Introduction and Setup

• StatQuest educational video from UNC Chapel Hill genetics department
• Topic: fitting line to data (least squares/linear regression)
• Starting with experimental data plotted on XY graph
• Goal: determine best line to show data trend

Horizontal Line Baseline

• Horizontal line through average y-value (3.5) as starting point
• Called point B (varies by dataset)
• Line equation: y = B where B = 3.5 for this dataset
• Worst fit but good reference for optimization discussion

Distance Measurement Method

• Measure fit by calculating distances from data points to line
• Point coordinates: (X1,Y1), (X2,Y2), etc.
• Distance calculations:
  • First point: B - Y1
  • Second point: B - Y2
  • Third point: B - Y3
  • Fourth point: B - Y4 (negative since Y4 > B)
  • Fifth point: very negative (point high above line)

Sum of Squared Residuals

• Problem: negative distances cancel positive ones
• Solution: square each distance term (avoids absolute value complexity)
• Sum of squared residuals = Σ(B - Yi)²
• For horizontal line example: 24.62
• Residuals = differences between real data and line values

Line Rotation Testing

• Rotating line slightly: sum of squared residuals = 18.7205 (improvement)
• Continued rotation keeps decreasing value
• Over-rotation: sum = 31.71 (worse fit)
• Sweet spot exists between horizontal and vertical

Generic Line Equation

• Standard form: y = ax + B
  • a = slope of line
  • B = y-intercept (where line crosses y-axis at x=0)
• Goal: find optimal values for a and B to minimize sum of squared residuals

Mathematical Framework

• Sum of squared residuals equation: Σ(aXi + B - Yi)²
• Method called "least squares" (minimizes sum of squares)
• Each term calculates distance between line value and observed value at Xi

Optimization Process

• Plot sum of squared residuals vs. line rotation
• Creates parabolic curve
• Minimum point represents optimal line rotation
• Use derivatives to find minimum (slope = 0 at optimal point)

3D Visualization

• 3D graph showing slope (a) vs intercept (B) vs sum of squares
• Z-axis represents intercept going into screen
• Can fix intercept (e.g., = 3) and vary slope to see effect
• Derivatives of both slope and intercept parameters locate optimal values

Key Concepts and Implementation

• Concept 1: minimize squared distance between observed values and line
• Concept 2: use derivatives to find where slope equals zero
• Final optimal line minimizes sum of squares (least squares)
• Computer-based solution (not solved by hand)
• Final equation for this example: y = 0.77x + 0.66