Reply #188: Statewide

[Febble]

certainly wouldn't be the appropriate unit of analysis for a congressional district. But I am wondering whether, after all, the most appropriate unit of analysis is the unit that determines the outcome of any given election. For local elections it would be local. For a presidential election it would be state.

Of course if you have an a priori reason to suspect a specific jurisdiction, then you would want an audit at that level, with an appropriate sample size.

And I agree, one if the issues that has to be determined is what constitutes a "clean" audit - how much tolerance do you allow?

But statistically, I think it is more useful to ask: what could I conclude from an audit of N precincts that found no anomalies, than to ask what N you need to detect a certain percentage of corrupt precincts. I.e. to decide what is the highest acceptable percentage you are prepared to tolerate, then compute the statistical power (i.e. the sample size) you need to detect anything higher with a given degree of confidence.

But the more I think about it, I think that a percentage is the wrong thing to mandate. I think sample size, or even statistical power is what needs to be specified.