0

I have a set of numbers $S$, and for each $s_i\in S$, $0\lt s_i \lt 1$. I would like to transform them so that they sum up to $1$.

An obvious way to do it is to calculate $t_i=\frac{s_i}{\sum_i{s_i}}$, so that $\sum_i{t_i}=1$. I wonder if there is any other way to achieve the same effect, and what are the differences between them.

MLister
  • 175
  • 1
  • 6

1 Answers1

4

A little generalization of this would be: $t_i=\frac{f(s_i)}{\sum f(s_i)}$ with $\sum f(s_i)$ not equal to zero. $f$ would allow you to "bend" $S$ (give different weights).

Arthur
  • 396
  • 3
  • 8
  • I almost updated the same. I am deleting my post. – vinux Jan 10 '12 at 16:05
  • @Arthur, thanks for the generalization, could you please elaborate with an example for what you mean by "bending" the set $S$? Appreciated. – MLister Jan 10 '12 at 16:07
  • 2
    +1. But the "little generalization" is *too* general: given any target set of values $t_i$ (summing to unity), you can find such an $f$ for which each $s_i$ is transformed into $t_i$, effectively losing all information about the original $s$'s. Normally, then, we restrict $f$ to be *positive,* *continuous,* and *strictly monotonically increasing* so that at least the order of the original sequence is preserved. – whuber Jan 10 '12 at 16:57
  • MLister, some illustrated examples of such re-expressions appear at http://stats.stackexchange.com/a/10979. When $f$ is linear (with zero constant term), its effect is *identical* to your "obvious" formula. When $f$ is nonlinear, it expands or contracts the differences between the $t_i$ by different amounts: this is the "bending" Arthur refers to. – whuber Jan 11 '12 at 13:17
  • @whuber that's true. However it really depends on what you're looking for. You could imagine that $s_i$ is the proportion of women in profession $i$, and use $f(x)=|x-0.5|$ as the "distance from perfect diversity". Moreover, nothing in the question forbids choosing $f=1_{\mathbb{Q}}$, although I agree my generalization is not "little" at all. – Arthur Jan 11 '12 at 14:59
  • MLister, If you look at @whuber's examples, what I mean by bending $S$ is that the "distance" between values is changing, e.g. the distance between $f(0.98)$ and $f(0.99)$ is huge as compared to the distance between $f(0.5)$ and $f(0.51)$, whereas the "original" distances (in $S$) were the same ($0.01$). But there is an endless number of choosing $f$, so it really depends on what $S$ is and what you're looking for exactly. – Arthur Jan 11 '12 at 15:14