2
$\begingroup$

How to prove that:

$$P_{avg} = \dfrac{w_{1}P_{A} + w_{2}P_{B} + w_{3}P_{C}}{w_{1} + w_{2} + w_{3}}$$

lies between 0 and 1, where $P_{i}$ corresponds to probability scores and $w_{j}$ are real numbers

$\endgroup$
  • 5
    $\begingroup$ The conclusion cannot be proven unless you assume the $w_j$ are nonnegative real numbers. $\endgroup$ – whuber Jan 17 '16 at 21:06
  • $\begingroup$ The answers and the comment by @whuber are true, under the presumption that all the probabilities lie in the range 0 to 1. As I have shown at stats.stackexchange.com/questions/4220/… , this presumption is not always true in the "real world". $\endgroup$ – Mark L. Stone Aug 9 '16 at 22:01
1
$\begingroup$

It is the same way in which one proves that any weighted average lies between its maximum and minimum component. Consider the set $$A=(P_{A},P_{B},P_{C})$$ Without loss of generality, let ${max{A}=P_{A}}$and ${min{A}=P_{B}}$

Now, consider weights $$0\leq\omega_{i}\leq1$$ such that $$\sum_{i}\omega_{i}=1$$ . Therefore, $$0\leq P_{B}\leq\tilde{P}=\sum_{i}\omega_{i}P_{i}\leq P_{A}\leq1$$

$\endgroup$
  • $\begingroup$ I am not sure, if I understand the part in the equality where you put weighted sum between min and max. I understand intuitively that if you replace all probabilities by max probability and then take a weighted sum, it will be equal to max. As all other probabilities are less than max, hence the score will be less.. $\endgroup$ – The Wanderer Jan 17 '16 at 20:45
  • $\begingroup$ Your intuition is correct- the max probability with weight 1 will be weakly higher than a sum of weighted probabilities which contain elements (weakly) lesser than the max probability. $\endgroup$ – ChinG Jan 17 '16 at 20:50
1
$\begingroup$

There are probably a lot of ways to prove this, but here's my shot.

Let $\tilde{w_i} = w_i/(w_1+w_2+w_3)$, $W = (\tilde{w_1}, \tilde{w_2}, \tilde{w_3})'$, $P = (P_A,P_B,P_C)'$ and $|| \cdot ||_2$ denote the $L2$ norm. Then $$ P_{avg} = |W'P| \le ||W||_2 |||P||_2 \le 1 $$

by the Cauchy-Schwarz inequality.

$\endgroup$
  • 1
    $\begingroup$ (1) This does not appear to be an accurate statement of the Cauchy-Schwarz Inequality, unless you mean some unusual things by "$||\ ||$". (2) You still need to prove that $||W||$ and $||P||$ do not exceed $1$. Why not just use Holder's Inequality? $\endgroup$ – whuber Jan 17 '16 at 21:16
  • $\begingroup$ 1) you're right; I changed the notation a bit. 2) That's pretty much by assumption. CS is just a special case of Holder. $\endgroup$ – Taylor Jan 17 '16 at 21:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.