如何基于总均值±2SD筛选排除被试RT数据？（R语言）

Hey there! Let's work through this step by step to get your subject exclusion logic working correctly. First, I spotted a small misstep in your current code for calculating the grand mean and standard deviation—you’re using sd(sd_sub), which computes the standard deviation of subject-level RT standard deviations, but what we actually need is the standard deviation of the subject-level average RTs to set our exclusion threshold.

Here's a complete, corrected workflow using dplyr:

Step 1: Load required library and define your raw data #

First, make sure dplyr is loaded, and structure your raw data (I’ve included your sample data here for completeness):

library(dplyr)

# Raw sample data
df <- tibble(
  Subj = rep(1:3, each = 10),
  Trial = rep(1:10, 3),
  factor = c("Early","Late","Early","Late","Early","Early","Late","Late","Late","Early",
             "Early","Late","Early","Late","Early","Early","Late","Late","Late","Early",
             "Early","Late","Early","Late","Early","Early","Late","Late","Late","Early"),
  rt = c(324,405,1093,738,310,389,1050,782,513,401,
         920,1230,1509,1456,2341,900,1100,1329,1189,1030,
         999,499,1003,389,356,365,1234,345,300,402)
)

Step 2: Calculate subject-level statistics #

Compute each subject’s average RT (and other useful stats) as you started:

subj_stats <- df %>%
  group_by(Subj) %>%
  summarise(
    N = n(),
    avgrt = mean(rt),
    sd_sub = sd(rt),
    min_rt = min(rt),
    median_rt = median(rt),
    max_rt = max(rt)
  ) %>%
  ungroup() # Ungroup to avoid issues with subsequent calculations

Step 3: Compute grand mean and threshold values #

Calculate the overall mean of subject-level average RTs, plus the 2-standard-deviation threshold for exclusion:

# Get grand mean and grand standard deviation (of subject averages)
grand_stats <- subj_stats %>%
  summarise(
    grand_mean = mean(avgrt),
    grand_sd = sd(avgrt)
  )

# Extract threshold values for filtering
grand_mean <- grand_stats$grand_mean
two_sd_range <- 2 * grand_stats$grand_sd

Step 4: Filter valid subjects #

Identify which subjects have average RTs within the grand_mean ± 2*grand_sd range:

valid_subjs <- subj_stats %>%
  filter(avgrt >= grand_mean - two_sd_range,
         avgrt <= grand_mean + two_sd_range)

# View which subjects are retained
valid_subjs

Step 5: Get clean dataset with valid subjects only #

Filter your original raw data to keep only the subjects that passed the exclusion criteria:

data_clean <- df %>%
  filter(Subj %in% valid_subjs$Subj)

# Inspect the cleaned data
data_clean

Key Notes #

• The critical fix here is using sd(avgrt) instead of sd(sd_sub)—we need the spread of subject averages, not the spread of subject-level RT variability, to set our exclusion threshold.
• This workflow ensures you exclude entire subjects whose overall average RT falls outside the 2SD range relative to the grand mean, regardless of individual trial or factor-level data.

内容的提问来源于stack exchange，提问作者arvag19