0

I had to normalize my data using this equation :

$m ↦ (m - r_{min}) / (r_{max} - r_{min}) * (t_{max} - t_{min}) + t_{min}$

as seen here: scale a number between a range

If I calculate the mean of these data and I want then to calculate the standard error, how do i do?

Do I only take into account the multiplication/division? when I do the standard error with the data taking into account only multiplication/division factor (m/(rmax-rmin)*(tmax-tmin)) I end up with negative values and have a bigger standard error in comparison to the value I have with raw data.

Could someone help me with this please?

Best Jean

Jeremy Miles
  • 17,812
Jean Jacques
  • 1
  • Calculate this intervals at the original domain and then transform those estimates using your equation at hand. – usεr11852 Aug 01 '23 at 01:17
  • @usεr11852 How do you propose to incorporate the potentially very large contribution to the standard error introduced by the uncertainties in the extreme values of the data? – whuber Aug 01 '23 at 18:46
  • @whuber Same way we would do it in any other case. As the OP realises, calculating SE in a bounded domain is messy (values outside the expected support) so using the unbounded original domain is the way to go. Is there something else you would do? :) – usεr11852 Aug 01 '23 at 19:48
  • @usεr11852 My point is that the $r_{min}$ and $r_{max}$ are random variables, not constants. I would consider that with great care! – whuber Aug 01 '23 at 20:30

0 Answers0