It's surprisingly non-trivial to pin down exactly what the 'correct' scale[0] for a given measurement is. I do agree about "roughly" generally being up to a factor of two in the appropriate scale[1], although I'm the sort who thinks a 19% increase (or 16% decrease) should be called a quarter of a factor of two.
0: Uniform, linear and logarithmic are obvious candidates, but depending on the domain you can end up with some really wierd scales (eg floating-point ULPs, which can look logarithmic or linear, but aren't either).
1: hence > So you're off by more than a factor of two.
I'd call that half a order of magnitude, considering log₁₀(3) is around 0.477. So you're off by more than a factor of two.