如何针对数据框中特定日期绘制线性模型残差图？

Let's walk through refining your function to reliably plot model residuals for a specific date, without relying on dplyr. First, let's break down why the original approach had issues, then jump to the corrected code.

What Was Tripping Up the Original Code? #

• Your initial plot attempt compared a character string (from substr(MyData$Time,1,10)) to a POSIXct date object—R can't properly match these types, leading to incorrect filtering or errors.
• The first version of your plot_one_day function used character-string dates to filter a POSIXct Time column, which works unpredictably. It also pulled the title from the first row of filtered data, which would fail if no data matched the date.

Corrected plot_one_day Function #

Here's the robust, dplyr-free version:

plot_one_day <- function(LM, Day, DF) {
  # Attach residuals to the original data frame (aligns perfectly with na.exclude)
  DF$Residuals <- resid(LM)
  
  # Create proper POSIXct start/end times for the target day
  day_start <- as.POSIXct(paste(Day, "00:00:00"), format = "%Y-%m-%d %H:%M:%S")
  day_end <- as.POSIXct(paste(Day, "23:59:59"), format = "%Y-%m-%d %H:%M:%S")
  
  # Filter to the target day, removing rows with NA residuals
  filtered_data <- DF[DF$Time >= day_start & DF$Time <= day_end & !is.na(DF$Residuals), ]
  
  # Handle empty data to avoid plot errors
  if (nrow(filtered_data) == 0) {
    message(paste("Oops, no data available for", Day))
    return(invisible(NULL))
  }
  
  # Build the plot
  plot(Residuals ~ Time, 
       data = filtered_data,
       main = paste("Residuals for", Day),
       xlab = "Time of Day",
       ylab = "Model Residuals",
       xlim = c(day_start, day_end),
       ylim = range(filtered_data$Residuals),
       col = "darkslateblue", pch = 16)
  
  # Add a clear reference line at y=0
  abline(h = 0, col = "firebrick", lwd = 2)
  
  # Optional: Format x-axis to show hours (easier to read than full timestamps)
  axis.POSIXct(1, at = seq(day_start, day_end, by = "2 hours"), format = "%H:%M")
}

What Makes This Better? #

• Type-safe filtering: We convert the input Day string to POSIXct times, so comparisons with your Time column (also POSIXct) are 100% reliable.
• Residual alignment: Since you used na.action = na.exclude in your model, resid(LM) includes NA values for excluded rows—directly attaching it to DF ensures every residual lines up with its original row.
• Error protection: The function checks if filtered data is empty and sends a friendly message instead of crashing.
• Cleaner visualization: Added a y-axis label, formatted the x-axis to show hours, and tweaked colors/markers for readability.

Test It Out #

Call the function exactly as you intended:

plot_one_day(LM_1, "2015-05-07", MyData)

You'll get a clear plot of residuals for the specified day, with the red line at y=0 making it easy to spot positive vs. negative residuals.

内容的提问来源于stack exchange，提问作者Saalva