Two features in a regression model are highly correlated. What effect does this have on their individual coefficient estimates?