{"id":{"repo_id":"plymouth","oai_identifier":"oai:pearl.plymouth.ac.uk:pms-theses-1059"},"canonical_url":"https://search.dev.ndltd.org/etd/plymouth/oai:pearl.plymouth.ac.uk:pms-theses-1059","repository":{"repo_id":"plymouth","name":"University of Plymouth","base_url":"https://pearl.plymouth.ac.uk/do/oai"},"display":{"title":"Improving the efficiency of modelling complex physical activity data using longitudinal compositional data analysis","abstract":"Physical activity (PA) data are often objectively collected using an accelerometer worn by the participant, with data collected over a pre-specified period. PA has historically been summarised as a single numerical measure, often the number of minutes spent in moderate-to-vigorous PA, disregarding PA performed at lower intensities. The number of minutes spent in PA during the collection period is also associated with wear time of the accelerometer. Therefore, modelling all PA intensities as proportions of the total is more appropriate.
Standard statistical methods cannot model more than one intensity of PA within a single model, as the intensities are likely to be highly correlated. Compositional data analysis (CoDA) provides a way of modelling the multiple intensities of PA within a single (regression) model. However, there has been little progress in modelling compositional PA data as an outcome variable over multiple points in time.
To expand the use of CoDA for PA data, a compositional multivariate mixed effects model has been developed to enable two outcome variables (i.e. three PA intensities represented as two isometric log ratios) to be modelled compositionally over multiple points in time. This new model was developed using data (participants aged 5 followed up until age 16 years) from a longitudinal observational study (the Early Bird cohort study) and then modified for use in intervention studies (using data (participants aged 16-74 years) from the e-coachER randomised controlled trial). The model incorporates the covariance between the two correlated outcome variables to provide greater precision (compared to univariate methods) around point estimates, using elliptical confidence regions. In addition to modelling a compositional outcome, the newly developed methodology enables more than two time points to be modelled as independent variables. Being able to model a compositional outcome and/or independent variables in a single model, rather than using multiple models, provides point estimates with greater precision.
The CoDA methods developed for modelling complex PA data could be applied to any compositional data set, providing an opportunity for use and further development of compositional methodology in other research fields.","abstract_html":"Physical activity (PA) data are often objectively collected using an accelerometer worn by the participant, with data collected over a pre-specified period. PA has historically been summarised as a single numerical measure, often the number of minutes spent in moderate-to-vigorous PA, disregarding PA performed at lower intensities. The number of minutes spent in PA during the collection period is also associated with wear time of the accelerometer. Therefore, modelling all PA intensities as proportions of the total is more appropriate.<br/><br/>Standard statistical methods cannot model more than one intensity of PA within a single model, as the intensities are likely to be highly correlated. Compositional data analysis (CoDA) provides a way of modelling the multiple intensities of PA within a single (regression) model. However, there has been little progress in modelling compositional PA data as an outcome variable over multiple points in time. <br/><br/>To expand the use of CoDA for PA data, a compositional multivariate mixed effects model has been developed to enable two outcome variables (i.e. three PA intensities represented as two isometric log ratios) to be modelled compositionally over multiple points in time. This new model was developed using data (participants aged 5 followed up until age 16 years) from a longitudinal observational study (the Early Bird cohort study) and then modified for use in intervention studies (using data (participants aged 16-74 years) from the e-coachER randomised controlled trial). The model incorporates the covariance between the two correlated outcome variables to provide greater precision (compared to univariate methods) around point estimates, using elliptical confidence regions. In addition to modelling a compositional outcome, the newly developed methodology enables more than two time points to be modelled as independent variables. Being able to model a compositional outcome and/or independent variables in a single model, rather than using multiple models, provides point estimates with greater precision. <br/><br/>The CoDA methods developed for modelling complex PA data could be applied to any compositional data set, providing an opportunity for use and further development of compositional methodology in other research fields.","abstract_has_math":false,"creators":["Chynoweth, Jade"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Joanne Hosking, Adam Streeter, Jonathan Pinkney, Siobhan Creanor"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-01-01T08:00:00Z","date_published":"2025-01-01T08:00:00Z","updated_at":"2026-07-24T03:49:42Z","subjects":["Physical Activity","Statistics","Compositional data analysis","Longitudinal","Accelerometry"],"languages":["eng"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://pearl.plymouth.ac.uk/pms-theses/60","outbound_label":"Repository record","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Joanne Hosking, Adam Streeter, Jonathan Pinkney, Siobhan Creanor"]},{"key":"dc:creator","label":"Author","values":["Chynoweth, Jade"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.available","label":"Dc Date Available","values":["2025-09-24T07:00:00Z"]},{"key":"dc:date.issued","label":"Date","values":["2025-01-01T08:00:00Z"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Physical Activity","Statistics","Compositional data analysis","Longitudinal","Accelerometry"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://pearl.plymouth.ac.uk/pms-theses/60"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Physical activity (PA) data are often objectively collected using an accelerometer worn by the participant, with data collected over a pre-specified period. PA has historically been summarised as a single numerical measure, often the number of minutes spent in moderate-to-vigorous PA, disregarding PA performed at lower intensities. The number of minutes spent in PA during the collection period is also associated with wear time of the accelerometer. Therefore, modelling all PA intensities as proportions of the total is more appropriate.
Standard statistical methods cannot model more than one intensity of PA within a single model, as the intensities are likely to be highly correlated. Compositional data analysis (CoDA) provides a way of modelling the multiple intensities of PA within a single (regression) model. However, there has been little progress in modelling compositional PA data as an outcome variable over multiple points in time.
To expand the use of CoDA for PA data, a compositional multivariate mixed effects model has been developed to enable two outcome variables (i.e. three PA intensities represented as two isometric log ratios) to be modelled compositionally over multiple points in time. This new model was developed using data (participants aged 5 followed up until age 16 years) from a longitudinal observational study (the Early Bird cohort study) and then modified for use in intervention studies (using data (participants aged 16-74 years) from the e-coachER randomised controlled trial). The model incorporates the covariance between the two correlated outcome variables to provide greater precision (compared to univariate methods) around point estimates, using elliptical confidence regions. In addition to modelling a compositional outcome, the newly developed methodology enables more than two time points to be modelled as independent variables. Being able to model a compositional outcome and/or independent variables in a single model, rather than using multiple models, provides point estimates with greater precision.
The CoDA methods developed for modelling complex PA data could be applied to any compositional data set, providing an opportunity for use and further development of compositional methodology in other research fields."]},{"key":"dc:title","label":"Title","values":["Improving the efficiency of modelling complex physical activity data using longitudinal compositional data analysis"]}]}],"canonical_facts":{"dc:contributor":["Joanne Hosking, Adam Streeter, Jonathan Pinkney, Siobhan Creanor"],"dc:creator":["Chynoweth, Jade"],"dc:date.available":["2025-09-24T07:00:00Z"],"dc:date.issued":["2025-01-01T08:00:00Z"],"dc:description.abstract":["Physical activity (PA) data are often objectively collected using an accelerometer worn by the participant, with data collected over a pre-specified period. PA has historically been summarised as a single numerical measure, often the number of minutes spent in moderate-to-vigorous PA, disregarding PA performed at lower intensities. The number of minutes spent in PA during the collection period is also associated with wear time of the accelerometer. Therefore, modelling all PA intensities as proportions of the total is more appropriate.
Standard statistical methods cannot model more than one intensity of PA within a single model, as the intensities are likely to be highly correlated. Compositional data analysis (CoDA) provides a way of modelling the multiple intensities of PA within a single (regression) model. However, there has been little progress in modelling compositional PA data as an outcome variable over multiple points in time.
To expand the use of CoDA for PA data, a compositional multivariate mixed effects model has been developed to enable two outcome variables (i.e. three PA intensities represented as two isometric log ratios) to be modelled compositionally over multiple points in time. This new model was developed using data (participants aged 5 followed up until age 16 years) from a longitudinal observational study (the Early Bird cohort study) and then modified for use in intervention studies (using data (participants aged 16-74 years) from the e-coachER randomised controlled trial). The model incorporates the covariance between the two correlated outcome variables to provide greater precision (compared to univariate methods) around point estimates, using elliptical confidence regions. In addition to modelling a compositional outcome, the newly developed methodology enables more than two time points to be modelled as independent variables. Being able to model a compositional outcome and/or independent variables in a single model, rather than using multiple models, provides point estimates with greater precision.
The CoDA methods developed for modelling complex PA data could be applied to any compositional data set, providing an opportunity for use and further development of compositional methodology in other research fields."],"dc:identifier":["https://pearl.plymouth.ac.uk/pms-theses/60"],"dc:language":["eng"],"dc:subject":["Physical Activity","Statistics","Compositional data analysis","Longitudinal","Accelerometry"],"dc:title":["Improving the efficiency of modelling complex physical activity data using longitudinal compositional data analysis"],"dc:type":["Thesis"]},"updated_at":"2026-07-24T03:49:42Z"}