This consulting project was part of my graduate consulting curriculum, where students provide statistical consulting for the statistics department.
The client (Noah Jeffery) was a Biomedical Engineering M.S. student working on their thesis project. Noah had come to the consulting department a few weeks earlier to help design his study, and has returned with collected data. He put three devices on his subjects to measure the angle of their spine while sitting, with the goal of showing that all three devices are necessary to measure different postures.
The first part of his study was 15 minutes of unstructured time. From this data he wanted summary statistics. The second part of the study had the subjects following a guided video where they would perform multiple postures (good, laying back, lumbar slouch, etc.). From this data he wanted what angle thresholds and combinations that resulted in poor posture (anything that wasn’t good posture) and to mark these postures in the 15 minutes of unstructured time.
My group did the following:
Used linear mixed models to determine the confidence and prediction intervals for each posture for each device in the structured time (guided video section)
Created visualizations showing that each device was able to identify different postures, requiring all three devices to detect all postures
Applied the prediction intervals to the 15 minutes of unstructured time to create plots for each subject, as well as summary statistics across all subjects (e.g. time spent in each posture)