s-news
[Top] [All Lists]

Re: expected value of a normal variable with integrate()

To: Agin.Patrick@hydro.qc.ca
Subject: Re: expected value of a normal variable with integrate()
From: Ping Zhang <pingzhang@avaya.com>
Date: 22 Dec 2003 16:29:25 -0500
Cc: s-news@lists.biostat.wustl.edu
In-reply-to: <3B7812F13593D411979200508BE3A75C10A23380@msxcentral2.hydro.qc.ca>
Organization:
References: <3B7812F13593D411979200508BE3A75C10A23380@msxcentral2.hydro.qc.ca>
Your second expression evaluates a mathematical "expectation".
In other words, you are calculating, through Monte Carlo simulation, the
expectation of the random variable X*1(X<0), where X follows a standard
normal distribution. If you let Y=X*1(X<0), you'll find that the density
function of Y is 2*dnorm(x).


On Mon, 2003-12-22 at 16:10, Agin.Patrick@hydro.qc.ca wrote:
> Hi,
> 
> Could anyone tell me why the two following expressions don't give the
> same result:
> 
> integrate(function(x) x*dnorm(x),-Inf,0) equals -0.3989
> and 
> tmp <- rnorm(1000000); mean(tmp[tmp<0]) approximately equals -0.79
> (two times -0.3989)
> 
> I noticed that -0.3989 is the result of mean(pmin(0,tmp)), the mean of
> all the normal deviates with a cap of zero.
> 
> If I'm interested in the expected value of all the negatives under the
> standard normal, the good result seems to be -0.79. Am I wrong? Why
> the integrate() expression above doesn't give this number?
> 
> Thank you,
> Patrick
> 
> P.S. I'm using S-Plus 4.5 on Windows XP.
-- 
Ping Zhang
Data Analysis Research Department
Avaya Labs
233 Mt.Airy Rd., Basking Ridge, NJ
(908)696-5138


<Prev in Thread] Current Thread [Next in Thread>