Wisdom Without Waiting: Statistical Variability and the Presidential Election
© John L. Mariotti 2000
The credit for this column goes to a friend of mine who reminded me of an important fact at a dinner party tonight. No one can ever truly know which Presidential candidate got more votes in the Florida election! Forget all the pleas to count all the votes and make every vote count. Forget all the lawyers and judges and appeals. The total process is simply not within enough process control to reliably determine the outcome in a race this close.
This is an immutable statistical fact about the Florida Presidential election. No matter how the votes are counted (or even interpreted) the inherent error in the process–quality managers call it process variability–is greater than the "winning" candidate’s margin of victory.
I am not a statistician, so some specialist might pick some nits in my story, but the conclusion is inescapably true. All naturally occurring things have an inherent variability. The familiar "bell curve" describes it graphically. Measure the sizes of people, the weights of fruit, the dimensions of parts made by the finest, most precise machine, and guess what--each of them has inherent variability–its own bell curve.
Statisticians know how to compute indicators about how much variability a given set of data has. They calculate things like the mean (average), the median (middle number–equal points on each side of it), the range (how widely distributed the extremes are), and the ultimate measure of central tendency–sigma–the standard deviation. Then they can calculate upper and lower control limits which, simply stated, tell you just how accurately you can conclude anything about the process–with statistical certainty.
You’ve heard of "Six Sigma", unless you’ve been locked away in a Florida basement dimpling chads for the past ten years. Six Sigma describes a process with very little variation or error–about three parts per million that may be out of the specified tolerance. This is a very, very, very, very precise level of quality and variability.
Now, back to the Florida election. If everything else were perfect, neither the punch card counting machines nor the most careful hand counters can approach anything close to six-sigma accuracy. Quality assurance people have known for years that 100% inspection–by people–is far from 100% accurate. It’s more like 98-99% accurate, and that is if the standards are clear, the people well trained, the gauges used well calibrated–you get the idea.
If we consider that the Florida election counted approximately 6 million votes (or attempted), then a near perfect (six sigma) process would still have a variation error of 20 votes.
Does anyone who has watched this debacle believe that anything done there even approached Six Sigma? I sure don’t. More likely, this election was counted with the accuracy of your normal airline baggage handlers, or about 4-5 sigma–which is around 1000-2500 parts per million variation–due to inherent flaws in the process, machine errors, human errors, and so forth. That means anyone attempting to tally the Florida closer than 6,000 votes has as good a chance to be wrong as right. Scary? Sure. Stupid? Sure. But it happened, and the same logic applies to all those other places with close races.
Only if hand counts are impeccably controlled, to standards that are clearly established and imposed by well trained, diligent (and not fatigued) counters, can the 6 sigma level be approached–and that is with reliable mechanical assistance and careful re-verification. (This means that people count, then have machines verify, then people verify again–and only accept the tally if all 3 agree exactly!) That is why data input in the punch card era required the data entry person to enter the data twice and have it match exactly.
But the technology exists to achieve six sigma–in the absence of human meddling, fraud or electronic malfunction. This technology is all over the U. S. and it is called an ATM (Automatic Teller Machine)–which dispenses cash, accepts deposits, and must balance exactly to accounts in a parent bank each and every day. The technology exists to verify that the holder of an ATM card is who they claim to be–called a photo ID–required to board a commercial airline flight or drive a car (or if you want to get fancy methods like retinal scans, fingerprints, etc. could be used)
If this is so, why are we still spending hundreds of millions of dollars a month after the election trying to figure out who the next President will be? If we invested the rounding error in the U. S. budget (a few $billion here or there) we could equip every voting precinct in the U. S. with voting machines like ATMs.
We could link them to computers, require every registered voter to acquire a photo ID voting card. (I’d bet Wal*Mart would help with this. Maybe even the US Postal Service could chip in, too!) Voters could validate the voting card like a credit card upon receipt (via an 800 call with help from Visa or MasterCard?). Then election officials could identify voters, electronically, and visually when they came to vote and voters could cast their votes with a confirmation question after each vote via ATM like keys.
Tallies could be nearly instantaneous, and uniform polling place hours, irrespective of time zones could be imposed. Absentee ballots would have to carry the imprint of the voter’s card and a PIN (Personal Identification Number) along with the voter’s signature. Can fraud still happen–sure. Will systems break down? Sure–backups must be available and in place. Might someone still have to fall back on old-fashioned optical readable, hand marked ballots? Sure. But when all is done isn’t selecting the leaders of the largest, wealthiest, and most emulated democracy in the world worth that investment.
The last point, for heaven’s sake, don’t expect the government to manage such a process–let the contract to a consortium of private sector companies who know what they are doing. Then process variability would not lead to a media circus and legions of lawyers deciding who was elected President of the United States.
This should be the first official act of President Bush and Congress followed shortly after by the McCain-Feingold Campaign Reform Bill and a tax cut for all Americans.
