According to the Drudge Report, Democrats are predicting a 210 vote win for Scott Murphy, after absentees and military votes are counted. They base this on a model they produced assigning where absentee ballots came from relative to who won which county on election day. In other words, they assume if Murphy won 51% of the vote in Columbia Country, then Murphy should win 51% of the absentee ballots from that county.
No, really. That's how they did their model.
Even though, according to actual statistics, there are more Republicans that submitted absentee and military ballots than Democrats.
After all, why use actual data when you can make up models that predict whatever you prefer the outcome to be?