Is reading for pleasure in adolescence good for mental health? A counterfactual and within-person analysis in a large longitudinal study

doi:10.21203/rs.3.rs-2636095/v1

Download PDF

Research Article

Is reading for pleasure in adolescence good for mental health? A counterfactual and within-person analysis in a large longitudinal study

https://doi.org/10.21203/rs.3.rs-2636095/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Purpose: Reading has been proposed to be a protective factor in mental health; however, testing this is made challenging by the vulnerability of the reading-mental health association to confounding.

Methods: We used two complementary approaches to address this: propensity score matching and random intercepts cross-lagged panel models (RI-CLPM) in a large longitudinal study: the Zurich Project on Social Development from Childhood to Adulthood (z-proso). For the counterfactual analyses, mental health outcomes of anxiety, depression, and psychosis-like symptoms were measured at ages 17 and 20 and reading engagement was measured at ages 15 and 17. Matching variables included a wide range of factors potentially related to reading and mental health outcomes.

Results: There was no consistent effect of reading for pleasure on anxiety, depression, and psychosis. The only significant effects were detrimental effects of reading on anxiety and psychosis at age 20 when using non-bi-partite matching.

Conclusion: Reading for pleasure without considering content and style of engagement may not be a protective factor in adolescent mental health.

reading

mental health

counterfactual analysis

anxiety

depression

psychosis

Engagement with reading has been proposed to act as a protective factor in mental health and forms the basis of a number of ‘bibliotherapeutic’ approaches, and other reading-based interventions [1–3]. Interest in reading-based interventions also stems from the fact that they are likely to be highly scale-able and may carry less stigma than more clinical approaches, thus overcoming two primary identified barriers to young people seeking and receiving mental health support. As an enjoyable activity, interventions based on reading for pleasure would also have the potential to address the poor compliance that is often observed with self-help mental health interventions [4].

However, a major challenge is that neither reading nor mental health is randomly distributed in the population [5]. This makes any associations identified in observational data vulnerable to confounding; yet experimental designs to examine the effects of reading on mental health are logistically challenging and expensive, and correspondingly rare. A small number of studies have sought to address this challenge using counterfactual analysis approaches, such as propensity score matching [5–7]. Counterfactual analysis approaches are based on ‘re-constructing’ a situation akin to a randomised controlled trial using observational data, through the construction of counterfactuals. For example, in propensity score matching, individuals who are similar in their ‘propensity’ for experiencing a treatment but differing in their actual experience of that treatment are compared on the outcomes of interest.

Using this approach in data from the Millennium Cohort Study, [5] found that reading for pleasure at age 11 was associated with higher self-esteem. In later analyses following a similar approach in the same dataset, these authors found that reading for pleasure at age 7 was associated with lower levels of attention deficit hyperactivity disorder (ADHD) symptoms and emotional problems at age 11 and higher levels of prosociality as measured by the Strengths and Difficulties Questionnaire [6]. However, there was no effect on peer problems or conduct problems.

These findings indicate the promise of reading for pleasure as a protective factor in mental health; however, there remains a need to replicate these findings in other datasets. Further, since reading engagement is not a single event but may show change over developmental time, complementary approaches that leverage within-person change in reading engagement over time and examine corresponding changes in mental health are helpful for illuminating this association. For example, approaches such as the random intercepts cross-lagged panel model (RI-CLPM) can partial out between-person time-stable confounds and provide information about how within-person variation in a candidate exposure variable is reciprocally related to within-person changes in an outcome variable [8]. Utilising these complementary approaches to addressing potential confounding also helps highlight potential unmeasured confounding. For example, propensity score matching aims to comprehensively measure and model between-person confounds but may miss some such confounds, while RI-CLPM relies on exploiting within-person comparisons over time but can be vulnerable to unmeasured time-varying confounds.

Past research has also left it unclear whether reading for pleasure remains a protective factor in and across adolescence. Adolescence is a particularly critical period for mental health, being a time of peak vulnerability for the onset and escalation of mental health issues [9, 10]. Mental health issues at this stage of development often persist into adulthood where they can lead to lifelong distress, impairment and disadvantage [11–13]. Conversely, establishing healthy mental health-related behaviours and habits in adolescence can help minimise lifelong risk. Given this, the goal of the current study was to establish whether reading engagement in adolescence can support better mental health in the transition to adulthood.

Participants

Participants were from the longitudinal Zurich Project on Social Development from Childhood to Adulthood (z-proso) study[14]. Z-proso began in 2004 when the participants were aged 7 and entering primary school. For the present analyses we focus on the most recent waves of data when participants were aged 15, 17, and 20. Information on the sociodemographic composition of the sample and the Ns available for each variable (smallest N = 762) are provided in Table S1 of Supplementary Materials. The sample included approximately equal numbers of males and females and from a wide range of sociodemographic backgrounds, reflecting the diversity of the underlying same-aged population of Zurich, Switzerland. Previous research has suggested that the sample achieved good representativeness of the underlying population and has been subject to little non-random attrition over time [15].

Relevant data were available for two sets of waves for the counterfactual analysis. Specifically, for age 17 anxiety and depression we had treatment (reading engagement) data at age 15 and covariate data at age 13. In addition, for age 20 anxiety, depression, and psychosis-like symptoms we had treatment (reading engagement) data at age 17 and covariate data at age 15. We, therefore, conducted two sets of analyses using data from the age 13–17 waves and age 15–20 waves.

For the RI-CLPM analysis, data were available for 4 waves (age 13, 15, 17, 20] for the anxiety and depression outcomes; however, psychosis-like symptoms were available at only age 20. As such, only the effects of reading on depression and anxiety could be explored using this approach.

Measures

Outcomes

Anxiety and depression were measured using the self-reported Social Behaviour Questionnaire [16] administered at the age 17 and 20 wave of z-proso. At age 20, four items captured anxiety (e.g., fearfulness) and nine items captured depression (e.g., unhappiness, lethargy). Responses were recorded on a five-point Likert-type scale from never to very often.

Slightly different sets of items were available at different ages for these outcomes.. At ages 11–17, four items each measured anxiety and depression. Previous psychometric analyses have supported the reliability and validity of the scores from the SBQ in the current sample [17–20]. Omega reliability for anxiety and depression at age 11 was .69 and .65 respectively; at age 13 was .72 and.74; at age 15 was .74 and .74; at age 17 was .77 and .77 and at age 20 was .80 and .90. Psychosis-like symptoms were measured using an abbreviated six-item version of the Community Assessment of Psychic Experiences Scale [21] at age 20 only. Psychotic-like experiences were measured with six items (e.g., hearing voices). Omega reliability for this scale was .73.

Treatment Variable

Reading engagement was measured using a single-item self-report of reading frequency at the ages 13, 15 and 17 waves of z-proso and asks participants how often they ‘read a book or magazine’ on a five-point scale from never to (almost) every day between Monday and Friday.

For the purposes of the counterfactual analyses, we dichotomised the item such that those who answered ‘never’ or ‘less than once a week’ were designated the control group and those who answered ‘on 1 to 2 days’, ‘on 3 to 4 days’, or ‘(almost) every day’ were designated the treated group. As a sensitivity analysis, we also implemented non-bipartite matching which treats the treatment variable as ordinal

Counterfactual Analysis Matching Variables

Matching variables were selected based on plausibly having an association with the treatment and outcome and included sociodemographic, mental health and wellbeing markers, prosocial/antisocial traits, social risk/protective factors, and school-specific risk/protective factors. The measures are used for matching are comprehensively described in Supplementary Materials and included gender, migration background, socioeconomic status, internalising problems, ADHD symptoms, general trust, self-control, substance use, aggression, prosociality, delinquency, parental involvement, social support, bullying victimisation, teacher bond, class bond, school difficulties, and school achievement in German, maths, and school motivation. All covariates used for the age 20 outcome analyses were measured prior to age 17 (at the age 15 wave or earlier) and all covariates used for the age 17 outcomes were measured prior to age 13 except ISEI and migration status, which were based on data from both age 13 and 15. School achievement variables were measured at age 12 because once youth transition to secondary school, achievement ratings are only valid when comparing youth within the same stream.

Statistical Procedure

R code is provided at: https://osf.io/f23jb

Missingness

Unless data are missing completely at random (MCAR), treatment effects estimated using propensity scores to match or weight the data will be biased and less efficient than alternative methods that attempt to model the missingness [22, 23]. We, therefore, where possible implemented methods that can provide unbiased parameter estimates under a missing at random assumption (MAR) (see Bottigliengo et al., 2021]. Specifically, for our matching-based analyses with our dichotomised treatment variable, we used multiple imputation using a ‘within’ approach that conducts the propensity score analysis within each dataset and pools the results using Rubin’s rules. For this we used a Bayesian multivariate chained equations approach (a fully conditional specification approach) implemented in the mice package in R statistical software [25], with 10 imputations. All covariates and outcome variables were included in the imputation model [26]; however, we focused on the cases with complete treatment data given that missingness in these variables was fairly minimal. All variables in the imputation model were predicted from all other covariates measured at all waves using predictive mean matching. A separate imputation model was built for each set of outcome analyses (age 17 versus 20).

We used a single imputation approach for the non-bipartite matching-based analyses as multiple imputation is not straight-forward to implement for non-bipartite matching. For our RI-CLPM analysis we used full information maximum likelihood estimation (FIML) to deal with missingness. This also provides unbiased parameter estimates under the assumption of MAR.

Group Comparisons

We began by conducting independent samples t-tests to compare those with and without frequent reading engagement. We also examined group differences using nearest neighbour propensity score matching based on a logistic regression propensity model. Given that we had relatively even numbers of participants in each group, we used 1:1 matching, implemented using the MatchIt package in R statistical software [27].

If matching could be achieved (defined by standardised mean differences between the matched treated and control groups of <|.15| for the covariates), we used a separate linear regression for each outcome: anxiety, depression, and psychosis. There has been some debate regarding whether paired or ‘matched’ analyses are optimal given that the dependency in the data is induced by the propensity model and this does not necessarily imply that outcomes will be highly correlated within pairs [28]. Indeed, this approach can lead to inflated type 2 error rates and alternative approaches such as including the matching variables in the main analysis model can lead to better type 2 error rates [28]. We therefore implemented the approach suggested by [28] in which the mental health outcomes were regressed on the treatment and all matching variables in the matched sample. This also has the advantage of helping to remove residual confounding related to the measured variables after matching. We implemented two versions of these analyses: in which prior reading engagement was versus was not included in the set of matching variables.

We also implemented an optimal non-bipartite matching version of the analysis where the treatment variable is modelled as ordinal [29]. In this method, participants with different levels of the treatment variable but similar propensity scores can be matched. Further, in optimal non-bipartite matching, rather than using a greedy matching approach that can result in poor matches for later-matched pairs, an algorithm based on Derigs shortest augmentation path algorithm [30] is used. To implement the non-bipartite matching method, we first estimated treatment propensity scores for the participants using an ordinal logistic regression model. We then used these to derive the following distance measure:

$$\varDelta \left({x}_{i},{x}_{j}\right)= \frac{{\left({\widehat{\beta }}^{T}{x}_{i}-{\widehat{\beta }}^{T}{x}_{j}\right)}^{2}+ϵ}{{\left({Z}_{i}-{Z}_{j}\right)}^{2}}$$

(1)

Where ${\widehat{\beta }}^{T}x$ is the propensity score, $Z$ is the treatment dose level, and $ϵ$is a vanishingly small but strictly positive number. This disallows matches between individuals with the same treatment dose. To control the amount of residual confounding, matches were also disallowed between pairs differing in distance by more than 0.15SD units by setting the relevant entries in the distance matrix to very large numbers. We then implemented optimal non-bipartite matching based on the distance matrix constructed using this measure using the R package nbpMatching [31].

RI-CLPM

Random-intercept cross-lagged panel models (RI-CLPMs)were fit to the longitudinal reading engagement and mental health data. The RI-CLPM can be thought of as an extension of the standard CLPM, which models the lag-1 autoregressive and cross-lagged relations between constructs over time. However, it has been noted that the standard CLPM conflates between-person and within-person effects, where it is the latter that is typically of interest in developmental theory and for informing interventions [8]. The RI-CLPM was proposed as a way to disaggregate these between- and within-person effects. It does so by specifying random intercept factors for each variable (person random effects representing individual differences in the overall levels of each variable) and fitting the cross-lagged structure to the resulting time-specific residuals. The random intercept factors are allowed to covary; they estimate the between-person association between constructs (e.g., the extent to which people with overall higher levels of reading engagement tend to have lower overall levels of mental health issues). The time-specific residuals after modelling these random intercepts represent individuals’ deviations from their baseline level of a construct and can be used to test whether being higher or lower than usual on one construct is associated with escalations or declines in another. For example, it can be used to test whether reading more at a given time is associated with better mental health concurrently and at the next wave. This represents the within-person cross-lagged effects of reading on mental health. All lag-1 autoregressive and cross-lagged effects between the reading and relevant mental health variables were modelled, as were the concurrent (residual) covariances between reading and the relevant mental health variable at each time point. Models were fit using the lavaan package in R statistical software with maximum likelihood estimation and separate models fit for each mental health outcome.

Descriptive statistics

Descriptive statistics for the reading engagement and mental health outcomes are provided in Table 1. At age 11, n = 253 read less than once a week and n = 884 read more once a week or more; at age 13, n = 659 read less than once a week and n = 704 read once a week or more; at age 15, 922 read less than once a week and n = 522 read once a week or more; and at age 17 n = 833 read less than once a week and n = 418 read once a week or more. Descriptive statistics for the matching variables are provided in Tables S1 (and S2 of Supplementary Materials).

Table 1

Descriptive statistics for mental health variables
	N	Mean	Min	Max
Age 11 Anxiety	1126	8.18	4	20
Age 13 Anxiety	1350	8.59	4	20
Age 15 Anxiety	1436	9.17	4	20
Age 17 Anxiety	1295	9.5	4	20
Age 20 Anxiety	1180	9.61	4	20
Age 11 Depression	1129	8.09	4	19
Age 13 Depression	1354	8.88	4	20
Age 15 Depression	1442	9.49	4	20
Age 17 Depression	1299	9.8	4	20
Age 20 Depression	1179	20.31	9	45
Age 20 Psychosis	1177	8.79	6	30
Note. The possible score range for Age 11–20 Anxiety and Age 11–17 Depression is 4–20; For Age 20 Depression it is 4–45 and for Age 20 Psychosis it is 4–30.

T-tests

T-tests comparing those reading more versus less than once a week at the prior wave suggested no effect of reading on age 13 anxiety.

[mean treated = 8.70, mean control = 8.37, t(315.18)=-1.32, p = .19]; however, who engaged in more prior reading were higher in anxiety at age 15 [mean treated = 9.44, mean control = 8.88; t(1305.1)=-2.99, p = .003] and at age 17 [mean treated = 9.30, mean control = 9.81; t (1023) = -2.46, p = .014].

There were no significant differences on age 13 depression symptoms by prior reading engagement group [mean treated = 8.94, mean control = 8.93; t (323.99), p = .97], nor on age 15 depression symptoms [mean treated = 9.64, mean control = 9.38; t (1318.9) = -1.46, p = .14], age 17 depression symptoms [mean treated = 10.00, mean control = 9.67, t(1000.8)= -1.69, p = .09], nor age 20 depression symptoms [mean treated = 20.44, mean control = 20.24; t(752.25)=-0.418, p = .68].

There were no significant differences in psychosis scores based on prior reading engagement [mean treated = 8.72, mean control = 8.800; t(896.82) = 0.41, p = .68].

Matching-based Analyses

Age 17 outcomes analysis

Nearest neighbour matching with treatment

Propensity score matching with nearest neighbour matching and a logistic regression propensity model resulted in an average of 522 matched treated and control units across the imputed datasets. Most covariates had standardised mean differences (SMD) of <|.15| across the treated vs control groups, the exceptions being: ISEI (SMD = .19), parental involvement (SMD = .17), and school achievement in the domain of motivation (SMD = .16). Full match statistics are provided in Table S3 of Supplementary Materials.

Using these matches there was no significant effect of reading engagement on age 17 anxiety, without (b = .15, p = .50) or with adjustment for the matching variables (b = .06, p = .72). There was also no effect on age 17 depression either without (b = .19, p = .39) or with adjustment for the matching variables (b = .13, p = .49).

Repeating the procedure including prior reading engagement in the set of matching variables resulted again in the successful matching of 522 treated and control units though there were several covariates with SMDs > |.15|: ISEI (SMD = .26), parental involvement (SMD = .24), social support from adults (SMD = .18), school achievement in maths (SMD = .16), school achievement in German (SMD = .21), school achievement in motivation (SMD = .18), and prior reading (SMD = .17). Full match statistics are provided in Table S4 of Supplementary Materials.

Using these matches, there was again no significant effect of reading engagement on age 17 anxiety (b = .12, p = .60 without adjustment for matching variables; b = .04, p = .90 with adjustment for matching variables) nor depression (b = .15, p = .51 without adjustment for matching variables; b = .07, p = .73 with adjustment for matching variables).

Non-bipartite Matching

The non-bipartite optimal matching procedure resulted in 510 matched units. Full matching statistics are provided in Supplementary Materials Table S5. There were three covariates with SMDs >|.15|: low self-control (SMD=-.17), physical aggression (SMD=-.21), and delinquency (SMD=-.15), with all others showing SMDs <|.15| between the treated and control groups.

Using these matches, there was a significant effect of reading engagement on age 17 anxiety without adjustment (b=-.42, p = .03); however, this was no longer significant with adjustment for the matching variables (b=-.23, p = .16). There was no significant effect of reading engagement on age 17 depression either prior to (b=-.30, p = .11) or after (b = − .22, p = .21) adjusting for matching variables as covariates.

Age 20 Outcomes Analysis

Nearest neighbour matching with treatment

Matching using nearest neighbour matching resulted in an average of 418 cases successfully matched across the imputed datasets. Match statistics are provided in Supplementary Materials Table S6 and suggested that the average standardised mean differences in covariates across group was <|.15| for all covariates, except low self-control (SMD=-0.163).

Using these matches there was no significant effect of reading engagement on age 20 anxiety either before (b = .14, p = .61) or after (b = .07, p = .75) inclusion of the matching variables as covariates in the linear regression to test reading engagement effects. There was also no significant effect of reading engagement on depression before (b = .29, p = .58) or after (b = .19, p = .68) adjustment for the matching variables as regression covariates. Finally, there was no significant effect of reading engagement on psychosis either before (b = .27, p = .24) or after (b = .31, p = .14) regression adjustment for the matching variables.

Repeating the procedure additionally matching on prior reading engagement led to an average of 418 matches across the imputed datasets. Full matching statistics are provided in Table S7. Three covariates had SMD>|.15|: ISEI (SMD = − .18), low self-control (SMD = − .20), and age 15 reading engagement (SMD = .25).

Using these matches, there was no significant effect of reading engagement on age 20 anxiety either before (b = .06, p = .82) or after (b = .04, p = .86) additionally adjusting for the matching variables as regression covariates. There was also no effect on age 20 depression before (b = .19, p = .73) or after (b = .16, p = .76) and no effect on age 20 psychosis before (b = .17, p = .48) or after (b = .26, p = .26) additionally adjusting for the matching variables as regression covariates.

Non-bipartite Matching

Non-bipartite matching for the age 20 outcome analyses led to 408 matches (See Supplementary Materials Table S8 for balance statistics). All matching variables had SMDs <|.15|. Using these matches, there was a significant effect of reading engagement on age 20 anxiety prior to regression adjustment for the matching variables (b=-.59, p = .006) and after covariate adjustment (b=-.58, p = .002). There was no significant effect of reading engagement on age 20 depression before (b=-.61, p = .15) or after (b=-.68, p = .07) covariate adjustment. There was a significant effect of reading engagement on age 20 psychosis before (b=-.53, p = .004) and this effect increased in magnitude following regression adjustment for the matching variables as covariates (b=-.60, p = .0008).

RI-CLPM

The RI-CLPM for reading and depression fit well (CFI = .992, TLI = .974, RMSEA = .037, SRMR = .019). Full results are provided at: https://osf.io/qu93e. There were no significant within-person effects of reading on depression, nor any significant concurrent (same-wave) within-person associations. There was, however, a significant covariance between the reading and depression person random effects, representing their between-person association (r = .24 p = .002).

The RI-CLPM for reading and anxiety also fit well (CFI = .991, TLI = .971, RMSEA = .041, SRMR = .022). Full results are provided at: https://osf.io/ahyfb. There were no significant within-person effects of reading on anxiety, nor any significant concurrent (same-wave) within-person associations. There was a significant covariance between the reading and anxiety person random effects (r = .31 p < .001).

Comparison Of Approaches

Table 2 summarises the findings across the approaches.

Table 2

Summary of findings on the effect of prior reading engagement on mental health outcomes across approaches
	T-test	Nearest neighbour matching	Non-bipartite optimal matching	RI-CLPM
Age 13 anxiety	No effect	Not tested	Not tested	No effect
Age 15 anxiety	Detrimental effect	Not tested	Not tested	No effect
Age 17 anxiety	Detrimental effect	No effect	No effect	No effect
Age 20 anxiety	Detrimental effect	No effect	Detrimental effect	No effect
Age 13 depression	No effect	Not tested	Not tested	No effect
Age 15 depression	No effect	Not tested	Not tested	No effect
Age 17 depression	No effect	No effect	No effect	No effect
Age 20 depression	No effect	No effect	No effect	No effect
Age 20 psychosis	No effect	No effect	Detrimental effect	Not tested
Note. The summary of nearest neighbour and non-bipartite optimal matching effects are based on the analyses additionally adjusted for the matching variables as covariates.

In this study we examined whether reading for pleasure in adolescence is a protective factor in mental health. Using several complementary approaches, we found no evidence for an effect of reading engagement on adolescent depression and inconsistent evidence for a detrimental effect of reading engagement on anxiety and psychosis. Specifically, reading for pleasure at age 17 had inconsistent effects on age 20 mental health across different counterfactual analysis approaches: based on nearest neighbour matching on a dichotomised reading variable, no significant effects on anxiety, depression, and psychosis, but significant detrimental effects on anxiety and psychosis based on optimal non-bipartite analyses. In RI-CLPM analysis, there were no within-person cross-lagged effects of reading engagement on either depression or anxiety across ages 11, 13, 15, and 17. Taken together, results suggest that for this age range, general reading for pleasure - without consideration of the content or nature of engagement – is not a promising target for intervention.

For the few significant effects observed, in contrast to our hypothesis, our findings pointed to a potential detrimental effect of reading engagement on anxiety and psychosis in later adolescence. While these findings should be interpreted with caution due to their reliance on approaches that can produce incorrect standard errors (optimal non-bipartite matching with single imputation) and biased treatment effects (raw independent samples t-tests on unmatched data), they are potentially explainable in terms of current understanding of adolescent development. In particular, adolescence is a time in which the development of peer relationships and associated skills become particularly important [32, 33] and spending large amounts of time engaging in reading could represent displacement from engaging in these developmental tasks. Similarly, spending large amounts of time reading may not itself be a causal factor but could represent a marker of social exclusion or lack of social motivation that could be an early sign for later-emerging anxiety or depression.

As noted, analysing the impact of reading for pleasure on mental health must take into account confounding due to the social patterning of reading for pleasure [5]. For example, factors such as language ability, attention difficulties, socioeconomic background, and pre-existing mental health symptoms may impact motivation and/or ability to engage with reading and subsequent mental health. Counterfactual methods are valuable for providing a solution to this issue when only observational data are available and can help inform which protective factors are worth further exploring in interventions. Approaches such as RI-CLPM provide a useful complementary approach for treatments that show variation over time because they can leverage these within-person changes to illuminate whether individuals’ changes in their reading habits relative to their typical habits are associated with changes in their mental health[8, 34, 35].

Previous studies have applied methods that seek to address confounding and have identified an effect of reading for pleasure on mental health using counterfactual approaches. However, these have been focused on younger age groups, up to early adolescence [5, 6]. Taken together with our findings, this suggests that reading for pleasure could be effective in supporting mental health up to and into the transition to adolescence. Indeed, the early adolescence stage of development has been identified as a particularly critical period of development characterised by rapid, marked, and diverse social, emotional, and biological changes that young people must navigate [36]. Unsurprisingly, it is also a peak period of vulnerability for the onset of new mental health disorders and escalation of symptoms where there are pre-existing difficulties. As such, it may be a period where there is also greater sensitivity to potential protective factors such as reading for pleasure.

It is important to note the limitations of the current study. First, we had a long time lag between the treatment and outcomes which meant that we could only detect effects that persisted over a two year period between ages 11, 13, 15, and 17 and a three year period between age 17 and 20. These may not be the optimal times lag for detecting reading effects [37] and future research examining the impact of reading on mental health with shorter temporal delays could help clarify whether reading for pleasure has a short- term impact on mental health. Second, counterfactual methods rely on the assumption that all important confounding effects are captured directly, or indirectly via their correlations with the measured covariates. It is, therefore, possible that relevant confounding effects were not included in our models. However, given the non-significant treatment effects and the rich set of relevant potential confounding factors modelled, we think it is unlikely that the inclusion of additional covariates would substantially change our conclusions. Similarly, the RI-CLPM is vulnerable to the effects of within-person time-varying confounds. Finally, our measure of reading was coarse.and we were unable to examine the effects of content, style of engagement, motivation, or other aspects of how adolescents engage with reading on mental health.

When using a counterfactual approach to address potential confounding, reading for pleasure at ages 15 and 17 were not consistently positively associated with later (ages 17 and 20) mental health. Similarly, there was no evidence of an effect of reading on mental health within an RI-CLPM approach. This suggests promoting general reading for pleasure without consideration of content or style of engagement may not be a promising intervention target for long term effects on mental health in this age group.

Acknowledgements

We are grateful to the research assistants and participants of the z-proso study. Funding from the Swiss National Science Foundation (Grants 405240 − 69025, 100013_116829, 100014_132124, 100014_149979, 100014_149979, 10FI14_170409/1, 10FI14_170409/2, 10FI14_198052/1), the Jacobs Foundation (Grants 2010 − 888, 2013-1081-1), the Jacobs Center for Productive Youth Development, the Swiss Federal Office of Public Health (Grants 2.001391, 8.000665), the Canton of Zurich’s Department of Education, the Swiss Federal Commission on Migration (Grants 03-901 (IMES), E-05-1076), the Julius Baer Foundation, and the Visana Foundation is gratefully acknowledged.

Preprint DOI: 10.31234/osf.io/dsbec

Conflicts of interest

The authors have declared that they have no competing or potential conflicts of interest.

Arslan G, Yıldırım M, Zangeneh M, Ak İ.(2022) Benefits of Positive Psychology-Based Story Reading on Adolescent Mental Health and Well-Being. Child indicators research. 2022;1–13.
Carney J, Robertson C.(2022) Five studies evaluating the impact on mental health and mood of recalling, reading, and discussing fiction. Plos one. 2022;17(4):e0266323.
Dowrick C, Billington J, Robinson J, Hamer A, Williams C.(2012) Get into Reading as an intervention for common mental health problems: exploring catalysts for change. Medical Humanities. 2012;38(1):15–20.
Baumel A, Muench F, Edan S, Kane JM.(2019) Objective user engagement with mental health apps: systematic search and panel-based usage analysis. Journal of medical Internet research. 2019;21(9):e14567.
Mak HW, Fancourt D.(2019) Arts engagement and self-esteem in children: results from a propensity score matching analysis. Annals of the New York Academy of Sciences. 2019;1449(1):36–45.
Mak HW, Fancourt D.(2020) Longitudinal associations between reading for pleasure and child maladjustment: Results from a propensity score matching analysis. Social Science & Medicine. 2020;253:112971.
Mak HW, Fancourt D.(2020) Reading for pleasure in childhood and adolescent healthy behaviours: Longitudinal associations using the Millennium Cohort Study. Preventive medicine. 2020;130:105889.
Hamaker EL, Kuiper RM, Grasman RP.(2015) A critique of the cross-lagged panel model. Psychological methods. 2015;20(1):102.
Kessler RC, Berglund P, Demler O, Jin R, Merikangas KR, Walters EE.(2005) Lifetime prevalence and age-of-onset distributions of DSM-IV disorders in the National Comorbidity Survey Replication. Archives of general psychiatry. 2005;62(6):593–602.
Murray AL, Ushakova A, Speyer L, Brown R, Auyeung B, Zhu X.(2021) Sex/gender differences in individual and joint trajectories of common mental health symptoms in early to middle adolescence. JCPP Advances. 2021;e12057.
Fergusson DM, Boden JM, Horwood LJ.(2007) Recurrence of major depression in adolescence and early adulthood, and later mental health, educational and economic outcomes. The British Journal of Psychiatry. 2007;191(4):335–42.
Gargano LM, Locke S, Li J, Farfel MR.(2018) Behavior problems in adolescence and subsequent mental health in early adulthood: results from the World Trade Center Health Registry Cohort. Pediatric research. 2018;84(2):205–9.
Johnson JG, Cohen P, Kasen S.(2009) Minor depression during adolescence and mental health outcomes during adulthood. The British Journal of Psychiatry. 2009;195(3):264–5.
Ribeaud D, Murray A, Shanahan L, Shanahan MJ, Eisner M.(2022) Cohort Profile: The Zurich Project on the Social Development from Childhood to Adulthood (z-proso). Journal of Developmental and Life-Course Criminology [Internet]. 2022 Feb 21; Available from: https://doi.org/10.1007/s40865-022-00195-x
Eisner NL, Murray AL, Eisner M, Ribeaud D.(2018) A practical guide to the analysis of non-response and attrition in longitudinal research using a real data example. International Journal of Behavioral Development. 2018;0165025418797004.
Tremblay RE, Loeber R, Gagnon C, Charlebois P, Larivee S, LeBlanc M.(1991) Disruptive boys with stable and unstable high fighting behavior patterns during junior elementary school. Journal of abnormal child psychology. 1991;19(3):285–300.
Murray AL, Eisner M, Obsuth I, Ribeaud D.(2017) Situating violent ideations within the landscape of mental health: Associations between violent ideations and dimensions of mental health. Psychiatry research. 2017;249:70–7.
Murray AL, Eisner M, Ribeaud D.(2017) Can the Social Behavior Questionnaire help meet the need for dimensional, transdiagnostic measures of childhood and adolescent psychopathology? European Journal of Psychological Assessment. 2017;
Murray AL, Booth T, Eisner M, Ribeaud D, McKenzie K, Murray G.(2019) An Analysis of Response Shifts in Teacher Reports Associated with the Use of a Universal School-Based Intervention to Reduce Externalising Behaviour. Prevention Science. 2019;1–9.
Murray AL, Obsuth I, Eisner M, Ribeaud D.(2017) Evaluating longitudinal invariance in dimensions of mental health across adolescence: An analysis of the Social Behavior Questionnaire. Assessment. 2017;1073191117721741.
Mark W, Toulopoulou T.(2016) Psychometric properties of “community assessment of psychic experiences”: review and meta-analyses. Schizophrenia Bulletin. 2016;42(1):34–44.
Leyrat C, Seaman SR, White IR, Douglas I, Smeeth L, Kim J, et al.(2019) Propensity score analysis with partially observed covariates: How should multiple imputation be used? Statistical methods in medical research. 2019;28(1):3–19.
Rubin DB.(1976) Inference and missing data. Biometrika. 1976;63(3):581–92.
Bottigliengo D, Lorenzoni G, Ocagli H, Martinato M, Berchialla P, Gregori D.(2021) Propensity Score Analysis with Partially Observed Baseline Covariates: A Practical Comparison of Methods for Handling Missing Data. International Journal of Environmental Research and Public Health. 2021;18(13):6694.
Buuren S van, Groothuis-Oudshoorn K.(2010) mice: Multivariate imputation by chained equations in R. Journal of statistical software. 2010;1–68.
Crowe BJ, Lipkovich IA, Wang O.(2010) Comparison of several imputation methods for missing baseline data in propensity scores analysis of binary outcome. Pharmaceutical statistics. 2010;9(4):269–79.
Stuart EA, King G, Imai K, Ho D.(2019) MatchIt: nonparametric preprocessing for parametric causal inference. Journal of statistical software. 2011;
Wan F. Matched or unmatched analyses with propensity-score–matched data? Statistics in medicine. 2019;38(2):289–300.
Lu B, Greevy R, Xu X, Beck C.(2011) Optimal nonbipartite matching and its statistical applications. The American Statistician. 2011;65(1):21–30.
Derigs U.(1988) Solving non-bipartite matching problems via shortest path techniques. Annals of Operations Research. 1988;13(1):225–61.
Beck C, Lu B, Greevy R, Beck MC.(2014) Package ‘nbpMatching.’ CRAN; 2016.
Blakemore SJ, Mills KL. Is adolescence a sensitive period for sociocultural processing? Annual review of psychology. 2014;65:187–207.
Steinberg L, Monahan KC.(2007) Age differences in resistance to peer influence. Developmental psychology. 2007;43(6):1531.
Berry D, Willoughby MT.(2017) On the practical interpretability of cross-lagged panel models: Rethinking a developmental workhorse. Child Development. 2017;88(4):1186–206.
Murray AL, Eisner M, Ribeaud D.(2019) Within-person analysis of developmental cascades between externalising and internalising problems. Journal of Child Psychology and Psychiatry. 2019;
Rapee RM, Oar EL, Johnco CJ, Forbes MK, Fardouly J, Magson NR, et al.(2019) Adolescent development and risk for the onset of social-emotional disorders: A review and conceptual model. Behaviour Research and Therapy. 2019;103501.
Dormann C, Griffin MA.(2015) Optimal time lags in panel studies. Psychological methods. 2015;20(4):489.

No competing interests reported.

SupplementaryMaterials.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Is reading for pleasure in adolescence good for mental health? A counterfactual and within-person analysis in a large longitudinal study

Status:

Version 1

Abstract

Introduction

Methods

Participants

Measures

Outcomes

Treatment Variable

Counterfactual Analysis Matching Variables

Statistical Procedure

Missingness

Group Comparisons

RI-CLPM

Results

Descriptive statistics

T-tests

Matching-based Analyses

Age 17 outcomes analysis

Nearest neighbour matching with treatment

Non-bipartite Matching

Age 20 Outcomes Analysis

Nearest neighbour matching with treatment

Non-bipartite Matching

RI-CLPM

Comparison Of Approaches

Discussion

Limitations

Conclusions

Declarations

Acknowledgements

Conflicts of interest

References

Additional Declarations

Supplementary Files

Status:

Version 1