Missouri Increases Security Checks (Manual Audits) on Voting Equipment

[Wilms]

Pretty good news out of MO.

Audits can be good things, particularly when done by hand and when done to enough ballots. Now, to be sure, a flat 5% audit, while better than a flat 1% audit, may not give good confidence levels in a really tight statewide race. (Forget about congressional or more local contests.)

Perhaps one of the forum geeks can crunch the numbers to give an estimate on how "loose" the race would have to be in order for the 5% audit to yield a 99% chance (confidence) that a wrong result would be detected.

E-Vote: Missouri Increases Security Checks on Voting Equipment

September 16, 2008

snip

Carnahan issued procedures to local election authorities that expand the manual audit conducted after Election Day to ensure that voting equipment is accurate and secure.

The required post-election audit will increase from 1 percent to 5 percent of precincts, giving Missouri one of the highest audit standards in the country. The audit compares electronically tabulated results with a hand recount of votes in randomly selected precincts in each county to ensure the accuracy of voting equipment.

snip

http://www.govtech.com/gt/412367?topic=117671

[Gore1FL]

1. It has been a while...

but technically it should be 5% +1

In any event:

Finite population correction factor = SQRT((N-n)/(N-1)) where n= sample size and N= population size.

[Bill Bored]

2. Need to know how many precincts they got in MO and how many voters in each of 'em. nt

[Stevepol]

3. If this audit is carried out properly, this is wonderful news,

the best I've heard in a long long time.

[OnTheOtherHand]

4. some guesswork, but MO has around 5500 precincts

so a 5% sample is around 275. That has about a 99% chance of detecting a magnitude of error that occurs in up to 90 of the precincts. How much difference could 90 precincts out of 5500 make in the outcome? Conceivably a few points' worth -- it depends on assumptions about how much error can be packed in each precinct.

If MO law and/or these new provisions have some way of auditing precincts with anomalous results, that would help a lot.

By the way, 99% is really high. If there were large miscounts in 40 precincts, a simple random sample of this size would still have about 7 chances in 8 of including at least one of them.

[Wilms]

5. Yeah. What happens if an anomoly is detected.

That's a really important part. As is the margin at which the state recount law, if any, kicks in.

[Bill Bored]

6. Depending on the prcinct size distribution and all that, a race closer than a 1.5% margin

or so might not meet the 99% detection/escalation standard.

To answer Wilms' orginal question, I think wider margins than that would be safe, assuming there is proper escalation in the event discrepancies are found.