Question

此示例来自Faraway＆＃39; Extending the Linear Model with R。

data(esoph)

modl <- glm(cbind(ncases,ncontrols) ~ agegp + alcgp + tobgp, family=binomial, data=esoph)

summary(modl)
Call:
glm(formula = cbind(ncases, ncontrols) ~ agegp + alcgp + tobgp, 
family = binomial, data = esoph)

Deviance Residuals: 
Min       1Q   Median       3Q      Max  
-1.6891  -0.5618  -0.2168   0.2314   2.0642  

Coefficients:
        Estimate Std. Error z value Pr(>|z|)    
(Intercept) -1.77997    0.19796  -8.992  < 2e-16 ***
agegp.L      3.00534    0.65215   4.608 4.06e-06 ***
agegp.Q     -1.33787    0.59111  -2.263  0.02362 *  
agegp.C      0.15307    0.44854   0.341  0.73291    
agegp^4      0.06410    0.30881   0.208  0.83556    
agegp^5     -0.19363    0.19537  -0.991  0.32164    
alcgp.L      1.49185    0.19935   7.484 7.23e-14 ***
alcgp.Q     -0.22663    0.17952  -1.262  0.20680    
alcgp.C      0.25463    0.15906   1.601  0.10942    
tobgp.L      0.59448    0.19422   3.061  0.00221 ** 
tobgp.Q      0.06537    0.18811   0.347  0.72823    
tobgp.C      0.15679    0.18658   0.840  0.40071    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

Null deviance: 227.241  on 87  degrees of freedom
Residual deviance:  53.973  on 76  degrees of freedom
AIC: 225.45

Number of Fisher Scoring iterations: 6

> modl$dev
[1] 227.2406

> modl$null
[1] 227.2406

> modl$df.residual
[1] 76

> modl$df.null
[1] 87

?glm说＆＃34; deviance：最大为常数，减去最大对数似然的两倍。在合理的情况下，选择常数使得饱和模型的偏差为零。＆＃34;

我认为model$dev与Residual deviance的值相同。 53.973，在这种情况下。我不明白为什么这个例子的modl$dev = modl$null。有人能告诉我我做错了吗？

来自glm：glm $ dev的R意外输出等于glm $ null

0 个答案: