Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Configuring Data to percentage ratios


From   Joseph Coveney <[email protected]>
To   Statalist <[email protected]>
Subject   Re: st: Configuring Data to percentage ratios
Date   Mon, 06 Dec 2004 15:38:10 +0900

Nick's suggestion sounds very much worth looking into.  I don't have timely
access to the book, but from what's said about it (Google for "Re:
Confidence Interval on Quotient" and "Rolf Turner"), it bears a striking
resemblance to Fieller's methods.

If John McCloskey and his audience don't mind working with transformed
values, then a simple approach to the pharmacologist's 'fold above control'
penchant is to logarithmically transform the fluorescence values and submit
them to ordinary least squares regression with a dummy (indicator) variable
for the control group (0) and drug group (1).  The Student's t-based
confidence interval reported by -regress- should be useable as-is if you're
willing to keep in the transformed scale for reporting and interpretation.
I'd hesitate to advocate back-transforming (exponentiating) the regression
coefficient and confidence limits:
www.stata.com/statalist/archive/2002-12/msg00193.html .  Once you've made
the commitment to transform the values, it would seem better to stay with
the transformed scale thereafter.

Generalized linear modeling with a logarithmic link function would be
another approach, one that could give Wald confidence intervals in the
untransformed metric, e.g., -glm , family(gaussian) link(log) eform-, but
this would seem to need larger sample sizes than what John has.

Joseph Coveney

----------------------------------------------------------------------------

Nick Cox wrote:

Not my field at all, but the chapter on ratios
in

Rupert G. Miller.
1986 (reissue 1997).
Beyond ANOVA.
New York: John Wiley (London: Chapman and Hall)

looks relevant.


Nick
[email protected]

David Airey

> > Pharmacology journals, at least in the past, allowed you to divide
> > individual
> > fluorescence values in each of the drug groups by the mean of the
> > control group
> > values as the divisor, and then calculate SEs or SDs from the
> > quotients.  This
> > treats the control group's mean as if it were fixed, known.
>  It gives
> > an SE
> > that is symmetric about the ratio of the means.
>
>
> This is still allowed. I would love to read a reference in
> which it is
> shown to hurt rather than help. Being in a pharmacology
> department but
> coming from other fields, the practice bothers me.
>







*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index