使用nlme的lme模型调用ggpredict报错，lmer模型正常运行

问题重现 #

使用nlme拟合含随机截距和斜率的混合模型：

set.seed(1)
Sex = sample(rep(c(0,1), times = 50))
set.seed(2)
BT = sample(rep(c(0,1), times = 50))
AT = abs(round(rnorm(100, 6.5, 4),0))
Age = abs(round(rnorm(100, 3.8, 2),2))
outcome <- abs(round(rnorm(100, 85, 10), 0))
ID <- seq(1,100)

data <- data.frame(ID = ID, Sex = Sex,Age = Age, BT = BT, AT = AT, outcome = outcome)

ids_list <- list()
ids <- seq(1,100)
set.seed(1)
reps <- sample(rep(2:5, each = 25))
for (i in 1:length(ids)){
  ids_list[[i]] <- rep(i, each = reps[i])
}

time_list <- list()
for (i in 1:length(ids)){
  time_list[[i]] <- sort(round(runif(reps[i], 0, 15),0), decreasing = F)
}

new_df <- data.frame(ID = unlist(ids_list), Time = unlist(time_list))
df <- merge(data, new_df, by = "ID")

model <- lme(outcome ~ Sex + BT + Age + AT*Time, random = ~ 1 + Time|ID,
                   data = df, na.action = na.exclude, method = "REML")

调用ggpredict(model, terms = c("AT", "Time"), type = "re")时触发错误：

Error in tapply(x$predicted, list(x$group), NULL) : 
  arguments must have same length

更换其他变量仍报错，但lme4::lmer拟合的模型可正常使用ggpredict。

解决方法 #

1. 移除数据中的重复观测 #

你的数据集df存在同一ID-Time组合的重复行（比如ID=1的Time=4出现两次），这是导致ggpredict处理随机效应分组时长度不匹配的核心原因。先去重后重新拟合模型：

# 去重：保留每个ID-Time的唯一行
df_clean <- df[!duplicated(df[, c("ID", "Time")]), ]
# 重新拟合模型
model_clean <- lme(outcome ~ Sex + BT + Age + AT*Time, random = ~ 1 + Time|ID,
                   data = df_clean, na.action = na.exclude, method = "REML")
# 重新运行ggpredict
ggpredict(model_clean, terms = c("AT", "Time"), type = "re")

2. 手动计算随机效应预测值并可视化 #

如果去重不符合分析需求，可手动构建预测网格，结合predict.lme()计算预测值后用ggplot可视化：

library(ggplot2)

# 创建预测网格：选取AT的分位数、Time的全范围，其他协变量取均值/众数
pred_grid <- expand.grid(
  AT = quantile(df$AT, c(0.25, 0.5, 0.75)),
  Time = seq(min(df$Time), max(df$Time), by = 1),
  Sex = round(mean(df$Sex)),
  BT = round(mean(df$BT)),
  Age = mean(df$Age),
  ID = unique(df$ID) # 包含所有个体以计算随机效应
)

# 计算含随机效应的预测值（level=1表示个体水平）
pred_grid$predicted <- predict(model, newdata = pred_grid, level = 1)

# 绘制可视化图
ggplot(pred_grid, aes(x = Time, y = predicted, color = factor(AT))) +
  geom_line(aes(group = interaction(AT, ID)), alpha = 0.3) + # 个体趋势线
  stat_summary(aes(group = AT), fun = mean, geom = "line", size = 1.2) + # 平均趋势线
  labs(x = "Time", y = "Predicted Outcome", color = "AT") +
  theme_minimal()

3. 转换nlme模型为lmer对象（临时方案） #

若需快速复用ggpredict的功能，可使用nlme4lmer包将lme模型转换为lmer对象：

library(nlme4lmer)
model_lmer <- as_lmer(model)
ggpredict(model_lmer, terms = c("AT", "Time"), type = "re")

内容的提问来源于stack exchange，提问作者yentl02