Reliability and Validity of the Computer Science and Applications Accelerometer in a Mechanical Setting

Abstract

Accelerometry has been proposed as an objective method for measurement of physical activity. Intra- and inter-instrument reliability and validity of the Computer Science and Applications (CSA) Model 7164 accelerometer, along with the nature of the analogue filter in the device, was studied with sinusoidal movements in a mechanical setting. Six CSA units were tested with 17 different frequencies (0.5 - 4 Hz) on 3 radius settings (0.022 - 0.049 m), yielding 51 different acceleration settings (0.1 - 19.7 m· sec-2). Intra-instrument reliability was relatively good (mean CV of 4.4%), although questionable on extreme values of acceleration (less than 1 m · sec-2 and higher than 16 m · sec-2). Analyses on inter-instrument reliability revealed both overall systematic bias and acceleration-specific differences between units, approaching unit error magnitudes of 20% from the mean of all 6 units. The correlation between CSA output and acceleration (i.e., validity) was significant but the relation was clearly nonlinear. Applying equations to cancel the frequency-dependent filtering restored this linearity (R2 = 0.97). In conclusion, the CSA exhibits reasonable intra-instrument reliability but that large unit differences call for unit-specific calibration or alternatively statistical adjustment. Linearity between CSA output and acceleration is achieved only by correcting for movement frequency filtering.