[R] Using known errors and error bars

[Prof Brian Ripley]

What do you actually know?

You don't have `known errors', as if you did you could correct the values.
I doubt if you actually have a known range, more likely a standard error 
or a confidence interval.  (If you think you do have a known range, how do 
you know?)

And if A is not known exactly, linear regression is not fully appropriate.  
If you know standard errors, then you need a homoscedastic 
errors-in-variables formulation.  One early account is 

Ripley, B. D. and Thompson, M.(1987) Regression techniques for the 
detection of analytical bias. Analyst 112, 177-183.

and its Fortran program is still available, and although I have never
coded it in R, I believe others have.

On Sun, 9 May 2004, Sebastian Schubert wrote:

> I'm now to R and hope (actually, I'm quite sure) you can help me. I made 
> an experiment and measured two values. As I know the errors of these 
> values I want to use them with the linear regression, 
> eg 
> Value A 
> 1.1+-0.02 
> 1.9+-0.05 
> 3.05+-0.03 
> 4.0+-0.01 
> 5.1+-0.06 
>  
> Value B 
> 4.2+-0.14 
> 5.3+-0.05 
> 6.8+-0.11 
> 7.9+-0.01 
> 8.5+-0.02 
>  
> lm(B~A) does the linear regression but how can I use the fact that 1.1 is 
> between 1.12 and 1.08? 
> How can I put error bars for A and B in the plot (like Excel is capable 
> of)? 

Many ways, for example using arrows() or plotCI in package gregmisc.

> I hope I made myself understandable, 
> thanks 
> Sebastian 

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595