Re: Testing SD2? - Absolutely! / scientific method

Mark parashakti108 at yahoo.com
Wed, 13 Apr 2005 01:20:25 -0000

--"denis bider" wrote:
> > -M: *IF* you want to be *scientific*, the two processes should 
only differ by ONE variable. In this case, that variable would only 
be the algorithm employed to analyze the data. The data should be 
identical if you want to use only one experimental group.
 
>D: The Condorcet portion of the collected data can be analyzed with 
either Condorcet or PageRank, and I'm planning on doing both. So I 
think that satisfies your above criterion (only 1 variable, for that 
part).

-M: Yes. This would be my mentioned groups #1 and #2.
 
>D: Additionally, I'm planning on collecting the non-candidate-
restricted data specifically for analysis with PageRank. True, the 
question will be on the same sheet (most likely), but I think it's 
sufficiently orthogonal to the first one that it won't corrupt the 
data that would be collected if there was just question 1 (ordering 
of candidates). Do you see a significant fault in this reasoning, or 
do you think it will be acceptable to have both questions on the same 
sheet?

-M: Yes, with your approach the non-candidate data, when filled out 
by a candidate, no endorsement pathways are allowed back to the 
candidate as a node - this is an anomalous constraint to place on 
PageRank. This reduces the sample size and introduces an unnessicary 
variable. So the third group, "experimental group #2" should have 
both candidate and non-candidate data analyzed by PageRank.

-M: If you want to analyze non-candidates endorsing non-candidates, 
this could be a *fourth* group, "experimental group #3". This would 
have one variable different from "experimental group #2".

-M: All data should be on the same sheet.

-Shanti
-Mark, Seattle WA USA