Skip to main content
. 2021 Jan 28;12:625247. doi: 10.3389/fpsyt.2021.625247

Table 3.

A comparison of MLM model performance on the prediction of depression.

Model Fixed effects Goodness of fit and Comparison*
Estimate SE t-value Df p-value F df1, df2 P-value
Baseline model
   Intercept 0 0.13 0 105 > 0.999
EMA model
   Intercept 0 0.11 0.00 104 > 0.999 10.52 1, 172 0.001a
   Valence −0.39 0.11 −3.49 55 0.001
GPS model
   Intercept 0 0.12 0 104 > 0.999 4.574751 1, 568 0.033a
   Variance −0.21 0.10 −2.15 81 0.035
Extended digital phenotyping model: GPS and wearable data
   Intercept 0 0.11 0.00 103 > 0.999 5.23 1, 136 0.024b
   Variance −0.21 0.10 −2.17 78 0.033
   Time in bed 0.25 0.10 2.43 58 0.018
Combined model: EMA and digital phenotyping model
   Intercept 0 0.10 0.00 102 > 0.999 11.47 1, 176 0.001c
   Variance −0.21 0.09 −2.32 72 0.023 5.42 2, 560 0.005d
   Time in bed 0.24 0.09 2.54 59 0.014
   Arousal −0.38 0.10 −3.61 52 0.001
*

For more details on the likelihood ratio test see chapter 5.3.1 in Van Buuren (66).

a

Comparison against baseline,

b

Comparison against GPS,

c

Comparison against extended sensing model,

d

Comparison against EMA model.