How technostress and self-control of social networking sites affect academic achievement and wellbeing

2.1 SNS technostress

Interacting with IT can be stressful. This phenomenon has given rise to a significant volume of research into the concept of technostress, which has been defined as stress that individuals experience due to their use of IT (Tarafdar et al., 2019).

Technostressors are the creators of stress (Ragu-Nathan et al., 2008). The design and features of SNS applications, such as real-time notifications and emotion displays, strongly influence the potency of technostressors. While research into the stressors associated with SNS use is relatively recent, six SNS stressors have been identified and validated in prior literature (Maier et al., 2012, 2015). These are social overload which emerges when users feel that they are experiencing excessive social demands through the SNS (Maier et al., 2014, 2015); disclosure which refers to the stress-creating condition of individuals feeling that they get too much information on SNS, which prevents them from effectively using the application (Maier et al., 2015; Tarafdar et al., 2020); pattern which describes the stressor the individual experiences when they adapt their use of SNS to conform to his or her friends' use, for example, keeping up with friends' postings and informing friends of their activities (Maier et al., 2015; Tarafdar et al., 2020); complexity which reflects the extent to which users perceive SNS as technically difficult to use because they do not understand enough about it – this difficulty could arise from, for instance, features that may be complicated to navigate through, such as privacy and data sharing features (Maier et al., 2015; Tarafdar et al., 2020); uncertainty which describes the constant changes and updates to the SNS applications that are stressful (Maier et al., 2015; Tarafdar et al., 2020) and invasion, which reflects the stress-creating condition wherein individuals feel that their personal life is being invaded by SNS because they use SNS in inopportune situations such as during vacations (Maier et al., 2012, 2015; Zhang and Fu, 2020).

The outcomes linked to technostress are called strain responses and can be psychological, behavioral or physiological (Ayyagari et al., 2011; Ragu-Nathan et al., 2008). Studies conclude that people reporting elevated levels of technostress are more likely to suffer the psychological strains of diminished commitment (Tarafdar et al., 2010, 2015), poor self-esteem (Korzynski et al., 2020), dissatisfaction with the IT system (Tams et al., 2020), harmful psychological responses (Riedl et al., 2012), burnout (Afifi et al., 2018) and lower job satisfaction (Califf and Sarker, 2020; Suh and Lee, 2017). While studies of technostress generally rely on self-reported data, physiological studies have also demonstrated that stress hormones cortisol and alpha amylase increase with IT-mediated interruptions (Tams et al., 2018), IT system breakdown (Riedl et al., 2012) and extensive exposure to IT applications (Afifi et al., 2018). For these reasons, technostress has been firmly housed within research pertaining to the dark side of IT use (Salo et al., 2018).

As engagement with digital technology has rippled beyond the workplace to influence our personal lives, so too has the technostress literature evolved to evaluate voluntary interactions with hedonic technologies, specifically SNS. This body of work also confirms the significant stress associated with the use of SNS. Through the distractions inherent within SNS, technostress has been likened with the emerging problem of technology addiction (Brooks et al., 2017; Tarafdar et al., 2020) which is exacerbated by habitual SNS use (Tarafdar et al., 2020; Vaghefi et al., 2020). Problems with concentration, sleep, identity and social relationships have been linked with SNS features which trigger technostress (Kaur et al., 2021; Salo et al., 2019). Emerging research also reveals that the formation of stress associated with using personal IT systems such as SNS occurs at different rates. For instance, the invasion technostressor builds up gradually, while a technostressor such as complexity emerges in a more punctuated manner (Salo et al., 2021). Technostress also has implications on intentions to continue using SNS. In response to SNS stressors and exhaustion, users develop discontinuous usage intentions to avoid such stress (Luqman et al., 2017; Maier et al., 2015).

Despite the recent advancements in SNS technostress research, the findings from prior studies are limited (in at least one way) as they tend to focus on one specific SNS platform, i.e. Facebook. There are different categories of SNS and users engage with these platforms for different purposes. While Facebook can be considered a general SNS, other categories of SNS are more heavily used today, including media sharing networks (e.g. Instagram, Snapchat), discussion forums (e.g. Reddit, Digg), consumer review networks (e.g. Yelp, TripAdvisor), content curation networks (e.g. Pinterest, Flipboard) and blogging networks (e.g. Tumblr, Medium). Indeed, the insights from Facebook stressor studies may have limited applicability to university students today as a recent Pew Research Center study reports that 18- to 24-year-olds are more likely to use the media sharing networks Instagram and Snapchat, ahead of Facebook (Pew Research Center, 2021). Therefore, in our study we requested student participants to consider the SNS platform they most frequently use when providing responses.

Underpinning the majority of technostress studies is the transactional theory of stress (Lazarus and Folkman, 1984), which purports that people respond to a stressor through a process of evaluating and implementing available coping resources. Coping mechanisms such as distractions (Tarafdar et al., 2020), distancing from IT (Pirkkalainen et al., 2019), mindfulness (Pflügner et al., 2021) and instrumental and emotional support (Weinert et al., 2020) have been examined in the technostress literature to date. Yet, recent interpretivist research demonstrates how difficult it is for users to mitigate technostress as considerable self-control is required (Salo et al., 2021). Indeed, a proactive personality is important for successful coping (Tiwari, 2020). Therefore, to provide a deeper understanding of how SNS technostress impacts self-control resources, this study views SNS through the strength model of self-control lens.

2.2 The strength model of self-control

Self-control refers to the capacity for altering one's own behavior, especially to bring them into line with standards such as ideals, values, morals and social expectations, and to support the pursuit of long-term goals (Baumeister and Tierney, 2012). In everyday life, self-control is required to overcome “dominant responses” (i.e. impulses, habits, urges, cravings and drives) that tempt us to procrastinate, overeat, drink too much alcohol, spend beyond our means or say hurtful things. Self-control is particularly important in managing activities that generate hedonic gratifications and are socially favored (Baumeister and Heatherton, 1994). Engaging in SNS would be consistent with such activities. Many social and behavioral problems can be traced back to a deficiency in self-control (Hagger et al., 2010). Analogously, effective self-control has been shown to contribute to many adaptive outcomes in society, such as success in education, better physical and mental health, career advancement, flourishing interpersonal relationships and the ability to deal with problems (see Nielsen et al., 2020 for a review). The notion of self-control is particularly important in this study as the behavioral outcomes arising from SNS technostress depend on the habitual use of SNS (Tarafdar et al., 2020).

The human capacity to exhibit self-control in the face of temptations is not infinite and depletes over time. In the strength model of self-control, Baumeister and colleagues compare self-control to a muscle (Baumeister et al., 2007). When muscles become stressed from continued use, such as lifting weights repetitiously, they fatigue and eventually fail. Like muscles, self-control requires energy to perform but depletes when exerted. Stressing self-control resources incapacitates the ability to override subsequent acts requiring self-control, a phenomenon called ego depletion (Baumeister et al., 2007). For example, a student may have a goal of performing well in forthcoming exams and so wants to avoid spending time consuming entertainment videos on SNS. An impulse to engage with SNS is triggered every time they see another person using such platforms (Tandon et al., 2021; Whelan et al., 2020). Extinguishing these impulses continuously draws strength from their self-control reserves. When these reserves become exhausted and weak, the person is less likely to maintain the self-control needed to avoid wasting time on SNS (Bright and Logan, 2018).

While we may believe certain people possess tremendous self-control, research suggests that everyone's self-control diminishes when called upon; some people are just better at shaping their environments in a way that limits exposure to temptation, thus preserving their reserves of self-control (Hofmann et al., 2012). A number of lab experiments have confirmed that the ability to regulate oneself becomes impaired following initial efforts to instigate self-control. For example, participants who completed a stressful task designed to deplete self-control (following a set of confusing instructions) committed more errors during a subsequent color-word Stroop task than the nonstressed control group (Hagger et al., 2010). In an important extension to the muscle metaphor, self-control can be improved with dedicated practice (Baumeister and Tierney, 2012). Targeted efforts to control behavior in one area, such as healthy eating, lead to improvements in unrelated areas, such as studying or household chores. While the published literature on ego depletion has been found to be influenced by publication bias (Carter and McCullough, 2014) (i.e. the outcome of an experiment or research study influences the decision whether to publish or otherwise distribute it), overall, a considerable amount of research supports the primary tenet of the strength model of self-control (Englert, 2019; Ginis and Bray, 2009; Hagger et al., 2010).

Students, in particular, are exposed to lots of external distracting stimuli which can impact their ability to develop a successful academic career. In educational settings, the strength of self-control as a means of preventing ego depletion has received some support as a mechanism enabling students to neutralize the impulses and desires not conducive to academic success (Duckworth and Yeager, 2015; Englert et al., 2017; Panek, 2014). Yet, the question of whether the depletion of self-control explains the association between stressors and life outcomes has yet to be examined. We address this question by specifically examining how a deficiency in self-control mediates the relationship between a prominent stressor in today's world (SNS stressors) and student outcomes (achievement and wellbeing). Thus, the strength model of self-control can provide new insights into why SNS stressors are associated with psychological responses, which in turn can inform the design of targeted interventions.

2.3 Academic outcomes

Studying at university brings not only a wealth of opportunities but also significant challenges. The factors that influence the success and wellbeing of university students have received much scholarly attention. In the majority of studies on students’ use of SNS and its impact on academic achievement, student performance has been measured on the basis of GPA (Cao et al., 2018; Giunchiglia et al., 2018; Junco, 2012; Kirschner and Karpinski, 2010; Lambic, 2016; Michikyan et al., 2015; Rashid and Asghar, 2016). However, more recent research has sought a more holistic measure of academic achievement by measuring learning performance (Al-Rahmi et al., 2018), academic engagement (Datu et al., 2018) and other important aspects of achievement, such as leadership development and extracurricular activities, which potential employers look for when assessing the suitability of a graduating student (Whelan et al., 2020). In this study, we adopt the broader conceptualization of “academic achievement”, a multifaceted construct that comprises the different domains of learning available to students while in university (Steinmayr et al., 2018). As such, we define academic achievement as the extent to which a student has attained their goals in terms of study performance and career development.

The pressure to achieve academically, while also transitioning into the university environment, can contribute to suboptimal wellbeing for young students. In recent times, the uptake of student counseling services within universities has risen in parallel with reported mental health difficulties (Kaparounaki et al., 2020). Wellbeing difficulties have become even more pronounced for students during the COVID-19 pandemic as campuses closed, depriving students of social interaction and other support networks (Burns et al., 2020). Indeed, 34% of Australian university students reported low wellbeing during COVID-19, with 32% reporting very low wellbeing (Dodd et al., 2021). Stemming from World Health Organization's (WHO) definition that health “is a state of complete physical, mental and social wellbeing and not merely the absence of disease or infirmity” (World Health Organization [WHO], 1948) it is now widely accepted that wellbeing is a multidimensional construct (Wills-Herrera et al., 2009). Subjective wellbeing consists of three strands – evaluative wellbeing (or life satisfaction), eudemonic wellbeing (or vitality) and hedonic wellbeing (feelings of happiness, sadness, anger, stress and pain) (Steptoe et al., 2015). In this study, we adopt two common measures of subjective wellbeing in student populations: satisfaction with academic life and subjective vitality (Burns et al., 2020). Satisfaction with academic life encompasses mood, satisfaction with relationships, self-concepts and self-perceived ability to cope with one's daily academic life (Diener et al., 1985). A student could be physically healthy, but not leading a full and vibrant life. The concept of subjective vitality accounts for these aspects of life, and differs from life satisfaction in that it refers to the state of feeling alive and alert, and having energy available to the self (Vallerand, 2015).

4.1 Participants

The research model was tested using survey data collected over two points in time from students of a large (450 students) introductory-level course in a university in Ireland. The study began with a pilot study involving 42 SNS users who were university students. Results demonstrated sufficient validity and reliability. Participants for the main study were recruited through an in-class announcement. To be included participants had to be at least 18 years of age, currently using SNS and must consent to participation. Participation was voluntary and was encouraged with entry into a draw for one of five EURO 25 gift vouchers. Study procedures were approved by the research ethics board of the university.

The time-lagged design separated predictor variables from outcomes which reduces common method bias (CMB; Podsakoff et al., 2003) and lends stronger (but imperfect) support for causal arguments (Turel and Qahri-Saremi, 2016). The predictors and control variables were measured at t1. Two weeks later, the outcome variables, mediator and moderator were measured at t2. The two-week time lag was considered sufficient for participants to retain interest in the study, but not too long as they lose interest (Turel and Qahri-Saremi, 2016).

The first survey was completed by 372 participants (83% response rate) with the follow-up survey completed by 356 (79% response rate). When matched, 318 participants had completed both surveys. Of the matched sample, 37 did not fully complete both surveys, resulting in unusable data that was discarded. Another 99 responses were discarded as they incorrectly answered at least one of the four attention check questions included in the surveys. After this data cleaning process, the final sample consisted of 220 participants. We used the G-power sampling size information (Faul et al., 2009) to determine if the sample size of 220 is adequate. Since our model has three predictors, using G-power with an effect size of 0.10, alpha of 0.05 and a power of 0.95, the minimum sample size needed was 132. Thus, we can conclude that the sample size of 220 provides sufficient power to enable the findings to be used with confidence.

In terms of sample characteristics, 60% were female with the remaining 40% classifying themselves as male. The mean age was 19.84 years (SD = 1.36). Snapchat was by far the most preferred SNS platform (68%), followed by Instagram (23%), Twitter (3%), WhatsApp (2%), Facebook (1%) and others (3%). This aligns with recent data from the USA which confirms that 18–24 years old mostly use the media sharing networks Snapchat and Instagram (Pew Research Center, 2021). Of the participants, 75% had been using SNS for more than five years, 19% for four to five years, with the remaining 6% for three to four years. Participants were asked to answer all questions in relation to their use of their preferred SNS platform.

4.2 Measures

All measurement items were taken from prior validated studies. A seven-point Likert scale was used to measure the key constructs of the research model, except for academic achievement which was measured with a five-point scale as per the original construct design. All survey items are provided in Table A1.

Following the approach of Tarafdar et al. (2020), SNS stressor is specified as a second-order reflective latent construct with six first order dimensions, that are also reflective. The first-order constructs are social overload, disclosure, pattern, complexity, uncertainty and invasion. Deficient SNS self-control was measured using the scale developed by LaRose et al. (2003). The moderating variable, academic SNS use, was measured using a scale taken from (Chang et al., 2019). Student vitality was measured using a well-established six-item scale (Bostic et al., 2000; Ryan and Frederick, 1997). Satisfaction with academic life was adapted from the commonly used life satisfaction scale (Diener et al., 1985). Finally, academic achievement is specified as a second order reflective construct comprising the two first order dimensions of study performance and career development, also reflective (Whelan et al., 2020). When completing this scale, participants were instructed to reflect specifically on their academic life as opposed to their general life satisfaction.

4.3 Controls

As a personal trait, self-control may influence the extent of their SNS self-control, we controlled for this effect by using an established four-item scale (Soror et al., 2015). We also controlled for academic motivation, using the same scale as Klobas et al. (2018) as previous research suggests academic motivation drives many outcomes relevant to student life (Wentzel and Wigfield, 1998). Gender and age were also included in the model as control variables.

4.4 Research model results

The hypotheses were tested using SmartPLS 3.3.3 (Ringle et al., 2015). The Partial least-squares (PLS) approach to structural equation modeling (SEM) enables researchers to estimate complex models with many constructs, indicator variables and structural paths without imposing distributional assumptions on the data (Hair et al., 2011). PLS is an appropriate methodology when the goal of the study is both to evaluate the validity of a research model, and to test new theoretical relationships within that model (Hair et al., 2017). As the purpose of this study is to test a new theoretical explanation (i.e. the strength model of self-control) linking SNS stressors to academic achievement and wellbeing outcomes, PLS-SEM is deemed an appropriate approach. In SEM analyses, a two-stage approach, first examining the measurement model and then the structural model is recommended (Hair et al., 2011). The assessment of the measurement model, provided in the supplementary information, demonstrates sufficient convergent validity, discriminant validity and reliability for all scales.

4.5 Assessing the measurement model

As all constructs in the model were reflective, we followed the procedures recommended by Hair et al. (2019) to test convergent and discriminant validity. The first step involves the examination of indicator loadings. Loadings above 0.708 are recommended. Two items on the seven-item deficient SNS self-control scale and one item on the six-item student vitality scale fell below this threshold. However, the item for student vitality was retained as the average variance extracted (AVE) for the construct remained above 0.50 (Hair et al., 2019). The two deficient SNS self-control items were removed from the analysis as their inclusion would have resulted in the AVE falling below 0.50.

The second step tests for construct reliability, which indicates the extent to which items on a scale are related in reality. We evaluated the internal consistency of the scale items by examining composite reliabilities (CR) and the convergent validity with the AVE values. The CRs being above 0.7 suggests satisfactory construct reliability (Fornell and Larcker, 1981). No CR value was above 0.95, which can be an indication of redundant items, thereby reducing construct validity. Convergent validity is the extent to which the construct converges to explain the variance of its items. AVE is the metric used for evaluating a construct's convergent validity. A construct that explains at least 50% of the variance of its items is considered acceptable (Hair et al., 2019). All AVEs in our data exceed 0.5 which further establishes satisfactory construct reliability (Fornell and Larcker, 1981). The descriptive statistics, loadings, CR and AVEs are shown in Table A1.

Discriminant validity, the extent to which a construct is empirically distinct from other constructs in the model, is assessed in the third step. For appropriate discriminant validity, the heterotrait-monotrait (HTMT) ratio, which is an estimate of the true correlation between two constructs if they were perfectly measured, should be below 0.90 for all constructs, or 0.85 for the more conservative 0.85 threshold (Henseler et al., 2015). Based on the HTMT values in Table 1, all the values did not exceed the threshold of 0.85, indicating that the variables in the dataset have sufficient discriminant validity. Henseler et al. (2015) also suggest that the traditionally used Fornell and Larcker criterion (Fornell and Larcker, 1981) also be employed as a support to HTMT values. This measure is obtained by comparing the square root of the AVE to the inter-construct correlation of that same construct and all other reflectively measured constructs in the structural model. If the square root of AVE (diagonal values in Table 1) is greater than the corresponding construct correlations, this suggests discriminant validity is present (Fornell and Larcker, 1981). This was the case for all our data.

As we used self-reported surveys to gather data, we tested for the potential influence of CMB. All CMB detection techniques have limitations, so we used a number of methods to assess for CMB. First, the occurrence of a variance inflation factor (VIF) greater than 3.30 for any latent variable is proposed as an indication that a model may be contaminated by CMB (Kock, 2015). Therefore, if all VIFs resulting from a full collinearity test are equal to or lower than 3.30, the model can be considered free of CMB. The VIF matrix confirmed all values were less than 3.30 (see Table A1 for values for each latent variable). Second, we conducted a single factor test (Harman, 1976) to check for CMB. We examined the unrotated factor solution in SPSS for all the items of our first-order constructs. The factor analysis revealed 12 distinct factors with the largest factor accounting for only 25.75% of the variance and found no single construct accounted for a majority of the total variance. These tests, along with the two-wave survey, ensure that CMB is not a major concern in this study.

4.6 Assessing the structural model

When the measurement model assessment is satisfactory, the next step in evaluating PLS-SEM results is to assess the structural model. Statistical significance and relevance of the path coefficients, coefficient of determination (R²), the blindfolding-based cross validated redundancy measure Q² and the model's out-of-sample predictive power should all be assessed at this stage (Hair et al., 2019).

The significance of path coefficients was determined via a bootstrapping procedure by setting the number of cases equal to the sample size (Tenenhaus et al., 2005) and the number of bootstrap samples to 5,000 (Hair et al., 2017). While the conventional approach for mediation testing begins with establishing the direct relationship between the independent and dependent variables (Baron and Kenny, 1986), contemporary thinking suggests this first step has no bearing on whether mediation is present or not (Hayes, 2017). Indeed, terms such as full or partial mediation are redundant. The indirect effect is either significant or not, regardless of the significance of the total effect (Hayes, 2017). Adopting the contemporary mediation approach, Figure 2 indicates that deficient SNS self-control mediates the relationship between SNS stressors and student vitality (H1 supported), satisfaction with academic life (H2 supported) and academic achievement (H3 supported) [1], [2]. The t scores and p values for each hypothesis are also provided in Table 2.

As hypothesized, the intensity with which a student used SNS for academic purposes moderated the relationship between SNS stressors and deficient SNS self-control (H4 supported). As depicted in Figure 3, the positive relationship between SNS stressor and deficient SNS self-control is stronger for those students who use SNS more intensely for academic purposes.

With respect to R², a variance of 16.6% is explained for deficient SNS self-control; 15.2% is explained for student vitality; 18.2% of the variance of satisfaction with academic life and 16.7% for academic achievement is explained in our research model. Of the control variables, academic motivation was positively related to all three outcomes. The demographic controls of age and gender were not significantly related to any construct. Similarly, trait self-control was not significantly related to any construct, including deficient SNS self-control. All four hypotheses remained supported when the control variables were excluded from the analysis, with only modest changes to the path coefficient strengths.

The f² effect size for the mediated relationships in H1, H2 and H3 are 0.07, 0.09 and 0.05 respectively. These effect sizes indicate that deficient SNS self-control has a substantive mediation impact on all three outcomes, albeit one that is in the low to moderate effect range (Hair et al., 2017). The f² value for the moderated relationship in H4 is 0.16, which is in the small effect range. These effect sizes are in line with the literature on digital media impacts where effect sizes are typically low and close to null (Johannes et al., 2021). Concerning the control variables, trait self-control had a weak effect on deficient SNS self-control (f² = 0.09). Academic motivation had a moderate effect on academic achievement (f² = 0.17) and weak to moderate effects on student vitality (f² = 0.12) and satisfaction with academic life (f² = 0.09).

Calculating the Q² is another means to assess the predictive power of the PLS path model (Hair et al., 2019). This measure indicates the model's out-of-sample predictive power and relevance. In a structural model, Q² values larger than zero for reflective latent constructs indicate predictive relevance for a particular dependent construct (Hair et al., 2017). In our model, all four dependent variables are above zero, with academic achievement the highest Q² (0.12), followed by student vitality (0.11), satisfaction with academic life (0.10) and deficient SNS self-control (0.07). Thus, these results provide support for predictive relevance of the latent constructs.

In a final assessment of the model's predictive validity, we conducted a PLSpredict analysis which uses a subsample of the data to predict another sample in a random repeated process. Researchers are advised to focus on the prediction errors of the model's key endogenous construct when interpreting PLSpredict results (Hair et al., 2019). If none of the prediction errors for each construct indicator are higher for the PLS-SEM assessment when compared to a linear regression version, then a high predictive power can be assumed (Hair et al., 2019). In our PLSpredict assessment, we selected deficient SNS self-control as the central construct. The linear regression prediction errors for all indicators were higher than the PLS-SEM version, thus confirming the model exhibits high predictive power.

It is also possible that our theoretical construction of the model, with the strength model of self-control explaining why SNS stressors lead to maladaptive academic outcomes, is inappropriate. To alleviate such concerns, we tested for potential reverse relationships in our model. First, we tested the alternative hypothesis that SNS stressors acted as the mediating variable between deficient SNS self-control and the academic outcomes. This hypothesis was rejected for each of the paths to the three academic outcomes. Second, we considered whether the relationship from deficient SNS self-control to SNS stressors was moderated by academic SNS use. This path too was not significant. Third, although not compatible with the strength model of self-control, we also tested if academic SNS use moderated the relationship between deficient SNS self-control and academic outcomes. Each of these moderated relationships was insignificant, which validates the positioning of academic SNS use in our research model. Together, these tests show that alternative relationships are unlikely to be an issue in our research model.

5.1 Research contributions

Our first theoretical contribution is to extend recent discussions on the impact of SNS on student outcomes (Muraven, 2010). Unlike previous studies which mostly consider how excessive time spent on SNS impacts academic outcomes (Junco, 2012; Kirschner and Karpinski, 2010), we provide a more nuanced understanding of how SNS use unfolds for students using the technostress and self-control perspectives. Such theory-driven approach in educational Information systems (IS) perspective is rare, and causal pathways developed in prior studies lack a proper theoretical foundation (Masood et al., 2020). Our study contributes by shifting the discussion away from a focus on the amount of time students spend on SNS, to an appreciation of students' perception of SNS (i.e. the stress arising from SNS use) and how that perception can influence important academic outcomes. Extensive use of SNS may only be a problem in that it takes time away from other activities (e.g. studying for exams). In the context of video games for example, increasing evidence supports the view that intense play may involve patterns of gaming that are characterized by high involvement but are not problematic (Billieux et al., 2019; Griffiths, 2010). This study places emphasis on the stressors resulting from SNS use which indirectly affect student wellbeing and success variables through self-control deficiencies. Our findings are consistent with the recent Troll et al.’s (2020) study of student attachment to smartphones, in that self-control over personal technology is central in explaining why some students fare better than others. While time spent on SNS use does play a role, it needs to be considered in conjunction with moderating variables, such as SNS self-control and gratification sought, to better understand how time spent on SNS influences important life outcomes. Furthermore, our study extends prior research which stops at the performance level (Cao et al., 2018; Luqman et al., 2017; Masood et al., 2020) by incorporating student psychological wellbeing into the educational SNS research agenda.

Second, this study extends previous works by yielding a more nuanced understanding of the moderating factors that bound the applicability of SNS stressor effects. We found that the use of SNS for academic purposes enhances the effect of SNS stressors on deficient SNS self-control. In other words, the stress generated from using SNS will weaken a student's self-control, and this weakening of self-control is even more pronounced when the student also uses SNS intensely for academic purposes. This finding sheds light on the boundary conditions, or contextual factors, on which the stress-related effects of SNS depends, a critical contribution to theory development and testing (Hayes, 2017). In explaining the observed association between SNS stressors and SNS addiction, Tarafdar et al. (2020) argue that users continue to engage in the SNS platform for alternative tasks in order to take their mind off the stress they experience when using SNS. Following this line of thought, as a coping mechanism, students may switch to using SNS to achieve academic goals when SNS stressors materialize. However, the technostress experienced from using SNS for their personal lives may leave self-control resources in a weakened state. Therefore, as our results suggest, a heavy reliance on SNS for academic purposes only serves to exacerbate the effects of SNS stress. For example, if using Snapchat for social purposes generates stress for a student, a switch to using Snapchat intensely to coordinate university assignments will result in less self-control and ultimately poorer academic outcomes.

Our literature search only returned one previous study (i.e. Chang et al., 2019) which considered how both academic and nonacademic SNS uses affect students' outcomes. In that particular study, the dual uses of SNS were examined in separate research models with compulsive nonacademic SNS use found to be negatively associated with academic achievement, and compulsive academic SNS usage positively associated with academic achievement. When examined in isolation, high academic use of SNS may well be positively correlated with academic achievement. However, people engage with SNS for multiple purposes which often interact. These interaction effects are encapsulated in our research model, the results of which suggest that high use of SNS for academic purposes indirectly leads to diminished student achievement and wellbeing. Therefore, our findings question those existing studies which report a direct positive relationship between academic SNS use and student performance (Chang et al., 2019; Lambic, 2016). Students can use the same SNS for both social and academic activities. Extrapolating from our findings, it is possible that the boundaries between hedonic (i.e. social interactions with friends) and utilitarian (i.e. developing academic knowledge) pursuits become blurred and confused when SNS is used for dual purposes. SNS was traditionally an environment for hedonic experiences. For students, the recent adoption of SNS to satisfy utilitarian needs may conflict with their perceptions of what SNS provides them. Indeed, they may feel they never get a break from SNS as it has become central to their social and academic lives. This conflict could lead to a lack of focus which strengthens the relationship between stressors and academic outcomes.

Our third contribution is to introduce the strength model of self-control into the technostress literature. This model advocates that self-control is analogous to the muscles in our body in that repeated stress (e.g. through lifting weights or avoiding dominant responses) depletes the energy needed to override subsequent acts requiring self-control (Baumeister et al., 2007; Baumeister and Tierney, 2012). For students, self-control seems to be particularly important (Duckworth et al., 2014). In experimental settings, ego-depleted undergraduate students performed significantly worse on tasks based on cognitive ability and fluency (Schmeichel et al., 2003, 2006). When stressed by exams, students' self-control performance was less efficient leading to a less healthy lifestyle during the exam period (Oaten and Cheng, 2006). Our findings confirm existing studies which suggest university students are particularly susceptible to SNS stressors (Cao et al., 2018; Masood et al., 2020). We go a step further and show that it is through the process of diminishing self-control over SNS use that SNS stressors inhibit performance and wellbeing outcomes. This is a new conceptual development because it validates an important causal pathway in technostress research and adds to our understanding of how the damaging effects of technostress can be mitigated.

5.2 Practical contributions

This study has important practical implications. Deficient control over SNS use is central in explaining how SNS stressors lead to problems. Extending the muscle analogy, research has shown that self-control can be strengthened in the long run by regularly performing self-control acts. Such training improves the efficiency or availability of self-control resources (Hagger and Chatzisarantis, 2013), and should be considered by individuals and their caregivers as a focal point for effective intervention strategies. For instance, research has shown that engaging in short-term acts (e.g. controlling speech, using the nondominant hand) lead to improved performance on self-control tasks under depletion (Muraven, 2010). Similarly, SNS users may believe that switching their use to different SNS features or purposes will enable them to cope with SNS stressors. However, even a low use of SNS for academic purposes will only partially offset the effects of SNS stressors on self-control. Users are thus advised to take a complete break from using SNS on the onset of stress. Insights from the boundary management literature (Bulger et al., 2007) are pertinent here too. To alleviate SNS technostress, students should separate the personal life use of SNS from the academic use, rather than continuously switching between the two.

Providers of SNS applications can also benefit from this study. Problematic users of SNS are more likely to quit using the app, which would adversely affect the providers’ revenue model. While numerous digital self-control tools exist to temper SNS use, such as feature minimization, app blocking and gamification awards, to the best of our knowledge none directly addresses the stress generated by SNS use. Numerous stress-sensing technologies can now be embedded in devices accessing SNS (Whelan et al., 2018). These technologies can be exploited by SNS providers to alert users to rising stress levels and “nudge” them to less stressful activities.

Finally, because the COVID-19 mandated remote teaching, SNS applications are increasingly being used in universities to disseminate learning content and facilitate interactions between students and instructors. While SNS provides efficiencies and is often the preferred mode of interaction by students, educators also have a duty of care to students. Educators need to be aware that using SNS for academic purposes is cognitively taxing on students. Mandating that students use these technologies may only serve to diminish self-control even further (Englert et al., 2017). Students should be offered the opportunity to choose to some degree when and how to use SNS to support specific learning content.

5.3 Limitations and future research

This study is subject to a number of limitations. First, our sample had an overwhelming preference for using media sharing SNS, which is only one category of SNS. A fruitful area of future enquiry would be to compare the association between SNS stressors and self-control across different categories of SNS, for example, professional SNS like LinkedIn compared to discussion SNS like Digg. Second, our study relied on self-reported data which can be subject to some biases, ranging from a misunderstanding of what a proper measurement is, to social-desirability bias. Prior technostress research suggests that the combination of physiological measures of stress along with psychological measures yields a more valid representation of the stress construct (Tams et al., 2014). To advance our understanding of SNS stressors, future studies should aim to incorporate physiological measures of stress into their investigations. This need not be an expensive proposition. Wearable technologies such as the Fitbit now capture heart rate variability, which is a reliable indicator of stress. Third, the wellbeing outcomes measured in this study were customized specifically to student academic life. It is possible that a student could thrive in their academic life, but struggle in other aspects such as family life. Future studies could examine if and how SNS stressors contribute to conflict between the academic and family life domains. Finally, our study only considered between-person effects. Cross-sectional self-reported psychological measures can fluctuate from day to day and even over the course of a day. Future studies should adopt an experience sampling study which would also assess within-person effects over time. Such an approach would allow researchers to assess how the perception of SNS stressors in an individual, at a point in time, influence other variables at a later point in time.