[R] In fact this is a Stats question, but... "The return."... Again!

[David Winsemius]

eesteves at ualg.pt wrote in 
news:20080530143958.243337zvyt3epsao at wmail.ualg.pt: 

> Dear All, I suspect I'm becoming a bit anoying but... following the 
> "question" about the low r2 vs signific regression (thru ANCOVA),
> the reviewer suggested (and I quote here): 
> 
> 'If you used RNA/DNA ratios as the dependent variable, but gut 
> content  as the independent variable, and fish length as the 
> covariate you  would be able to state that ?Gut content had no 
> significant influence  on RNA/DNA ratios after controlling for fish 
> length?'. 
> 
> Is it possible/sensible to perform ANCOVA with a indep (categ) var  
> (e.g. factor GUT in the previous msg) and a continuous covariate  
> (larval length)? If yes, how can I do it in R? I can not figure it 
> out  by myself! 

It sounds sensible and ... I thought you already did it:

RNADNA.sl.gut<-lm(sqrt(RNADNA) ~ gut*sl, subset=gut!="Yolk-sac", 
data=cond)
summary(RNADNA.sl.gut)

The "*" operator gave you the interaction (a single term since "gut had 
only two levels) and the main effects estimates. If you (or the 
reviewer) want the reduced model with only main effects:

RNADNA.sl.gut.red <- lm(sqrt(RNADNA)~gut+sl, subset=gut!="Yolk-sac", 
data=cond)



> 
> Thanks again. Regards, Eduardo 
> 
> Quoting Peter Dalgaard <p.dalgaard at biostat.ku.dk>: 
> 
>> eesteves at ualg.pt wrote: 
>>> *Thanks* all those who took the time to help me (even if the  
>>> "question" was not related to - the use of - R). 
>>>
>>> Now I think I can soundly make my point w/ the referee (can I use 
>>> your replies? If so I intend to properly cite its use?!?). 
>> In general, I think it is best not to cite this kind of replies, at
>> least not in publications. Steve E's note is a bit of an eye-opener
>> in that regard: There could in fact be serious problems in your  
>> analysis without respondents realizing it (e.g., you could have 
>> 1700 larvae, but they came from only 10 batches of eggs with a 
>> strong within-batch correlation). Judging from the text below I
>> wouldn't  expect that this is the case, but the risk is there. The
>> general  problem is that it is very difficult to give credit
>> without also  as signing some level of responsibility. 
>>
>>    -pd 
>>>
>>> Regards, Eduardo Esteves 
>>>
>>> ps - Sorry for not explaining the "biological details" of my  
>>> posting: RNA/DNA is the ratio of RNA content to DNA content  
>>> obtained for individual fish larvae (plus for each one I noted if 
>>> it had visible prey items in the gut or not, thus the levels Prey 
>>> and Empty of factor Gut); and sl is the standard length (distance 
>>> from the tip of snout to the posterior extremity of the hypurals, 
>>> the expanded bones at the end of the backbone that support the  
>>> caudal fin, in mm) of the specimens. In the MS, I consider the  
>>> relationship RNA/DNA to sl to be biologically irrelevant (due to  
>> the very low r2) although statistically significant. Furthermore, 
>>> no effect of gut content upon that relationship is significant  
>>> (facilitating further analysis of pooled data). 
>>>