home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.math.stat
- Path: sparky!uunet!caen!roy
- From: roy@biostat.med.umich.edu (Roy St. Laurent)
- Subject: Re: Calculating variances
- Message-ID: <TrG=-G=@engin.umich.edu>
- Date: Fri, 20 Nov 92 10:47:07 EST
- Organization: University of Michigan Biostatistics, Ann Arbor
- References: <luke.722264309@barney>
- Reply-To: roy@sol.biostat.med.umich.edu (Roy St. Laurent)
- Keywords: variance
- Originator: roy@sol
- Nntp-Posting-Host: sol.biostat.med.umich.edu
- Lines: 37
-
-
- In article <luke.722264309@barney>, luke@cs.city.ac.uk (Luke Whitaker) writes:
- |>
- |> It seems that the only general rule for computing variances is for the sum or
- |> difference of random variables (ie var(x+/-y) = var(x) + var(y)). What about
- |> products, reciprocals and other arbitrary combinations of random variables.
- |> I realise that this is non-trivial and that there is no simple answer as
- |> there is for sums and differences but there must be techniques to get a
- |> handle on this.
- |>
- |> Could someone point me in the right direction, references etc for this problem.
-
- An approximate answer to your question is given by the delta-method -- a method
- for finding the asymptotic variance of f(x) for some smooth function f(.), as a
- function of var(x).
- In the simplest case, the delta method gives:
- approx var(f(x)) = f'(mu)^2 var(x)
- where f'(mu) is the first derivative of f evaluated at mu=E(x).
- For ratios of random variables and other functions, a more general
- (multivariate) version of the delta method may be used. See either of the
- references below for the details.
-
- Ref'ns
- Bishop,YMM, Fienberg,SE, & Holland, PW (1975). Discrete Multivariate Analysis:
- Theory and Practice. MIT Press: Cambridge, Mass. p486-502.
- Serfling, RJ (1980). Approximation Theorems of Mathematical Statistics. Wiley:
- New York.
-
- -- Roy St. Laurent
- Biostatistics
- Univ. of Mich.
- roy@sol.biostat.med.umich.edu
- --
- -- Roy St. Laurent (roy@biostat.med.umich.edu)
-
- "Even $10 calculators have sigma buttons. [I] haven't seen one with an
- M-estimator button, yet." --Jerry Dallal in sci.math.stat
-