Discussion

This study aimed to explore the longitudinal associations between loneliness and smoking-related outcomes among adolescents (specifically smoking initiation) and among university students (specifically smoking status). Convergent evidence from our multi-dataset approach did not provide strong support for loneliness predicting smoking initiation in adolescents. Additionally, although one suggestive finding was observed in one of the two samples (for the long-term effect only), no compelling support was found for baseline loneliness predicting later smoking status in university students based on the main analyses. Complementary analyses suggested that loneliness might be associated with being a regular smoker but not an occasional smoker.

Reflection on Results Adolescent Samples

The findings of the three adolescent samples converged, providing very little support for loneliness predicting smoking initiation in adolescents. Previous studies focusing on longitudinal associations were scarce. The only study that explored the longitudinal relationship between loneliness (in school) and smoking initiation also did not find support for this association using a multivariate analysis approach, but only in univariate analyses (Park, 2009). While other longitudinal (Qualter et al., 2013) and cross-sectional studies have addressed smoking behaviours (Dyal & Valente, 2015; McClure-Thomas et al., 2022; Shadmehr et al., 2019), such as smoking quantity but not initiation, their findings have been mixed. Differences in study designs (e.g. sample characteristics, follow-up periods, cross-sectional designs) and the use of varying smoking measures (e.g. frequency, number of cigarettes, smoking status) limit the comparability of their results with our study.

Interestingly, our study revealed a significant impact of educational type on the likelihood of smoking initiation among adolescents. Specifically, adolescents enrolled in pre-university education programs were less likely to start smoking compared to the average likelihood across all educational types. There could be different explanations for this observation. A study among 1860 adolescents showed that in vocational tracks, popularity norms for smoking and alcohol were more positive and predicted classroom differences in smoking and alcohol (Peeters et al., 2021). In general, attending lower education increases the likelihood of initiating smoking because of limited health literacy (Sørensen et al., 2015).

Reflection on Results in University Student Samples

The current study did not provide compelling support for baseline loneliness predicting smoking status at 6 and 18 months. However, one suggestive finding was observed in the HSL sample, which showed that baseline loneliness might predict smoking status 18 months later. No previous studies assessing these longitudinal associations in university students were available. Longitudinal studies in (dwelling) older adults do suggest loneliness might predict current smoking (Adebisi et al., 2024; Yang et al., 2022). However, the comparability might be limited given the geographical differences and age of the participants.

The lack of convergence between the two university student samples on baseline loneliness predicting smoking status at 18 months might be explained by the sample size of the HSL project. The association between smoking and loneliness has been proposed to have a small effect size (DeWall & Pond Jr., 2011). Larger samples are more likely to find associations between loneliness and smoking (Dyal & Valente, 2015). This seemed to be the case in the current study: A significant association between loneliness and smoking status was observed in the largest dataset (HSL: n = 2636) and not in the smaller one (al-RISCO: n = 404), but the effects were in the same direction. Further support for this explanation is that sensitivity analyses with complete cases (i.e. reduced sample size) showed that the significant effect was not observable anymore in the HSL sample. In addition, smoking prevalence in both samples was relatively low (decreasing power); for example, at T1, non-smokers constituted 73%–86% of both samples. Additionally, drop-out rates were acceptable, but attrition analyses for the al-RISCO sample indicated that more frequent smokers were more likely to discontinue participation (see Supplementary Materials X.2). This probably further contributes to limited power in this sample.

To better understand the suggestive finding, we have complemented the main analyses with nominal regression analyses, leading to separate odds ratios for occasional smokers (compared to non-smokers) and regular smokers (compared to non-smokers). Baseline loneliness was associated with being a regular smoker but not an occasional smoker in the HSL dataset. A similar pattern, although not significant, was seen in the al-RISCO dataset. Occasional and regular smokers have different characteristics. In general, regular (daily) smokers score higher on nicotine dependence and smoke for negative reinforcement motives (to avoid withdrawal symptoms) (Mathew et al., 2014). The occasional smokers might smoke more often in social situations, for example, with friends, and for positive reinforcement motives (Oksuz et al., 2007). Therefore, loneliness at baseline could form a higher risk of smoking regularly (coping with loneliness) than occasional smoking (social smoking).

General Reflection

Several explanations could clarify the limited support for longitudinal associations between loneliness and smoking in both the adolescent and university student samples. One explanation pertains to the underlying motivation for smoking (Dyal & Valente, 2015). Peer influence plays a crucial role in shaping young people’s health behaviour, including smoking (Montgomery et al., 2020). For instance, previous studies suggested that young people may smoke to increase their social acceptance and sense of belonging (Brown et al., 2011; DeWall & Pond Jr., 2011). However, peer influence on motivation to smoke might be particularly relevant in cultures where the behaviour is viewed as a means of gaining popularity. This study focuses on the Netherlands, where smoking rates have been declining since 1999. The decline in smoking rates has also been reflected in the current study as the smoking prevalence at baseline was higher among the adolescents in the F&H study (35.7%; conducted between 2002 and 2003) in comparison to the smoking prevalence at baseline among adolescents in the KLS (4.1%; conducted between 2012 and 2014) and GFT studies (4.2%; conducted between 2017 and 2020). Although smoking among young people remains undoubtedly worrisome, awareness of the health impacts of smoking appears to be effective. This is also reflected in current views on smoking. For example, a Dutch report (n = 1008) showed that 80% of youngsters considered smoking as ‘not cool’ any longer and not sociably desirable (Gezondheidsfondsen voor Rookvrij, 2020). To further support this observation, a study of adolescents in the Netherlands from four secondary schools (n = 875) concluded that smoking initiation is significantly influenced by friends’ attitudes towards smoking (i.e. what they think or say about it) (Huisman, 2014). Cultural acceptance of smoking, or general lack thereof, among young people in the Netherlands, might explain why the current study did not find strong support for loneliness predicting smoking.

The lack of strong support for loneliness predicting smoking among general adolescent and university student samples, as included in the current study, does not necessarily mean that this is also the case for certain subgroups who are more at risk of experiencing loneliness or initiating smoking. In our samples, the observed levels of loneliness were relatively low, and most participants did not smoke. Regarding subgroups with a higher risk for loneliness: A review by Bayat et al. (2021) indicated that contextual factors like being bullied at school, poor student–teacher relationships, parental divorce, experiencing illness of a close family member, and problematic use of social media were positively associated with loneliness. Alternatively, being non-native and having lower social connection were also positively associated with loneliness (Bayat et al., 2021; Moore et al., 2023), as well as personal characteristics such as shyness, low self-esteem, and poor social skills (Mahon et al., 2006). Regarding subgroups with a higher risk for smoking: Although general views on smoking have changed over the last years (Gezondheidsfondsen voor Rookvrij, 2020), certain subgroups might still smoke to gain social acceptance, and relief from loneliness, for young people who wish to belong to a peer group where smoking is considered ‘cool’. Especially since previous work has acknowledged that young people who have friends who smoke are more likely to initiate smoking themselves (O’Loughlin et al., 2017; Wellman et al., 2016). Hence, future research could examine whether a longitudinal link between loneliness and smoking exists in higher-risk samples, such as those susceptible to peer influence or experiencing poorer mental health. These studies could also control for potential confounders of the link between loneliness and smoking, such as socioeconomic position, depressive symptoms, or stress, as previously suggested by Philip et al. (2022), Dyal and Valente (2015), and Wootton et al. (2021), respectively.

Alternatively, the instruments previously used to assess loneliness might have contributed to the mixed findings from earlier work. Of the six cross-sectional studies supporting a positive association between loneliness and smoking among adolescents in the systematic review by Dyal and Valente (2015), five used a one-item measure of loneliness (including the word ‘lonely’). Marangoni and Ickes (1989) have raised concerns about this, as not everyone recognises themselves as feeling lonely, and others might want to avoid identifying themselves as lonely, given the stigma associated with loneliness (Marangoni & Ickes, 1989). Single-item loneliness measures might assess a specific sub-dimension or variant of loneliness (which might be associated with smoking (initiation)). In contrast, our three adolescent samples used a 12-item scale focusing on loneliness in peer relations, and the two university student samples an 8-item (HSL) or 3-item scale (al-RISCO) reflecting more general feelings of loneliness. Possibly, these instruments capture different aspects of loneliness (which might not be associated with smoking initiation in general adolescent samples or smoking status in university students).

Strengths and Limitations

In this study, we used a multiple-dataset approach where we answered the same research question within three adolescent and two university student samples. Given that no single study can provide a definite answer to the question of whether longitudinal associations between loneliness and smoking (initiation/status) exist, convergence across multiple datasets provides stronger confidence in our conclusions and provides more robust insights (Hammerton & Munafò, 2021).

Several limitations need to be acknowledged as well. First, the sample sizes varied between the included samples, and although we tried to align the included measures as much as possible, the loneliness and smoking measures were not completely the same for all adolescent or all university student samples. Concerning loneliness, in some samples, brief loneliness scales were used while in others, more comprehensive measures were used. Specifically, among students, the HSL sample used the UCLA scale with eight items, while the al-RISCO sample used the short version with three items (Hughes et al., 2021; Russell et al., 1980). Although both scales exhibit acceptable psychometric properties, the 8-item version tends to have better psychometric properties compared to the 3-item version, offering a more reliable measure of loneliness (Lin, 2022). Consequently, the slightly different measures might limit the compatibility of the findings. However, the scales are conceptually comparable overall, as all capture the subjective experience of loneliness (Maes et al., 2022). Also, previous research has shown a high degree of convergence between RULS-8 and the original UCLA-Revised scale (r = 0.92), highlighting that both scales assess the same underlying construct (Goossens et al., 2014). Concerning smoking measures, we used different measures per examined developmental stage. For adolescents, we zoomed in on smoking initiation, given that at this stage, most adolescents have not yet progressed to regular smoking. For young adults, we zoomed in on regular smoking, given that at this stage, most young adults have initiated or experimented with smoking. The use of different measures across the utilized samples might cause discrepancies and reduce the comparability of the findings across the samples. However, we do believe this impact is limited, given the high level of observed convergence in findings across the samples.

Additionally, for university students, we used a categorical outcome with three categories to assess smoking status. Combined with the low prevalence of smokers, this may have limited the statistical power of our analyses. Another methodological limitation involves the time intervals between surveys for one of the adolescent samples (GFT), which varied and potentially affected the detection of smoking initiation. Future studies should carefully consider the time scale at which the process (of loneliness leading to smoking initiation) would take place. Also, it is important to note that the young adult samples in our study were imbalanced in favour of women, and our findings are, thus, generalizable with gender constraints. Future studies should strive for more gender-balanced young adult samples.

Another possible limitation is that we used datasets collected in schools and universities in the Netherlands specifically. Thus, our results may not be generalizable in cultures that have different social norms and societal perceptions. Future studies can test if socio-cultural factors play a role in the relation between loneliness and smoking, and compare the findings across cultures.

Moreover, reliance on self-reported data for smoking and loneliness introduces the risk of social desirability bias and recall inaccuracies (Grimm, 2010), which can affect the validity of the findings. In line with this, attrition analyses among the young adult samples showed that variables like loneliness, smoking status, and age (only sample 4) were linked to a higher risk of dropping out (see Supplemental Material X). While it is difficult to explicitly distinguish between missing at random (MAR) and missing not at random (MNAR) missingness patterns (Hughes et al., 2021; Little et al., 2014), the findings of the attrition analyses suggest that missingness may have depended on prior loneliness or smoking scores. This could have led to underreporting of loneliness and smoking, and potentially, biased findings (Hughes et al., 2021; Little et al., 2014). Specifically, when individuals with more feelings of loneliness or individuals who smoke were more likely to drop out, averages and associations involving these variables may be attenuated due to the underrepresentation of higher-risk individuals (Weuve et al., 2012). Additionally, this could reduce the generalizability of the findings, as the sample might not have been fully representative of the general student population in the Netherlands. Future studies should be mindful of missing data patterns. Statistical procedures such as Bayesian Models or Heckman Selection Models can help to account for missing data patterns (Galimard et al., 2018; Linero & Daniels, 2018). It is also important to clarify what constitutes smoking initiation. While one puff of a cigarette may not fully equate to smoking initiation, it is often considered the beginning of experimenting with smoking. Consequently, a more robust measure, such as smoking at least one whole cigarette, may more accurately define smoking initiation.

Furthermore, the COVID-19 pandemic covered some periods of the study for samples such as from GFT, HSL, and al-RISCO, which might have affected loneliness experiences and smoking behaviours. However, the latter might not be affected to a high extent, as previous work among university students indicated that smoking behaviours remained stable when comparing before and during the first COVID-19 lockdown (van Hooijdonk et al., 2022). In contrast, several reviews have indicated that mental health problems as well as loneliness increased among younger people during the COVID-19 pandemic (Ernst et al., 2022; Pai & Vella, 2021; Saulle et al., 2022).

Last, in the current study, we focused on assessing whether loneliness predicted smoking, while the temporal relationship could also occur in the other direction (smoking predicting more loneliness). The relationship between these variables could be reciprocal, with loneliness influencing smoking behaviour and smoking behaviour potentially exacerbating feelings of loneliness (Wootton et al., 2021).

Conclusions and Implications for Adolescents and University Students

Overall findings from our multi-dataset approach did not provide compelling support for (baseline) loneliness predicting smoking initiation (among adolescents) or smoking status (among university students). Complementary analyses in the student samples suggested that baseline loneliness might be associated with later regular smoking but not occasional smoking. Future studies could focus on exploring longitudinal associations in higher-risk groups for either loneliness or smoking, such as those susceptible to peer influence or experiencing poorer mental health. Our findings contribute to the limited longitudinal research on the association(s) between loneliness and smoking (Dyal & Valente, 2015). However, more longitudinal research with larger sample sizes is needed to make stronger claims.