Content uploaded by Scott Caplan
Author content
All content in this area was uploaded by Scott Caplan on Dec 19, 2019
Content may be subject to copyright.
234
Relations Among Loneliness, Social Anxiety,
and Problematic Internet Use
SCOTT E. CAPLAN, Ph.D.
ABSTRACT
The model of problematic Internet use advanced and tested in the current study proposes
that individuals’ psychosocial well-being, along with their beliefs about interpersonal
communication (both face-to-face and online) are important cognitive predictors of negative
outcomes arising from Internet use. The study examined the extent to which social anxiety
explains results previously attributed to loneliness as a predictor of preference for online
social interaction and problematic Internet use. The results support the hypothesis that the
relationship between loneliness and preference for online social interaction is spurious, and
that social anxiety is the confounding variable.
CYBERPSYCHOLOGY & BEHAVIOR
Volume 10, Number 2, 2007
© Mary Ann Liebert, Inc.
DOI: 10.1089/cpb.2006.9963
INTRODUCTION
PROBLEMATIC INTERNET USE (PIU) is a multidi-
mensional syndrome consisting of cognitive
and behavioral symptoms that result in negative
social, academic, and professional conse-
quences.1–6 The cognitive-behavioral model of PIU
proposes that psychosocial problems (i.e., loneli-
ness, low social skills) predispose some Internet
users to develop cognitions and behaviors involv-
ing their online activity that ultimately result in
negative outcomes.1–5 Researchers have found
substantial support for the association between
psychosocial health and PIU. Several studies re-
port a significant positive correlation between
loneliness and negative outcomes due to Internet
use.1,6–8 Although the empirical link between lone-
liness and PIU is well-documented, current theo-
retical explanations for why loneliness correlates
with PIU are less developed and warrant further
examination. The study reported here examined
the extent to which social anxiety might better ex-
plain results previously attributed to loneliness as
a predictor of preference for online social interaction
and PIU.
A closer look at the relationship between
loneliness and PIU
To explain the empirical association between
loneliness and the negative outcomes of PIU, some
researchers,have advanced the lonely drawn to the
Internet hypothesis, proposing that, compared to
non-lonely people, the lonely are particularly
drawn to certain unique features of synchronous
online social interaction that are not available
in normal face to face (FtF) conversations.6,9,10
McKenna et al. suggest that lonely individuals are
“somewhat more likely to feel that they can better
express their real selves with others on the Internet
than they can with those they know offline.”11 Such
arguments are consistent with a growing line of
literature suggesting that FtF interaction and
synchronous computer-mediated communication
(CMC) differ from one another in important ways
that may be especially appealing to those with
Department of Communication, University of Delaware, Newark, Delaware.
LONELINESS, SOCIAL ANXIETY, AND PROBLEMATIC INTERNET USE 235
interpersonal difficulties.2,3,12–15 Online social inter-
action affords greater anonymity,11,16–18 greater
control over self-presentation,14,19 and less per-
ceived social risk, than in traditional FtF communi-
cation.10,14,20–21 As a result of the apparent social
benefits of online social interaction, Caplan has
argued that the association between loneliness and
negative outcomes of Internet use is mediated by a
preference for online social interaction (POSI).2,3
POSI is cognitive individual-difference character-
ized by beliefs that one is safer, more efficacious,
more confident, and more comfortable with online
interpersonal interactions and relationships than
with traditional FtF interaction.2,3,9 In one study,
Caplan found that participants’ self-reported level
of POSI mediated the association between their
level of loneliness and the extent to which they re-
ported experiencing negative outcomes due to their
Internet use.2Such findings lend support to the ar-
gument that lonely individuals are drawn to the in-
terpersonal advantages offered by online social
interaction, which in turn predicts problematic out-
comes. However, as the next section explains, our
understanding of the relationship between loneli-
ness and PIU is still quite undeveloped and limited.
Limitations of the lonely drawn to the
Internet hypothesis
Upon closer examination, there is a problem with
the logic of the lonely drawn to the Internet hy-
pothesis: people are lonely for a wide variety of dif-
ferent reasons, some of which have nothing to do
with their attitudes about interpersonal behavior or
their perceived social skills (e.g., job relocation, liv-
ing in a nursing home, insufficient time for social
activities, frequent travel for work). In such cases,
why would these lonely individuals be especially
likely to prefer the unique communicative charac-
teristics of online social interaction over regular FtF
communication? The lonely drawn to the Internet
hypothesis does not offer a satisfactory answer to
such a question because it confounds situational
loneliness with dispositional loneliness.22 To ad -
dress this problem, the following section presents
another potential explanation that may help
advance our understanding of psychosocial well-
being and PIU.
An alternative account: the socially anxious
drawn to the Internet hypothesis
An alternative to the lonely drawn to the
Internet hypothesis is that social anxiety, rather
than loneliness, leads to a preference for online
social interaction and, in turn, negative outcomes
of Internet use. The argument advanced here is
that social anxiety represents a theoretical motivat-
ing dynamic linking psychosocial well-being to
POSI and, consequently, offering a more detailed
account of the cognitive processes described by the
cognitive behavioral model of PIU.
In almost all social interactions, people are moti-
vated to engage in strategic self presentation and
identity management and to avoid making unde-
sired impressions on others.23–28 Social anxiety
arises from the desire to create a positive impres-
sion of one’s self in others along with a lack of self-
presentational confidence.27–30 The need to reduce
their anxiety motivates socially anxious people to
minimize their chances of making undesired im-
pressions on others. Most importantly, for the pur-
poses of the current study, the self-presentational
theory of social anxiety posits that, in order to
increase their perceived self-presentational efficacy,
socially anxious individuals are highly motivated
to seek low-risk communicative encounters.27–30
To reduce perceived social risks, Leary argues
that socially anxious people restrict their self-
presentational behaviors to situations perceived as
“relatively safe bets” and “will want to convey self
images that carry little risk and will want to avoid
jeopardizing their images if they can help it.”29
Thus, the hypothesis advanced here is that the so-
cially anxious should be more likely than those
who are not socially anxious to prefer online social
interaction because anxious people perceive their
self-presentational efficacy online to be greater
than in FtF interaction.19 Some evidence already
supports this claim: Caplan found that partici-
pants’ level of self-perceived self-presentational
skill was a direct negative predictor of preference
for online social interaction.3Thus, POSI should
arise from a perceived increase in self-presenta-
tional efficacy available online, along with a reduc-
tion in perceived threat, that socially anxious
people experience when engaged in online social
interaction.
Focus of the current study
The current study sought to help clarify the con-
ceptual and empirical ambiguity surrounding the
associations among social anxiety, loneliness, and
POSI. First, the current study sought to answer the
question of whether social anxiety or loneliness is a
better predictor of POSI. Based on the literature
reviewed above, it was predicted that:
H1: The relationship between loneliness and POSI is
spurious, and social anxiety is the confounding variable.
236 CAPLAN
Additionally, the cognitive-behavioral model of
PIU, and results from Caplan2suggest the following:
H2: Social anxiety is a positive direct predictor of POSI.
H3: POSI is a positive predictor of negative outcomes
resulting from one’s Internet use.
H4: There is an indirect relationship between social
anxiety and outcomes resulting from one’s Internet use
that is fully mediated by a POSI.
In sum, the current study had two goals: (1) to
clarify the relationships among social anxiety, lone-
liness, and POSI, and (2) to simultaneously test the
relationships among social anxiety, POSI, and
negative outcomes predicted by the cognitive
behavioral model of PIU. The following section
reports the methods used to test the hypotheses
presented above.
METHODS
Participants
Participants were 343 undergraduate students (239
females and 104 males) ranging in age from 18 to 28
years old (M = 19.4 Years; SD = 1.37). In the current
sample, participants reported having 1–18 years of
experience using computers (M = 8.80 years; SD =
2.99 years) and 1–12 years of experience using email
(M = 5.49 years; SD = 1.93).
Variables and measures
Loneliness. Loneliness was measured with the
20-item UCLA Loneliness Scale31 (in the current
study, = 0.92).
Social anxiety. Social anxiety was assessed with
the Social Avoidance and Distress (SAD) scale32
(= 0.93 in the current study). Participants’ scores
on the scale’s items were averaged to produce an
overall social anxiety index score (for descriptive
statistics, see Table 1). The social anxiety index
score was treated as an observed variable in the
structural equation modeling (SEM) analysis.
Preference for online social interaction. The items
employed as indicators of POSI were written based
on a measure developed by Caplan.2Participants
rated the extent they agreed, on a scale ranging
from 1 (strongly disagree) to 5 (strongly agree), with
four statements reflecting a POSI (Table 1 presents
TABLE 1. DESCRIPTIVE STATISTICS AND ITEM WORDING
Variable M SE SD Variance Item wording
Playing interactive
games online 1.63 0.06 1.03 1.07 —
Viewing sexually explicit
material online 1.77 0.06 1.15 1.33 —
Gambling online 1.13 0.03 0.47 0.22 —
Loneliness 1.77 0.03 0.62 0.38 —
Social anxiety 2.25 0.03 0.58 0.33 —
POSI 1 1.83 0.05 0.87 0.76 I am more confident socializing
online than I am offline.
POSI 2 1.70 0.05 0.84 0.71 I feel safer relating to other people
online rather face-to-face.
POSI 3 1.29 0.03 0.60 0.35 I prefer communicating with other
people online rather than face-to-face.
POSI 4 1.37 0.04 0.65 0.43 Meeting and talking with people is
better when done online rather than
in face-to-face situations.
NEGOUT 1 1.29 0.03 0.61 0.37 I have missed classes or work
because of online activities.
NEGOUT 2 1.15 0.03 0.47 0.22 I have gotten into trouble with my
employer or school because of
being online.
NEGOUT 3 1.14 0.02 0.45 0.20 I have missed social engagements
because of online activities.
POSI, Preference for Online Social Interaction; NEGOUT, Negative Outcomes of Online Activity.
LONELINESS, SOCIAL ANXIETY, AND PROBLEMATIC INTERNET USE 237
the exact wording of each item along with descrip-
tive statistics). In the current study, reliability
among the scale items was = 0.80. In the SEM
analysis, each item served as an indicator of the
latent POSI construct.
Negative outcomes of Internet use. The survey
items used to operationalize negative outcomes as-
sociated with one’s Internet use were drawn from
measures employed in previous studies.1–3,10 Partic-
ipants rated the extent they agreed, on a scale rang-
ing from 1 (strongly disagree) to 5 (strongly agree),
with three statements indicating that they had ex-
perienced negative outcomes due to their Internet
use (Table 1 presents the exact wording of each
item, along with descriptive statistics). In the cur-
rent study, the reliability coefficient for the negative
outcomes subscale was = 0.70. In the SEM analy-
sis, each item served as an indicator of the latent
negative outcomes construct.
Exogenous variables. In order to create a stricter
test for the endogenous model hypothesized in the
current paper, several exogenous variables that
previous research suggested may influence social
anxiety and negative outcomes associated with
Internet use were measured. The exogenous vari-
ables measured for the current study were partici-
pants’ gender and their self-reported frequency of
three content-specific PIU behaviors thought to
influence negative outcomes (i.e., how often they
engaged in online gambling, playing online inter-
active games, and viewing sexually explicit materi-
als online). First, Leary and Kowalski review
research indicating that gender may influence social
anxiety.30 Additionally, other literature suggests that
negative outcomes of one’s Internet use may be in-
fluenced by content-specific Internet uses such as
gambling online, viewing sexual material online,
and playing online games.10,33–35 Including these ex-
ogenous variables in the model helped to clarify re-
lations among social anxiety, POSI, and negative
outcomes of Internet use. Participants indicated the
frequency with which they engaged in gambling on-
line, viewing sexually explicit materials online, and
playing interactive games online on a scale ranging
from 1 (not at all) to 5 (a lot). The descriptive statis-
tics for each exogenous variable appear in Table 1.
Statistical analysis
The question of whether the link between loneli-
ness and POSI is spurious (i.e., H1) was tested with
a path analytic technique recommended by Cohen
and Cohen.36 Additionally, structural equation
modeling (SEM) was used to simultaneously test
H2, H3, and H4, while also accounting for the effects
of exogenous variables on the endogenous model.
RESULTS
Before testing H1, a correlation analysis
checked to ensure that social anxiety, loneliness,
and POSI were, in fact, significantly correlated
with one another. Consistent with H1, the zero-
order Pearson correlation between social anxiety
and loneliness was r= 0.64, p < 0.01. Additionally,
both loneliness, r = 0.34, p < 0.001, and social
anxiety, r= 0.45, p< 0.001, were significantly cor-
related with POSI. After establishing that the vari-
ables of interested were correlated, a path analysis
tested Hypothesis 1.
Path analysis
Hypothesis 1 predicted that the relationship be-
tween loneliness and POSI is spurious, and that so-
cial anxiety confounds that association. To test such
a hypothesis, Cohen and Cohen recommend a hier-
archical path analytic procedure in which “each
variable should be entered only after other vari-
ables that may be the source of spurious relation-
ship have been entered. This leads to an ordering of
the variables that reflects their presumed causal
priority.”36 H1 was tested with a path analysis that
employed two hierarchical multiple regression
equations where the order of entry of the predictors
was reversed in the second analysis.
Regression equation 1. In the first regression
analysis (Table 2), the dependent variable was
POSI, sex was treated as an exogenous variable,
and loneliness and social anxiety were predictors.
At Step 1, gender was entered, alone, into the equa-
tion in order to statistically control for its influence
on the relationships under study. Results indicated
that gender accounted for 2% of the explained vari-
ance in POSI scores, R2 = 0.02, F(1, 341) = 6.76, p=
0.01. Next, loneliness was entered at the Step 2,
which increased the predictive power of the model
significantly, R2Change = 0.11, FChange (1,340) =
40.94, p< 0.001, and eliminated the previously ob-
served gender effect. Next, social anxiety was intro-
duced at Step 3. The addition of social anxiety
increased the percentage of explained variance by
another 9%, R2Change = 0.09, FChange (1, 339) =
36.62, p < 0.001. These results indicate that social
anxiety predicts additional variance in POSI be-
yond that shared with loneliness.
238 CAPLAN
Regression model 2. Next, a second regression
analysis was performed in which the order of entry
for loneliness and social anxiety was reversed
(Table 3); gender was treated as an exogenous vari-
able and was entered at the first step, social anxiety
was entered second, and loneliness was added on
the third step. Entering loneliness at the third step,
after social anxiety had already been entered on
Step 2, identified the variance in the dependent
variable that loneliness explained, both uniquely
and in conjunction with the social anxiety. At Step 2,
loneliness was a significant predictor of POSI, =
0.33, t= 6.40, p< 0.001, and once again, the gender
effect was no longer significant.
If the relationship between loneliness and POSI is
spurious (i.e., confounded by social anxiety) as pre-
dicted by Hypothesis 1, then the addition of loneli-
ness at the Step 3 of the second equation should not
result in a significant increase in explained variance.
However, if loneliness predicts additional variance
in POSI, after social anxiety has been taken into ac-
count (i.e., the relationship is not spurious), then the
addition of loneliness at the third step in the second
equation should result in a statistically significant
increase in explained variance.
The current data supported Hypothesis 1; after
social anxiety had been taken into account in Step 2,
loneliness added no significant predictive value
to the model at Step 3, R2Change = 0.00, FChange
(1, 339) = 1.98, p= 0.16. These results support
the claim made earlier that the relationship be-
tween loneliness and POSI is spurious, and that
this relationship is confounded by both variables
association with social anxiety. Together, the two
regression analyses suggest that social anxiety is a
significant positive predictor of POSI, accounting
for approximately 19% of the explained variance in
POSI scores. Thus, the current results support the
socially anxious drawn to the Internet hypothesis.
Structural equation modeling analysis
Although the path analysis reported above was
useful for clarifying the psychosocial predictors of
POSI, an SEM analysis was employed to test the
remaining hypotheses. One of the important
TABLE 2. HIERARCHICAL REGRESSION EQUATIONS PREDICTING PREFERENCE FOR ONLINE SOCIAL INTERACTION:
EQUATION 1 (LONELINESS ENTERED BEFORE SOCIAL ANXIETY)
Variables Std. R2F Total
Step entered t change change (df) R2F total (df)
1 Gender 0.14 2.60** 0.02 6.76 (1, 341)** 0.02 6.76 (1, 341)**
2 Gender 0.10 1.86 0.11 40.94 (1, 340)*** 0.13 24.25 (2, 340)***
Loneliness 0.33 6.40***
3 Gender 0.08 1.62 0.09 36.62 (1, 339)*** 0.21 30.06 (3, 339)***
Loneliness 0.09 1.41
Social anxiety 0.38 6.05***
*p< 0.05; **p< 0.01; ***p< 0.001.
TABLE 3. HIERARCHICAL REGRESSION EQUATIONS PREDICTING PREFERENCE FOR ONLINE SOCIAL INTERACTION:
EQUATION 2 (SOCIAL ANXIETY ENTERED BEFORE LONELINESS)
Variables Std. R2F Total
Step entered t change change (df) R2F total (df)
1 Gender 0.14 2.60** 0.02 6.76 (1, 341)** 0.02 6.76 (1, 341)**
2 Gender 0.08 1.73 0.19 79.64 (1, 340)*** 0.21 43.98 (2, 340)***
Social anxiety 0.44 8.92***
3 Gender 0.08 1.62 0.01 1.98 (1, 339) 0.21 30.06 (3, 339)***
Social anxiety 0.38 6.05***
Loneliness 0.09 1.41
*p< 0.05; **p< 0.01; ***p< 0.001.
LONELINESS, SOCIAL ANXIETY, AND PROBLEMATIC INTERNET USE 239
advantages of SEM is that it allows for the simulta-
neous assessment of multiple hypothesized direct
and indirect effects. Hypotheses 2–4 predicted that
social anxiety would have a direct effect on POSI,
that POSI would have a direct positive influence on
negative outcome scores, and that social anxiety
would have an indirect effect on negative outcomes
of Internet use that is fully mediated by POSI.
A full-information maximum-likelihood SEM
analysis simultaneously tested all of these hypothe-
ses, while also accounting for the effects of the ex-
ogenous variables. The SEM analysis employed
several exogenous variables thought to influence
social anxiety and negative outcomes. Gender was
included as an exogenous predictor of social anxi-
ety. Additionally, content-specific Internet uses
(gambling online, viewing sexually explicit materi-
als online, and playing interactive games online)
were included as direct predictors of negative out-
comes. All exogenous variables were treated as ob-
served variables and all were allowed to covary
with one another. Additionally, social anxiety was
treated as an observed variable (i.e., participants’
scores on the social anxiety scale). Both POSI and
negative outcomes were treated as latent variables
with their respective scale items serving as indica-
tors. A correlation matrix of all observed measures
included in the SEM analysis appears in Table 4.
The SEM analysis reported below was computed
with AMOS 5.0.37
Model fit. Overall, the hypothesized model fit
the current data very well, = 52.89 d.f. = 47, N =
343, p= 0.26; CFI = 1.00, RMSEA = 0.02 (90% C.I:
0.00 – 0.04), pclose to fit = 0.99, SRMR = 0.05. The
model accounted for 24% of the variance in POSI
and 31% of the variance in negative outcomes of
Internet use.
Exogenous variable effects. Consistent with the lit-
erature cited earlier, all but one of the exogenous
predictor variables had and influence on the en-
dogenous variable. First, gender was a significant
direct predictor of social anxiety, = 0.13, p < 0.05.
Men’s social anxiety scores tended to be higher than
women’s scores. Two of the three behavioral predic-
tors had a significant influence on negative outcome
of Internet use. Both online gambling, = 0.21,
p< 0.001, and playing interactive games online,
= 0.14, p < 0.05, were significant positive predic-
tors of negative outcomes. However, viewing sexu-
ally explicit materials online was not a significant a
predictor of negative outcomes, = 0.12, p = 0.053.
Endogenous variable effects. Figure 1 summarizes
the results for the endogenous model. As predicted
by the socially anxious drawn to the Internet hy-
pothesis and H2, results of the SEM analysis indi-
cated that social anxiety was a significant direct
predictor of POSI, = 0.49, p < 0.001. Additionally,
as predicted by H3, POSI was a direct predictor of
TABLE 4. ZERO-ORDER PEARSON CORRELATIONS AMONG VARIABLES IN STRUCTURAL
EQUATION MODELING ANALYSIS
1234567891011
1. Playing 1.00
interactive
games online
2. View sexually 0.20** 1.00
explicit
materials
online
3. Gambling 0.14** 0.37** 1.00
online
4. Gender 0.20** 0.71** 0.30** 1.00
5. Social anxiety 0.01 0.12* 0.03 0.13* 1.00
6. POSI 1 0.08 0.12* 0.08 0.07 0.36** 1.00
7. POSI 2 0.10 0.14** 0.06 0.13* 0.40** 0.60** 1.00
8. POSI 3 0.07 0.09 0.10 0.12* 0.30** 0.42** 0.46** 1.00
9. POSI 4 0.14** 0.19** 0.08 0.13* 0.34** 0.47** 0.55** 0.50** 1.00
10. NEGOUT 1 0.21** 0.23** 0.24** 0.20** 0.13 0.29** 0.31** 0.31** 0.27** 1.00
11. NEGOUT 2 0.12* 0.17** 0.25** 0.14** 0.04 0.22** 0.23** 0.20** 0.17** 0.50** 1.00
12. NEGOUT 3 0.14** 0.21** 0.18** 0.16** 0.06 0.24** 0.29** 0.29** 0.24** 0.56** 0.47**
*p< 0.05, **p< 0.01.
negative outcomes, = 0.44, p < 0.001. In fact, in the
current data, POSI was the strongest direct predic-
tor of negative outcomes, with a path estimate more
than twice the size of any of the content-specific
exogenous predictors.
The hypothesized indirect effect of social anxiety
on negative outcomes (mediated by POSI) was as-
sessed with the distribution of product coefficients
(P).38,39 The current data support the hypothesized
indirect effect, P= 49.65, p < 0.001. The following
section interprets these results in light of the litera-
ture presented earlier and describes limitations of
the current study along with recommendations for
future research.
DISCUSSION
A number of scholars have noted the need for
better theoretical accounts, along with more de-
tailed empirical evidence, of how psychosocial well
being is associated with PIU. The purpose of the
current study was to shed further light on the pro-
cesses that constitute PIU by explaining how and
why social anxiety and POSI might predict nega-
tive outcomes of Internet use. More specifically, the
current study had two goals: (1) to assess the claim
that the association between loneliness and POSI is
spurious, and (2) to simultaneously test the
relationships between social anxiety, POSI, and
negative outcomes predicted by the cognitive be-
havioral model of PIU.
Hypothesis 1: Results of path analysis
One important theoretical question that has re-
mained unclear in much of the literature is whether
the motivating dynamic drawing people to CMC is
loneliness or aversion to the distress associated with
social anxiety, or some combination of both. The
results reported above support H1, which predicted
a spurious relationship between loneliness and PIU
that is confounded by social anxiety. As the results
presented above clearly demonstrate, once social
anxiety was taken into account, the predictive and
explanatory power of loneliness became statisti-
cally insignificant. One the other hand, social anxi-
ety predicted additional variance, beyond that
explained by loneliness, in POSI scores.
As explained earlier, the socially anxious drawn to
the Internet hypothesis proposes that socially anx-
ious individuals may develop a preference for online
social interaction because they perceive greater con-
trol over self-presentation online than they do in FtF
encounters. According to the self-presentational the-
ory, social anxiety should diminish as an individual
experiences greater self-presentational confidence in
a communication situation. Although the results
reported here support these propositions, further
research is needed to determine if socially anxious
people actually perceive greater control over inter-
personal interactions online, and if so, if their social
anxiety applies more to FtF interactions than it
does to CMC.
Hypotheses 2–4: Results of the SEM Analysis
The SEM analysis provided a rigorous simultane-
ous test of aspects of the cognitive behavioral model
described in H2, H3, and H4. First, as predicted by
H2, social anxiety was a significant direct predictor
of POSI. Second, as H3 predicted POSI was a strong
direct predictor of negative outcomes. Third, the
data also support H4, which hypothesized an indi-
rect effect of social anxiety on negative outcomes.
The results linking POSI to negative outcomes,
even after taking content-specific Internet behav-
iors into account, are consistent with previous re-
search3and should be of particular interest because
they indicate that one’s interpersonal communica-
tive preferences and perceived self-presentational
efficacy may play an important role in the etiology
of PIU. Moreover, the current results indicate that
one’s thoughts and attitudes about FtF and online
social interaction are important mediators of the re-
lationship between social anxiety and the negative
outcomes associated with Internet use.
In sum, the findings presented above suggest sev-
eral implications for research on PIU and on interper-
sonal Internet use. First, the findings indicate that
social anxiety may be better than loneliness at
predicting POSI. Besides the statistical evidence sup-
porting that conclusion, social anxiety also repre-
sents a more theoretically powerful predictor in that
it provides a clear motivation for why some people
might prefer the altered interpersonal context avail-
able in CMC. Second, the results lend further empiri-
cal support to the cognitive behavioral model of PIU.
The study reported here represents another step
240 CAPLAN
.49***
.44***
Social Anxiety Negative
Outcomes
Preference for Online
Social Interaction
FIG. 1. Standardized path coefficients for hypothesized
endogenous model. *p< 0.05; **p< 0.01; ***p< 0.001.
toward a more detailed, testable, and empirically
supported theory of PIU. Finally, the results also sug-
gest that PIU is an area that may be especially inter-
esting to interpersonal communication researchers
because it involves the communicative context online,
rather than the content-specific functions that are
often the subject of media studies.
Limitations and directions for future research.
The results reported above require qualification
in light of a number of limitations in the current
study. As explained at the outset of this paper, re-
search on PIU is only just beginning to develop
testable theories. The current study represents one
effort toward advancing that literature and devel-
oping more detailed methods for testing the cogni-
tive behavioral model of PIU. One general
qualification is that, although these results can and
do support hypotheses of direct and indirect ef-
fects, the cross-sectional research design does not
allow for a formal assessment of causal relations.
Another limitation of the current study is that the
sample of college students represents a small, but
highly relevant, section of the general population.
The use of college student samples is common in
many of our journals and represents a pragmatic
starting point for developing a more extensive pro-
gram of research. However, one benefit of the sam-
ple was that the participants in the current study
were experienced and frequent users of the Internet,
which is an important requirement for a study on
PIU. Moreover, given that most of the variables of
interest involved cognition, it is not unreasonable
to speculate that similar cognitive processes are at
work in the broader population of heavy Internet
users. Of course, future studies should seek to
confirm the findings reported above with a more
diverse sample.
The study was also limited in that it did not fully
measure generalized PIU behaviors thought to be
related to POSI (i.e., online social behaviors). How-
ever, the inclusion of content-specific uses (e.g., on-
line gambling, gaming, and viewing sexually
explicit material) in analysis as exogenous vari-
ables demonstrates that a significant proportion of
variance in negative outcomes is due to factors
other than content-specific Internet uses. Indeed,
the coefficient for the path from preference to on-
line social interaction to negative outcomes was
almost twice the size of the coefficients for the con-
tent-specific predictors. The results reported here
may indicate that the behavioral aspects of PIU
have more to do with patterns of CMC use, not
the amount of it. For instance, one hypothesis
suggested by Caplan2,40 and worth exploring in fu-
ture studies, is that that patterns of CMC use that
are defined as compulsive are probably better pre-
dictors of negative outcomes than those that are de-
fined simply as use that exceeds a certain amount of
time (i.e., frequency of use). Similarly, La Rose et al.
present a social cognitive model of PIU which argues
that problematic use results “deficient self-regulation
[of online activity] . . . in which conscious self-control
is relatively diminished.”41 These authors contend
that “what others have termed ‘Internet addiction’
can be redefined as deficient self-regulation.”41 In
their study, La Rose et al. found that depression pre-
dicted deficient Internet self-regulation, which in
turn, predicted Internet usage.
Finally, the results of the current study suggest
additional directions for future research; they fur-
ther illustrate the need for more detailed, parsimo-
nious, and predictive theories of psychosocial well
being and PIU. Although the current study shed
some light on the cognitive aspects of PIU, it also
raises questions regarding the maladaptive behav-
ioral processes involved in PIU. In the SEM analy-
sis reported above, POSI accounted for a significant
amount of variance in negative outcome scores
beyond that explained by content-specific behav-
iors. One important question for future research to
address is: how and why does POSI ultimately result
in negative outcomes associated with Internet use?
REFERENCES
1. Caplan, S.E. (2002). Problematic Internet use and
psychosocial well-being: development of a theory-
based cognitive-behavioral measure. Computers in
Human Behavior 18:533–575.
2. Caplan, S.E. (2003). Preference for online social
interaction: a theory of problematic Internet use
and psychosocial well-being. Communication Research
30:625–648.
3. Caplan, S.E. (2005). A social skill account of
problematic Internet use. Journal of Communication
55:721–736.
4. Davis, R.A. (2001). A cognitive-behavioral model of
pathological Internet use. Computers in Human Behavior
17:187–195.
5. Davis, R.A., Flett, G.L., & Besser, A. (2002). Validation
of a new measure of problematic Internet use: Implica-
tions for pre-employment screening. CyberPsychology
& Behavior 5:331–346.
6. Morahan-Martin, J., & Schumacher, P. (2003). Loneli-
ness and social uses of the Internet. Computers in
Human Behavior 19:659–671.
7. Amichai-Hamburger, Y., & Ben-Artzi, E. (2003). Lone-
liness and Internet use. Computers in Human Behavior
19:71–80.
LONELINESS, SOCIAL ANXIETY, AND PROBLEMATIC INTERNET USE 241
242 CAPLAN
8. Kubey, R.W., Lavin, M.J., & Barrow, J.R. (2001). Inter-
net use and collegiate academic performance decre-
ments: early findings. Journal of Communication
51:366–382.
9. Morahan-Martin, J. (1999). The relationship between
loneliness and Internet use and abuse. CyberPsychol-
ogy & Behavior 2:431–440.
10. Morahan-Martin, J., & Schumacher, P. (2000). Inci-
dence and correlates of pathological Internet use
among college students. Computers in Human Behavior
16:13–29.
11. McKenna, K.Y.A., Greene, A.S., & Gleason, M.E.J.
(2002). Relationship formation on the Internet: what’s
the big attraction? Journal of Social Issues 58:9–31.
12. Hancock, J.T., & Dunham, P.J. (2001). Impression
formation in computer-mediated communication
revisited: an analysis of the breadth and intensity of
impressions. Communication Research 28:325–347.
13. Ramirez, A., Walther, J.B., Burgoon, J.K., et al. (2002).
Information-seeking strategies, uncertainty, and
computer-mediated communication: toward a con-
ceptual model. Human Communication Research
28:213–228.
14. Walther, J.B. Computer-mediated communication:
impersonal, interpersonal, and hyperpersonal inter-
action. (1996). Communication Research 23:3–43.
15. Walther, J.B., & Parks, M.R. (2002) Cues filtered out,
cues filtered in: computer-mediated communication
relationships. In: Knapp, M.L., Daly, J.A., & Miller,
G.R. (eds.), The handbook of interpersonal communication,
3rd ed. Thousand Oaks, CA: Sage, pp. 529–563.
16. Bargh, J.A., McKenna, K.Y.A., & Fitzsimmons, G.M.
(2002). Can you see the real me? Activation and
expression of the “true self” on the Internet. Journal of
Social Issues 58:33–48.
17. McKenna, K.Y.A., & Bargh, J.A. (1999). Causes and
consequences of social interaction on the Internet: a
conceptual framework. Media Psychology 1:249–269.
18. McKenna, K.Y.A., & Bargh, J.A. (2000). Plan 9 from
cyberspace: the implications of the Internet for per-
sonality and social psychology. Journal of Personality
and Social Psychology 75:681–694.
19. Noonan, R.J. (1998). The psychology of sex: a mirror
from the Internet. In: Gackenbach, J. (ed.), Psychology
and the Internet: intrapersonal, interpersonal, and
transpersonal implications. San Diego, CA: Academic
Press, pp. 143–168.
20. Turkle, S. (1995). Life on the screen. New York: Simon
& Schuster.
21. Wallace, P.M. (1999). The psychology of the Internet.
New York: Cambridge University Press.
22. Jones, W.H. (1987). Research and theory on loneli-
ness: a response to Weiss’s reflections. Journal of
Social Behavior and Personality. Special Issue: Loneliness:
Theory, Research, and Applications 2:27–30.
23. Brown, P., & Levinson, S.C. (1987). Politeness: some
universals in language use. New York: Cambridge
University Press.
24. Cupach, W.R., & Metts, S. (1994). Facework. Thousand
Oaks, CA: Sage.
25. Goffman, E. (1959). The presentation of self in everyday
life. New York: Doubleday.
26. Goffman, E. (1967). Interaction ritual. New York:
Doubleday.
27. Schlenker, B.R., & Leary, M.R. (1982). Social anxiety
and self-presentation: a conceptualization and
model. Psychological Bulletin 92:641–669.
28. Schlenker, B.R., & Leary, M.R. (1985). Social anxiety
and communication about the self. Journal of
Language and Social Psychology 45:1347–1354.
29. Leary, M.L. (1983). Understanding social anxiety: social,
personality, and clinical perspective. Newbury Park,
CA: Sage.
30. Leary, M.R., & Kowalski, R.M., (1995). Social anxiety.
New York: Guilford Press.
31. Russell, D., Peplau, L.A., & Cutrona, C.E. (1980). The
revised UCLA loneliness scale: concurrent and dis-
criminant validity evidence. Journal of Personality and
Social Psychology 39:472–480.
32. Watson, D., & Friend, R. (1969). Measurement of
social evaluation anxiety. Journal of Consulting and
Clinical Psychology 33:448–457.
33. Beard, K.W., & Wolf, E.M. (2001). Modification in the
proposed diagnostic criteria for Internet addiction.
CyberPsychology & Behavior 4:377–383.
34. Griffiths, M.D., & N. Hunt, N. (1998). Computer
game “addiction” in adolescence? A brief report.
Psychological Reports 82:475–480.
35. Pratarelli, M.E., & Browne, B.L. (2002). Confirmatory
factor analysis of Internet use and addiction.
CyberPsychology & Behavior 5:53–64.
36. Cohen, J., & Cohen, P. (1983). Applied multiple
regression/correlation analysis for the behavioral sciences,
2nd ed. New York: Wiley.
37. Arbuckle, J.L. (2003). Amos for Windows. Analysis of
moment structures (Version 5.0). Chicago: Small Waters.
38. MacKinnon, D.P., Lockwood, C., & Hoffman, J.
(1998). A new method to test for mediation. Presented at
the Annual Meeting of the Society for Prevention
Research, Park City, UT.
39. MacKinnon D.P., Lockwood C.M., Hoffman J.M.,
et al. (2002). A comparison of methods to test media-
tion and other intervening variable effects. Psycholog-
ical Methods 7:83–104.
40. Caplan, S.E., & High, A.C. (2007). Beyond excessive
use: the interaction between cognitive and behav-
ioral symptoms of problematic Internet use. Commu-
nication Research Reports 23:265–271.
41. La Rose, R., Lin, C.A., & Eastin, M.S. (2003). Unregu-
lated Internet usage: addiction, habit, or deficient
self-regulation? Media Psychology 5:225–253.
Address reprint requests to:
Dr. Scott E. Caplan
Department of Communication
University of Delaware
250 Pearson Hall
Newark, DE 19716
E-mail: caplan@udel.edu