130 likes | 340 Views
Design, Findings, and Lessons Learned: Sample Audit Recounts in 2006 North Carolina Elections . William D. Kalsbeek Lei Zhang University of North Carolina, Survey Research Unit, Department of Biostatistics E-mail: bill_kalsbeek@unc.edu. Background.
E N D
Design, Findings, and Lessons Learned: Sample Audit Recounts in 2006 North Carolina Elections William D. Kalsbeek Lei Zhang University of North Carolina, Survey Research Unit, Department of Biostatistics E-mail: bill_kalsbeek@unc.edu
Background • The NC Board of Elections asked the UNC Survey Research Unit (SRU) to design and conduct an election recount audit for 2006 primary and general elections • A 2006 bill passed by the NC Legislature now mandates that a “ hand-to-eye” recount be done for national/statewide offices in each election • Little mention of how the recount data are to be analyzed • Recounts completed thus far: • May 2006 primary – State supreme court associate justice seat (five candidates) • November 2006 general election – State supreme court chief justice seat (two candidates)
Sampling Precincts/Places • Stratified random sample precincts/places • NC has 3,047 precincts/places overall • 100 counties as strata (sampling in each county required by NC-BOE) • Total precinct/place sample sizes: • n = 200 (6.6%) for May primary election • 2 per county • n = 264 (8.7%) for November general election • 2 or more per county • More than 2 to the extent of May discrepancies)
Recounting the Votes • Selected precincts/places announced after each election • Bi-partisan recount: • Generally followed hand-count procedures • Teams of 3-4 from each political party • Team members rotate duty as “tallier” and “caller”
Two Types of Vote Count Discrepancies in Precincts/Places • Discrepancy in Candidate Count (DCC) • In vote count for each candidate on ballot • Discrepancy = [Election Count] – [Recount] • Discrepancy in Total Count (DTC) • In total vote count for all candidates • Discrepancy = [Election Count] – [Recount]
ELECTION: Candidate E1 E2 E3 E4 ___________ Total E Count RECOUNT: Candidate R1 R2 R3 R4 ___________ Total R Count All DCC Discrepancies DTCDiscrepancy Discrepancies at Each Precinct/Place
TABLE 1Estimated % Distribution of DTCsAmong All Precincts/Placesfor Five Statewide Candidates in May 2006 PrimarySupreme Court Associate Justice (Wainwright Seat) • FINDINGS: • Total votes reported in the election = 519,615 • Range of DTCs: -4 to +4 • Over-count vs. under-count: favors undercount somewhat • Discrepancies much less likely for iVotronic than M100
TABLE 2Estimated % Distribution of All DCCs Among All Precincts/Placesfor Five Statewide Candidates in May 2006 Primary Supreme Court Associate Justice (Wainwright Seat) • FINDINGS: • Total votes reported in the election = 519,615 • Range of All DCCs: -2 (undercount) to +3 (overcount) • Overall over-count vs. under-count -- very slightly favoring undercount • Discrepancies equally rare for iVotronic and M100
TABLE 3Estimated % Distribution of DTCsAmong All Precincts/Placesfor Two Statewide Candidates in November General ElectionElection for State Supreme Court Chief Justice • FINDINGS: • Total reported votes = 1,707,326; 2 to 1 margin of victory = 569,366 • Range of DTCs --- mostly -13 to +13 • Discrepancies of this type are more likely than in May primary • Over-count vs. under-count: slightly favors undercount • Discrepancies much less likely for iVotronic than M100
TABLE 4Estimated % Distribution of AllDCCs Among All Precincts/Places for Two Statewide Candidates in November 2006 General ElectionElection for State Supreme Court Chief Justice • FINDINGS: • Total election votes = 1,707,326 ; 2 to 1 margin of victory = 569,366 • Range of AllDCCs: mostly -14 to +12; with outlier at +86 • Discrepancies of this type are more likely than in May primary • Overall over-count vs. under-count: slightly favors undercount • Discrepancies much less likely for iVotronic than M100
Summary of Key Findings • May primary and November general election: DTCs and DCCs in precincts/places • Both + (indicating overcount) and - (indicating undercount) • Slightly favoring – (undercount) • Greater discrepancies in November general election than May primary • >3 time as many votes cast in November • Greater discrepancies (in both directions) in precincts/places using M100 voting machines than in those using IVotronic machines
Lessons Learned in Conducting North Carolina Election Audits • Sampling • Must be random • What do we need to learn from an audit? Sample design must be responsive to this. • Drop county focus? • Sample more intensively where there has been disparity • Sample locations not announced until after the election • Data gathering • Recounting should be “blinded” to election count • Think about other practical ways to make the recount a better gold standard • Expect a few process “glitches”