We have examined the behavior of the compressibility, the dc conductivity, the single-particle gap, and the Drude weight as probes of the density-driven metal-insulator transition in the Hubbard model on a square lattice. These quantities have been obtained through determinantal quantum Monte Carlo simulations at finite temperatures on lattices up to $16\ifmmode\times\else\texttimes\fi{}16$ sites. While the compressibility, the dc conductivity, and the gap are known to suffer from ``closed-shell'' effects due to the presence of artificial gaps in the spectrum (caused by the finiteness of the lattices), we have established that the former tracks the average sign of the fermionic determinant ($\ensuremath{\langle}\text{sign}\ensuremath{\rangle}$), and that a shortcut often used to calculate the conductivity may neglect important corrections. Our systematic analyses also show that, by contrast, the Drude weight is not too sensitive to finite-size effects, being much more reliable as a probe to the insulating state. We have also investigated the influence of the discrete imaginary-time interval ($\ensuremath{\Delta}\ensuremath{\tau}$) on $\ensuremath{\langle}\text{sign}\ensuremath{\rangle}$, on the average density ($\ensuremath{\rho}$), and on the double occupancy ($d$): we have found that $\ensuremath{\langle}\text{sign}\ensuremath{\rangle}$ and $\ensuremath{\rho}$ are more strongly dependent on $\ensuremath{\Delta}\ensuremath{\tau}$ away from closed-shell configurations, but $d$ follows the $\ensuremath{\Delta}{\ensuremath{\tau}}^{2}$ dependence in both closed- and open-shell cases.