|dc.description.abstract||There is growing evidence for mindfulness-based interventions in alleviating the symptoms and enhancing the coping abilities of people suffering from psychological and physical health conditions and improving overall well-being. In essence, mindfulness is our immediate, instant contact with our internal and external environments that is not contaminated by judgmental attitudes or habitual cognitions, and is associated subjectively with a greater clarity of consciousness. With increased application of mindfulness-based interventions, evaluation of their effectiveness requires more accurate measurement of both mindfulness and associated health-related outcomes. In particular, issues with measurement precision (e.g. ordinal rather than interval scaling), item functioning and the state-trait distinction have not been sufficiently addressed or resolved using appropriate modern statistical methods. Ordinal measures have limited precision, and using them with parametric statistical techniques violates the basic assumptions of these tests. The accurate distinction of state from trait and establishing measurement at an interval level are two essential steps for rigorously validating mindfulness and health outcome measures.
The initial part of this thesis focused on applying Rasch analysis to improve the scaling properties of ordinal mindfulness and outcome measures to interval-level scales suitable for parametric statistics. Four studies improved the psychometric properties of widely used and recently developed mindfulness measures including the Mindful Attention and Awareness Scale (MAAS), the Kentucky Inventory of Mindfulness Skills (KIMS), the Five Facet Mindfulness Questionnaire (FFMQ) and the Comprehensive Inventory of Mindfulness Experiences (CHIME). Three further studies improved the scaling properties of the Functional Assessment Measure UK FIM+FAM, the Oxford Happiness Questionnaire (OHQ), and the Perceived Stress Scale (PSS). These studies all employed Rasch analysis and developed conversion algorithms to transform ordinal responses into interval-level data. The second part of the thesis applied Generalisability Theory for the first time to distinguish quantitatively between state and trait components in a mindfulness measure. This study demonstrated that Generalisability Theory can be successfully applied to accurately distinguish between state and trait components in a psychometric measure, and it is recommended as the most applicable psychometric method to validate state and trait questionnaires in the future. Until now the distinction between state and trait has typically based upon a single correlation between total test scores on two different occasions. Consequently, poor items could ‘hide’ behind the other items undetected by test-retest correlation and may affect the overall performance of a scale. The proposed method estimates the extent to which a scale and every individual item are each measuring a state and a trait. Findings of this study have far-reaching implications to help improve the accuracy of distinction between state and trait in measurement of mindfulness and other areas of psychological assessment. Together, these studies analysed data representing 2,551 participants including community and clinical populations, as well as university students. Overall, this work contributed practical solutions and innovative methods to improve the reliability, validity and scaling properties of psychometric measures with a range of implications for mindfulness and health research practice.||en_NZ