Subjective wellbeing, a conceptualization of hedonic experience, can be defined as when an individual feels and evaluates their lives as pleasant and gratifying (Diener et al., 1999). Subjective wellbeing is said to be high when emotional/affective experiences are more positive than negative (‘happiness’) and when, concurrently, an individual’s cognitive judgments about their life circumstances relative to personal criteria like goals and values are good (‘satisfaction’) (Diener et al., 2002). Thus, subjective wellbeing has been widely conceptualized over the last 30 years as having a tripartite structure including positive affect, lack of negative affect, and satisfaction with life (Busseri and Sadava, 2011). Numerous studies have shown that these three components are highly correlated (Keyes et al., 2002) and tend to load on a general subjective wellbeing factor (Jovanovic, 2015). As such, it is common in the literature for studies to calculate and use composite subjective wellbeing scores (e.g., Cloninger and Zohar, 2011).

Promoting health and subjective wellbeing in university students is now widely recognized as being essential for promoting various other outcomes. For example, research suggests that wellbeing is an important antecedent to better academic performance (El Ansari and Stock, 2010), reducing mental health risks (Campbell et al., 2022), fostering personal development, growth, and self-discovery (Yu et al., 2018), and strengthening social connections and support (Cobo-Rendón et al., 2020). Furthermore, promoting subjective wellbeing in university students extends beyond the individual, contributing to a positive and thriving campus culture (Travia et al., 2020). Therefore, subjective wellbeing is pivotal for helping university students excel academically, emotionally, and socially, thus preparing them for success in their university years and beyond.

Given emotion regulation is frequently done to promote a positive rather than negative emotional/affective experience, it is unsurprising that past research has revealed that certain individual ‘adaptive’ and ‘maladaptive’ emotion regulation strategies are differently related to indicators of wellbeing. A classic study by Gross and John (2003), for example, found that cognitive reappraisal was significantly positively correlated with life satisfaction and various indicators of psychological wellbeing. Expression suppression, conversely, was significantly negatively correlated with the same outcomes. However, as far as we are aware, despite research testing the implication of different emotion regulation profiles on clinical variables such as depression and anxiety (Chesney et al., 2019), we are unaware of any study that has examined how such profiles relate to subjective wellbeing.

Internet usage has seen a remarkable global surge, with one recent estimate classifying 62.5% of all people as internet users (We are Social and Hootsuite, 2022). This widespread use of the internet offers benefits for societies and individuals. For example, internet use can alleviate feelings of loneliness (Phu and Gow, 2019) and has been linked to elevations in university student satisfaction with campus life (Kim et al., 2020). Unfortunately, with the dramatic growth of internet use in the last decades there has also been a significant rise in the proportion of individuals with difficulties regulating their internet use (Pan et al., 2020; Lozano-Blasco et al., 2022). In extreme cases, such difficulties can lead to highly excessive use with characteristics of clinical addiction (Block, 2008), leading some countries to consider it a serious public health issue. Indeed, the concept of internet addition has been widely studied and shown to be comorbid with anxiety and depression (Lee et al., 2008) and other issues such as alcohol dependence (Ko et al., 2008). However, there remains a controversy over whether internet addiction is a true addiction or distinct from other compulsive-impulsive disorders (Yellowlees and Marks, 2007). As such, some researchers use the term “Problematic Internet Use” interchangeably with “Internet Addiction” (Shapira et al., 2003; Montag et al., 2010).

Research indicates that university students may be more susceptible to internet addiction than the general population (Anderson et al., 2017). This trend is concerning because internet addiction is associated with adverse psychosocial outcomes for students, including academic difficulties (Marahatta et al., 2015) and a heightened risk for mental health issues and psychopathology [see meta-analysis of 223 studies by Mannino et al. (2021) and Cai et al. (2023)]. Further empirical evidence underscores the negative impacts of internet addiction on young adults’ subjective wellbeing (Zhao, 2021), with some studies revealing a reciprocal relationship between internet addiction and indicators or illbeing and psychopathology (Zhu et al., 2021).

Several studies have identified a significant positive association between difficulties in emotion regulation and internet addiction among young people [see, for example, the review of 23 studies published between 2010 and 2020 by Gioia et al. (2021)], with internet addiction suggested as a coping mechanism for these difficulties (Caplan, 2010). In other words, emotion dysregulation may be a risk factor for internet addiction because internet use can offer a way of managing or escaping negative emotions. Consistent with this, research on young adults has shown that individuals with internet addiction tend to report greater difficulties in emotion regulation (Evren et al., 2018; Pettorruso et al., 2020). Furthermore, individuals with internet addiction are found to often employ ‘maladaptive’ emotion regulation strategies (Caplan, 2002). Interestingly, some research has suggested that internet addiction could be both a consequence of, and contributor to, emotion dysregulation (Hormes et al., 2014). Despite these insights, there remains a notable gap in the literature: no study has explicitly examined how different emotion regulation profiles in university students relate to their levels of internet addiction.

Characterized by its recurrent nature, intrusiveness, and difficulty to disengage from (Ehring and Watkins, 2008), repetitive negative thinking often stems from distressing experiences and can significantly disrupt daily life (Magson et al., 2019). The two most intensively studied expressions of repetitive negative thinking are worry (linked to anxiety) and rumination (linked to depression). Worry has been defined as a predominantly verbal thought activity, which is negatively affect-laden, relatively uncontrollable, and focused on uncertain events with the potential for a future negative outcome (Borkovec et al., 1983). In turn, rumination can be defined as a maladaptive and repetitive thinking style that is activated in response to negative internal and/or external triggers (Nolen-Hoeksema et al., 2008).

Repetitive negative thinking is involved in the maintenance of multiple types of emotional problems and has therefore been described by various researchers as a transdiagnostic cognitive processing style (Ehring and Watkins, 2008; McEvoy et al., 2013). Consistent with this perspective, repetitive negative thinking has been considered as one of the possible mechanisms through which emotion dysregulation occurs and is maintained (Gaynor and Gordon, 2019), and through which individuals may be at increased risk for heightened anxiety, reduced self-esteem, and emotional distress, thereby affecting life satisfaction and wellbeing (Berry et al., 2010; Peixoto and Cunha, 2022). Indeed, rumination, and therefore repetitive negative thinking, is frequently considered in research on emotion regulation as a maladaptive strategy (Kashdan et al., 2012). However, to the best of our knowledge, there are no studies on how emotion regulation profiles differ in repetitive negative thinking.

Participants were 261 university students from Portuguese universities across Portugal, of which 175 were undergraduates and 86 were postgraduates (see Table 1). Approximately half of the participants were Psychology students (n = 128), with the remainder enrolled in a variety of courses including architecture (n = 7), law (n = 15), nursing (n = 6), engineering (n = 9), marketing (n = 9) and history (n = 4), among others. The sample comprised 213 women (81.6%) and 44 men (16.9%). The mean age across participants was 22.51 years (SD = 5.30). On average, the participants reported spending 4.70 h per day (SD = 2.16) on their smartphone or computer for leisure (e.g., social media, gaming).

Participants responded to items that assessed demographic variables including gender (male, female, and other), age, nationality (Portuguese, other), and level of degree currently enrolled in (undergraduate, master, Ph.D.). Participants also indicated the average number of hours they spend per day on various social networks.

The PINTS is a unidimensional 5-item self-report measure of repetitive negative thinking (Magson et al., 2019). Participants respond to statements that evaluate three core features of repetitive negative thinking: (a) repetition, (b) intrusiveness, and (c) difficulty disengaging from such thoughts (Ehring et al., 2011). Participants rate each statement from 1 (never) to 5 (almost always). The sum scores for the PINTS thus fall within the range of 5–25, with higher scores indicating a higher presence of repetitive negative thinking. The Portuguese version of the PINTS used in this study has been shown to be a reliable (α = 0.88) and valid (positive correlation of r = 0.52 with a measure of Generalized Anxiety Order; negative correlation of r = −0.54 with a measure of Satisfaction with Life) measure in a sample of 432 Portuguese adults (age range = 18–73 years; Peixoto and Cunha 2022). Internal consistency reliability was also good in the present study sample (ω = 0.85).

The IAT is a 20-item self-report questionnaire. As described in its manual (Young, 2017), this test is designed to measure problematic behaviors associated with compulsive use of technology. For each item, participants rate the extent to which they endorse a particular addictive behavior on a Likert-type scale ranging from 0 (never) to 5 (always) (example item: How often… Do you find that you stay online longer than you intended?). Higher scores reflect greater evidence of internet addiction, but are not necessarily diagnostic of a psychopathological condition. Authors have debated what total score across IAT items would represent a threshold for a clinical diagnosis of “addiction” (Young, 2011) – although for the present study, we treated IAT scores as a continuous variable. The IAT is arguably the most widely validated measure in this domain, with demonstrated psychometric adequacy in various samples globally. The Portuguese version of the IAT used in this study has been shown to be valid (strong positive correlation of r = 0.82 with the Beck Depression Inventory-II) and reliable (α = 0.90) in a large sample of Portuguese students (age range = 15–39 years; Pontes et al., 2014). Consistent with these past findings, in our study sample, the internal consistency reliability was found to be good (ω = 0.88).

The DERS-SF (Kaufman et al., 2016) is a self-report measure of difficulties in emotion regulation. The 18 items of this scale load on six subscales: (1) Lack of emotional awareness; (2) Lack of emotional clarity; (3) Non-acceptance of emotional responses; (4) Limited access to emotion regulation strategies; (5) Difficulties engaging in goal-directed behavior; and (5) Impulse control difficulties. Participants indicate how much each item applies to themselves from 1 (almost never) to 5 (almost always). For this study, we calculated sum scores for each subscale. The Portuguese version of the DERS-SF we used was shown to be reliable (α across subscales = 0.70–0.91) in a sample of 646 Portuguese adults (age range = 18–66 years; Gouveia et al., 2022). Consistent with this finding, the internal reliability values of the six subscales in the present study all surpassed the required threshold of acceptability (ω = 0.74–0.89).

The PANAS has 20 items, each presenting an adjective that describes an emotional experience (Watson et al., 1988). These items load on two subscales that capture Positive Affect (example item: “Excited”) and Negative Affect (example item: “Afraid”). Participants rate the extent to which they had experienced each emotion in the past 2 weeks on a scale ranging from 1 (very slightly or not at all) to 5 (extremely). The Portuguese version of the PANAS is widely used and its structure and psychometric adequacy are well supported (e.g., Galinha et al., 2013). For example, in a sample of 348 Portuguese adults aged 18–50 years, both the positive and negative subscales were found to have high internal reliability (α = 0.86 and 0.89, respectively) (Galinha and Pais-Ribeiro, 2005). Consistent with other works, in our sample the internal reliability of the positive and negative affect scales was also excellent (ω = 0.91 and 0.88).

As is common in studies using the PANAS, we summarized each participant’s perceived affective wellbeing by forming a composite variable (or Happiness Index; Josefsson et al., 2011). We calculated this variable by subtracting the mean average of their scores across negative affect items from the mean average of their scores across positive affect items.

The WHOQOL-BREF is a short-form assessment of quality of life developed by the World Health Organization WHOQOL Group (WHOQOL Group, 1998). Conceptually, quality of life reflects an individual’s perceptions of their life position concerning the cultural and value systems they are part of, as well as their goals, expectations, standards, and concerns. In this way, quality of life captures the cognitive (non-affective) component of wellbeing.

Within this scale, 24 items group in four domains: physical health, psychological health, social relationships, and physical environment. Two additional items request global ratings for quality of life and satisfaction with health. Participants rate all 26 items from 1 to 5, with higher scores reflecting greater perceived quality of life. The Portuguese version of the WHOQOL-BREF has been shown to be reliable (α across all items = 0.92) and valid (significantly lower values in hospital patients vs. healthy controls) in a sample of 604 Portuguese adults (Vaz Serra et al., 2006).

To summarize each participants’ cognitive wellbeing, we formed a composite variable (or Composite Health Index; Josefsson et al., 2011) by calculating the mean average of their scores across all WHOQOL-BREF items, excluding item 26: “How often do you have negative feelings such as blue mood, despair, anxiety, depression?” We excluded this item because it contaminates the measure with depressive symptomatology (Aigner et al., 2006) and is conceptually closer to affective wellbeing than cognitive wellbeing. In our study sample, the internal reliability consistency across the considered 25 items was excellent (ω = 0.92).

The study had almost no missing data because of its online format. However, we identified one missing case for one item of the PANAS. For analysis, we replaced this missing using the series mean.

Subjective wellbeing is a multidimensional construct consisting traditionally of an affective and cognitive component (Diener et al., 1999). To summarize each participant’s subjective well-being, we computed a final composite score by calculating the mean average of their scores on both the Happiness Index (from PANAS) and the Composite Health Index (from WHOQOL-BREF). This approach has been widely demonstrated to be a valid way to study subjective wellbeing (Cloninger and Zohar, 2011; Josefsson et al., 2011).

Participant characteristics were analyzed using descriptive statistics, which included calculating means and standard deviations for continuous variables and frequencies (as percentages) for categorical variables. We calculated Pearson’s correlation coefficients to explore the relationships among study variables.

We used a common person-centered mixture modeling analysis (Oberski, 2016) to identify different latent types, or subgroups, of individuals based on their standardized mean scores for the six DERS-SF subscales. Specifically, we performed latent profile analysis (LPA) given the continuous nature of these observed variables.

The first step of the LPA process was to decide on the optimal number of latent profiles to explain the observed data. This was done by estimating the parameters, using maximum likelihood estimation, of a series of models that differed in their number of latent profiles (from a 2-profile through to 10-profile solution). Fit indices obtained from each of the models were then compared to determine the model with best fit. Research recommends making this decision based on multiple statistical indices and theoretical considerations (Spurk et al., 2020). For this study, we inspected and compared the following indices: Akaike information criterion (AIC), Bayesian information criterion (BIC; Schwartz, 1978), and sample-size adjusted BIC (SABIC; Sclove, 1987). For these three indices, the model with the lowest value was deemed to have the best fit (Nyland et al., 2007). We also used the bootstrap likelihood ratio test (BLRT; McLachlan and Peel, 2000) to compare models with increasing numbers of latent profiles. The solution prior to an occurring non-significant value (p > 0.05) was judged the best fitting. Considering recommendations for best practice we also inspected profile size (Lubke and Neale, 2006).

After deciding on the optimal number of profiles, we sought to validate the profiles by testing how they differed in terms of theoretically relevant outcomes (i.e., internet addiction, repetitive negative thinking, and subjective wellbeing). To do this, we used the WRS2 package in R to perform a series of robust ANOVAs (Mair and Wilcox, 2020). These robust procedures were employed to counter potentially problematic variances across classes of unequal sizes. As an indicator of effect size, WRS2 calculates an explanatory measure of effect size ξ (Wilcox and Tian, 2011). Values of 0.10, 0.30, and 0.50 correspond to small, medium, and large effects, respectively. Robust ANOVAs were followed by procedures for robust post-hoc comparisons.