GAM建模变量规格错误：添加timepoints后模型报错求助

原有正常运行的回归模型 #

model1 = bam(depend ~ s(indep, by = gender.x,bs = "tp") + 
    s(Age, by = gender.x, bs = "tp") + ti(Age, indep, by = gender.x) +
    gender.x + 
    s(Subject, bs = "re") + s(Age, Subject, bs = "re") + 
    s(ImagingCentreCity, bs = "re"), data = M2_SRS_Project,
  method = "fREML", family = "gaussian")

引入timepoints变量后的出错模型 #

model2 = bam(depend ~ s(Age, bs = "tp") + s(indep, bs = "tp") + 
    s(Age, indep, bs = "tp", by = gender.x) + gender.x + timepoints + 
    s(Subject, bs="re") + s(Age, Subject, bs = "fs", m = 1), 
  data = M2_SRS_Project)

报错信息 #

Erreur dans h(simpleError(msg, call)) :
erreur d'evaluation de l'argument 'x' lors de la selection d'une methode pour la fonction 'chol' : tableaux de tailles inadéquates

翻译：在为函数'chol'选择方法时评估参数'x'出错：数组大小不合适

timepoints变量统计 #

• BL: 510
• FU1: 162
• FU2: 457
• FU3: 509

解决建议 #

1. 补全模型参数并调整随机效应结构
   model2遗漏了原模型中的method = "fREML"和family = "gaussian"参数，且将s(Age, Subject, bs = "re")改为了bs = "fs"，先尝试换回原随机效应结构并补全参数：

   model2 = bam(depend ~ s(Age, bs = "tp") + s(indep, bs = "tp") + 
       s(Age, indep, bs = "tp", by = gender.x) + gender.x + timepoints + 
       s(Subject, bs="re") + s(Age, Subject, bs = "re"), 
     data = M2_SRS_Project, method = "fREML", family = "gaussian")
   

2. 确认timepoints的编码格式
   确保timepoints是因子类型，若为字符型则转换：

   M2_SRS_Project$timepoints <- factor(M2_SRS_Project$timepoints, levels = c("BL", "FU1", "FU2", "FU3"))
   

   避免因编码混乱导致矩阵维度不匹配。

3. 排查变量间的奇异矩阵问题

   • 检查每个Subject对应的timepoints分布，若存在Subject仅对应单个timepoint，可能导致随机效应矩阵奇异：

     table(M2_SRS_Project$Subject, M2_SRS_Project$timepoints)
     

   • 检查gender.x各水平下的样本量，若某一水平数据过少，s(Age, indep, bs = "tp", by = gender.x)的平滑项矩阵可能无法正常计算。

4. 逐步简化模型定位问题
   从最简模型开始逐步添加变量，确定是哪个部分导致错误：

   # 第一步：基础模型
   model_test1 <- bam(depend ~ gender.x + timepoints, data = M2_SRS_Project, method = "fREML", family = "gaussian")
   # 第二步：添加单变量平滑项
   model_test2 <- bam(depend ~ gender.x + timepoints + s(Age, bs = "tp") + s(indep, bs = "tp"), data = M2_SRS_Project, method = "fREML", family = "gaussian")
   # 第三步：添加交互平滑项
   model_test3 <- bam(depend ~ gender.x + timepoints + s(Age, bs = "tp") + s(indep, bs = "tp") + s(Age, indep, bs = "tp", by = gender.x), data = M2_SRS_Project, method = "fREML", family = "gaussian")
   # 第四步：添加随机效应
   model_test4 <- bam(depend ~ gender.x + timepoints + s(Age, bs = "tp") + s(indep, bs = "tp") + s(Age, indep, bs = "tp", by = gender.x) + s(Subject, bs="re") + s(Age, Subject, bs = "re"), data = M2_SRS_Project, method = "fREML", family = "gaussian")
   

   每一步运行后检查是否报错，定位到具体出错的变量或项。

内容的提问来源于stack exchange，提问作者Ruben MIRANDA MARCOS