Regression leverage
Regression leverage hᵢᵢ measures how far an observation's predictor values lie from the centre of the regressor space.
FrameworkInfluential observations
What it is
See it move
Loading infographic...
The hat matrix H maps outcomes onto fitted values, ŷ = Hy, and its diagonal element hᵢᵢ is the leverage of observation i. Leverage satisfies 0 ≤ hᵢᵢ ≤ 1 and sums to k + 1, giving an average of (k+1)/n. A point with hᵢᵢ above twice that average sits far from the rest of the predictors and is flagged, though it only distorts the fit alongside a large residual.
The formula
LaTeX
Variables
- hat (projection) matrix (dimensionless)
- n × (k+1) design matrix of predictors including intercept column (dimensionless)
- leverage of observation i; i-th diagonal element of H (dimensionless)
ŷ = Hy. Leverage satisfies 0 ≤ hᵢᵢ ≤ 1 and Σhᵢᵢ = k + 1. Average leverage is (k+1)/n; observations with hᵢᵢ > 2(k+1)/n are commonly flagged.