[R] Getting the same values of adjusted mean and standard errors as SAS

[Frank E Harrell Jr]

David J. Netherway wrote:
> Thanks for the help.
> 
> Both the "Design" package and the "effects" package look as though they are
> what I need although it will probably take me a while to get on top of 
> both.
> 
> I have had a brief  go at the Design package and the contrast function 
> is particularly useful.
> 
> A question on the Design package:
> 
> There are 5 types for factor "group", one is the reference - call it "a".
> 
> f <- ols(y ~ age + sex + group, data=dd)
> contrast(f, list(group='a'), list(group='b'))
> 
> I can use this to contrast pairs but can I use this to contrast b 
> against c,d, and e as a group.
> Also "a" against the rest?
> 

Type ?contrast.Design.  You'll see examples of 'vector' contrasts with 
and without weighted/unweighted averaging of effects.  E.g. contrast(f, 
list(group='b'), list(group=c('c','d','e'))) will give 3 contrasts. 
There is an option to average these.

Frank
> 
> Thanks, David
> 
> Frank E Harrell Jr wrote:
> 
>> On Thu, 27 May 2004 16:34:58 +0930
>> "David J. Netherway" <david.netherway at adelaide.edu.au> wrote:
>>
>>  
>>
>>> Hello,
>>>
>>> I am trying to get the same values for the adjusted means and 
>>> standard errors using R that are given in SAS for the
>>> following data. The model is Measurement ~ Age + Gender + Group. I 
>>> can get the adusted means at the mean age  by using predict. I do not 
>>> know how to get the appropriate standard errors at the adjusted means 
>>> for Gender
>>> using values from predict. So I attempted to get them directly from 
>>> the residuals as follows. The data is at the end
>>> of the email. While there is a match for the males there is a large 
>>> difference for the females indicating that what I am doing is wrong.
>>>
>>> #  meanAge <- mean(dd$Age)
>>> meanAgeM <- mean(dd$Age[d$Gender=="M"])
>>> meanAgeF <- mean(dd$Age[d$Gender=="F"])
>>>   
>>
>> . . . .
>>
>> By using sex-specific means of age you are not getting adjusted estimates
>> in the usual sense.
>>
>> I prefer to think of effects as differences in predicted values rather
>> than as complex SAS-like contrasts. The Design package's contrast 
>> function
>> makes this easy (including SEs and confidence limits):
>>
>> library(Design)   # also requires Hmisc
>> d <- datadist(dd); options(datadist='d')
>> f <- ols(y ~ age + sex + group, data=dd)
>> contrast(f, list(sex='M'), list(sex='F'))   # usual adjusted difference M
>> vs F
>> contrast(f, list(sex='M',age=mean(dd$age[dd$sex=='M']),
>>            list(sex='F',age=mean(dd$age[dd$sex=='F')) # M vs F not
>> holding age constant
>>
>> You can also experiment with specifying age=tapply(age, sex, mean,
>> na.rm=TRUE) using some of the contrast.Design options.
>> ---
>> Frank E Harrell Jr   Professor and Chair           School of Medicine
>>                     Department of Biostatistics   Vanderbilt University
>>  
>>
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html
> 


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