我的研究设计包含三个相互作用的因素,并且我有 16S 微生物群落数据。我正在尝试建立一个模型,在其中可以测试所有主效应(A、B、C)、双向交互作用(A:B、A:C、B:C)和三个效应的效果-way 交互 (A:B:C),不指定因素的顺序(默认情况下)。理想情况下,我会得到与 III 型方差分析类似的输出。
使用边际检验(
by = "margin"
)是行不通的,因为当它这样写时,它只显示了交互项的重要性(没有主效应)。
下面是一个仅包含两个因素的可重现示例。
library(vegan)
data(dune)
data(dune.env)
adonis2(dune~Moisture*Manure,
data = dune.env,
permutations = 999,
method = "bray",
strata = dune.env$Management,
by = "margin")
输出:
Permutation test for adonis under reduced model
Marginal effects of terms
Blocks: strata
Permutation: free
Number of permutations: 999
adonis2(formula = dune ~ Moisture * Manure, data = dune.env, permutations = 999, method = "bray", by = "margin", strata = dune.env$Management)
Df SumOfSqs R2 F Pr(>F)
Moisture:Manure 4 0.4678 0.10881 0.9213 0.451
Residual 8 1.0154 0.23620
Total 19 4.2990 1.00000
当交互被写出来时也是如此
adonis2(dune~Moisture+Manure+Moisture:Manure,
data = dune.env,
permutations = 999,
method = "bray",
strata = dune.env$Management,
by = "margin")
输出:
Permutation test for adonis under reduced model
Marginal effects of terms
Blocks: strata
Permutation: free
Number of permutations: 999
adonis2(formula = dune ~ Moisture + Manure + Moisture:Manure, data = dune.env, permutations = 999, method = "bray", by = "margin", strata = dune.env$Management)
Df SumOfSqs R2 F Pr(>F)
Moisture:Manure 4 0.4678 0.10881 0.9213 0.424
Residual 8 1.0154 0.23620
Total 19 4.2990 1.00000
是否有一种合理的方法可以同时测试主要效果及其相互作用?
谢谢你
边际检验之所以称为边际检验,是因为它们只检验边际效应。如果有交互项,这些交互作用是边际的,但它们的组成主效应不是边际的,也没有经过测试。
adonis2
中没有“Type III”(有趣的名字)。没有人写过这些。