GLM模型拟合报错‘NA/NaN/Inf in 'y'’排查求助

Hey there, let's work through this GLM error you're facing. Even though you confirmed there are no NA values in your dataset, the "NA/NaN/Inf in 'y'" error can pop up from a few hidden issues—let's break them down one by one:

• Check for Inf/NaN in your response variable
  The table(is.na()) check only catches NA values, not infinite (Inf) or not-a-number (NaN) values. Your response variable rope.directional.change might have these hidden issues. Run these quick checks to confirm:

  # Check for infinite values in the response
  any(is.infinite(rope_complete$rope.directional.change))
  # Check for NaN values in the response
  any(is.nan(rope_complete$rope.directional.change))
  

  If either returns TRUE, you'll need to clean those values (e.g., filter out rows with Inf/NaN or correct the source data).

• Investigate your offset term
  Your offset uses log(rope_complete$rope.Time.of.the.shark.in.the.video)—and here's a common gotcha: if any value in rope.Time.of.the.shark.in.the.video is 0, log(0) returns -Inf, which will throw off the model. Check for zero values with:

  any(rope_complete$rope.Time.of.the.shark.in.the.video == 0)
  

  If you find zeros, you have a few fixes:

  1. Filter out rows where the time is 0 (if those observations aren't meaningful for your analysis)
  2. Add a tiny constant to the time values before taking the log (e.g., log(rope.Time.of.the.shark.in.the.video + 0.01) to avoid log(0))

• Verify your code is complete and clean
  The code snippet you shared cuts off mid-offset definition: offset( log(rope_complete$rope.Tim.... Make sure you've written the full offset term (it should reference rope.Time.of.the.shark.in.the.video properly). Also, using the data parameter will make your code cleaner and less prone to typos:

  # Corrected, cleaner code example (with offset adjustment if needed)
  directional_turn_fit <- glm(rope.directional.change ~ rope.X...Sound + offset(log(rope.Time.of.the.shark.in.the.video + 0.01)),
                              data = rope_complete)
  

• Double-check variable types
  Ensure rope.directional.change is a numeric variable (not a factor or character string). A wrong variable type can sometimes trigger unexpected errors. Confirm with:

  class(rope_complete$rope.directional.change)
  

Start with checking the offset term first—odds are that zero time values are causing the Inf values that break your model.

内容的提问来源于stack exchange，提问作者Jessi