If you take your 90%+ as a baseline (ie zero) then you can fiddle with the stats to get a massive number.
Let's say the best effort is 91.001% and our smart new process is 92.001%. Now set the baseline at 91.000% This is the sleight of hand bit: If we say that "normal" is 91.000, we now set the best effort as 0.001 and our effort as 1.001. That can seem quite reasonable when trotted out by a news reader or reporter.
So we are (1.001/0.001) x 100 = 100,100% better than the previous best!
The real improvement is more like: 92.001/91.001 x 100 = 101.098% which is a bit obvious when you look at the numbers involved.
With a single quite clumsy move and a bit of misdirection you can turn 101 into 100,100 and sound quite convincing. Now that's only using the very basic arithmetic operations and a bit of bullshit. The clever kids can really go to town.
We need some sort of laws about the presentation of statistics because it really is getting out of hand. The nonsense I've just presented isn't fiction. I saw something similar recently, can't remember what for but it took a while for me to calm down afterwards 8)
But in your example it's more sane if you flip it.
Survival rate from a new medicine may be 99.99% and only allows 1.0001 multiple of improvement. For severe outcomes, not allowing expansive language to describe improvements is a legitimate problem. i.e. if we insisted on this, it would be hard to motivate improvements in safety from 99% to 99.999% since they are "only <1% improvements".
So in cases like this I think of it in reverse: the death rate of 0.01% can be improved by 90% (or 10x) to 0.001%. This does run the risk of inflating the importance of the fix when the base rate is extremely low.
My rule is: articles which use multiples or percentages must give enough detail to exactly reconstruct the calculation method used.
And don't even get started on "Improved by 40%" (1.4x) being occasionally referred to as "Improved to 140%" (1.4x), which then is backported "Improved by 140%" (to 2.4x).
The real problem is percentages; people also intuitively assume "50% less than (50% more than X) is X"
Let's say the best effort is 91.001% and our smart new process is 92.001%. Now set the baseline at 91.000% This is the sleight of hand bit: If we say that "normal" is 91.000, we now set the best effort as 0.001 and our effort as 1.001. That can seem quite reasonable when trotted out by a news reader or reporter.
So we are (1.001/0.001) x 100 = 100,100% better than the previous best!
The real improvement is more like: 92.001/91.001 x 100 = 101.098% which is a bit obvious when you look at the numbers involved.
With a single quite clumsy move and a bit of misdirection you can turn 101 into 100,100 and sound quite convincing. Now that's only using the very basic arithmetic operations and a bit of bullshit. The clever kids can really go to town.
We need some sort of laws about the presentation of statistics because it really is getting out of hand. The nonsense I've just presented isn't fiction. I saw something similar recently, can't remember what for but it took a while for me to calm down afterwards 8)