R help - Aug 2010 - Plotting confidence bands around regression line

Dear R-helpers and graphics gurus,

I have two problems with plotting confidence bands:

1. First is relatively simple. I am using the Passing-Bablok procedure 
to obtain "unbiased" regression coefficients. This procedure yields
the
"a" & "b" coefficient values along with their confidence
intervals. I
then plot the raw data with the regression line, but I would like to add 
the confidence band for the line... and I can't figure out how to do it.

In other words, given:

            Estimate      5%CI     95%CI
Intercept -4.305562 -9.931152 -1.381792
Slope      1.257318  1.053025  1.678516

How to plot the regression line with confidence band?

2. Second problem is plotting confidence bands along fitted "nls" 
regression line. I tried "predict(nls.object, int='c')" - but
doesn't
work. Later I figured in the documentation that the 'int' parameter is 
currently ignored. I guess this means it's not a trivial thing to do. 
Does anyone have a suggestion on how to obtain confidence predictions 
for such a model?

Please cc my email address when you reply. Thanks and best regards,

-- 
Michal J. Figurski, PhD
HUP, Pathology & Laboratory Medicine
Biomarker Research Laboratory
3400 Spruce St. 7 Maloney
Philadelphia, PA 19104
tel. (215) 662-3413

On Aug 10, 2010, at 10:56 AM, Michal Figurski wrote:
> Dear R-helpers and graphics gurus,
>
> I have two problems with plotting confidence bands:
>
> 1. First is relatively simple. I am using the Passing-Bablok  
> procedure to obtain "unbiased" regression coefficients. This  
> procedure yields the "a" & "b" coefficient values
along with their
> confidence intervals. I then plot the raw data with the regression  
> line, but I would like to add the confidence band for the line...  
> and I can't figure out how to do it.
>
> In other words, given:
>
>           Estimate      5%CI     95%CI
> Intercept -4.305562 -9.931152 -1.381792
> Slope      1.257318  1.053025  1.678516
>
> How to plot the regression line with confidence band?
Take a look at plotCI in either gplots or plotrix packages.

Harrell's rms/Hmisc packages are nicely integrated with lattice and  
encourage you to create effective displays of models that remove  
simplistic linearity assumptions.

-- 
David.
>
> 2. Second problem is plotting confidence bands along fitted "nls"
> regression line. I tried "predict(nls.object, int='c')" -
but
> doesn't work. Later I figured in the documentation that the
'int'
> parameter is currently ignored. I guess this means it's not a  
> trivial thing to do. Does anyone have a suggestion on how to obtain  
> confidence predictions for such a model?
>
> Please cc my email address when you reply. Thanks and best regards,
>
> -- 
> Michal J. Figurski, PhD
> HUP, Pathology & Laboratory Medicine
> Biomarker Research Laboratory
> 3400 Spruce St. 7 Maloney
> Philadelphia, PA 19104
> tel. (215) 662-3413
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
David Winsemius, MD
West Hartford, CT

>>> Frank Harrell <f.harrell at vanderbilt.edu> 11/08/2010
17:02:03 >>>
> This problem seems to cry out for one of the many available robust 
> regression methods in R.
Not sure that would be much more appropriate, although it would
_appear_ to work. The P&B method is a sort of nonparametric/robust
approach to an errors-in-variables problem, intended to provide an
indication of consistency of results between two different measurement
methods, often with similar error variance. So the aim is to handle the
error-in-variable problem at least consistently, to avoid the bias that
results from assuming no error in predictors. The M-estimator and
related robust regression methods in things like MASS and robustbase
don't handle errors in the predictors. Of course, with small errors in
predictors that won't matter much; rlm and the like will be pretty much
as defensible then as they ever are.

But perhaps one could construct a more formal robust equivalent of
error-in-variable regression by using a max likelihood functional
relationship model with bivariate t (choosing arbitrarily low df)
instead of bivariate gaussian errors? Unfortunately I haven't tried
that, so no help beyond the thought ...

Steve Ellison

*******************************************************************
This email and any attachments are confidential. Any use...{{dropped:8}}

I may be missing something but I don't see how P&B handles errors in 
variables any differently than other regression methods that ignore 
this problem.

Frank

Frank E Harrell Jr   Professor and Chairman        School of Medicine
                      Department of Biostatistics   Vanderbilt University

On Wed, 11 Aug 2010, S Ellison wrote:
>
>
>>>> Frank Harrell <f.harrell at vanderbilt.edu> 11/08/2010
17:02:03 >>>
>
>> This problem seems to cry out for one of the many available robust
>> regression methods in R.
>
> Not sure that would be much more appropriate, although it would
> _appear_ to work. The P&B method is a sort of nonparametric/robust
> approach to an errors-in-variables problem, intended to provide an
> indication of consistency of results between two different measurement
> methods, often with similar error variance. So the aim is to handle the
> error-in-variable problem at least consistently, to avoid the bias that
> results from assuming no error in predictors. The M-estimator and
> related robust regression methods in things like MASS and robustbase
> don't handle errors in the predictors. Of course, with small errors in
> predictors that won't matter much; rlm and the like will be pretty much
> as defensible then as they ever are.
>
> But perhaps one could construct a more formal robust equivalent of
> error-in-variable regression by using a max likelihood functional
> relationship model with bivariate t (choosing arbitrarily low df)
> instead of bivariate gaussian errors? Unfortunately I haven't tried
> that, so no help beyond the thought ...
>
> Steve Ellison
>
> *******************************************************************
> This email and any attachments are confidential. Any u...{{dropped:9}}