Skip to content

How to do log normal regression?

3 messages · Haibo Huang, Spencer Graves, Frederico Zanqueta Poleto

Original
Haibo Huang
# 
I want to fit a Log-Normal CDF function between two
variables, and estimate the parameters. Is there any
package/functions designed for this purpose? 

Basically, I have data for Y and X, and I suspect the
relationship between Y and X is Y = CDF Log-Normal
(X), and I want to run this regression to verify this
and estimate the parameters. Anyone has any thoughts?

Any input is valuable to me, so please do not hesitate
to share your thoughts. Thank you!

Ed.
6 days later
Spencer Graves
# 
1.  Have you considered "nls" or "optim"?  The documentation for both 
includes useful examples.

	  2.  What do your Y values represent?  The almost universal standard 
today is maximum likelihood estimation.  If you tell us what the Y 
values are, someone might help you write a likelihood function that 
could then be maximized using "optim".

	  3.  I notice this is the third post I've answered for you in the past 
month.  I've been so busy recently, I tend to limit myself only to posts 
that are several days old for which I have not seen a reply.  I believe 
you might get quicker and more useful replies to your questions if you 
please read and try to follow the posting guide! 
"http://www.R-project.org/posting-guide.html".

	  Good luck.
	  spencer graves
Haibo Huang wrote:

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Frederico Zanqueta Poleto
    

    

      
      #
    

  


  

      
A very simple and useful relation between the normal and log-normal distributions is that if log(Y) follows the normal distribution, then Y follows the log-normal distribution, and vice-versa.

Then, if you want to do a log-normal regression, you can take the natural logarithm of your data and do a normal regression.

Take a look on any book about statistical distributions, e.g. the one from Evans, Hastings and Peacock or Johnson and Kotz or Patel, Kapadia and Owen,  because I am not so sure, but I think the exponential of the fitted values will be the estimated median of the log-normal distribution.

Regards,
--
Frederico Zanqueta Poleto
fred at poleto.com
--
"All knowledge is, in final analysis, History. All sciences are, in the abstract, Mathematics. All judgements are, in their rationale, Statistics." Radhakrishna Rao


------------- Segue mensagem original! -------------

De: Spencer Graves <spencer.graves at pdf.com>
Data: Wed, 24 Aug 2005 21:18:48 -0700
Para: Haibo Huang <edhuang00 at yahoo.com>
Assunto: Re: [R] How to do log normal regression?

	  1.  Have you considered "nls" or "optim"?  The documentation for both 
includes useful examples.

	  2.  What do your Y values represent?  The almost universal standard 
today is maximum likelihood estimation.  If you tell us what the Y 
values are, someone might help you write a likelihood function that 
could then be maximized using "optim".

	  3.  I notice this is the third post I've answered for you in the past 
month.  I've been so busy recently, I tend to limit myself only to posts 
that are several days old for which I have not seen a reply.  I believe 
you might get quicker and more useful replies to your questions if you 
please read and try to follow the posting guide! 
"http://www.R-project.org/posting-guide.html".

	  Good luck.
	  spencer graves

        
Haibo Huang wrote: