例を示してみよう.
Laplace 近似 (nAGQ = 1
)
の場合はこんなかんぢなのだが,
Generalized linear mixed model fit by the Laplace approximation
Formula: f
Data: d
AIC BIC logLik deviance
166 212 -73.8 148
Random effects:
Groups Name Variance Std.Dev.
ID (Intercept) 0.000 0.00
Residual 0.109 0.33
Number of obs: 1353, groups: ID, 1353
Fixed effects:
Estimate Std. Error t value
(Intercept) -4.777170 0.097535 -49.0
SeasonSummer 0.591766 0.025425 23.3
SeasonAutumn 0.338857 0.027257 12.4
SeasonWinter -0.051675 0.027241 -1.9
AreaSouthern 0.583936 0.014148 41.3
Depth -0.001259 0.000117 -10.7
SL 0.011122 0.000430 25.9
Gauss-Hermite 近似
(nAGQ = 10
, ただし分割点数とはあまり関係ない)
ではこうなった.
Generalized linear mixed model fit by the adaptive Gaussian Hermite approximation
Formula: f
Data: d
AIC BIC logLik deviance
1920 1966 -951 1902
Random effects:
Groups Name Variance Std.Dev.
ID (Intercept) 7.40 2.72
Residual 2.78 1.67
Number of obs: 1353, groups: ID, 1353
Fixed effects:
Estimate Std. Error t value
(Intercept) -4.78e+00 2.84e+15 -1.70e-15
SeasonSummer 5.91e-01 7.03e+14 8.00e-16
SeasonAutumn 3.39e-01 7.22e+14 5.00e-16
SeasonWinter -5.23e-02 6.78e+14 -1.00e-16
AreaSouthern 5.83e-01 4.32e+14 1.40e-15
Depth -9.82e-01 3.30e+12 -2.97e-13
SL -1.85e-01 1.23e+13 -1.51e-14
係数の推定値の Std. Error
の異常なる値をみてやってください.
ついでにいうと,
じつは推定値自体もけっこうヘンだったり.