It may be helpful to describe the process and the information that comes out of the daily absentee ballot analysis that I have been posting, which some would describe as 'early' (because this type of voting comes 'early' in terms of before Election Day, which the NC State Board of Elections would classify in-person absentee voting as 'early').
First, the data comes from the North Carolina State Board of Elections "Data Download" section (go to "ENRS", then scroll down to the file "Absentee11xx08xx2016.zip" (A WARNING: this is a huge file that is zipped, and unloading it makes it even bigger).
Upon opening that file in SPSS, I do a couple of things to first 'clean up the data.' First, I record the "Return Status" for "Accepted" and code that as a 1. This allows me to isolate the records for just accepted ballots, which are counted for the November 8th election.
Next, I run an identification for duplicate cases using the "County Description" and "Voter Registration" number, with a filter for "Accepted" and based on "Sent Date." Sometimes a voter will submit a ballot that is defective in some way, and the county board of elections will reissue a ballot to the voter. This identifies the duplicates in the data file so that the same voter is only counted once (insert "rigged election" joke here). Those duplicate records are then deleted.
Then, I recode the fields to create numeric identifiers for voter party, generation, and type of ballot, either a mail-in ballot or in-person ballot.
Then, I can begin running different analyses on the types of ballots, which is important to understand. North Carolina has two types of absentee ballots.
The first type of absentee ballots are mail-in ballots, which can be requested by mail, in-person, fax, or electronic delivery. Those are requested, sent out by the county board of elections, and then have to be returned and reviewed as "accepted" ballots.
The second type of absentee ballots that North Carolina uses is in-person ballots, in which a voter goes to a voting site and requests a ballot, just as if it's election day. The voter fills out the ballot, then submits it 'early' before election day.
Sometimes these two types of ballots aren't "accepted," for a variety of reasons: spoiled, no witness signature, no voter signature, etc. I post each type of ballot and the various categories that the ballots are in: accepted, spoiled, etc.
So, when you have the cleaned-up data file, you may have all of the ballots--sent out mail-in ballots, returned mail-in ballots that are 'accepted' or 'rejected', in-person ballots that are 'accepted' or 'rejected,' all with a requested, sent, and returned dates.
The first number that I use when I create the analysis is to look at the two different types of ballots--whether sent, returned, or accepted--and see how many there are; that's the overall ballots that are in North Carolina. I usually use the 'sent' date for this analysis, simply to make it comparable to the 2012 absentee ballot data that I use for comparison purposes. I also use the 'return' date, especially for analyzing mail-in accepted ballots.
Then, I analyze each type of ballots (mail-in or in-person) for the ones that have been 'accepted'; remember, that in the case of mail-in ballots, there are still outstanding ballots that haven't been returned and evaluated to be accepted.
Next, I begin the different types of analysis--by gender, race, age/generation, native/born-out-of-state, region (urban, suburban, rural) and most importantly, party registration for all of the ballots and then for each type of ballot (mail-in or in-person).
After that, I merge this data with the voter registration data and information on whether the voter voted in a past election, like 2012. This information is based on matching both files using the county description and the voter registration number within both files.
While others are using different methods and categories for classifying the absentee ballots, my main approach is broken down into: all ballots, the types of ballots based on date sent, and the accepted ballots using the return dates.