Leverage in linear regression measures how unusual an observation's values of the independent variables
are relative to those of the other observations. If 
is the full-column-rank design
matrix, the projection matrix
 |
maps the observed response vector to the vector of fitted values. The leverage of observation 
is the diagonal entry 
. If 
has 
columns and 
rows, the leverages sum to 
and therefore have average 
.
An observation having unusually large leverage is called a leverage
point.
