J Sleep Res.

(2016) 25, 458–465 Insomnia and internet addiction

Sleep problems and internet addiction among children and

adolescents: a longitudinal study
Y I - L U N G C H E N 1 , 2 and S U S A N S H U R - F E N G A U 1 , 2
1
Department of Psychiatry, National Taiwan University Hospital and College of Medicine, Taipei, Taiwan; 2Graduate Institute of Epidemiology
and Preventive Medicine, College of Public Health, National Taiwan University, Taipei, Taiwan

Keywords SUMMARY
children and adolescents, internet addiction, Although the literature has documented associations between sleep
sleep problems, Taiwan
problems and internet addiction, the temporal direction of these
Correspondence relationships has not been established. The purpose of this study is to
Susan Shur-Fen Gau, MD, PhD, Department of evaluate the bidirectional relationships between sleep problems and
Psychiatry, National Taiwan University Hospital internet addiction among children and adolescents longitudinally. A four-
& College of Medicine, No. 7, Chung-Shan
wave longitudinal study was conducted with 1253 children and adoles-
South Road, Taipei 10002, Taiwan.
Tel.: +886-2-2312-3456 ext. 66802; cents in grades 3, 5 and 8 from March 2013 to January 2014. The sleep
fax: +886 2 23812408; problems of the student participants were measured by parental reports
e-mail: [email protected] on the Sleep Habit Questionnaire, which catalogues early insomnia,
middle insomnia, disturbed circadian rhythm, periodic leg movements,
Accepted in revised form 12 December 2015;
received 7 August 2015 sleep terrors, sleepwalking, sleep talking, nightmares, bruxism, snoring
and sleep apnoea. The severity of internet addiction was measured by
DOI: 10.1111/jsr.12388 students’ self-reports on the Chen Internet Addiction Scale. Based on the
results of time-lag models, dyssomnias (odds ratio = 1.31), especially
early and middle insomnias (odds ratio = 1.74 and 2.24), sequentially
predicted internet addiction, and internet addiction sequentially predicted
disturbed circadian rhythm (odds ratio = 2.40), regardless of adjustment
for gender and age. This is the first study to demonstrate the temporal
relationship of early and middle insomnia predicting internet addiction,
which subsequently predicts disturbed circadian rhythm. These findings
imply that treatment strategies for sleep problems and internet addiction
should vary according to the order of their occurrence.

20-item version of the Internet Addiction Test (Widyanto and

INTRODUCTION
Mcmurran, 2004), Chen Internet Addiction Scale (CIAS;
Sleep problems and internet addiction are common among Chen et al., 2003) and Internet Use Survey (Rotunda et al.,
children and adolescents, and are a main concern of parents. 2003). Cain and Gradisar (2010) provided three possible
Approximately 20% of youth have been reported to have mechanisms: (1) media use displaces sleep; (2) media use
internet addiction (Ko et al., 2005), and approximately 45% of causes increased arousal problems; and (3) bright light
elementary and junior high students were reported by their exposure delays the circadian rhythm. In contrast, Becker
parents to have had at least one kind of sleep problem (Gau, et al. (2015) proposed a biopsychosocial and contextual
2006). Addiction-related sleep problems and addiction are model in which sleep is not just the consequence of other
prevalent, and contribute to a notable fraction of the disease factors but has reciprocal relationships with biological,
burden in mental and neurological disorders in established psychosocial and contextual factors. However, reverse cau-
market economies (Fineberg et al., 2013). Examining the sation where the sleep problems lead to internet overuse and
longitudinal relationships between sleep problems and inter- then addiction is also possible. In the only cohort study on
net addiction could help develop preventions and treatment this topic, Tavernier and Willoughby (2014) found that sleep
strategies for these two problems. problems predicted a longer time spent on social networking
Sleep problems are usually considered negative outcomes sites among young university students. But the converse was
or complications of internet addiction, which was supported not true: internet use did not predict the development of sleep
by many assessment tools of internet addiction, such as the difficulties.

458 ª 2016 European Sleep Research Society

 Sleep and internet addiction 459

However, not all sleep problems have been studied

Sleep problems and duration
thoroughly and simultaneously for their associations with
internet addiction. In addition this issue has not yet been The Sleep Habit Questionnaire (SHQ) is designed to inves-
studied in child and adolescent populations where internet tigate the sleep habits and sleep problems in students. For
addiction and some sleep problems may be more common the section on sleep habits, there are four items containing
(Young, 1998). self-reported sleep duration (the average total sleep time and
This study aims to explore the possible longitudinal required sleep time to maintain normal daily function), self-
relationships between a range of sleep problems and internet reported sleep-related time (bedtime and rise time on school
addiction using a longitudinal study in a school-based sample days and weekends), multiple choice questions about sleep
of children and adolescents whilst addressing key con- patterns (i.e. regular sleep pattern: 22:00–00:00 hours for
founders. bedtime and 06:00–08:00 hours for wake-up time; early-
risers: after 00:00 hours for bedtime and after 09:00 hours for
wake-up time; late sleepers: before 22:00–00:00 hours for
MATERIALS AND METHODS
bedtime and before 06:00 hours for wake-up time). An
indicator of poor sleep was calculated using the required
Description of the study
sleep time to maintain normal daily function minus the
The present study is part of a longitudinal school-based average total sleep time. A higher score indicates a more
epidemiological study that aims to investigate the possible severe subjective experience of sleep deprivation.
risk factors and negative consequences of internet use and For the section regarding sleep problems, a binary
addiction among children and adolescents. Part of this work response (yes/no) with 11 items of current (within the past
on a different topic has been published elsewhere (Chen 6 months) sleep problems were asked. These sleep prob-
et al., 2015). lems were determined according to the definitions from
Diagnostic and Statistical Manual of Mental Disorders-IV
(American Psychiatric Association, 1994), including early
Participants
insomnia, middle insomnia, disturbed circadian rhythm, sleep
This study consists of four-wave data within an 11-month terrors, sleepwalking, sleep talking, nightmares, bruxism,
research period (early March 2013 to January 2014). Grade 3 snoring, sleep apnoea and periodic leg movements (PLMs),
and 5 students were recruited from six elementary schools, all of which were recorded as current sleep problems only if
and grade 8 students were from one junior high school they lasted longer than 1 month in the past 6 months.
in Northern Taiwan. The age ranges were 8–12 years Moderate to almost perfect agreement in reliability between
(mean  SD = 8.6  0.3 years), 10–12 years (mean  clinical interviews and parent reports on children
SD = 10.6  0.3 years) and 13–16 years (mean  SD (kappa = 0.54–1) for all sleep items has been reported
= 13.6  0.3 years) for grade 3, 5 and 8 students, respec- (Gau and Soong, 1995).
tively. These schools were selected based on the coopera- A post-classification of 11 sleep problems was further
tion of principals. conducted using factor structures to reduce the problems to
The study sample consisted of 1253 eligible students and three constructs according to the authors’ previous study
1128 parents (90.0%) of these students (one student (Gau, 2006): (1) parasomnias, including nightmares, sleep
reported by either or both parents in one parental question- talking, bruxism, sleepwalking and sleep terrors; (2) dyssom-
naire), who consented to the study. There were 638 boys nias, including early insomnia, middle insomnia, disturbed
and 614 girls in Wave 1 (n = 1253); the follow-up rates were circadian rhythm and PLM; (3) sleep-disordered breathing
93.1% (n = 1166), 89.6% (n = 1123) and 84.1% (n = 1054) (SDB), including snoring and sleep apnoea.
in the second, third and fourth waves, respectively. The Table S1 presents the items for the SHQ used in the study
numbers of parents for the first to the fourth wave were (Gau and Soong, 1995). Only the data on the nocturnal sleep
1128, 1005 (89.1%), 941 (83.4%) and 849 (75.3%). More duration (the average total sleep time, the required sleep time
detailed demographic information about participants is to maintain normal daily function, and indicators of poor
presented in Table 1. No significant differences among sleep) and sleep problems in the SHQ were utilized in this
participating students were found between the respondents study. The informant source for sleep problems was based
and the non-respondents based on gender in Wave 2 on parental reports because some sleep disorders, such as
(v2 = 0.95, P = 0.329); however, by treating the participants SDB (e.g. snoring and sleep apnoea) are difficult to self-
in Wave 1 as a reference group, the dropouts were identify, and parental reports have been commonly used in
significantly fewer in girls than boys in Wave 3 assessing paediatric sleep problems (Montgomery-Downs
(v2 = 13.78, P < 0.001) and Wave 4 (v2 = 13.33, et al., 2004). Sleep duration was based on student reports
P < 0.001). The dropouts were significantly fewer in grade because the sleep duration, including bedtime, wake-up time
5 than grades 3 and 8 for Waves 2 and 3 (P < 0.001), and the required sleep time to maintain normal daily
but not in Wave 4 (v2 = 4.17, P = 0.125) compared with function, are more likely to be self-monitored subjectively by
Wave 1. students.

ª 2016 European Sleep Research Society

460 Y.-L. Chen and S. Shur-Fen Gau

Table 1 Sample characteristic of students and their self-reported for sleep duration and internet addiction

Variables/student report Time 1 (n = 1253) Time 2 (n = 1166) Time 3 (n = 1123) Time 4 (n = 1054)

Gender
Male, n (%) 638 (50.9) 589 (50.5) 549 (48.9) 509 (48.3)
Female, n (%) 615 (49.1) 577 (49.5) 573 (51) 545 (51.7)
Grade
Grade 3, n (%) 254 (20.3) 219 (18.8) 207 (18.4) 212 (20.1)
Grade 5, n (%) 281 (22.4) 273 (23.4) 270 (24) 249 (23.6)
Grade 8, n (%) 718 (57.3) 674 (57.8) 646 (57.5) 593 (56.3)
CIAS (mean  SD) 47.26 (14.4) 45.05 (15.4) 42.72 (15.3) 42.44 (16.0)
Internet addiction, n (%) 152 (12.1) 128 (11.0) 102 (9.1) 98 (9.3)
Father’s education level
Elementary school or below, n (%) 4 (0.4) – – –
Middle school, n (%) 41 (4.1) – – –
High school, n (%) 220 (22.2) – – –
College, n (%) 532 (53.6) – – –
Graduate school, n (%) 196 (19.7) – – –
Mother’s education level
Elementary school or below, n (%) 6 (0.6) – – –
Middle school, n (%) 28 (3.0) – – –
High school, n (%) 240 (25.8) – – –
College, n (%) 567 (60.9) – – –
Graduate school, n (%) 90 (9.7) – – –
Nocturnal sleep hours (mean  SD) 7.63 (1.38) 7.64 (1.28) 7.42 (1.44) 7.51 (1.41)
Father’s job
Professional, n (%) 108 (11.4) – – –
Skilled, n (%) 714 (75.3) – – –
Not skilled, not professional, n (%) 126 (13.3) – – –
Mother’s job
Professional, n (%) 40 (4.5) – – –
Skilled, n (%) 533 (59.4) – – –
Not skilled, not professional, n (%) 325 (36.2) – – –
Nocturnal sleep hours required (mean  SD) 8.40 (1.53) 8.40 (1.40) 8.28 (1.67) 8.44 (1.61)
Indicator of poor sleep (mean  SD) 0.77 (1.44) 0.75 (1.45) 0.85 (1.64) 0.92 (1.55)
SAICA
Mother–child relationship (mean  SD) 4.63 (1.92) 4.56 (1.90) 4.63 (2.04) 4.56 (1.90)
Father–child relationship (mean  SD) 5.24 (2.30) 5.12 (2.28) 5.16 (2.34) 5.12 (2.28)
Parents–child relationship (mean  SD) 4.94 (1.88) 4.84 (1.88) 4.91 (1.98) 4.84 (1.88)

Internet addiction was measured by the CIAS with a cut-off score of 63/64 to determine non-cases and cases. Indicator of poor sleep was
calculated by nocturnal sleep hours required minus nocturnal sleep hours.
CIAS, Chen Internet Addiction Scale; SAICA, Social Adjustment Inventory for Children and Adolescents.

The CIAS and the penetration of internet use The Chinese version of the Social Adjustment Inventory
for Children and Adolescents (SAICA)
The CIAS is a 26-item self-reported measure using a four-
point Likert scale to assess internet-related symptoms. The participants’ parent–child relationship quality was
CIAS has high internal consistency (Cronbach’s alpha, assessed by six questions in the Chinese SAICA, which
0.79–0.93; Chen et al., 2003), and the Cronbach’s alpha was designed to assess the adaptive functioning of children
value for the overall CIAS was 0.94 based on Wave 1 data and adolescents aged 6–18 years in four major areas,
for this study. The CIAS scores range from 26 to 96, in including school, spare time activities, peer relations and
which a higher score indicates the more severe and home life (John et al., 1987). The six parent–child relation-
problematic internet usage. The cut-off point of 63/64 was ship questions were in the subscale of home behaviours; the
found to be the best in discriminating the cases of internet Chinese SAICA was used to assess children’s perceptions of
addiction from non-cases – with a sensitivity of 87.6% and active interactions with their mother and father, such as
specificity of 92.6% (Ko et al., 2005). One question was willingness to talk or do activities with their parents and be
added: ‘Have you ever used the internet within a 6-month with them. The Chinese SAICA was found to have satisfac-
period?’ to assess the percentage of internet users of the tory psychometric properties (Gau et al., 2006). A combined
participants. score of the subscales of mother–child and father–child

ª 2016 European Sleep Research Society

 Sleep and internet addiction 461

relationships was presented as the index of the parent–child and job were 24.7% (310/1253) and 28.7% (360/1253),
relationship quality, with a higher score indicating a more respectively.
severely impaired parent–child relationship. Finally, the amount of missing data for the penetration of
internet use (only asked in Wave 1) was 9.0% (1129/1253).
The maximum likelihood methods for the estimation of
Socioeconomic status (SES)
missing values were evaluated by the generalized model
Socioeconomic status was measured by parental education and linear mixed model using the SAS 9.4 version GLIMMIX
level and job only for Wave 1. There are five education levels: and MIXED procedures (SAS Institute, Cary, NC, USA),
elementary school or below; middle school; high school; respectively.
college; and graduate school. The three job groups are Linear trend tests for nocturnal sleep hours, CIAS and the
professional (e.g. professor and judge), skilled (e.g. techni- case number of internet addiction among four time points
cian and cook), and not skilled or professional (e.g. vendor). were conducted using linear mixed model analyses adjusting
for the lack of independence for classes (students in the
same class). Trend tests were conducted using Cochran–
Procedures
Armitage trend test and a general linear model to address
The Research Ethics Committee of National Taiwan categorical (the case number of internet addiction) and
University Hospital approved the ethics of this study continuous (nocturnal sleep hours and raw score of CIAS)
(approval number: 201212010RINC) before study imple- variables, respectively, for detecting the change over the four
mentation. Written informed consent was obtained from time points.
each of the study participants and their parents after an To test the longitudinal relationships between sleep dura-
explanation of the purpose and procedure of the study and tion and internet addiction, linear mixed model analyses were
reassurance of confidentiality. The parents were invited to performed with sleep hours as an outcome variable and
attend several occasions of the introduction of the study status of internet addiction as a predictor, taking into account
presented by the corresponding author (SSG) hosted by the lack of independence for the classes (students in the
the schools. The parents received the informed consent in same class) and individuals (repeated-measures in the same
paper format from the student participants. Only if the student). For assessing longitudinal associations between
parents agreed to participate in the study were they asked sleep problems and internet addiction, time-lag models were
to complete the questionnaire at home and return it in a constructed using the generalized linear mixed model with a
sealed envelope within 1 week. The students completed binomial distribution and logit link function with adjustment for
the questionnaires during class with the supervision of dependences of the same class and the repeated-measures
research assistants and their teachers. Demographics, within the same individuals. In the current model, each
sleep duration and internet addiction data assessed by internet addiction was regressed on the earlier measures of
the Chinese version of the CIAS were collected from the sleep problems, which Wave 2 internet addiction
student participants, and sleep problems were assessed by predicted by Wave 1 sleep problems, Wave 3 internet
the SHQ from their parents. addiction predicted by Wave 2 sleep problems, and Wave 4
internet addiction predicted by Wave 3 sleep problems
(Fig. 1). Further analyses were conducted after adjustment
Statistical analyses
for age, gender, SES and parent–child relationship covari-
The descriptive results were displayed as frequencies and ates, and the changes in the estimates were reported to
percentages for demographics as well as the case numbers detect the independent effects from the predictors. Subse-
of internet addiction and sleep problems, and as mean and quently, the pathways of previous analyses were reversed to
standard deviation for continuous variables, including the explore the prediction of internet addiction from sleep
sleep duration and raw score of the CIAS. No missing data problems.
were found in demographics. For exploring whether grade level would moderate the
A small amount of missing data on the 26 CIAS items and time-lag effect between sleep problems and internet addic-
the parent–child relationship subscale of SAICA were found: tion, moderation analyses were conducted based on a time-
<0.01% (32/4596)–0.01% (50/4596) and <0.01% (41/4596). lag model by further adding the main effect of grade and its
Mean imputation was conducted to obtain the total score of interaction terms with predictive variables.
the CIAS and the parent–child relationship subscale of
SAICA, whereas the amount of missing data for SHQ on
RESULTS
nocturnal sleep hours needed, nocturnal sleep hours
required and sleep problems were 4.5% (206/4596), 5.2%
Sample characteristics
(243/4596) and 5.3% (405/4596)–9.0% (414/4596), respec-
tively. For SES, the amount of missing data for father’s Table 1 presents the demographics, SES, nocturnal sleep
education level and job were 21.1% (265/1253) and 26.1% duration, CIAS total scores, the rates of internet addiction,
(327/1253), respectively, and for the mother’s education level and the parent–child relationships score of SAICA. There was

ª 2016 European Sleep Research Society

462 Y.-L. Chen and S. Shur-Fen Gau

Figure 1. Graphical displays of the

relationships between sleep problems and
internet addiction. For each model, only
three-wave data were used (Time 1 to Time
3 for predictive variables, and Time 2 to Time
4 for outcome variables).

a significant negative linear trend in the CIAS scores of sleep problems and internet addiction from Time 1 to Time
(P < 0.001). The internet penetration rate was 96.3%. The 4.
rates of internet addiction were 12.1, 11.0, 9.1 and 9.3% with
a significant decline from Time 1 to Time 4 (Z = 2.45,
Internet addiction and sleep duration
P = 0.014). The mean nocturnal sleep duration ranged from
7.42 to 7.64 h with a significantly negative linear trend Table S2 shows the sleep duration and internet addiction
(P < 0.001), suggesting decreasing actual nocturnal sleep from Time 1 to Time 4. The linear mixed model analysis
hours over time by the student’s report. The mean duration of revealed that students with internet addiction compared with
required nocturnal sleep-hours ranged from 8.28 to 8.44 h those without internet addiction had shorter nocturnal sleep
with a non-significant linear trend across the four time points duration by 0.14 h (8.4 min; F1,4399 = 5.61, P = 0.018),
(P = 0.696), suggesting unchanged required nocturnal sleep longer required nocturnal sleep duration by 0.27 h
duration over time. Finally, the mean indicator of poor sleep (16.2 min; F1,4399 = 12.6, P < 0.001) and poorer sleep by
(difference between required sleep duration and actual 0.41 h (24.6 min; F1,4399 = 29.25, P < 0.001).
nocturnal sleep duration) ranged from 0.77 to 0.92 h with a
significant positive linear trend across the four time points
Time-lag models of sleep disturbances and internet
(P < 0.001), suggesting increasingly poor sleep over time.
addiction and moderation analyses
Table 2 presents the prevalence for each sleep problem. In
general, there were significantly increasing rates of sleep Table S3 shows the earlier measures of sleep problems and
problems from Time 1 to Time 2, reaching a plateau from internet addiction from Time 2 to Time 4. Table 3 shows the
Time 2 to Time 4 for dyssomnias (19.8–28.1%), parasomnias time-lag models with earlier measures of sleep problems as
(19.3–34.1%), SDB (13.7–17.6% up to 21.6% at Time 4), and predictors and internet addiction as an outcome variable. For
any sleep problem (35.1–51.8%). Table S1 shows the rates crude and adjusted models, only dyssomnias, consisting of

Table 2 Prevalence of sleep problems for children based on parent’s report

Parents’ report on child Time 1 (n = 1128) Time 2 (n = 1005) Time 3 (n = 941) Time 4 (n = 849)

Any dyssomnias 223 (19.8) 282 (28.1) 265 (28.2) 245 (28.9)
Early insomnia 27 (2.4) 19 (1.9) 17 (1.8) 15 (1.8)
Middle insomnia 11 (1.0) 13 (1.3) 9 (1.0) 8 (0.9)
Disturbed circadian rhythm 16 (1.4) 15 (1.5) 14 (1.5) 14 (1.6)
PLMs 189 (16.8) 256 (25.5) 227 (24.1) 183 (21.6)
Any parasomnias 218 (19.3) 343 (34.1) 322 (34.2) 296 (34.9)
Sleep terror 17 (1.5) 14 (1.4) 11 (1.2) 12 (1.4)
Sleep walking 11 (1.0) 13 (1.3) 15 (1.6) 20 (2.4)
Sleep talking 131 (11.6) 212 (21.1) 192 (20.4) 169 (19.9)
Nightmare 34 (3.0) 35 (3.5) 25 (2.7) 23 (2.7)
Bruxism 101 (9.0) 182 (18.1) 170 (18.1) 141 (16.6)
Any SDB 155 (13.7) 177 (17.6) 177 (18.8) 183 (21.6)
Snore 150 (13.3) 162 (16.1) 164 (17.4) 144 (17.0)
Sleep apnoea 8 (0.7) 8 (0.8) 12 (1.3) 15 (1.8)
Any sleep problem 396 (35.1) 521 (51.8) 500 (53.1) 469 (55.2)

Values are n (%).

PLMs, periodic leg movements SDB, sleep-disordered breathing.

ª 2016 European Sleep Research Society

 Sleep and internet addiction 463

Table 3 Time-lag models for sleep problems predicting internet Table 4 Time-lag models for internet addiction predicting sleep
addiction with and without adjusting for demographics problems with and without adjusting for demographics

OR with 95% CI OR with 95% CI

Model of sleep Time-lag prediction Time-lag model of internet Time-lag prediction

Sleep problems predicting model including Sleep addiction predicting sleep model including
problems internet addiction covariates† problems problems covariates†

At least one 1.30 (1.01–1.69)* 1.38 (1.01–1.91)* At least one 1.11 (0.84–1.46) 1.10 (0.73–1.27)
kind of kind of
Dyssomnias Dyssomnias
Early insomnia 1.66 (0.81–3.40) 1.62 (0.77–3.41) Early insomnia 1.05 (0.44–2.48) 1.12 (0.47–2.65)
Middle 2.18 (0.99–4.77) 2.20 (0.97–4.98) Middle 0.55 (0.13–2.33) 0.59 (0.14–2.51)
insomnia insomnia
Disturbed 0.89 (0.31–2.56) 0.95 (0.32–2.80) Disturbed 2.27 (1.13–4.58)* 2.11 (1.08–4.11)*
circadian circadian
rhythm rhythm
PLMs 1.17 (0.90–1.53) 1.19 (0.90–1.57) PLMs 1.10 (0.67–1.20) 1.10 (0.64–1.16)
At least one 0.86 (0.65–1.13) 0.83 (0.63–1.10) At least one 0.80 (0.61–1.05) 0.77 (0.59–1.02)
kind of kind of
Parasomnias Parasomnias
Sleep terror 1.27 (0.50–3.25) 1.23 (0.47–3.23) Sleep terror 0.80 (0.24–2.63) 0.85 (0.26–2.79)
Sleep walking 0.39 (0.08–1.88) 0.37 (0.07–1.87) Sleep walking 0.17 (0.02–1.24) 0.17 (0.02–1.22)
Sleep talking 0.85 (0.62–1.17) 0.82 (0.59–1.14) Sleep talking 1.10 (0.67–1.21) 1.10 (0.67–1.20)
Nightmare 1.62 (0.95–2.76) 1.61 (0.89–2.67) Nightmare 1.59 (0.86–2.93) 1.58 (0.85–2.93)
Bruxism 0.84 (0.59–1.18) 0.82 (0.57–1.17) Bruxism 1.10 (0.76–1.39) 1.10 (0.75–1.37)
At least one 1.09 (0.78–1.53) 1.05 (0.75–1.48) At least one 1.16 (0.85–1.58) 1.10 (0.67–1.23)
kind of SDB kind of SDB
Snore 1.12 (0.80–1.56) 1.08 (0.77–1.51) Snore 1.10 (0.80–1.51) 1.05 (0.76–1.44)
Sleep apnoea 1.16 (0.39–3.44) 1.08 (0.36–3.25) Sleep apnoea 1.85 (0.75–4.53) 1.78 (0.72–4.40)
At least one 1.01 (0.79–1.30) 1.00 (0.77–1.29) At least one 1.10 (0.74–1.18) 1.10 (0.72–1.16)
kind of sleep kind of sleep
problems problems

CI, confidence interval; OR, odds ratio; PLMs, periodic leg CI, confidence interval; OR, odds ratio; PLMs, periodic leg
movements SDB, sleep-disordered breathing. movements SDB, sleep-disordered breathing.
*P < 0.05. *P < 0.05.
† †
Gender, age, parental relationship and SES are the covariates. Gender, age, parental relationship and SES are the covariates.

early insomnia, middle insomnia, disturbed circadian rhythm regardless of whether the sleep problems predicted internet
and PLM, could predict internet addiction at the next wave addiction or internet addiction predicted the sleep problems.
[crude odds ratios (ORs) = 1.30 and 95% confidence interval This indicates that the time-lag effect between sleep prob-
(CI) = 1.01–1.69; adjusted ORs = 1.38 and 95% CI = 1.01– lems and internet addiction did not differ across grades.
1.91]. Furthermore, the effect of all sleep problems did not
change much in their estimated effects on internet addiction
DISCUSSION
between the crude and adjusted models, indicating that the
effects of sleep problems on internet addiction are highly This is the first study to delineate the temporal relationships
independent of gender and age. between a variety of sleep problems and internet addiction in
Table S3 shows the sequential measures of sleep prob- a child and adolescent sample. It was found that internet
lems and internet addiction from Time 2 to Time 4. Table 4 addiction was associated with decreased sleep duration
presents the opposite path of time-lag models with the earlier during the night-time, increased sleep need and prospec-
measure of internet addiction as a predictor and sleep tively predicted disturbed circadian rhythm; internet addiction
problems as outcome variables. Internet addiction at earlier was also predicted by dyssomnias within the follow-up time
time points significantly predicted a later disturbed circadian frame.
rhythm (crude OR = 2.27 and 95% CI = 1.14–4.58), and the For the pathway from dyssomnias to internet addiction, the
effect of internet addiction is independent of gender and age possible explanation is that people who have difficulty falling
(adjusted OR = 2.11, 95% CI = 1.08–4.11). asleep during the night-time or remaining sleep are likely to
According to the moderation analysis, no moderated effect engage in internet use, which may cause people to develop
(P > 0.05) was found for the longitudinal association rela- internet addiction, which was also found by Tavernier and
tionships between sleep problems and internet addiction, Willoughby (2014). Together these two studies provide

ª 2016 European Sleep Research Society

464 Y.-L. Chen and S. Shur-Fen Gau

refinement of the previous cross-sectional studies, finding gaming addiction and generalized internet addiction, in
associations between sleep problems (mainly insomnia) and which people use the internet for a broad range of
internet addiction (Yen et al., 2008; Choi et al., 2009; activities. However, it is probable that sleep problems
Cheung and Wong, 2011; Cheng et al., 2012; Canan et al., relate differently across a variety of subtypes of internet
2013). addiction (e.g. video games, social networking, cybersex,
The results also show that internet addiction prospectively watching films and cyber gambling). Some researchers
predicted disturbed circadian rhythm. A few studies docu- have proposed that the consequences of internet use may
menting similar findings may help explain the results. For differ depending on the content of internet use (Padilla-
example, the use of computers has physiological impacts on Walker et al., 2010). Further studies should examine the
participants due to the exposure to the light source generated relationships between sleep problems and the specific
from computers or electronic devices (Higuchi et al., 2005; subtypes of internet addiction.
Chuang, 2006), which produced a sleep phase shift to later
hours (Duffy and Czeisler, 2009). Furthermore, excessive
Methodological consideration
internet use leads to irregular sleep patterns due to an
irregular bedtime schedule (Kim et al., 2010). Furthermore, This study has several strengths. First, no other study has
the characteristics of the circadian rhythm sleep disorders simultaneously examined as many sleep problems using a
that follow internet addiction resemble an irregular sleep– longitudinal study design. Some sleep problems and their
wake pattern. The core clinical features of a disturbed sleep– relationship with internet addiction have never been exam-
wake pattern are a lack of a clearly defined circadian rhythm ined or reported before, including PLMs and sleep terrors,
of sleep compared with other possible phase types, such as walking, and talking. Moreover, due to four-wave follow-up,
delayed sleep or advanced sleep (Thorpy, 2012). Research the temporal relationship between sleep problems and
into whether there is a relationship between irregular sleep– internet addiction is more solid than the only two-wave
wake rhythms and internet addiction would be recom- studies. Furthermore, this study adjusted for possible con-
mended. founders, i.e. gender, age, SES and parent–child relation-
The current findings did not replicate findings of prior ships.
studies indicating an association between internet addiction However, several major limitations should be considered
and several sleep problems (Yen et al., 2008; Choi et al., while interpreting these results. First, this study is a ques-
2009; Cheung and Wong, 2011; Cheng et al., 2012; Canan tionnaire survey. Objective measures such as polysomnog-
et al., 2013). Choi et al. (2009) found that parasomnias, raphy or actigraphy should be included in future studies.
including sleep snoring, sleep apnoea, bruxism and night- Second, although this is a large-scale study, some preva-
mares, were associated with internet addiction. However, the lence rates of sleep problems are low (e.g. early and middle
associations were not found by Li et al. (2007) or the current insomnia, 0.9–2.4%). Hence, the study may not have enough
study. This may be explained by methodological differences statistical power to detect the effect of uncommon sleep
(cross-sectional versus longitudinal study designs), different problems on internet addiction (see specific testing for the
statistical approaches (correlations versus time-lag models) effects of early insomnia or middle insomnia). The general-
or assessment tools. ization to clinical populations may be limited because there
Internet addicts had longer required sleep (16.2 min), are not structured clinical interviews and the participants are
shorter reported sleep (8.4 min), and consequently a larger not from a clinical population. Moreover, this study did not
deficit in the difference between actual and required sleep investigate the content of internet that the participants used
(sleep deficit = 24.6 min) than non-internet addicts over the and when the participants uses. Further studies should
waves of the cohort follow-up. The current results support consider the content of the internet use as a potential
previous studies reporting that excessive computer use or mediating factor. Lastly, some possible confounders such as
internet addiction is associated with decreased sleep dura- child behavioural and emotional problems are not controlled
tion (King et al., 2013; Nuutinen et al., 2014), but this is not in this study.
always found (Cheung and Wong, 2011). Sleep deficit may
be a better index to examine these relationships than simple
CONCLUSION
self-reported sleep duration.
Some features of sleep problems and internet addiction This is a 1-year longitudinal study to evaluate the bidirectional
may result in difficulties interpreting their relationships and relationships between sleep problems and internet addiction
should be treated with caution in future studies. First, there among children and adolescents. The present findings
is a wide range of prevalence rates of internet addiction showed that dyssomnias sequentially predicted internet
reported by previous studies. Second, the definitions or addiction, and internet addiction sequentially predicted dis-
criteria for both sleep problems and internet addiction are turbed circadian rhythm. Young people with dyssomnias may
diverse across different studies, both having different fill the time where they struggle to sleep with internet use, but
classifications or subtypes. The existing literature mostly this in turn can lead to circadian rhythm disturbances,
focuses on the sleep problems associated with video possibly through the effects of light at adverse times.

ª 2016 European Sleep Research Society

 Sleep and internet addiction 465

Higuchi, S., Motohashi, Y., Liu, Y. and Maeda, A. Effects of playing a

ACKNOWLEDGEMENTS computer game using a bright display on presleep physiological
variables, sleep latency, slow wave sleep and REM sleep. J. Sleep
This work was supported by grants from the Ministry of
Res., 2005, 14: 267–273.
Education, Taiwan (MOE102-A060), and Ministry of Science John, K., Gammon, D. G., Prusoff, B. A. and Warner, V. The social
and Technology, Taiwan (MOST103-2314-B-002-055-MY3). adjustment inventory for children and adolescents (SAICA): testing
of a new semistructured interview. J. Am. Acad. Child Adolesc.
Psychiatry, 1987, 26: 898–911.
AUTHOR CONTRIBUTIONS Kim, Y., Park, J. Y., Kim, S. B., Jung, I.-K., Lim, Y. S. and Kim, J.-H.
The effects of Internet addiction on the lifestyle and dietary
SS-FG designed the study and collected the data, and both behavior of Korean adolescents. Nutr. Res. Pract., 2010, 4: 51–57.
Y-LC and SS-FG contributed to the hypothesis generation, King, D. L., Gradisar, M., Drummond, A., Lovato, N., Wessel, J.,
data analyses and drafting the manuscript. SS-FG revised Micic, G. and Delfabbro, P. (2013). The impact of prolonged violent
and submitted the manuscript to the journal. video‐gaming on adolescent sleep: an experimental study. J.
Sleep Res., 22: 137–143.
Ko, C.-H., Yen, C.-F., Yen, C.-N., Yen, J.-Y., Chen, C.-C. and Chen,
REFERENCES S.-H. Screening for internet addiction: an empirical study on cut-off
points for the Chen Internet Addiction Scale. Kaohsiung J. Med.
American Psychiatric Association. Diagnostic and Statistical Manual Sci., 2005, 21: 545–551.
of Mental Disorders, 4th edn. American Psychiatric Association, Li, S., Jin, X., Wu, S., Jiang, F., Yan, C. and Shen, X. The impact of
Washington DC, 1994 (4th edn). media use on sleep patterns and sleep disorders among school-
Becker, S. P., Langberg, J. M. and Byars, K. C. Advancing a aged children in China. Sleep, 2007, 30: 361–367.
biopsychosocial and contextual model of sleep in adolescence: A Montgomery-Downs, H. E., O Brien, L. M., Holbrook, C. R. and
review and introduction to the special issue. J Youth Adolesc., Gozal, D. Snoring and sleep-disordered breathing in young
2015, 44: 239–270. children: subjective and objective correlates. Sleep, 2004, 27:
Cain, N. and Gradisar, M. Electronic media use and sleep in school- 87–94.
aged children and adolescents: a review. Sleep Med., 2010, 11: Nuutinen, T., Roos, E., Ray, C. et al. Computer use, sleep duration
735–742. and health symptoms: A cross-sectional study of 15-year olds in
Canan, F., Yildirim, O., Sinani, G., Ozturk, O., Ustunel, T. Y. and three countries. Int J Public Health, 2014, 59: 619–628.
Ataoglu, A. Internet addiction and sleep disturbance symptoms Padilla-Walker, L. M., Nelson, L. J., Carroll, J. S. and Jensen, A. C.
among Turkish high school students. Sleep Biol. Rhythms, 2013, More than just a game: video game and internet use during
11: 210–213. emerging adulthood. J. Youth Adolesc., 2010, 39: 103–113.
Chen, S.-H., Weng, L., Su, Y., Wu, H. and Yang, P. Development of a Rotunda, R. J., Kass, S. J., Sutton, M. A. and Leon, D. T. Internet use
Chinese Internet addiction scale and its psychometric study. Chin. and misuse preliminary findings from a new assessment instru-
J. Psychol., 2003, 45: 279. ment. Behav. Modif., 2003, 27: 484–504.
Chen, Y.-L., Chen, S.-H. and Gau, S. S.-F. ADHD and autistic traits, Tavernier, R. and Willoughby, T. Sleep problems: predictor or
family function, parenting style, and social adjustment for Internet outcome of media use among emerging adults at university? J.
addiction among children and adolescents in Taiwan: a longitudi- Sleep Res., 2014, 23: 389–396.
nal study. Res. Dev. Disabil., 2015, 39: 20–31. Thorpy, M. J. Classification of sleep disorders. Neurotherapeutics,
Cheng, S. H., Shih, C.-C., Lee, I. H. et al. A study on the sleep quality of 2012, 9: 687–701.
incoming university students. Psychiatry Res., 2012, 197: 270–274. Widyanto, L. and Mcmurran, M. The psychometric properties of the
Cheung, L. M. and Wong, W. S. The effects of insomnia and internet internet addiction test. Cyberpsychol. Behav., 2004, 7: 443–450.
addiction on depression in Hong Kong Chinese adolescents: an Yen, C.-F., Ko, C.-H., Yen, J.-Y. and Cheng, C.-P. The multidimen-
exploratory cross-sectional analysis. J. Sleep Res., 2011, 20: 311– sional correlates associated with short nocturnal sleep duration
317. and subjective insomnia among taiwanese adolescents. Sleep,
Choi, K., Son, H., Park, M. et al. Internet overuse and excessive 2008, 31: 1515–1525.
daytime sleepiness in adolescents. J. Neuropsychiatry Clin. Young, K. S. Internet addiction: the emergence of a new clinical
Neurosci., 2009, 63: 455–462. disorder. Cyberpsychol. Behav., 1998, 1: 237–244.
Chuang, Y.-C. Massively multiplayer online role-playing game-
induced seizures: a neglected health problem in internet addiction.
Cyberpsychol. Behav., 2006, 9: 451–456. SUPPORTING INFORMATION
Duffy, J. F. and Czeisler, C. A. Effect of light on human circadian
physiology. Sleep Med. Clin., 2009, 4: 165–177. Additional Supporting Information may be found in the online
Fineberg, N. A., Haddad, P. M., Carpenter, L. et al. The size, burden version of this article:
and cost of disorders of the brain in the UK. J. Psychopharmacol., Table S1. Items of Sleep Habit Questionnaire.
2013, 27: 761–770. Table S2. Internet addiction for sleep problems at each
Gau, S. S. F. Prevalence of sleep problems and their association with
inattention/hyperactivity among children aged 6–15 in Taiwan. J.
Time.
Sleep Res., 2006, 15: 403–414. Table S3. Earlier measure of sleep problems for subse-
Gau, S.-F. and Soong, W.-T. Sleep problems of junior high school quent Internet addiction from time 2 to time 4.
students in Taipei. Sleep, 1995, 18: 667–673. Table S4. Internet addiction for subsequent sleep prob-
Gau, S. S., Shen, H.-Y., Chou, M.-C., Tang, C.-S., Chiu, Y.-N. and lems from time 2 to time 4.
Gau, C.-S. Determinants of adherence to methylphenidate and the
impact of poor adherence on maternal and family measures. J.
Child. Adolesc. Psychopharmacol., 2006, 16: 286–297.

ª 2016 European Sleep Research Society