Re: st: RE: Interval regression with skewed data

[Gillian.Frost@hsl.gov.uk]

Hello Ronán,

Thank you for your informative reply.  I think that your approach to 
dealing with zeros seems very sensible, and I will look into using this. 
It is also reassuring to know that predictions from -intreg- seem 
reasonable.

Thank you for your help.

Gillian




From:   Ronan Conroy <rconroy@rcsi.ie>
To:     "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Date:   12/01/2012 16:49
Subject:        Re: st: RE: Interval regression with skewed data
Sent by:        owner-statalist@hsphsun2.harvard.edu



On 2012 Ean 10, at 08:40, <Gillian.Frost@hsl.gov.uk> 
<Gillian.Frost@hsl.gov.uk> wrote:

> Nick, I apologise for not being clear in my original posting.  My 
> outcome/dependent variable is the number of colony forming units per ml, 

> and my predictor/independent variable is the region (North West, North 
> East, South East England,...) within which the sample was taken.

The approach I use is to express the colony forming units (CFU) as log10 
units. I do work with rather contaminated samples, but the approach may 
well work well in your case.

The problem of zero having no log is resolved when you note that a zero 
reading means no CFU were detected in 100 ml of sampled water; it does not 
mean that the water contains no bacteria. For this reason, I define zero 
CFU as having an upper limit of log10(1) and no lower limit (.).

Likewise, though you may not have seen them, you will get water samples 
where the CFU are too numerous to count, and these can be treated 
likewise. Unlike you, I have worked on datasets in which 40% of the data 
were so contaminated that the bugs were too numerous to count - and a lot 
of the world's population is still reliant on water like that!