My premise is that given a set of medium-sized towns which includes a statistically significant amount of both hand and machine counting, the ratio of votes received by each candidate from AccuVote machines should match the overall ratio for the towns. For example, for towns with 500-800 votes, 41% of total votes were by AccuVote, so I would expect the individual candidates to show 41% of their votes from AccuVote.
However Obama's AccuVote numbers come in significantly lower, and are consistently lowest among the candidates for various vote ranges I've tried. Conversely, Clinton's machine vote numbers are significantly higher than the overall ratio, and consistently highest among the candidates.
Data sources:
Stats for New Hampshire 2008 Democratic Primary
111 of 237 towns (46%) used AccuVote machines.
239632 of 284104 total votes (84%) by AccuVote machines.
Considering set of towns with 200-1100 votes.
135 towns (56%) had votes in this range, representing 79883 (28%) of total votes.
Within this set, 52 towns (38%) used AccuVote, representing 41129 (51.5%) of votes.
Comparison of hand and machine votes by candidate:
Clinton Obama Edwards Richardson
total: 28970 30397 14575 4278
hand: 13121 (45.3%) 15561 (51.2%) 6890 (47.3%) 2227 (52.1%)
accuvote: 15849 (54.7%) 14836 (48.8%) 7685 (52.7%) 2051 (47.9%)
Stats for New Hampshire 2008 Republican Primary
111 of 237 towns (46%) used AccuVote machines.
197251 of 233424 total votes (84%) by AccuVote machines.
Considering set of towns with 200-1100 votes.
130 towns (54%) had votes in this range, representing 66394 (28%) of total votes.
Within this set, 49 towns (37%) used AccuVote, representing 34714 (52.3%) of votes.
Comparison of hand and machine votes by candidate:
McCain Romney Huckabee Giuliani Paul
total: 26423 18012 8859 5733 6059
hand: 13029 (49.3%) 8130 (45.1%) 4227 (47.7%) 2736 (47.7%) 2940 (48.5%)
accuvote: 13394 (50.7%) 9882 (54.9%) 4632 (52.3%) 2997 (52.3%) 3119 (51.5%)