Does internet use benefit the mental health of older adults? Empirical evidence from the China health and retirement longitudinal study

Zhang, Lixia; Li, Shaoting; Ren, Yanjun Article — Published Version Does internet use benefit the mental health of older adults? Empirical evidence from the China health and retirement longitudinal study Heliyon Provided in Cooperation with: Leibniz Institute of Agricultural Development in Transition Economies (IAMO), Halle (Saale) Suggested Citation: Zhang, Lixia; Li, Shaoting; Ren, Yanjun (2024) : Does internet use benefit the mental health of older adults? Empirical evidence from the China health and retirement longitudinal study, Heliyon, ISSN 2405-8440, Elsevier, Amsterdam, Vol. 10, Iss. 3, pp. 1-15, https://doi.org/10.1016/j.heliyon.2024.e25397 , https://www.sciencedirect.com/science/article/pii/S2405844024014282 This Version is available at: https://hdl.handle.net/10419/282018 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/4.0/ Heliyon 10 (2024) e25397 Available online 1 February 2024 2405-8440/© 2024 Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Research article Does internet use benefit the mental health of older adults? Empirical evidence from the China health and retirement longitudinal study Lixia Zhang a , b , Shaoting Li a , Yanjun Ren a , c , * a College of Economics and Management, Northwest A&F University, 3 Taicheng Rd, Yangling 712100, Shaanxi, China b Six Industrial Research Institute, Northwest A&F University, 3 Taicheng Rd, Yangling 712100, Shaanxi, China c Leibniz Institute of Agricultural Development in Transition Economies (IAMO), Theodor-Lieser-Str. 2, 06120, Halle (Saale), Germany ARTICLE INFO Keywords: Internet use Mental health Older adults China ABSTRACT The mental health (MH) of older adults is a prominent public health concern. However, research regarding the impact of emerging Internet use on MH among older adults remains limited, particularly in transitional economies experiencing a rapidly aging population such as China. Thus, to address this research gap, this study uses data from the 2013–2018 waves of the China Health and Retirement Longitudinal Study. To investigate the causal relationship between Internet use and MH among older adults and explore the underlying channels through which this relationship operates. The results reveal a notable positive association between Internet use and MH among older adults. Furthermore, the study highlights social interaction, social trust, traveling expenses, and healthy habits as crucial channels through which Internet use can impact MH among older adults. The analysis also reveals how Internet use demonstrates a stronger positive effect on older individuals who have fewer chronic diseases and live with their offspring compared with their counterparts. These findings have significant policy implications, which thus emphasizes the need to enhance Internet use among older adults as a means of improving their MH. 1. Introduction An increasingly aging population has emerged as a substantial challenge across the world; consequently, promoting healthy aging to maintain a high quality of life and independence has become a global concern. The proportion of individuals aged 65 years and above in developed regions is projected to rise from 14.3 % in 2000 to 25.9 % in 2050 [1]. With the advancement of Internet technology, this proportion will grow rapidly. The older population becomes increasingly vulnerable to various health risks and diseases as their life expectancy increases [2]. The existing statistics reveal that the prevalence rates of chronic diseases, disability rates, and the per capita consumption of health resources among older adults are 3.2, 3.6, and 1.9 times higher than those of the total population, respectively [3]. This situation is particularly challenging for China because of its rapid population growth compared with other developing countries. By the end of 2020, the population aged 60 or above in China will reach 264 million, with an aging rate of 18.70 % [4]. The current projections indicate that the proportion of the population over 60 years old in China will reach 34.9 % by 2050, * Corresponding author. College of Economics and Management, Northwest A&F University, 3 Taicheng Rd, Yangling 712100, Shaanxi, China. E-mail address: [email protected] (Y. Ren). Contents lists available at ScienceDirect Heliyon journal homepage: www.cell.com/heliyon https://doi.org/10.1016/j.heliyon.2024.e25397 Received 19 June 2023; Received in revised form 22 January 2024; Accepted 25 January 2024 Heliyon 10 (2024) e25397 2 ushering the country into a stage of profound aging. Older adults have placed considerable pressure on the healthcare system [5,6]. However, mental health (MH) issues are more serious and less noticeable than physical health problems in older adults. Statistically, half of the adults aged 65–84 years old have experienced MH issues throughout their lives, including depression, loneliness, anxiety, and stress [7]. Poor depression significantly affects the quality of life of older adults and is associated with an elevated risk of mortality and morbidity [8]. Therefore, exploring effective strategies to enhance the MH of older adults is essential. At present, the Internet has become increasingly popular among older adults following the development of technology and the popularity of smartphones. For instance, 7.3 million senior adults (1.3 % of the senior population) used the Internet in 2009, and the number s went up to 940 million, of which older users (i.e., those over 60 years old) account for 10.3 % [9,10]. The Internet penetration rate of the older adult population aged 60 years and above is 43.2 %. Internet use has substantial effects on reemployment, income enhancement, dietary improvement, and CO 2 emissions among older adults [11–14]. Internet use-based interventions are promising in terms of supporting the health of older adults and have proven to be effective in facilitating lifestyle changes and disease management in countries such as Australia, the Netherlands, and Switzerland [15]. Thus, such interventions might constitute a promising approach for bridging the existing gaps in MH care for older adults in China [16]. Thus far, the existing research has provided some correlational evidence but has yet to arrive at consistent conclusions. Most studies have reliably demonstrated that Internet use offers a convenient and effective means of communication among older adults, thereby enhancing various aspects of their well-being [17–21]. For example, using OLS regressions, Zhang et al. [22] found that Internet use is significantly associated with better MH for older Chinese residents. Clinical data also indicate that Internet use has been effective in reducing anxiety levels among patients with cardiomyopathy (Minto et al. [18]. The reason for such findings is that Internet use could increase life satisfaction, happiness, and overall well-being while concurrently reducing feelings of loneliness, anxiety, stress, and depressive symptoms [23]. Furthermore, the Internet provides an efficient and convenient channel through which older adults can communicate with their friends, fulfill their spiritual needs, enhance their cognitive abilities, and reduce their likelihood of experiencing depression [24]. However, some studies have shown that Internet use has negative implications for older adults in that it reduces social participation, leads to a narrowing of social circles, and weakens friendship networks and the sense of community belonging [4,20,25]. Although existing studies have analyzed the impact of Internet use on MH, few empirical tests have been conducted on the specific channels by which Internet use improves MH and reduces the risk of depression. Social participation serves as an intermediary factor in the relationship between Internet use and the MH of older adults [26–28]. Empirically, Internet use not only improves the self-reported health, MH, and social adaptation of older adults but also enhances their social participation. In particular, Internet use promotes community engagement among older adults, thereby enhancing their life satisfaction and overall health. For instance, Heo et al. [29] report that Internet use reduces loneliness and enhances the life satisfaction and well-being of older adults with social support acting as a mediating variable. Other mediating effects have also been proposed. For example, Cotten et al. [30] demonstrate that household size mediates the positive impact of Internet use among retired adults in the United States, especially for those living alone. Similarly, Yuan [31] reveals that older adults who engage in frequent Internet use are less likely to experience MH issues and identify the moderating effects of chronic diseases and household income on this relationship. The literature review reveals that there is no current consensus on the influence of Internet use on the MH of older adults, and some limitations are involved. First, most studies have focused on developed countries, such as the United States and New Zealand, and have paid limited attention to developing countries. Such studies have also been based on small samples, thereby limiting the statistical sophistication and robustness of the findings. Second, while earlier studies have separately focused on influence channels by considering the impact of Internet use on social participation, social support, or income, insufficient attention has been paid to social trust and health habits. Previous studies also did not consider these aspects within the same framework to analyze the potential channels of influence. Third, concerning research methodology, the existing studies have mainly focused on simple descriptive statistics and multiple linear regression analysis, which have been criticized for their estimation effectiveness, especially in terms of endogeneity and heterogeneity. Finally, most studies treat older adults as a homogeneous group and report varied, even conflicting, outcomes. Few studies have explored the uneven impact of Internet use on subgroups with different MH conditions. Therefore, the present research aims to investigate how Internet use affects the MH of older adults based on the 2013–2018 waves of the China Health and Retirement Longitudinal Study (CHARLS). Moreover, we examine whether Internet use has a mediating effect on MH by influencing four key aspects: social interaction, social trust, traveling expenses, and health habits. This study also considers the heterogeneity of older adults in China and analyzes the impact of Internet use on MH among different groups. Unlike previous studies, this research contributes to the existing literature in the following three aspects. First, using the nationally representative data of the 2013–2018 waves of CHARLS, this study enhances the validity of the estimated causal relationship between Internet use and MH among older adults by addressing endogeneity concerns using the instrumental variable (IV) method. Second, to the best of our knowledge, this is the first study to examine the mediating effect of traveling expenses and consider four key aspects in the same framework to further the existing understanding of the underlying channels between Internet use and MH. Third, we analyze the heterogeneity of the results across the number of chronic diseases that older adults have and their living arrangements (i.e., living with or without their children) to capture the effect of Internet use on these specific groups of older adults. This study offers valuable insights for future research endeavors that aim to explore the relationship between Internet use and MH among older adults in China. The study is organized as follows. Section 2 presents the data and variables, while Section 3 describes the empirical methods. Section 4 provides the estimation results, and Section 5 discusses the findings based on the empirical results. Finally, Section 6 concludes by summarizing the main findings and proposing policy implications. L. Zhang et al. Heliyon 10 (2024) e25397 3 2. Data 2.1. Sample The data used herein are taken from the 2013–2018 waves of CHARLS, a nationally representative longitudinal survey conducted by the Institute of Social Science Survey at Peking University. CHARLS focuses on households and individuals over 45 years of age in China. The survey employs a multistage sampling method with probability proportional to size. The national baseline survey was first conducted in 2011, followed by subsequent waves every 2–3 years. The data from the 2013–2018 waves are the most current data available. Encompassing 150 counties and 450 villages or resident committees in 28 provinces, these waves of the survey included approximately 17,000 individuals from approximately 10,000 households. The questionnaire collects a variety of information, including details on the respondents’ socioeconomic status and health circumstances. The CHARLS database is widely recognized as a reliable and valuable source of data within the academic community. According to the ‘Law of Protection of Rights and Interests of Older Adults’ in the People’s Republic of China, individuals over the age of 60 years are considered older adults. Therefore, for the sample, individuals aged 60 years and above in the 2013–2018 waves of CHARLS data are selected. After screening and eliminating samples that lacked relevant variables, a total of 27,561 valid samples were obtained. All data analyses are performed using STATA 16.1 software (STATA Corp. LLC, College Station, TX, USA). 2.2. Variables 2.2.1. Dependent variable In this study, the dependent variable is MH, specifically referred to as “depression” in the CHARLS dataset. To measure depression, CHARLS uses a simplified version of the Center for Epidemiologic Studies Depression Scale (CES-D10) [32]. This scale comprises eight questions regarding negative emotions and two questions addressing positive emotions. The scale encompasses diverse aspects such as daily mood, loneliness, sleep sufficiency, and the life situation of older adults. The response options, namely, “rarely or none of the time,” “some or a little of the time,” “occasionally or a moderate amount of the time,” and “most or all of the time” are assigned values of 0, 1, 2, and 3, respectively. In contrast, the scores for the positive emotion questions are inversely coded. The sum of the scores for the 10 items ranged from 0 to 30, with higher values indicating higher depression levels. A score of 10 or higher is considered a high depressive level and assigned a value of 1, whereas a score below 10 indicates no depression or a low depressive level and is assigned a value of 0 [33,34]. It should be noted that CES-D10 serves as a preliminary depression screening tool and cannot replace clinical diagnostic conclusions. The reliability coefficients for the CES-D10 items as investigated by Cronbach in 1999 and 2003 are 0.84 and 0.86, respectively [35]. 2.2.2. Independent variable The independent variables examined in this study are Internet use and frequency of Internet use. Owing to data limitations, we examined Internet use and frequency of Internet use for the 2013–2018 waves of the CHARLS questionnaire using the following question, “Have you done any of these activities in the last month?” The question presented 12 response options. The respondents who selected the option “Used the Internet” are classified as Internet users in this study and assigned a value of 1. Otherwise, the respondents are assigned a value of 0 [28,36,37]. The frequency of Internet use variable, as robustness, was examined by the following question, “How often did you use the Internet in the last month?” The response options for this question are categorized as follows: “almost daily” and “almost every week” are assigned a value of 1, whereas “not regularly” is assigned a value of 0. 2.2.3. Channel variables To comprehend the underlying channels through which Internet use (frequency of Internet use) may impact depression among older adults, variables associated with social interaction, social trust, traveling expenses, and health habits are examined. In the 2013–2018 waves of the CHARLS questionnaire, the variable of social interaction is assessed using the question, “Have you done any of these activities in the last month?” The question included various response options such as “Interacted with friends,” “Played Ma-jong, played chess, played cards, or went to a community club,” “Provided help to family, friends, or neighbors who do not live with you,” “Went to a sport, social, or other kinds of the club,” “Took part in a community-related organization,” and “Attended an educational or training course.” If respondents select any of these options, they are assigned a value of 1; otherwise, they are assigned a value of 0. The variable of social trust is assessed by asking the following question: “Have you done any of these activities in the last month?” In CHARLS, respondents are asked, “In the last year, how much did your household spend on the following items?” If the answer to this question is greater than 0, a value of 1 is assigned; otherwise, a value of 0 is assigned. For the variable of travel expenses, we focused on enjoyable consumption such as traveling expenses. The question is “Have you done any of these activities in the last month?” The responses included options such as “Done voluntary or charity work” and “Cared for a sick or disabled adult who does not live with you.” Respondents who selected either of these options are assigned a value of 1, whereas others are assigned a value of 0. The variables of health habits included smoking and physical activity. The responses for smoking are derived from questions such as “Have you ever chewed tobacco, smoked a pipe, smoked self-rolled cigarettes, or smoked cigarettes/cigars?” and “Do you still have the habit, or have you quit?” The status of physical activity is derived by asking, “Do you usually engage in this type of activity for at least 10 min every week?” To measure the variable of health habits, “yes” responses are assigned a value of 1, whereas “no” responses are assigned a value of 0. L. Zhang et al. Heliyon 10 (2024) e25397 4 2.2.4. Covariates According to Grossman’s health needs theory [38], various factors—such as biological factors, income, social status, education, culture, early child development, health systems, social support, population profiles, and geography—play significant roles in influencing depression. In related studies on Internet use (frequency of Internet use) and depression among older adults, control variables are selected from three dimensions: individual, family, and social characteristics. In this study, the control variables included gender, age, education, marital status, the logarithm of income, the number of chronic diseases, urban residence, insurance, and province and time controls [25,32,36,37]. In addition, given the increasing number of “empty nesters,” we took the influence of children on older adults into consideration, and variables such as cohabitation with children and the number of children are also included. These variables are selected because of their close associations with the occurrence of depression symptoms. The data used herein can be obtained from the CHARLS 2013–2018 dataset, although further details are not provided. 3. Empirical method 3.1. Effect of internet use (frequency of internet use) and MH in older adults To investigate how Internet use (frequency of Internet use) could affect older adults’ depression, the panel random effects (RE) Probit model is employed to examine the causality between MH and Internet use (frequency of Internet use) among older adults. This model is established on the basis of the standard normal distribution equation. P(Hit =1) = Φ(β0+β1Iit +β2Zit)(1) In eq (1), H i is the MH of the ith older adults in t year, I it represents Internet use (frequency of Internet use), and Z it indicates several control variables to account for potential confounding factors. Eq (1) can be given by: Hit∗= α 0+ α 1Iit + α 2Zit +δt+ξit,Hi=1(Hi ∗>0)(2) In eq (2), Hi∗is a latent variable, δt is a dummy variable that controls the time trend and province, ξi symbolizes random disturbance terms, and we can obtain the unbiased estimator Cov(Iit,ξit) = 0. To ensure unbiased and consistent estimation results while addressing potential endogeneity issues, this study used the IV-RE Probit model. The IV-RE Probit model, which corresponds to binary factor variation, is selected as the benchmark for the bivariate panel Probit model. Under the assumption of convergence, employing the IVRE Probit model for estimation proved to be more efficient. The fundamental concept is to identify instrumental variables that are associated with Internet use (frequency of Internet use) but not with MH and construct an instrumental variable equation to replace the endogenous variables in the equation. The specific implementation process is outlined as follows: Hit∗=β0+β1Iit +β2Zit +δt+ξit Hit =1(Hit ∗>0)(3)  Iit =γ0+γ1IVit +γ2Zit +ξ(4) In eq (3) – eq (4), IVi indicates the instrumental variables and Cov(IVit,ξit) = 0, Cov(ξ,ξit) = 0, and Cov(IVit,Iit) ∕= 0. 3.2. Possible channels explaining the health effects of internet use and frequency of internet use The other aim of this study is to shed light on the potential channels through which Internet use (frequency of Internet use) affects depression. Some scholars have proposed that the use of stepwise tests for mediating effects should be discontinued and that researchers should refrain from estimating the size of indirect effects or testing their statistical significance [39]. Instead, according to such scholars, the focus of research should shift toward enhancing the credibility of the identified causal relationship between explanatory variable I and explained variable H. As per economic theory, one or several intermediary variables, M, are introduced to reflect the causal pathway from I to H. The impact of M on H should be direct and evident, with an emphasis on identifying the correlation between I and M. In this study, four primary channels, that is, social interaction, social trust, traveling expenses, and health habits, are examined. The model is given as follows: Mit =β0+β1Iit +β2IVit +β3Zit +δt+ τ it (5) In eq (5), Zit represents the control variable, and Mit denotes the mediator variable. 3.3. Identification strategy Given the longitudinal structure of our data, we can simply run a RE estimation and a fixed effects estimation and use the Hausman test to check which model is preferred. However, we have unbalanced panel data, which suggests that participants are not consistently interviewed or did not consistently respond to the CHARLS questions we are interested in. The fixed effects model assumes that some time-invariant individual characteristics correlate with the predictor or independent variables and may be biased in the estimates. The RE model assumes that variation between individuals is random and uncorrelated with the predictor or independent variables included in the model, unlike the fixed effects model. In this study, the explanatory variable “depression” is a dichotomous variable, so it is L. Zhang et al. Heliyon 10 (2024) e25397 5 estimated using a panel probit. However, it will lead to an inconsistent estimation of the coefficient α _1 if the fixed effect of the panel probit is estimated. At the same time, the LR test of the RE model in Table 2 rejects the hypothesis of the mixed probit regression model. Therefore, the panel RE probit model is finally selected for estimation. However, the potential endogeneity problem still affects our primary results. Namely, the decision to use the Internet, along with an older individual’s depressive symptoms may be determined or influenced by some missing or unobserved variable. In addition, MH conditions affect an older individual’s behavior, which in turn may affect their online behavior, creating a reverse causality problem. To overcome these problems and test the main results, we employ the IV-RE probit method to re-estimate the model. This study uses the average Internet use levels of older adults in each village as the instrumental variable of whether the subjects used the Internet or not. The IV-RE probit method is a two-stage regression. In the first stage, the instrumental variables of the principal independent variables are determined and their correlations are evaluated. In the second stage, the instrumental variables are substituted into the model for regression, and the externality of Internet use (frequency of Internet use) is verified according to the endogeneity test parameters. 4. Results 4.1. Statistical description Table 1 displays the descriptive statistics of the variables, including the number of observations, means, and standard deviations. 36 % of the respondents reported experiencing depression. The independent variable, “Internet”, is utilized by 26 % of the older population in the last month, indicating a relatively low prevalence of Internet use among older adults in China. The average frequency of Internet use of the sample is 1.1, with a standard deviation of 0.36. The participants’ educational levels range from 0 to 22 years, with an average of 6.6 years of education. The majority of the participants (76 %) are married. Further details of the descriptive statistics are presented in Table 1. Table A1 provides a visual comparison of the differences between Internet users and non-users. The results indicate a statistically significant difference in the level of depression between Internet users (mean =0.39) and non-users (mean =0.28). In addition, statistically significant differences are noted in 8 covariates between the aforementioned two groups. It suggests that a causal link between Internet use and depression levels in older adults cannot be extrapolated. 4.2. Baseline regression Using the panel RE Probit model, Table 2 presents the impact of Internet use and frequency of Internet use on MH. In Model (1), Internet use demonstrates a significant negative coefficient of −0.401 (p <0.01), indicating that it reduces depression levels. Model (2) yields consistent results even after incorporating covariates, thus highlighting the significantly negative relationship between the selected control variables and depression with a coefficient of −0.300. To minimize the bias of analysis results, we also consider the frequency of Internet use in the baseline regression analysis. Model (3) in Table 2 presents the estimated results of the frequency of Internet use variable. The results show that the coefficient is estimated as −0.432 thus indicating a statistically significant impact at the 1 % level. Upon incorporating covariates, Model (4) displays consistently significantly negative results with a coefficient of −0.261. It Table 1 Descriptions and basic statistics of the selected variables. Variables Definition Mean S.D. Dependent variable MH If high depressive level =1, otherwise =0 0.36 0.48 Independent variable Internet If individual uses Internet =1, otherwise =0 0.26 0.44 Fre_internet The frequency of Internet use, If not regularly =1, If almost every week =2, If almost daily =3 1.1 0.36 Channel variables Social Joining social activities 0.46 0.50 Trust Trust others including strangers 0.03 0.16 Traveling expenses Long-distance traveling expenses 0.37 0.48 Smoke Smoking 0.34 0.47 Physical activities Doing physical activities every week 0.57 0.49 Control variables Gender If male =1, otherwise =0 0.50 0.50 Age Years of age 70 7.40 Age 2 Age squared 4885 1079 Education Years of education 6.60 5.00 Married If married =1, otherwise =0 0.76 0.42 Lincome The Logarithm of respondent’s income in the last year (yuan) 5.50 3.10 Illness Numbers of chronic diseases 1.30 1.40 Urban If currently living in urban =1, otherwise =0 0.39 0.49 Living with child If living with child =1, otherwise =0 0.40 0.49 Child number Numbers of children 1.90 2.00 Insurance If had basic endowment insurance =1, otherwise =0 0.32 0.46 Observations: 27,561 Source: Authors’ calculation based on the CHARLS data from the 2013–2018 waves. L. Zhang et al. Heliyon 10 (2024) e25397 6 suggests that introducing control variables can effectively reduce estimation bias. This finding supports the notion that Internet use can effectively decrease depression levels and enhance MH. 4.3. Accounting for the endogeneity of internet use and frequency of internet use For the instrumental variable, this study selects the average Internet use levels of older adults in each village or community, excluding the individuals being analyzed. Using high-level average values (e.g., province- or community-level ones) as instruments for lower levels (e.g., the household level) is a common technique in the literature [40–42]. The reliability of the IV estimation depends on the validity of the instrumental variable. The instrumental variable should be highly correlated with the endogenous variable but has Table 2 Baseline regression results of the effect of Internet use and frequency of Internet use on MH. Variables (1) (2) (3) (4) Internet −0.401*** −0.300*** (0.027) (0.028) Fre_internet −0.432*** −0.261*** (0.036) (0.036) Gender −0.413*** −0.408*** (0.026) (0.026) Age 0.155*** 0.160*** (0.026) (0.026) Age 2 −0.001*** −0.001*** (0.000) (0.000) Education −0.012*** −0.012*** (0.003) (0.003) Married −0.166*** −0.171*** (0.031) (0.031) Lincome −0.018*** −0.018*** (0.005) (0.005) Illness 0.131*** 0.132*** (0.008) (0.008) Urban −0.334*** −0.363*** (0.028) (0.028) Living with child 0.066*** 0.013 (0.023) (0.023) Child number −0.026*** −0.022** (0.010) (0.010) Insurance −0.254*** −0.252*** (0.034) (0.034) Province and time fixed effect Yes Yes Yes Yes lnsig2u 0.032 −0.221*** 0.056 −0.203*** (0.045) (0.049) (0.045) (0.049) LR test 1741.13*** 1221.59*** 1800.37*** 1252.35*** N 27,561 27,561 27,561 27,561 Note: Standard errors are in parentheses. *p <0.1, **p <0.05, ***p <0.01. Source: Authors’ calculation based on the CHARLS data from the 2013–2018 waves. Table 3 IV estimation of the impact of Internet use and frequency of Internet use on MH. Variables (1) (2) Second stage: Internet −0.581*** (0.077) Fre_internet −1.351*** (0.117) Controls a Yes Yes Province and time fixed effect Yes Yes First stage: IV −0.739*** −0.325*** (0.016) (0.014) Wald test 19.52*** 43.95*** AR 57.65*** 60.08*** Wald 57.35*** 55.50*** N 27,561 27,561 Note: Standard errors are in parentheses. *p <0.1, **p <0.05, ***p <0.01. a Controls include all control variables as in Table 2. Source: Authors’ calculation based on the CHARLS data from 2013 to 2018 waves. L. Zhang et al. Heliyon 10 (2024) e25397 7 no direct correlation with the dependent variable. First, it satisfies the correlation requirement. Following behavioral imitation theory, a higher Internet use rate among older adults in a given village increases the likelihood and the frequency of individual Internet use due to spillover effects. Second, it is deemed exogenous. Internet use by others in the same village is unlikely to have a direct impact on individual behavior and depression. Hence, the average Internet use rate is considered an ideal instrumental variable. The CHARLS samples are categorized according to village/residence code, and the average Internet use rate of older adults in each village or neighborhood committee is calculated as the regional-level Internet use rate. The strength of the instrumental variable is assessed using the p-value. If it is significant at the 1 %, 5 %, or 10 % levels, the absence of a weak instrumental variable problem in the regression is indicated. This approach ensures that the instrumental variable selected is strongly correlated with the dependent variables and does not introduce bias into the result estimators. Of the various methods used to establish causality, IV analysis offers distinct advantages in terms of addressing endogeneity. To address endogenous issues, this study employs an IV-RE approach for a comprehensive analysis. In the first stage of the regression, as shown in Table 3, the instrumental variable is statistically significant and negatively correlated with Internet use and the frequency of Internet use thus meeting the dependency condition of the instrumental variables. The p-value of the weak IV test is significant at the 1 % level. Rejecting the null hypothesis (null hypothesis H 0 : endogenous variables are not correlated with instrumental variables), indicates that the instrumental variables are not weak. Thus, the instrumental variable is valid and can be used in further estimations. Following the findings in Model (1)–(2) of Table 3, Internet use and the frequency of Internet use exhibit statistical significance (p <0.01) for depression among older adults. These findings indicate that a higher level of Internet use or a higher frequency of Internet use has a negative impact on depression among older adults, which is consistent with the results obtained from the baseline regression analysis. It is similar to the results of previous studies [36,43]. 4.4. Robustness test To ensure the robustness of the findings, three additional methods are used to conduct robustness tests. To explore the robustness of the depression variable, the measurement of the independent variable is changed to a continuous variable Table 4 presents the estimated results after substituting the measurement of the dependent variable for robustness testing. According to the results, the effect is statistically significant at the 1 % level. This finding further supports the notion that Internet use (frequency of Internet use) enhances the MH of older adults, which is consistent with the results obtained from the baseline regression analysis. According to the age criteria defined by the World Health Organization for the older population, individuals aged 85 years and above are considered to be long-lived older adults. Thus, to account for the potential impact of extreme age values, individuals exceeding 85 years of age are excluded from this analysis, and the robustness of the results is re-estimated. Table 5 shows that the estimated marginal effects are −0.570 and −0.811, demonstrating statistical significance at the 1 % level. These results are consistent with previous research findings; thus reinforcing the robustness of the observed relationship between Internet use (frequency of Internet use) and MH outcomes among the older population. To address potential self-selection bias and reverse causality, the endogenous switching probit (ESP) method is employed for indepth analysis. This approach accounts for the possibility that Internet use among older adults may result from self-selection and that there could be a bidirectional relationship between Internet use and MH. The ESP model has been designed to mitigate bias due to observable and unobservable factors [44,45]. To control for self-selection bias, the IV method is employed to correct for any potential bias [46]. The estimation results of the ESP model are presented in Table A2. The average Internet use of other older adults within the same village had a significant positive effect on an individual’s likelihood of using the Internet. This suggests that a higher average level of Internet use among older adults in the same village increases the probability of individual Internet adoption. The parameter ρ 1 is statistically significant at the 1 % level, indicating the presence of selective bias in the regression model. Unobservable factors impact both the level of Internet access and the MH of older adults. Furthermore, ρ 1 determined to be positive, suggesting that the MH level of Table 4 IV estimation of the impact of Internet use and frequency of Internet use on MH. Variables (1) (2) Second stage: Internet −0.248** (0.232) Fre_internet −0.522*** (0.109) Controls a Yes Yes Province and time fixed effect Yes Yes First stage: IV −0.353*** −0.353*** (0.247) (0.446) F value 20.79 62.87 N 27,561 27,561 Note: Standard errors are in parentheses. *p <0.1, **p <0.05, ***p <0.01. a Controls include all control variables as in Table 2. Source: Authors’ calculation based on the CHARLS data from 2013 to 2018 waves. L. Zhang et al. Heliyon 10 (2024) e25397 8 older adults using the Internet is higher than that of the overall sample. The goodness-of-fit test and model independence are statistically significant at the 1 % level, indicating that the use of the ESP model is appropriate. The ESP model is selected to empirically analyze the impact within a “counterfactual” framework, allowing for the calculation of the average treatment effect on the treated (ATT), average treatment effect on the untreated (ATU), and average treatment effect (ATE) [47]. However, because the ATU and ATE parameters include the effects on samples unaffected by joining cooperatives, this study focuses solely on the ATT, representing the mean change in the MH status of the sample using the Internet compared with that if they did not use the Internet. Table 6 presents the estimated treatment effects of Internet use on MH behaviors among older adults. The treatment effect is consistent with the previously estimated results, indicating that Internet use can significantly improve MH among older adults. The results obtained from the ESPmodel provide additional support for the robustness of the findings. 4.5. Heterogeneity analysis In the previous study, researchers treated all older adults as a homogeneous group when examining the average effect of Internet use (frequency of Internet use) on MH. However, the Internet use (frequency of Internet use) of different older groups with varying individual characteristics is heterogeneous is notable. Previous studies have demonstrated the impact of Internet use (frequency of Internet use) on depression across various demographic groups [44,48,49], which differ in terms of income level, gender, education level, age, and number of accompanying persons [50–52]. However, limited evidence exists regarding the differences observed among groups based on their physical health status or living arrangements. To investigate the effects of Internet use (frequency of Internet use) on MH among older individuals with different characteristics, we examine the regression results of variables such as the number of chronic diseases and living arrangements (specifically, living with children). Moreover, the analysis employs the IV-RE method to address any potential endogeneity issues. Table 7 presents the findings of the heterogeneity analysis between Internet use and depression. In Model (1), the effect of the interaction between Internet use and living arrangements (whether the respondents are living with their children) on depression is examined. The findings demonstrate a significant negative interaction effect, suggesting that the relationship between Internet use and depression is attenuated among older adults who cohabit with their children. In other words, Internet use is more likely to be associated with better MH among individuals living with their children. In Model (2), the effect of the interaction between Internet use and the number of chronic diseases on depression is investigated. The results reveal a significant positive interaction, suggesting that the association between Internet use and depression is stronger among older adults with a lower number of chronic diseases, which indicates that healthier older individuals tend to experience greater benefits from Internet use than those with poorer health conditions. Model (3) incorporates both of these interaction terms, and the results align with those of Models (1) and (2). Thus, it can be concluded that the Internet has the potential to reduce MH disparities among the older adult population. Notably, this effect tends to be significantly positive for older adults with fewer chronic diseases who live with their children. Table A3 presents the findings of the Table 5 Robustness test results of MH consequences of Internet use and frequency of Internet use. Variables (1) (2) Second stage: Internet −0.570*** (0.083) Fre_internet −1.263*** (0.186) Controls a Yes Yes Province and time fixed effect Yes Yes First stage: IV −0.773*** −0.555*** (0.018) (0.025) Wald test 18.86*** 37.97*** AR 47.80*** 49.99*** Wald 47.43*** 46.09*** N 21,833 21,833 Note: Standard errors are in parentheses. *p <0.1, **p <0.05, ***p <0.01. a Controls include all control variables as in Table 2. Source: Authors’ calculation based on the CHARLS data from 2013 to 2018 waves. Table 6 The treatment effect of Internet use on older adults’ MH. Variables Observations Mean S.D. Min Max ATT 7053 −0.192 0.039 −0.334 −0.050 ATU 20,508 −0.167 0.042 −0.296 0.001 ATE 27,561 −0.173 0.039 −0.298 0.003 Source: Authors’ calculation based on the CHARLS data from the 2013–2018 waves. L. Zhang et al. Heliyon 10 (2024) e25397 15 [54] T.C. Lang, The effect of the Internet on travel consumer purchasing behaviour and implications for travel agencies, J. Vacat. Mark. 6 (4) (2000) 368–385, https://doi.org/10.1177/135676670000600407. [55] R. Law, A. Law, E. Wai, The impact of the internet on travel agencies in Hong Kong, J. Trav. Tourism Market. 11 (2–3) (2002) 105–126, https://doi.org/ 10.1108/09596110410519982. [56] M. Kroesen, J. De Vos, Does active travel make people healthier, or are healthy people more inclined to travel actively? J. Transport Health 1610 (2020) https:// doi.org/10.1016/j.jth.2020.100844. [57] A. Zamanzadeh, T. Cavoli, R. Banerjee, Travelling, anxiety and the impact of COVID-19: evidence from Italy, Curr. Issues Tourism 8 (2022), https://doi.org/ 10.1080/13683500.2022.2153651. [58] C.E. Moge, D.M. Romano, Contextualising video game engagement and addiction in mental health: the mediating roles of coping and social support, Heliyon 6 (11) (2020) 10, https://doi.org/10.1016/j.heliyon.2020.e05340. [59] R.F. McCloud, C.A. Okechukwu, G. Sorensen, K. Viswanath, Entertainment or health? Exploring the internet usage patterns of the urban poor: a secondary analysis of a randomized controlled trial, J. Med. Internet Res. 18 (3) (2016), https://doi.org/10.2196/jmir.4375. [60] K. Mikkelsen, L. Stojanovska, M. Polenakovic, M. Bosevski, V. Apostolopoulos, Exercise and mental health, Maturitas (2017) 10648–10656, https://doi.org/ 10.1016/j.maturitas.2017.09.003. L. Zhang et al.