A leverage point in linear regression is an observation having unusually large leverage. If 
is the full-column-rank design
matrix, the diagonal entries 
of
 |
measure leverage. Their average is 
, where 
is the number of fitted regression
coefficients and 
is the number of observations, so unusually large 
values identify high-leverage points.
Leverage describes the potential to affect a fit, not the actual effect. A high-leverage point that closely follows the fitted relationship may not be an influential point, while a high-leverage point with a large residual can be strongly influential.
