Table 1.
Mixed effect regression models used to analyse (1) the change in difference between GPAQ and SenseWear measurements; (2) the effect of personal attributes on the difference between both methods.
Table 2.
Median and IQR of the PA measures aggregated over three sessions per participant for both measurement methods (GPAQ and SenseWear).
Number of participants included in the analysis is 122.
Fig 1.
Boxplots of MVPA energy expenditure (EE), SB, moderate EE and vigorous EE per method and session.
For each session, the difference Δ and Spearman correlation coefficient r is specified. Δ was calculated as the mean difference between both methods and was tested for significance using the Wilcoxon signed rank sum test. rrm is the overall Spearman correlation adjusted for repeated measures (rm) [33]; pΔ(t) = the p-value of the effect of session in the Δ(t) model which indicates if the difference between GPAQ and SenseWear measurements changes per repeated measurement. Statistical significance is expressed as *p<0.05, **p<0.01, and *** p<0.001.
Fig 2.
Bland-Altman plots comparing MVPA, moderate and vigorous energy expenditure (METminutes/week) and SB (minutes/day) measured by the SenseWear armband (SW) and the GPAQ.
All percentage differences on the Y-axis are calculated by subtracting GPAQ from SenseWear results divided by their average. Moderate- and vigorous-intensity activities included influential observation. The red, dashed lines represent the mean difference and 95% limits of agreement excluding these observations. EE = energy expenditure.
Table 3.
β1-coefficients of the attribute models to indicate the effect of sex, age, BMI, body fat and PA domain on the absolute and percentage difference (Δ) between measured (SenseWear) and reported (GPAQ) METminutes/week, minutes PA/day and sedentary minutes/day.
The difference between both methods is used as the dependent variable. Absolute differences were calculated as SenseWear minus GPAQ results. Percentage differences are calculated by dividing the absolute difference by the average of the measurements from both methods. Separate models were fitted for each characteristic. Each model included random participant effects clustered per city.