TREC-6: Comments
- Best model for each site; cross-site ANOVA finds some system difference(s) significant; multiple comparisons test does not.
- Half the effort expended on the control, which was not of special interest
- Two attempts to confirm the effectiveness of the control for cross-site comparisons did not succeed - samples too small and confounding searcher characteristics
- Questions about the effect of disagreements between searchers’ and assessors’ view of the aspect space.