolved. Given the low purchase rate, another caveat is the over-sampling of users who made mobile purchases. Yet oversamplingalso implies a high chance of capturing the true purchase reasonsamong all purchasers. Still, we have a representative sample ofnonpurchasers similar to the general users in our data in terms ofmobile use behavior characteristics. Also, the surveys help rule outseveral alternative explanations such as social anxiety, preventionmindsets, price consciousness, and deal proneness (see Appendix Bfor the constructs).

our corporate partner. We report the results in Table 6.In model 1, we entered mobile use behaviors, alongwith survey responses to questions related to pref-erence for the promotion, perceptions of missed callfrequency, a prevention focus, deal proneness, andprice consciousness. As expected (and as shown inmodel 1), preference for the promotion had a sig-nificant impact on purchase likelihood (p < 0005). Inmodel 2 we entered the crowdedness variable fromour field data. The results confirm that crowdednesshad a significant impact on mobile purchase (p < 0005).This corroborates our field data evidence for the effectof crowdedness. Interestingly, as shown in model 3,mobile immersion had a significant impact, whilecrowdedness remains significant, albeit with a smallereffect size (all p < 0005). As expected, in model 4,

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


Table 6 Field Survey Results

Parameter Model 1 Model 2 Model 3 Model 4

Ln(ARPU) 00066 00067 −00116 000334003645 4003675 4004235 4003725

Ln(MOU) −00127 −00135 −00124 −001284001625 4001645 4001855 4001655

Ln(SMS) −00145 −00132 00228 −001594001905 4001915 4002135 4001965

Ln(GPRS) −00067 00060 00119 000634000675 4000685 4000755 4000685

Missed call preference 00741∗∗ 00723∗∗ 00766∗∗ 00752∗∗

4001225 4001225 4001345 4001295Missed call frequency 00023 00026 00125 00029

4001405 4001415 4001535 4001435Prevention focus 00061 −00029 00042 00082

4001465 4001715 4002145 4002025Price consciousness −00041 −00047 −00074 −00016

4001455 4001465 4001625 4001485Deal proneness 00080 00060 00112 00116

4001415 4001435 4001595 4001485Crowdedness 00238∗∗ 00219∗∗ 00152∗

4001165 4001185 4000975Mobile immersion 00821∗∗ 00725∗∗

4003525 4003585Social mimicry −00203 −00165

4002455 4002315Down time 00144 00220

4001435 4001345Social anxiety −00097 −00059

4001605 4001525Show off −00076 −00186

4001615 4001505Mobile involvement 00882∗∗

4002185Day effects (weekend dummy) Yes Yes Yes YesTime effects (peak hour dummy) Yes Yes Yes Yes

Observations 235 235 235 235

Notes. The dependent variable here is purchase or not. ARPU, Average rev-enue per user; MOU, minutes of use; SMS, number of texts sent and receivedper user; GPRS, data use with the wireless provider.

∗p < 0010; ∗∗p < 0005.

mobile involvement is positively related to purchaselikelihood (p < 0005). Also, via bootstrap mediationtests (Preacher and Hayes 2004), we find that crowded-ness is positively related to mobile immersion (0.465,p < 0001), which is positively related to involvementwith SMS and phones (1.375, p < 0001). In both cases,zero is not included in the 95% confidence intervalestimates of the bootstrap mediation tests. Together,these findings support a mobile immersion explanation:As crowding invades one’s physical space, peopleadaptively turn inwards and become more involvedwith their personal mobile devices. In turn, the moreconsumers are involved with their mobile devices, themore likely they are to purchase (Petty et al. 1983).

4. ConclusionThis research tests whether consumers are more likelyto respond to mobile ads in the hyper-contextualdimension of a crowded environment. It analyzes theimpact of crowdedness in subway trains on users’propensity to respond to targeted mobile ads. Anideal way to gauge the effect of crowdedness wouldbe to randomize consumers into different conditionsof crowdedness. Because this is practically impossi-ble to achieve in an actual field setting, we dependon field data (a quasi-field experiment combinedwith natural experiments) and multiple identificationstrategies to address potential endogeneity and selec-tion issues.

Mobile technology provides novel measures ofphysical crowdedness. In the unique marketing envi-ronment of our data, the underground subway isequipped with subway-specific cellular lines that runalong the tunnel walls. Crowdedness is then gaugedas the volume of mobile users in each subway trainwho automatically connect to this underground lineas a function of the size of the train compartment.Our data analyses: (1) control for selection on observ-ables that include peak versus nonpeak traffic hours,weekdays versus weekends, mobile use behaviors,and randomly sending mobile ads, (2) exploit sud-den changes in crowdedness induced by street clo-sures aboveground and train delays belowground,and (3) use a PSM estimator to further eliminate selec-tion bias. With over 15,000 mobile users, our datadocuments that crowdedness has a positive effect onmobile ad purchase likelihood after a lower threshold,and that this crowding effect is quite robust to differ-ent modeling techniques. Further analyses with fieldsurveys suggest that a mobile immersion accountsfor why crowding affects response to mobile ads:Consumers may be more deeply immersed in theirmobile phones when in a crowded environment. Thatis, consistent with the social psychology literature, ascrowding invades one’s physical space, people adap-tively turn inwards and become more susceptible tomobile ads.

Our results offer several theoretical and managerialimplications. Theoretically, we deepen understandingof consumer reactions to crowding. The consumerbehavior literature reveals how people react to crowd-ing. Crowding can trigger nervousness and decreasecreativity (Maeng et al. 2013). In retail settings it canprovoke negative attitudes towards the store (Huiand Bateson 1991). Crowding has also been shownto induce avoidance behaviors and prompt shop-pers to rely more on familiar brands, avoid inter-action with store employees and others, and reduceshopping time (Harrell et al. 1980). Overall, thesestudies have largely shown the negative effects of

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


crowding. However, we find that targeting consumersin crowded environments with mobile promotionsmay serve as a welcome relief. That is, in the contextof the consumers’ physical environment, our workreveals positive effects of crowding on mobile ads.

We also extend the mobile commerce literature byshowing that environmental contexts such as crowd-ing in public transit are significant determinants ofmobile ad effectiveness. Physical crowding affects con-sumer receptivity to mobile messages. We advanceprior research on mobile targeting (Danaher andDagger 2013; Dickinger and Kleijnen 2008; Ghose et al.2013a, b; Fong et al. 2015; Luo et al. 2014; Molitoret al. 2014) by revealing the new hyper-contextualdimension of crowding, a novel way to target mobileusers. For marketers, the ability to reach consumersanytime, anywhere, offers substantial opportunities(Kim et al. 2011). The potency of mobile ads hingeson understanding consumers’ real-time experiencesand the hyper-contextual environment (Kenny andMarshall 2000). Marketers must not only compre-hend “how customers behave with respect to themobile medium” (Shankar and Balasubramanian 2009,p. 128) but also how they behave with respect to theirdynamic environment.

The average public transportation commute timefor Americans is 48 minutes each way (McKenzie andRapino 2011). This considerable amount of consumerdown time during daily commutes can be a gold minefor marketers. Grocery retailers in countries such asSouth Korea have created virtual stores in under-ground subways by superimposing product imageswith Quick Response (QR) codes over the platformwalls. While waiting for their train, commuters canpurchase groceries on their mobiles for convenientdelivery (Solon 2011). In the Netherlands, infraredsensors inside trains are used to send mobile alertsto boarding passengers about each car’s occupancy toimprove commuting efficiency (Wokke 2013). Indeed,marketers are adopting new mobile technologies suchas iBeacon (a short-distance pulsing device that oper-ates on the Bluetooth low energy of any iOS7 device).Retailers can send within-store mobile promotions toconnect with shoppers in real time. Sports stadiumsand entertainment venues have also adopted iBea-con to help spectators find their seats and receiveconcession-stand offers (Grobart 2013). Understand-ing consumers’ quotidian routines and contextualactivities enables marketers to deliver relevant mes-sages that cater to customer needs in the moment.Especially as Apple’s iPhone 6 hits the market withbigger screen sizes for more vivid viewing, mar-keters can deliver more media-rich mobile ads (Sloane2014). Indeed, consumers are receptive to receiving“messages from particular brands at specific timesand under certain conditions” (Johnson 2013). To the

extent that consumers have a greater involvementwith their mobile phones in relatively more crowdedcontexts, mobile marketing may see higher effective-ness in crowded settings with in-app advertising anddisplay ads (Bart et al. 2014).

However, there is a catch. Although crowding isan inherent part of everyday life, marketers shouldacknowledge that personal space preferences and tol-erance for crowding may vary by settings (subway,concert hall, and restaurant). For example, crowdedrestaurants more likely comprise self-selected atten-dance and can boost people’s feelings of excitementor signal quality (Knowles 1980, Tse et al. 2002). Inthat context, targeting consumers on their personaldevices may be construed as an annoying and unwel-come invasion of privacy. By contrast, our subway isless likely to involve self-selecting into desired crowdsbecause public transit often denotes commutes withstrangers. Also, in the subway setting, consumerscould instantly purchase and activate the promotedservice. Future research could investigate differentconsumption settings to reveal more nuanced effectsof crowdedness. Research could also study whetherdifferences in product conspicuousness matter whentargeting consumers in crowded versus noncrowdedsettings. Finally, future research may examine whetherconsumers can regulate their negative moods such asstress in crowded environments by purchasing hedo-nic goods or services.

In conclusion, this research serves as an initial stepto measure crowdedness with mobile technologiesand identify the effects of crowdedness on purchases.Managers may consider gauging the crowdedness ofa consumer’s physical environment as a new way toboost the effectiveness of mobile ads.

Supplemental MaterialSupplemental material to this paper is available at http://dx.doi.org/10.1287/mksc.2015.0905.

AcknowledgmentsThe authors acknowledge the immense and invaluablesupport from one of the world’s largest mobile serviceproviders. The authors thank the editor, associate editor,and anonymous reviewers for their constructive comments.This article is partly based on an essay in the first author’sdissertation at Temple University chaired by the secondauthor. The authors thank seminar participants at Tem-ple University, Northwestern University, New York Univer-sity, University of Utah, University of Texas at Austin, andthe Indian School of Business. The corresponding authorof this publication is Zheng Fang. This research is fundedby the National Natural Science Foundation of China[Grant 71172030, 71202138, 71472130], the Youth Founda-tion for Humanities and Social Sciences of the Ministryof Education of China [Grant 12YJC630045, 14YJA630024,14YJC630166], and Sichuan University [Grant skqy201423,skqy201502].

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


Appendix A. Histogram of ARPU Mobile UsageBehavior

2.00

600

400

Freq

uenc

y

200

03.00 4.00

Larpu

5.00 6.00 7.00

Appendix B. Field Survey ItemsThe corporate partner’s customer service call-center con-ducted the field surveys. The surveys were conductedthe day after the first part of the field study. This daywas selected because the first part of the field data wascompleted after 22:00, thus making it too late to callrespondents. To avoid demand effects, the surveys werepresented under the guise of a customer satisfaction sur-vey intended to assess satisfaction with the subway wirelesssignal. Before answering the survey questions, mobile userswere asked to recall when they received and purchased theSMS on the subway the preceding day to remind them ofthe crowdedness they experienced. Respondents were askedto confirm whether they had received the promotional SMSwhile in the subway and whether they had read and replied(or not) to it while in the subway.

Mobile immersion (adapted from Burger 1995)When surrounded by a lot of people in the subway, I am usually

eager to get away by myself.During a subway ride, I would like to spend the time quietly.

Social mimicry (Tanner et al. 2008)When I see others looking at their mobiles, I usually look at

my own mobile.Passing down time

I prefer to pass down time by fiddling with my mobile phonein the subway car.Social anxiety (Fenigstein et al. 1975)

Large groups in the subway make me nervous.During the subway ride, it is embarrassing to look at strangers.

Show off (Richins and Dawson 1992)I like my cellular phone to be noticed by other people.

Mobile involvement (Olsen 2007, Warrington and Shim2000)

In the subway, I am involved with my cell phone.I pay attention to incoming SMS during the subway ride.

Several alternative explanations to the results are tested.First, commuters may wish to occupy themselves duringthe down time of their travel. Crowdedness would restrictthe activities commuters could engage in to make thisdown time productive (i.e., it would be harder to use a

laptop). Also, due to crowded environments, the invasionof personal space would mean an increased likelihood ofcatching unwanted gazes, which can heighten social stressand embarrassment. This is tested for with a measure ofsocial anxiety (Fenigstein et al. 1975). Also, in crowded envi-ronments, there may be a higher number of consumerswho use their mobiles, which could visibly and subcon-sciously prompt others to do likewise. Thus, social mimicry(Chartrand and Bargh 1999, Tanner et al. 2008) is tested for.In public environments, some people may wish to flaunttheir smartphones to others. This alternative explanation istested by measuring the desire to show off mobile devices(Richins and Dawson 1992). All of these explanations mayaccount for why crowdedness may lead consumers tobecome more involved with their mobile devices. To ruleout additional explanations, several control variables wereincluded in the survey instrument. The covariates includeddeal proneness (Lichentenstein et al. 1995), price conscious-ness (Dickerson and Gentry 1983), a prevention-focusedmindset (per prior research which finds that crowded-ness triggers a prevention-focused mindset (Maeng et al.2013)), the perceived frequency of missed calls, and thepreference for missed call alerts. In addition, we controlfor mobile use behaviors with the company’s customerrecords.

ReferencesAgarwal A, Hosanagar K, Smith MD (2011) Location, location, loca-

tion: An analysis of profitability of position in online advertis-ing markets. J. Marketing Res. 48(6):1057–1073.

Agence France-Presse (2014) Hong Kong metro seats may bescrapped for smartphone space. Agence France-Presse. AccessedMarch 7, 2015. http://tinyurl.com/kltqspl.

Aiello JR, DeRisi DT, Epstein YM, Karlin RA (1977) Crowdednessand the role of interpersonal distance preference. Sociometry40(3):271–282.

Bart Y, Stephen A, Sarvary M (2014) Which products are best suitedto mobile advertising? A field study of mobile display adver-tising effects on consumer attitudes and intentions. J. MarketingRes. 51(3):270–285.

Brehm JW (1966) A Theory of Psychological Reactance (AcademicPress, New York).

Bull M (2005) No dead air! The iPod and the culture of mobilelistening. Leisure Stud. 24(4):343–355.

Burger J (1995) Individual differences in preference for solitude.J. Res. Personality 29(1):85–108.

Chartrand TL, Bargh JA (1999) The chameleon effect: The percep-tion–behavior link and social interaction. J. Personality Soc.Psych. 76(6):893–910.

Collette J, Webb SD (1976) Urban density, household crowdednessand stress reactions. J. Sociol. 12(3):184–191.

Danaher P, Dagger T (2013) Comparing the relative effectivenessof advertising channels: A case study of a multimedia blitzcampaign. J. Marketing Res. 50(4):517–534.

Deng C, Graz J (2002) Generating randomization schedules usingSAS programming. Proc. 27th Annual SAS Users Group Internat.Conf. (SAS Institute, Inc., Cary, NC), 267–270.

Dickerson MD, Gentry JW (1983) Characteristics of adopters andnon-adopters of home computers. J. Consumer Res. 10(9):225–235.

Dickinger A, Kleijnen M (2008) Coupons going wireless: Deter-minants of consumer intentions to redeem mobile coupons.J. Interactive Marketing 22(3):23–39.

eMarketer (2012) Mobile banners continue to boost high clickrates. Emarketer. Accessed March 7, 2015. http://tinyurl.com/oynne4h.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

http://tinyurl.com/kltqspl

http://tinyurl.com/oynne4h

http://tinyurl.com/oynne4h


eMarketer (2014) Driven by Facebook and Google, mobile ad mar-ket soars 105% in 2013. Emarketer. Accessed March 7, 2015.http://tinyurl.com/ns9s2hh.

Evans GW, Wener RE (2007) Crowdedness and personal space inva-sion on the train: Please don’t make me sit in the middle.J. Environ. Psych. 27(1):90–94.

Fenigstein A, Scheier MF, Buss AH (1975) Public and private self-consciousness: Assessment and theory. J. Consulting ClinicalPsych. 43(4):522–527.

Flegenheimer M (2013) Wi-Fi and cell phone service on subwaytrains? M.T.A. leader says it may happen. New York Times.Accessed March 7, 2015. http://tinyurl.com/mcmc5gh.

Fong N, Fang Z, Luo X (2015) Geo-conquesting: Competitivelocational targeting of mobile promotions. J. Marketing Res.Forthcoming.

Ghose A, Han SP (2011) An empirical analysis of user content gen-eration and usage behavior on the mobile Internet. ManagementSci. 57(9):1671–1691.

Ghose A, Han SP (2014) Estimating demand for mobile applicationsin the new economy. Management Sci. 60(6):1470–1488.

Ghose A, Goldfarb A, Han SP (2013a) How is the mobile Internetdifferent? Search costs and local activities. Inform Systems Res.24(3):613–631.

Ghose A, Han SP, Park S (2013b) Analyzing the interdepen-dence between Web and mobile advertising: A random-ized field experiment. Working paper, New York University,New York.

Griffitt W, Veitch R (1971) Hot and crowded: Influence of popula-tion density and temperature on interpersonal affective behav-ior. J. Personality Soc. Psych. 17(1):92–98.

Grobart S (2013) Apple’s location-tracking iBeacon is poised for usein retail sales. Businessweek. Accessed March 7, 2015. http://tinyurl.com/lhf2p8c.

Guadagni PM, Little JDC (1983) A logit model of brand choice cal-ibrated on scanner data. Marketing Sci. 2(3):203–238.

Harrell G, Hutt M, Anderson J (1980) Path analysis of buyer behav-ior under conditions of crowdedness. J. Marketing Res. 17(1):45–51.

Huang Q, Nijs VR, Hansen K, Anderson ET (2012) Wal-mart’simpact on supplier profits. J. Marketing Res. 49(2):131–143.

Hui M, Bateson J (1991) Perceived control and the effects of crowd-edness and consumer choice on the service experience. J. Con-sumer Res. 18(2):174–184.

Hui SK, Inman JJ, Huang Y, Suher J (2013) The effect of in-storetravel distance on unplanned spending: Applications to mobilepromotion strategies. J. Marketing 77(2):1–16.

Jarrell S, Howsen R (1990) Transient crowdedness and crime: Themore “strangers” in an area, the more crime except for murder,rape, and assault. J. Econom. Sociol. 49(4):483–494.

Johnson L (2013) Consumers are driving need for contextual mobileexperiences: Token exec. Mobile Commerce Daily. AccessedMarch 7, 2015. http://tinyurl.com/lrqrkf2.

Kenny D, Marshall JF (2000) Contextual marketing: The real busi-ness of the Internet. Harvard Bus. Rev. 78(6):119–125.

Kim JB, Albuquerque P, Bronnenberg BJ (2011) Mapping online con-sumer search. J. Marketing Res. 48(1):13–27.

Knowles ES (1980) An affiliative conflict theory of personal andgroup spatial behavior. Paulus PB, ed. Psychology of Group Influ-ence (Lawrence Erlbaum, Hillsdale, NJ), 133–188.

Levav J, Zhu R (2009) Seeking freedom through variety. J. ConsumerRes. 36(4):600–610.

Lichentenstein DR, Netemeyer RG, Burton S (1995) Assessing thedomain specificity of deal proneness: A field study. J. ConsumerRes. 22(3):314–326.

Luo X, Andrews M, Fang Z, Phang CW (2014) Mobile targeting.Management Sci. 60(7):1738–1756.

Maeng A, Tanner RJ (2013) Construing in a crowd: The effects ofsocial crowding on mental construal. J. Experiment. Soc. Psych.49(6):1084–1088.

Maeng A, Tanner RJ, Soman D (2013) Conservative when crowded:Social crowdedness and consumer choice. J. Marketing Res.50(6):739–752.

Marshall J (2014) Weather-informed ads are coming to Twitter. WallStreet Journal. Accessed March 7, 2015. http://tinyurl.com/mq7f8z6.

McKenzie B, Rapino M (2011) Commuting in the United States:2009. U.S. Census Bureau, American Community SurveyReports.

Milgram S (1970) The experience of living in cities. Science167(3924):1461–1468.

Milgram S, Sabini J (1978) On maintaining social norms: A field ex-periment in the subway. Baum A, Singer JE, Valins S, eds. Ad-vances in Environmental Psychology, Vol. 1 (Lawrence Erlbaum,Hillsdale, NJ), 31–40.

Molitor D, Reichhart P, Spann M (2013) Location-based advertising:Measuring the impact of context-specific factors on consumers’choice behavior. Working paper, Munich School of Manage-ment, Munich, Germany.

Molitor D, Reichhart P, Spann M, Ghose A (2014) Measuringthe effectiveness of location-based advertising: A random-ized field experiment. Working paper, New York University,New York.

Olsen SO (2007) Repurchase loyalty: The role of involvement andsatisfaction. Psych. Marketing 24(4):315–341.

Petty R, Cacioppo JT, Schumann D (1983) Central and periph-eral routes to advertising effectiveness: The moderating role ofinvolvement. J. Consumer Res. 10(2):135–146.

Preacher KJ, Hayes AF (2004) SPSS and SAS procedures for esti-mating indirect effects in simple mediation models. BehavioralRes. Methods, Instruments, Comput. 36(4):717–731.

Richins ML, Dawson S (1992) A consumer values orientation formaterialism and its measurement: Scale development and val-idation. J. Consumer Res. 19(3):303–316.

Rosenbaum PR, Rubin DB (1983) The central role of the propen-sity score in observational studies for causal effects. Biometrika70(1):41–55.

Rubin DB (2006) Matched Sampling for Causal Effects (CambridgeUniversity Press, New York).

Schmitt RC (1966) Density, health, and social disorganization.J. Amer. Inst. Planners 32(1):38–40.

Shankar V, Balasubramanian S (2009) Mobile marketing: A synthe-sis and prognosis. J. Interactive Marketing 23(2):118–129.

Sherrod DR (1974) Crowdedness, perceived control, and behavioralaftereffects. J. Appl. Soc. Psych. 4(2):171–186.

Shriver SK, Nair HS, Hofstetter R (2013) Social ties and user-generated content: Evidence from an online social network.Management Sci. 59(6):1425–1443.

Sloane G (2014) The iPhone 6 should make mobile ads a big-ger deal. Adweek. Accessed March 7, 2015. http://tinyurl.com/nrxp6xp.

Solon O (2011) Tesco brings the supermarket to time-poor com-muters in South Korea. Wired. Accessed March 7, 2015. http://tinyurl.com/6bbr8cm.

Sommer R (2009) Personal space. Reis HT, Sprecher SK, eds. Ency-clopedia of Human Relationships (Sage, Thousand Oaks, CA),1227–1229.

Sonnier GP, McAlister L, Rutz OJ (2011) A dynamic model of theeffect of online communications on firm sales. Marketing Sci.30(4):702–716.

Stephen AT, Toubia O (2010) Deriving value from social commercenetworks. J. Marketing Res. 47(2):215–228.

Stokols D (1972) On the distinction between density and crowd-ing: Some implications for future research. Psych. Rev. 79(3):275–277.

Tanner R, Ferraro R, Chartrand T (2008) Of chameleons and con-sumption: The impact of mimicry on choice and preferences.J. Consumer Res. 34(6):754–767.

Tse ACB, Sin L, Yim FH (2002) How a crowded restaurant affectsconsumers’ attribution behavior. Internat. J. Hospitality Manage-ment 21(4):449–454.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

http://tinyurl.com/ns9s2hh

http://tinyurl.com/mcmc5gh

http://tinyurl.com/lhf2p8c

http://tinyurl.com/lhf2p8c

http://tinyurl.com/lrqrkf2

http://tinyurl.com/mq7f8z6

http://tinyurl.com/mq7f8z6

http://tinyurl.com/nrxp6xp

http://tinyurl.com/nrxp6xp

http://tinyurl.com/6bbr8cm

http://tinyurl.com/6bbr8cm


Warrington P, Shim S (2000) An empirical investigation of therelationship between product involvement and brand commit-ment. Psych. Marketing 17(9):761–782.

Wicklund RA (1974) Freedom and Reactance (Lawrence Erlbaum,Potomac, MD).

Wokke AD (2013) NS-APP laat realtime drukte in treinen zien.Tweakers. Accessed March 7, 2015. http://tinyurl.com/cxxhnad.

Xu J, Shen H, Wyer RS (2012) Does the distance between us matter?Influences of physical proximity to others on consumer choice.J. Consumer Psych. 22(3):418–423.

Zimbardo P (1969) The human choice: Individuation, reason, andorder versus deindividuation, impulse, and chaos. Arnold W,Levine D, eds. Nebraska Sympos. Motivation, Vol. 17 (Universityof Nebraska Press, Lincoln, NE), 237–307.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

2 N

ovem

ber

2015

, at 0

5:03

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

http://tinyurl.com/cxxhnad

mobile ad effectiveness: hyper-contextual targeting with ... · mobile ad effectiveness:...

Documents