[R] Understanding lm-based analysis of fractional factorial experiments

[Peter Claussen]

On Mar 7, 2013, at 4:47 AM, Kjetil Kjernsmo <kjekje at ifi.uio.no> wrote:

> On Wednesday 6. March 2013 16.33.34 Peter Claussen wrote:
>> But you don't have enough data points to estimate all of the possible
>> interactions; that's why you have NA in your original results.
> 
> Yes, but it seems to me that lm is doing the right thing, or at least the 
> expected thing, here, the NA's are simply telling me these are aliased, which 
> is correct and expected.
> 

I ran across the text you reference while looking up a couple other references yesterday; having read the appropriate section I can better understand your questions.

I would say in this case that what R is doing is expected, but not necessarily correct, for this problem. 

The authors go into some detail about aliasing treatments, and that there are different choices for aliasing. I didn't have my computer handy, but does R choose the same set of aliases as the authors?

Peter


>> You could
>> add the just the first order interactions manually, i.e.,  + B:C + B:D …
> 
> Yeah, I tried that, but then it returns to the "unexpected" result, i.e., I 
> get the same result as with the yavg ~ B * C * D * E * Q formula. Therefore, I 
> think the problem doesn't lie with the formula, nor does it lie with any of 
> the code, it is just a matter of understanding defaults...
> 
> I have consulted local help (of course), but what they say is that "R has some 
> odd defaults, you need to ask them or use something different". I don't want to 
> use something different, I like R, I have contributed to R in the past and will 
> do so again if only I can get my head around this... :-)
> 
> Kjetil
> 
> ______________________________________________
> 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.