XI Conference – European Congress of Methodology

Analysis of long-term and feedback effects in individual physical activity patterns

23 Jul 2025, 08:45

15m

Faculty of Social Sciences and Communication. (The Pyramid)/11 - Room (Faculty of Social Sciences and Communication. (The Pyramid))

Oral Presentation Statistical analyses Session 10 : "Statistical Applications in Social Sciences and Sports"

Speaker

Marek Brabec (Institute of Computer Science)

Abstract

This methodological study has been conducted as part of the „Research of Excellence on Digital Technologies and Wellbeing CZ.02.01.01/00/22_008/0004583“ co-financed by the European Union and is based on analyzing large-scale ecological momentary assessment data of daily step counts from the „Healthy Aging in Industrial Environment HAIE CZ.02.1.01/0.0/0.0/16_019/0000798“ project. It starts with traditional mixed effects linear hierarchical Markovian (AR-type) model allowing for the separation of between- and within-subject relationships to important covariates (such as sex, age, exercise identity, education, SES level) and their interactions. We show that the traditional additive modeling is not sufficient and there are important interactions with place of residence. Then, we carefully explore seasonality of the physical activity at two scales (weekly and annual) proving their significance and quantifying their relative contributions. We demonstrate that corrections for seasonality are important in that if unadjusted, they distort systematically individual dynamics and give false impression of external controllability of the activity (e.g. by intervention) where important parts are pre-determined by largely unchangeable seasonal patterns. Further insight is based on modeling autoregressive component in a nonlinear way (accounting for local saturation effects connected with over- or under-exercising that cannot be captured by linear models) utilizing the flexible generalized additive modeling approach. Since the long-term effects (related e.g. to fatigue after cumulative over-exercising spanning several days) in the AR dynamical structure are notoriously difficult to model, we approach the problem from the complexity-penalizing viewpoint and regularize the higher lag coefficient behavior based on flexible (soft) constraints. This allows us to demonstrate long-term dynamic details that operate above the seasonal behavior and hence are controllable in principle. Since our model is hierarchical with individual-specific random effects, we are able to separate and quantify several sources of inter-individual variability and show that specific parts of inter-individual structural variability in the physical activity dynamics is an important feature to consider when designing new interventions programs.