This is not a matlab question but a numerical analysis problem. Since someone already generated the results for you, it's a matter of understanding that typical discretization errors scale with the discretization size raised to some power, which is the order of accuracy
where h is the discretization size, so
So within a shift, you can find the order of accuracy p
h = [0.2 0.1 0.05 0.025 ]
e = [6.98e-8 1.4e-9 2.5e-11 4.2e-13]
pfit = polyfit(log(h),log(e),1)
p = pfit(1)
