Making connections: statistical methods to investigate health-risk behaviours in adolescents

0
  • Clinical psychology research is increasingly looking to use information from multiple sources to understand the processes behind known relationships. Dr Ginger Lockhart, Assistant Professor of Quantitative Psychology at Utah State University is developing innovative statistical methods to investigate the mechanisms by which adolescents develop health-risk behaviours.

    In the world of clinical research, there is currently a great interest in establishing not only relationships between variables, but also how and why one variable may affect another. A good understanding of such causal processes allows effective prevention methods to be identified and can enhance our understanding of how treatment methods operate. Statistical mediation analysis investigates how an independent variable (X) influences a mediator (M), which in turn affects a dependent outcome variable (Y). This method explores the underlying mechanism behind a known relationship and is employed when there appears to be no definite connection between the independent and dependent variables.

    In clinical psychology research, the use of multiple informant (MI) data is considered best practice to assess variables X, Y and M. An informant can be anyone who reports on a case, for example a patient (self-report), parent, or healthcare worker. Thus, MI data describes the combination of reports from multiple sources. This approach is widely used and deemed best methodological practice as it allows for the collection of different perspectives and the determination of convergent validity (the level of agreement between reporters).

    In her research, Dr Lockhart aims to develop innovative statistical techniques that can be used to address complex problems in prevention and intervention science. Her current work combines MI data with statistical mediation analysis to discover the mechanisms through which adolescents develop health-risk behaviours. Results from Dr Lockhart’s research have important implications for informing prevention programme strategies.

    Uncovering the links
    Dr Lockhart’s research on statistical methods has a wide range of applications. However, the primary focus is how these methods can be used to understand how young people develop health-related habits. In a number of previous studies, Dr Lockhart has applied varying statistical approaches to identify the processes by which adolescents develop a range of health-risk behaviours.

    TV and unhealthy snacks
    In one such study, Dr Lockhart used a single, multilevel multivariate path model to explore connections between adolescents’ personal characteristics, technology use and food cravings. The study found that ethnicity was a significant factor, with TV exposure in non-Hispanic youths associated with cravings for sweet snacks, sweetened drinks and salty snacks more so than for Hispanic youths. Hispanic youths showed stronger associations between phone messaging and sweet snack, sweetened drink and salty snack cravings. Gender was also found to be a significant mediating factor, with stronger associations between video games and salty snack cravings shown for males.

    Alcohol intake
    A separate study investigated how adolescent alcohol use is influenced by the dynamics of social networks. In this study, Dr Lockhart used a network dynamic model (longitudinal stochastic actor-based model) to differentiate between peer selection and social assimilation. Adolescents were found to choose their friendship groups based on common alcohol intake. However, this study also found that adolescents do not clearly assimilate their alcohol intake to that of their friends. This study implies that prevention programmes targeting alcohol use among young people would benefit from a focus on preventing the formation of negative friendships.

    Disordered eating
    In order to examine the relationships between weight stigmatisation and disordered eating, Dr Lockhart tested the presence of two distinct mediated pathways: stress as a mediator between weight stigmatisation and emotional eating; and social withdrawal as a mediator between weight stigmatisation and dietary restraint. The hypothesised pathways were tested with a path analysis followed by a random sampling method, known as bias-corrected bootstrap mediation analysis.

    Dr Lockhart aims to develop innovative statistical techniques that can be used to address complex problems in prevention and intervention scienceQuote_brain

    Results showed that both stress and social withdrawal partially mediated the pathways between weight stigmatisation and emotional eating and dietary restraint, respectively. As weight stigmatisation is a widespread social issue, targeting these person-level mediating factors is a practical way of addressing disordered eating in the study population.

    Risk behaviours
    A further study published by Dr Lockhart earlier this year examined how self-worth mediates between adolescent attachment security and three distinct risk behaviours (physical fights, weapon carrying and substance abuse). A longitudinal study was employed to examine how these variables interact over time in a study group of adolescent impoverished African Americans. Results were obtained using a method of investigation (ordinal logistic path analysis) which simultaneously specified the three mediated pathways linking attachment security to the three risk behaviours. The results confirmed the hypothesis: greater attachment security is associated with future higher self-worth and subsequent lower substance use and weapons carrying later in adolescence, although the findings for violence were inconclusive. The confirmation of these mediated pathways suggests that adult mentorship is an important component for effective prevention programmes.

    Combining methods
    Whilst efforts have previously been made to combine MI data with mediation analysis, this task presents many methodological and practical challenges and previously developed approaches have significant limitations. One commonly used approach involves taking the average of scores from multiple informants and using this average to create a composite score. The mediation analysis is then performed on this new composite score. Although this approach is simple, it does not allow for the determination of convergent validity and it assumes all informant scores should be equally weighted. An alternative approach reports separate mediation analyses for each informant type, providing interesting informant-level details but failing to provide a single comprehensive statistical model. A third commonly used approach takes the observed scores from informants and uses them as indicators of a latent (unobserved) variable. The modelling of mediating effects is carried out on the latent variables. This has the advantage of allowing different informant reports to be given different factor loadings. However, these models are susceptible to overestimations of random error variance and underestimation of the reliability of observed variables, resulting in bias in the model.

    In order to address these limitations, Dr Lockhart has developed an innovative new approach for combining MI into statistical mediation analysis. The new approach is based on a CT-C(M-1) model. Within a CT-C(M-1) approach, each informant type reports on each trait (e.g., X, Y and M for mediation studies). Multiple observed variables are used as latent variable indicators for each trait and each informant type. When MI data is combined with this approach, one informant serves as the reference informant. Non-reference reports are contrasted against reference reports, enabling researchers to examine how well informants’ reports converge to the ‘gold standard’ reference report.

    Dr Lockhart’s new approach expands on the CT-C(M-1) model by combining it with mediation models that are already commonly used in path analysis. This latent variable approach allows for the correction of random error. Furthermore, it allows trait effects, informant effects and measurement error to be separated, facilitating the quantification of convergent validity, informant specificity and informant reliability. Each informant type and trait has multiple indicators allowing the study of whether discrepancies between informants are shared across different informants or generalised for different traits. Another advantage of this method is that differences in wording or content that result in method effects can be properly accounted for due to the use of variable-specific trait factors. In addition, the model provides the flexibility to be used in situations where either all or only some constructs (X, Y and M) are assessed with MI.

    Steps to prevention
    Dr Lockhart’s new approach provides a methodology that is appropriate to the analysis of mediating effects in the context of a MI study and will have significant impacts for adolescent health-risk prevention programmes. The successful dissemination of these results will provide crucial knowledge for those working in the field of prevention when designing effective interventions and informing prevention programme strategies.

  • Your research focuses on adolescents, what particularly interests you about this study group?
    This developmental stage fascinates me because of the rapid brain and social changes that are taking place. These changes create a complex web of both strengths and vulnerabilities that challenge us to optimise positive health outcomes as adolescents transition into early adulthood and beyond.

    When assessing how a given variable may affect a certain behaviour, why is it important to consider reports from multiple informants?
    With few exceptions, considering only self-report of a measure such as depression introduces bias, because people (particularly kids), tend to do a rather poor job of reporting on their own psychological and behavioural functioning. Moreover, it gives us the opportunity to discover the extent of overlap vs non-overlap between reporters and across contexts, which can be interesting in and of itself. For example, some kids may show more outwardly obvious signs of depression, and they may therefore show strong agreement in their depression scores as provided by their teachers and parents. Other kids may behave differently at school vs home and show disparate scores between reporters. These differences in reporters’ overlap are clinically meaningful because they may uncover the extent to which psychological problems are consistent across contexts or specific to a situation (e.g., a child shows depressive symptoms at school but not at home).

    Dr Lockhart’s new approach is set to have a significant impact on adolescent health-risk prevention programmesQuote_brain

    You have shown that, for many health-risk behaviours, the cause and the outcome are linked by a mediating variable: M. When conducting a study, how do you choose M?
    The primary reason to perform a mediation analysis in the context of prevention science is to identify variables in a causal sequence that can potentially be acted upon to effect change in an outcome. Mediators must therefore be malleable, meaning that they have the potential to change if targeted in an intervention, and they must be intermediate in that causal sequence, such that changes in a predictor happen first, which then cause changes in the mediator, which in turn cause changes in a health-risk behaviour.

    Why is it effective to focus on the mediating variable in prevention programmes?
    This theory-driven approach to preventing health problems is effective because it is typically not possible to directly change a health behaviour; practitioners are not in a position to physically keep kids from smoking or using drugs when they are living their daily lives. But we can identify malleable variables, that, if changed in a healthy way, can therefore exert changes in their behaviour. For example, we have found that working on kids’ self-worth reduces the likelihood of substance use over time.

    How do you ensure the successful dissemination of your research findings so that they can be employed practically in prevention programmes?
    We publish our findings in applied journals with a wide reach covering both researchers and practitioners and present our work at clinical conferences. Beginning this year, we will begin producing animated video shorts to disseminate through social media to communicate our findings to an even wider audience.

  • Research Objectives
    Dr Ginger Lockhart has been developing innovative statistical methods to investigate the ways in which adolescents develop health-risk behaviours.

    Funding
    National Institute of Drug Abuse and United States Department of Justice

    Collaborators
    Lab Associates:

    • Tyson Barrett
    • Amanda Hagman, MS
    • Morgan Kawamura
    • Emily Long, MEd
    • Melissa Simone, MS

    Collaborators:

    • John Bolland, PhD
    • Laurie Chassin, PhD
    • Rick Cruz, PhD
    • Christian Geiser, PhD
    • Nicholas Ialongo, PhD
    • Michael Levin, PhD
    • David MacKinnon, PhD
    • Greg Madden, PhD
    • Kim Reynolds, PhD
    • Michael Twohig, PhD

    Bio
    Ginger Lockhart is an Assistant Professor of Quantitative Psychology at Utah State University. Her research is driven by two primary goals: firstly, to discover the processes through which young people develop health-related habits and secondly, to develop innovative statistical techniques to answer complex questions in prevention science.

    Contact
    Ginger Lockhart, PhD
    Assistant Professor
    Department of Psychology
    Program in Quantitative Psychology
    Utah State University
    USA

    E: ginger.lockhart@usu.edu
    W: lockhartlab.org

    https://www.researchgate.net/profile/Ginger_Lockhart

  • Making connections: statistical methods to investigate health-risk behaviours in adolescents
Share.

Leave A Reply

%d bloggers like this: