Reply #29: depends...

[OnTheOtherHand]

If the canvassed margin of victory ends up being around 0.5%, then in principle a ballot-level audit of 1000 individual ballots could give you about a 90% "confidence" level (see Stark's S4RLA paper, Table 3) -- probably somewhat larger, allowing for some marking errors. But I doubt that a ballot-level audit is feasible in Wisconsin (unless someone could do a Humboldt-style rescan of all the ballots).

Wisconsin has something like 3500 reporting units, a bunch of which reported over 2,000 votes in the gubernatorial election. (I say "a bunch" because it would take a bit of work to get the data file into a format where I could say exactly....) So, on the most conservative assumptions, probably just one or two miscounts could account for the apparent margin. Not much reason not to go for a full recount, on those assumptions. I'm curious what one could get by with if willing to use a within-precinct miscount assumption, although at this point I don't think enough data are available to say. One could do pretty well with maybe a 20% sample, depending on how it was drawn.