Advancing health-related cluster analysis methodology: quantification of pairwise activity cluster similarities

J Phys Act Health. 2015 Mar;12(3):395-401. doi: 10.1123/jpah.2014-0008. Epub 2014 May 9.

Abstract

Background: To date, most health-related time-use research has investigated behaviors in isolation; more recently, however, researchers have begun to conceptualize behaviors in the form of multidimensional patterns or clusters.

Methods: The study employed 2 techniques: radar graphs and centroid vector length, angles and distance to quantify pairwise time-use cluster similarities among adolescents living in Australia (N = 1853) and in New Zealand (N = 679).

Results: Based on radar graph shape, 2 pairs of clusters were similar for both boys and girls. Using vector angles (VA), vector length (VL) and centroid distances (CD), 1 pair for each sex was considered most similar (boys: VA = 63°, VL = 44 and 50 units, and CD = 48 units; girls: VA = 23°, VL = 65 and 85 units, and CD = 36 units). Both methods employed to determine similarity had strengths and weaknesses.

Conclusions: The description and quantification of cluster similarity is an important step in the research process. An ability to track and compare clusters may provide greater understanding of complex multidimensional relationships, and in relation to health behavior clusters, present opportunities to monitor and to intervene.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Adolescent
  • Adolescent Behavior*
  • Australien
  • Cluster Analysis*
  • Data Collection
  • Female
  • Health Behavior*
  • Humans
  • Life Style
  • Male
  • Neuseeland