Consider that the monitor misses a vote, in the sense that between the previous and current refresh two new votes were cast on your image and let's say that the votes it missed were a 6 and a 4. The program has no access to the actual votes, so it can only infer that 10 points over two votes were added to your score. So it guesses which votes were cast based on a simple vote estimation model that I wrote. It might guess that 6 and 4 were cast, or it might guess that 1 and 9 were cast, or maybe 2/8, or 3/7, etc. The model uses your current average and variance of votes in order to make it more likely that 6/4 is guessed rather than 9/1, but this doesn't make it right. If it misses 10 votes or more this vote estimator needs to guess a lot.
Furthermore, there is a tendency for voters to change their votes after initial casting. In this case one vote in the total set of votes you have got changed, and once more I have not an iota which vote that was, so I guess and adjust a vote which is more likely than not not to have been the one changed by the margin that I inferred from the change in score.
So, when I wrote the program I said that because I didn't know any of these details, that it was not appropriate for me to make it seem as though I did, so every time the refresh button is pressed, you get different estimates of the vote distribution that results to the same average score. This means the program works best when left on 24-7, and pretty inaccurately when only turned on a few times a day.
|
