Does Religion Suppress Internet Addiction? An Instrumental Variable Approach Using Data from China

Abstract

Internet addiction has become a significant concern due to its detrimental effects on individual well-being, leading to emotional, psychological, and social challenges. Scholars have long explored various strategies to mitigate the risk of internet addiction. Recently, scholars have argued that religion is a protective factor against internet addiction. However, empirical research has reported a mix of negative and nonsignificant main effects. In this study, we used an instrumental variable quantile analysis to re-estimate the impact of religion (religious belief and religiosity) on internet addiction (generalized internet addiction and short video addiction) based on data from a nationally representative survey and (N = 2337; M_age = 42.03, SD = 14.15, range = 18 to 86) an online survey (N = 441, M_age = 28.98, SD = 7.59, range = 18 to 59). The results indicated that religious belief could suppress generalized internet addiction and short video addiction when endogeneity was not considered. However, when endogeneity was taken into account, the impacts of religious belief on generalized internet addiction and short video addiction were not significant. In addition, the impact of religiosity on short video addiction changed from significant to nonsignificant when endogeneity was considered. Our findings revealed that the protective effect of religion on internet addiction was very limited. Our study also provides a possible explanation for the existing mixed conclusions about religion and internet addiction.

1. Introduction

The development of digital technology was among the most profound events influencing humanity in the late 20th century. With the innovation of digital technology, the internet has become an indispensable tool for human existence and even a second life apart from daily real life (Whitty and McLaughlin 2007). There were 5.44 billion internet users worldwide, which amounted to 67.1% of the global population (Petrosyan 2024). Internet use facilitates learning, communication, and information transmission. Despite the obvious benefits of internet use, the high profitability (e.g., satisfaction) and low cost of internet use may make it easy for individuals to excessively use the internet and develop addictive behaviors (Tao et al. 2010). Internet addiction (IA) can be defined as the uncontrollable and problematic use of internet information technologies in such a way that it disrupts different life domains (Davis 2001; Pan and Yeh 2018; Poon 2018). IA encompasses both generalized and specific forms (Montag et al. 2015). Generalized IA refers to the excessive use of the internet covering a broad range of internet-related activities. Specific IA comprises strongly directed online activities, such as internet gaming addiction, social media addiction, and short video addiction. Studies have shown that engaging in beneficial online activities can promote learning, education, work, and daily functioning, thereby positively influencing well-being and mental health (Castellacci and Tveito 2018; Lippke et al. 2021). However, once IA occurs—whether generalized or specific—it can severely affect individuals’ physical and mental health (Aghasi et al. 2020; Huang 2017; Ko et al. 2012), leading to a significant negative impact on their well-being (Çikrıkci 2016). For example, a meta-analysis with 223 studies showed that prolonged and uncontrolled internet use was moderately and positively associated with loneliness, anxiety, and depressive symptoms, and negatively related to subjective well-being (Cai et al. 2023).

To enhance people’s well-being and mental health in the digital age, scholars have long been interested in understanding the risk and protective factors for IA (Choi et al. 2015; Lin et al. 2018; Lozano-Blasco et al. 2022). Recently, scholars have begun to recognize the role of religion in preventing IA (Braun et al. 2016; Charlton et al. 2013; Grubbs et al. 2017; Nadeem et al. 2019). However, there is no consensus on the relationship between religion and IA. One line of research suggested a negative correlation between religion and IA. For example, based on a sample of 350 Israeli–Palestinian Muslim college students from Israel, Agbaria and Bdier (2019) revealed a negative association between religiosity and IA. This finding is corroborated by a study involving 800 Muslim college students from Pakistan, which found that intrinsic religious orientation was linked to reduced IA, while anti-religious activities increased IA (Nadeem et al. 2019). Another study using participants from Korea indicated that teens in the high-risk group for smartphone addiction had lower levels of spiritual well-being and positive images of God than did teens in the potential risk and control groups (Shim 2019). In addition, a previous study established a negative association between religious belief and online gaming addiction (Braun et al. 2016). The question arises as to why there might be a negative association between religion and IA. Previous studies have shown that faith-based and spiritual approaches help people feel grounded, calm, and present in difficult environments (Ozcan et al. 2021). Religious individuals are more likely to cultivate generosity, gratitude, and optimism (Koenig et al. 2014). Religion serves as a meaning-making system. People who embrace religious beliefs often develop a sense of meaning, which helps them understand the world, the self, and their relationships with others. This may foster a sense of purpose in life, a feeling of control over the future, and positive emotional states, thereby enhancing overall well-being (Van Cappellen et al. 2016). IA has been described as a maladaptive coping strategy employed to alleviate stress stemming from physically or mentally taxing events. Those with strong religious convictions may be more adept at managing life’s pressures and challenges, possessing a heightened sense of self-efficacy (van Olphen et al. 2003). Thus, they could be less likely to resort to escaping into a self-created virtual online world.

Other evidence has shown that religion may not play a significant role in mitigating IA. For example, a study using a sample of 389 high school students demonstrated no statistically significant correlation between religion and IA (Ekşi and Ciftci 2017). Based on a nationally representative Polish sample, Lewczuk et al. (2021) found that while religiosity was significantly associated with pornography addiction, it did not exhibit a substantial link to generalized IA, social networking addiction, or online video game addiction.

These inconsistent conclusions drawn from various studies indicate a nuanced relationship between religion and IA (Dossi et al. 2022). The substantial variation in these conclusions may stem from several factors. One possible reason is that existing studies do not consider the endogeneity of religion, which leads to estimation bias and an inability to estimate the causal effect of religion on IA. The endogeneity of the existing research regarding the impact of religion on IA might arise from omitted variable bias and simultaneity. Specifically, omitted variable bias arises when some unobserved heterogeneity that might simultaneously influence both religion and IA is not controlled for. For instance, religion and IA may be simultaneously influenced by insecurity and socialization pressure (Arslan and Coşkun 2022; Mercier et al. 2018). If not properly controlled for, omitted variable bias may lead to an incorrect causal relationship between religion and IA. This simultaneity means that causality may also run from IA to religion. For example, McClure (2017) found that internet use significantly influenced religious beliefs, indicating that the relationship between the two is not unidirectional. The effect of religion on IA can be estimated more accurately when endogeneity is controlled for.

To address potential endogeneity and determine the causal relationship between religion and IA, this study used data from a nationally representative survey and an online survey to re-estimate the impact of religion on IA based on instrumental variable (IV) quantile analysis. Specifically, in Study 1, we used a nationally representative survey to examine the impact of religious belief on generalized IA. In Study 2, we used an online survey to investigate the impacts of religious belief and religiosity on a specific form of IA (i.e., short video addiction).

2. Study 1

2.1. Method

2.1.1. Participants

To examine the impact of religious belief on generalized IA, we obtained data from the China General Social Survey (CGSS). Initiated in 2003, the CGSS is the longest-standing continuous national representative survey project encompassing both urban and rural households. For our study, we focused on the 2017 wave of the CGSS, as this particular survey year investigated both religious belief and generalized IA. The CGSS 2017 conducted interviews with 12,582 individuals across 28 provinces. Within the pool, a subset of 4219 respondents was randomly selected to participate in the Social Network Survey module, which inquired about internet usage. After removing the observations that had never used the internet and those with missing values, we obtained a final sample of 2337. All participants were adults aged between 18 and 86 years (M_age = 42.03, SD = 14.15).

2.1.2. Measures

Generalized internet addiction. In the CGSS 2017, a list of 11 statements was designed to measure respondents’ generalized IA. The respondents were instructed to assess their agreement with these statements using a five-point scale (1 = strongly disagree, 5 = strongly agree). The statements were derived from the Chinese Internet Addiction Scale (Chen et al. 2003) and covered various symptoms (e.g., “I often spend more time on the internet than I plan to”) and adverse effects of IA (e.g., “Because of using the internet, my shoulders and cervical vertebrae are sore”). The scores of the 11 statements were averaged to create a composite measure of generalized IA. This scale has been used in previous studies to measure generalized IA (Dong et al. 2023; Wu et al. 2023). In the present study, the scale had good reliability (Cronbach’s alpha = 0.87).

Religious belief. Following the research by Zeng et al. (2020), religious belief was measured by a single question that asked whether the respondents had a religious belief or not. We coded religious belief as a dummy variable that took a value of 0 if the respondents did not believe in any religion and 1 if the respondents had a religious belief.

Control variables. Several demographic variables were included in the estimated models as control variables, including age, gender, education, marital status, employment, ethnicity, social class, and Hukou status. Age was calculated by subtracting the birth year from the survey year. Gender was coded as 0 if the respondent was female; otherwise, it was coded as 1. Education was determined based on the highest educational qualification reported in the survey year and was coded as follows: 1 = primary school education and below, 2 = junior high school education, 3 = senior high school education, 4 = college education, and 5 = postgraduate education and above. Marital status was coded as 0 = single, 1 = married/cohabiting, 2 = divorced, and 3 = widowed. Employment was coded as 0 = retirement/unemployed, 1 = agricultural work, and 2 = nonagricultural work. Ethnicity was coded as 0 = minority and 1 = Han Chinese. Social class was measured by the MacArthur Scale of Subjective Social Status, represented by a ten-rung ladder (1 = lowest, 10 = highest). The respondents positioned themselves at the appropriate point on this ladder. Hukou status was coded as 0 if the respondent had rural Hukou status; otherwise, it was coded as 1.

2.1.3. Analytical Strategy

Our empirical analysis consisted of two steps, which we will now elaborate on in further detail. First, we employed ordinary least squares (OLS) to examine the association between religious belief and generalized IA. The specified model is outlined as follows:

$${IA}_i={\beta }_0+{\beta }_1{RB}_i+{\beta }_2{X}_i+{\epsilon }_i,$$

(1)

where ${IA}_i$ is the generalized IA of respondent i, ${RB}_i$ is religious belief, $X_i$ is a vector of control variables, ${\beta }_0$, ${\beta }_1$, and ${\beta }_2$ are parameters to be estimated, and ${\epsilon }_i$ is the error term. The potential endogeneity of our independent variable posed a challenge to identifying a causal relationship between religious belief and generalized IA in traditional OLS estimation. This endogeneity can be attributed to omitted variable bias and simultaneity.

To address the possible endogeneity issue and more accurately establish a causal association between religious belief and generalized IA, we used the two-stage residual inclusion (2SRI) method, also known as the control function approach. 2SRI is a widely employed econometric approach that addresses the endogeneity of independent variables and enables researchers to make causal inferences in models with binary endogenous independent variables (Wooldridge 2015). 2SRI consisted of two stages. In the first stage, we specified religious belief as a function of all control variables and an IV. Since religious belief was a binary variable, we used a probit model in the first stage:

$${RB}_i^{*}=a_1{IV}_i+a_2{X}_i+{\rho }_i, {RB}_i=\left\{\begin{array}{c}1, if {RB}_i^{*}\ge 0\\ 0, if {RB}_i^{*}<0\end{array}\right.$$

(2)

where ${RB}_i^{*}$ is a latent variable indicating the likelihood of respondent i having religious belief, ${RB}_i^{*}$ is determined by the observed dichotomous variable ${RB}_i$, $a_1$ and $a_2$ are unknown parameters to be estimated, and ${\rho }_i$ is the error term. IV refers to the instrumental variable of religious belief. A valid IV necessitates a strong correlation with the endogenous variable (meeting the relevance condition) while having no direct effect on the dependent variable (satisfying the exclusion condition). In this study, we used the number of religious venues in a specific province as the IV. A religious institution (e.g., church, temple, or mosque) is a nonprofit organization established with the purpose of practicing and promoting a specific religious belief. In the context of China, religious activities are predominantly organized and overseen by these religious institutions. Therefore, the greater the number of religious venues in a given province is, the more likely the people in that province are to have religious beliefs (Xu et al. 2017). Nevertheless, the number of religious venues in a given province obviously does not directly impact generalized IA. The data on religious venues were obtained from the Center on Religion and Chinese Society at Purdue University (Yang et al. 2019). The natural logarithm of the number of religious venues in a given province was used to reduce the skewness.

Following the first-stage estimation, we obtained the predicted generalized residuals of Equation (2). Subsequently, we included the generalized residuals in Equation (1) as a control. The incorporation of generalized residuals can address the endogeneity problem related to religious belief (Wooldridge 2015).

2.2. Results

Columns 1 and 2 in

Table 1. Demographic characteristics.

Variable	Sample 1		Sample 2
	N/M ± SD	%/Range	N/M ± SD	%/Range
Religious belief
With religious belief	215	9.21	132	29.93
No religious belief	2119	90.79	309	70.07
Religiosity			2.32 ± 0.83	1–4.73
Internet/short video addiction	2.66 ± 0.74	1–5	4.01 ± 1.45	1–6.67
Age	42.08 ± 14.12	18–86	28.98 ± 7.59	18–58
Gender
Female	1174	50.30	290	65.76
Male	1160	49.70	151	34.24
Education	2.80 ± 1.05	1–5	4.16 ± 0.49	2–5
Marital status
Single	373	15.98	195	44.22
Married/Cohabiting	1814	77.72	246	55.78
Divorced	87	3.73
Widowed	60	2.57
Employment
Retirement/unemployed	783	33.55
Agricultural work	195	8.35
Non-agricultural work	1356	58.10
Ethnicity
Minority	154	6.60
Han Chinese	2180	93.40
Social class	2.34 ± 0.86	1–5
Income
≤2000			166	37.64
2001–5000			147	33.33
5001–10,000			94	21.32
≥10,001			34	7.71
Hukou status
Rural	964	41.30	69	15.65
Urban	1370	58.70	372	84.35

Note. Sample 1 = 2334, Sample 2 = 441. Sample 1 was extracted from the Chinese General Social Survey. Sample 2 was collected by an online survey.

display the descriptive characteristics of the sample used in Study 1. The mean age of the respondents was 42.03 years (SD = 14.15, range = 18–86), and 50.41% of the respondents (N = 1178) were female. Among the 2337 respondents, 9.16% (N = 214) had religious beliefs. The mean generalized IA was 2.66 (SD = 0.74).

Table 2. Results of the ordinary least-squares and two-stage residual inclusion analyses of the impact of religious belief on internet addiction.

Variables	OLS	2SRI
		First Stage	Second Stage
Religious belief	−0.103 *		0.137
	(0.052)		(0.342)
Number of religious sites		0.094 **
		(0.029)
Age	−0.013 ***	0.006	−0.014 ***
	(0.001)	(0.004)	(0.001)
Gender	−0.009	−0.153 *	−0.004
	(0.030)	(0.077)	(0.031)
Education	0.085 ***	−0.199 ***	0.094 ***
	(0.017)	(0.044)	(0.021)
Marital status
Married/Cohabiting	−0.151 **	−0.083	−0.149 **
	(0.049)	(0.137)	(0.049)
Divorced	−0.051	0.034	−0.054
	(0.089)	(0.219)	(0.089)
Widowed	−0.207	−0.065	−0.207
	(0.109)	(0.266)	(0.109)
Employment
Agricultural work	−0.078	−0.453 *	−0.063
	(0.060)	(0.187)	(0.064)
Non-agricultural work	−0.000	0.131	−0.006
	(0.034)	(0.090)	(0.035)
Ethnicity	0.010	−1.127 ***	0.076
	(0.060)	(0.114)	(0.111)
Social class	−0.002	0.039	−0.003
	(0.017)	(0.049)	(0.017)
Hukou type	0.019	0.199 *	0.012
	(0.037)	(0.095)	(0.038)
Residual			−0.125
			(0.176)

Note. N = 2334. OLS = Ordinary least-squares, 2SRI = Two-stage residual inclusion. * p < 0.1, ** p < 0.05, *** p < 0.01.

presents the OLS and 2SRI estimation results. Specifically, the OLS estimation results are displayed in Column 1 of Table 2. We started by introducing the results for the control variables. Age showed a negative association with generalized IA, and education was positively correlated with generalized IA. Respondents who were married/cohabiting reported significantly lower levels of generalized IA. For the main results, religious belief was significantly and negatively associated with generalized IA (β = −0.102, p < 0.05), indicating that respondents with religious belief had significantly lower levels of generalized IA than those without religious belief. Nevertheless, due to the potential statistical issues caused by omitted variable bias and reverse causality, it was not appropriate to interpret this association as causal. To enhance our understanding of the causal connection between religious belief and generalized IA, we now turn to the estimation results derived from 2SRI.

Columns 2 and 3 show the estimation results for 2SRI. For the first-stage regression, we found that the number of religious venues had a significant positive effect on the likelihood of being religious (β = 0.097, p < 0.001). The significant coefficient of the IV indicated that the IV was valid (Gielens et al. 2021). After taking into account the endogeneity of religious belief, the estimation results of the second-stage regression indicated that religious belief had no significant impact on generalized IA (β = 0.101, p = 0.766).

In addition, we further explored the role of religious involvement. The results revealed that the degree of religious involvement was not significantly associated with IA (β = −0.014, p = 0.315).

2.3. Summary

In Study 1, we employed both OLS and 2SRI to estimate the impact of religious belief on generalized IA. Our initial analysis utilizing OLS indicated that religious belief significantly inhibited generalized IA. However, when endogeneity was considered through the 2SRI method, the impact of religious belief on generalized IA was not significant.

Although we investigated the impact of religious belief on IA in Study 1, two limitations should be noted. First, due to the constraints of the data, we only explored the causal relationship between religious belief and IA without examining the impact of religiosity on IA. Second, the form of IA measured in Study 1 was generalized IA. With the rapid advancement and widespread availability of internet technology, different types of IA are emerging (e.g., social media addiction, smartphone addiction, and internet gaming disorder). Short video addiction, as a specific form of IA, has become a prominent and salient issue in contemporary society (Mu et al. 2024). To address these limitations, Study 2 was designed with a revised approach. We recollected data through an online survey, which allowed a more nuanced examination of the relationship between religiosity and various forms of IA, including the emerging issue of short video addiction.

3. Study 2

3.1. Method

3.1.1. Participants and Procedure

The ethical considerations for this study were rigorously reviewed and approved by the Human Subjects Ethics Sub-Committee of Wuhan University To recruit adult participants, we utilized the dataset marketplace of the Credamo platform “https://www.credamo.com (accessed on 2 September 2024)”, an online integrated data platform used in China for data collection. The data collection process took place in October 2024. We began the survey by presenting participants with an informed consent form. By clicking on “Continue to participate”, the participants agreed with the terms outlined in the consent form. In the survey, we included an attention-check item to screen for careless responses. Participants in the online survey were asked to answer each question before submission, thereby guaranteeing the absence of missing values. Upon completion, each participant was awarded compensation of 2 RMB (approximately $0.28).

The initial group of participants included 465 individuals. However, 24 participants were excluded from the study due to failing attention checks. One participant who reported an incorrect age was removed. This exclusion resulted in a final dataset of 441 participants, consisting of 290 women (65.76%) and 151 men (34.24%). The participants’ ages varied from 18 to 59 years, with an average age of 28.98 years (SD = 7.59).

3.1.2. Measures

Short video addiction. The 6-item Short Video Addiction Scale (SVAS-6, Zhang et al. 2019) was adopted to measure SVA. The SVAS-6 was adapted from the Social Network Service Addiction Scale (Choi and Lim 2016) and has been widely used in previous studies (Mu et al. 2024; Zhang et al. 2019). An example item was “I feel anxious if I cannot access this short video app”. Participants responded on a 7-point scale (1 = strongly disagree, 7 = strongly agree). In this study, the SVAS-6 had good reliability (Cronbach’s alpha = 0.90).

Religious belief. The measure of religious belief used was consistent with that used in Study 1.

Religiosity. Religiosity was measured by the Centrality of Religiosity Scale (CRS, Huber and Huber 2012). The CRS measures religiosity in terms of five core dimensions: public practice, private practice, ideology, intellect, and religious experience. Each dimension had three items, and participants rated the 15 items on a 5-point scale (1 = never/not important/very unlikely/not at all interested, 5 = very often/very important/very likely/very interested). A higher mean score suggested a greater degree of religiosity. Cronbach’s alpha for the CRS was 0.96, indicating excellent reliability.

Control variables. Similar to Study 1, we controlled for several demographic variables, including age, gender, education, marital status, income, and Hukou status. The measures of age, gender, education, and Hukou status were consistent with those used in Study 1. Marital status was coded as 0 = single and 1 = married/cohabiting. Income refers to the monthly income of the participants and was coded as 1 = less than or equal to 2000 RMB, 2 = 2001–5000 RMB, 3 = 5001–10,000 RMB, and 4 = equal to or greater than 10,000 RMB.

3.1.3. Analytical Strategy

To estimate the impact of religiosity on SVA, we first used the OLS approach. Subsequently, we used two-stage least squares (2SLS) to estimate causality. The reason for using 2SLS was that the endogenous variable (i.e., religiosity) was a continuous variable rather than a discrete variable. Unlike 2SRI, 2SLS used the predicted values from the first stage as the independent variable in the second stage. Certainly, 2SRI can also address causal estimation when the endogenous variable is continuous. Therefore, we used 2SRI to test the robustness of the results. We similarly used the parent’s religious belief as an IV.

3.2. Results

Columns 3 and 4 in Table 1 present the descriptive characteristics of the sample used in Study 2. Among the 441 participants, 29.93% (N = 132) had religious beliefs. The mean religiosity was 2.32 (SD = 0.83), and the mean SVA was 4.01 (SD = 1.45).

Table 3. Results of the ordinary least-squares and two-stage least squares analyses of the impact of religious belief on short video addition.

Variables	OLS	2SLS
		First Stage	Second Stage
Religious belief	−0.478 **		−0.103
	(0.149)		(0.240)
Parent’s religious belief		1.835 ***
		(0.156)
Age	−0.014	0.029 *	−0.018
	(0.012)	(0.013)	(0.012)
Gender	−0.315 *	−0.120	−0.291 *
	(0.139)	(0.162)	(0.139)
Education	0.157	−0.059	0.182
	(0.141)	(0.150)	(0.141)
Marital status	−0.260	0.348	−0.294
	(0.183)	(0.209)	(0.183)
Income	−0.180 *	−0.036	−0.193 *
	(0.089)	(0.097)	(0.089)
Hukou type	−0.161	−0.368	−0.125
	(0.191)	(0.211)	(0.191)
Residual			−0.337 *
			(0.170)

Note. N = 441. OLS = Ordinary least-squares, 2SLS = two-stage least squares. * p < 0.1, ** p < 0.05, *** p < 0.01.

displays the OLS and 2SLS estimation results for the association between religious belief and SVA. Specifically, the OLS estimation results (Column 1 in Table 3) showed a significant and negative association between religious belief and SVA (β = −0.478, p < 0.001), indicating that participants with religious belief had significantly lower levels of SVA than those without religious belief. Columns 2 and 3 in Table 3 present the estimation results of 2SRI. The significant positive effect of parents’ religious belief on participants’ religious belief (β = 1.835, p < 0.001) demonstrated that the IV was valid. After accounting for the endogeneity of religious belief, the estimation results of the second-stage regression indicated that religious belief had no significant impact on SVA (β = −0.103, p = 0.669).

Table 4. Results of the ordinary least-squares and two-stage least squares analyses of the impact of religiosity on short video addition.

Variables	OLS	2SLS
		First Stage	Second Stage
Religiosity	−0.274 ***		−0.032
	(0.081)		(0.168)
Parent’s religious belief		0.816 ***
		(0.080)
Age	−0.014	0.014 *	−0.019
	(0.012)	(0.013)	(0.012)
Gender	−0.308 *	−0.031	−0.287
	(0.138)	(0.075)	(0.147)
Education	0.206	0.135	0.191
	(0.141)	(0.081)	(0.145)
Marital status	−0.261	0.114	−0.298
	(0.183)	(0.101)	(0.191)
Income	−0.209 *	−0.107	−0.198 *
	(0.089)	(0.050)	(0.088)
Hukou type	−0.133	−0.026	−0.117
	(0.190)	(0.085)	(0.191)

Note. N = 441. OLS = Ordinary least-squares, 2SLS = two-stage least squares. * p < 0.1, *** p < 0.01.

displays the OLS and 2SLS estimation results for the association between religiosity and SVA. The results of OLS estimation indicated that religiosity was significantly and negatively associated with SVA (β = −0.274, p < 0.001). The results of the first-stage estimation of the 2SLS model indicated that parents’ religious belief was significantly and positively correlated with participants’ religiosity (β = 0.816, p < 0.001). In addition, the Cragg–Donald Wald F statistic was 124.45, which was greater than 10 (Staiger and Stock 1997). These results strongly support the relevance of the IV. After controlling for the endogeneity of religiosity, we found that religiosity had no significant impact on SVA (β = −0.032, p = 0.851). The results of 2SRI showed consistent findings.

3.3. Summary

Based on data from an online survey, Study 2 further examined the impact of religious belief on a specific form of IA (i.e., SVA). Additionally, we explored the impact of religiosity on SVA in Study 2. The results indicated that both religious belief and religiosity could significantly curb SVA when the OLS approach was used. However, religious belief and religiosity had no meaningful impact on SVA when endogeneity was taken into account.

4. Discussion and Conclusions

Religion is a fundamental social component, as it holds a prominent position among the most extensive and influential social structures, impacting people’s attitudes and behavioral patterns both individually and collectively (Nadeem et al. 2019). Religion constructs a system of meaning, and those who hold religious beliefs often experience an unprecedented sense of fulfillment and well-being, along with enhanced perceived control over life and reduced stress. Given the negative impact of IA on well-being, some studies have begun to investigate the role of religion in relation to Internet addiction and regard it as an important preventive measure (Mestre-Bach et al. 2021). However, empirical studies have not reached a consensus on the relationship between religion and IA. We argue that the mixed findings of negative and null results might be due to the failure to control for endogeneity between religion and IA.

This paper was designed to examine the causal relationship between religion and IA using samples from China. In Study 1, we estimated the impact of religious belief on generalized IA using OLS and 2SRI. Initial findings from Study 1 suggested that religious belief significantly inhibited IA when endogeneity was not considered. However, when endogeneity issues were considered, the impact of religion on IA was not significant. To provide more evidence for the impact of religion on IA, Study 2 was designed to explore the effects of religious belief and religiosity on a specific form of IA (i.e., SVA). The results indicated that both religious belief and religiosity could significantly curb SVA when the OLS approach was used. However, when controlling for endogeneity, the effects of religious belief and religiosity on SVA were rendered nonsignificant.

The results indicated that religion did not have a significant impact on either generalized IA or specific IA (i.e., SVA) when endogeneity was considered, which implied that there was no relationship between religion and IA in China. We believe this finding may have been reached for the following reasons. First, we acknowledge that a faith-based and spiritual approach helps people feel grounded, calm, resilient, and present in difficult environments (Ozcan et al. 2021). The relatively low level of religiosity in China limits the capacity of religion to exert substantial influence on values and IA, especially in the face of the strong allure of the internet. Second, the separation between religious identity and religious practice is a common phenomenon. Having a religious identity does not necessarily imply a belief in the existence of God or deities; rather, it often reflects a desire to seek peace and blessings through religious affiliation. Therefore, individuals with a religious identity may not adhere to the moral precepts of religion, thus diminishing its efficacy in preventing IA. Third, in the course of modernization, most countries in the world today are secularized rather than religious (Berger 1997). Although religion has historically played an important role, its influence has gradually diminished (Zeng et al. 2021), particularly with regard to its impact on addictive behavior. In a sense, religion, as a cultural factor, is insufficient as a means to help those with religious beliefs to resist the temptations that come with internet use, although it can have a subtle influence on human behavior and well-being.

Overall, the conclusion implies that the influence of religion on IA is quite limited. Existing studies also indirectly confirm that the positive effects of religion on well-being and mental health may operate through alternative pathways, rather than through its impact on IA behaviors.

5. Limitations

As with any empirical endeavor, our study is not free of limitations. First, although we used multiple data sources to test the impacts of religious belief and religiosity on generalized IA and SVA, our findings were limited to the Chinese context. Therefore, future research should delve more deeply into the impacts of religious belief and religiosity on IA across cultural settings. Second, we explored only the direct impacts of religious belief and religiosity on IA. Although the direct effects of religious belief and religiosity on IA were not significant after controlling for endogeneity, future research could further explore potential indirect pathways through which religious belief and religiosity impact IA. Third, our study used self-reported instruments to measure IA, which might have led to additional measurement errors. Therefore, future research might benefit from using other-rated IA or objective indicators of IA (e.g., the duration of internet usage or short video usage). Fourth, this study did not include religious groups that restrict access to technology, either through (a) bans on internet use imposed by religious elders, (b) national restrictions on technology access through infrastructural blocking, or (c) religious rituals such as prayers or days of rest that limit opportunities to use the internet. Excluding these groups may bias the measurement of religiosity, and future research should consider incorporating them to re-examine the impact of religiosity on IA. Finally, our study mainly examined whether religious belief and religiosity affected IA. Future research would benefit from assessing whether different forms of religious belief have differential impacts on IA.