options(scipen = 999)

library(dplyr)
library(car)
library(lmtest)
library(stargazer)
library(sandwich)
library(lmtest)

setwd("C:/Users/irina/Documents/DND/EOHI/eohi1")

df <- read.csv("ehi1.csv")

data <- df %>%
  select(eohiDGEN_mean, ehi_global_mean, demo_sex, demo_age_1) %>%
  filter(demo_sex != "Prefer not to say")

str(data)
colSums(is.na(data))
sapply(data, class)

# Create dummy variable for sex (0 = Male, 1 = Female)
data$sex_dummy <- ifelse(data$demo_sex == "Female", 1, 0)

# Verify the dummy coding
print(table(data$demo_sex, data$sex_dummy))

#descriptives

# Descriptives for age
print(summary(data$demo_age_1))
print(sd(data$demo_age_1, na.rm = TRUE))

# Center demo_age_1 (subtract the mean)
data$age_centered <- data$demo_age_1 - mean(data$demo_age_1, na.rm = TRUE)

# Verify the centering
print(summary(data$age_centered))

# Descriptives for sex (frequency table)
print(table(data$demo_sex))
print(prop.table(table(data$demo_sex)))

# Descriptives for sex dummy variable
print(table(data$sex_dummy))

#### REGRESSION MODELS ####
# MODEL 1: Age only - EOHI
age_DGEN <- lm(eohiDGEN_mean ~ age_centered, data = data)
par(mfrow = c(2, 2))
plot(age_DGEN)
print(shapiro.test(residuals(age_DGEN)))
print(summary(age_DGEN))
print(AIC(age_DGEN))

# MODEL 1: Age only - EHI
age_domain <- lm(ehi_global_mean ~ age_centered, data = data)
par(mfrow = c(2, 2))
plot(age_domain)
print(shapiro.test(residuals(age_domain)))
print(summary(age_domain))
print(AIC(age_domain))

# MODEL 2: Sex only - EOHI
sex_DGEN <- lm(eohiDGEN_mean ~ sex_dummy, data = data)
par(mfrow = c(2, 2))
plot(sex_DGEN)
print(shapiro.test(residuals(sex_DGEN)))
print(summary(sex_DGEN))
print(AIC(sex_DGEN))
# P1 (res vs fitted) + P3 (scale location): test for homoscedasticity. relatively flat red line = homoscedasticity. relatively scattered points = homoscedasticity. this assumption is met.
# P2 (qq plot): test for normality. points scattered around a relatively straight line = normality. this assumption is violated but large sample is robust.
# P4 (residuals vs leverage): test for outliers. high leverage points = outliers. leverage > 2p/n. 
  # p = parameters; for this model p = 2 (intercept + sex_dummy). n = 1061 (removed prefer not to say). threshold = 2*2/1061 = 0.00377. maximum leverage in plot is ~ 0.002 therefore no points have concerning leverage.
# across the plots, there are 3 outliers: 258, 670, 872. this represents 0.28% of the data (much less than the acceptable threshold of 5%). therefore, analysis can proceed. 

# MODEL 2: Sex only - EHI
sex_domain <- lm(ehi_global_mean ~ sex_dummy, data = data)
par(mfrow = c(2, 2))
plot(sex_domain)
print(shapiro.test(residuals(sex_domain)))
print(summary(sex_domain))
print(AIC(sex_domain))

# MODEL 3: Age + Sex + Interaction - EOHI
interaction_DGEN <- lm(eohiDGEN_mean ~ age_centered + sex_dummy + age_centered:sex_dummy, data = data)
par(mfrow = c(2, 2))
plot(interaction_DGEN)
print(shapiro.test(residuals(interaction_DGEN)))
vif(interaction_DGEN)
print(summary(interaction_DGEN))
print(AIC(interaction_DGEN))

# MODEL 3: Age + Sex + Interaction - EHI
interaction_domain <- lm(ehi_global_mean ~ age_centered + sex_dummy + age_centered:sex_dummy, data = data)
par(mfrow = c(2, 2))
plot(interaction_domain)
print(shapiro.test(residuals(interaction_domain)))
vif(interaction_domain)
print(summary(interaction_domain))
print(AIC(interaction_domain))