exploring trust of mobile applications based on user behaviors: an empirical study
TRANSCRIPT
Exploring trust of mobile applications based on userbehaviors: an empirical studyZheng Yan1,4, Yan Dong2, Valtteri Niemi3, Guoliang Yu2
1The State Key Lab of ISN, Xidian University2Institute of Psychology, Renmin University of China3Department of Mathematics and Statistics, University of Turku4Department of Communications and Networking, Aalto University
Correspondence concerning this article shouldbe addressed to Zheng Yan, The State KeyLaboratory of ISN, Xidian University, Box 119,No. 2 South Taibai Road, 710071 Xi’an, China.E-mail: [email protected]
Author NotesZheng Yan is a computer scientist with aninterest in trust and security; she is a professorin XiDian University, China, and a docent inthe Aalto University, Finland; she received herPhD in Electrical Engineering from HelsinkiUniversity of Technology. Yan Dong is apsychological researcher with an interest inacademic emotions, mental health, andinterpersonal trust; she is a lecturer in theInstitute of Psychology, Renmin University ofChina; she received her PhD in psychology fromChinese Academy of Sciences. Valtteri Niemi isa professor in University of Turku, Finland,doing privacy and security research. Hereceived PhD in Mathematics from University ofTurku in 1989. Guoliang Yu is a psychologistwith an interest in psychology of developmentand education, and personality and socialdevelopment of adolescent; he is a professor inthe Institute of Psychology, Renmin Universityof China; he received his PhD in Psychologyfrom Beijing Normal University.
doi: 10.1111/j.1559-1816.2013.01044.x
Abstract
This paper explores trust of mobile applications based on users’ behaviors. It pro-poses a trust behavior construct through principal component analysis, reliabilityanalysis, and confirmatory factor analysis based on the data collected from a ques-tionnaire survey with more than 1,500 participants. It is indicated that a user’s trustbehavior is composed of three principal constructs: using behavior, reflectionbehavior, and correlation behavior. They are further delineated into 12 measurablesub-constructs and relate to a number of external factors. The data analysis showedthat the questionnaire has positive psychometric properties with respect to con-struct validity and reliability. We also discuss the practical significance and limita-tions of our work toward usable trust management.
Introduction
Mobile device has evolved into an open platform capable ofexecuting various applications. A mobile application is a soft-ware package that can be installed and executed in the mobiledevice, for example, a mobile e-mail client to access e-mailsvia a mobile phone. Generally, mobile applications developedby various vendors can be downloaded for installation.
Whether a mobile application is trustworthy for a user to pur-chase, download, install, consume, or recommend becomes acrucial issue that impacts its final success.
Trust is a multidimensional, multidisciplinary, and multi-faceted concept. The concept of trust has been studied indisciplines ranging from economics to psychology, fromsociology to medicine, and to information and computerscience. Trust has been defined by researchers in many
bs_bs_banner
Journal of Applied Social Psychology 2013, 43, pp. 638–659
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
different ways. Common to many definitions of trust are thenotions of confidence, belief, expectation, dependence, andreliance on the goodness, reliability, integrity, ability, or char-acter of an entity (i.e., trustee) (Yan, 2007). Herein, we definea user’s trust in a mobile application as his/her belief onwhether the application could fulfill a task as expected.The trustworthiness of mobile applications relates to theirdependability, security, and usability (Avizienis, Laprie,Randell, & Landwehr, 2004).
A user’s trust in a mobile application is highly subjectiveand inherently hard to measure (Antifakos, Kern, Schiele, &Schwaninger, 2005). It is built up over time and changes withthe use of the application due to the influence of manyfactors.As it is seen as an internal“state”of the user, there is noway of measuring it directly. However, trust is an importantfactor that impacts usage. In the literature, numerousresearchers have found a positive correlation between trustand use (Lee & Moray, 1992; Muir, 1994; Muir & Moray, 1996;Sheridan, 1980). Trust is important because it helps consum-ers overcome perceptions of uncertainty and risk and engagein“trust-related behaviors”(Lewis & Weigert, 1985); in short,trust behavior. Trust behaviors are a trusting subject’s actionsto depend on or make him/her vulnerable to a trusting object(McKnight, Choudhury, & Kacmar, 2002). Examples areusing the application regularly to fulfill a routine task, or con-tinuing to consume the application even while facing someerrors.After the trusting subject (i.e., trustor) and the trustingobject (i.e., trustee) have interacted for some time, credibleinformation has been gleaned.
One issue that contributes to whether the users would liketo use a new product is how much they trust it. Muir is one ofthe first researchers to look at a decision process betweensupervisors and automated systems. Her study found a posi-tive correlation between trust and use (Muir & Moray, 1996).Lee and Moray (1992) found that trust in a system partiallyexplained system use. McKnight et al. (2002) also proposedthat a consumer’s “subjective probability of depending” (i.e.,the perceived likelihood that one will depend on the other)involves the projected intention to engage in some specificbehaviors.
Marsh (1994) reasoned that it might be more appropriateto model trust behavior rather than trust itself, removing theneed to adhere to specific definitions. Furthermore, modelingtrust behavior overcomes the challenges of measuring a sub-jective concept. Instead, measuring may be done throughobjective behavior observation, which provides a concreteclue of trust. Regarding mobile application usage, we positthat credible information is gained only after a mobile userhas engaged in trust behaviors (e.g., acting on using a mobileapplication) and assessed the trustworthiness of the applica-tion by observing the consequences of its performance andrelying on it in his/her routine life. The goal of our research isto study what interaction behaviors are related to the user’s
trust in a mobile application. We hypothesize that theuser’s trust in a mobile application can be studied throughthe user’s behaviors, which can be monitored via the user–device interaction during the application usage.
However, few existing trust models explore trust in theview of human behaviors (Grabner-Kräuter & Kaluscha,2003; Yan, 2010; Yan & Holtmanns, 2008). Some studies focuson human’s trust in an automatic and intelligent machine(Lee & Moray, 1992; Muir, 1994; Muir & Moray, 1996). Anumber of trust models have been proposed in the context ofe-commerce (Grabner-Kräuter & Kaluscha, 2003; McKnightet al., 2002) while little work has been done in the context ofmobile applications. Numerous researchers have conceptual-ized trust as a behavior which has been validated in thecontext of work collaboration and social communications(Anderson & Narus, 1990; Deutsch, 1973; Fox, 1974). Priorresearch has also confirmed a strong correlation betweenbehavioral intentions and actual behavior, especially for soft-ware system usage (Sheppard, Hartwick, & Warshaw, 1988;Venkatesh & Davis, 2000). However, still very few studiesexamined trust from the view of trust behaviors. Some worksstudy the trust behavior in e-banking (Grabner-Kräuter &Kaluscha, 2003). To our knowledge, no existing work explorestrust behavior of mobile application usage, which is a differ-ent context from the above research domains as regards appli-cation execution environment and user interface, as well asusage experiences.
In the work presented in this paper, we first review the lit-erature of trust model and trust behavior study. Then, wepropose a trust behavior research model based upon theo-retic justification and practical experiences. We continueexploring and confirming the construct of trust behavior formobile application usage through a large-scale user survey.Furthermore, the findings and implications, as well as the sig-nificance and limitations of current empirical study, are dis-cussed. Finally, conclusions and future work are presented inthe last section.
Literature background
Trust model
The method to specify, evaluate, set up, and ensure trust rela-tionships among entities is referred to as a trust model. Trustmodeling is the technical approach used to represent trust.One of the earliest formalizations of trust in computingsystems was done by Marsh (1994). He integrated the variousfacets of trust from the disciplines of economics, psychology,philosophy, and sociology. Since then, many trust modelshave been constructed for various computing paradigmsincluding ubiquitous computing, distributed systems(e.g., peer-to-peer systems, ad hoc networks, GRID virtual
Yan et al. 639
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
organizations), multi-agent systems, Web services, e-commerce (e.g., Internet services), and component software(Yan & Holtmanns, 2008). In almost all of these studies, trustis accepted as a subjective notion, thus bringing us to aproblem of how to measure trust. Translation of this subjec-tive concept into a machine readable language is the mainobjective of trust modeling.
In information and computer science, a trust model aimsto process and/or control trust digitally. Most of the trustmodels are designed based on the understanding of trustcharacteristics (e.g., subjective and dynamic) by consideringthe factors that influence trust. Current work covers a widearea, mostly in distributed systems and e-commerce. Acommon approach that can be found in the literature is com-putational trust, e.g., Song, Hwang, Zhou, and Kwok (2005);Sun, Yu, Han, and Liu (2006); Theodorakopoulos and Baras(2006); and Xiong and Liu (2004). Although a variety of trustmodels have been proposed, it is still not well understoodwhat fundamental criteria trust models must follow. Withouta good answer to this question, the design of trust models isstill at an empirical stage (Sun et al., 2006). Current workfocuses on concrete solutions in specific systems.
One promising approach of trust modeling aims to con-ceptualize trust based on user studies through a psychologi-cal or sociological approach (e.g., using a measurementscale). This kind of research aims to prove the complicatedrelationships among trust and other multiple factors indifferent facets. Two typical examples are the initial trustmodel proposed by McKnight et al. (2002) and the technol-ogy trust formation model (TTFM) studied by Li, Valacich,and Hess (2004). Initial trust refers to trust in an unfamiliartrustee, a relationship in which the actors do not yet havecredible, meaningful information about, or affective bondswith, each other (Bigley & Pearce, 1998). McKnight et al.proposed and validated the measures for a multidisciplinaryand multidimensional model of initial trust in e-commerce(McKnight et al., 2002). The TTFM is a comprehensivemodel of initial trust formation. It is used to explain andpredict people’s trust toward a specific information system(Li et al., 2004). Both models made use of the framework ofthe theory of reasoned action (TRA) created by Fishbeinand Ajzen (1975) to explain how people form initial trust inan unfamiliar entity, and both models integrated importanttrusting antecedents into their frameworks in order to effec-tively predict people’s trust. Since the objective of bothstudies was to predict initial trust (trusting intention) beforeany actual interaction with the trusting object, trust behav-ior was not considered.
In addition to the initial trust, there are short-term trustthat is built up over the first interactions with a system andlong-term trust that is developed with the continuous use of asystem over a longer period of time. Ongoing trust appearedin McKnight et al. (2002) and it consists of the short-term
trust and the long-term trust. In our study, we mainly focuson the ongoing trust evaluation based on the usage behaviorswith regard to mobile applications. In particular, the ongoingtrust could contribute to the trustee’s reputation and thusgreatly help other entities to generate their initial trust.
For other examples of psychometrical studies ontrust, Gefen (2000) proved that familiarity builds trust;Pennington, Wilcox, and Grover (2004) tested that one trustmechanism, vendor guarantees, has direct influence onsystem trust; Bhattacherjee (2002) studied three key dimen-sions of trust: trustee’s ability, benevolence, and integrity;Pavlou and Gefen (2004) explained that institutional mecha-nisms engender a buyer’s trust in the community of onlineauction sellers.
Existing studies have recognized that trust is a multidimen-sional construct and examined the types of trust (e.g.,knowledge-based trust, cognition-based trust, calculative-based trust, institution-based trust, and personality-basedtrust) (Gefen, Karahanna, & Straub, 2003; Wu & Chen, 2005;Zucker, 1986). Most of existing trust studies fall into theareas of organizational settings and electronic commerce(Abrams, Cross, Lesser, & Levin, 2003; Corritore, Kracher, &Wiedenbeck, 2003; Gefen et al., 2003; McAllister, 1995;Parkhe, 1998; Paul & McDaniel, 2004; Ratnasingam, 2005;Zucker, 1986). However, few studies explore trust within thecontext of mobile applications, especially through trustbehavior experiments.
Current trust models have been developed by consideringspecific security issues and also knowledge, experience,practices, and performance history (Daignault & Marche,2002). Much of the prior research in trust of automationhas focused primarily on its psychological aspects (Muir,1994). A number of proposals have been presented to linksome of the psychological aspects of trust with engineeringissues. For example, attempts have been made to map psy-chological aspects of trust (e.g., reliability, dependability,and integrity) to human–computer trust associated withengineering trust issues such as reliability and security(Crocker & Algina, 1986). Lance, Hoffman, Kim, and Blum(2006) studied trust from a number of influencing factorsfrom both the engineering and psychological points of viewand tried to combine these factors in order to provide acomprehensive model.
Quite a number of computational trust models have beendeveloped in the literature of computer science and engineer-ing (Yan, 2010). However, some factors that influence trustare hard to model, e.g., the subjective factor of a user. Forassessing a user’s trust in a computer product, the user’s trustcriteria have to be understood in different contexts. Thisintroduces additional challenges on human–computer inter-action (e.g., usability). Therefore, in order to overcome theabove challenges, it might be more appropriate to model trustbehavior rather than trust itself.
640 Trust can be evaluated based on trust behaviors
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
Trust behavior study
Trust enables people to live in a risky and uncertain situation(Mayer, Davis, & Schoorman, 1995). It provides the means todecrease complexity by reducing the number of options onehas to consider in a given situation (Barber, 1983; Lewis &Weigert, 1985; Luhmann, 1979). Trust is the key to decisionmaking. Research in ergonomics has examined how trust isestablished, maintained, lost, and regained in human–computer systems (Corritore et al., 2003). Trust has been seenas an intervening variable that mediates the users’ behaviorswith computers (Muir, 1994).
McKnight et al. (2002) proposed that consumer subjectiveprobability of depending involves the projected intention toengage in three specific risky behaviors—provide the vendorpersonal information, engage in a purchase transaction, oract on vendor information (e.g., financial advice). Followingthis theory, a mobile user gains experience with a specificmobile application, and this experience may be the dominantinfluence on trusting beliefs, instead of dispositional orinstitution-based trust. In addition, perceived mobile appli-cation quality should relate positively to both trusting beliefsand trusting intentions. McKnight et al. (2002) also pointedout a number of opportunities for future research. Some ofthem relate directly to overcoming the limitations of theirstudies. One particular suggestion was to conduct a study inwhich the ultimate outcome of interest—trust behavior—isdirectly measured. They identified a handful of commonlydiscussed trust-related behaviors in e-commerce—there maybe others. Because of the difficulty of asking subjects toundertake such behavior, they did not measure actual behav-ior in their study. Note that trust and trust behavior are differ-ent concepts: Trust is the willingness to assume risk; trustbehavior is the assuming of risk (Mayer et al., 1995, p. 724).Whether or not the trustor will take a specific risk is influ-enced by trusting beliefs, trusting intentions, and the per-ceived risk of the trust behavior (Grabner-Kräuter &Kaluscha, 2003).
The TRA explains how people form initial trust in an unfa-miliar entity (Fishbein & Ajzen, 1975). It posits that beliefslead to attitudes, which lead to behavioral intentions, whichlead to the behavior itself (Fishbein & Ajzen, 1975). Applyingthis theory, we propose that trusting beliefs (perceptions ofspecific mobile application attributes) lead to trusting inten-tions (intention to engage in using a mobile applicationthrough user–computer interaction), which in turn result intrust behaviors (trying to use and continuously using theapplication in various contexts). Additionally, numerousresearchers have conceptualized trust as a behavior. Valida-tion has been done in the context of work collaboration andsocial communications (Anderson & Narus, 1990; Deutsch,1973; Fox, 1974), software system usage (Sheppard et al.,1988; Venkatesh & Davis, 2000), and e-banking and economy
(Grabner-Kräuter & Kaluscha, 2003). However, no priorwork explores trust behavior of mobile application usagebased on our current knowledge.
The trust and technology acceptance model (Gefen et al.,2003) has been well studied in online shopping and the resultsshowed that understanding both the Internet technology andissues around trust is important in determining behavioralintention to use (Gefen et al., 2003; Wu & Chen, 2005). Thismodel placed the use of online system into two system fea-tures: ease of use and usefulness (related to importance andurgency) (Davis, 1989) as well as trust in e-vendors. The resultindicated that these variables are good predictors for behaviorintention to use online shopping (Wu & Chen, 2005). Wecould try to further extend the above conclusions to thecontext of mobile applications.
One important issue that contributes to whether the userswould like to use a new product or system (e.g., a mobileapplication) is how much they trust it. Muir is one of the firstresearchers to look at a decision process between supervisorsand automated systems. She verified the hypothesis proposedby Sheridan et al. that the supervisor’s intervention behavioris based upon his/her trust in automation (Sheridan, 1980). Apositive correlation between trust and use has been proved(Lee & Moray, 1994; Muir, 1994; Muir & Moray, 1996). Leeand Moray (1992) found that trust in a system partiallyexplains system use, but other factors (such as the user’s ownability to provide manual control) also influence the systemuse. Muir and Moray (1996) further argued that trust in auto-mated machines is based mostly on users’ perceptions on theexpertise of the machine, i.e., the extent to which the automa-tion performs its function properly. The relationship betweentrust and interaction behavior is obvious. Trust has a directinfluence on behavioral intention to use (Bandura, 1986;Davis, 1989). Wu and Chen (2005) also pointed out that trustis apparently an important antecedent of attitude toward theonline transaction behavior. All above studies lay the founda-tion for our hypothesis: A user’s trust in a mobile applicationcan be evaluated based on the user’s trust behaviors.
Many studies have empirically examined the determinantsof continued usage (continuance) (Hsu, Yen, Chiu, & Chang,2006). These studies provided preliminary evidence that con-tinued usage behaviors are determined by different factors.Most studies aimed to examine the change of users’ cognitivebeliefs and attitudes from pre-usage stage to usage stage andhow they influence users’ intentions to continue their usage,especially within the context of online shopping. Forexample, expectancy disconfirmation theory (Oliver, 1980),which is a consumer behavior model, explained and pre-dicted consumer satisfaction and repurchase intentions.Based on this theory, satisfied consumers form intentions toreuse a product or service in the future, while dissatisfiedusers discontinue subsequent use. Users would evaluatewhether their initial cognitions are consonant or dissonant
Yan et al. 641
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
with actual experience (Bhattacherjee & Premkumar, 2004).In our study, we examine the relationship between continu-ance and trust since trust is obviously relevant to use.Continuance implies users’ ongoing trust based on actualexperiences.
Grabner-Kräuter and Kaluscha (2003) reviewed 11 trustconstruct studies. Trusting intentions and their antecedentsare examined in several studies (in Bhattacherjee, 2002;Gefen, 2000; Gefen & Straub, 2003; Jarvenpaa, Tractinsky, &Saarinen, 1999; Jarvenpaa, Tractinsky, & Vitale, 2000)whereas trust-related behaviors are only investigated in twostudies (in Kim & Prabhakar, 2004; Pavlou, 2003). In order toinvestigate trust-related behavior as a consequent of trust, theformer has to be measured in terms of actual behavior, notwillingness to engage in behavior (Mayer et al., 1995). Insome of the past studies (e.g., Jarvenpaa et al., 1999, 2000;Pavlou, 2003; de Ruyter et al., 2001), company reputationwas discussed and included, but none of the 11 studiesexplicitly investigated the interrelationship between trustand branding. Another largely unexplored factor is productquality. The interdependencies between consumers’ trust andproduct performance could be a basis for upcoming research(Shankar, Urban, & Sultan, 2002). Varying the trustor ortrustee or context may result in different sets of antecedentand consequence trust.All of the above offer additional direc-tions for our research.
Existing studies focus on human’s trust in an automationand intelligent machine. A number of trust models have beenproposed in the context of e-commerce. Little work has beendone in the context of mobile applications although the sig-nificance of such study is obvious. Prior arts also lackstudy on the influence of recommendations, personality, andusage context with regard to human–computer trust. Withthe rapid development of mobile computing technology, amobile device has become a multi-application system formultipurpose and multi-usage. It is an open platform thatallows deploying new or upgraded applications at anytimeand anywhere through a network connection. Therefore,such a dynamically changed system, different from the exist-ing research domains, introduces new challenges for human–computer trust. We believe that research should go into depthin the newly thrived domain of mobile applications.
Research model andscale development
We applied a psychometric method to examine our hypoth-esis, i.e., the user’s trust in a mobile application can be studiedthrough the user’s behaviors, which can be monitored via theuser–device interaction during the application usage. Wedesigned a questionnaire (i.e., measures or a measurementscale), taking short message service (SMS) as a concreteexample of mobile application. Each item in the question-
naire is a statement for which the participants need to indi-cate their levels of agreement. The questionnaire is anchoredusing a 7-point Likert scale ranging from “strongly disagree”to “strongly agree.” First, a preexperiment with 318 partici-pants was conducted in order to optimize our questionnaire(Yan, Niemi, Dong, & Yu, 2008). Further, we ran a formalexperiment with more than 1,500 participants to explore andevaluate the trust behavior construct for mobile applications.
Because there are scarce theory and evidence about whatusers’ trust behaviors are regarding mobile applications, westarted our study in an exploratory manner. Thus, we soughtfirst to generate a measurement scale that would define thenumber and the nature of the dimensions that underlie theusers’ perceptions of trust behaviors. For this purpose, weused exploratory factor analysis in the preexperiment study(Yan et al., 2008). Exploratory factor analysis is particularlysuitable to identify “a set of latent constructs underlyinga battery of measured variables” (Fabrigar, Wegener,MacCallum, & Strahan, 1999, p. 275). Once we obtained sucha set of latent constructs, we conducted the formal experi-ment study that used both exploratory factor analysis andconfirmatory factor analysis (CFA) to refine and test themodel that emerged from our preceding study (Anderson &Gerbing, 1988; Gerbing & Anderson, 1988; MacCallum &Austin, 2000).
From our preexperiment study, we found that three con-structs play significant roles as direct determinants of trustbehavior (Yan et al., 2008).We call them using behavior (UB),reflection behavior (RB), and correlation behavior (CB).They comprise the user’s trust behavior and contribute to thecalculation of the estimate (from the device point of view) onthe user’s trust in the mobile application. Meanwhile, weposit that trust behavior is also related to or impacted by anumber of external factors that could play as the basis of theuser’s initial or potential trust. The labels used for the con-structs describe the essence of the construct and are meant tobe independent of any particular theoretical perspective. Inthe remainder of this section, we define each of the determi-nants, specify the role of external factors, and provide thetheoretical justification for the hypotheses. Figure 1 presentsour research model. The definitions of constructs and meas-urement scale items are described in Table 1.
Personal
motivation
Brand impact
Perceived quality
Personality
Using behavior
Reflection behavior
Correlation
behavior
Trust
Behavior
Figure 1 Research model.
642 Trust can be evaluated based on trust behaviors
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
Tab
le1
Con
stru
cts,
Defi
nitio
ns,a
ndSc
ales
Con
stru
ctD
efini
tion
Item
s
UB:
usin
gbe
havi
or(r
ootc
onst
ruct
)Th
ese
tofb
ehav
iors
abou
tnor
mal
appl
icat
ion
usag
eU
B1:n
orm
alus
age
beha
vior
(sub
-con
stru
ct)
The
elap
sed
usag
etim
e,th
enu
mbe
rofu
sage
,and
usag
efr
eque
ncy
1.Th
em
ore
times
you
use
the
mes
sagi
ng,t
hem
ore
you
trus
tit.
2.Th
em
ore
freq
uent
lyyo
uus
eth
em
essa
ging
,the
mor
eyo
une
edit.
3.Th
elo
nger
time
you
use
the
mes
sagi
ng,t
hem
ore
you
trus
tit.
UB2
:beh
avio
rrel
ated
toco
ntex
t(s
ub-c
onst
ruct
)Th
epe
rspe
ctiv
eth
atth
ech
arac
teris
tics
ofus
age
cont
ext
such
asus
age
risk,
impo
rtan
ce,a
ndur
genc
yco
uld
influ
ence
the
trus
tbeh
avio
r
1.Yo
udo
mor
eim
port
antt
asks
thro
ugh
the
mes
sagi
ngif
you
trus
titm
ore.
2.Yo
udo
mor
eris
kyta
sks
thro
ugh
the
mes
sagi
ngif
you
trus
titm
ore
(e.g
.,SM
Spa
ymen
t).
3.Yo
udo
mor
eur
gent
task
sth
roug
hth
em
essa
ging
ifyo
utr
usti
tmor
e.U
B3:f
eatu
re-r
elat
edus
age
beha
vior
(sub
-con
stru
ct)
The
pers
pect
ive
that
the
appl
icat
ion
feat
ures
expe
rienc
edby
aus
erim
ply
the
profi
cien
cyof
the
user
inth
eap
plic
atio
nus
age
1.Yo
uw
ould
try
mor
efe
atur
esof
the
mes
sagi
ngif
you
trus
titm
ore.
2.A
fter
tryi
ngm
ore
feat
ures
ofth
em
essa
ging
,you
gain
mor
eex
pert
ise
onit.
3.G
ood
qual
ityof
the
mes
sagi
ngw
ould
enco
urag
eyo
uto
try
new
feat
ures
ofit.
RB:r
eflec
tion
beha
vior
(roo
tco
nstr
uct)
The
usag
ebe
havi
ors
afte
rthe
user
conf
ront
sap
plic
atio
npr
oble
ms/
erro
rsor
has
good
/bad
usag
eex
perie
nces
RB1:
bad
perf
orm
ance
refle
ctio
nbe
havi
or(s
ub-c
onst
ruct
)H
owth
eba
dpe
rfor
man
ceof
anap
plic
atio
nin
fluen
ces
aus
er’s
trus
tin
it1.
You
coul
dde
crea
seth
etim
esof
usin
gth
em
essa
ging
due
toits
bad
perf
orm
ance
.2.
Your
usag
ein
tere
stan
dus
age
freq
uenc
yco
uld
bede
crea
sed
due
toth
eba
dm
essa
ging
perf
orm
ance
.3.
You
coul
dde
crea
seth
etim
eof
usin
gth
em
essa
ging
due
toits
bad
perf
orm
ance
.RB
2:ba
dpe
rfor
man
cere
flect
ion
beha
vior
rela
ted
toco
ntex
t(s
ub-c
onst
ruct
)
How
the
bad
perf
orm
ance
ofan
appl
icat
ion
influ
ence
sa
user
’str
usti
ndi
ffer
entc
onte
xts
1.Ba
dpe
rfor
man
ceof
the
mes
sagi
ngco
uld
disc
oura
geyo
uto
doim
port
antt
hing
sw
ithit.
2.Ba
dpe
rfor
man
ceof
the
mes
sagi
ngco
uld
disc
oura
geyo
uto
dohi
ghly
risky
thin
gsw
ithit.
3.Ba
dpe
rfor
man
ceof
the
mes
sagi
ngco
uld
disc
oura
geyo
uto
dour
gent
thin
gsw
ithit.
RB3:
good
perf
orm
ance
refle
ctio
nbe
havi
or(s
ub-c
onst
ruct
)H
owth
ego
odpe
rfor
man
ceof
anap
plic
atio
nin
fluen
ces
aus
er’s
trus
tin
it1.
You
coul
din
crea
seth
etim
eof
usin
gth
em
essa
ging
due
toits
good
perf
orm
ance
.2.
You
coul
din
crea
seth
etim
eof
usin
gth
em
essa
ging
due
toits
good
perf
orm
ance
.3.
Your
usag
ein
tere
stan
dus
age
freq
uenc
yco
uld
bein
crea
sed
due
toth
ego
odm
essa
ging
perf
orm
ance
.RB
4:go
odpe
rfor
man
cere
flect
ion
beha
vior
rela
ted
toco
ntex
t(s
ub-c
onst
ruct
)
How
the
good
perf
orm
ance
ofan
appl
icat
ion
influ
ence
sa
user
’str
usti
ndi
ffer
entc
onte
xts
1.G
ood
perf
orm
ance
ofth
em
essa
ging
coul
den
cour
age
you
todo
high
lyris
kyth
ings
with
it.2.
Goo
dpe
rfor
man
ceof
the
mes
sagi
ngco
uld
enco
urag
eyo
uto
doim
port
antt
hing
sw
ithit.
3.G
ood
perf
orm
ance
ofth
em
essa
ging
coul
den
cour
age
you
todo
urge
ntth
ings
with
it.RB
5:ba
dex
perie
nce
refle
ctio
nto
cont
ext(
sub-
cons
truc
t)H
owth
eba
dus
age
expe
rienc
eof
anap
plic
atio
nin
fluen
ces
aus
er’s
trus
tin
diff
eren
tcon
text
s1.
Aft
erve
ryba
dex
perie
nces
ofus
ing
the
mes
sagi
ng,y
ouco
uld
use
itto
dole
ssris
kyta
sk.
2.A
fter
very
bad
expe
rienc
esof
usin
gth
em
essa
ging
,you
coul
dus
eit
todo
less
impo
rtan
tta
sk.
3.A
fter
very
bad
expe
rienc
esof
usin
gth
em
essa
ging
,you
coul
dus
eit
todo
less
urge
ntta
sk.
RB6:
good
expe
rienc
ere
flect
ion
toco
ntex
t(su
b-co
nstr
uct)
How
the
good
usag
eex
perie
nce
ofan
appl
icat
ion
influ
ence
sa
user
’str
usti
ndi
ffer
entc
onte
xts
1.A
fter
very
good
expe
rienc
esof
usin
gth
em
essa
ging
,you
coul
dus
eit
todo
mor
eris
kyta
sks.
2.A
fter
very
good
expe
rienc
esof
usin
gth
em
essa
ging
,you
coul
dus
eit
todo
mor
eim
port
ant
task
s.3.
Aft
erve
rygo
odex
perie
nces
ofus
ing
the
mes
sagi
ng,y
ouco
uld
use
itto
dom
ore
urge
ntta
sks.
Yan et al. 643
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
Tab
le1
(Con
tinue
d)
Con
stru
ctD
efini
tion
Item
s
CB:
corr
elat
ion
beha
vior
(roo
tco
nstr
uct)
The
usag
ebe
havi
ors
corr
elat
edto
mob
ileap
plic
atio
nsw
ithsi
mila
rfun
ctio
nalit
ies
CB1
:com
paris
onof
norm
alus
age
beha
vior
(sub
-con
stru
ct)
How
the
diff
eren
tusa
gebe
havi
ors
ofsi
mila
rlyfu
nctio
ned
appl
icat
ions
indi
cate
trus
t1.
Usi
ngth
em
essa
ging
mor
etim
esth
anan
othe
rsim
ilarly
func
tione
dm
obile
appl
icat
ion
mea
nsyo
utr
usti
tmor
e.2.
Usi
ngth
em
essa
ging
mor
efr
eque
ntly
than
anot
hers
imila
rlyfu
nctio
ned
mob
ileap
plic
atio
nm
eans
you
trus
titm
ore.
3.Sp
endi
ngm
ore
time
inus
ing
the
mes
sagi
ngth
anan
othe
rsim
ilarly
func
tione
dm
obile
appl
icat
ion
mea
nsyo
utr
usti
tmor
e.C
B2:c
ompa
rison
rela
ted
toco
ntex
t(s
ub-c
onst
ruct
)H
owth
edi
ffer
entu
sage
deci
sion
sar
ere
late
dto
trus
tin
vario
usco
ntex
tsw
ithre
gard
tosi
mila
rlyfu
nctio
ned
appl
icat
ions
1.U
sing
the
mes
sagi
ng,n
otan
othe
rsim
ilarly
func
tione
dm
obile
appl
icat
ion,
tofu
lfill
am
ore
impo
rtan
ttas
km
eans
you
trus
titm
ore.
2.U
sing
the
mes
sagi
ng,n
otan
othe
rsim
ilarly
func
tione
dm
obile
appl
icat
ion,
tofu
lfill
am
ore
risky
task
mea
nsyo
utr
usti
tmor
e.3.
Usi
ngth
em
essa
ging
,not
anot
hers
imila
rlyfu
nctio
ned
mob
ileap
plic
atio
n,to
fulfi
lla
mor
eur
gent
task
mea
nsyo
utr
usti
tmor
e.C
B3:r
ecom
men
datio
nbe
havi
or(s
ub-c
onst
ruct
)Th
epe
rspe
ctiv
eth
atpo
sitiv
ere
com
men
datio
nbe
havi
orim
plie
sth
eus
er’s
trus
t1.
Ifyo
uha
veve
rygo
odex
perie
nces
inus
ing
the
mes
sagi
ng,y
ouge
nera
llyw
ould
like
tore
com
men
dit.
2.Fo
rtw
osi
mila
rlyfu
nctio
ned
mes
sagi
ngap
plic
atio
ns,y
outr
ustm
ore
inth
eon
eyo
uw
ould
like
tore
com
men
d.3.
Aft
erve
ryba
dex
perie
nces
inus
ing
the
mes
sagi
ng,y
ouge
nera
llydo
n’tw
antt
ore
com
men
dit.
Exte
rnal
fact
ors
The
fact
ors
orva
riabl
esth
atin
fluen
cetr
ustb
ehav
iors
BI:b
rand
impa
ctTh
epe
rspe
ctiv
eth
attr
ustb
ehav
iori
sin
fluen
ced
bybr
and
and
repu
tatio
ns1.
You
like
am
obile
appl
icat
ion
deve
lope
dby
afa
mou
sve
ndor
.2.
You
like
usin
ga
mob
ileph
one
with
afa
mou
sbr
and.
3.Yo
uw
ould
like
tore
com
men
da
mob
ileap
plic
atio
nde
velo
ped
bya
fam
ous
vend
or.
4.Fo
ram
obile
appl
icat
ion
deve
lope
dby
anin
fam
ous
vend
or,y
ouw
illca
utio
usly
use
itor
stop
usin
git
ifyo
um
eets
ome
prob
lem
s.PM
:per
sona
lmot
ivat
ion
The
pers
pect
ive
that
the
user
’sge
nera
lpro
pens
ityto
trus
tis
one
ofim
port
ante
xter
nalf
acto
rsre
late
dto
trus
tbeh
avio
rs
1.A
mob
ileph
one
isve
ryim
port
antt
oyo
urlif
e.2.
You
ofte
nus
ea
mob
ileph
one
totr
ansf
erim
port
anti
nfor
mat
ion.
3.Be
caus
ea
mob
ileph
one
isim
port
antt
oyo
u,yo
uw
illco
ntin
uous
lyus
eit
insp
iteof
itsso
ftw
are
havi
ngso
me
prob
lem
s.P:
pers
onal
ityTh
epe
rspe
ctiv
eth
atpe
rson
ality
ison
eof
the
impo
rtan
tfa
ctor
sth
atm
ayaf
fect
trus
tbeh
avio
rs1.
Inyo
urop
inio
n,m
obile
appl
icat
ions
have
apr
omis
ing
futu
re.
2.Yo
ube
lieve
the
mob
ileap
plic
atio
nw
illbe
cont
inuo
usly
impr
ovin
gan
dup
grad
ing.
3.Yo
uha
veco
nfide
nce
onth
efu
ture
ofm
obile
phon
eus
age.
4.Yo
uth
ink
the
mob
ileph
one
isa
pers
onal
com
mun
icat
orlik
edby
mos
tpeo
ple.
5.Yo
uth
ink
the
mob
ileap
plic
atio
nsbe
nefit
your
life
and
stud
yve
rym
uch.
DQ
:per
ceiv
edde
vice
qual
ityTh
epe
rspe
ctiv
eth
atth
equ
ality
ofus
edm
obile
devi
ces
coul
dbe
ane
cess
ary
(but
nots
uffic
ient
)con
ditio
nfo
rm
obile
appl
icat
ion
trus
t
1.G
ener
ally
spea
king
,the
phon
eyo
uar
eus
ing
isw
orki
ngw
ell.
2.U
sing
am
obile
phon
efo
ryou
isea
syan
dco
nven
ient
.3.
Usi
nga
mob
ileph
one
isec
onom
ical
and
usef
ulfo
ryou
.4.
Usi
nga
mob
ileph
one
can
satis
fyyo
urpe
rson
alde
man
dsw
ell.
SMS
=sh
ortm
essa
gese
rvic
e.
644 Trust can be evaluated based on trust behaviors
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
Using behavior
UB is defined as the set of behaviors about normal applica-tion usage. Trust has tight correlation with use (Muir &Moray, 1996). Trust in a system partially explains system use(Lee & Moray, 1992). Using an application is a behaviorimplying trust in it (McKnight et al., 2002), thus it is highlyrelated to trust behavior.We posit that trust could be reflectedby normal usage behavior (UB1), which is indicated byelapsed usage time, the number of usage, and usage fre-quency. Meanwhile, characteristics of the usage contextsuch as risk, importance, and urgency could also influencethe trust behavior (UB2) (Grabner-Kräuter & Kaluscha,2003). McKnight et al. proposed some high-risk andhigh-importance task-related behaviors as trust behaviors(McKnight et al., 2002). Herein, we try to examine trust couldleverage consumers to overcome perceptions of uncertaintyand risk and engage in “trust-related behaviors.” In addition,the ability of a user to use a mobile application also influencesthe usage (Lee & Moray, 1992). Generally, a mobile applica-tion provides a number of functionalities, i.e., features. Themore features experienced by the user, the more proficienthe/she is in the application usage (UB3). In summary, wepropose that:
Hypothesis 1a. Trust in a mobile application will havea positive influence on the user’s usage reflectedby elapsed usage time and number and frequency ofusages.
Hypothesis 1b. Trust in a mobile application will have apositive influence on the user’s behavior regardingrisky, urgent, or important tasks.
Hypothesis 1c. Experiences on the features of a mobileapplication will have a positive influence on the user’sproficiency in using the application.
Reflection behavior
RB consists of usage behaviors after the user confrontsapplication problems/errors or has good/bad usage experi-ences. Notably, the difference of the RB and the UB lies inthe fact that the first one is related to the type of eventwhereas the second one is related to general usage statistics.Their contributions to trust evaluation could be different.Some researchers have found the strong effect of computererrors or performance on trust (Corritore et al., 2003;Kantowitz, Hankowski, & Kantowitz, 1997; Lee & Moray,1992; Muir, 1994; Muir & Moray, 1996). Empirical studies oftrust in automated machines show that performance andtrust increase following a similar learning curve as long asthere are no errors (Lee & Moray, 1992). But machine errorshave a strong effect on trust. The magnitude of an error isan important factor in loss of trust (Lee & Moray, 1992;
Muir & Moray, 1996). On the other hand, trust couldrecover even when small errors continue, if the user is ableto understand and compensate for the errors; but trust maynot be restored to its level prior to the series of errors (Lee &Moray, 1992; Muir & Moray, 1996). Errors encountered inone function of a system can lead to distrust of related func-tions, but do not necessarily generalize to the entire system(Muir & Moray, 1996). However, even in the face of compu-ter errors, a user may continue to trust a computer system incertain situations, for example, if workload is high or if theerrors are predictable (Muir & Moray, 1996). From the lit-erature, we can see that the RB can imply the user’s trustand continuous trust in an application, especially in thecontext when he/she is facing an application problem. Thus,it is one of important constructs of trust behavior. Herein,we propose that:
Hypothesis 2a. Good/bad performance of a mobileapplication will have a positive/negative influence onthe user’s trust in the application.
Hypothesis 2b. Good/bad application performance orusage experience will positively/negatively influencethe user’s behavior related to risky, urgent, or impor-tant tasks.
We designed a number of items (as shown in Table 1) toexamine bad and good application performance’s influenceon usage (RB1, RB3) and context (e.g., importance, risk, andurgency), related usage decision (RB2, RB4), as well as theimpact of user experience on contextual usage decision(RB5, RB6).
Correlation behavior
Future mobile market would probably be very competitive. Anumber of similarly functioned mobile applications devel-oped by different vendors would be available at the same timefor consumption. CB concerns the usage behaviors correlatedto similarly functioned mobile applications. Since trust isobviously related to use (Lee & Moray, 1992; Muir, 1994;Muir & Moray, 1996), the usage implies trust. The usagebehavior differences with regard to similar applications implythe user’s different trust in them (Yan et al., 2008). Thereby,the CB is also one important construct of trust behavior.Herein, we propose that for two similarly functioned applica-tions, higher usage rate (i.e., elapsed usage time and fre-quency, the number of usages) means more trust. Meanwhile,trust is also influenced by various contexts (Grabner-Kräuter& Kaluscha, 2003; Yan, 2010; Yan & Holtmanns, 2008). There-fore, we expect that for two similarly functioned applications,the user would like to use more trustworthy one to do risky,urgent, or important tasks. We designed a number of items(as shown in Table 1) about similarly functioned applications
Yan et al. 645
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
to examine the correlation of different usage behaviors totrust (CB1) and the relation of different usage decisions totrust in various contexts (CB2).
Hypothesis 3a. For two similarly functioned applica-tions, higher usage rate (i.e., elapsed usage time andfrequency, the number of usages) implies more trust.
Hypothesis 3b. For two similarly functioned applica-tions, the user would like to use more trustworthy oneto do risky, urgent, or important tasks.
In addition, we posit that a positive recommendation (abehavior to suggest other people using a mobile application)also implies trust (CB3). Current literature has widely studiedrecommendation’s impact on trust (Koh & Sundar, 2010;Wang & Benbasat, 2005). We would like to argue that positiverecommendation behavior indicates a recommender’s trustin a recommended entity. For example, a person who highlyrecommends his/her friend to install and use an applicationindicates his/her favorite and trust in it (Yan et al., 2008). Thisbehavior can be observed by the mobile device since quite anumber of mobile applications provide a possibility to share(and, hence, recommend) via SMS, or multimedia service, orshort-range connection (e.g., Bluetooth or infrared), a linkfor an application installation with other mobile devices.Thus, we propose that:
Hypothesis 3c. Positive recommendation behaviorimplies the user’s trust.
External factors
Apart from the trust behavior exploration, we also designeda number of items in order to do external nomological vali-dation. It is widely accepted that trust is influenced by brandand reputations (Corritore et al., 2003; Grabner-Kräuter &Kaluscha, 2003; Yan, 2010). Trustor’s general propensity totrust is one of important external factors related to trust(Corritore et al., 2003). Personality is one of the importantfactors that may affect usage decision and usage behavior.Different personalities attribute different importance levelsto each of the accepted trust cues (such as branding andprofessional user interface design, etc.) Extroversion andopenness to experience lead to a higher disposition to trust.However, neuroticism and conscientiousness lead to a lowerdisposition to trust (Lumsden & MacKay, 2006). On theother hand, the quality of used mobile devices could be anecessary (but not sufficient) condition for mobile applica-tion trust (Kim & Prabhakar, 2004; Pavlou, 2003). Hence,we attempt to study the influence of the following fourexternal variables on the user’s trust behavior: (a) brandimpact (BI); (b) personal motivation (PM); (c) personality(P); and (d) perceived device quality (DQ). We proposethat:
Hypothesis 4a. The brand of mobile application willinfluence the user’s trust behavior.
Hypothesis 4b. The PM will influence the user’s trustbehavior.
Hypothesis 4c. The user’s personality will influence theuser’s trust behaviors.
Hypothesis 4d. The perceived DQ will be a necessarycondition for application trust.
As shown in Table 1, most items about the external factorsare adapted from prior related research conducted in thefields of e-commerce, and are modified to fit the mobileapplication context. The items on PMs and perceived DQare designed based on the definitions in McKnight et al.(2002). The items about BI and P are designed on the basisof the theoretic results achieved in Corritore et al. (2003);Grabner-Kräuter and Kaluscha (2003); and Yan andHoltmanns (2008).
Analysis and results
Data collection
The experiment was conducted by three psychologists inBeijing, China. The participants were chosen from BeijingJiao Tong University, Renmin University of China, andBeijing International Studies University. The questionnairewas provided to the participating undergraduates enrolled inpsychology lectures. At the beginning, the participants werearranged to answer the questionnaires in a group. Then,the questionnaires were collected and each participant wasoffered a small gift. Almost all participants had past experi-ences of answering a questionnaire survey. They were familiarwith the basic rules of such a survey. Meanwhile, the conduc-tors explained the basic concepts in the questionnaire beforethe survey was carried out. It took the participants about 20minutes to answer the questionnaire.
Totally, 1,575 subjects participated; 1,120 responses(71.1%) were valuable and usable based on three selectioncriteria: (1) no missed item response; (2) no regular patterncan be found from the responses; and (3) the responses on allitems are not the same (i.e., serious response). Among theselected subjects with valid responses, 671 (59.9%) werewomen and 449 (40.1%) were men; 43 participants werebelow 18 years and others were between 19 and 35 years.There were 502 (44.8%) participants who majored in scienceor technology, while 480 (42.9%) in arts. Except for onesample where information was missing, the rest majored inhumanities and social science. The personality types of theparticipants were presented in Table 2 based on the partici-pants’ personal specifications. About half of the participantshad sanguine personality type. One fourth of the participants
646 Trust can be evaluated based on trust behaviors
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
were phlegmatic facing different issues. Thus, most of partici-pants had positive personality types.
Table 3 provides the information about the participants’experiences on the usage of mobile phones and SMS.According to the survey, 419 (37.4%) participants hadexperiences of using the Internet-accessed applications (e.g.,a Web browser), 864 (77.1%) had experiences of usingmobile network-accessed applications (e.g., SMS), and 796(71.1%) had that of non-network-accessed applications(e.g., Profile). Most of the participants (87.9%) used mobile
phones more than half an hour and 62.1% more than 1hour/day. This indicates that mobile phone usage was verycommon and popular in Chinese universities, and Chineseundergraduates and graduates are typical samples of mobileusers. In addition, SMS was regularly and frequently used byChinese university students. Altogether, 71.4% of partici-pants sent or received SMS more than 10 times per day. Thisimplies that adopting SMS as an example of mobile applica-tion in our experiment was appropriate and easy to be fol-lowed by the participants.
Table 4 provides the information about the general experi-ences of weekly usage of some specific mobile applications.We observed that most of Chinese university students hadexperiences on different kinds of mobile applications, amongwhich chatting with friends and mobile communicationswere the most popular ones, whereas mobile advertisementand mobile commerce were not popular at the time of ourexperiment. The statistical data indicated that the mobilephone mainly served as a communication tool for Chineseuniversity students. Thereby, it was suitable to examine thetrust behavior via SMS in our experiment.
Table 2 Personality Type of Participants Based on Direct Indication
Personality type Frequency Percent
Sanguine 506 45.2Melancholy 91 8.1Choleric 76 6.8Phlegmatic 282 25.2Others 142 12.7Missing 23 2.1Total 1,120 100
Table 3 Experience on Mobile Phone and SMS Usage
The experience on mobile phone and SMS No. of subjects Percent
Mobile phone usage time Below 0.5 hour/day 131 11.70.5–1 hour/day 289 25.81–5 hours/day 335 29.9More than 5 hours/day 361 32.2Missing 4 .3Total 1,120 100
SMS usage times Below 10 times/day 320 28.610–30 times/day 581 51.930–50 times/day 144 12.9More than 50 times/day 74 6.6Missing 1 .1Total 1,120 100.0
SMS = short message service.
Table 4 General Experience on Mobile Applications
General experience
Chatting with friends Mobile communicationsReceiving serviceinformation Shopping and business
Frequency Percent Frequency Percent Frequency Percent Frequency Percent
Times Uncertain 193 17.2 173 15.4 159 14.2 347 31.0<5 times/week 257 22.9 238 21.3 418 37.3 695 62.15–10 times/week 209 18.7 204 18.2 258 23.0 24 2.111–15 times/week 88 7.9 109 9.7 116 10.4 12 1.116–20 times/week 44 3.9 50 4.5 28 2.5 1 .121–25 times/week 37 3.3 45 4.0 21 1.9 1 .1>25 times/week 280 25.0 294 26.3 105 9.4 10 .9Missing 12 1 7 .6 15 1. 30 2.Total 1,120 100 1,120 100 1,120 100 1,120 100
Yan et al. 647
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
Data processing and analysis
Data analysis took place in three phases, as described below.In the process, we analyzed three types of validity: conver-gent, discriminant, and nomological. Together, these types ofvalidity constitute construct validity, or “the extent to whichan operationalization measures the concept it is supposed tomeasure” (Bagozzi, Yi, & Phillips, 1991, p. 421). Convergentvalidity (CV) means the extent to which the measures for avariable act as if they are measuring the underlying theoreti-cal construct because they share variance (Schwab, 1980).Internal consistency reliability is generally considered a nec-essary, but not sufficient, condition for CV (Schwab, 1980).Discriminant validity (DV) investigates the degree to whichmeasures of two constructs are empirically distinct (Bagozziet al., 1991; Davis, 1989). Nomological validity refers towhether the construct performs as expected within its nomo-logical network (Schwab, 1980), such as relating to other con-structs as theory suggests (Boudreau, Gefen, & Straub, 2001;Webster & Martocchio, 1992).
We leveraged large sample size to provide better confi-dence in the results. The samples were randomly dividedinto two approximately equal parts. One part (n = 567) wasused for principal component analysis (PCA), which is oneimportant method of exploratory factor analysis, while theremaining samples (n = 553) were used for CFA. Mean-while, we also conducted correlation analysis and reliabilityanalysis.
Phase 1: PCA
Because some items were added and revised according to theresults of the preexperiment (Yan et al., 2008), in the firstphase, exploratory, principal components factor analysis andinternal consistency reliability analysis were conducted todetermine the extent to which trust constructs were discrimi-nant (using SPSS v11.5, SPSS Inc., Chicago, IL). The purposeof using PCA was to cull out the items that did not load on theappropriate high-level construct and extract principal factorsfor making a predictive model. Kaiser’s criterion was appliedin the PCA, which considers factors with an eigenvaluegreater than 1 as common factors (Nunnally, 1978). Inaddition, the items’ loadings should be greater than .4, andno cross-loading above .4 on the “wrong” trust construct(Boudreau et al., 2001; Hair, Anderson, Tatham, & Black,1998). If the item had a cross-loading above .4, and droppingthe item did not improve the reliability of its sub-constructs,the item was retained in its main sub-construct.
PCA was performed using an orthogonal rotation withquartimax. McKnight, Cummings, and Chervany (1998)argued that if the trust constructs form a model of causallylinked factors/variables (which implies positive correla-tions), oblique rotation should be applied in the PCA. The
orthogonal rotation assumes that constructs are not corre-lated. Selecting orthogonal rotation is based on the resultsachieved in our preexperiment: no much correlation amongfactors/variables (Yan et al., 2008).
Phase 2: CFA
The second phase was a CFA using structural equation mod-eling to assess the CV and the DV of the latent sub-constructsin each of the three high-level trust behavior constructs (i.e.,UB, RB, and CB). We conducted this analysis by creating aLISREL v8.53 (Scientific Software International, Inc., Skokie,IL) path diagram for each construct, its constituent sub-constructs, and their items. We applied the following indicesand criteria to assess model fitness: goodness-of-fit index(GFI) and normed fit index (NFI) greater than .90, adjustedgoodness-of-fit index (AGFI) greater than .80 (Gefen, Straub,& Boudreau, 2000), comparative fit index (CFI) greater than.90 (Jiang & Klein, 1999/2000), and root mean square error ofapproximation (RMSEA) lower than .08 for a good fit andlower than .05 for an excellent fit (Browne & Cudeck, 1993).The c2 statistic is particularly sensitive to sample size (i.e., theprobability of model rejection increases with an increasingsize of samples, even though the model is minimally false),and hence adjusted c2 (c2/df; df is the degree of freedom) issuggested as a better fit metric (Bentler & Bonnett, 1980). It isrecommended that this metric should not exceed 5 for amodel with good fitness (Bentler, 1989).
If the fitness of the model is good, we further assessedthe CV and the DV of the latent sub-constructs insideeach of the three root constructs. CV was assessed usingthree criteria: (a) individual item lambda coefficients aregreater than .5 (Fornell & Larcker, 1981); (b) t statistic has asignificant .05 level for each path (Gefen et al., 2000); and(c) each path’s loading is greater than twice its standarderror (Anderson & Gerbing, 1988). DV among the latentvariables is shown without question if the intercorrelationbetween different latent variables is less than .6 (Carlsonet al., 2000).
Phase 3: discriminate validity analysis andnomological validity analysis
In the third phase, we used second-order models (throughLISREL 8.53) to analyze cross-construct relationships andreaffirm DV among the constructs. This analysis was per-formed at both the item and scale levels to ensure cross-validation and interpretation of results. We assessed internalnomological validity by testing proposed relationshipsamong the trust constructs themselves, and external nomo-logical validity by examining relationships between the trustbehavior constructs and other factors or variables influencingthe trust behavior.
648 Trust can be evaluated based on trust behaviors
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
Results
PCA results
Table 5 shows each item’s loading and extracted factorsachieved from PCA, exploratory factor analysis. The UB2.1had a cross-loading above .4. The reliability analysis showedthat dropping this item reduced the alpha value for the UBfrom .734 to .701. (Note that alpha is a reliability coefficient,which is an index to retain the item.) In addition, this item’sloading on its expected/intended construct was higher thantheir loadings on a “wrong” construct. Hereby, we decided toretain the item for the UB.
As shown in Table 5, all the item loadings were over .486regarding the UB. Five factors formed the RB. Their itemloadings were over .553. No items had a cross-loading above.4. However, RB1 and RB2 were merged. The CB was in formof two well-defined factors regarding behavior comparison(CB1 and CB2 were merged) and recommendation behavior.All their item loadings were over .589.
CFA results
Table 6 provides the results of CFA analysis on three root trustbehavior constructs based on the extracted components in
Table 5 Rotated Component Matrix of Exploratory Factor Analysis for Trust Behavior
Items
Loadings of components
UB1 UB2 UB3 RB1/2 RB4 RB6 RB3 RB5 CB1/2 CB3
UB1.1 .768UB1.3 .766UB1.2 .704UB3.3 .783UB3.2 .727UB3.1 .700UB2.2 .831UB2.3 .756UB2.1 .438 .486RB1.3 .776RB1.2 .770RB1.1 .743RB2.2 .737RB2.1 .714RB2.3 .553RB4.2 .746RB4.3 .671RB4.1 .653RB6.2 .742RB6.3 .718RB6.1 .711RB3.2 .764RB3.1 .758RB3.3 .661RB5.2 .820RB5.3 .747RB5.1 .697CB1.2 .770CB1.3 .743CB2.1 .711CB1.1 .688CB2.3 .685CB2.2 .676CB3.1 .778CB3.2 .744CB3.3 .589
Note. Extraction method: principal component analysis. Rotation method: quartimax with Kaiser normalization.
Yan et al. 649
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
PCA. Table 7 shows the results of CV and DV for each of sub-constructs of UB, RB, and CB.
Using behavior
Figure 2 shows the result of CFA analysis of the using behav-ior model (Model-UB). CV for each of the three sub-constructs was conducted according to the criteria discussedabove. The lambda coefficients of items were above .54 (.54–.75), each path was significant (t values are between 8.45 and10.36, p is at .05 level), and each path loading was greater than
twice its associated standard error (.08–.12). With GFI, NFI,and GFI statistics above .9, AGFI above .8, and RMSEA below.08, Model-UB fit was good (as shown in Table 6).
Reflection behavior
Figure 3 shows the result of the CFA analysis of the reflectionbehavior model (Model-RB). CV for each of its six sub-constructs was conducted based on the criteria discussedabove. As shown in Figure 3, each item’s lambda coefficientwas above .57 (.57–.85), each path was significant (t values are
Table 6 CFA Indices of Using Behavior, Reflection Behavior, and Correlation Behavior
c2 df c2/df GFI AGFI RMSEA NFI CFI IFI
Model-UB 100.59 24 4.19 .96 .93 .076 .91 .93 .93Model-RB 323.96 120 2.97 .94 .91 .055 .95 .97 .97Model-CB 60.96 24 2.54 .98 .96 .053 .97 .98 .98
AGFI = adjusted goodness-of-fit index; CB = correlation behavior; CFI = comparative fit index; GFI = goodness-of-fit index; IFI = incremental fit index;NFI = normed fit index; RB = reflection behavior; RMSEA = root mean square error of approximation; UB = using behavior.
Table 7 Results of Convergent Validity and Discriminant Validity
Lambdacoefficientsof items
t value ofeach path
p level ofeach path
Path loading(standard error)
Intercorrelation betweendifferent latent variables
Model-UB >.54 (.54–.75) 8.45–10.36 .05 level .54–.75 (.08–.12) <.30 (Table 8)Model-RB >.57 (.57–.85) 11.06–17.98 .05 level .57–.85 (.06–.12) <.60 (Table 8)Model-CB >.50 (.50–.80) 7.61–14.43 .05 level .50–.80 (.09–.21) <.56 (Table 8)
CB = correlation behavior; RB = reflection behavior; UB = using behavior.
UB1.1
UB1
UB1.2
UB1.3
UB2.1
UB2.2
UB2.3
UB3.1
UB3.2
UB3.3
UB2
UB3
0.68
0.54
0.75
0.68
0.54
0.56
0.60
0.62
0.63
0.54
0.71
0.43
0.53
0.71
0.68
0.64
0.61
0.60
1.00
1.00
1.00
0.45
0.48
0.32
Figure 2 The confirmatory factor analysis of using behavior. Note: UB1 = normal usage behavior; UB2 = usage behavior related to context;UB3 = usage behavior related to application features.
650 Trust can be evaluated based on trust behaviors
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
between 11.06 and 17.98, p is significant at .05 level), and eachpath loading was greater than twice its associated standarderror (.06–.12). With GFI, NFI, and GFI statistics above .9,AGFI above .8, and RMSEA below .08, Model-RB fit was good(as shown in Table 6). From the results of CFA, we found thatthe results of PCA and CFA are not so well matched. One pos-sible reason is that the model of RB is not stable; anotherreason may be caused by the difference of subject samples.
The covariance matrix of latent variables regarding the RBthat is not displayed in Figure 3 is shown below.
RB1 RB2 RB3 RB4 RB5 RB6
RB1 1RB2 .72 1RB3 .43 .44 1RB4 .17 .13 .38 1RB5 .31 .32 .25 .34 1RB6 .32 .29 .42 .65 .47 1
Correlation behavior
Figure 4 shows the result of the CFA analysis of the correla-tion behavior model (Model-CB). CV for each of its three
sub-constructs was inferred based on the criteria discussedabove. As can be seen from Figure 4, each item’s lambda coef-ficient was above .50 (.50–.80), each path was significant (tvalues are between 7.61 and 14.43, p values are at .05 level),and each path loading was greater than twice its associatedstandard error (.09–.21). With GFI, NFI, and GFI statisticsabove .9, AGFI above .8, and RMSEA below .08, Model-CB fitwas good (as shown in Table 6).
Reliability analysis results
Next, internal consistency reliability analyses were conductedusing Cronbach’s alphas. Table 8 reports results based on thefactor pattern in the CFA. The three behavior constructs werereliable in Nunnally’s (1978) heuristics.
Table 8 Cronbach’s Alphas of Internal Consistency Reliability
Behavior No. of subjects No. of items a
Using behavior 553 9 .71Reflection behavior 553 18 .85Correlation behavior 553 9 .79Trust behavior 553 36 .90
RB1
RB2
RB3
0.75
0.730.85
0.790.800.66
0.660.840.66
1.00
1.00
1.00
0.43 RB1.1
0.28 RB1.2
0.47 RB1.3
0.37 RB2.1
0.36 RB2.2
0.56 RB2.3
0.57 RB3.1
0.29 RB3.2
0.57 RB3.3
0.67 RB4.1
0.33 RB4.2
0.40 RB4.3
0.58 RB5.1
0.46 RB5.2
0.56 RB5.3
0.58 RB6.1
0.45 RB6.2
0.63 RB6.3
RB4
RB5
RB6
0.57
0.780.82
0.65
0.660.74
0.65
0.610.74
1.00
1.00
1.00
Figure 3 The confirmatory factor analysis of reflection behavior. Note: RB1 = bad performance reflection behavior; RB2 = bad performance reflectionbehavior related to context; RB3 = good performance reflection behavior; RB4 = good performance reflection behavior related to context; RB5 = badexperience reflection to context; RB6 = good experience reflection to context.
Yan et al. 651
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
Correlations of constructs
Table 9 shows the correlations of the sub-constructs compos-ing the UB, the RB, and the CB. We found that the sub-constructs had significant correlations with their rootconstructs at .01 level, which indicates that the sub-constructscan represent their root constructs (or the root constructs aremainly affected by their sub-constructs). The correlationbetween each internal sub-construct (e.g., UB1, UB2, andUB3) and its corresponding root construct (e.g., UB) wasalmost at the same level (except CB3’s correlation with CBwas a bit lower than CB1-CB’s and CB2-CB’s). This correla-tion was also higher than the correlations among the sub-constructs. This indicates that the sub-constructs belongingto a concrete root construct can measure not only the generalaspects but also the specific aspects of the represented type oftrust behavior. The above results show that the current sub-constructs of UB, RB, and CB are feasible and reasonable.
In particular, all three root constructs of trust behaviors(UB, RB, and CB) had correlations with each other at .01level. But these correlations are less significant than their cor-relations with the trust behavior. This indicates that thesethree root constructs can represent not only the generalaspects but also the specific aspects of the trust behavior.
External nomological validation
Based on the theoretical development of relationshipsbetween the trust behavior constructs and other variables,external nomological validity was assessed through the fol-lowing variables: (a) BI; (b) PM; (c) P; and (d) perceived DQ.
The results are shown in Table 10, which also includes thecorrelation coefficients between the external variables and theroot constructs of trust behaviors.We found all expected rela-tionships. Thus, the achieved results showed that the con-structs have adequate external nomological validity.
Discussions
Findings and implications
According to the aforementioned results, a 36-item scale wascreated that measures the UB, the RB, and the CB of the trustbehaviors. Meanwhile, we also developed items for studyingthe influence of external variables on the trust behaviors (asshown in the Appendix).
A trust behavior construct (the trust model) for mobileapplications is achieved based on the above data analysisaccording to the listed criteria with sound reliability (UB:a = .71; RB: a = .85; CB: a = .79; overall trust behavior:a = .90), as shown in Figure 5. The relationships of differentcomponents (i.e., the edge values in Figure 5) are set based onthe correlation analysis. Notably, the mutual correlations ofthe root constructs are around .5, which implies that theseconstructs may influence or impact each other. But theassumed relationships cannot be well proved by internalnomological validity of our experiment and the theories fromthe literature. This means that these factors could be corre-spondingly in parallel, without any causal relationships. Wealso found the influence of a number of external variables(i.e., PM, BI, perceived DQ, and P) on UB, RB, and CB; their
CB1.1
CB1
CB1.2
CB1.3
CB2.1
CB2.2
CB2.3
CB3.1
CB3.2
CB3.3
CB2
CB3
0.64
0.75
0.77
0.76
0.72
0.67
0.51
0.80
0.50
0.59
0.43
0.40
0.43
0.48
0.55
0.74
0.36
0.75
1.00
1.00
1.00
0.75
0.41
0.34
Figure 4 The confirmatory factor analysis of correlation behavior. Note: CB1 = comparison of normal usage behavior; CB2 = comparison related tocontext; CB3 = recommendation behavior.
652 Trust can be evaluated based on trust behaviors
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
correlations are also shown in Figure 5. Figure 6 further illus-trates the sub-constructs of the UB, RB, and CB according tothe CFA and correlation analysis.
Based on the achieved results, we proved most of ourhypotheses. Concretely, all hypotheses related to the UB werewell proved, i.e., Hypothesis 1a was proved by UB1; Hypoth-esis 1b was proved by UB2; and Hypothesis 1c was proved byUB3. In addition, Hypothesis 2a was well proved by RB1 andRB3; and Hypothesis 2b was well proved by RB2, RB4, RB5,and RB6. Regarding the CB, all hypotheses were well proved:Hypothesis 3a was proved by CB1; Hypothesis 3b was provedby CB2; and Hypothesis 3c was proved by CB3. In addition,Hypothesis 4a–Hypothesis 4d were verified by externalnomological validation.
We also detected the internal nomological validity of trustbehavior constructs although we cannot find theoreticalsupport on their causal relationships.The detection was basedonourhypothesesduetothe lackof existingtheoreticsupport.First, we set up a model with UB, RB, and CB. The model had abad fit according to the criteria indicated in “Data processingand analysis” section because GFI, AGFI, NFI, CFI, and incre-mental fit index were all below .9, and RMSEA was above .08.We further explored the causal relationships among RB, UB,and CB based on different hypotheses. However, the casualrelationships were not so clear among RB,UB,and CB becauseit lacked sound confirmatory analysis support. This resultindicated that the main three types of trust behavior (UB, RB,and CB) could be correspondingly independent.
In summary, the UB, RB, and CB represent the user’s trustbehaviors. They are further delineated into 12 measurablesub-constructs and relate to a number of external factors. ThePCA, CFA, and reliability analysis showed that the question-naire has positive psychometric properties with respectto model construct validity and reliability. We statisticallyproved the proposed research model.
Practical significance
Exploring such a trust behavior construct (i.e., a conceptualtrust model) has practical significance. First, our model pro-vides a valuable guideline on what kind of user data should bemonitored and collected for the purpose of user trust evalua-tion. Second, applying this model helps us ease the loadof extra human–device interaction that may be requiredby some existing trust management solutions (Yan &Holtmanns, 2008). This is because it is possible to monitorthe trust behavior through an auto-observation mechanismlocated at the mobile device. No additional usability study isneeded if such a trust management solution is deployed basedon this model. Thus, through auto-monitoring users’ trustbehaviors via user–device interactions during applicationconsumption, we can automatically extract useful informa-tion for trust evaluation. In Yan and Yan (2009), weTa
ble
9C
orre
latio
nsof
the
Fact
ors
Com
posi
ngTr
ustB
ehav
iors
UB1
UB2
UB3
UB
RB1
RB2
RB3
RB4
RB5
RB6
RBC
B1C
B2C
B3C
BTB
UB1
1U
B2.2
79**
1U
B3.2
35**
.296
**1
UB
.714
**.7
44**
.690
**1
RB1
1RB
2.5
94**
1RB
3.3
34**
.383
**1
RB4
.132
**.1
04*
.307
**1
RB5
.232
**.2
48**
.209
**.2
68**
1RB
6.2
33**
.223
**.3
31**
.514
**.3
52**
1RB
.561
**.6
63**
.671
**.6
59**
.595
**.5
99**
.676
**1
CB1
1C
B2.5
60**
1C
B3.2
31**
.302
**1
CB
.493
**.5
38**
.798
**.8
25**
.653
**1
TB.7
76**
.897
**.7
78**
1
TB=
trus
tbe
havi
or;
RB=
refle
ctio
nbe
havi
or;
CB
=co
rrel
atio
nbe
havi
or;
UB
=us
ing
beha
vior
.*C
orre
latio
nis
sign
ifica
ntat
the
0.05
leve
l(tw
ota
iled)
.**
Cor
rela
tion
issi
gnifi
cant
atth
e0.
01le
vel
(tw
ota
iled)
.
Yan et al. 653
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
Table 10 External Nomological Validity and Its Relationship with Trust Behavior with Path Coefficients and Correlation Coefficients
Personal motivation Brand impact Perceived quality Personality
Lambda coefficients(selected value>0.3) basedon CFA
PM.1 .57 BI.1 .63 PQ.1 .58 P.1 .69PM.2 .77 BI.2 .85 PQ.2 .71 P.2 .67PM.3 .32 BI.3 .68 PQ.3 .64 P.3 .73
BI.4 .31 PQ.4 .65 P.4 .59P.5 .45
Pathcoefficients
Correlationcoefficients
Pathcoefficients
Correlationcoefficients
Pathcoefficients
Correlationcoefficients
Pathcoefficients
Correlationcoefficients
Using behavior .00 .264** .15 .348** .21 .342** .38* .386**Reflection behavior .03 .355** .20** .464** .13 .453** .50** .536**Correlation behavior .06 .307** .16* .379** .14 .385** .27 .436**
CFA = confirmatory factor analysis. *p < .05, correlation is significant at the .05 level (two tailed). **p < .01, correlation is significant at the .01 level(two tailed).
Personal
motivation
Brand impact
Perceived quality
Personality
Using behavior
Reflection behavior
Correlation
behavior
.264**
.355**
.307**
.348**
.464**
.379**
.342**
.453**
.385**
.386**
.536**
.436**
.561**
.538**
.493**Trust
Behavior
.776**
.897**
.778**
Figure 5 Trust behavior construct of mobile applications. Note that ** indicates correlation is significant at the .01 level (two tailed).
RB1 BR2 RB3 RB4
Reflection Behavior
(RB)
.663** .671** .659** .595**
.594 ** .383** .307**
.334**
.104 **
.132**
RB5 RB6
.599** .676**
.268** .352**
.209**
.514**
.248**
.331**
.232 **
.223**
.233**
(a)
(c)
(b)
Using Behavior (UB)
UB1 UB2 UB3
.714** .744** .690**
.279** .296**
.235**
Correlation Behavior
(CB)
CB1 CB2 CB3
.798** .825** .653**
.560** .302**
.231**
Figure 6 Internal relationships of (a) using behavior (UB), (b) reflection behavior (RB), and (c) correlation behavior (CB). Note that ** indicates correla-tion is significant at the .01 level (two tailed).
654 Trust can be evaluated based on trust behaviors
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
formalized this trust behavior construct with a mathematicmeasure and proposed a computational trust model. Thismodel can be directly used to evaluate an individual user’strust in a mobile application, thus assisting the evaluationand management of the mobile application’s trust in a user-friendly manner. Third, this model is examined through auser study. The trust explanation mechanism based on thismodel could be easily understood and accepted by the users(Yan & Niemi, 2009). Meanwhile, a recommendation from auser or a mobile application service provider can be furtherassessed and explained with this trust behavior construct inorder to help other users in selecting a trustworthy mobileapplication. Therefore, conducting such an empirical studyhelps us explore a trust model based on human behaviors inorder to realize usable trust management in practice.
Limitations and suggestions forfuture research
The formal experiment was conducted in three Chinese uni-versities. There were more than 1,500 students with differ-ent majors who participated in the experiment. But only1,120 responses (71.1%) were valuable and usable. Most ofthe invalid and blank questionnaires were dug out from thesubjects in the Renmin University of China. After theexperiment, we interviewed some of the participants.They commented that some items were so similar andconfusing and they were afraid of providing impreciseanswers.
Some results of PCA and CFA are not so well matched, e.g.,RB1/RB2 and CB1/CB2 were merged as an independent com-ponent in PCA. The main reasons could be the following. (a)The measurement scale is still not stable enough. The CFAresults indicated that the explored model by PCA is notperfect. (b) The selected samples used for PCA and CFA aredifferent. The mismatch of the result could be caused by sam-pling errors. This kind of inconsistency occurred often in pre-vious researches. That is also the reason we need to multi-prove a measurement scale. No matter which reason, weshould further confirm our measurement scale. We may takeadditional samples to repeat CFA if we think CFA results aremore reasonable based on theoretic analysis.
We found that the questionnaire has good convergent andDV regarding the latent sub-constructs within each of theroot constructs. The achieved results were not good enoughregarding the causal relationship assumption. This maybe caused by two reasons. First, the number of variables/principal factors in the trust model was small. We onlyselected three principal variables and other variables may alsocontribute to the model. Second, the path of the trust modelwe identified was possibly not good enough.
The internal nomological validity was examined throughthe causal relationships among the three root constructs of
trust behaviors. In our reported results, the fit of causal rela-tionship model was not good. In the future, we will furtherexplore these causal relationships. If the result is still notgood, it will imply that there are no causal relationships, orthe relationships are in parallel, or not linear.
In our study, we used samples made up of Chinese univer-sity students. Note that China is the biggest mobile phonemarket in the world and it is very common for Chinese uni-versity students to use mobile phones in their routine life(refer to Tables 3 & 4), the samples we adopted have certainuniversality. As students are not representative of the entiremobile user population, the results may not be generalizableto other types of users. Future study would be useful tofurther prove the result with other representative samples. Inaddition, we plan to conduct our empirical study in differentcountries with different cultures in order to explore the cul-tural influence on trust behaviors.
Conclusions
User–application trust is becoming more and more impor-tant for developing and facilitating mobile applicationsand mobile Internet-based services. Studying the trustbehavior greatly helps in explaining trust status becausereal behavior-based explanation is more convincing. In thispaper, we explored a conceptual trust model for mobileapplications based on trust behaviors. This model is a trustbehavior construct achieved from a large-scale user experi-ment. The construct has been examined and proved withsound validity and reliability by PCA, reliability analysis,and CFA. It provides the main factors and constructs oftrust behavior that contribute to the calculation of the user’strust in a mobile application. More significantly in practice,the investigated trust behaviors can be auto-monitored bythe mobile device. Thus, applying our model for trust evalu-ation could greatly reduce the need on explicit human–computer interaction.
Regarding the future work, we plan to continue alongseveral directions. First, we plan to further improve thecurrent measures and conduct further empirical studiesbased on the suggestions discussed in the previous section.Second, we are going to prototype a secure trust evaluatorin a mobile device on the basis of the computationally for-malized trust behavior model with usage privacy preserva-tion. Additionally, we plan to develop a credible and usablereputation system for mobile applications by aggregatingindividual trust calculated based on the model exploredherein.
Acknowledgment
This work is sponsored by the Fundamental Research Fundsfor the Central Universities under Grant No. K5051201032.
Yan et al. 655
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
The authors thank Prof. Rong Yan for his efforts on projectcoordination and user experiments. The authors would like
to thank the anonymous reviewers’ valuable comments forthe improvement of the paper.
References
Abrams, L. C., Cross, R., Lesser, E., & Levin,D. Z. (2003). Nurturing interpersonaltrust in knowledge-sharing networks.Academy of Management Executive, 17,64–77.
Anderson, J. C., & Gerbing, D. W. (1988).Structural equation modeling in prac-tice: A review and recommended two-step approach. Psychological Bulletin,103, 411–423.
Anderson, J. C., & Narus, J. A. (1990). Amodel of distributor firm and manufac-turer firm working partnerships. Market-ing, 54, 42–58.
Antifakos, S., Kern, N., Schiele, B., & Schwa-ninger, A. (2005). Towards improvingtrust in context-aware systems by display-ing system confidence. Proceedings of the7th International Conference on HumanComputer Interaction with MobileDevices & Services MobileHCI’05, 9–14.doi: 10.1145/1085777.1085780
Avizienis, A., Laprie, J. C., Randell, B., &Landwehr, C. (2004). Basic concepts andtaxonomy of dependable and securecomputing. IEEE Transactions onDependable and Secure Computing, 1,11–33.
Bagozzi, R. P., Yi, Y., & Phillips, L. W. (1991).Assessing construct validity in organiza-tional research. Administrative ScienceQuarterly, 36, 421–458.
Bandura, A. (1986). Social foundations ofthought and action: A social cognitivetheory. Englewood Cliffs, NJ: PrenticeHall.
Barber, B. (1983). The logic and limits oftrust. New Brunswick, NJ: RutgersUniversity Press.
Bentler, P. M. (1989). EQS Structural Equa-tions Program Manual. BMDP StatisticalSoftware, Los Angeles.
Bentler, P. M., & Bonnett, D. G. (1980). Sig-nificance tests and goodness of fit in theanalysis of covariance structures. Psycho-logical Bulletin, 88, 588–606.
Bhattacherjee, A. (2002). Individual trust inonline firms: Scale development and
initial test. Journal of Management Infor-mation Systems, 19, 211–241.
Bhattacherjee, A., & Premkumar, G. (2004).Understanding changes in belief andattitude toward information tech-nology usage: A theoretical model andlongitudinal test. MIS Quarterly, 28,229–254.
Bigley, G.A., & Pearce, J. L. (1998). Strainingfor shared meaning in organizationscience: Problems of trust and distrust.Academy of Management Executive, 23,405–421.
Boudreau, M. C., Gefen, D., & Straub, D. W.(2001). Validation in IS research: A state-of-the-art assessment. MIS Quarterly, 25,1–24.
Browne, M. W., & Cudeck, R. (1993). Alter-native ways of assessing model fit. In K.A.BoUen & J. S. Long (Eds.), Testing struc-tural equation models. Beverly Hills, CA:Sage.
Carlson, D., Kacmar, K., & Williams, L.(2000). Construction and initial valida-tion of a multidimensional measure ofwork-family conflict. Journal of Voca-tional Behavior, 56, 249–276.
Corritore, C. L., Kracher, B., & Wiedenbeck,S. (2003). On-line trust: Concepts,evolving themes, a model. InternationalJournal of Human-Computer Studies,Trust and Technology, 58, 737–758.
Crocker, L., & Algina, J. (1986). Introductionto classical and modern test theory. NewYork: Holt, Rinehart, and Winston.
Daignault, M., & Marche, S. (2002). Ena-bling trust online. Proceedings of theThird International Symposium on Elec-tronic Commerce, p. 3. doi: 10.1109/ISEC.2002.1166905
Davis, F. D. (1989). Perceived usefulness,perceived ease of use, and user accept-ance of information technology. MISQuarterly, 13, 319–340.
Deutsch, M. (1973). The resolution of con-flict: Constructive and destructive proc-esses. New Haven, CT: Yale UniversityPress.
Fabrigar, L. R., Wegener, D. T., MacCallum,R. C., & Strahan, E. J. (1999). Evaluating
the use of exploratory factor analysisin psychological research. PsychologicalMethods, 4, 272–299.
Fishbein, M., & Ajzen, I. (1975). Beliefs, atti-tude, intention and behavior: An introduc-tion to theory and research. Reading, MA:Addison-Wesley.
Fornell, C., & Larcker, D. F. (1981). Evaluat-ing structural equation models withunobservable variables and measure-ment error. Journal of MarketingResearch, 18, 39–50.
Fox, A. (1974). Beyond contract: Work,power, and trust relations. London, UK:Faber.
Gefen, D. (2000). E-commerce: The roleof familiarity and trust. Omega-International Journal of ManagementScience, 28, 725–737.
Gefen, D., Karahanna, E., & Straub, D.(2003). Trust and TAM in online shop-ping: An integrated model. MIS Quar-terly, 27, 51–90.
Gefen, D., & Straub, D. (2003). Managinguser trust in B2C e-services. E-ServiceJournal, 2, 7–24.
Gefen, D., Straub, D., & Boudreau, M.(2000). Structural equation modelingand regression: Guidelines for researchpractice. Communications AIS, 7, 1–78.
Gerbing, D.W., & Anderson, J. C. (1988). Anupdated paradigm for scale developmentincorporating unidimensionality andits assessment. Journal of MarketingResearch, 15, 186–192.
Grabner-Kräuter, S., & Kaluscha, E. A.(2003). Empirical research in on-linetrust: A review and critical assessment.International Journal of Human-Computer Studies, 58, 783–812.
Hair, J. F., Anderson, R. E., Tatham, R. L., &Black, W. C. (1998). Multivariate dataanalysis with readings (5th ed.). Engle-wood Cliffs, NJ: Prentice Hall.
Hsu, M. H., Yen, C. H., Chiu, C. M., &Chang, C. M. (2006). A longitudinalinvestigation of continued onlineshopping behavior: An extension ofthe theory of planned behavior.
656 Trust can be evaluated based on trust behaviors
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
International Journal of Human-Computer Studies, 64, 889–904.
Jarvenpaa, S., Tractinsky, N., & Saarinen, L.(1999). Consumer trust in an Internetstore: A cross-cultural validation. Journalof Computer-Mediated Communication,5(2). Retrieved from www.onlinelibrary.wiley.com/doi/10.1111/j.1083-6101.1999.tb00337.x/full.
Jarvenpaa, S., Tractinsky, N., & Vitale, M.(2000). Consumer trust in an Internetstore. Information Technology and Man-agement, 1, 45–71.
Jiang, J. J., & Klein, G. (1999/2000). Supervi-sor support and career anchor impacton the career satisfaction of the entry-level information systems professional.Management Information Systems, 16,219–240.
Kantowitz, B. H., Hankowski, R. J., & Kan-towitz, S. C. (1997). Driver acceptance ofunreliable traffic information in familiarand unfamiliar settings. Human Factors,39, 164–176.
Kim, K. K., & Prabhakar, B. (2004). Initialtrust and the adoption of B2Ce-commerce: The case of Internetbanking. ACM SIGMIS Database, 35,50–64.
Koh, Y. J., & Sundar, S. S. (2010). Effects ofspecialization in computers, Web sites,and Web agents on e-commercetrust. International Journal of Human-Computer Studies, 68, 899–912.
Lance, J., Hoffman, L. J., Kim, L. J., & Blum,J. (2006). Trust beyond security: Anexpanded trust model. Communicationsof the ACM, 49, 94–101.
Lee, J., & Moray, N. (1992). Trust, controlstrategies and allocation of function inhuman–machine systems. Ergonomics,35, 1243–1270.
Lee, J. D., & Moray, N. (1994). Trust, selfconfidence and operators’ adaptationto automation. International Journal ofHuman-Computer Studies, 40, 153–184.
Lewis, D., & Weigert, A. (1985). Trust as asocial reality. Social Forces, 63, 967–985.
Li, X., Valacich, J. S., & Hess, T. J. (2004). Pre-dicting user trust in information systems: Acomparison of competing trust models.Proceedings of the 37th AnnualHawaii International Conference on
System Sciences, 10pp. doi: 10.1109/HICSS.2004.1265619
Luhmann, N. (1979). Trust and power.Chichester, England: Wiley.
Lumsden, J., & MacKay, L. (2006). How doespersonality affect trust in B2Ce-commerce? Proceedings of 8th Interna-tional Conference on ElectronicCommerce, 471–481. doi: 10.1145/1151454.1151526
MacCallum, R. C., & Austin, J. T. (2000).Applications of structural equation mod-eling in psychological research. AnnualReview of Psychology, 51, 201–226.
Marsh, S. (1994). Formalising trust as a com-putational concept. PhD Thesis, Univer-sity of Stirling.
Mayer, R. C., Davis, J. H., & Schoorman,F. D. (1995). An integrative model oforganizational trust. Academy of Man-agement Review, 20, 709–734.
McAllister, D. J. (1995). Affect andcognition-based trust as foundations forinterpersonal cooperation in organiza-tions. Academy of Management Journal,38, 24–59.
McKnight, D. H., Choudhury, V., &Kacmar, C. (2002). Developing and vali-dating trust measures for e-commerce:An integrative typology. InformationSystems Research, 13, 334–359.
McKnight, D. H., Cummings, L. L., &Chervany, N. L. (1998). Initial trust for-mation in new organizational relation-ships. Academy of Management Review,23, 473–490.
Muir, B. M. (1994). Trust in automation:Part I. Theoretical issues in the study oftrust and human intervention in auto-mated systems. Ergonomics, 37, 1905–1922.
Muir, B. M., & Moray, N. (1996). Trustin automation: Part II. Experimentalstudies of trust and human interventionin a process control simulation. Ergo-nomics, 39, 429–460.
Nunnally, J. C. (1978). Psychometric theory(2nd ed.). New York: McGraw-Hill.
Oliver, R. L. (1980). A cognitive model forthe antecedents and consequences of sat-isfaction. Journal of Marketing Research,17, 460–469.
Parkhe, A. (1998). Building trust in interna-tional alliances. Journal of World Business,33, 417–437.
Paul, D. L., & McDaniel, R. R., Jr. (2004). Afield study of the effect of interpersonaltrust on virtual collaborative relationshipperformance. MIS Quarterly, 28, 183–227.
Pavlou, P. A. (2003). Consumer acceptanceof electronic commerce—integratingtrust and risk with the technologyacceptance model. International Journalof Electronic Commerce, 7, 69–103.
Pavlou, P. A., & Gefen, D. (2004). Buildingeffective online marketplaces withinstitution-based trust. InformationSystems Research, 15, 37–59.
Pennington, R., Wilcox, H. D., & Grover,V. (2004). The role of system trustin business-to-consumer transactions.Journal of Management InformationSystems, 20, 197–226.
Ratnasingam, P. (2005). Trust in inter-organizational exchanges: A case study inbusiness to business electronic com-merce. Decision Support Systems, 39, 525–544.
de Ruyter, K., Wetzels, M., & Kleijnen, M.(2001). Customer adoption of eservice:An experimental study. InternationalJournal of Service Industry Management,12, 184–207.
Schwab, D. P. (1980). Construct validity inorganizational behavior. In B. M. Staw &L. L. Cummings (Eds.), Research inorganizational behavior (Vol. 2, pp. 3–43).Greenwich, CT: JAI Press.
Shankar, V., Urban, G. L., & Sultan, F.(2002). Online trust: A stakeholderperspective, concepts, implications, andfuture directions. Journal of StrategicInformation Systems, 11, 325–344.
Sheppard, B. H., Hartwick, J., & Warshaw, P.R. (1988). The theory of reasoned action:A meta-analysis of past research withrecommendations for modifications infuture research. Consumer Research, 15,325–343.
Sheridan, T. (1980). Computer control andhuman alienation. Technology Review, 83,61–73.
Song, S., Hwang, K., Zhou, R., & Kwok, Y. K.(2005). Trusted P2P transactions withfuzzy reputation aggregation. IEEE Inter-net Computing, 9, 24–34.
Sun, Y., Yu, W., Han, Z., & Liu, K. J. R.(2006). Information theoretic frame-work of trust modeling and evaluation
Yan et al. 657
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
for ad hoc networks. IEEE Journal onSelected Area in Communications, 24,305–317.
Theodorakopoulos, G., & Baras, J. S. (2006).On trust models and trust evaluationmetrics for ad hoc networks. IEEEJournal on Selected Areas in Communica-tions, 24, 318–328.
Venkatesh, V., & Davis, F. D. (2000). A theo-retical extension of the technologyacceptance model: Four longitudinalfield studies. Management Science, 46,186–204.
Wang, W., & Benbasat, I. (2005). Trust inand adoption of online recommendationagents. Journal of the Association for Infor-mation Systems, 6, 72–101.
Webster, J., & Martocchio, J. J. (1992).Microcomputer playfulness: Develop-ment of a measure with workplaceimplications. MIS Quarterly, 16, 201–226.
Wu, I., & Chen, J. (2005). An extension oftrust and TAM model with TPB in theinitial adoption of on-line tax: Anempirical study. International Journal ofHuman-Computer Studies, 62, 784–808.
Xiong, L., & Liu, L. (2004). PeerTrust:Supporting reputation-based trust forpeer-to-peer electronic communities.IEEE Transactions on Knowledge andData Engineering, 16, 843–857.
Yan, Z. (2007). Trust management formobile computing platforms. PhDThesis, Helsinki University ofTechnology.
Yan, Z. (2010). Trust modeling and manage-ment in digital environments: fromsocial concept to system development. IGIGlobal. doi: 10.4018/978-1-61520-682-7
Yan, Z., & Holtmanns, S. (2008). Trust mod-eling and management: From social trustto digital trust. In R. Subramanian (Ed.),Computer security, privacy and politics:
Current issues, challenges and solutions(pp. 290–323). Hershey, PA: IGI Global.
Yan, Z., & Niemi, V. (2009). A methodologytowards usable trust management.ATC’09, LNCS 5586, 179–193. doi:10.1007/978-3-642-02704-8_14
Yan, Z., Niemi, V., Dong, Y., & Yu, G. (2008).A user behavior based trust model formobile applications. ATC’08, LNCS 5060,455–469. doi:10.1007/978-3-540-69295-9_36
Yan, Z., & Yan, R. (2009). Formalizingtrust based on usage behaviours formobile applications. ATC’09, LNCS 5586,194–208. doi: 10.1007/978-3-642-02704-8_15
Zucker, L. G. (1986). Production of trust:Institutional source of economic struc-ture, 1840–1920. In B. M. Staw & L. L.Cummings (Eds.), Research in organiza-tional behavior (pp. 53–111). Greenwich,CT: JAI Press.
Appendix
Measures
1. Using behavior (UB)
UB1: normal usage behavior1. The more times you use the messaging, the more you
trust it.2. The more frequently you use the messaging, the more you
need it.3. The longer time you use the messaging, the more you trust
it.UB2: behavior related to context1. You do more important tasks through the messaging if you
trust it more.2. You do more risky tasks through the messaging if you trust
it more (e.g., SMS payment).3. You do more urgent tasks through the messaging if you
trust it more.UB3: feature-related usage behavior1. You would try more features of the messaging if you trust
it more.2. After trying more features of the messaging, you gain more
expertise on it.3. Good quality of the messaging would encourage you to try
new features of it.
2. Reflection behavior (RB)
RB1: bad performance reflection behavior1. You could decrease the times of using the messaging due to
its bad performance.2. Your usage interest and usage frequency could be
decreased due to the bad messaging performance.3. You could decrease the time of using the messaging due to
its bad performance.RB2: bad performance reflection behavior related tocontext1. Bad performance of the messaging could discourage you
to do important things with it.2. Bad performance of the messaging could discourage you
to do highly risky things with it.3. Bad performance of the messaging could discourage you
to do urgent things with it.RB3: good performance reflection behavior1. You could increase the time of using the messaging due to
its good performance.2. You could increase the times of using the messaging due to
its good performance.3. Your usage interest and usage frequency could be
increased due to the good messaging performance.RB4: good performance reflection behavior related tocontext1. Good performance of the messaging could encourage you
to do highly risky things with it.
658 Trust can be evaluated based on trust behaviors
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659
2. Good performance of the messaging could encourage youto do important things with it.
3. Good performance of the messaging could encourage youto do urgent things with it.
RB5: bad experience reflection to context1. After very bad experiences of using the messaging, you
could use it to do less risky task.2. After very bad experiences of using the messaging, you
could use it to do less important task.3. After very bad experiences of using the messaging, you
could use it to do less urgent task.RB6: good experience reflection to context1. After very good experiences of using the messaging, you
could use it to do more risky tasks.2. After very good experiences of using the messaging, you
could use it to do more important tasks.3. After very good experiences of using the messaging, you
could use it to do more urgent tasks.
3. Correlation behavior (CB)
CB1: comparison of normal usage behavior1. Using the messaging more times than another
similarly functioned mobile application means you trustit more.
2. Using the messaging more frequently than another simi-larly functioned mobile application means you trust itmore.
3. Spending more time in using the messaging than anothersimilarly functioned mobile application means you trust itmore.
CB2: comparison related to context1. Using the messaging, not another similarly functioned
mobile application, to fulfill a more important task meansyou trust it more.
2. Using the messaging, not another similarly functionedmobile application, to fulfill a more risky task means youtrust it more.
3. Using the messaging, not another similarly functionedmobile application, to fulfill a more urgent task means youtrust it more.
CB3: recommendation behavior1. If you have very good experiences in using the messaging,
you generally would like to recommend it.2. For two similarly functioned messaging applications, you
trust more in the one you would like to recommend.3. After very bad experiences in using the messaging, you
generally don’t want to recommend it.PM: personal motivation1. A mobile phone is very important in your life.2. You often use a mobile phone to transfer important
information.3. Because a mobile phone is important to you, you will con-
tinuously use it in spite of having some problems.BI: brand impact1. You like a mobile application developed by a famous
vendor.2. You like using a mobile phone with a famous brand.3. You would like to recommend a mobile application devel-
oped by a famous vendor.4. For a mobile application developed by an infamous
vendor, you will cautiously use it or stop using it if youmeet some problems.
DQ: perceived device quality1. Generally speaking, the phone you are using is working
well.2. Using a mobile phone for you is easy and convenient.3. Using a mobile phone is economical and useful for you.4. Using a mobile phone can satisfy your personal demands
well.P: Personality1. In your opinion, mobile applications have a promising
future.2. You believe the mobile application will be continuously
improving and upgrading.3. You have confidence on the future of mobile phone usage.4. You think the mobile phone is a personal communicator
liked by most people.5. You think the mobile applications benefit your life and
study very much.
Yan et al. 659
© 2013 Wiley Periodicals, Inc. Journal of Applied Social Psychology 2013, 43, pp. 638–659