Propagating errors

Gerard Tromp tromp at sanger.med.wayne.edu
Mon Jul 8 12:28:36 EST 1996


Helo Michael,

	The standard error of a quotient (sigma-quotient) of two means (x-bar1 and x-bar2)
which are mutually independent is:

	[sigma-quotient/quotient]^2 = (sigma1/x-bar1)^2 + (sigma2/x-bar2)^2

	sigma-quotient^2 = quotient^2 * [(sigma1/x-bar1)^2 + (sigma2/x-bar2)^2]

 	sigma-quotient  = {quotient^2 * [(sigma1/x-bar1)^2 + (sigma2/x-bar2)^2]}^1/2

Hope it helps.


Michael Lonetto wrote:
> 
> Hi,
> 
> I have a question about calculations on datasets, each with separate
> errors.  For example :-)
> 
> I have three data sets, each of which has a mean and a standard
> deviation.
> 
> Set 1 is a control (eg, no trascription activator)
> 
> Sets 2 and 3 are results with mutant and wild type activators,
> respectively.
> 
> Dividing the means of sets 2 and 3 each by set 1 provides a measure
> of transcription activation.
> 
> The question is, what to do about the errors:
> 
> 1) since I'm using the same set as control for both sets 2 and 3 can
> I ignore the error on set 1?
> 
> 2) If not, how can I combine the standard deviations?  Simple combinations
> (divide, multiply) do not produce reasonable numbers :-)  Adding them
> isn't too bad, but not of obvious validity.  Should I average the
> coefficients of variability and normalize them to the activation ratio?
> 
> Any way that I do this I see differences well beyond the spread of the
> data, the question is what is the proper thing to do?
> 
> Sorry for posting this to bionet.software, but I remember seeing a
> similar query some time back (and wondering at the time why anyone
> would care :-0)  I can't find any discussion of this problem in
> BIOMETRY or my other statistics book.  Any pointers to an appropriate
> source will be appreciated.
> 
> Thanks, Mike
> 
> --
> -=-=-=-=-=-=-            Mike Lonetto                   lonetto at cgl.ucsf.edu
> UCSF Dept of Stomatology, 513 Parnassus Avenue, San Francisco, CA 94143-0512
>                                                               =-=-=-=-=-=-=-

Gerard
-- 
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Gerard Tromp, Ph.D.
CMMG, Wayne State University    vox:	313-577-8773
3116, Scott Hall		fax: 	313-577-5218
540 E Canfield Ave		e-mail: tromp at sanger.med.wayne.edu
Detroit, MI 48201                       gtromp at cmb.biosci.wayne.edu



More information about the Methods mailing list