466 lines
18 KiB
R
466 lines
18 KiB
R
# Mixed ANOVA Analysis for DGEN Variables
|
||
# EOHI Experiment 2 Data Analysis - DGEN Level Analysis with TIME, DOMAIN, and INTERVAL factors
|
||
# Variables: DGEN_past_5_Pref, DGEN_past_5_Pers, DGEN_past_5_Val, DGEN_past_10_Pref, DGEN_past_10_Pers, DGEN_past_10_Val,
|
||
# DGEN_fut_5_Pref, DGEN_fut_5_Pers, DGEN_fut_5_Val, DGEN_fut_10_Pref, DGEN_fut_10_Pers, DGEN_fut_10_Val
|
||
|
||
# Load required libraries
|
||
library(tidyverse)
|
||
library(ez)
|
||
library(car)
|
||
library(afex) # For aov_ez (cleaner ANOVA output)
|
||
library(nortest) # For normality tests
|
||
library(emmeans) # For post-hoc comparisons
|
||
library(purrr) # For map functions
|
||
library(effsize) # For Cohen's d calculations
|
||
library(effectsize) # For effect size calculations
|
||
|
||
# Global options to remove scientific notation
|
||
options(scipen = 999)
|
||
|
||
# Set contrasts to sum for mixed ANOVA (necessary for proper interpretation)
|
||
options(contrasts = c("contr.sum", "contr.poly"))
|
||
|
||
setwd("C:/Users/irina/Documents/DND/EOHI/eohi2")
|
||
|
||
# Read the data
|
||
data <- read.csv("eohi2.csv")
|
||
|
||
# Display basic information about the dataset
|
||
print(paste("Dataset dimensions:", paste(dim(data), collapse = " x")))
|
||
print(paste("Number of participants:", length(unique(data$pID))))
|
||
|
||
# Verify the specific variables we need
|
||
required_vars <- c("DGEN_past_5_Pref", "DGEN_past_5_Pers", "DGEN_past_5_Val",
|
||
"DGEN_past_10_Pref", "DGEN_past_10_Pers", "DGEN_past_10_Val",
|
||
"DGEN_fut_5_Pref", "DGEN_fut_5_Pers", "DGEN_fut_5_Val",
|
||
"DGEN_fut_10_Pref", "DGEN_fut_10_Pers", "DGEN_fut_10_Val")
|
||
|
||
missing_vars <- required_vars[!required_vars %in% colnames(data)]
|
||
if (length(missing_vars) > 0) {
|
||
print(paste("Warning: Missing variables:", paste(missing_vars, collapse = ", ")))
|
||
} else {
|
||
print("All required DGEN variables found!")
|
||
}
|
||
|
||
# Define variable mapping for the three within-subjects factors
|
||
variable_mapping <- data.frame(
|
||
variable = required_vars,
|
||
TIME = c(rep("Past", 6), rep("Future", 6)),
|
||
DOMAIN = rep(c("Preferences", "Personality", "Values", "Preferences", "Personality", "Values"), 2),
|
||
INTERVAL = rep(c("5", "5", "5", "10", "10", "10"), 2),
|
||
stringsAsFactors = FALSE
|
||
)
|
||
|
||
# Variable mapping created
|
||
print("Variable mapping:")
|
||
print(variable_mapping)
|
||
|
||
# Efficient data pivoting using pivot_longer
|
||
long_data <- data %>%
|
||
select(pID, ResponseId, temporal_DO, interval_DO, all_of(required_vars)) %>%
|
||
pivot_longer(
|
||
cols = all_of(required_vars),
|
||
names_to = "variable",
|
||
values_to = "DGEN_SCORE"
|
||
) %>%
|
||
left_join(variable_mapping, by = "variable") %>%
|
||
# Convert to factors with proper levels
|
||
mutate(
|
||
TIME = factor(TIME, levels = c("Past", "Future")),
|
||
DOMAIN = factor(DOMAIN, levels = c("Preferences", "Personality", "Values")),
|
||
INTERVAL = factor(INTERVAL, levels = c("5", "10")),
|
||
pID = as.factor(pID),
|
||
temporal_DO = as.factor(temporal_DO),
|
||
interval_DO = as.factor(interval_DO)
|
||
) %>%
|
||
# Select final columns and remove any rows with missing values
|
||
select(pID, ResponseId, temporal_DO, interval_DO, TIME, DOMAIN, INTERVAL, DGEN_SCORE) %>%
|
||
filter(!is.na(DGEN_SCORE))
|
||
|
||
print(paste("Long data dimensions:", paste(dim(long_data), collapse = " x")))
|
||
print(paste("Number of participants:", length(unique(long_data$pID))))
|
||
print("Factor levels:")
|
||
print(paste("TIME:", paste(levels(long_data$TIME), collapse = ", ")))
|
||
print(paste("DOMAIN:", paste(levels(long_data$DOMAIN), collapse = ", ")))
|
||
print(paste("INTERVAL:", paste(levels(long_data$INTERVAL), collapse = ", ")))
|
||
print(paste("temporal_DO:", paste(levels(long_data$temporal_DO), collapse = ", ")))
|
||
print(paste("interval_DO:", paste(levels(long_data$interval_DO), collapse = ", ")))
|
||
|
||
# =============================================================================
|
||
# DESCRIPTIVE STATISTICS
|
||
# =============================================================================
|
||
|
||
# Overall descriptive statistics by TIME, DOMAIN, and INTERVAL
|
||
desc_stats <- long_data %>%
|
||
group_by(TIME, DOMAIN, INTERVAL) %>%
|
||
summarise(
|
||
n = n(),
|
||
mean = round(mean(DGEN_SCORE, na.rm = TRUE), 5),
|
||
variance = round(var(DGEN_SCORE, na.rm = TRUE), 5),
|
||
sd = round(sd(DGEN_SCORE, na.rm = TRUE), 5),
|
||
median = round(median(DGEN_SCORE, na.rm = TRUE), 5),
|
||
q1 = round(quantile(DGEN_SCORE, 0.25, na.rm = TRUE), 5),
|
||
q3 = round(quantile(DGEN_SCORE, 0.75, na.rm = TRUE), 5),
|
||
min = round(min(DGEN_SCORE, na.rm = TRUE), 5),
|
||
max = round(max(DGEN_SCORE, na.rm = TRUE), 5),
|
||
.groups = 'drop'
|
||
)
|
||
|
||
print("Descriptive statistics by TIME, DOMAIN, and INTERVAL:")
|
||
print(desc_stats)
|
||
|
||
# Descriptive statistics by between-subjects factors
|
||
desc_stats_by_between <- long_data %>%
|
||
group_by(temporal_DO, interval_DO, TIME, DOMAIN, INTERVAL) %>%
|
||
summarise(
|
||
n = n(),
|
||
mean = round(mean(DGEN_SCORE, na.rm = TRUE), 5),
|
||
variance = round(var(DGEN_SCORE, na.rm = TRUE), 5),
|
||
sd = round(sd(DGEN_SCORE, na.rm = TRUE), 5),
|
||
.groups = 'drop'
|
||
)
|
||
|
||
print("Descriptive statistics by between-subjects factors:")
|
||
print(desc_stats_by_between)
|
||
|
||
# Calculate mean differences for key comparisons
|
||
print("\n=== KEY MEAN DIFFERENCES ===")
|
||
|
||
# Past vs Future differences for each DOMAIN × INTERVAL combination
|
||
past_future_diffs <- long_data %>%
|
||
group_by(DOMAIN, INTERVAL, pID) %>%
|
||
summarise(
|
||
past_score = DGEN_SCORE[TIME == "Past"],
|
||
future_score = DGEN_SCORE[TIME == "Future"],
|
||
difference = past_score - future_score,
|
||
.groups = 'drop'
|
||
) %>%
|
||
group_by(DOMAIN, INTERVAL) %>%
|
||
summarise(
|
||
n = n(),
|
||
mean_diff = round(mean(difference, na.rm = TRUE), 5),
|
||
sd_diff = round(sd(difference, na.rm = TRUE), 5),
|
||
se_diff = round(sd(difference, na.rm = TRUE) / sqrt(n()), 5),
|
||
.groups = 'drop'
|
||
)
|
||
|
||
print("Past vs Future differences by DOMAIN × INTERVAL:")
|
||
print(past_future_diffs)
|
||
|
||
# 5 vs 10 interval differences for each TIME × DOMAIN combination
|
||
interval_diffs <- long_data %>%
|
||
group_by(TIME, DOMAIN, pID) %>%
|
||
summarise(
|
||
interval_5_score = DGEN_SCORE[INTERVAL == "5"],
|
||
interval_10_score = DGEN_SCORE[INTERVAL == "10"],
|
||
difference = interval_5_score - interval_10_score,
|
||
.groups = 'drop'
|
||
) %>%
|
||
group_by(TIME, DOMAIN) %>%
|
||
summarise(
|
||
n = n(),
|
||
mean_diff = round(mean(difference, na.rm = TRUE), 5),
|
||
sd_diff = round(sd(difference, na.rm = TRUE), 5),
|
||
se_diff = round(sd(difference, na.rm = TRUE) / sqrt(n()), 5),
|
||
.groups = 'drop'
|
||
)
|
||
|
||
print("\n5 vs 10 interval differences by TIME × DOMAIN:")
|
||
print(interval_diffs)
|
||
|
||
# =============================================================================
|
||
# ASSUMPTION TESTING
|
||
# =============================================================================
|
||
|
||
# Remove missing values for assumption testing
|
||
long_data_clean <- long_data[!is.na(long_data$DGEN_SCORE), ]
|
||
print(paste("Data after removing missing values:", paste(dim(long_data_clean), collapse = " x")))
|
||
|
||
# 1. Missing values check
|
||
missing_summary <- long_data %>%
|
||
group_by(TIME, DOMAIN, INTERVAL) %>%
|
||
summarise(
|
||
n_total = n(),
|
||
n_missing = sum(is.na(DGEN_SCORE)),
|
||
pct_missing = round(100 * n_missing / n_total, 2),
|
||
.groups = 'drop'
|
||
)
|
||
|
||
print("Missing values by TIME, DOMAIN, and INTERVAL:")
|
||
print(missing_summary)
|
||
|
||
# 2. Outlier detection
|
||
outlier_summary <- long_data_clean %>%
|
||
group_by(TIME, DOMAIN, INTERVAL) %>%
|
||
summarise(
|
||
n = n(),
|
||
mean = mean(DGEN_SCORE),
|
||
sd = sd(DGEN_SCORE),
|
||
q1 = quantile(DGEN_SCORE, 0.25),
|
||
q3 = quantile(DGEN_SCORE, 0.75),
|
||
iqr = q3 - q1,
|
||
lower_bound = q1 - 1.5 * iqr,
|
||
upper_bound = q3 + 1.5 * iqr,
|
||
n_outliers = sum(DGEN_SCORE < lower_bound | DGEN_SCORE > upper_bound),
|
||
.groups = 'drop'
|
||
)
|
||
|
||
print("Outlier summary (IQR method):")
|
||
print(outlier_summary)
|
||
|
||
# 3. Anderson-Darling normality test
|
||
normality_results <- long_data_clean %>%
|
||
group_by(TIME, DOMAIN, INTERVAL) %>%
|
||
summarise(
|
||
n = n(),
|
||
ad_statistic = ad.test(.data$DGEN_SCORE)$statistic,
|
||
ad_p_value = ad.test(.data$DGEN_SCORE)$p.value,
|
||
.groups = 'drop'
|
||
)
|
||
|
||
print("Anderson-Darling normality test results:")
|
||
# Round only the numeric columns
|
||
normality_results_rounded <- normality_results %>%
|
||
mutate(across(where(is.numeric), ~ round(.x, 5)))
|
||
print(normality_results_rounded)
|
||
|
||
# 4. Homogeneity of variance tests
|
||
# Test homogeneity across TIME within each DOMAIN × INTERVAL combination
|
||
homogeneity_time <- long_data_clean %>%
|
||
group_by(DOMAIN, INTERVAL) %>%
|
||
summarise(
|
||
levene_F = leveneTest(DGEN_SCORE ~ TIME)$`F value`[1],
|
||
levene_p = leveneTest(DGEN_SCORE ~ TIME)$`Pr(>F)`[1],
|
||
.groups = 'drop'
|
||
)
|
||
|
||
print("Homogeneity of variance across TIME within each DOMAIN × INTERVAL:")
|
||
print(homogeneity_time)
|
||
|
||
# Test homogeneity across DOMAIN within each TIME × INTERVAL combination
|
||
homogeneity_domain <- long_data_clean %>%
|
||
group_by(TIME, INTERVAL) %>%
|
||
summarise(
|
||
levene_F = leveneTest(DGEN_SCORE ~ DOMAIN)$`F value`[1],
|
||
levene_p = leveneTest(DGEN_SCORE ~ DOMAIN)$`Pr(>F)`[1],
|
||
.groups = 'drop'
|
||
)
|
||
|
||
print("Homogeneity of variance across DOMAIN within each TIME × INTERVAL:")
|
||
print(homogeneity_domain)
|
||
|
||
# Test homogeneity across INTERVAL within each TIME × DOMAIN combination
|
||
homogeneity_interval <- long_data_clean %>%
|
||
group_by(TIME, DOMAIN) %>%
|
||
summarise(
|
||
levene_F = leveneTest(DGEN_SCORE ~ INTERVAL)$`F value`[1],
|
||
levene_p = leveneTest(DGEN_SCORE ~ INTERVAL)$`Pr(>F)`[1],
|
||
.groups = 'drop'
|
||
)
|
||
|
||
print("Homogeneity of variance across INTERVAL within each TIME × DOMAIN:")
|
||
print(homogeneity_interval)
|
||
|
||
# =============================================================================
|
||
# HARTLEY'S F-MAX TEST WITH BOOTSTRAP CRITICAL VALUES
|
||
# =============================================================================
|
||
|
||
# More efficient bootstrap function for Hartley's F-max test
|
||
bootstrap_hartley_critical <- function(data, group_var, response_var, n_iter = 1000) {
|
||
# Get unique groups and their sample sizes
|
||
groups <- unique(data[[group_var]])
|
||
|
||
# Calculate observed variances for each group
|
||
observed_vars <- data %>%
|
||
dplyr::group_by(!!rlang::sym(group_var)) %>%
|
||
dplyr::summarise(var = var(!!rlang::sym(response_var), na.rm = TRUE), .groups = 'drop') %>%
|
||
dplyr::pull(var)
|
||
|
||
# Handle invalid variances
|
||
if(any(observed_vars <= 0 | is.na(observed_vars))) {
|
||
observed_vars[observed_vars <= 0 | is.na(observed_vars)] <- 1e-10
|
||
}
|
||
|
||
# Calculate observed F-max ratio
|
||
observed_ratio <- max(observed_vars) / min(observed_vars)
|
||
|
||
# Pre-allocate storage for bootstrap ratios
|
||
bootstrap_ratios <- numeric(n_iter)
|
||
|
||
# Get group data once
|
||
group_data_list <- map(groups, ~ {
|
||
group_data <- data[data[[group_var]] == .x, response_var]
|
||
group_data[!is.na(group_data)]
|
||
})
|
||
|
||
# Bootstrap with pre-allocated storage
|
||
for(i in 1:n_iter) {
|
||
# Bootstrap sample from each group independently
|
||
sample_vars <- map_dbl(group_data_list, ~ {
|
||
bootstrap_sample <- sample(.x, size = length(.x), replace = TRUE)
|
||
var(bootstrap_sample, na.rm = TRUE)
|
||
})
|
||
bootstrap_ratios[i] <- max(sample_vars) / min(sample_vars)
|
||
}
|
||
|
||
# Remove invalid ratios
|
||
valid_ratios <- bootstrap_ratios[is.finite(bootstrap_ratios) & !is.na(bootstrap_ratios)]
|
||
|
||
if(length(valid_ratios) == 0) {
|
||
stop("No valid bootstrap ratios generated")
|
||
}
|
||
|
||
# Calculate critical value (95th percentile)
|
||
critical_95 <- quantile(valid_ratios, 0.95, na.rm = TRUE)
|
||
|
||
# Return only essential information
|
||
return(list(
|
||
observed_ratio = observed_ratio,
|
||
critical_95 = critical_95,
|
||
n_valid_iterations = length(valid_ratios)
|
||
))
|
||
}
|
||
|
||
# Hartley's F-max test across between-subjects factors within each within-subjects combination
|
||
print("\n=== HARTLEY'S F-MAX TEST RESULTS ===")
|
||
set.seed(123) # For reproducibility
|
||
|
||
# Test across temporal_DO within each TIME × DOMAIN × INTERVAL combination
|
||
print("F-max test across temporal_DO within each TIME × DOMAIN × INTERVAL combination:")
|
||
hartley_temporal_results <- long_data_clean %>%
|
||
group_by(TIME, DOMAIN, INTERVAL) %>%
|
||
summarise(
|
||
hartley_result = list(bootstrap_hartley_critical(pick(temporal_DO, DGEN_SCORE), "temporal_DO", "DGEN_SCORE")),
|
||
.groups = 'drop'
|
||
) %>%
|
||
mutate(
|
||
observed_ratio = map_dbl(hartley_result, ~ .x$observed_ratio),
|
||
critical_95 = map_dbl(hartley_result, ~ .x$critical_95),
|
||
significant = observed_ratio > critical_95
|
||
) %>%
|
||
select(TIME, DOMAIN, INTERVAL, observed_ratio, critical_95, significant)
|
||
|
||
print(hartley_temporal_results)
|
||
|
||
# Test across interval_DO within each TIME × DOMAIN × INTERVAL combination
|
||
print("F-max test across interval_DO within each TIME × DOMAIN × INTERVAL combination:")
|
||
hartley_interval_results <- long_data_clean %>%
|
||
group_by(TIME, DOMAIN, INTERVAL) %>%
|
||
summarise(
|
||
hartley_result = list(bootstrap_hartley_critical(pick(interval_DO, DGEN_SCORE), "interval_DO", "DGEN_SCORE")),
|
||
.groups = 'drop'
|
||
) %>%
|
||
mutate(
|
||
observed_ratio = map_dbl(hartley_result, ~ .x$observed_ratio),
|
||
critical_95 = map_dbl(hartley_result, ~ .x$critical_95),
|
||
significant = observed_ratio > critical_95
|
||
) %>%
|
||
select(TIME, DOMAIN, INTERVAL, observed_ratio, critical_95, significant)
|
||
|
||
print(hartley_interval_results)
|
||
|
||
# =============================================================================
|
||
# MIXED ANOVA ANALYSIS
|
||
# =============================================================================
|
||
|
||
# Check data dimensions and structure
|
||
print(paste("Data size for ANOVA:", nrow(long_data_clean), "rows"))
|
||
print(paste("Number of participants:", length(unique(long_data_clean$pID))))
|
||
print(paste("Design factors: TIME (", length(levels(long_data_clean$TIME)), "), DOMAIN (",
|
||
length(levels(long_data_clean$DOMAIN)), "), INTERVAL (",
|
||
length(levels(long_data_clean$INTERVAL)), "), temporal_DO (",
|
||
length(levels(long_data_clean$temporal_DO)), "), interval_DO (",
|
||
length(levels(long_data_clean$interval_DO)), ")", sep = ""))
|
||
|
||
# Check for complete cases
|
||
complete_cases <- sum(complete.cases(long_data_clean))
|
||
print(paste("Complete cases:", complete_cases, "out of", nrow(long_data_clean)))
|
||
|
||
# Check if design is balanced
|
||
design_balance <- table(long_data_clean$pID, long_data_clean$TIME, long_data_clean$DOMAIN, long_data_clean$INTERVAL)
|
||
if(all(design_balance %in% c(0, 1))) {
|
||
print("Design is balanced: each participant has data for all TIME × DOMAIN × INTERVAL combinations")
|
||
} else {
|
||
print("Warning: Design is unbalanced")
|
||
print(summary(as.vector(design_balance)))
|
||
}
|
||
|
||
# =============================================================================
|
||
# MIXED ANOVA WITH SPHERICITY CORRECTIONS
|
||
# =============================================================================
|
||
|
||
print("\n=== MIXED ANOVA RESULTS (with sphericity corrections) ===")
|
||
|
||
# Mixed ANOVA using ezANOVA with automatic sphericity corrections
|
||
# Between-subjects: temporal_DO (2 levels) × interval_DO (2 levels)
|
||
# Within-subjects: TIME (2 levels: Past, Future) × DOMAIN (3 levels: Preferences, Personality, Values) × INTERVAL (2 levels: 5, 10)
|
||
mixed_anova_model <- ezANOVA(data = long_data_clean,
|
||
dv = DGEN_SCORE,
|
||
wid = pID,
|
||
between = .(temporal_DO, interval_DO),
|
||
within = .(TIME, DOMAIN, INTERVAL),
|
||
type = 3,
|
||
detailed = TRUE)
|
||
|
||
print("ANOVA Results:")
|
||
anova_output <- mixed_anova_model$ANOVA
|
||
rownames(anova_output) <- NULL # Reset row numbers to be sequential
|
||
print(anova_output)
|
||
|
||
# Show Mauchly's test for sphericity
|
||
print("\nMauchly's Test of Sphericity:")
|
||
print(mixed_anova_model$Mauchly)
|
||
|
||
# Show sphericity-corrected results (Greenhouse-Geisser and Huynh-Feldt)
|
||
if(!is.null(mixed_anova_model$`Sphericity Corrections`)) {
|
||
print("\nGreenhouse-Geisser and Huynh-Feldt Corrections:")
|
||
print(mixed_anova_model$`Sphericity Corrections`)
|
||
|
||
# Extract and display corrected degrees of freedom
|
||
print("\n=== CORRECTED DEGREES OF FREEDOM ===")
|
||
|
||
sphericity_corr <- mixed_anova_model$`Sphericity Corrections`
|
||
anova_table <- mixed_anova_model$ANOVA
|
||
|
||
corrected_df <- data.frame(
|
||
Effect = sphericity_corr$Effect,
|
||
Original_DFn = anova_table$DFn[match(sphericity_corr$Effect, anova_table$Effect)],
|
||
Original_DFd = anova_table$DFd[match(sphericity_corr$Effect, anova_table$Effect)],
|
||
GG_DFn = anova_table$DFn[match(sphericity_corr$Effect, anova_table$Effect)] * sphericity_corr$GGe,
|
||
GG_DFd = anova_table$DFd[match(sphericity_corr$Effect, anova_table$Effect)] * sphericity_corr$GGe,
|
||
HF_DFn = anova_table$DFn[match(sphericity_corr$Effect, anova_table$Effect)] * sphericity_corr$HFe,
|
||
HF_DFd = anova_table$DFd[match(sphericity_corr$Effect, anova_table$Effect)] * sphericity_corr$HFe,
|
||
GG_epsilon = sphericity_corr$GGe,
|
||
HF_epsilon = sphericity_corr$HFe
|
||
)
|
||
|
||
print(corrected_df)
|
||
|
||
print("\n=== CORRECTED F-TESTS ===")
|
||
for(i in seq_len(nrow(corrected_df))) {
|
||
effect <- corrected_df$Effect[i]
|
||
f_value <- anova_table$F[match(effect, anova_table$Effect)]
|
||
|
||
print(sprintf("\n%s:", effect))
|
||
print(sprintf(" Original: F(%d, %d) = %.3f",
|
||
corrected_df$Original_DFn[i], corrected_df$Original_DFd[i], f_value))
|
||
print(sprintf(" GG-corrected: F(%.2f, %.2f) = %.3f, p = %.6f",
|
||
corrected_df$GG_DFn[i], corrected_df$GG_DFd[i], f_value, sphericity_corr$`p[GG]`[i]))
|
||
print(sprintf(" HF-corrected: F(%.2f, %.2f) = %.3f, p = %.6f",
|
||
corrected_df$HF_DFn[i], corrected_df$HF_DFd[i], f_value, sphericity_corr$`p[HF]`[i]))
|
||
}
|
||
} else {
|
||
print("\nNote: Sphericity corrections not needed (sphericity assumption met)")
|
||
}
|
||
|
||
# =============================================================================
|
||
# EFFECT SIZES (GENERALIZED ETA SQUARED)
|
||
# =============================================================================
|
||
|
||
print("\n=== EFFECT SIZES (GENERALIZED ETA SQUARED) ===")
|
||
|
||
# Extract generalized eta squared from ezANOVA (already calculated)
|
||
effect_sizes <- mixed_anova_model$ANOVA[, c("Effect", "ges")]
|
||
effect_sizes$ges <- round(effect_sizes$ges, 5)
|
||
print("Generalized Eta Squared:")
|
||
print(effect_sizes) |