编写多元二项Logistic回归分析循环及对应数据框技术咨询

Got it, let's tackle this problem step by step. You want to run multiple binary logistic regression models in a loop using your data frame, where disease is the binary outcome variable. Here's how you can do it properly in R:

First, let's confirm your data frame is set up correctly (I'll include it here for reference):

df <- data.frame( 
  disease = c(0,1,0,1), 
  var1 = c(0,1,2,0), 
  var2 =c(0,1,2,0), 
  var3 = c(0,1,2,0), 
  var40 = c(0,1,2,0), 
  Bi = c(0,1,0,1), 
  gender = c(1,0,1,0), 
  P1 = c(-0.040304832,0.006868288,0.002663759,0.020251087), 
  P2 = c(0.010566526,0.002663759,0.017480721,-0.008685749), 
  P3 = c(-0.008685749,0.020251087,-0.040304832,0.002663759), 
  P4 = c(0.017480721,0.024306667,0.002663759,0.010566526), 
  stringsAsFactors = FALSE
)

Option 1: Loop Through Single Predictors + Control Variables #

If you want to run a model for each predictor while including a fixed set of control variables (e.g., Bi and gender), this code will work perfectly:

# Define control variables and list of predictors to iterate over
control_vars <- c("Bi", "gender")
predictors <- setdiff(names(df), c("disease", control_vars))

# Initialize a list to store all model results (clean way to keep track)
model_list <- list()

# Loop through each predictor
for (pred in predictors) {
  # Build the model formula dynamically
  formula_str <- paste("disease ~", pred, "+", paste(control_vars, collapse = " + "))
  formula <- as.formula(formula_str)
  
  # Fit the binary logistic regression model
  model <- glm(formula, data = df, family = binomial(link = "logit"))
  
  # Store the model in the list with a descriptive name
  model_list[[paste0("model_", pred)]] <- model
  
  # Optional: Print summary for each model as it runs (great for quick checks)
  cat("\n--- Summary for Model with Predictor:", pred, "---\n")
  print(summary(model))
}

# Later, you can access individual models like this: model_list$model_var1

Option 2: Loop Through Multivariate Model Variations #

If you want to test different full multivariate models (e.g., exclude one predictor at a time from the full model), here's how to do that:

# Get all predictor variables (excluding the outcome)
all_predictors <- setdiff(names(df), "disease")

# Initialize list to store these models
full_model_list <- list()

# First, fit the full model with all predictors
full_formula <- as.formula(paste("disease ~", paste(all_predictors, collapse = " + ")))
full_model_list[["full_model"]] <- glm(full_formula, data = df, family = binomial(link = "logit"))
cat("\n--- Summary for Full Multivariate Model ---\n")
print(summary(full_model_list[["full_model"]]))

# Now loop to fit models excluding one predictor at a time
for (pred in all_predictors) {
  # Create a subset of predictors excluding the current one
  subset_predictors <- setdiff(all_predictors, pred)
  formula_str <- paste("disease ~", paste(subset_predictors, collapse = " + "))
  formula <- as.formula(formula_str)
  
  # Fit the model
  model <- glm(formula, data = df, family = binomial(link = "logit"))
  full_model_list[[paste0("model_excluding_", pred)]] <- model
  
  # Print summary
  cat("\n--- Summary for Model Excluding:", pred, "---\n")
  print(summary(model))
}

Quick Notes to Keep in Mind #

• Categorical Variables: Your categorical variables like Bi and gender are already coded as 0/1, which works for glm. If you had unordered factors with more levels, convert them to factors first with df$var <- as.factor(df$var) to generate proper dummy variables.
• Sample Size: Heads up—your dataset only has 4 rows, which is way too small for reliable logistic regression results. This code is for demonstration; you'll need much more data in practice to get meaningful coefficients and p-values.
• Result Storage: Using lists to store models is efficient—you can easily extract coefficients, p-values, or run further analyses on all models at once (e.g., lapply(model_list, coef) to get all coefficients in one go).

内容的提问来源于stack exchange，提问作者Maneesh Mishra