Skip to main content

Table 3 Latent class analysis model summary for WA variables (N subjects = 72, N observations = 3716, with best fitting solution in bold) % subjects in class

From: Latent class analysis of actigraphy within the depression early warning (DEW) longitudinal clinical youth cohort

Variable

N of Classes

Log Likelihood

BIC

AIC

1

2

3

4

5

6

Acceleration

1

−15757.1

31544.1

31528.2

100.0

     

2

−15304.7

30652.1

30629.3

37.5

62.5

    

3

−15205.6

30466.7

30437.1

18.1

52.8

29.2

   

4*

−15154.5

30377.4

30341.0

23.6

41.7

11.1

23.6

  

5”

−15166.1

30413.4

30370.1

9.7

22.2

40.3

22.2

5.6

 

6

−15138.1

30370.3

30320.2

23.6

37.5

11.1

11.1

1.4

15.3

Sleep duration

1

−22589.1

45208.1

45192.2

100.0

     

2

−22417.5

44877.7

44854.9

86.1

13.9

    

3

−22336.4

44728.3

44698.7

41.7

47.2

11.1

   

4

−22251.9

44572.3

44535.9

34.7

51.4

11.1

2.8

  

5*

−22234.2

44549.7

44506.4

51.4

34.7

6.9

5.6

1.4

 

6

−22226.5

44547.0

44497.0

34.7

51.4

6.9

4.2

1.4

1.4

Sleep efficiency

1

2133.4

−4215.6

−4242.9

100.0

     

2

2340.9

−4617.7

−4651.9

75.0

25.0

    

3*

2400.2

−4723.4

−4764.4

27.8

54.2

18.1

   

4

2405.6

−4721.5

−4769.3

12.5

23.6

18.1

45.8

  

5

2414.6

−4726.5

−4781.2

27.8

52.8

5.6

8.3

5.6

 

6

2414.7

−4713.9

−4775.3

27.8

45.8

18.1

2.8

5.6

0.0

  1. Selected latent class models based on BIC, AIC, Lo-Mendel test and class distributions
  2. *Lo-Mendell-Rubin ad-hoc adjusted likelihood ratio test results not significant (at alpha = 0.05) between class # and the next (e.g. p = 1.00 between 4 and 5 classes)
  3. Model failed to converge at global maximum, hence Log Likehood is smaller than previous class