Reliability of Body Composition Assessment Using Generalizability Theory (G-Theory).

Abstract

PURPOSE: Adequate reliability of body composition (BC) assessments is a requirement before such measures can be considered valid. Many studies to date have only examined a single source of measurement error such as that from trials (test-retest). Generalizability theory (G-theory) is a statistical technique that allows for the examination of different sources of measurement error simultaneously in a single analysis. Therefore, the aim of this study was to examine the different sources of error seen in the assessment of BC. A secondary purpose was to determine the appropriate number of trials required of each assessment in order to gain a reliable BC measure. METHODS: This measurement study included 38 participants who had been assessed on two different occasions (in the same week) and on each of four different BC field METHODS: percent body fat (PBF) by skinfold technique (SF), waist circumference (WC), body mass index (BMI), and PBF by handheld bioelectrical impedance (HH). Two different G-theory designs were used in this research. First, a two facet crossed p×t×m design was analyzed treating all facets as random. Then, the same design was performed treating BC method as a fixed facet. In both designs, a generalizability study (G-study) and decision study (D-study) were conducted. Three different software packages were used to ensure consistent and valid results (GENOVA, SPSS macro, and SAS GLM). RESULTS: The completely random design showed the largest variance component for persons (p) (57.8%). Variance components for both trials (t) and BC method (m) were negligible. However, the interaction between persons and method (p×m) was substantial (38.6%). D-study results indicated reliable BC scores for measurement designs administered once using three different methods (G=.803). The mixed design, averaging over BC method, showed majority of variance due to persons (98.5%) and each of the four BC methods showed reliable scores with a single trial (G’s>.945). CONCLUSIONS: Results from this G-theory study indicate that the measurement equivalence of commonly administered BC assessments may be inadequate. Although different BC assessments individually are reliable, for dependable BC trait generalization to the universe, a minimum of three different methods administered once may be required.