[R] package "rms" nomogram

[David Winsemius]

Putting context back in.

>> On Jul 31, 2012, at 9:01 AM, B787s wrote:
>>
>>> Dear R-Help,
>>>
>>> I am using 'rms' package to draw nomogram. I wonder how is the  
>>> "Points"
>>> determined for each predictor in the model? Is it by the coefficient
>>> estimate (beta) relative to the highest effect in the model or?
>>
>> It would be better if you asked this question about a specific  
>> example because the rms package has many sorts of regression fit- 
>> objects for which nomogram will provide results. The linear  
>> predictor in a regression method will have contributions from each  
>> of the terms, so I would have said that the variate scales were  
>> being displayed relative to the mean values rather than relative to  
>> the "highest effect" ... what ever that term means to you.
>>
>> The upper portion of a nomogram is used to calculate "Points",  
>> while the lower portion is used to calculate probability of event  
>> by transforming from the linear predictor scale to the response  
>> scale. A unit-increment in "Points" displayed by 'plot.nomogram'  
>> for one variate will be related to a unit increment of another  
>> variate by the ratios of their coefficients.
>>
>>
>>> Thanks
>>> Lin
>>>
>> David Winsemius, MD

On Aug 1, 2012, at 6:22 PM, B787s wrote:

> Thanks, it is helpful.
> I knew there were several modeling capacities built into this  
> package "rms".
> I would just like to have a general ideal how the points for each  
> predictor
> determined. I read a paper by Lasonos et al 2008. It mentioned it  
> was by the
> size of the effect.
>

Let me guess.... it's the one lying behind the request for US$ at:

http://jco.ascopubs.org/content/26/8/1364.long


It seems possible the conceptual gaps may be in the degree to which  
you understand how glm() functions work. Do you have a working  
understanding of what a linear predictor is? Do you understand what a  
link function does? Do you understand that a unit change in the linear  
predictor will not imply a unit change in the response unless the link  
function is "identity"? If the article did not cover those topics then  
you were ill-served.

-- 
David Winsemius, MD
Alameda, CA, USA