Quantifying Side Bias Through ELO Regression

Code

The following Python code demonstrates how we extracted the necessary attributes from our DynamoDB table.

import boto3
from decimal import Decimal
import numpy as np
from scipy import stats

def calculate_aff_bias():
    # Initialize the DynamoDB resource
    dynamodb = boto3.resource('dynamodb', region_name='us-east-1') # Change to your region
    table = dynamodb.Table('Debate_Rounds_2025-26_HS')

    print("Fetching data from DynamoDB...")
    
    # Scanning the table
    # Note: If you have > 1MB of data, you'd need to handle pagination
    response = table.scan(
        ProjectionExpression="AffirmativeWin, PreliminaryRound"
    )
    items = response.get('Items', [])

    # Counters
    stats_dict = {
        "Prelim": {"wins": 0, "total": 0},
        "Elim": {"wins": 0, "total": 0}
    }

    for item in items:
        # Determine if it's a prelim or elim
        is_prelim = item.get('PreliminaryRound') == 'Yes'
        category = "Prelim" if is_prelim else "Elim"
        
        # Determine if Aff won
        aff_won = item.get('AffirmativeWin') == 'Yes'
        
        stats_dict[category]["total"] += 1
        if aff_won:
            stats_dict[category]["wins"] += 1

    # Print Results
    print("\n--- Debate Statistics 2025-26 ---")
    total_all = len(items)
    
    for cat, data in stats_dict.items():
        if data["total"] > 0:
            win_rate = (data["wins"] / data["total"]) * 100
            bias = win_rate - 50
            print(f"{cat} Rounds:")
            print(f"  Total: {data['total']}")
            print(f"  Aff Win Rate: {win_rate:.2f}%")
            print(f"  Aff Bias: {bias:+.2f}%") # Shows + or - prefix
        else:
            print(f"{cat} Rounds: No data found.")

    print(f"\nProcessed {total_all} total rounds.")


def calculate_elo_regression():
    dynamodb = boto3.resource('dynamodb', region_name='us-east-1') # Change to your region
    
    # 1. Fetch all Teams to get their ELOs
    team_table = dynamodb.Table('Team_Info_2025-26_HS')
    teams_items = team_table.scan(ProjectionExpression="SchoolTeamCode, OverallELO").get('Items', [])
    elo_map = {item['SchoolTeamCode']: float(item['OverallELO']) for item in teams_items}

    # 2. Fetch all Rounds
    round_table = dynamodb.Table('Debate_Rounds_2025-26_HS')
    rounds = round_table.scan(ProjectionExpression="AffirmativeSchool, NegativeSchool, AffirmativeWin").get('Items', [])

    x_expected = [] # ELO Predicted Win Probabilities
    y_actual = []   # Actual Outcomes (1 for Win, 0 for Loss)

    for r in rounds:
        aff_code = r.get('AffirmativeSchool')
        neg_code = r.get('NegativeSchool')
        
        if aff_code in elo_map and neg_code in elo_map:
            ra = elo_map[aff_code]
            rb = elo_map[neg_code]
            
            # Standard ELO Expected Win Probability for Aff
            expected_win = 1 / (1 + 10 ** ((rb - ra) / 400))
            actual_win = 1 if r.get('AffirmativeWin') == 'Yes' else 0
            
            x_expected.append(expected_win)
            y_actual.append(actual_win)

    # 3. Linear Regression
    # We are regressing Actual Outcome vs Expected Probability
    slope, intercept, r_value, p_value, std_err = stats.linregress(x_expected, y_actual)

    print(f"--- ELO Regression Analysis ---")
    print(f"Sample Size: {len(y_actual)} rounds")
    print(f"Slope: {slope:.4f} (Ideal: 1.0)")
    print(f"Intercept: {intercept:.4f} (Ideal: 0.0)")
    print(f"R-Squared: {r_value**2:.4f}")
    
    # Interpretation
    aff_advantage = intercept * 100
    print(f"\nStatic Aff Bias: {aff_advantage:+.2f}%")
    if intercept > 0:
        print("Interpretation: Even when ELO predicts a 0% win chance, Aff starts with a baseline advantage.")
    
    return slope, intercept

if __name__ == "__main__":
    calculate_aff_bias()
    calculate_elo_regression()

  

This raw data was then cross-referenced with our team ELO table to calculate the residuals used in our final regression model.