In the past 24 hours or so there have been, oh, a few diaries about the electoral reversal of fortune in Waukesha County, Wisconsin. Like many posters here, I have seen nothing that leads me to believe that there was foul play in Waukesha County. ("Nothing to see here, move along" comments begin in 4... 3....) That doesn't mean that I'm sure that there wasn't -- and it certainly doesn't mean that I think Kathy Nickolaus and her Access database did a great job. Based on the evidence I've seen so far, however, it appears at least strongly plausible that the Waukesha County canvassing board corrected an egregious error in the unofficial returns.
That raises some excellent questions: (1) How could the error happen in the first place? (2) How could we possibly know what really happened -- how people actually voted? (3) What are the implications for election verification, not just in Wisconsin but around the country?
In this diary, I will discuss my present understanding of Waukesha, but the main point is to think beyond the present controversy to broader systemic issues.
A good place to start is with this Journal-Sentinel article. Some points:
- Where did the votes come from? According to county clerk Nickolaus -- and corroborated by the Democratic member of the canvass board -- the election night totals she provided to the media omitted the votes from Brookfield city. Nickolaus had Brookfield's vote report, but for reasons she does not entirely understand, it was not included in the totals. (Smaller discrepancies occurred in two other places.)
- "Unofficial" means unofficial. I am not an expert in Wisconsin election law (feel free to dive in with me, here, especially chapters 5 and 7), but as far as I can tell, counties report election-night totals strictly as a courtesy to the media. It is the official canvass that they really have to get right.
- The county canvass is detailed. Under Wisconsin statute 7.60 (4)(c) [see link above], the county canvass board is supposed to check the tally tapes and inspector reports, then report how many votes each candidate received in each "election district, ward or combination of wards." So, the canvass would be a good place to discover that returns from an entire city were omitted from the unofficial totals.
- Wisconsin elections have paper trails. Most Wisconsin voters use hand-marked paper ballots with optical scanners. Some Wisconsin counties also provide touchscreen Direct Recording Electronic machines, to satisfy the accessibility requirements of the Help America Vote Act. (Those machines do provide a voter-verifiable "paper trail," which is not a great solution.) Inspectors also issue each voter a serial number, written on the poll list, which can be used to check the turnout at each polling place.
- These paper trails aren't necessarily used very well. As I understand it, a recount in Wisconsin ordinarily is a machine recount: the ballots are fed back through the scanners. Even though statute 7.50 provides standards for establishing voter intent, the scanners are not familiar with its provisions. There is a vague legal requirement for the Government Accountability Board, after general elections, to "audit the performance of each voting system used in this state to determine the error rate of the system in counting ballots that are validly cast by electors" (7.08 (6)). I'm not at all familiar with all the provisions in WI law for contesting elections, but if it were up to me, I would change these laws regardless.
As far as I can tell, Nickolaus's story seems to make sense. (The least coherent part is how the uploaded Brookfield data failed to end up in her database. She could be fudging on the details, or she may honestly not know exactly what happened.) The Brookfield city clerk reported the city results on election night, and apparently the same totals are reflected in the county canvass. As Nate Silver and Craig Gilbert have pointed out, the unofficial figures actually give Waukesha County surprisingly low turnout compared with other similar counties. There might be a good fraud scenario that connects these dots, but I haven't seen anyone present one yet. (It has been suggested that Brookfield was some sort of decoy, but it isn't obvious to me how attracting nationwide attention was a masterstroke of misdirection -- unless, perhaps, the election was actually stolen in some other county, which doesn't seem to be the prevailing idea.)
But how do we know what happened -- or, perhaps better, how could we know what happened? Here are some of the crucial criteria, in Waukesha County and elsewhere.
- Paper, paper, paper. Paper ballots aren't perfect, but most voters can see how their votes are recorded on the ballots -- whereas they cannot verify whether or how a Direct Recording Electronic machine has recorded their votes. (There are some proposed workarounds, which I won't discuss here.) Many voters are stuck voting on paperless DREs. Others use DREs with some sort of voter-verifiable paper trail; these paper trails tend to be much less verifiable and durable than paper ballots. See Verified Voting for details on who uses what. The tally printouts, poll lists and other written documents from polling places provide another important "paper trail."
- Security and speed. A paper trail may be worse than useless if it can be corrupted before it is checked. Careful security measures provide one mitigation; I have seen little if any information about Waukesha County's chain of custody provisions. Expeditious processes are also an important mitigation. The canvass board began reviewing the records on Wednesday, limiting the opportunity to do whatever one imagines might have been done to them.
- Provide detailed data early. How does someone manage to supply totals that exclude an entire city? By not looking at subtotals. If Waukesha County had a policy of publishing unofficial results at the election district or even municipality level, probably this mistake never would have happened -- and if it had, it would have been detected and understood very quickly. Minnesota publishes precinct-level results for the entire state in close to real time. It's a great model. As the Coleman/Franken recount underscored, even when the unofficial totals turn out to be wrong, having posted the details can provide a basis for public confidence that the corrected totals are in fact correct.
- Use the paper: good audit and recount policies. A paper trail doesn't do much good if it is practically inaccessible. I favor routine, rigorous audits to check that votes have been counted substantially correctly, combined with easy recourse to count ballots and inspect other records in particular places where the results seem questionable. In general, closer election contests deserve more extensive auditing. The closest election contests deserve 100% hand recounts. (Some folks would like to see 100% hand counts for all election contests -- and some places do use 100% hand counts, although I don't really expect that practice to spread widely.)
These are simple ideas, but they are hard to get right. Wisconsin does seem to have at least enough of them right that the upcoming investigation in Waukesha County has some prospect of clarifying what did and didn't go wrong there. But Wisconsin, and every other state I can think of, could implement election verification a lot better. (Every state could also do much worse, and may yet -- internet voting, anyone?) So, if we're interested in getting elections right, we should be busy for a while.