r - 在因子 : GAM 内添加两个随机效应

Question

从这个数据:

UQdata  MudUQ   Estuary   Site
7.00    10.9    NoriPau   A
6.00    13.9    NoriPau   A
5.00    10.3    NoriPau   B
4.00    7.9     Porirua   A
4.00    8.3     Porirua   A
4.00    8.7     Porirua   A
4.00    10.9    NoriPau   B
3.00    9.8     Porirua   B
3.00    9.8     Porirua   B
3.00    11.5    Porirua   B

我正在使用 mgcv 包拟合以下 GAM 模型:

aa2.estuary <- gam(UQdata~s(MudUQ, bs="ps", k=5) + s(Estuary, bs="re"), 
                   family=Gamma(link=log),data=Antho)

问题:我想加 Estuary和 Site内 Estuary作为两个随机效应(即 s(Estuary ~ Site + Estuary, bs="re") )，但是当我尝试这样做时，它会引发此错误:

aa2.estuary <- gam(UQdata ~ s(MudUQ,bs="ps", k=5) + 
                   s(Estuary~Estuary+Site, bs="re"),
                   family=Gamma(link=log),data=Antho)

Error in `contrasts<-`(`*tmp*`, value = contr.funs[1 + isOF[nn]]) : 
  contrasts can be applied only to factors with 2 or more levels
In addition: Warning message:
In s(Estuary ~ Estuary + Site, bs = "re") :
  number of items to replace is not a multiple of replacement length

任何帮助在这里将不胜感激。

mnel，我试过你的建议字符串:

> aa1.estuary<-gam(UQdata~s(MudUQ,bs="ps", k=5) + s(Estuary, bs="re") + s(Site, Estuary, bs = 're'),family=binomial, gamma=1,data=Antho)
Error in while (mean(ldxx/(ldxx + ldss)) < 0.4) { : 
  missing value where TRUE/FALSE needed

有什么想法吗？

Answer 1

通过返回原始数据并重命名 Sites与每个河口相关(见下文):

UQdata  MudUQ   Estuary   Site
7.00    10.9    NoriPau   Nori1
6.00    13.9    NoriPau   Nori1
5.00    10.3    NoriPau   Nori2
4.00    7.9     Porirua   Pori1
4.00    8.3     Porirua   Pori1
4.00    8.7     Porirua   Pori1
4.00    10.9    NoriPau   Nori2
3.00    9.8     Porirua   Pori2
3.00    9.8     Porirua   Pori2
3.00    11.5    Porirua   Pori2

包括 Site作为另一个随机效应:

aa2.estuary <- gam(UQdata ~ s(MudUQ,bs="ps", k=5) + s(Estuary, bs="re") + s(Site, bs="re"),family=Gamma(link=log),data=Antho)

并且不用担心哪个嵌套在哪个中。这兼顾了 Site相关性，以及内部 Estuary相关性。