They also pick different comparison groups for the projected income -- likely to give the appearance of linearity where there is some evidence the predictions aren't actually linear over the quintiles.