Table 2. Model fitting summary.
Model | Goodness of fit | Deviance | Transfer to KSS | ||||||
No | Description | log L | ε | η | vs | chi2 | a | b | |
Models with observed sleeps in the home base time zone (n = 5744) | |||||||||
1a | Null model | −11798.9 | 1.887 | 4.58 | |||||
1b | Random intercept | −11174.1 | 1.656 | 0.809 | 1a | 1250 | 4.62 | ||
2a | C | −11125.7 | 1.642 | 0.813 | 1b | 97 | 4.67 | −0.14 | |
2b | S | −10559.8 | 1.485 | 0.808 | 1b | 1229 | 9.82 | −0.45 | |
2c | SB | −10520.5 | 1.474 | 0.828 | 2b | 79 | 9.70 | −0.46 | |
3a | SC | −10315.0 | 1.422 | 0.819 | 1b | 1718 | 10.19 | −0.47 | |
3b | SCW | −10629.3 | 1.504 | 0.815 | 3a | −629 | 8.51 | −0.35 | |
3c | SCU | −10298.2 | 1.418 | 0.818 | 3a | 34 | 10.03 | −0.48 | |
3d | SCUW | −10590.4 | 1.493 | 0.814 | 3c | −584 | 8.57 | −0.36 | |
4a | SCp15h | −10151.9 | 1.382 | 0.810 | 9.46 | −0.40 | |||
4b | SCp15U | −10163.8 | 1.385 | 0.809 | 9.13 | −0.39 | |||
4c | SCTU | −10114.0 | 1.372 | 0.811 | 9.41 | −0.42 | |||
4d | SCT | −10102.6 | 1.370 | 0.811 | 4c | 23 | 9.75 | −0.43 | |
5a | SBC | −10313.8 | 1.421 | 0.838 | 3a | 2 | 9.83 | −0.46 | |
5b | SBCW | −10629.5 | 1.503 | 0.823 | 5a | −631 | 8.24 | −0.33 | |
5c | SBCU | −10297.7 | 1.417 | 0.839 | 5a | 32 | 9.68 | −0.46 | |
5d | SBCUW | −10593.0 | 1.494 | 0.823 | 5c | −591 | 8.28 | −0.35 | |
6a | SBCp15h | −10118.6 | 1.373 | 0.827 | 5c | 358 | 9.30 | −0.41 | |
6b | SBCp15hU | −10128.0 | 1.375 | 0.826 | 5c | 339 | 8.99 | −0.40 | |
6c | SBCTU | −10083.5 | 1.364 | 0.829 | 5c | 428 | 9.24 | −0.42 | |
6d | SBCT | −10074.5 | 1.362 | 0.829 | 5c | 446 | 9.56 | −0.43 | |
Models with generated sleeps in the home base time zone (n = 5744) | |||||||||
7a | SBCU | −10464.8 | 1.460 | 0.826 | 5c | −334 | 9.68 | −0.47 | |
7b | SBCp15hU | −10468.2 | 1.461 | 0.812 | 7a | −7 | 9.03 | −0.42 | |
7c | SBCp15hU (new thresholds) | −10279.4 | 1.413 | 0.809 | 7a | 371 | 8.71 | −0.37 | |
7d | SBCTU (new thresholds) | −10245.1 | 1.405 | 0.812 | 7a | 439 | 9.04 | −0.40 | |
8a | SBC | −10488.8 | 1.466 | 0.822 | 7a | −48 | 9.77 | −0.46 | |
8b | SBCp15h | −10477.6 | 1.464 | 0.811 | 7a | −26 | 9.30 | −0.42 | |
8c | SBCp15h (new thresholds) | −10273.7 | 1.412 | 0.808 | 7a | 382 | 8.99 | −0.38 | |
8d | SBCT (new thresholds) | −10241.6 | 1.404 | 0.811 | 7a | 446 | 9.33 | −0.40 | |
Models on data in all time zones (n = 8040) | |||||||||
10a | SBC | −15344.6 | 1.449 | 0.816 | 9.74 | −0.45 | |||
10b | SBCp15h | −15081.1 | 1.404 | 0.807 | 10a | 527 | 9.30 | −0.41 | |
10c | SBCT | −15027.5 | 1.395 | 0.807 | 10a | 634 | 9.52 | −0.42 | |
11a | SBCTA100% | −15086.0 | 1.405 | 0.806 | 10c | −117 | 9.49 | −0.42 | |
11b | SBCTA50% | −15003.8 | 1.391 | 0.806 | 10c | 47 | 9.54 | −0.43 | |
11c | SBCTA30% | −14999.1 | 1.391 | 0.806 | 10c | 57 | 9.55 | −0.43 |
Note. Description lists model components (SCUWA) used to calculate the alertness score. Subscripts indicate the presence of the “brake function” described in equation 2–3 for process S (B), if the phase of C was different from the default p = 16.8 h based on circadian type (T) or set at 15 h (p15h) for all subjects, and the daily acclimatization rate for process A (%). log L = log likelihood. ε = residual standard deviation. η = subject level random effect standard deviation. All models (except 1a & 1b) are based on equation 11 and differ only in how the alertness score was calculated. Deviance indicate 2*- log likelihood difference (chi2) compared to selected models (vs). Transfer to KSS describes the best fitting constant (a) and coefficient (b) to transfer the alertness score to KSS (equation 9).