On Friday, November 12, 2010, I observed the random sampling for the New Mexico post election ballot audit. It took about 2 hours to randomly select (using 7 differently colored 10 sided dice) the 42 statewide precincts to audit. All ballots, including early, Election Day and absentee will be audited from each of the 42 precincts for Judge of the Court of Appeals-Position 1. This stateside ballot item represents the closest state contest (1.7% margin) and as such has the largest number of precincts for hand recounting. Other offices that will be audited include all three of the US House contests (CD1 22 precincts, CD2 5 precincts and CD3 4 precincts) and the statewide gubernatorial contest (9 precincts). Each of these offices will be recounted using a subset of the 42 precincts that were picked for the smallest margin.
The law is meant to ensure with at least 90% probability that faulty tabulators did not elect the wrong person for these offices. To my knowledge, New Mexico is the only state to base their audit on a risk limiting premise.
The precinct selection process unfolded in the following way. The auditors hired to do the sampling had 2 spreadsheets. The first, the “precinct” spreadsheet, contained the voter registration totals for each precinct in the state and a column that cumulates the total number of voters in the order they appear in the file. A second spreadsheet recorded each dice roll and identified the precinct selected by matching the resulting normalized number to the precinct whose cumulative number the normalized dice roll number is within. The normalized formula equals the number rolled on the dice +1 divided by 10,000,000 and then multiplied by the total number of registered voters (about 1.15 million voters). My understanding is that this method of selection creates an increased probability for larger precincts to be selected than smaller precincts.
In watching and thinking about the process, I think the biggest risk is that voter registration numbers are used for precincts instead of turnout totals. Although New Mexico had 1,152,821 million registered voters for the 2010 general election only about 602,285 participated in the election (about 52% turnout) and these are not randomly distributed across precincts since some precincts turn out at higher percentages than other precincts. These numbers were used, however, because they were available and there is likely a high correlation between precinct size and precinct turnout. However, and we had a lively and productive discussion about this, it would be much better to use actual turnout in selecting precincts. One possibility is that unofficial turnout data could be obtained for each precinct and these numbers would be much closer to actual turnout than voter registration numbers. Unofficial turnout changes only slightly and is mostly due to the addition of provisional voters to precincts and to data entry errors.
Hand recounting will begin this week and the law provides for an escalation procedure if discrepancies are found. Although there might be some problems with implementation and some may have some issues with the nature of the post election law, there is no doubt that New Mexico is a leader in election reform in this area and has made substantial progress over the last 4 year in implementing new voter systems and post election ballot audits.
I have attached a link that is a copy of the handout provided to us at the State Capitol where the random selection process took place. It includes the law, the procedures followed and an example of how the two files were set up to select precincts.