Not Quite a Blog

I've just finished an update to my Election Audit Bibliography.

Find a PDF and BibTeX here:

http://josephhall.org/eamath/bib.pdf
http://josephhall.org/eamath/eamath.bib

Let me know if you see any changes I need to make.

Some of you may know that I've been working, for over two years now, with a bunch of other people to develop a set of generic procedures that California counties can use to conduct their manual tally. (The manual tally is a legally-mandated "automatic recount" of ballots cast in 1% of precincts.)

Well, I've finished this set of procedures and you can find them here:

Comments welcome!

I have an academic paper outlining how we constructed these procedures but I can't yet share that document.

Note that we worked closely with officials in San Mateo county and more indirectly with Alameda, Marin and Yolo counties to do this research. San Mateo uses a more specific set of procedures in their elections, adjusted for their county and voting system.

I've been involved in a lengthy project to create procedures that any California County can use to do their 1% Manual Tally (as prescribed by CA Election Code §15360). In the process, I've observed the selection and manual tally in a few California counties and have taken some photos and video.

Here are some photos and video from Marin, Alameda and San Mateo County's random selection and manual tally.

Warning: These images and videos will probably be abysmally boring unless you both appreciate elections and are really interested in the intricacies of election procedure.

San Mateo County Random Selection and Tally

Photoset:

Videos:

• "San Mateo 1% Selection -- roll 028"
• "San Mateo 1% Selection -- misroll, roll 166"
• "San Mateo 1% Selection -- Using dicebins.php"
• "San Mateo 1% Tally -- Team 2 Unsorted, cross-stack"
• "San Mateo 1% Tally -- Team 2 Sorted, cross-stack"
• "San Mateo 1% Tally -- Tallying Nos for Proposition"
• "San Mateo 1% Tally -- Team 1, Obama tally"

Marin County Random Selection and Tally

Photoset:

Videos:

• "Marin 1% Tally -- Procedure (sort and stack)"
• "Marin 1% Tally -- Tallier"
• "Marin 1% Tally -- Showing sorting by partisan ballot"

Alameda County Random Selection and Tally

Photosets:

Videos:

• "Alameda 1% Tally -- Selecting a Precinct with Tumbler"
• "Alameda 1% Tally -- Choosing a Thousandths Place Digit"
• "Alameda 1% Tally -- Team 9, tallying proposition"
• "Alameda 1% Tally -- Team 11, tallying obama "
• "Alameda 1% Tally -- Team 2, no cross-stack"
• "Alameda 1% Tally -- Team 2, with cross-stack"
• "Alameda 1% Tally -- Team 7, cutting VVPATs"
• "Alameda 1% Tally -- Team 7, cutting totals tape"
• "Alameda 1% Tally -- Team 7, counting VVPATs"

This is probably most interesting to the election officials out there. (although I point out a few points of academic interest at the end)

I've written a quick calculator based on the Cordero, Wagner and Dill paper (the 10-sided dice paper) that election officials can use to produce "bins" of random numbers for manual tally spreadsheets. That is, say you have to choose one of 13 precincts, but if you use 2 dice and just read the numbers off as corresponding to the precinct number, you might be rolling for a while. This calculator will produce ranges (or "bins") to help you optimize your dice rolling. To use it, you enter the number of dice you'll be using and the number of precincts (or units, whatever) that you have to select. The program will output bins so that you don't have to roll the dice as much. It also has a mode that allows for easy copying and pasting into a column of a spreadsheet.

Comments, patches, etc. welcome.

Why did I do this? Can't you just roll dice without binning? Well, to give you some perspective, it took us 30-45 minutes in November of last year to select 20 precincts for San Mateo County's manual tally. We did a lot of re-rolling. In fact, we got so frustrated at having to reroll so much that we started to make up rules (one die would be even-odd, one would be divide-by-3, etc.). I realized that this frustration was actually dangerous for the random selection as it injects a good deal of ad-hockery into a security-relevant process. I can imagine that it might also be useful as we start to do larger selections in more rigorous manual tallies.

A few interesting comments:

• The tool defaults to 2 dice and 13 precincts. This is an interesting case because 1) you would normally have to reroll 87% of the time and 2) with a number of precincts that is a prime, it's difficult to calculate these bins in one's head (it is for me).

• If you input 3 dice and 501 precincts, you see that you'll have to reroll roughly 50% of the time:

  However, if you add one more die (for 4 die total), it drops the number of rerolls by a factor of 10 to roughly 5%. To do this in practice, you'd want a forth die along with the recommended red-white-blue set of three... I suggest a clear die.

• Here's what LA would have to do with 4 die and 4766 precincts and it gets them a ~5% reroll rate (with no bins, 4 dice would have a ~52% reroll rate for LA):