Yves - I don't understand your point. In your scenario, by the time the minimum 5 rankings have been received, assuming the first was a very low one from a troll and the remainder were all nearer the high end, then the troll's marking would become an outlier beyond the limits of the bell curve and would be omitted from the overall ranking. This would occur irrespective of when the troll posted, first, last or somwehere between.

Of course it isn't perfect. As has been pointed out, "block voting" from a group of trolls or more likely a troll with "multiple personalities" could still sway the ranking - in the extreme case more strongly than before, as a mass attack could cause the fair votes to actually be discounted as lying outside the accepted region thus having absolutely no impact on the overall result! Hopefully though such cases would be glaringly obvious and would be reported for suitable exclusion action against the individual(s) concerned.

I do agree with Nish's point though. The lowest few terms are so pejorative that I think few other than trolls would consider using them, preferring not to rank in such cases. Obviously since you are computing standard deviations, you are converting these verbal rankings into numerical equivalents. My guess is that these are in a simple arithmetic progression by unit decrement. If so this could be construed as a bit misleading, as the verbal representations don't seem to me to form an especially linear progression between the extremes, and in any case are open to individual interpretation, a particular problem in a multi-lingual community such as this. I would have thought simple numeric marks out of ten would be easier to grasp. I am still of the opinion that removing anonymity from the coting would be an effective deterrent to trolls.