View Full Version : Log Normal Distribution wtf?
I was looking it up on Wiki, which says it was this:
http://img464.imageshack.us/img464/1786/371418813ae6c9e52069fcbdt5.png
I've never seen the lognormal distribution before, but this can't be right can it? Is the x in the divisor a mistake? wtf?
linky (http://en.wikipedia.org/wiki/Log-normal_distribution)
anyone?
the long normal distribution should not be undefined @ x=0
however the above equation is, right?
okay I just checked this (http://mathworld.wolfram.com/LogNormalDistribution.html) website and it says the same. I will be in deep thought trying to understand why the function is 0 @ x = 0
any help appreciated..
edit: nevermind. exp(-infinity) is 0, not 1
all is well with the universe
PairTheBoard
04-26-2007, 04:22 PM
The pdf for the Log Normal is non-zero only for x>0.
You should be able to see it's correct by looking at the change of variable y=ln(x), dy=dx/x, x=e^y and integrating your f(x) from 0 to X for its Cumulative Distribution. After the change of variable the integral should equate to the Integral from -infinity to ln(X) of p(y) where p(y) is the standard normal pdf.
PairTheBoard
jay_shark
04-26-2007, 04:22 PM
This is a typo !! Ln x should be replaced with x and the denominator should not have an x .
This is the correct formula .
f(x) = 1/[sqrt(2*pi)sigma]*e^[-(x-mu)^2/(2*sigma^2)]
where sigma =standard deviation
mu is the mean and x takes on the values from negative infinity to positive infinity . Think of f(x) as the y-value and naturally x is on the x-axis . We can evaluate the height of this function by plugging in the appropriate numbers .
In this case x takes on the mean mu , then all you have to do is compute the standard deviation which is simple to do .
It is almost the exact same formula . The only difference is that they have lnx instead of x and that there is an x in the denominator which mine doesn't have .
PairTheBoard
04-26-2007, 04:23 PM
Also, see this Link.
Log Normal (http://mathworld.wolfram.com/LogNormalDistribution.html)
PairTheBoard
[ QUOTE ]
The pdf for the Log Normal is non-zero only for x>0.
[/ QUOTE ]
yeah yeah, I meant the limit as x--> 0 you nit /images/graemlins/grin.gif edit: uh, nevermind thought you said something else
jay shark: I think that's the normal distribution you are talking about there
thylacine
04-26-2007, 04:25 PM
[ QUOTE ]
The pdf for the Log Normal is non-zero only for x>0.
You should be able to see it's correct by looking at the change of variable y=ln(x), dy=dx/x, x=e^y and integrating your f(x) from 0 to X for its Cumulative Distribution. After the change of variable the integral should equate to the Integral from -infinity to ln(X) of p(y) where p(y) is the standard normal pdf.
PairTheBoard
[/ QUOTE ]
I 2nd this.
PairTheBoard
04-26-2007, 04:30 PM
[ QUOTE ]
This is a typo !! Ln x should be replaced with x and the denominator should not have an x .
This is the correct formula .
f(x) = 1/[sqrt(2*pi)sigma]*e^[-(x-mu)^2/(2*sigma^2)]
where sigma =standard deviation
mu is the mean and x takes on the values from negative infinity to positive infinity . Think of f(x) as the y-value and naturally x is on the x-axis . We can evaluate the height of this function by plugging in the appropriate numbers .
In this case x takes on the mean mu , then all you have to do is compute the standard deviation which is simple to do .
It is almost the exact same formula . The only difference is that they have lnx instead of x and that there is an x in the denominator which mine doesn't have .
[/ QUOTE ]
Doesn't that just give you the pdf for the Normal Distribution?
PairTheBoard
ApeAttack
04-26-2007, 04:43 PM
[ QUOTE ]
anyone?
the long normal distribution should not be undefined @ x=0
however the above equation is, right?
[/ QUOTE ]
Notice how there is an e^[-(lnx)^2] term.
As x-->0, this term --> e^[-(-inf)^2] = e^-inf = 0
Use L'Hopital's rule to see what this whole function converges to at x-->0
yeah, what tripped me up was I thought the limit as x --> 0+ of exp(-(ln(x))^2) was 1. /images/graemlins/frown.gif
BruceZ
04-26-2007, 10:50 PM
[ QUOTE ]
[ QUOTE ]
This is a typo !! Ln x should be replaced with x and the denominator should not have an x .
This is the correct formula .
f(x) = 1/[sqrt(2*pi)sigma]*e^[-(x-mu)^2/(2*sigma^2)]
where sigma =standard deviation
mu is the mean and x takes on the values from negative infinity to positive infinity . Think of f(x) as the y-value and naturally x is on the x-axis . We can evaluate the height of this function by plugging in the appropriate numbers .
In this case x takes on the mean mu , then all you have to do is compute the standard deviation which is simple to do .
It is almost the exact same formula . The only difference is that they have lnx instead of x and that there is an x in the denominator which mine doesn't have .
[/ QUOTE ]
Doesn't that just give you the pdf for the Normal Distribution?
PairTheBoard
[/ QUOTE ]
Yes, the original reference is correct for the LOG normal. Just as the sum of a large number of i.i.d. random variables has a distribution that converges to a normal distribution by the central limit theorem, a product of such random variables follows the log normal distribution.
PairTheBoard
04-26-2007, 11:24 PM
[ QUOTE ]
[ QUOTE ]
[ QUOTE ]
This is a typo !! Ln x should be replaced with x and the denominator should not have an x .
This is the correct formula .
f(x) = 1/[sqrt(2*pi)sigma]*e^[-(x-mu)^2/(2*sigma^2)]
where sigma =standard deviation
mu is the mean and x takes on the values from negative infinity to positive infinity . Think of f(x) as the y-value and naturally x is on the x-axis . We can evaluate the height of this function by plugging in the appropriate numbers .
In this case x takes on the mean mu , then all you have to do is compute the standard deviation which is simple to do .
It is almost the exact same formula . The only difference is that they have lnx instead of x and that there is an x in the denominator which mine doesn't have .
[/ QUOTE ]
Doesn't that just give you the pdf for the Normal Distribution?
PairTheBoard
[/ QUOTE ]
Yes, the original reference is correct for the LOG normal. Just as the sum of a large number of i.i.d. random variables has a distribution that converges to a normal distribution by the central limit theorem, a product of such random variables follows the log normal distribution.
[/ QUOTE ]
These would have to be iid random variables which only take on positive values though. Right?
PairTheBoard
arahant
04-27-2007, 12:37 AM
[ QUOTE ]
okay I just checked this (http://mathworld.wolfram.com/LogNormalDistribution.html) website and it says the same. I will be in deep thought trying to understand why the function is 0 @ x = 0
any help appreciated..
edit: nevermind. exp(-infinity) is 0, not 1
all is well with the universe
[/ QUOTE ]
Yeah, I read this and would have replied, but I honestly figured you'd see the error in whatever you were thinking before I hit 'post'. Plus, I didn't know what your problem was with it exactly. You just said 'wtf'...next time, try being more explicit /images/graemlins/smile.gif
vBulletin® v3.8.11, Copyright ©2000-2024, vBulletin Solutions Inc.