If X and Y are log-normal, then log(X) and log(Y) are normal, hence
log(X)-log(Y)=log(X/Y) is normal too, and so X/Y is log-normal.
Best,
Bendix
----------------------
Bendix Carstensen
Senior Statistician
Steno Diabetes Center
Niels Steensens Vej 2
DK-2820 Gentofte
Denmark
tel: +45 44 43 87 38
mob: +45 30 75 87 38
fax: +45 44 43 07 06
[log in to unmask]
www.biostat.ku.dk/~bxc
----------------------
> -----Original Message-----
> From: A UK-based worldwide e-mail broadcast system mailing
> list [mailto:[log in to unmask]] On Behalf Of Allan Reese (Cefas)
> Sent: Tuesday, November 15, 2005 12:11 PM
> To: [log in to unmask]
> Subject: Query re Ratio statistics
>
>
> I'm working with ratio data. The statistic of interest is
> the ratio of two measurements. Both measurements are counts
> but are large with proportional accuracy, and each follows a
> lognormal distribution. Typical values for the ratio are
> 10^4 to 10^8. I'm looking for any theory to support
> distributional assumptions.
>
> I across some SPSS documentation on a procedure I'd not
> previously noticed. "Ratio statistics" is in the base
> module.
> http://www.rrz.uni-hamburg.de/RRZ/Software/SPSS/Algorith.115/ratio.pdf
> lists the formulae.
>
> The reason for writing is that the documentation, and a
> footnote in the output, state the assumption that the ratios
> follow a *normal* distribution. The standard I'm working to
> assumes that ratios will follow a *log-normal* distribution.
>
> I found plenty of references to the ratio of two *standard*
> normal variates (Cauchy) and plenty to ratios of quadratic
> functions of normal variates (F). Kendall (Advanced Theory
> of Statistics) gives a general integral for X/Y in terms of
> distributions and characteristic functions. Attempting to
> apply this, the ratio of two independent normal variates
> seems to integrate back to a normal distribution (dF
> proportional to exp(-x^2)).
>
> However, running simulations (500 repeats of 2000 sample
> pairs), suggests pretty conclusively that the ratio does not
> follow a normal distribution but is fully consistent with a
> lognormal. From simulations: normal/normal -> lognormal;
> lognormal/lognormal -> lognormal.
>
> I think this happens because both counts (called N and Na)
> are definitely positive numbers with pretty small CVs.
> Taking the inverse of a standard normal distribution gives a
> distribution with an asymptote at zero, but 1/Na gives
> another simple peaked distribution. Hence N/Na is very like
> a product of independent normal variates and tends to a
> lognormal distribution.
>
> Would anyone like to comment or point to references on
> whether it is more correct to assume that the ratio of two
> variates will follow a normal, lognormal or other distribution?
>
> Allan
>
>
> **************************************************************
> *********************
> This email and any attachments are intended for the named
> recipient only. Its unauthorised use, distribution,
> disclosure, storage or copying is not permitted. If you have
> received it in error, please destroy all copies and notify
> the sender. In messages of a non-business nature, the views
> and opinions expressed are the author's own and do not
> necessarily reflect those of the organisation from which it
> is sent. All emails may be subject to monitoring.
> **************************************************************
> *********************
>
|