For this project, I chose the dataset “Social Media and Mental Health Balance” from Kaggle. The data consists of 500 rows and 10 columns, including demographic information, screen time, sleep quality, and other factors. By performing data manipulation and statistical analysis on this dataset, we can explore the relationship between social media usage and mental wellbeing.

The goal of this project is to identify patterns that highlight the influence that social media has on factors affecting mental health. This topic is particularly meaningful to me, as I use social media daily for both personal and professional purposes and have observed both its positive and negative effects on my own wellbeing.

Link to dataset: https://www.kaggle.com/datasets/ayeshaimran123/social-media-and-mental-health-balance/data

The Social Media and Mental Health Balance dataset has the following columns:

• User_ID
• Age
• Gender
• Daily_Screen_Time(hrs)
• Sleep_Quality(1-10)
• Stress_Level(1-10)
• Days_Without_Social_Media
• Exercise_Frequency(week)
• Social_Media_Platform
• Happiness_Index(1-10)

• Load and clean the data

# load libraries
library(ggplot2)
library(plotly)
library(dplyr)
library(knitr)

# load dataset + clean by renaming messy column names (data wrangling)
data <- read.csv("Mental_Health_and_Social_Media_Balance_Dataset.csv") %>%
  rename(
    Screen_Time = Daily_Screen_Time.hrs.,
    Sleep_Quality = Sleep_Quality.1.10.,
    Stress_Level = Stress_Level.1.10.,
    Days_Off_Social_Media = Days_Without_Social_Media,
    Exercise_Freq = Exercise_Frequency.week.,
    Happiness = Happiness_Index.1.10.,
    Platform = Social_Media_Platform
  )

• Get statistics

# summary statistics
summary(data)
numeric_data <- data %>% 
  select(Age, Screen_Time, Sleep_Quality, 
         Stress_Level, Days_Off_Social_Media, 
         Exercise_Freq, Happiness)

# calculate mean, min, max
stats_summary <- data.frame(
  Variable = c("Age", "Daily Screen Time (hrs)", "Sleep Quality (1-10)",
               "Stress Level (1-10)", "Days Without Social Media",
               "Exercise Frequency (week)", "Happiness Index (1-10)"),
  Mean = sapply(numeric_data, mean, na.rm = TRUE),
  Min = sapply(numeric_data, min, na.rm = TRUE),
  Max = sapply(numeric_data, max, na.rm = TRUE)
)
stats_summary

gender_counts <- as.data.frame(table(data$Gender))
colnames(gender_counts) <- c("Gender", "Number of Users")

kable(gender_counts, col.names = c("Gender", "Number of Users"), align = c("c", "c"))

platform_counts <- as.data.frame(table(data$Platform))
colnames(platform_counts) <- c("Social Media Platform", "Number of Users")

kable(platform_counts,
col.names = c("Social Media Platform", "Number of Users"), align = c("c", "c"))

Since I based my HW3 on Hypothesis Testing (particularly z-test) and found it interesting, I decided to explore this dataset further by using different(and more complex) types of hypothesis tests for the various problems I wanted to find the conclusions to in this dataset.

Different types of hypothesis tests include (one-sample and two-sample):

• z-test
• t-test
• F-test
• \(\chi^2\)-test
• linear regression

• \(H_0\): There is no significant correlation between age and stress level.
• \(H_1\): There is a significant positive correlation between age and stress level.
• Statistical Test: Linear regression

model_age <- lm(Stress_Level ~ Age, data = data)
slope <- round(coef(model_age)[2], 4)
p_val <- summary(model_age)$coefficients[2, 4]
r_squared <- round(summary(model_age)$r.squared, 3)
# extract key values
cat("\nRegression equation: Stress =", 
    round(coef(model_age)[1], 2), "+", 
    round(coef(model_age)[2], 4), "* Age\n")
# scatter plot code (reduced spaces to fit on slide)
ggplot(data, aes(x = Age, y = Stress_Level)) +
  geom_point(alpha = 0.6, color = "#0B968E", size = 2.5) +geom_smooth(method = "lm", color = "black", 
              se = TRUE, linewidth = 1) +
  labs(title = "Linear Regression: Age vs Stress Level",
       subtitle = paste0("Slope = ", slope, ", p = ", format.pval(p_val, digits = 3),", R² = ", r_squared),
       x = "Age (years)", y = "Stress Level (1-10)") +
  theme_minimal(base_size = 16) + theme(plot.title = 
        element_text(face = "bold", hjust = 0.5),plot.subtitle = element_text(hjust = 0.5))

• \(H_0\): Gender and social media platform preference are independent (no association).
• \(H_1\): There is a significant association between gender and social media platform preference.
• Statistical Test: \(\chi^2\)-test

# filter to include only Male and Female 
data_filtered <- data %>%
  filter(Gender %in% c("Male", "Female"))
# create contingency table
cont_table <- table(data_filtered$Gender, data_filtered$Platform)
# chi-square test
chi_test <- chisq.test(cont_table)
# important values
cat("\nChi-square statistic:", round(chi_test$statistic, 3), "Degrees of freedom:", chi_test$parameter, "P-value:", 
    format.pval(chi_test$p.value, digits = 4), "Sample size after filtering:", nrow(data_filtered), "users")

## 
## Chi-square statistic: 7.182 Degrees of freedom: 5 P-value: 0.2075 Sample size after filtering: 477 users

• \(H_0\): There is no significant difference in happiness index between individuals with low and high screen time.
• \(H_1\): Individuals with lower screen time have significantly higher happiness scores than those with higher screen time.
• Statistical Test: Two-sample t-test

# create screen time categories
data <- data %>%
  mutate(Screen_Category = case_when(
    Screen_Time < 4 ~ "Low",
    Screen_Time >= 4 & Screen_Time <= 6 ~ "Medium",
    Screen_Time > 6 ~ "High"
  ))

# low and high only for this test
low_group <- data %>% filter(Screen_Category == "Low")
high_group <- data %>% filter(Screen_Category == "High")

# two-sample t-test
t_test <- t.test(low_group$Happiness, high_group$Happiness,
                 alternative = "two.sided")

cat("P-value:", format.pval(t_test$p.value, digits = 4))

## P-value: < 2.2e-16

# display results
cat("Mean Happiness (Low Screen Time):", 
    round(mean(low_group$Happiness), 2), "\n")

## Mean Happiness (Low Screen Time): 9.74

cat("Mean Happiness (High Screen Time):", 
    round(mean(high_group$Happiness), 2), "\n")

## Mean Happiness (High Screen Time): 7.25

cat("T-statistic:", round(t_test$statistic, 3), "\n")

## T-statistic: 21.303

• \(H_0\): There is no significant relationship between daily screen time and the combination of sleep quality and stress level.
• \(H_1\): Higher screen time is associated with lower sleep quality and higher stress levels.
• Statistical Test: Multiple linear regression

# multiple linear regression models
model_sleep <- lm(Sleep_Quality ~ Screen_Time, data = data) # Sleep Quality ~ Screen Time
model_stress <- lm(Stress_Level ~ Screen_Time, data = data) # Stress Level ~ Screen Time

# display results
cat(
"Sleep Quality - Coefficient:", round(coef(model_sleep)[2], 4),
", P-value:", format.pval(summary(model_sleep)$coefficients[2,4], digits = 4),
", R^2:", round(summary(model_sleep)$r.squared, 3), "\n",
"Stress Level - Coefficient:", round(coef(model_stress)[2], 4),
", P-value:", format.pval(summary(model_stress)$coefficients[2,4], digits = 4),
", R^2:", round(summary(model_stress)$r.squared, 3), "\n"
)

## Sleep Quality - Coefficient: -0.6692 , P-value: < 2.2e-16 , R^2: 0.576 
##  Stress Level - Coefficient: 0.6581 , P-value: < 2.2e-16 , R^2: 0.547