使用lmer拟合带方差结构的混合模型遇错误，求修正方法

Hey there! The error you're hitting comes from mixing syntax between two different mixed model packages: lmer() is from the lme4 package, but varWeights() is a function from the nlme package—they don't play nicely together out of the box. R is misinterpreting your varWeights() call as the control argument for lmer(), hence the confusing error message.

Here's how to fix this properly, depending on which package you want to use:

Option 1: Use nlme::lme() (Recommended for Variance Structures) #

The lme() function was designed specifically to handle mixed models with flexible variance structures, so it's the perfect fit for your goal. Here's the adjusted code:

First, install/load the nlme package if you haven't already:

install.packages("nlme")  # Only run once
library(nlme)

Then fit your model with the correct syntax:

# Your original data frame
DTF <- data.frame(Y = rnorm(7*4, 0, 1), TIME = rep(1:7, 4), GROUP = c(rep(1,7), rep(2,7), rep(3,7), rep(4,7)), X = rep(rnorm(7, 20, 5), 4))

# Fitted model with variance structure
model <- lme(
  Y ~ TIME:GROUP + X,
  random = ~1 | GROUP,  # Same as (1 | GROUP) in lmer()
  data = DTF,
  method = "ML",        # Equivalent to REML = FALSE in lmer()
  weights = varFixed(~ TIME)  # Specify your time-dependent variance structure
)

Key Notes: #

• random = ~1 | GROUP replaces the (1 | GROUP) syntax from lmer()
• method = "ML" matches your original REML = FALSE; use method = "REML" if you want restricted maximum likelihood instead
• weights = varFixed(~ TIME) directly implements the variance structure you were targeting with varWeights()

Option 2: Stick with lme4 Ecosystem via glmmTMB #

If you prefer to stay in the lme4-style syntax, the glmmTMB package supports heteroscedastic (non-constant variance) mixed models. Here's how to use it:

install.packages("glmmTMB")  # Only run once
library(glmmTMB)

model <- glmmTMB(
  Y ~ TIME:GROUP + X + (1 | GROUP),
  data = DTF,
  REML = FALSE,
  dispformula = ~ TIME  # Define variance structure here
)

The dispformula argument lets you specify how the residual variance varies with your TIME variable, which achieves the same goal as your original varWeights() call.

内容的提问来源于stack exchange，提问作者Emiliano Traini