s-news
[Top] [All Lists]

Re: LME- log-normal distribution of parameters

To: david.thompson@mnr.gov.on.ca, jadhavpr@vcu.edu, s-news@wubios.wustl.edu
Subject: Re: LME- log-normal distribution of parameters
From: Andrew Robinson <andrewr@uidaho.edu>
Date: Fri, 26 Mar 2004 08:49:58 -0800
Cc: slarsen@insightful.com
In-reply-to: <3B1DF09C9A08D3118D310008C7913C450F51036D@rlc00aex006.mnr.gov.on.ca>
Organization: University of Idaho
References: <3B1DF09C9A08D3118D310008C7913C450F51036D@rlc00aex006.mnr.gov.on.ca>
User-agent: KMail/1.5.4
Although it's pragmatic, this approach leads to a solution that depends on the 
magnitude of the correction.  Why, after all, use log(predictor+1)?  Why not 
log(predictor+0.0001)?  The log transform makes no sense when there are zeros 
unless you're sure that the zeros are due to some measurement error.

Another, less arbitrary, avenue to explore is the so-called hurdle or two-step 
models.

Andrew

On Friday 26 March 2004 08:20, david.thompson@mnr.gov.on.ca wrote:
> Does log(predictor+1) seem a reasonable alternative?
>
> DaveT.
> **********************************************************
> David J. Thompson
> Silviculture Data Analyst
> Ontario Forest Research Institute
> Ontario Ministry of Natural Resources
> 1235 Queen Street East
> Sault Ste. Marie, Ontario, P6A 2E5
>
> (705) 946-7433
> (705) 946-2030 Fax
> david.thompson@mnr.gov.on.ca
> **********************************************************
>
>
> -----Original Message-----
> From: Pravin [mailto:jadhavpr@vcu.edu]
> Sent: March 25, 2004 6:42 PM
> To: s-news@wubios.wustl.edu
> Cc: slarsen@insightful.com
> Subject: LME- log-normal distribution of parameters
>
>
>
> Hello,
>
> >lme(response ~ time, data=data.g, random = ~ 1+time|ID)
>
> ##"time" is used a predictor for the "response" in the data frame "data.g"
> ##Random regression intercepts and slope on "time"
> ##Random effects vary over  individual ID
>
> The above model assumes that the intercept and the slope are normally
> distributed. How can I specify log-normal distribution for these
> parameters. But the residual error can be normally distributed-- that is
> fine. One suggestion was to use log transformation. But I cannot use log
> transformation in the fixed effects model (log(response)~log(time)) because
> there are a few 0's in the predictor column.
>
> Thank you,
>
> Pravin
>
> Pravin Jadhav

-- 
Andrew Robinson                      Ph: 208 885 7115
Department of Forest Resources       Fa: 208 885 6226
University of Idaho                  E : andrewr@uidaho.edu
PO Box 441133                        W : http://www.uidaho.edu/~andrewr
Moscow ID 83843                      Or: http://www.biometrics.uidaho.edu
No statement above necessarily represents my employer's opinion.


<Prev in Thread] Current Thread [Next in Thread>