循环从population data frame创建多data frame并批量提取逻辑回归beta₀

Got it, let's walk through how to solve this in R—whether you prefer the tidyverse workflow (clean, readable) or base R (no extra packages needed), I've got you covered.

Tidyverse Approach (Recommended for Readability) #

First, make sure you have the tidyverse package installed (run install.packages("tidyverse") if you haven't already). This method uses dplyr for grouping and purrr for applying functions to each group:

library(tidyverse)

# Replace `your_formula_here` with your actual logistic regression formula
# Example: if your outcome is `binary_outcome` and predictors are `var1` + `var2`, use binary_outcome ~ var1 + var2
beta0_results <- pop %>%
  group_by(replicate) %>%
  nest() %>%  # Nest each replicate's data into a list column
  mutate(
    # Fit logistic regression for each nested dataset
    model = map(data, ~ glm(your_formula_here, data = ., family = binomial(link = "logit"))),
    # Extract the intercept (beta₀) from each model
    beta0 = map_dbl(model, ~ coef(.)["(Intercept)"])
  ) %>%
  # Keep only the replicate ID and beta₀ values
  select(replicate, beta0) %>%
  ungroup() %>%
  # Sort by replicate to ensure order 1-110
  arrange(replicate)

Quick Explanation: #

• nest(): Packages all rows for each replicate into its own mini-data frame, making it easy to process groups individually.
• map(): Applies the glm() function to each nested data frame to fit the logistic regression.
• map_dbl(): Pulls out the intercept (labeled (Intercept) in model coefficients) as a numeric value, so we end up with a clean column of beta₀s.

Base R Approach #

If you'd rather avoid extra packages, here's how to do it with base R functions:

# Split the original data frame into a list of sub-data frames by replicate
split_pop <- split(pop, pop$replicate)

# Define a helper function to fit the model and extract beta₀
get_beta0 <- function(df) {
  # Again, replace `your_formula_here` with your actual formula
  model <- glm(your_formula_here, data = df, family = binomial)
  # Return the intercept value
  coef(model)["(Intercept)"]
}

# Apply the helper function to all sub-data frames
beta0_values <- sapply(split_pop, get_beta0)

# Convert results to a sorted data frame
beta0_results <- data.frame(
  replicate = as.integer(names(beta0_values)),
  beta0 = as.numeric(beta0_values)
) %>%
  arrange(replicate)

Important Notes #

• Replace the Formula: Don't forget to swap your_formula_here with your actual logistic regression formula. Your outcome variable must be binary (0/1 or TRUE/FALSE) for logistic regression to work.
• Handle Small Groups: If some replicate groups have too few observations (fewer than the number of predictors + 1), glm() will throw an error. To avoid breaking the whole process, use possibly() in the tidyverse approach to return NA for problematic groups:

  model = map(data, ~ possibly(glm, otherwise = NA)(your_formula_here, data = ., family = binomial))
  

内容的提问来源于stack exchange，提问作者wild west