[R] Force regression line to a 1:1 relationship

[David Winsemius]

On Sep 14, 2011, at 10:52 AM, Patrick Breheny wrote:

> On 09/13/2011 04:27 PM, David Winsemius wrote:
> ...
>> Also should be plotting against fitted() rather than regressors.
> ...
>
> Both types of plots (vs. fitted values and vs. regressors) are very  
> common.  The former is a default plot in R, the first one that  
> appears if you submit:
>
> fit <- lm(y~X)
> plot(fit)
>
> The latter type of plot is called a "partial regression plot" or  
> "added variable plot".  They are discussed in any regression  
> textbook, as well as wikipedia and probably dozens of other web sites.

You appear to be criticizing a single sentence after an egregious  
removal of context. The context for that sentence was considering the  
graphical examination of the constant variance assumption, and for  
that question I still think the proper procedure is to use residuals  
vs. fitted.  I did not say that plotting residuals against regressors  
should not be done, and I was in fact was endorsing exactly that sort  
of graphical examination at two other points above that sentence for  
the purpose of assessing departures from linearity of individual  
regressors.

The Wikipedia entry for "partial regression plot" does not agree with  
the formulation you offered. It describes plotting residuals of Y  
around fit(y) versus residuals of X_i around fit(X_i ~ other X's).

>
> -- 
> Patrick Breheny
> Assistant Professor
> Department of Biostatistics
> Department of Statistics
> University of Kentucky

David Winsemius, MD
West Hartford, CT