A Practical Guide to Analyzing Time-Varying Associations between Physical Activity and Affect Using Multilevel Modeling

Figure 1

Analytic techniques for physical activity. (a) The temporal associations of depressive mood and local mean of physical activity which evaluates lower/higher mean activity levels. Estimated values of the univariate multilevel model coefficient for the associations are shown in a colored matrix form consisting of 25 columns (different location) and 12 rows (different size) in older adults (n = 9). Each grid cell indicates specific location and size of a time frame used for calculating the local mean of physical activity surrounding each EMA recording of depressive mood. A color in each cell represents the value of the model coefficient (γ₁₀) of the predictors. The false discovery rate with the q value of .05 was used as the multiple comparison adjustment. Only the significant cases were shown by colors. Note that the univariate model used for the analysis is as follows. Depressive mood score_tj = γ₀₀ + γ₁₀ (local statistics of physical activity_tj) + ζ_0j + ε_tj [see [85] for details]. (b) The same is shown in panels (a), except for the local mean. Local skewness of physical activity, which evaluates asymmetry of a distribution (i.e., occasional bursts of physical activity in a time window), was used in this panel. (c) A raw physical activity time series for 120 min and the second-order polynomial line (red). (d) The detrended physical activity derived by subtracting the fitted line for the original data.

(a) Mean

(b) Detrended skewness

(c) Raw physical activity

(d) Detrended physical activity