Extension: Error Analysis in Linear Approximation

When using a linear approximation of $x^{\frac23}$ at $x=9$, why is the error larger for the estimate of $9.2^{\frac23}$ than for the estimate of $9.1^{\frac23}$?