E-commerce transactions, the installed base of credit ... · review the empirical literature on the...

13
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=raec20 Download by: [Curtin University Library] Date: 27 October 2016, At: 18:46 Applied Economics ISSN: 0003-6846 (Print) 1466-4283 (Online) Journal homepage: http://www.tandfonline.com/loi/raec20 E-commerce transactions, the installed base of credit cards, and the potential mobile E-commerce adoption Gary Madden, Aniruddha Banerjee, Paul N. Rappoport & Hiroaki Suenaga To cite this article: Gary Madden, Aniruddha Banerjee, Paul N. Rappoport & Hiroaki Suenaga (2017) E-commerce transactions, the installed base of credit cards, and the potential mobile E- commerce adoption, Applied Economics, 49:1, 21-32, DOI: 10.1080/00036846.2016.1189507 To link to this article: http://dx.doi.org/10.1080/00036846.2016.1189507 Published online: 27 May 2016. Submit your article to this journal Article views: 101 View related articles View Crossmark data

Transcript of E-commerce transactions, the installed base of credit ... · review the empirical literature on the...

Page 1: E-commerce transactions, the installed base of credit ... · review the empirical literature on the growth and ... brand-ing, awareness, ... to the literature in international economics

Full Terms & Conditions of access and use can be found athttp://www.tandfonline.com/action/journalInformation?journalCode=raec20

Download by: [Curtin University Library] Date: 27 October 2016, At: 18:46

Applied Economics

ISSN: 0003-6846 (Print) 1466-4283 (Online) Journal homepage: http://www.tandfonline.com/loi/raec20

E-commerce transactions, the installed base ofcredit cards, and the potential mobile E-commerceadoption

Gary Madden, Aniruddha Banerjee, Paul N. Rappoport & Hiroaki Suenaga

To cite this article: Gary Madden, Aniruddha Banerjee, Paul N. Rappoport & Hiroaki Suenaga(2017) E-commerce transactions, the installed base of credit cards, and the potential mobile E-commerce adoption, Applied Economics, 49:1, 21-32, DOI: 10.1080/00036846.2016.1189507

To link to this article: http://dx.doi.org/10.1080/00036846.2016.1189507

Published online: 27 May 2016.

Submit your article to this journal

Article views: 101

View related articles

View Crossmark data

Page 2: E-commerce transactions, the installed base of credit ... · review the empirical literature on the growth and ... brand-ing, awareness, ... to the literature in international economics

E-commerce transactions, the installed base of credit cards, and the potentialmobile E-commerce adoptionGary Maddena, Aniruddha Banerjeeb, Paul N. Rappoportc and Hiroaki Suenagad

aDepartment of Economics and Property, Curtin University, Perth, Australia; bSSRS, Media, PA, USA; cDepartment of Economics, TempleUniversity, Philadelphia, PA; dDepartment of Finance and Banking, Curtin University, Perth, WA, Australia

ABSTRACTMobile e-commerce (m-commerce) relaxes consumers’ temporal and geographic purchasingconstraints and encourage the establishment of omnichannel markets. It is often argued thatrapid increase in smartphone penetration is the primary driver of m-commerce adoption,whereas others contend that early adoption of m-commerce applications are mostly by “relativelyheavy” Internet commerce users. Brynjolfsson et al. (2013) argue that rapid increase in smart-phone penetration is the primary driver of m-commerce adoption, whereas Einav et al. (2014)contend that early adoption of m-commerce applications are mostly by ‘relatively heavy’ Internetcommerce users. This article explores strength of the influences within a nested multiple-serviceframework, where the reduced-form econometric analysis allows for interdependency betweenm-commerce and e-commerce services, and the installed base of credit cards. The results reveal acomplex situation in which credit cards facilitate e-commerce services, whereas m-commerceadoptions are driven by prior e-commerce and online transaction activity. Also, higher respon-dent incomes are negatively associated with proposed m-commerce adoption. Surprisingly,privacy concerns do not affect proposed adoption independently; however, an interaction termsuggests privacy remains an adoption barrier for the older persons.

KEYWORDSM-commerce adoption;multi-service framework;U.S. consumer preferences;trivariate probit model;survey data

JEL CLASSIFICATIONC25; C93; D12; L96; O51

I. Introduction

The rapid adoption of smartphones has createdmany new and potentially disruptive commercialopportunities (Brynjolfsson, Yu, and Rahman2013). For instance, smartphones provide customerswith an ability to conduct personal communications,business transactions and banking operations in amore flexible manner (viz. less restrictive temporaland geographic environment). Furthermore, toenhance productivity, organisations are integratingwireless and mobile technology into business opera-tions. Important emerging m-commerce (mobilecommerce) applications include banking, entertain-ment, information and ticketing services.

More formally, m-commerce refers to businessmodel which allow consumers to complete commercialtransactions using mobile devices, either at a point-of-sale (POS) (e.g. payments made through near field

communications technology) or remotely (e.g. SMSpayments or payments charged on mobile operators’bills) (OECD 2012).1 Enabled devices can be waved atPOS terminals equipped to read information fromdigital wallets, and provide fast, private and secure(and often doubly authenticated) transactions, withno customer information revealed to merchants.

Potential m-commerce adoption depends,among other things, on consumers’ currentInternet (e-commerce) commerce activity and theinstalled base of credit cards.2 In a recent study,Einav et al. (2014) document the early effects ofhow mobile devices might change Internet andretail commerce. In particular, they report, via ananalysis of eBay’s mobile shopping application, thatearly adopters of mobile e-commerce applicationsare ‘relatively heavy’ e-commerce users. Thus,e-commerce experiences probably influence earlym-commerce activity. Furthermore, online

CONTACT Gary Madden [email protected] Department of Economics and Property, Curtin University, Perth, WA, Australia1While it goes by many names, a ‘digital wallet’ containing encrypted credit or debit card information is loaded into mobile devices.2The U.S. Census Bureau defines e-commerce as ‘transactions sold on-line whether over open networks such as the Internet or proprietary networks runningsystems such as Electronic Data Interchange.’ Electronic data interchange is defined as ‘the structured transmission between organizations by electronicmeans . . . without human intervention.’ Electronic data interchange is more applicable to B2B e-commerce than the B2C transactions that define the retailsector (Lieber and Syverson 2012).

APPLIED ECONOMICS, 2017VOL. 49, NO. 1, 21–32http://dx.doi.org/10.1080/00036846.2016.1189507

© 2016 Informa UK Limited, trading as Taylor & Francis Group

Page 3: E-commerce transactions, the installed base of credit ... · review the empirical literature on the growth and ... brand-ing, awareness, ... to the literature in international economics

transactions are primarily facilitated through creditcard payments. This suggests that m-commercedepends both on received e-commerce activity andthe installed base of credit cards.3 While alternativemeans of payment (such as PayPal and online storetransaction cards) are emerging, credit cardsremain the most often used payment tool.

Not surprisingly, there is an absence of empiricalresearch exploring early m-commerce adoptionwithin a more realistic multi-service framework. Animpediment to the conduct of such research is thepaucity of available data comprised of paired Internet(including actual and intended mobile) commerceactivity with detailed consumer (household) socio-demographic composition information. Importantly,our econometric analysis is conducted using Centris©household survey data that contains information onsmartphone ownership, and several demographic andtrust variables. While the survey is cross-sectional,there are retrospective questions asking when(whether) the respondent started purchasing productsor services online via the regular Internet (viz. e-com-merce), and history of prior credit card adoption andusage. Respondents were also asked about whetherthey intended to purchase products and servicesonline via smartphone (viz. adopt m-commerce).

Methodologically, we empirically examine the inter-dependency betweenm-commerce, e-commerce (trans-actions conducted through desktop or laptop computersvia the Internet) and credit card based transactionswithin a trivariate probit reduced-form framework. Inparticular, a nested dependence relationship is specifiedwhere m-commerce depends on current e-commerceactivity and the credit card installed base, and e-com-merce depends on the credit card installed base.

Our results reveal a complex situation in whichcredit cards facilitate e-commerce services, whereasm-commerce adoptions are driven by prior e-com-merce and online transaction activity. Also, higherrespondent incomes are negatively associated withproposed m-commerce adoption. Surprisingly, priv-acy concerns do not affect proposed adoption

independently; however, an interaction term sug-gests privacy remains an adoption barrier for theolder respondents.

Additionally, we analysed the robustness of ourresults by estimating separate univariate equationprobit models for all endogenous variables. We alsoestimated an eight-dimensional multinomial prob-ability model that specified the joint probabilities ofthe three binomial variables as functions of exogen-ously determined demographic and trust variables.Comparison of results from the estimations revealedthe univariate binary models and the eight-dimen-sional multinomial model yielded coefficient esti-mates of similar magnitudes, and the samestatistical significance as those obtained for the tri-variate probit model.

Prior to our results it was thought that earlym-commerce adoption depended, at least in part,on the installed base of credit cards and prior e-com-merce activity. However, these beliefs were merelymarket speculations based on anecdotal data ratherthan rigorous analysis. Our findings have implica-tions for mobile platforms that want to extend theirreach, viz. they should innovate to better promotethe utility of the mobile platforms to credit cardholders and regular Internet transactors.

We organise the evidence in five sections. We firstreview the empirical literature on the growth andimpacts of U.S. e-commerce markets. Then inSection III, we describe the descriptive statistics forsample data, and variables for econometric analysisare defined. The trivariate probit and multinomialmodels are specified in Section IV, while empiricalresults are reported in Section V. The final sectionsummarises and suggests modelling extensions.

II. Selective literature review

Einav et al. (2014, 489) report that ‘early adopters ofmobile e-commerce applications appear to be peoplewho already were relatively heavy Internet com-merce users.’ They also argue that mobile shopping

3While contactless transactions are convenient in a variety of uses, the two most popular forms of m-commerce globally are undoubtedly POS payments andfund transfers. Starting in late 2014, a variety of high-profile smartphone makers launched own versions of mobile POS payment systems using similar NFCand authentication technologies. These include Apple Pay from Apple, Inc., Android Pay from Google, Inc. and Samsung Pay from Samsung (presentlylimited only to selected Samsung mobile devices). Although these systems are not yet universally available, significant use of them is already being (or willshortly be) made in the U.S., U.K. and elsewhere. Implementing systems require collaboration and revenue-sharing among smartphone makers, mobilenetwork operators, credit card-issuing banks and financial institutions and other intermediaries. PayPal Holdings, Inc., a leading international onlinepayments company, has just become the latest entrant into the mobile payments field. With already millions of users and several hundred thousands ofparticipating retailers and merchants, the prognosis for mobile POS payment growth worldwide is encouraging.

22 G. MADDEN ET AL.

Page 4: E-commerce transactions, the installed base of credit ... · review the empirical literature on the growth and ... brand-ing, awareness, ... to the literature in international economics

is ‘associated with both an immediate and sustainedincrease in total platform purchasing,’ and that suchmobile application purchases would not have beenmade via the regular Internet platform. Takentogether, these points suggest that analysis of theimpacts of e-commerce could provide useful insightsinto the probable consequences from wider m-com-merce adoption. Thus, we chronologically and selec-tively review the economic literature that questionsthe impact of e-commerce on retail price levels anddispersions, geographic patterns of trade andimpacts on the supply side of the retail industry.

Early studies of U.S. residential customer behaviorshowed that e-commerce affected both price levelsand dispersions (see, for example, Brynjolfsson andSmith 2000; Clay, Krishnan, and Wolff 2001; Brownand Goolsbee 2002). In particular, Brynjolfsson andSmith (2000) argue that ‘while there is lower frictionin many dimensions of Internet competition, brand-ing, awareness, and trust remain important sourcesof heterogeneity among Internet retailers.’ Morerecently, Gorodnichenko and Talavera (2014) estab-lished that prices in online stores, relative to pricesin brick-and-mortar stores, have shorter spell dura-tions; smaller changes; larger pass-throughs and fas-ter adjustment speeds. These findings are consistentwith the earlier ‘reduced friction’ thesis, and alsoindicate the increased integration of online markets.

Further analysis by Hortaçsu, Martínez-Jerez andDouglas (2009) used price data from online auctionsites (eBay and MecadoLibre) to identify the pre-sence of any online geographic trade patterns.Importantly, they (p. 53) found (in a manner similarto the literature in international economics litera-ture) that ‘distance continues to be an importantdeterrent to trade between geographically separatedbuyers and sellers.’ Interestingly, they isolate a ‘homebias’ for trading within the same city.

Yet another dimension to be considered is thatwhile the (U.S. Census Bureau tabulated) fraction ofall retail sales accounted for by e-commerceincreased monotonically from 0.9% (2000) to 6.4%(2014), these sales are concentrated in a narrowproduct range. For instance, Hortaçsu andSyverson (2015, p. 96) report that about 85% of retaile-commerce sales occur in the electronic shopping

and mail-order house industry (NAICS 45411).Further (p. 97), the product categories that accountfor most sales are online retail clothing, accessoriesand footwear (18%), ‘other merchandise’ (15%) andfurniture (10%).

Accordingly, Goldmanis et al. (2009) selectivelyexamined the effect of the diffusion of e-commerceon the supply-side structure of the U.S. retailindustry.4 Theoretically (p. 651) by interpreting theintroduction of e-commerce as reducing consumers’search costs, they argue that the primary competitiveimpact of e-commerce would be to ‘reallocate mar-ket shares’ from high-cost (‘mom and pop’ stores) tolow-cost producers (viz. large stores).5 Their esti-mated models on U.S. travel agency, bookstore andnew car dealer sales and firm count data conform tothese predictions.

Our empirical literature review clearly illustratedthe wide-ranging and fundamental nature of theconsequences from the diffusion of e-commerce. Inthis situation, a question that naturally arises is: whatfactors would determine the early adoption ofm-commerce? Thus, we empirically test whetherprior e-commerce and credit card activity are asso-ciated with proposed m-commerce adoption.Importantly, our econometric analysis is conductedusing household survey data that contains informa-tion on smartphone ownership, and several demo-graphic and trust variables. While the survey iscross-sectional, conveniently for us there are retro-spective questions asking when (whether) therespondent started purchasing products or servicesonline via the regular Internet (viz. e-commerce),and history of prior credit card adoption andusage. The respondent was also asked about whetherthey intended to purchase products and servicesonline via smartphone (viz. adopt m-commerce).We construct from these responses data (detailedbelow) for econometric analysis.

III. Survey, data and variables

Centris© is a U.S.-based market research companythat surveys micro-level consumer demand forinformation and communications technology, andmedia products. Centris© conducted a pan-U.S

4The interested reader should also consult Bronnenberg and Ellickson (2015) for an excellent review.5Support for the ‘search cost’ approach can also be found in Ghose, Goldfarb and Han (2012).

APPLIED ECONOMICS 23

Page 5: E-commerce transactions, the installed base of credit ... · review the empirical literature on the growth and ... brand-ing, awareness, ... to the literature in international economics

survey of mobile phone usage, current e-commerceactivity and proposed m-commerce activity. Thesurvey identified mobile phone, Internet, onlinesocial network, e-commerce and proposed m-com-merce activity.

The survey is a stratified random sample with strati-fication by the key demographics of gender, age, house-hold income, race/ethnicity and region (balanced toapproximate national proportions). Operationally,households are drawn from Internet panel maintainedby Luth Research, Inc., San Diego, CA. Questionnaireresponses are provided by 4743 individuals (whichinclude the primary decision maker and other house-hold members aged at least 19 years) residing in 2584households. The sample for estimation is comprised of4560 individual survey responses.6 The primary deci-sion maker provides household demographic informa-tion and answers decision-maker only questions.Additionally, the primary decision maker returns thecompleted questionnaires.

Questionnaire response formats include: Likertscale, dichotomous, categorical and open-ended.The survey questionnaire is sent to primary decisionmakers via an e-mail interface. Survey questionselicited information on the: household structure(size, composition and decision maker); householdinventory of wireless service plans, providers and

equipment (including smartphones); wireless serviceusage and satisfaction (indicating switchingbehaviour); inventory and usage of wireline,Internet, TV and other communications/entertain-ment services; and consumer preference foradvanced and proposed wireless services (includingm-commerce).

The variables constructed from these survey datainclude respondent: current online transaction activ-ity (TRANSACT); credit card usage (CREDIT); age(AGE); income band (INCOME); technology owner-ship (SMART); online privacy concerns (PRIVACY);online account security concerns (SECURITY) andproposed m-commerce activity (MCOM).7

Tables 1 and 2, respectively, present the defini-tion, mean and standard deviations of endogenousand exogenous variables with the latter classifiedvariously as demographic, technology ownershipand trust measures that are deterrents to conducting

Table 1. Endogenous variable summary statistics.

Variable Definition MeanStd.dev.

MCOM = 1, if respondent proposed to adoptm-commerce services; = 0, otherwise

0.26 0.44

TRANSACT = 1, if respondent conducted onlinetransactions; = 0, otherwise

0.41 0.49

CREDIT = 1, if respondent used credit cards for onlinetransactions; = 0, otherwise

0.80 0.40

Table 2. Exogenous variable summary statistics.Variable Definition Mean Std. dev.

DemographicAGE = respondent age (years) 41.89 15.42INCOME = 5, if income > 250; = 4, if 200 < income < 250; = 3,

if 100 < income < 200; = 2, if 50 < income < 100; = 1,if 25 < income < 50; = 0, if income < 25 (1000 dollars)

1.92 1.40

TechnologySMART = 1, if respondent owns smartphone or PDA; = 0, otherwise 0.58 0.49

TrustPRIVACY = 4, if strongly deterred m-commerce adoption; = 3,

if somewhat deterred; = 2, if neither deterred norundeterred; = 1, if somewhat undeterred; = 0, if undeterred

2.38 1.33

SECURITY = 4, if strongly deterred m-commerce adoption; = 3,if somewhat deterred; = 2, if neither deterrednor undeterred; = 1, if somewhat undeterred; = 0, if undeterred

2.44 1.33

PRIVACY×AGE = Online privacy and age interaction 100.29 72.14

6The primary decision maker is previously identified by Luth Research, Inc. The survey was conducted in February 2013. 183 responses are omitted. 180responses have no mobile phone and three questionnaires are incomplete.

7The variables SECURITY and PRIVACY are constructed from the sample data. Section VI (Electronic and Mobile Commerce) of the questionnaire (p. 39)contains questions concerned with online security and privacy. The categorical variable SECURITY is obtained from Question 06-A-6, which asks: ‘Using acell phone to make purchases or bill payments may involve some risks. Knowing your personal habits and circumstances, are you likely to experience thefollowing risk.’ Five scenarios are available to be presented, however, only one scenario is presented to each respondent. For example, ‘I, or someone usingmy cell phone, may charge something by accident (tap when not intended).’ The scenarios are rotated and presented to different respondents. Availableresponses are: very likely; somewhat likely; neither likely nor unlikely; somewhat unlikely and very unlikely. Data for the variable PRIVACY is obtained fromQuestion 06-A7B, which asks: ‘How much are you deterred (or prevented) from making payments or transferring funds over the Internet by you concernsabout ..’ Seven scenarios are available to be presented. Again the available responses are rotated and presented to different respondents. For example,Scenario 2 is ‘unauthorized payments’, whilst Scenario 6 is ‘Theft by fraudulent websites.’

24 G. MADDEN ET AL.

Page 6: E-commerce transactions, the installed base of credit ... · review the empirical literature on the growth and ... brand-ing, awareness, ... to the literature in international economics

transactions online. Respondent age and income areincluded as control variables to account for theeffects of demographic variables on the potentialdemand for m-commerce.

The reduced-form econometric model attemptsto provide estimates of the impact of demo-graphic control (AGE, INCOME), technologyownership (SMART), current online activity andprerequisite service adoption (TRANSACT andCREDIT, respectively) and trust indicator(PRIVACT, SECURITY) variables on the prob-ability of adopting m-commerce (MCOM). Inparticular, the nested specification is in partmotivated by Einav et al. (2014) observation thatthe early adopters of mobile e-commerce applica-tions appear to be relatively heavy Internet com-merce users.8

To obtain consistent parameter estimates, ancil-lary equations are required for TRANSACT andCREDIT. For TRANSACT, it is widely recognizedthat credits cards, as an enabling technology, pro-mote the conduct of Internet transactions.Furthermore, the empirical analysis recognises thatonline transaction activity is in part based on demo-graphic control and trust variables. Goldfarb andTucker (2012) also find that older persons aremore privacy protective than younger people. Thisacknowledged pattern of distrust is also apparent inonline environments (Chong 2013; Lu et al. 2011;Zhang, Chen, and Lee 2013).9 Finally, Belanger,Hiller, and Smith (2002) suggest that consumers,when purchasing online, value security featuresmore than third-party privacy seals or privacystatements.10

For CREDIT, several papers analysed theeffects of individual consumers’ socio-demo-graphic characteristics on the adoption of pay-ment instruments by demographic cohort (see,e.g. Bertaut and Haliassos 2005; Zinman 2009).Importantly, they found credit card use washigher for older, higher income, more educatedand wealthier consumers. Accordingly, anotherequation is specified with CREDIT explained bydemographic controls.

IV. Trivariate probit and multinomial models

The binary dependent variables modelled are pro-posed m-commerce activity (MCOM), currentonline transaction activity (TRANSACT), and creditcard usage for online transactions (CREDIT). Thevariables are potentially related via a nested relation-ship. In particular, MCOM is specified as deter-mined by TRANSACT and CREDIT, whileTRANSACT is determined by CREDIT.TRANSACT and CREDIT is explained by othervariables. This structure identifies a multivariate bin-ary choice model. As TRANSACT and CREDITenter the MCOM equation, the recursive simulta-neous structure is consistently and efficiently esti-mated by FIML probit:

y�1;i ¼ β01x1;i þ δ12y2;i þ δ13y3;i þ ε1;i; y1;i ¼ 1ðy�1;i > 0Þ

y�2;i ¼ β02x2;i þ δ23y3;i þ ε2;i; y2;i ¼ 1ðy�2;i > 0Þ

y�3;i ¼ β03x3;i þ ε3;i; y3;i ¼ 1ðy�3;i > 0Þ

(1)

where the latent variables y�m;i can be given usualinterpretation as representing individual i’s netutility from selecting the three binary decisions(MCOM, TRANSACT and CREDIT for m = 1, 2and 3, respectively) and xm;i represents the vectorof exogenous variables affecting each binaryvariable.

The model arises when disturbances arecorrelated,

ε1;iε2;iε3;i

0@

1A,N

000

0@

1A;

1 ρ12 ρ13ρ21 1 ρ23ρ31 ρ32 1

0@

1A

24

35,N3 0;R½ �

(2)

where ρ is the tetrachoric pairwise correlationbetween disturbances εm;iðm ¼ 1; 2; 3Þ.

The recursive structure in Equation (1) imposesthe unidirectional causality from the credit transac-tions to the online transactions and also from thesetwo variables to the proposed future m-commerceusage. This recursive structure is necessary for themodel to be internally consistent and identifiable as

8Additionally, they argue that mobile application purchases are not simply purchases that would have been made otherwise on the regular Internetplatform, viz. regular online and mobile e-commerce are not substitutes.

9Importantly, smartphones are inherently less secure than wired desktop or laptop computers as limited bandwidth; memory and processing capabilityinhibit the installation of conventional data encryption and security software.

10On the role of trust in online environments, see Belanger, Hiller and Smith (2002), Lu et al. (2011) and Huang, Lu and Ba (2016).

APPLIED ECONOMICS 25

Page 7: E-commerce transactions, the installed base of credit ... · review the empirical literature on the growth and ... brand-ing, awareness, ... to the literature in international economics

shown by Amemiya (1985), Greene (2008) andMaddala (1984).11 Additionally to this econometricrequirement, the first causality is motivated by theprevious findings in the literature that credit cardspromote internet transactions as the enabling tech-nology. The second causality follows from the tem-poral nature of the survey questions; the respondentswere asked about their past experience of credit cardand online transactions and their proposed use ofm-commerce transactions in the future.

As an alternative to the trivariate probit modeldefined in (1) and (2), we also considered an eight-dimensional multinomial model for joint outcomesof the three binary variables,12

y�j;i ¼ ZiΦj þ vj;i (3)

where y�j;i represents the individual i’s net utilityfrom choosing the choice j which is indexed by thethree-digit binary numbers; 000, 001, 010, 011, 100,101, 110 and 111, representing the outcome of thethree binary choices (MCOM, TRANSACT andCREDIT, respectively) and yi ¼ m is observed ify�m;i > y�j;i for all j � m. The vector of exogenous

variables Zi should contain all the variables thataffect some or all of the three endogenous binaryvariables and hence be identical for all eight multi-nomial equations.

Two distributional assumptions are commonlyconsidered for the error term, vj;i, in the multinomialmodel (3). The probit model assumes the multivari-ate normal distribution, which allows non-zero cov-ariance of vj;iacross the choices j while assuming zerocorrelation across individuals i, i.e. covðvi;k; vj;lÞ ¼ 0"i � j and covðvi;k; vi;lÞ ¼ σkl. The logit modelassumes the extreme value distribution of the errorterm with zero correlation both across individualsand across choices.

Comparison of multinomial probit and logitmodels in the literature agrees on two points. First,estimating a multinomial probit model requiresnumerically evaluating the cumulative distributionfunction of multivariate normal distribution, which

is computationally intensive and often impracticalwhen the model involves a large number of choices.Multinomial logit model is less computationallyintensive, due primarily to the assumed indepen-dence of the error terms across choices. Yet, theassumption imposes the independence of irrelevantalternative property, which is often consideredunrealistic. Generally, the probit approach is oftenpreferred to logit approach in this regard. However,since the model (3) contains only the individual-specific attributes and not the choice-specific attri-butes on the right-hand side of the equation, thecovariances of the error term vj;i cannot be identifiedeven when they are allowed to be non-zero in thecase of the probit model. Thus, the probit modelloses its advantage over the logit model on the sec-ond of the above two points. We therefore assumethe extreme distribution of the error term.

Assuming the extreme value distribution of vj;i,the model becomes,

Pðyi ¼ kÞ ¼ expðZiΦkÞPjexpðZiΦjÞ (4)

where the coefficient vector Φj needs to be con-strained to be zero for any arbitrarily chosen choice j.

The multinomial model (3) predicts the jointprobability of the three binary choices as a functionof individual specific attributes, Z. It is more flexiblethan a multivariate binomial model, which predictsonly the marginal probability of each binary decisionseparately from other binary variables. As discussed,the three binary decisions follow clear temporalorder that respondents conduct credit and onlinetransactions prior to declaring their proposed futureuse of m-commerce at the time of the survey. Due tothis temporal order, our primary interest is on howindividual’s propensity for future m-commerceusage depends on his/her past experience of creditand online transactions. This can be assessed bycalculating the difference in the conditional prob-ability of proposing the future m-commerce usagebetween the two states of each conditional variable,

11The model would be consistent and identified and hence would be estimated by the method such as the instrumental variable method, were it to includethe latent variables (y�j;i) on the right-hand side of the equations instead of the observed binary variables (yj;i). In our context, such model would specify thenet utility from the proposed m-commerce use to depend not on the past experience of online and credit card transactions but on the unobserved netutility from these experiences. We found such specification unrealistic and hence decided to specify the net utility from m-commerce as a function of thepast experience of the online and credit card transactions.

12This transformation of a multivariate binomial model into a multinomial model has been long known since Cox (1972) and Amemiya (1981). However, ithas not been appreciated widely with only a small number of its application to the analysis of empirical data, see for example, Velandia et al. (2009).

26 G. MADDEN ET AL.

Page 8: E-commerce transactions, the installed base of credit ... · review the empirical literature on the growth and ... brand-ing, awareness, ... to the literature in international economics

PðM ¼ 1jT ¼ 1Þ � PðM ¼ 1jT ¼ 0Þ

¼ PðM ¼ 1 \ T ¼ 1ÞPðT ¼ 1Þ � PðM ¼ 1 \ T ¼ 0Þ

PðT ¼ 0Þ(5a)

and

PðM ¼ 1jC ¼ 1Þ � PðM ¼ 1jC ¼ 0Þ

¼ PðM ¼ 1 \ C ¼ 1ÞPðC ¼ 1Þ � PðM ¼ 1 \ C ¼ 0Þ

PðC ¼ 0Þ(5b)

where

P M ¼ 1 \ T ¼ 1ð Þ ¼ Pð111Þ þ Pð110ÞP M ¼ 1 \ T ¼ 0ð Þ ¼ Pð101Þ þ Pð100ÞP M ¼ 1 \ C ¼ 1ð Þ ¼ Pð111Þ þ Pð101ÞP M ¼ 1 \ C ¼ 0ð Þ ¼ Pð110Þ þ Pð100Þ

P T ¼ 1ð Þ ¼ Pð111Þ þ Pð110Þ þ Pð011Þ þ Pð010ÞP C ¼ 1ð Þ ¼ Pð111Þ þ Pð101Þ þ Pð011Þ þ Pð001Þ:The effect of a change in each exogenous factor in

vector Z on the marginal probability of each binaryvariable is also obtained as,

@PðM ¼ 1Þ@Xj

¼ @

@XjPð111Þ þ Pð110Þ þ Pð101Þ þ Pð100Þð Þ

@PðT ¼ 1Þ@Xj

¼ @

@XjPð111Þ þ Pð110Þ þ Pð011Þ þ Pð010Þð Þ

@PðC ¼ 1Þ@Xj

¼ @

@XjPð111Þ þ Pð101Þ þ Pð011Þ þ Pð001Þð Þ:

(6)

The expressions in (5a) and (5b) are nonlinearfunctions of the joint probabilities, whereas thosedefined in (6) are linear in the marginal effects ofthe exogenous factors on the joint probabilities.These expressions can be evaluated on the estimatedmultinomial logit model, and statistical significancecan be assessed with standard errors obtainedthrough the delta method.

V. Empirical results

The trivariate probit model in (1) and (2) is esti-mated via LIMDEP version 9.0 by simulated max-imum likelihood using a smooth recursive Geweke-Hajivassiliou-Keane simulator to evaluate the

multivariate normal probabilities.13 The simulatedmaximum likelihood estimator is asymptoticallyconsistent as the number of observations anddraws tend to infinity.14 The multinomial logitmodel defined in (4) is estimated by the standardmaximum likelihood estimated method, whichyields a consistent and asymptotically normally dis-tributed estimate of the parameter vector Φ. Themodel estimation and evaluation of conditionalprobabilities (5a) and (5b) as well as the marginaleffects in (6) are conducted via Gauss version 12.

Table 3 shows the estimates of a trivariate probitspecification. Proposed m-commerce activity(MCOM), current online transaction activity(TRANSACT) and credit card usage (CREDIT) arespecified as endogenous variables. In Table 3, sys-tems estimation that allows correlations among thedisturbances, clearly identifies the important suppor-tive role played by TRANSACT and CREDIT ininitial MCOM adoption, with the estimated coeffi-cients of 0.5936 and 1.0842, respectively, which areboth statistically significant at 1% test size. The resultsupports the finding of Einav et al. (2014) that pro-posed early MCOM adoption appears higher forthose who are relatively heavy Internet commerceusers.

Among the exogenous variables, the smartphoneownership variable (SMART) has a positive coeffi-cient (0.2345) in the estimated MCOM equation.That is, early adopters of the installed technologylead m-commerce adoption. Finally, the controlvariables AGE (–0.0245) and INCOME (–0.0429)indicate that proposed m-commerce adoption isless likely for elderly persons and those with higherincomes. These coefficients in the MCOM equationare all strongly significant (at 1% test size).

The estimated TRANSACT equation indicatese-commerce transaction activity is more likely whenrespondents use credit cards (CREDIT = 2.2938) fortransacting online. That is, regular credit card usereinforces online transaction activity. Conversely, pro-posed e-commerce adoption is damped when respon-dents are concerned by system trust issues (SECURITY= –0.0978). Interestingly, the estimated coefficient isnot statistically significant for PRIVACY while it issignificant negative for PRIVACY×AGE (–0.0019).

13For details of the algorithm see Train (2003, p. 126–137).14Cappellari and Jenkins (2003) argue that when the number of draws is more than the square root of the sample size, the parameter estimates are robust tothe initial seed value. Furthermore, within this framework the variances of the disturbances are normalized to unity.

APPLIED ECONOMICS 27

Page 9: E-commerce transactions, the installed base of credit ... · review the empirical literature on the growth and ... brand-ing, awareness, ... to the literature in international economics

This result supports Goldfarb and Tucker (2012) whoargue that privacy concerns particularly inhibitInternet commerce adoption by older people. NeitherAGE nor INCOME has significant impact on the prob-ability of e-commerce transaction. The significant rela-tionship between higher credit card use and older(0.0132) and wealthier (0.0260) persons reported byBertaut and Haliassos (2005) and Zinman (2009) issubstantial by the CREDIT equation coefficient esti-mates. Finally, the correlations between the three dis-turbances are statistically significantly different fromzero. The result indicates that the unobserved factorsaffecting the three binary decisions are correlated.

The robustness of the system is examined byseparately estimating univariate equation probitmodels for each endogenous variable (see Table 4).

Comparison shows that the univariate equation esti-mations provide identical significant coefficientsigns to systems estimation, with the reported coeffi-cient magnitudes are similar. One noticeable differ-ence is that while AGE has no significant impact onTRANSACT under systems estimation, a positivecoefficient is reported for the single equationestimation.

Table 5 reports the marginal effects of a change ineach explanatory variable on the probabilities of thethree endogenous binary choices, as evaluated at thesample means of the explanatory variables. The tableindicates that the three variables, SMART,TRANSACT and CREDIT, have substantial impactson the proposed m-commerce use with the adoptionof each of these three conditions increasing the

Table 3. Trivariate probit model estimates.

Variables

MCOM TRANSACT CREDIT

Coefficient Std. error Coefficient Std. error Coefficient Std. error

Constant –0.8577*** 0.0845 –1.4993*** 0.1056 0.3820*** 0.0509DemographicsAGE –0.0245*** 0.0015 0.0004 0.0020 0.0132*** 0.0010INCOME –0.0429*** 0.0146 –0.0183 0.0127 0.0260** 0.0123TechnologySMART 0.2345*** 0.0418ActivityTRANSACT 0.5936*** 0.0505CREDIT 1.0842*** 0.0703 2.2938*** 0.0409TrustPRIVACY –0.0563 0.0365SECURITY –0.0978*** 0.0213PRIVACY×AGE –0.0019*** 0.0007

Observations 4560Log likelihood –7111.09CorrelationρMT 0.1297*** 0.0208ρMC –0.5125*** 0.0178ρTC –0.9178*** 0.0000

***, ** and * denote significance at 1%, 5% and 10%, respectively. All regressions are checked for heteroscedasticity.

Table 4. Univariate probit model estimates.

Variable

MCOM TRANSACT CREDIT

Coefficient Std. error Coefficient Std. error Coefficient Std. error

Constant –0.4573*** 0.0829 –0.9418*** 0.1309 0.0762 0.0670DemographicsAGE –0.0215*** 0.0015 0.0131*** 0.0027 0.0172*** 0.0014INCOME –0.0344** 0.0153 –0.0132 0.0146 0.04890*** 0.0154TechnologySMART 0.3002*** 0.0450ActivityTRANSACT 0.4726*** 0.0443CREDIT 0.3998*** 0.0613 1.1486** 0.0612TrustPRIVACY 0.0150 0.0507SECURITY –0.1451*** 0.0272PRIVACY×AGE –0.0047*** 0.0010

Observations 4560 4560 4560Log likelihood –2346.94 –2635.49 –2170.77

***, ** and * denote significance at 1%, 5% and 10%, respectively. All regressions are checked for heteroscedasticity.

28 G. MADDEN ET AL.

Page 10: E-commerce transactions, the installed base of credit ... · review the empirical literature on the growth and ... brand-ing, awareness, ... to the literature in international economics

probability of the proposed m-commerce use by 8.6,14.2 and 10.8 percentage point, respectively. Thepast experience of credit card transaction also has asubstantial positive impact (an increase of 34.2%) onthe online transaction. Two demographic variables,age and income, have significant impact on the threebinary decisions (except that the effect of income onTRANSACT is insignificant). However, the magni-tudes of these impacts are rather small.

Finally, an eight-dimensional multinomial logitmodel defined in (3) and (4) yields the estimationresults that are essentially identical to the trivariateprobit model.15 The model imposes no recursivestructure on the relationship of the three endogen-ous binary variables. It thus provides even strongersupport for the robustness of the results obtained forthe link between the proposed m-commerce adop-tion and the past experience of online and creditcard transactions.

Table 6 reports the change in the conditional prob-ability of the proposed future m-commerce usebetween the two states of each of the two conditionvariables; CREDIT and TRANSACT. They areobtained according to (5a) and (5b) by evaluating theestimated multinomial logit model at the samplemeans of the six exogenous variables. In the table,the probability of the proposed future m-commerceuse is higher by 0.18 percentage point for individualswith past experience of online transaction(TRANSACT = 1) than for those without past online

transaction (TRANSACT = 0). Similarly, the probabil-ity of the proposed future m-commerce use is higherby 0.16 percentage point for those with past experienceof credit card transaction (CREDIT = 1) than for thosewithout past credit card transaction (CREDIT = 0).These differences in the conditional probabilities ofthe proposed m-commerce use are statistically signifi-cant at 1% test size. The results are consistent withstatistically positive coefficients of TRANSACT andCREDIT in the estimated trivariate probit model.

In Table 7, the effects of the exogenous variableson the marginal probabilities of the three binaryvariables are also consistent with those obtainedfrom the estimated trivariate probit model. Of thesix exogenous variables, SMART has the largestand significant effect on the probability of theproposed m-commerce use. It also has the largestpositive impacts on TRANSACT and CREDIT,which reflect strong positive association of thesetwo endogenous variables with MCOM. AGE andINCOME lower the probability of MCOM whilethey increase the probability of credit card transac-tion. SECURTIY has negative impacts on probabil-ity of online transaction, and so does thePRIVACY variable, yet only for older persons.These impacts are statistically significant, yetmuch smaller in magnitudes than the effect ofTRANSACT, CREDIT and SMART on MCOMprobability.

VI. Conclusions

Ubiquitous mobile technology, in particular smart-phones, has made the spread of mobile e-commerceinevitable. This study addresses the potential demandfor mobile e-commerce services in the U.S. This studyimplicitly considers whether mobile e-commerce is

Table 5. Probit marginal effects.

Variable

MCOM TRANSACT CREDIT

Effect Std. error Effect Std. error Effect Std. error

AGE –0.0062*** 0.0004 0.0043*** 0.0009 0.0046*** 0.0004INCOME –0.0010** 0.0044 −0.0043 0.0048 0.0131*** 0.0041SMART 0.0865*** 0.0128TRANSACT 0.1421*** 0.0135CREDIT 0.1075*** 0.0150 0.3419*** 0.0138PRIVACY 0.0049 0.0167SECURITY –0.0478*** 0.0089PRIVACY×AGE –0.0015*** 0.0003

***, **, and * denote significance at 1%, 5%, and 10%, respectively.

Table 6. Effects of changes in conditional probabilities.Effect Std. error

P(M = 1|C = 1) − P(M = 1|C = 0) 0.1581 0.0160 ***P(M = 1|T = 1) − P(M = 1|T = 0) 0.1837 0.0181 ***

Standard errors are obtained by delta method. *, **, and ***, respectively,represent that the effect is statistically significant at 10%, 5% and 1% testsize.

15Full estimation results on the multinomial logit model are provided in Appendix, Table A1.

APPLIED ECONOMICS 29

Page 11: E-commerce transactions, the installed base of credit ... · review the empirical literature on the growth and ... brand-ing, awareness, ... to the literature in international economics

likely to evolve independently as a ‘new’ service or beconsidered by consumers as a complement to existingsystems such as e-commerce and credit cards.

Importantly, this study informs the debate on thebest means by which mobile e-commerce is encour-aged, and how mobile e-commerce, e-commerce andcredit cards might share an integrated market. In par-ticular, the study reveals that credit cards, e-commerceand mobile e-commerce are related services. Forinstance, credit cards facilitate e-commerce services,whereas mobile e-commerce adoptions are driven byprior e-commerce and online transaction activity.

Another norm this study empirically establishedis that young persons will be early adopters.Interestingly, the higher income groups view mobilee-commerce as an inferior good. Surprisingly, priv-acy concerns do not appear to generally affect pro-posed mobile e-commerce activity independently;however, privacy remains an issue of concernamong the older population.

Finally, future research could complement ourfindings by focusing on the risks of fraud and priv-acy violations. Thus far, the response to such hazardshas been mostly via technology, e.g. encryption andauthentication procedures to keep consumer data

private.16 For now, regulatory agencies in the U.S.and the European Union are relying on existing lawsto monitor and police m-commerce activity, andthey remain vigilant about consumer privacy protec-tion. This situation, and the acceptance and spreadof m-commerce, would rely on regulators ability toadapt to emerging concerns as they arise.

Disclosure statement

No potential conflict of interest was reported by the authors.

References

Amemiya, T. 1981. “Qualitative Response Models: A Survey.”Journal of Economic Literature 19: 1483–1536.

Amemiya, T. 1985. Advanced Econometrics. Cambridge:Harvard University Press.

Belanger, F., J. Hiller, and W. Smith. 2002. “Trustworthinessin Electronic Commerce: The Role of Privacy, Security,and Site Attributes.” The Journal of Strategic InformationSystems 11: 245–270. doi:10.1016/S0963-8687(02)00018-5.

Bertaut, C., and M. Haliassos. 2005. “Credit Cards: Facts andTheories.” In Economics of Consumer Credit, edited by G.Bertola, R. Disney, and C. Grant. Cambridge, MA:MIT Press.

Table 7. Marginal effects of exogenous variables on joint and marginal probabilities.

Probability at mean

Marginal effects

{MTC} AGE INCOME SMART PRIVACY SECURITY PRIVACY × AGE

Joint probability{111} 0.1409 –0.0003 –0.0080* 0.0953*** –0.0097 –0.0137* –0.0004

(0.0008) (0.0045) (0.0127) (0.0148) (0.0076) (0.0003){110} 0.0061 –0.0001 –0.0005 0.0004 0.0002 0.0009 0.0000

(0.0002) (0.0008) (0.0028) (0.0035) (0.0015) (0.0001){101} 0.0833 –0.0020*** 0.0007 0.0215** 0.0171 0.0076 –0.0002

(0.0007) (0.0029) (0.0093) (0.0113) (0.0060) (0.0003){100} 0.0141 –0.0015*** –0.0025** 0.0028 –0.0032 0.0018 0.0002*

(0.0003) (0.0010) (0.0029) (0.0039) (0.0023) (0.0001){011} 0.2536 0.0079*** 0.0035 0.0625*** 0.0155 –0.0432*** –0.0014***

(0.0010) (0.0059) (0.0168) (0.0213) (0.0126) (0.0004){010} 0.0179 0.0005* 0.0007 0.0050 0.0027 –0.0046 –0.0001

(0.0003) (0.0014) (0.0042) (0.0061) (0.0028) (0.0001){001} 0.3453 0.0034*** 0.0139** –0.0798*** 0.0199 0.0408*** 0.0005

(0.0011) (0.0056) (0.0173) (0.0204) (0.0148) (0.0004){000} 0.1388 –0.0079*** –0.0078*** –0.1076*** –0.0424*** 0.0104* 0.0013***

(0.0004) (0.0022) (0.0067) (0.0082) (0.0059) (0.0001)Marginal probabilityP(M = 1) 0.2444 –0.0039*** –0.0103* 0.1200*** 0.0044 –0.0034 –0.0004

(0.0011) (0.0054) (0.0161) (0.0193) (0.0114) (0.0004)P(T = 1) 0.4185 0.0080*** –0.0043 0.1631*** 0.0086 –0.0605*** –0.0019***

(0.0013) (0.0074) (0.0220) (0.0263) (0.0178) (0.0005)P(C = 1) 0.8231 0.0089*** 0.0101*** 0.0995*** 0.0427*** –0.0084 –0.0014***

(0.0005) (0.0026) (0.0077) (0.0100) (0.0057) (0.0002)

Numbers in parentheses are standard errors obtained by delta method. *, ** and ***, respectively, represent that the effect is statistically significant at 10%,5% and 1% test size.

16Regulators have mostly acted to protect competition and ensure unhindered access to money transfers across national borders and enforcement of privacyrequirements.

30 G. MADDEN ET AL.

Page 12: E-commerce transactions, the installed base of credit ... · review the empirical literature on the growth and ... brand-ing, awareness, ... to the literature in international economics

Bronnenberg, B., and P. Ellickson. 2015. “Adolescence andthe Path to Maturity in Global Retail.” Journal of EconomicPerspectives 29: 113–134. doi:10.1257/jep.29.4.113.

Brown, J., and A. Goolsbee. 2002. “Does the Internet MakeMarkets More Competitive? Evidence from the LifeInsurance Industry.” Journal of Political Economy 110:481–507. doi:10.1086/339714.

Brynjolfsson, E., and M. Smith. 2000. “FrictionlessCommerce? A Comparison of Internet Conventional andRetailers.” Management Science 46: 563–585. doi:10.1287/mnsc.46.4.563.12061.

Brynjolfsson, E., J. Yu, and M. Rahman. 2013. “Competing inthe Age of Omnichannel Retailing.” MIT SloanManagement Review 54: 23–29.

Cappellari, L., and S. Jenkins. 2003. “Multivariate ProbitRegression using Simulated Maximum Likelihood.” StataJournal 3: 278–294.

Chong, A. 2013. “Mobile Commerce Usage Activities: TheRoles of Demographic and Motivation Variables.”Technological Forecasting and Social Change 80: 1350–1359. doi:10.1016/j.techfore.2012.12.011.

Clay, K., R. Krishnan, and E. Wolff. 2001. “Prices and PriceDispersion on the Web: Evidence from the Online BookIndustry.” The Journal of Industrial Economics 49: 521–539. doi:10.1111/1467-6451.00161.

Cox, D. 1972. “Analysis of Multivariate Binary Data.” Journalof the Royal Statistical Society, Series C 21: 113–120.

Einav, L., J. Levin, I. Popov, and N. Sundaresan. 2014.“Growth, Adoption, and Use of Mobile E-Commerce.”American Economic Review: Papers & Proceedings 104:489–494. doi:10.1257/aer.104.5.489.

Ghose, A., A. Goldfarb, and S. Han. 2012. “How Is theMobile Internet Different? Search Costs and LocalActivities.” Information Systems Research 24: 613–631.doi:10.1287/isre.1120.0453.

Goldfarb, A., and C. Tucker. 2012. “Shifts in PrivacyConcerns.” American Economic Review 102: 349–353.doi:10.1257/aer.102.3.349.

Goldmanis, M., A. Hortaçsu, C. Syverson, and Ö. Emre.2009. “E-Commerce and the Market Structure of RetailIndustries.” The Economic Journal 120: 651–682.doi:10.1111/ecoj.2010.120.issue-545.

Gorodnichenko, Y., and O. Talavera (2014), ‘Price Setting inOnline Markets: Basic Facts, International Comparisons,and Cross-border Integration’, NBER Working Papers

20406, National Bureau of Economic Research, Inc. (forth-coming in American Economic Review).

Greene, W. 2008. “Discrete Choice Modeling.” In Handbookof Econometrics: Vol. 2, Applied Econometrics, edited by T.Mills and K. Patterson. London: Palgrave.

Hortaçsu, A., F. Martínez-Jerez, and J. Douglas. 2009. “TheGeography of Trade in Online Transactions: Evidencefrom eBay and MercadoLibre.” American EconomicJournal: Microeconomics 1: 53–74.

Hortaçsu, A., and C. Syverson. 2015. “The Ongoing Evolutionof US Retail: A Formal Tug-of-War.” Journal of EconomicPerspectives 29: 89–112. doi:10.1257/jep.29.4.89.

Huang, L., X. Lu, and S. Ba. 2016. “An Empirical Study of theCross-Chanel Effects between Web and Mobile ShoppingChannels.” Information and Management 53: 265–278.doi:10.1016/j.im.2015.10.006.

Lieber, E., and C. Syverson. 2012. “Online versus OfflineCompetition.” In Oxford Handbook of the DigitalEconomy, edited by M. Peitz and J. Waldfogel. Oxford:Oxford University Press.

Lu, Y., S. Yang, P. Chau, and Y. Cao. 2011. “Dynamicsbetween the Trust Transfer Process and Intention to UseMobile Payment Services: A Cross-EnvironmentPerspective.” Information & Management 48: 393–403.doi:10.1016/j.im.2011.09.006.

Maddala, G. S. 1984. Limited-Dependent and QualitativeVariables in Econometrics. New York: CambridgeUniversity Press.

OECD. 2012. Report on Consumer Protection in Online andMobile Payments. Paris: OECD. http://search.oecd.org/officialdocuments/publicdisplaydocumentpdf/?cote=DSTI/CP(2010)22/FINAL&docLanguage=En.

Train, K. 2003. Discrete Choice Models with Simulation.Cambridge: Cambridge University Press.

Velandia, M., R. Rejesus, T. Knight, and B. Sherrick. 2009.“Factors Affecting Farmers’ Utilization of AgriculturalRisk Management Tools: The Case of Crop Insurance,Forward Contracting, and Spreading Sales.” Journal ofAgricultural and Applied Economics 41: 107–123.

Zhang, R., J. Chen, and C. J. Lee. 2013. “Mobile Commerceand Consumer Privacy Concerns.” Journal of ComputerInformation Systems 53: 31–38. doi:10.1080/08874417.2013.11645648.

Zinman, J. 2009. “Debit or Credit?” Journal of Banking &Finance 33: 358–366. doi:10.1016/j.jbankfin.2008.08.009.

APPLIED ECONOMICS 31

Page 13: E-commerce transactions, the installed base of credit ... · review the empirical literature on the growth and ... brand-ing, awareness, ... to the literature in international economics

Appendix

Table A1. Multinomial logit model estimates.Coefficient Std. error Coefficient Std. error

{MTC} = (111} {MTC} = {011}Constant –2.031 0.330 *** –2.554 0.299 ***AGE 0.055 0.008 *** 0.088 0.006 ***INCOME –0.001 0.043 0.070 0.036 *SMART 1.452 0.125 *** 1.022 0.108 ***PRIVACY 0.237 0.143 * 0.367 0.132 ***SECURITY –0.172 0.089 * –0.245 0.089 ***PRIVACY × AGE –0.012 0.003 *** –0.015 0.002 ***

{MTC} = {110} {MTC} = {010}Constant –4.699 1.316 *** –5.204 0.778 ***AGE 0.044 0.032 0.085 0.017 ***INCOME –0.021 0.136 0.094 0.082SMART 0.833 0.474 * 1.052 0.253 ***PRIVACY 0.342 0.588 0.454 0.359SECURITY 0.081 0.258 –0.334 0.178 *PRIVACY × AGE –0.017 0.014 –0.014 0.007 **

{MTC} = {101} {MTC} = {001}Constant –2.649 0.425 *** –2.525 0.298 ***AGE 0.032 0.011 *** 0.066 0.007 ***INCOME 0.064 0.042 0.096 0.032 ***SMART 1.034 0.138 *** 0.544 0.100 ***PRIVACY 0.511 0.165 *** 0.363 0.128 ***SECURITY 0.016 0.101 0.044 0.092PRIVACY × AGE –0.012 0.004 *** –0.008 0.002 ***

{MTC} = {100} {MTC} = {000}Constant –1.115 0.732 Φ(8) = 0 for identificationAGE –0.050 0.024 **INCOME –0.122 0.075SMART 0.974 0.225 ***PRIVACY 0.081 0.300SECURITY 0.050 0.182PRIVACY × AGE 0.003 0.008

Number of observations 4560Log likelihood –7043.49

Standard errors are obtained by delta method. *, ** and ***, respectively,represent that the effect is statistically significant at 10%, 5% and 1% testsize.

32 G. MADDEN ET AL.