Time-based estimates are particularly prone to poor estimates.
The studies that show poor propensity for estimation focused on time. Relative sizing is allegedly a more accurate approach. When siding with time estimates, though, a common approach is to multiply the best guess by three.