The history of medicine and social science is littered with the wreckage of ostensibly good ideas that were founded on the belief that an epidemiological association was causal. When interventions are developed on this basis, the intervention often disappoints. Beware assumptions regarding causality!
This study was published last year and is based on a secondary analysis, mainly by UK researchers, of data available from a longitudinal cohort study of adolescents in the USA. The paper offers really interesting insights about the dynamic relationship between ‘hobbies’ (somewhat idiosyncratically defined, including, in later data points, ‘shopping for fun’) and the commonest forms of substance misuse in adolescents (tobacco, cannabis use and binge drinking alcohol). Unfortunately, these interesting findings are somewhat obscured by an assumption of causality that is repeatedly stated through the entire paper.
This paper explores the relationship between ‘hobbies’ and some of the most common forms of substance misuse in adolescents.
Methods
Data was provided from The National Longitudinal Study of Adolescent to Adult Health (ADD Health) cohort, which started collecting data 30 years ago. Participants and some of their parents provided data, which seems to have formed a single data set. ADD Health collected much more data than is reported here. The sample was taken from school rosters, said initially to be representative of the US population as a whole and demographic data is reported. Hobbies are defined as non-money generating purposeful activities pursued voluntarily for pleasure. Involvement in smoking, binge-drinking and using cannabis over time are the outcomes of interest. These common and, at the time, partially socially endorsed forms of substance misuse are intrinsically damaging and worth measuring, but it is important to recognise that this is not a study of heroin or cocaine use.
The sample was divided into the groups: young adolescents (11 to 14 years); middle adolescents (15-16 years); and late adolescents (17-20 years). Data was collected in five waves over the subsequent 22 years. There was adjustment of findings for potential confounders that had been measured. These included gender; level of education; race; first language spoken at home; urbanicity of home area; parent marital status; annual household income; parental education; parental smoking and alcohol use.
Results
At baseline there was a negative relationship between hobby involvement and substance misuse in all age groups. As the cohorts aged, this continued to hold for younger adolescents, but reversed for older adolescents; in other words involvement in hobbies in older adolescents was, to some extent, associated with substance misuse (as defined here).
Studies of hobbies and substance misuse have been conducted in other countries and cultures with variable results. This variability might be partly accounted for by the dynamic and changing nature of the relationship as individuals develop, which cross-sectional studies will not detect. The hobbies of younger adolescents (for example, learning an instrument or attending scouts) are organised by adults; the hobbies of older adolescents (playing in a band or participating in a sports team) involve a lot more autonomy and lead to greater exposure to peer pressure to abuse substances. This may explain the reversal of the relationship over time. In addition, differential effects at various time points may be due to changes in the general culture. Smoking and drinking were more socially acceptable at the beginning of the period studied than at the end of it; the opposite was true of cannabis use. Neither hobbies nor misuse of various substances are unitary phenomena and their nature changes at different times of life.
The authors used latent growth analysis to make sense of the trajectory of each type of substance misuse, involving following the (non-linear) trajectory of substance misuse and selecting one model of best fit. This allowed them to come to the conclusion that, fundamentally, hobbies seem to be protective against substance misuse, independent of the potential confounding factors they controlled for.
This study found that hobbies are protective against substance misuse in adolescents, independent of potential confounding factors.
Conclusions
The authors conclude that they have drawn out some of the complexities and changes over adolescent life course that are seen in the relationship between hobbies and substance misuse. Notwithstanding the dynamic nature of the relationship, the authors clearly believe that hobbies are a good thing in adolescents and prevent substance misuse. This conclusion is reached using statistical methods that are unlikely to be well understood by most readers, but nonetheless, the findings are interesting. The authors state “…these findings demonstrate the importance of considering developmental differences when doing research on hobby engagement”. It is impossible to disagree with this conclusion. What to do about the relationship between hobbies and substance misuse is another thing.
The findings of this study highlight the complexities and changes over the adolescent life course that are seen in the relationship between hobbies and substance misuse.
Strengths and limitations
The strengths of this study are:
- It is longitudinal
- A wide range of potential confounders have been taken into account
- Adjustments have been made to retain representativeness after attrition
- Substance misuse is not assumed to be a unitary phenomenon
- Degree of engagement with hobbies is taken into account
The weaknesses of the paper divide into two groups: problems associated with all secondary analysis of longitudinal cohort studies and problems associated with a priori beliefs of the research group. The former include:
- Attrition. Data was weighted for this, but weighting will only work if attrition affected those parts of the population randomly. If some types of people in the attrition group are more likely to drop out, weighting can make findings less reliable, not more, especially where that factor has not been measured; a type of uncontrolled statistical bias. There is no description of the demography of who dropped out, but the original ADD Health paper suggests that attrition was not random.
- In later waves, new prompts to participants were introduced about what was meant by ‘hobby’. We cannot be sure that ‘hobbies’ referred to the same activities earlier and later.
- Learning violin at 11 and playing in a punk band at 19 are not analogous activities, but both are ‘hobbies’ here.
- The data were not collected by the authors. As a secondary data analysis, the original data collection was not designed to test the specific association of interest.
The problems due to a priori beliefs are separate but common. The paper repeatedly assumes that there is a causal relationship between hobby participation and substance misuse. There is reference to developing a preventative intervention to promote hobby participation. There is a statement at the very beginning of the abstract “…evidence suggests doing hobbies can reduce substance use”. This contrasts with the statement in the conclusions “Although it remains unclear whether hobby engagement causally influences substance use…”. It is common for otherwise sound social research to be blighted by simplistic assumptions regarding causality. If interventions fail because of false causal assumptions, there is an opportunity cost and disillusionment can result. Peter Huxley and I have written about this with respect to social prescribing, which may be intrinsically worthwhile, but cannot possibly address the social determinants of health, as is frequently claimed.
This paper repeatedly assumes a causal relationship between hobby participation and substance misuse, which can lead to an opportunity cost and disillusion with resulting interventions.
Implications for practice
This paper contains much that is of interest, and there are certainly implications for any similar future research in the need for a proper developmental perspective. What is wrong with this paper is the assumption of simple causality between hobby engagement and substance misuse, either positive or negative. Such a causal relationship may appear self-evident, but it is not. There are recurrent statements that hobbies are ‘protective’, re-enforcing that they might be a target for intervention because of a putative causal relationship. In this paper, there are 17 statements that seem to me to support causality; just one statement in the conclusion warning about assuming causality, and two further statements in the body of the paper that are more cautious.
Inferring causality from epidemiological findings has been controversial for many years and led Bradford Hill to develop his criteria for causation 60 years ago; they were not designed for such complex phenomena as are explored here. More recently, directed acyclic graphs have been developed to explore causal relationships. Drawing up these graphs attempts to ensure that the full range of causal relationships have been considered. They allow factors such as collider bias, similar to confounding, to be taken into account. Social capital and social networks should be explored, which are plausible factors here, but not controlled for.
Consider the offspring of two inner city male bus drivers. One family are church goers with educational aspirations that the parents have not achieved. The other family contains a mentally ill mother, with a father preoccupied in caring for her. Consequently the adolescents in the latter family have little parental engagement or guidance. In this study, the families would look the same, and hobbies of the adolescents of the former family look protective, but in fact this might be an epiphenomenon of more plausible factors that have not been measured. I am not suggesting that this is necessarily the case, but that it must be considered. In a similar vein, we know that confounders were adjusted for, but we do not know what impact this had on odds ratios. It is possible that the adjusted supposedly causal effect is dwarfed by a much larger effect due to a confounding factor, which is therefore the signal, not just the noise.
It is understandable that researchers want to move quickly to intervene, but there is a belief that these are lifestyle choices rather than consequences of social structural factors. Put bluntly, do adolescents really decide not to smoke cannabis because they would miss their violin lessons? I do not think so.
In conclusion, this paper has much to commend it, but, as is commonly the case, it has got ahead of itself on potential intervention. Furthermore, a cautious statement at the end does not negate the assumptions of the other 12 pages of the paper.
Whilst social prescription is intrinsically worthwhile, it cannot possibly address the social determinants of health, as is frequently claimed.
Statement of interest
I am a researcher with an interest in structural social factors. I have written on this subject. I hold grants and am involved in research on the subject of substance misuse and social factors in mental health. I have no other personal or financial competing interests to declare.
Links
Primary paper
Bone, J. K., Fancourt, D., Sonke, J. K., & Bu, F. (2025). The changing relationship between hobby engagement and substance use in young people: latent growth modelling of the Add Health Cohort. Journal of Youth and Adolescence, 54(1), 133-145.
Other references
Harris, K. M., Halpern, C. T., Whitsel, E. A., Hussey, J. M., Killeya-Jones, L. A., Tabor, J., & Dean, S. C. (2019). Cohort profile: The national longitudinal study of adolescent to adult health (add health). International journal of epidemiology, 48(5), 1415-1415k.
Poole, R., & Huxley, P. (2024). Social prescribing: an inadequate response to the degradation of social care in mental health. BJPsych Bulletin, 48(1), 30-33.
