如何让modelsummary()对无统计量的brms模型跳过统计量展示而非报错？

问题场景 #

当同时使用modelsummary()展示线性回归模型（lm）和贝叶斯模型（brms）时，若为线性模型添加标准误、t值、p值等统计量，会因贝叶斯模型无对应统计量而报错。

模型代码 #

models <- list(
  "Linear" = lm(outcome ~ week * food, data = df ),
  "Bayesian" = brm(outcome ~ s(week, k = 4, fx = TRUE, by = food) + food, data = df, family = "zero_one_inflated_beta")
)

报错代码及信息 #

执行以下代码尝试添加统计量：

modelsummary(models,
             estimate = "estimate[{conf.low}, {conf.high}]",
             statistic = c("Std.Error" = "std.error", 
                           "t-value" = "statistic", 
                           "p-value" = "p.value"))

触发错误：

Error: std.error is not available. The estimate and statistic arguments must correspond to column names in the output of this command: get_estimates(model)

解决方案 #

通过为不同模型单独指定统计量的方式，让贝叶斯模型跳过统计量展示：

方法1：直接在statistic参数中按模型设置

modelsummary(models,
             estimate = "estimate[{conf.low}, {conf.high}]",
             statistic = list(
               "Linear" = c("Std.Error" = "std.error", 
                            "t-value" = "statistic", 
                            "p-value" = "p.value"),
               "Bayesian" = NULL
             ))

方法2：使用statistic_override参数自定义

# 定义线性模型需要的统计量
lm_statistics <- list(
  "Std.Error" = ~ get_estimates(.)$std.error,
  "t-value" = ~ get_estimates(.)$statistic,
  "p-value" = ~ get_estimates(.)$p.value
)

# 贝叶斯模型不设置统计量
brms_statistics <- NULL

modelsummary(models,
             estimate = "estimate[{conf.low}, {conf.high}]",
             statistic_override = list("Linear" = lm_statistics, "Bayesian" = brms_statistics))

两种方法都能实现：线性模型展示指定的统计量，贝叶斯模型仅展示估计值与置信区间，避免报错。

内容的提问来源于stack exchange，提问作者nerd