[R] robust method to obtain a correlation coeff?

[Bert Gunter]

Inline below.

Bert Gunter
Genentech Nonclinical Biostatisics

-----Original Message-----
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On
Behalf Of David Winsemius
Sent: Monday, August 24, 2009 8:53 AM
To: David Winsemius
Cc: r-help at r-project.org Help; ted.harding at manchester.ac.uk
Subject: Re: [R] robust method to obtain a correlation coeff?


On Aug 24, 2009, at 11:38 AM, David Winsemius wrote:

> ... but I do not know what those "certain situations" really are.
So there are some function that may be affected by settings of  
options("na.action") but I cannot tell you where to find a list of  
such functions.
---
Because this is up to the whim of package developers, there is no such list.
In general, most of the modeling functions of base R and recommended
packages, e.g. lm, glm, rlm in MASS, lme in nlme,... have an na.action
argument. But not universally: e.g. loess() has an na.action argument, but
lowess() does not.

A real gotcha with missing values is that some of R's core arithmetic
functions like mean(), median(), sum() etc. use a logical argument, na.rm =
TRUE or FALSE to control the handling of missings (as I'm sure you know).

-- Bert