>mathematician arbitrarily chose the expected value, E[X] of a given probability distribution X:S-> R to be the sum of all (value_i - mean)^2 times p_i because it is convenient to work with.
how is this allowed? Why not just take the absolute value of it instead of squaring it
>>8420658
>derivative of |x| is a fun thing to work with
>>8420660
it is true tho, how the fuck do you interpret the currently accepted E[X] in words? The squaring removes any meaning from it
That's a very strange definition you have there. Pretty sure you mixed something up.
Reference?
>>8420667
this is for when the domain is N, but you get the idea.
>>8420667
>>8420676
wait, I meant the variance, not the mean, the mean makes complete sense.
>>8420658
Also, on re-read, you fucked up OP:
[eqn]Var(X) := E[(x_i - \bar{X})^2] = \sum_i p_i (x_i - \bar{X})^2 [/eqn]
>>8420691
yea well whatever, i meant the variance.
>>8420695
>Perhaps you can say what is confusing you
Var(X) is just the 'sum' of of squared distances from the mean, weighted by the probability of observing a given [math]x_i[/math].
> sum here being a placeholder for either summation or integration
>>8420715
If X and Y are independent, then Var(X+Y)=Var(X)+Var(Y). This would not be true if you used the absolute value instead of the square.
>>8420715
What would the central limit theorem look like if our standard were based on the absolute value? From a practical perspective, we get a lot of mileage out of the CLT.
>>8420728 makes an equally good point, and is again grounded in the notion that 'it is easier to work with'.
Also, what you are proposing/asking about, [math]E[|x_i - \bar{X}|][/math], is sometimes used, depending on the context.
>>8420751
>What would the central limit theorem look like if our standard were based on the absolute value?
There is just a constant factor between the standard derivation and the average distance to the mean for a normal distributed variable. A normal distributed random variable [math]X [/math] has the property that
[eqn] V(X) = E[(X - E(X))^2] = \frac{2}{\pi} \left( E[|X - E(X)|] \right)^2 [/eqn]
>>8420788
But we don't parameterize the normal distribution this way.
>be pissed off at this thread
>forgot that I recently started browsing /g/ and installed Noscript