在R中用nlme和SSasympOrig为渐近回归模型加协变量遇问题

现有drc模型 #

已使用drc库拟合分组渐近模型，代码如下：

library(drc)
library(nlme)
library(dplyr)

df <- structure(list(iv = c(1, 2, 3, 4, 5, 6, 1, 2, 3, 4, 5, 6, 1, 
2, 3, 4, 5, 6, 1, 2, 3, 4, 5, 6, 1, 2, 3, 4, 5, 6, 1, 2, 3, 4, 
5, 6), dv = c(9.2, 8.5, 13.5, 15.8, 18.3, 17.7, 8.7, 10.8, 14.3, 
15, 18, 15.3, 8.7, 14.6, 14.8, 16.8, 15.8, 15.8, 14.2, 13.5, 
18.5, 20.8, 23.3, 22.7, 13.7, 15.8, 19.3, 20, 23, 20.3, 13.7, 
19.6, 19.8, 21.8, 20.8, 20.8), group = c("1", "1", "1", "1", 
"1", "1", "1", "1", "1", "1", "1", "1", "1", "1", "1", "1", "1", 
"1", "2", "2", "2", "2", "2", "2", "2", "2", "2", "2", "2", "2", 
"2", "2", "2", "2", "2", "2")), row.names = c(NA, -36L), class = c("tbl_df", 
"tbl", "data.frame"))

drm_mod <- drm(
  dv ~ iv,
  curveid = group,
  data = df,
  fct = AR.2(names = c("Upper Limit", "Steepness"))
)

报错情况 #

尝试用nlme复现模型时出现两类错误：

错误1：指定groups参数时 #

执行代码：

nlme_mod <- nlme(dv ~ SSasympOrig(iv, Asm, lrc), data=df, 
                    fixed= list(Asym + lrc ~ group),
                    groups = group,
                    random = ~ 1)

得到报错：

Error: object 'group' not found
2.
nlme.formula(dv ~ SSasympOrig(iv, Asm, lrc), data = df, fixed = list(Asym +
lrc ~ group), groups = group, random = ~1)
1.
nlme(dv ~ SSasympOrig(iv, Asm, lrc), data = df, fixed = list(Asym +
lrc ~ group), groups = group, random = ~1)

错误2：移除groups参数时 #

执行代码：

nlme_mod <- nlme(dv ~ SSasympOrig(iv, Asm, lrc), data=df, 
                    fixed= list(Asym + lrc ~ group),
                #    groups = group,
                    random = ~ 1)

得到报错：

Error in nlsList.formula(model = dv ~ SSasympOrig(iv, Asm, lrc), data = df) : 
  'data' must be a "groupedData" object if 'formula' does not include groups
8.
stop("'data' must be a \"groupedData\" object if 'formula' does not include groups")
7.
nlsList.formula(model = dv ~ SSasympOrig(iv, Asm, lrc), data = df)
6.
nlme::nlsList(model = dv ~ SSasympOrig(iv, Asm, lrc), data = df)
5.
eval(expr, p)
4.
eval(expr, p)
3.
eval.parent(nlsLCall)
2.
nlme.formula(dv ~ SSasympOrig(iv, Asm, lrc), data = df, fixed = list(Asym +
lrc ~ group), random = ~1)
1.
nlme(dv ~ SSasympOrig(iv, Asm, lrc), data = df, fixed = list(Asym +
lrc ~ group), random = ~1)

解决方法 #

报错根源在于参数格式错误、模型参数名写错，以及随机效应设定不符合“观测独立”的前提：

1. 修正参数名：SSasympOrig的渐近值参数是Asym，而非代码中的Asm
2. groups参数格式：必须传入公式对象~group，不能直接写变量名
3. fixed参数格式：无需用list包裹，直接用公式Asym + lrc ~ group指定两个参数随分组变化
4. 随机效应设定：因所有观测独立，无随机变异，将random设为~0

修正后的nlme代码 #

nlme_mod <- nlme(
  dv ~ SSasympOrig(iv, Asym, lrc),
  data = df,
  fixed = Asym + lrc ~ group,
  groups = ~group,
  random = ~0
)

# 查看结果
summary(nlme_mod)

更贴合drc模型的替代方案 #

由于drc的模型是分组固定效应模型（无随机效应），使用nlsList更简洁，结果与drc完全一致：

nls_mod <- nlsList(
  dv ~ SSasympOrig(iv, Asym, lrc) | group,
  data = df
)

# 查看分组参数
summary(nls_mod)

内容的提问来源于stack exchange，提问作者myfatson