Student Performance Analysis Using SAS

This project investigates how different demographic and educational factors affect student performance.

Project Overview

This project investigates factors influencing student academic performance using data from two Portuguese schools. The primary outcome variable is the final grade (G3) in a Portuguese language course. We employed descriptive analysis, multiple linear regression, Poisson log-linear modeling, and cluster analysis to identify significant predictors of student success.

Team

• Shuoyuan Gao: Descriptive analysis of categorical variables, Cluster analysis
• Yufei Duan: Numeric variable analysis, Linear modeling, Variable selection
• Zhongwen Shen: Multiple regression modeling, Model evaluation

Methodology

• Data cleaning: Removed students with a final grade (G3) of 0.
• Exploratory analysis: Examined both numeric (e.g., age, study time, parental education) and categorical variables (e.g., gender, family support).
• Modeling:
  • Multiple linear regression to predict G3.
  • Over-dispersed Poisson log-linear model.
  • Hierarchical cluster analysis based on selected predictors.

Key Findings

• Significant predictors of G3:
  • Failures: Each additional failure decreased G3 by ~1.72 points.
  • Study time: Positively associated with G3.
  • Mother’s education: Higher education linked with better student outcomes.
  • Workday alcohol consumption (Dalc): Negatively associated with G3.

• Best model: The linear regression model had the best balance of fit and complexity (Adjusted R² = 0.3461).

• Clustering revealed distinct student profiles based on performance-related features.

Code Part

This study identifies critical academic and socio-demographic factors affecting student performance. The analysis offers valuable insights for educators and policymakers aiming to support diverse student needs.

🔍 Methods

The analysis was performed using SAS. Below is a snippet of the SAS code used in the analysis:

proc import datafile="/path/student-marks.csv" out=student dbms=csv replace;
   getnames=yes;
run;

data student_clean;
   set student;
   if gender = "female" then gender_num = 1;
   else gender_num = 0;
run;

proc logistic data=student_clean;
   model passed(event='1') = gender_num reading_score writing_score / clodds=wald;
run;

📎 Full Report

📄 View Full Report (PDF)

Reference

Cortez, Paulo. (2014). Student Performance. UCI Machine Learning Repository. https://doi.org/10.24432/C5TG7T