Work Stress and Subsequent Risk of Internet Addiction Among Information Technology Engineers in Taiwan

Abstract

Work stress, as defined by the Demand–Control–Support (DCS) model and the Effort–Reward Imbalance (ERI) model, has been found to predict risks for depression, anxiety, and substance addictions, but little research is available on work stress and Internet addiction. The aims of this study are to assess whether the DCS and ERI models predict subsequent risks of Internet addiction, and to examine whether these associations might be mediated by depression and anxiety. A longitudinal study was conducted in a sample (N=2,550) of 21–55 year old information technology engineers without Internet addiction. Data collection included questionnaires covering work stress, demographic factors, psychosocial factors, substance addictions, Internet-related factors, depression and anxiety at wave 1, and the Internet Addiction Test (IAT) at wave 2. Ordinal logistic regression was used to assess the associations between work stress and IAT; path analysis was adopted to evaluate potentially mediating roles of depression and anxiety. After 6.2 months of follow-up, 14.0% of subjects became problematic Internet users (IAT 40–69) and 4.1% pathological Internet users (IAT 70–100). Job strain was associated with an increased risk of Internet addiction (odds ratio [OR] of having a higher IAT outcome vs. a lower outcome was 1.53); high work social support reduced the risk of Internet addiction (OR=0.62). High ER ratio (OR=1.61) and high overcommitment (OR=1.68) were associated with increased risks of Internet addiction. Work stress defined by the DCS and ERI models predicted subsequent risks of Internet addiction.

Introduction

The growing use of the Internet across the world has been accompanied by emerging concerns regarding Internet addiction (IA).¹ Many researchers have investigated addictive properties of the Internet, including its high degree of novelty and unpredictability, and its use as an escape mechanism to avoid real or perceived problems.² In the American Journal of Psychiatry, Block (2008) proposed including IA as a psychiatric disorder with four main components: (a) excessive use, a loss of sense of time or a neglect of basic drives; (b) withdrawal, including feelings of anxiety and depression when the Internet is inaccessible; (c) tolerance, including the need for more hours of use or better computer equipment; and (d) adverse social consequences.³ In the 5th edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5), IA related to online gaming was identified as a condition that merits further research and consideration to be included as a formal disorder.⁴ Most studies on IA have focused on adolescents or young adults, and more research is needed for the working-age population. Note that IA may result in a decline in employees' work performance and “Internet misuse”—spending much time online involved in nonwork activities (e.g., unauthorized access or transmission of confidential data), resulting in unnecessary costs and legal problems for organizations.⁵

While IA may influence the workplace, it remains unclear whether the workplace influences the risk of IA. “Work stress,” defined by the Demand–Control–Support (DCS) model and the Effort–Reward Imbalance (ERI) model, has received strong support to predict various health outcomes. In the DCS model, control consists of skill discretion (e.g., opportunities to learn new things) and decision authority (e.g., ability to make decisions); demands represent work stressors (e.g., time pressures or pace of work).⁶ The combination of high demands and low control (job strain) leads to poor health. In addition, work social support from supervisors and colleagues prevents poor health or buffers stress–health relationships.⁷

In the ERI model, work stress defined by violation of social reciprocity in terms of high extrinsic effort (e.g., time pressure or overtime work) and low reward (e.g., salary, esteem, and social status control) can elicit stress responses and health problems. A construct of personality–overcommitment (OC) characterized by excessive striving at work and strong motivation for esteem was incorporated into a situational model. In control-limiting conditions, a high OC person has coping patterns of frustrated attempts to regain or maintain control over the environment. OC was suggested to have a direct effect on health or to modify ERI–health relationships.⁸

Work stress defined by the DCS and ERI models has been found to predict risk for depression, anxiety, and substance addictions (drinking and smoking).^9–11 The mechanisms suggested are that work stress leads to biological responses (e.g., dysfunction of mesolimbic dopamine system) and psychological distress (depression and anxiety), which then cause substance addictions.¹² Behavioral addictions (e.g., IA, pathological gambling, or compulsive buying) are characterized by the failure to resist an impulse or drive to perform an act harmful to the self or others.¹³ Growing evidence suggests that behavioral addictions resemble substance addictions in many domains (e.g., phenomenology or neurobiology), supporting DSM-5 to propose a new category of “Addiction and Related Disorders” encompassing both addictions.¹⁴ It is reasonable to suggest that behavioral and substance addictions share common pathophysiology and risk factors (e.g., work stress). Furthermore, depression and anxiety are identified risk factors for IA; those with depression/anxiety may excessively use the Internet to relieve psychological distress or distract attention.² Thus, it is hypothesized that work stress leads to biological responses and depression/anxiety (mediators), which then result in IA. However, little research is available on the link between work stress and IA.

The aims of this study are: (a) to assess whether work stress defined by the DCS and ERI models predicts subsequent risk of IA; and (b) to examine whether the associations between work stress and IA might be mediated by depression and anxiety. To predict these associations accurately, potential “confounders” are selected from known risk factors for IA that are associated with work stress exposures in this sample, such as demography (male, young, single, and high educational level), psychosocial factors (life events and low social support),^15,16 Internet-related factors (more time spent online, online gaming, and social media),^17,18 personality, and neurobiological mechanisms. High comorbidity (60%) of substance addictions and IA may indicate common causes (e.g., personality and neurobiological processes), so proxy adjustment for substance addictions partly captures these unmeasured confounders.¹⁹

More specifically, this study targets information technology (IT) engineers in Taiwan. They are expected to represent the working populations that have increased remarkably in this globally competitive and informational economy—called the “core labor force” by sociologist Manuel Castells (e.g., engineer, manager, service occupation, civil servant, or researcher).²⁰ Core labor forces organize innovation by creating and applying knowledge-based information in whatever domain they work. They often use the Internet as a working tool because the Internet provides access to immediate information on almost any subject and enables international communication. Due to their characteristics (e.g., well-educated professionals with more time spent online), these working populations might be vulnerable to IA and thus deserve empirical scrutiny.

Materials and Methods

Study population

The 21–55 year old IT engineers who attended periodical health examinations regulated by Labor Safety and Health Law in the general hospital near Hsinchu Science Park (HSP) were invited to participate in the study. The HSP is the largest base for IT industries in Taiwan, with 482 companies in 2012. “Engineer” was defined as someone professionally engaged in engineering fields with an academic degree in an engineering discipline.

At wave 1 between January and June 2012, 5,065 IT engineers who attended the health examinations were randomly selected and screened using the Internet Addiction Test (IAT); 4,123 (81.4%) of subjects without IA (IAT 20–39) were asked to participate in the study. A total of 3,064 participants completed a self-report questionnaire on demography, psychosocial factors, substance addictions, Internet-related factors, depression, anxiety, and work stress. The overall response rate was 74.3%.

At wave 2 between July 2012 and January 2013 (about 6 months after participation), each subject was given an initial e–mail invitation to complete the IAT questionnaire via a hyperlink, and then by a follow-up e-mail. Those who did not respond to e-mails were then approached by telephone to finish the questionnaire. The participation rate at wave 2 was 83.2%. The final sample consisted of 2,550 subjects. The length of follow-up between the two waves was on average 6.2 months. The study was approved by ethical committees in the hospital. All subjects gave written informed consent.

Measurements

Outcome variable— IAT

As the first validated instrument to assess IA, the psychometric properties of the IAT have been shown to be reliable and valid.²¹ The IAT consists of 20 items based on 5-point scale (1=“rarely,” 2=“occasionally,” 3=“frequently,” 4=“often,” and 5=“always”), covering the degree to which Internet use affects one's daily routine, social life, productivity, sleep pattern, and feelings. The minimum IAT score is 20 and the maximum is 100. An IAT score of 20–39 represents a normal online user who has control over Internet usage; an IAT score of 40–69 indicates frequent problems due to Internet usage (problematic Internet user); an IAT score of 70–100 indicates significant problems due to Internet usage (pathological Internet user). Researchers have suggested that identification of the “problematic Internet user,” a possible case of IA, is particularly relevant to define target groups for early intervention of IA.²²

DCS model

The Job Content Questionnaire (JCQ) consists of 22 items for three major scales. “Control” is measured by skill discretion with six items (e.g., learning new things, ability to develop new skills, or job requiring creativity), and decision authority with three items (e.g., freedom to make decisions, choice about how to perform work). “Demands” is measured by five items (e.g., excessive work, conflicting demands, or insufficient time). The scores for control and demands were divided into “low” and “high” by their medians respectively. No strain (high control and low demands), active job (high control and high demands), passive job (low control and low demands), and job strain (low control and high demands) were categorized. Work social support consists of social support from supervisors (four items) and co-workers (four items); each item is rated on a 4-point scale (1=“strongly disagree” to 4=“strongly agree”).²³ This score was divided into tertiles as “low,”, “medium”, and “high.” Cronbach's alphas for control, demands, and work social support in this study were 0.72, 0.73, and 0.81 respectively.

ERI model

The ERI questionnaire consists of 23 items. “Extrinsic effort” is assessed by six items (e.g., time pressure, interruptions, overtime work, or increasing demands). “Reward” is assessed by one item on salary, five items on esteem (e.g., respect from superiors and colleagues, adequate support), and five items on social status control (e.g., promotion prospects, or adequate position). The effort–reward (ER) ratio is estimated as the ratio of respective scores taking into account the unequal number of items. OC is assessed by six items (e.g., I get easily overwhelmed by time pressures at work; work is still on my mind when I go to bed). Each item is rated on a 4-point scale (1=“strongly disagree” to 4=“strongly agree”).²⁴ The scores for ER ratio and OC were divided into tertiles. Cronbach's alphas for extrinsic effort, reward, and OC in this study were 0.79, 0.80, and 0.75, respectively.

Confounders

Demographic factors included sex, age (21–30, 31–40, and 41–55 age groups), education, and marital status. Stressful life events in the previous 6 months were measured by the List of Threatening Experiences with 12 categories (e.g., serious illness/injury in subject or close relative, or major financial crisis).²⁵ Social support from family and friends was measured with the Brief Social Support Questionnaire, with five items scored on a 4-point scale (1=strongly disagree to 4=“strongly agree”).²⁶ This variable was dichotomised into “low to medium” (5–14 points) and “high” social support (15–20 points). Problem drinking was screened by the CAGE questionnaire consisting of four items with two responses (0=“no” and 1=“yes”). With a cutoff point of 2, sensitivity and specificity are high in relation to alcohol abuse and dependence.²⁷ Current smokers were identified by one or more points on the Fagerström Test for Nicotine Dependence (FTND) consisting of six questions.²⁸ For Internet-related factors, the length of time spent online was measured by average hours per day of Internet use “related to work” and “not related to work.” Specific scopes of Internet use (online games and social media) were asked by two items with two responses (0=“no” and 1=“yes”).¹⁸

Potential mediators

The Beck Depression Inventory (BDI) is a self-report, 21-item instrument that assesses the existence and severity of depressive symptoms, with 16/17 as optimal cutoff for depressive disorders.²⁹ This variable was dichotomised into “having depression” (BDI ≥17) and “no depression” (BDI<17). The Beck Anxiety Inventory (BAI) is a self-report, 21-item questionnaire that evaluates the existence and severity of anxiety symptoms, with 13/14 as optimal cutoff for anxiety disorders.³⁰ This variable was dichotomised into “having anxiety” (BAI≥14) and “no anxiety” (BAI<14).

Statistical analysis

First, descriptive characteristics for this sample were presented. The IAT outcome was set as an ordinal categorical variable: normal (IAT 20–39), problematic Internet user (IAT 40–69), and pathological Internet user (IAT 70–100). Second, ordinal logistic regression was used to assess the associations between the DCS model at wave 1 and the IAT at wave 2 by three steps: adjusted for demographic factors at wave 1 (Model 1); additionally adjusted for other confounders at wave 1 (Model 2); and additionally adjusted for potential mediators—depression and anxiety—at wave 1 (Model 3). Third, the associations between the ERI model and the IAT were assessed by ordinal logistic regression with three similar models. ERI–OC interaction terms were added into regression models and tested for significance with the likelihood ratio (LR) test. All of the above statistics were performed using STATA v11 software.

For potentially mediating roles of depression and anxiety, path analyses with an ordinal categorical IAT outcome (a continuous and normally distributed latent variable responds to each category of the observed outcome) were conducted by Mplus v6.³¹ Each path coefficient was obtained by regressing IAT outcome on an independent variable after adjusted for confounders in a probit model. An indirect effect was estimated by multiplying path coefficients along the paths connecting two variables; the extent of mediating (indirect) effect on total effect was estimated by the path coefficient.

Three fit indexes were selected in these path analyses: root mean square error of approximation (RMSEA), comparative fit index (CFI), and weighted root mean square residual (WRMR). A RMSEA value of <0.06 indicates good fit, while a RMSEA value of 0.06–0.08 is considered acceptable fit. Furthermore, a CFI value of >0.95 indicates good fit.³² WRMR is a variance-weighted approach particularly suited for path analyses with non-normally distributed outcomes (e.g., ordinal categorical outcomes); a WRMR value of <1.0 is interpreted as good model fit.³³

Results

Descriptive characteristics of the 2,550 subjects (1,914 men and 636 women) are shown in Table 1. Overall, there were larger proportions of male (75.1%), 31–40 age group (53.3%), university educated (60.6%), and married subjects (63.3%). After an average of 6.2 months of follow-up, 356 (14.0%) subjects became problematic Internet users (IAT 40–69) and and 106 (4.1%) pathological Internet users (IAT 70–100).

Table 1.

Descriptive Characteristics of the Sample (N=2,550)

Variable	Category	N	%
1. Demographic factors
Sex	Female	636	24.9
	Male	1,914	75.1
Age group	21–30	594	23.3
	31–40	1,359	53.3
	41–55	597	23.4
Education	University	1,545	60.6
	Graduate school	1,005	39.4
Marital status	Single	842	33.0
	Married	1,615	63.3
	Divorce	93	3.7
2. Psychosocial factors
Life event	None	1,893	74.3
	≥1	657	25.7
Social support	Low/medium (<15)	570	22.4
	High (15–20)	1,980	77.6
3. Substance addictions
Problem drinking	No	2,325	91.2
	Yes (CAGE ≥2)	225	8.8
Current smoker	No	2,028	79.5
	Yes (FTND ≥1)	522	20.5
4. Internet-related factors
Time on Internet (related to work)	0–4 hours/day	1,466	57.5
	≥5 hours/day	1,084	42.5
Time on Internet (not related to work)	0–3 hours/day	2,111	82.8
	≥4 hours/day	439	17.2
Online games	No	2,109	82.7
	Yes	441	17.3
Online social media	No	1,581	62.0
	Yes	969	38.0
5. Potential mediators
Depression	No	2,305	90.4
	Yes (BDI ≥17)	245	9.6
Anxiety	No	2,334	91.5
	Yes (BAI ≥14)	216	8.5
6. IAT outcomes
Normal	IAT=20–39	2,088	81.9
Frequent problems	IAT=40–69	356	14.0
Significant problems	IAT=70–100	106	4.1

FTND, Fagerström Test for Nicotine Dependence; BDI, Beck Depression Inventory; BAI, Beck Anxiety Inventory; IAT, Internet Addiction Test.

Table 2 shows the associations between the DCS model and the IAT assessed by ordinal logistic regression. In Model 2 adjusted for all confounders, the odds of having a higher IAT outcome versus a lower outcome (IAT≥40 vs. 20–39; IAT≥70 vs. 20–69) were 1.76 times larger for job strain than no strain; these odds were 0.56 times smaller for high work social support than low work social support. In Model 3 additionally adjusted for depression and anxiety, the odds of having a higher IAT outcome versus a lower outcome were 1.53 times larger for job strain than no strain; these odds were 0.62 times smaller for high work social support than low work social support. Compared to Model 1 adjusted for demographic factors, there was attenuation in the associations of job strain or work social support with the IAT after adjusting for other confounders and depression/anxiety, implying confounding or mediating effects.

Table 2.

Ordinal Logistic Regression for the Association Between the Demand–Control–Support Model at Wave 1 and the Internet Addiction Test at Wave 2

Variable	Category	Model 1 OR (95% CI)	Model 2 OR (95% CI)	Model 3 OR (95% CI)
Job strain	No strain	1.00	1.00	1.00
	Active job	1.12 (0.80–1.57)	1.02 (0.73–1.43)	0.97 (0.69–1.37)
	Passive job	1.45 (1.02–2.06)^*	1.42 (0.99–2.01)	1.36 (0.95–1.96)
	Job strain	1.95 (1.45–2.66)^*	1.76 (1.25–2.47)^*	1.53 (1.08–2.17)^*
	p for trend	p<0.001	p<0.001	p=0.004
Work social support	Low	1.00	1.00	1.00
	Medium	0.80 (0.61–1.06)	0.81 (0.62–1.07)	0.88 (0.67–1.16)
	High	0.52 (0.37–0.72)^*	0.56 (0.41–0.77)^*	0.62 (0.46–0.85)^*
	p for trend	p<0.001	p<0.001	p=0.001
Demographic factors
Sex	Female	1.00	1.00	1.00
	Male	1.19 (0.91–1.56)	1.08 (0.82–1.42)	1.17 (0.88–1.55)
Age	21–30	1.00	1.00	1.00
	31–40	0.64 (0.48–0.84)^*	0.73 (0.55–0.96)^*	0.72 (0.54–0.95)^*
	41–55	0.58 (0.43–0.80)^*	0.71 (0.48–1.05)	0.70 (0.46–1.04)
Education	University	1.00	1.00	1.00
	Graduate school	0.83 (0.65–1.06)	0.84 (0.65–1.07)	0.93 (0.72–1.20)
Marital status	Single	1.00	1.00	1.00
	Married	0.51 (0.40–0.65)^*	0.59 (0.45–0.77)^*	0.60 (0.46–0.79)^*
	Divorce	1.13 (0.67–1.88)	1.11 (0.66–1.88)	1.13 (0.67–1.87)
Psychosocial factors
Life event	≥1 life event		0.98 (0.76–1.26)	0.92 (0.71–1.18)
Social support	High (15–20)		0.32 (0.25–0.41)^*	0.36 (0.29–0.45)^*
Substance addictions
Problem drinking	Yes (CAGE ≥2)	The variables are not adjusted in Model 1	2.12 (1.50–2.99)^*	2.04 (1.42–2.88)^*
Current smoker	Yes (FTND ≥1)		1.39 (1.05–1.82)^*	1.35 (1.03–1.77)^*
Internet-related factors
Time on Internet (related to work)	0–4 hours/day		1.00	1.00
	≥5 hours/day		1.21 (0.96–1.52)	1.18 (0.94–1.49)
Time on Internet (not related to work)	0–3 hours/day		1.00	1.00
	≥4 hours/day	The variables are not adjusted in Model 1	2.26 (1.71–2.97)^*	2.11 (1.58–2.76)^*
Online games	Yes		2.04 (1.55–2.66)^*	1.91 (1.46–2.51)^*
Online social media	Yes		1.67 (1.32–2.11)^*	1.71 (1.35–2.17)^*
Potential mediators
Depression	Yes (BDI >=17)			1.60 (1.13–2.27)^*
Anxiety	Yes (BAI >=14)			1.71 (1.17–2.49)^*

Model 1: Ordinal logistic regression of IAT outcomes at wave 2 by job strain, work social support, and demographic factors at wave 1. Model 2: additionally adjusted for other confounders at wave 1. Model 3: additionally adjusted for potential mediators—depression and anxiety—at wave 1.

p<0.05.

Table 3 shows the associations between the ERI model and the IAT assessed by ordinal logistic regression. In Model 3 additionally adjusted for depression and anxiety, the odds of having a higher IAT outcome versus a lower outcome were 1.61 times larger for high ER ratio than low ER ratio, and 1.68 times larger for high OC than low OC. Finally, the LR test showed that ERI–OC interaction terms were not significant (p>0.1) in the three models.

Table 3.

Ordinal Logistic Regression for the Association Between the Effort–Reward Imbalance Model at Wave 1 and the Internet Addiction Test at Wave 2

Variable	Category	Model 1 OR (95% CI)	Model 2 OR (95% CI)	Model 3 OR (95% CI)
Effort–reward ratio	Low	1.00	1.00	1.00
	Medium	1.26 (0.92–1.72)	1.24 (0.89–1.71)	1.24 (0.89–1.71)
	High	1.84 (1.35–2.49)^*	1.76 (1.29–2.41)^*	1.61 (1.18–2.22)^*
	p for trend	p<0.001	p<0.001	p=0.002
Overcommitment	Low	1.00	1.00	1.00
	Medium	1.12 (0.84–1.50)	1.19 (0.88–1.61)	1.20 (0.89–1.62)
	High	1.65 (1.21–2.22)^*	1.76 (1.31–2.37)^*	1.68 (1.25–2.27)^*
	p for trend	p=0.002	p<0.001	p=0.001
Demographic factors
Sex	Female	1.00	1.00	1.00
	Male	1.23 (0.94–1.61)	1.13 (0.86–1.49)	1.23 (0.93–1.63)
Age	21–30	1.00	1.00	1.00
	31–40	0.64 (0.49–0.84)^*	0.73 (0.55–0.96)^*	0.71 (0.53–0.94)^*
	41–55	0.54 (0.40–0.76)^*	0.70 (0.47–1.04)	0.69 (0.47–1.03)
Education	University	1.00	1.00	1.00
	Graduate school	0.80 (0.62–1.01)	0.78 (0.59–1.01)	0.87 (0.67–1.12)
Marital status	Single	1.00	1.00	1.00
	Married	0.51 (0.40–0.66)^*	0.62 (0.47–0.82)^*	0.63 (0.48–0.83)^*
	Divorce	1.05 (0.63–1.75)	1.02 (0.60–1.72)	1.11 (0.66–1.86)
Psychosocial factors
Life event	≥1		0.91 (0.70–1.17)	0.86 (0.67–1.11)
Social support	High (15–20)		0.30 (0.24–0.40)^*	0.33 (0.26–0.43)^*
Substance addictions
Problem drinking	Yes (CAGE ≥2)	The variables are not adjusted in Model 1	2.18 (1.52–3.11)^*	2.16 (1.50–3.08)^*
Current smoker	Yes (FTND ≥1)		1.15 (0.87–1.52)	1.12 (0.85–1.49)
Internet-related factors
Time on Internet (related to work)	0–4 hours/day		1.00	1.00
	≥5 hours/day		1.21 (0.96–1.52)	1.18 (0.94–1.50)
Time on Internet (not related to work)	0–3 hours/day		1.00	1.00
	≥4 hours/day	The variables are not adjusted in Model 1	2.25 (1.72–2.96)^*	2.12 (1.60–2.79)^*
Online games	Yes		2.07 (1.57–2.73)^*	2.01 (1.54–2.63)^*
Online social media	Yes		1.79 (1.42–2.27)^*	1.82 (1.44–2.30)^*
Potential mediators
Depression	Yes (BDI ≥17)			1.41 (1.01–2.00)^*
Anxiety	Yes (BAI ≥14)			1.91 (1.32–2.79)^*

Model 1: Ordinal logistic regression of IAT outcomes at wave 2 by effort–reward ratio, overcommitment, and demographic factors at wave 1. Model 2: additionally adjusted for other confounders at wave 1. Model 3: additionally adjusted for depression and anxiety at wave 1.

p<0.05.

Table 4 shows the results of path analyses for mediating roles of depression/anxiety in work stress–IAT relationships; the DCS and ERI models were assessed separately. In the path analysis for the DCS model, job strain was associated with higher risks of depression (standardized β=0.057) and anxiety (standardized β=0.344) at wave 1. Depression predicted higher IAT scores at wave 2 (standardized β=0.176); anxiety predicted higher IAT scores at wave 2 (standardized β=0.139). High work social support was associated with lower risks of depression and anxiety. The extent of mediating (indirect) effect on total effect was estimated by path coefficient. The effects of job strain on the IAT were partially mediated by anxiety (0.048/0.122=39%), while the mediating effect of depression was not significant. The effects of work social support on the IAT were partially mediated by depression (29%) and anxiety (35%). The fit indexes for the path model were considered acceptable (RMSEA=0.077) or close to the cutoffs for good fit (CFI=0.86; WRMR=1.15).

Table 4.

Path Analyses for Potentially Mediating Roles of Depression and Anxiety in the Effects of Work Stress on Internet Addiction

Parameter	Unstandardized coefficient	Standard error	Standardized coefficient
1. Path analysis—DCS model and IAT
Total effect: Job strain→IAT	0.134^*	0.029	0.122
Direct effect: Job strain→IAT	0.070^*	0.037	0.064
Indirect effect: Job strain→Depression→IAT	0.012	0.010	0.010
Job strain→Depression	0.060^*	0.040	0.057
Depression→IAT	0.199^*	0.040	0.176
Indirect effect: Job strain→Anxiety→IAT	0.053^*	0.022	0.048
Job strain→Anxiety	0.363^*	0.045	0.344
Anxiety→IAT	0.145^*	0.057	0.139
Total effect: Work support→IAT	−0.180^*	0.045	−0.133
Direct effect: Work support→IAT	−0.062^*	0.057	−0.048
Indirect: Work support→Depression→IAT	−0.054^*	0.016	−0.039
Work support→Depression	−0.255^*	0.056	−0.213
Indirect effect: Work support→Anxiety→IAT	−0.064^*	0.027	−0.046
Work support→Anxiety	−0.439^*	0.068	−0.332
2. Path analysis—ERI model and IAT
Total effect: ERI→IAT	0.186^*	0.044	0.130
Direct effect: ERI→IAT	0.055^*	0.054	0.038
Indirect effect: ERI→Depression→IAT	0.074^*	0.023	0.052
ERI→Depression	0.391^*	0.055	0.301
Depression→IAT	0.194^*	0.045	0.176
Indirect effect: ERI→Anxiety→IAT	0.057^*	0.019	0.040
ERI→Anxiety	0.278^*	0.053	0.216
Anxiety→IAT	0.205^*	0.053	0.179
Total effect: OC→IAT	0.161^*	0.041	0.111
Direct effect: OC→IAT	0.101^*	0.044	0.070
Indirect effect: OC→Depression→IAT	0.029	0.012	0.020
OC→Depression	0.151^*	0.050	0.115
Indirect effect: OC→Anxiety→IAT	0.030	0.013	0.021
OC→Anxiety	0.147^*	0.052	0.117

All effects were adjusted for covariates. Total effect is the sum of all direct and indirect effects of one variable on another. Indirect effect is estimated as the product of path coefficients along this path.

p<0.05.

In the path analysis for the ERI model (Table 4), a high ER ratio was associated with higher risks of depression and anxiety at wave 1. Depression and anxiety at wave 1 predicted higher IAT scores at wave 2. High OC was associated with higher risks of depression and anxiety. The effects of the ERI model on the IAT were partially mediated by depression (0.052/0.130=40%) and anxiety (31%). However, the effects of OC on the IAT were not significantly mediated by depression or anxiety. The fit indexes for the path model were interpreted as acceptable (RMSEA=0.072) or close to the cutoffs for good fit (CFI=0.89; WRMR=1.04).

Discussion

To our knowledge, this is the first study to show that work stress defined by the DCS and ERI models can predict the risk of IA after 6.2 months of follow-up. In the DCS model, we found that job strain was associated with an increased risk of IA (Table 2); the odds ratio (OR) of having a higher IAT outcome versus a lower outcome was 1.53. High work social support was associated with a reduced risk of IA (OR=0.62). With regards to the ERI model, a high ER ratio (OR=1.61) and high OC (OR=1.68) were associated with increased risks of IA (Table 3). Path analyses showed that the effects of job strain on the IAT might be partially mediated by anxiety, but the mediating effect of depression was not significant. The effects of work social support and ER ratio on the IAT might be partially mediated by depression and anxiety. Mediating effects of depression and anxiety were not significant in OC–IAT relationships (Table 4).

It is suggested that work stress can lead to biological responses and depression/anxiety, which then cause behavioral addictions (IA) or substance addictions. Mounting evidence shows that behavioral addictions resemble substance addictions in natural history (e.g., onset in young age), phenomenology (e.g., withdrawal or tolerance), personality (e.g., novelty seeking), genetics, and neurobiological processes (e.g., serotonin involved with behavioral inhibition and emotional regulation, dopamine involved with motivation and salience of stimuli, serotonin transporter gene, or dopamine receptor gene).^14,34,35 With similar pathophysiology, IA and substance addictions may share a number of common risk factors, such as work stress.

Despite the lack of literature on work stress and IA, our findings are compatible with those empirical studies on work stress and substance addictions. Siegrist's review of the literature on the DCS model and drinking found that four out of six longitudinal studies supported that job strain predicted high alcohol consumption or alcohol dependence, particularly in men.³⁶ Albertson et al. reviewed 22 prospective studies and found strong evidence for the effect of job strain on high smoking intensity.³⁷ Similarly, work stress defined by a high ER ratio can predict increased risks of high alcohol consumption and high smoking intensity.^10,11,38

The Transactional Model of Stress proposed by Lazarus and Folkman suggests that when faced with a stressor in the environment, one evaluates the severity of potential threats (primary appraisal; job strain or ER ratio) and the availability of coping options and resources to deal with the stressor (secondary appraisal).³⁹ Adaptive coping strategies—such as receiving social support from the workplace, family, and friends—are found to reduce risks of substance/behavioral addictions.^2,40 Maladaptive coping strategies—such as a high OC person's frustrated attempts to regain or maintain control over environment—might increase risks of substance/behavioral addictions. Our findings for direct effect of OC on IA are consistent with the review by VanVegchel et al. Direct effect of OC was supported in 17 of 27 studies, but ERI–OC interaction was not supported in 9 of 12 studies.⁴¹ In future research on work stress–IA relationships, it will be of value to examine more individual specific coping strategies besides social support and OC personality.

Practical implications focus on workplace prevention and intervention for IA. First, an organizational-level approach can be adopted, for example to reduce extrinsic effort by even distribution of workloads and reduction of overtime works. Second, an individual-level approach can target causal chains from work stress to IA. For work stress, cognitive psychotherapy can target subjective evaluation of work stress; adaptive coping strategies can be developed to manage job demands and utilize work social support. For potential mediators, screening and treatment of depression and anxiety may prevent IA. Third, other risk factors for IA identified in this study (Table 2 and 3) such as low social support or Internet-related factors were generally consistent with the existing literature.^1,2 For example, time spent on the Internet “not related to work” (general and personal use) was associated with IA, but time spent “related to work” (professional use) was not. These robust risk factors outside work can be tackled with potentially available resources.

Several limitations of this study need to be taken into account. First, the outcome was measured in a single follow-up assessment after 6 months. While the assumptions on induction/latent periods between “work stress exposures” and “clinical diagnosis” of diseases were suggested based on psychiatric research,⁴² repeated measurements of the outcome and analyses by varying assumptions on induction/latent periods might provide better information on work stress–IA associations. Second, although the outcome was measured at wave 2, temporal ordering between exposures and mediators (both assessed at wave 1) cannot be clarified in this two-wave study. An alternative interpretation of our findings in path analysis is that work stress may mediate the effects of depression and anxiety on IA. Thus, a three-wave cohort study might produce more solid evidence for mediating roles of depression/anxiety.

Finally, it is unclear to what extent our findings can be generalized beyond this sample of IT engineers in Taiwan. The prevalence of IA in this sample, albeit higher than Western populations, is close to the prevalence reported in East Asian populations who often use the Internet.^43,44 This working population might be vulnerable to higher risks of IA for two reasons: first, with high accessibility, IA has become a major problem in East Asian countries; second, the characteristics of IT engineers (e.g., well-educated professionals, male predominance, or more time spent online) are also risk factors for IA. In future research, examining diverse working populations would ensure the generalizability of the current findings.

Footnotes

Author Disclosure Statement

No competing financial interests exist.

References

Byun

, Ruffini

, Mills

, et al. Internet Addiction: metasynthesis of 1996–2006 Quantitative Research. CyberPsychology & Behavior, 2009; 12:203–207.

Young

, de Abreu

. (2011) Internet Addiction: A Handbook and Guide to Evaluation and Treatment. Hoboken: John Wiley & Sons.

Block

. Issues for DSM-V: Internet addiction. American Journal of Psychiatry, 2008; 165:306–307.

American Psychiatric Association. (2013) Diagnostic and statistical manual of mental disorders. 5th ed. Arlington, VA: American Psychiatric Publishing.

Beard

. Internet addiction: current status and implications for employees. Journal of Employment Counseling, 2002; 39:211–212.

Karasek

, Theorell

. (1990) Healthy work: stress, productivity and the reconstruction of working life. New York: Basic Books.

Johnson

, Hall

. Job strain, workplace social support, and cardiovascular disease: a cross sectional study of a random sample of the Swedish working population. American Journal of Public Health, 1988; 78:1336–1342.

Siegrist

, Peter

. Job stressors and coping characteristics in work-related disease: issues of validity. Work & Stress, 1994; 8:130–40.

Stansfeld

, Candy

. Psychosocial work environment and mental health—a meta-analytic review. Scandinavian Journal of Work & Environmental Health, 2006; 32:443–462.

10.

Head

, Stansfeld

, Siegrist

. The psychosocial work environment and alcohol dependence: a prospective study. Occupational & Environmental Medicine, 2004; 61:219–224.

11.

Bobak

, Pikhart

, Kubinova

, et al. The association between psychosocial characteristics at work and problem drinking: a cross-sectional study of men in three Eastern European urban populations. Occupational & Environmental Medicine, 2005; 62:546–550.

12.

Blum

, Cull

, Braverman

, et al. Reward deficiency syndrome: addictive, impulsive and compulsive disorders may have a common genetic basis. American Scientist, 1996; 84:132–145.

13.

Holden

. Behavioral addictions debut in proposed DSM-V. Science, 2010; 327:935.

14.

Grant

, Potenza

, Weinstein

, et al. Introduction to behavioral addictions. American Journal of Drug & Alcohol Abuse, 2010; 36:233–241.

15.

Young

. Cognitive-behavioral therapy with Internet addicts: treatment outcomes and implications. CyberPsychology & Behavior, 2007; 10:671–679.

16.

Ferraro

, Caci

, D'Amico

, et al. Internet addiction disorder: an Italian study. CyberPsychology & Behavior, 2007; 10:170–175.

17.

Bakken

, Wenzel

, Gotestam

, et al. Internet addiction among Norwegian adults: a stratified probability sample study. Scandinavian Journal of Psychology, 2009; 50:121–127.

18.

Tsitsika

, Critselis

, Kormas

. Internet use and misuse: a multivariate regression analysis of the predictive factors of Internet use among Greek adolescents. European Journal of Pediatrics, 2009; 168:655–665.

19.

Susser

, Schwartz

, Morabia

, et al. (2006) Psychiatric epidemiology: searching for the causes of mental disorders. New York: Oxford University Press.

20.

Castells

. (2010) The Information Age: Economy, Society and Culture Volume 1: The Rise of the Network Society. 2nd ed. Oxford: Wiley Blackwell.

21.

Widyanto

, McMurren

. The psychometric properties of the Internet addiction test. CyberPsychology & Behavior, 2004; 7:445–453.

22.

Guertler

, Rumpf

, Bischof

, et al. Assessment of problematic Internet use by the compulsive Internet Use Scale and the Internet Addiction Test: a sample of problematic and pathological gamblers. European Addiction Research, 2014; 20:75–81.

23.

Cheng

, Luh

, Guo

. Reliability and validity of the Chinese version of the Job Content Questionnaire in Taiwanese workers. International Journal of Behavioral Medicine, 2003; 10:15–30.

24.

, Yang

, Cheng

, et al. Effort–reward imbalance at work and job dissatisfaction in Chinese healthcare workers: a validation study. International Archives of Occupational & Environmental Health, 2005; 78:198–204.

25.

Brugha

, Bebbington

, Tennant

, et al. The List of Threatening Experiences: a subset of 12 life event categories with considerable long term contextual threat questionnaire. Psychological Medicine, 1985; 15:189–194.

26.

Wang

. (2002) Evaluation of the effect of the self-care educational intervention and construction of self-care model in older adults. Taipei: National Science Council.

27.

Rush

, Pincus

, First

. (2009) Handbook of psychiatric measures. Washington, DC: American Psychiatric Association.

28.

Heatherton

, Kozlowski

, Frecker

. The Fagerström Test for Nicotine Dependence: a revision of the Fagerström Tolerance Questionnaire. British Journal of Addiction, 1991; 86:1119–1127.

29.

, Che

, Chang

, et al. Reliability and validity of the Chinese Version of the Beck Depression Inventory-II. Taiwanese J Psychiatry, 2002; 16:301–310.

30.

Che

, Lu

, Chen

, et al. Validation of the Chinese Version of the Beck Anxiety Inventory. Formosan Journal of Medicine, 2006; 10:447–454.

31.

Muthen

. A general structural equation model with dichotomous, ordered categorical and continuous latent indicators. Psychometrika, 1984; 49:115–132.

32.

Hoyle

. (2012) Handbook of structural equation modeling. New York: Guilford Press.

33.

. Evaluating cutoff criteria of model fit indices for latent variable models with binary and continuous outcomes. Doctoral dissertation, University of California, Los Angeles, 2002.

34.

Hamidovic

, Dlugos

, Skol

, et al. Evaluation of genetic variability in the dopamine receptor D2 in relation to behavioral inhibition and impulsivity/sensation seeking: an exploratory study with d-amphetamine in healthy participants. Experimental & Clinical Psychopharmacology, 2009; 17:374–383.

35.

Lee

, Han

, Yang

, et al. Depression-like characteristics of 5HTTLPR polymorphism and temperament in excessive internet users. Journal of Affective Disorders, 2009; 109:165–169.

36.

Siegrist

, Rödel

. Work stress and health risk behavior. Scandinavian Journal of Work & Environmental Health, 2006; 32:473–481.

37.

Albertsen

, Borg

, Oldenburg

. A systematic review of the impact of work environment on smoking cessation, relapse and amount smoked. Preventive Medicine, 2006; 43:291–305.

38.

Kouvonen

, Kivimäki

, Virtanen

, et al. Work stress, smoking status, and smoking intensity: an observational study of 46190 employees. Journal of Epidemiology & Community Health, 2005; 59:63–69.

39.

Lazarus

, Folkman

. (1984) Stress, appraisal and coping. New York: Springer.

40.

Niedhammer

, Goldberg

, Leclerc

, et al. Psychosocial work environment and cardiovascular risk factors in an occupational cohort in France. Journal of Epidemiology & Community Health, 1998; 52:93–100.

41.

VanVegchel

, DeJonge

, Bosma

, Schaufeli

. Reviewing the effort–reward imbalance model: drawing up the balance of 45 empirical studies. Social Science & Medicine, 2005; 60:1117–1131.

42.

Ehlers

, Harvey

. (2000) Acute stress reactions. In Gelder

, Lopez-Ibor

, Andreasen

, eds. New Oxford Textbook of Psychiatry. 1st ed. Oxford: Oxford University Press, pp. 751–757.

43.

Bai

, Lin

, Chen

. The characteristic differences between clients of virtual and real psychiatric clinics. American Journal of Psychiatry, 2001; 158:1160–1161.

44.

Whang

, Lee

, Chang

. Internet over-user's psychological profiles: a behavior sampling analysis on Internet addiction. CyberPsychology & Behavior, 2003; 6:143–150.