We study three different random walk models on several two-dimensional lattices by Monte Carlo simulations. One is the usual nearest neighbor random walk. Another is the nearest neighbor random walk which is not allowed to backtrack. The final model is the smart kinetic walk. For all three of these models the distribution of the point where the walk exits a simply connected domain $D$ in the plane converges weakly to harmonic measure on $\partial D$ as the lattice spacing $δ\rightarrow 0$. Let $ω(0,|dz|;D)$ be harmonic measure for $D$, and let $ω_δ(0,|dz|;D)$ be the discrete harmonic measure for one of the random walk models. Our definition of the random walk models is unusual in that we average over the orientation of the lattice with respect to the domain. We are interested in the limit of $(ω_δ(0,|dz|;D)- ω(0,|dz|;D))/δ$. Our Monte Carlo simulations of the three models lead to the conjecture that this limit equals $c_{M,L} \, ρ_D(z) |dz|$, where the function $ρ_D(z)$ depends on the domain, but not on the model or lattice, and the constant $c_{M,L}$ depends on the model and on the lattice, but not on the domain. So there is a form of universality for this first order correction. We also give an explicit formula for the conjectured density $ρ_D$.