From the Stack Overflow list:
Answers with zero score having exactly two votes of +1, -1: 44,238 (1.1%)
Answers with zero score having exactly two votes in order of -1 then +1: 17,005 (0.4%)
is statistically significant if we simply assume counting statistics
sqrt(44237) = 210
sqrt(17005) = 130
(44237 \pm 210)/2 - (17005 \pm 130) = 5113 \pm 167
or about 31 sigma.
Of course it is still 5000/3.8e6 so it is a small correction, and it go against the direction of pity voting.
Following that model for all the above data
Stack Overflow Questions - 13
Answers + 31
Server Fault Questions - 0.7
Super User Questions + 0.5
The interpretation of the table is as follows:
- a pattern of pity voting would generate a negative value for sign
- significance less than ~3 isn't evidence for anything
- significance more than ~5 is evidence for something, but doesn't prove what.
Both stack overflow results show large significance, but of different signs. The Server Fault and Super User data show no significance.
Not sure what to make of that.
I have developed a half-assed theory (pure speculation, really) that could account for the difference between questions and answer. It may be the question-posters are (a) likely to vote for any answer they received and (b) likely to do so soon after it appears.
If there is anything to that it would mean that some fraction of the (+1,-1) answers are the result of people trying to de-emphasize so-so answers that the OP voted for.