The problem here is illustrated by some of the examples that came up in the case. In one state, Eastman’s experts are claiming 8000 people who registered to vote too young voted. They got that by running a query that did a date subtract on the wrong fields, because they didn’t understand how the fields are encoded. And it turns out that the mean age of those people was 42, and 99 percent of them had re-registered after they turned 18. Two of the people were 92 years old. If the “Experts” understood voting procedures they would have realized that just because you were first entered into the voting system when you were <18, that doesn’t mean you registered that same day. And even if you did, you can re-register years later, and subsequent votes are then legal.
Another example is that they found 140,000 votes to be illegal in Detroit because the votes were counted but not associated with any precinct number. The problem with that is it turns out that there is separate system, not involving precinct numbers, for absentee ballots - so the ”experts” had declared every absentee ballot to be illegal because they didn’t understand what the databases meant.
Another example is they declared a hundred thousand votes illegal in Pennsylvania because the “mail date” field and the “voted” date were the same. But they didn’t understand that people who vote overseas, the military, and anyone who shows up at the clerk’s office and does a same day registration and vote would have this happen. They declared “mail date means date it was postmarked” but had no evidence of that, and it wasn’t true.
Another example is the statistical “experts” who declared that votes must be rigged because Biden did better than Hillary, and significantly so in big democratic cities. They ignored the fact that if there is, say, a 5 percent shift across the board where Biden does better than Hillary, it would be a much bigger effect in big cities because more people vote there. (They looked at NUMBER of votes, not SHARE of votes). Worse, their premise was that any deviation from 2016 to 2020 vote share or quantity Is proof of fraud. They assumed that votes follow a gaussian distribution, but couldn’t explain why. When asked to name another statistical distribution, they couldn’t.
The point is, you need to understand the system you are evaluating otherwise your statistical analysis has no meaning.