我有一个二项式 GLM：
all.fit <- glm(Presence/Total~Season*ToD*Site, family = binomial, weights = Total, data = all.dt)
Call: glm(formula = Presence/Total ~ Season * ToD * Site, family = binomial,
data = all.dt, weights = Total)
Coefficients:
(Intercept) SeasonSpring
-4.66074 0.60478
SeasonSummer SeasonWinter
-0.82682 1.26868
ToDDay ToDDusk
0.09865 2.48420
ToDNight SiteKawau
2.65294 3.09495
SiteNoises SiteTawharanui
3.48048 2.94694
SiteTiritiri SeasonSpring:ToDDay
3.25976 0.90540
SeasonSummer:ToDDay SeasonWinter:ToDDay
2.16685 -0.12465
SeasonSpring:ToDDusk SeasonSummer:ToDDusk
-2.78750 -1.52114
SeasonWinter:ToDDusk SeasonSpring:ToDNight
-0.90286 -1.66162
SeasonSummer:ToDNight SeasonWinter:ToDNight
-1.85826 -0.21758
SeasonSpring:SiteKawau SeasonSummer:SiteKawau
-0.67217 1.24760
SeasonWinter:SiteKawau SeasonSpring:SiteNoises
-1.68237 -0.79023
SeasonSummer:SiteNoises SeasonWinter:SiteNoises
0.42570 -0.73055
SeasonSpring:SiteTawharanui SeasonSummer:SiteTawharanui
-0.25773 1.32456
SeasonWinter:SiteTawharanui SeasonSpring:SiteTiritiri
-0.51055 -0.51381
SeasonSummer:SiteTiritiri SeasonWinter:SiteTiritiri
0.95804 -1.29898
ToDDay:SiteKawau ToDDusk:SiteKawau
0.70933 -2.35320
ToDNight:SiteKawau ToDDay:SiteNoises
-5.33351 0.27852
ToDDusk:SiteNoises ToDNight:SiteNoises
-2.53814 -4.04432
ToDDay:SiteTawharanui ToDDusk:SiteTawharanui
-0.16455 -2.76975
ToDNight:SiteTawharanui ToDDay:SiteTiritiri
-2.62701 0.81969
ToDDusk:SiteTiritiri ToDNight:SiteTiritiri
-2.24361 -4.14062
SeasonSpring:ToDDay:SiteKawau SeasonSummer:ToDDay:SiteKawau
-0.19831 -1.61316
SeasonWinter:ToDDay:SiteKawau SeasonSpring:ToDDusk:SiteKawau
0.57193 2.48229
SeasonSummer:ToDDusk:SiteKawau SeasonWinter:ToDDusk:SiteKawau
1.37786 0.81702
SeasonSpring:ToDNight:SiteKawau SeasonSummer:ToDNight:SiteKawau
1.41405 2.06227
SeasonWinter:ToDNight:SiteKawau SeasonSpring:ToDDay:SiteNoises
0.08689 -0.28007
SeasonSummer:ToDDay:SiteNoises SeasonWinter:ToDDay:SiteNoises
-1.48112 -0.06044
SeasonSpring:ToDDusk:SiteNoises SeasonSummer:ToDDusk:SiteNoises
1.93847 0.63232
SeasonWinter:ToDDusk:SiteNoises SeasonSpring:ToDNight:SiteNoises
0.83278 0.69621
SeasonSummer:ToDNight:SiteNoises SeasonWinter:ToDNight:SiteNoises
0.06911 0.48690
SeasonSpring:ToDDay:SiteTawharanui SeasonSummer:ToDDay:SiteTawharanui
-0.23150 -1.96624
SeasonWinter:ToDDay:SiteTawharanui SeasonSpring:ToDDusk:SiteTawharanui
0.43487 2.04614
SeasonSummer:ToDDusk:SiteTawharanui SeasonWinter:ToDDusk:SiteTawharanui
0.59597 0.73280
SeasonSpring:ToDNight:SiteTawharanui SeasonSummer:ToDNight:SiteTawharanui
1.02467 1.19957
SeasonWinter:ToDNight:SiteTawharanui SeasonSpring:ToDDay:SiteTiritiri
-0.16571 -0.48960
SeasonSummer:ToDDay:SiteTiritiri SeasonWinter:ToDDay:SiteTiritiri
-1.63190 0.35354
SeasonSpring:ToDDusk:SiteTiritiri SeasonSummer:ToDDusk:SiteTiritiri
1.77115 0.79347
SeasonWinter:ToDDusk:SiteTiritiri SeasonSpring:ToDNight:SiteTiritiri
0.48411 1.51686
SeasonSummer:ToDNight:SiteTiritiri SeasonWinter:ToDNight:SiteTiritiri
1.27621 0.77385
Degrees of Freedom: 22998 Total (i.e. Null); 22919 Residual
Null Deviance: 105700
Residual Deviance: 63170 AIC: 87340
然后我做了一个方差分析：
> anova(all.fit,test = "Chisq")
Analysis of Deviance Table
Model: binomial, link: logit
Response: Presence/Total
Terms added sequentially (first to last)
Df Deviance Resid. Df Resid. Dev Pr(>Chi)
NULL 22998 105714
Season 3 1005.8 22995 104709 < 2.2e-16 ***
ToD 3 16404.9 22992 88304 < 2.2e-16 ***
Site 4 7664.5 22988 80639 < 2.2e-16 ***
Season:ToD 9 2793.5 22979 77846 < 2.2e-16 ***
Season:Site 12 3890.0 22967 73956 < 2.2e-16 ***
ToD:Site 12 9804.5 22955 64151 < 2.2e-16 ***
Season:ToD:Site 36 981.0 22919 63170 < 2.2e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
在这个例子中，所有的交互都是重要的。但是，我不想在我的报告中报告大量重要的因素组合。既然我想如果交互很重要，那么单一的影响并不重要？我怎样才能更简单地说明我的发现的重要性？
我可以使用 anova(all.fit) 的输出吗？如果我这样做了，我会报告什么？