Recount ‘Em All, or None at All

Edward Glaeser
Edward Glaeser Fred and Eleanor Glimp Professor of Economics - Harvard University

November 13, 2000

There is a well-known trick among statistical economists for biasing your data while looking honest. First, figure out which data points don’t agree with your theory. Then zealously clean up the offending data points while leaving the other data alone. The key to maintaining academic dignity is to ensure that you do nothing to the data other than eliminate errors.

But while this approach may seem to improve accuracy, it actually leads to biased results. If you only clean the offending data points, then you will disproportionately keep erroneous data that agrees with your prior views. This leads many scholars to believe that data that is partially cleaned at the discretion of a researcher is worse than bad data.

This lesson from the ivory tower has a clear implication for the current mess in Florida. Hand counting ballots in only a few, carefully chosen counties is a sure way to bias the results. Even if hand counting is more accurate than machine counting, there is a clear bias introduced because Al Gore chose which counties to hand count. Mr. Gore has selected the state and counties where recounting has the best chance of helping him.

This is exactly the same as cleaning other data selectively. Naturally, if this opportunity for selective recounting becomes the norm, the floodgates will open and any candidate who loses a close election would be foolish not to demand a recount.

The immediate implication of this is clear. If there is to be recounting by hand, it cannot be selective. There needs to be total hand counting, not just within Florida, but across the U.S. in any state that was close. One candidate cannot be allowed just to choose where he wants the data cleaned. If this is prohibitively expensive, or time consuming, then it is better to leave the process unchanged than to introduce the selective recounting bias.

More generally, one of the principal lessons of macroeconomics is that rules generally work better than discretion. This is as true in elections as any place else. Giving candidates influence over how election results are processed does not help democracy to accurately reflect the will of the people. Judicial discretion is not much better, as judges will be responding to cases selectively filed by candidates. Furthermore, judges determining elections will exalt the judiciary to a king-making role it should not have.

While it certainly may be appropriate to ban butterfly ballots for all of eternity, and while reform of balloting procedures seems like a must, it is also clearly wrong to selectively recount certain areas.