Voter intention, data and measurement

In the same article, David Dill's views on data are addressed:

But Dill noted that the data that voting machine vendors and academics generally use to evaluate the integrity of e-voting machines doesn't include the kinds of problems that voters have been reporting. Generally, the number of undervotes and overvotes on a machine are used to measure their effectiveness.

Undervotes occur when a machine records no choice for a particular race -- either because the machine failed to record it or the voter chose to skip the race. Overvotes occur when voters choose more options or candidates than the race allows. E-voting machines are supposed to make it impossible for voters to overvote.

But "recording a vote for a wrong candidate is not something that shows up in the statistics," Dill said. It doesn't show up in the statistics because officials have no way to know whether a machine incorrectly recorded votes. Without a paper trail or some other way to independently verify that the votes on the machines are the votes voters intended, there's no way to truly measure the accuracy of the machines.