ioi speaks out about ergh "VGC analysts"

[DonFerrari]

Alby_da_Wolf said:

DonFerrari said:

Alby_da_Wolf said:
To people suggesting a number like 370022 for a given platform in a given week be rounded to 370000: it wouldn't be correct. If calculations bring to a result of 370022 ± error margin, rounding it to 370000 ± error margin wouldn't be the same. Look at the first message of this thread, it's explained quite clearly: that 370022 value is the midpoint of the probability curve of the extrapolation of sales data collected that week. BTW most probably a rounding to the closest unit already happened, as sales are integer numbers, start values are integers and final values must be too, but intermediate values very often won't and calculations will be made keeping all the available decimals in every intermediate result, and rounding will happen only at the end to limit the growth of rounding error, that would just be added to the error already present in the extrapolation.
I'm puzzled, I thought that some basic rules for rounding and about measurement errors were taught also in high-school physics courses all around the world, not just at university.

And aparrently you haven't learn them since you want to use more signficance numbers than the tolerance permit.

Just no. When you receive sales data, you collect integer numbers and you aren't introducing yourself an additional measurement error like if you were measuring a length, say, with a 1mm graduated ruler. Those integer numbers can be precise, if the store can give them to you for a given week, or they'll be approximations plus or minus an error margin. You'll take them into account, and you'll have to take into account also the precision available for your internal calculations and the approximation error accumulation, plus obviously your estimate of the error in the formula you devised for the extrapolation of your data but even then, rounding the final result of the central value to an integer different from the closest one (or the closest greater or lower one) would be an error in every case.

Not sure if you know metrology, but you can also decentralize your "mean"... instead of saying 10.0+-0.05 you could say 9.95+0.1. In statistic don't know it would be acceptable... but I never saw a statistic report on election being 22.5+-.5% of intention (and they use integer numbers even tough they state results in percentages), and then they give margin of error as 5 pp (percentual points)... by your logic they should state it as 22.5341% (if that was the exact approximation value on their method on lets say 1430 people polled).