Kerkhoffs' Law
So two things that generally interest me are election theory and the news media not doing its job, and after a ten-year break they've gotten back together in Wisconsin.
Anyone still reading this probably already knows the context, but here it is anyway. Wisconsin, like way too many other states, chooses judges by popular election. Normally this is a waste of everybody's time and money as people either vote for the incumbent or do the rough equivalent of coin-flipping. Now apparently there's a one-vote conservative majority on the state Supreme Court, and David Prosser, one member of that majority, was up for re-election last week. Normally this would be a snoozer, but between the massive labor uprising in the state, and a couple of comments suggesting that Prosser would prejudge politically charged cases, and some other comments calling his general fitness for office into question, the thing started looking close. Naturally, everyone decided this was a proxy fight, and both sides of the labor fight threw lots of resources at it. (I was rooting pretty hard for the challenger, Joanne Kloppenburg, about whom I knew and know nothing.)
Anyway, on election night - or, rather, Wednesday morning - unofficial counts suggested the challenger had won by roughly 200 votes. Then, during the canvass Thursday evening, the Waukesha county (exurban Milwaukee) clerk announced that she had forgotten to count the city of Brookfield, the second-largest city in the county. Adding them back in swung 7500 (=11000-3500) votes to Prosser.
Naturally, Democrats cried foul. There were reasons for this, other than the obvious: First, the clerk in question is a known Republican partisan who had been involved in election probes before. Second, she was using a highly nonstandard, jealously guarded setup to do the counting (so the public couldn't see what was going on; this made errors like omitting an entire city possible, and also may have made wholesale fraud possible), for which she'd come under heavy fire in the recent past. Third, 7500 votes was almost exactly what Prosser needed to move out of automatic-recount territory (Kloppenburg made up ground elsewhere in the canvass, so that didn't quite happen).
So there are two equally implausible stories out there now:
1. Kathy Nickolaus made a major data entry error Tuesday night, and didn't notice until much later. The official version here is that she failed to save (but that's apparently impossible in Microsoft Access, so more likely she failed to paste or equivalent, and doesn't know the difference), discovered the error around noon Wednesday, then didn't tell anyone until early Thursday evening.
2. Kathy Nickolaus decided to count the votes from Brookfield City twice, then spent all day Wednesday and most of Thursday covering her tracks.
There's also a third story, which is even more absurd, about which more later:
3. Kathy Nickolaus intentionally modified the count from Brookfield City, while leaving the rest of Waukesha County untouched ... and spent all day Wednesday and most of Thursday covering her tracks.
If Story One is true, then Nickolaus is at least the moral equivalent of criminally negligent, for creating an opaque system where this sort of error is possible. If mistakes of this magnitude can happen, smaller mistakes can go undetected, and how can anyone in Wisconsin trust the results of a close election ever again? She needs to be replaced, and quickly. Meanwhile, Story Two needs to be debunked so people can have confidence in the results of this election. That means all relevant records have to be made public, immediately. This hasn't happened, which again amounts to negligence on Nickolaus' part, and is another reason she needs to be replaced.
If Story Two is true...
Here's how I would steal an election, assuming I had access to a completely closed database of votes. First, I'd decide how many votes my candidate needed, then I'd go find a precinct or city or something which was giving him as close to that margin as possible. Now I need to go about double-counting that city. Suppose the city actually cast {K1,K2} votes, and the rest of the county cast {N1,N2} votes for the respective candidates. I need to distribute those K votes among the rest of the county, so I'm going to multiply all the precincts by (K1+N1)/N1 and (K2+N2)/N2. Of course this won't give integers, which is a problem, so now I'll round everything down, then distribute the missing votes at random. At this point I'm reasonably confident that I can't be caught with the power of Statistics (any statisticians know?), but I'm worried for two reasons. First, this is ridiculously easy to reverse-engineer, and I'm afraid that the number theorists or whoever might still be able to find a signature. Second, I have a sense that some form of variance was probably supposed to have increased as the numbers went up, and it almost certainly didn't. So, I'm going to shoot a couple thousand votes back and forth among the precincts (one at a time, again at random), in an attempt to kill any signature I didn't know about. At this point everything should be more or less exactly as in line with demography and local history as the true results, except that turnout may be a little high - but that's unavoidable. If turnout in the copied city is now obviously low, I'll give it the same treatment. If not, better leave well enough alone. Now I announce that I mysteriously didn't count my city, and release the new numbers. I'm reasonably confident that the mathematicians won't be able to touch get me ... provided that nobody anywhere has any of the original data.
So of course people are (or were at first) trying to use advanced math to prove that something's wrong. There was talk, for example, of using Benford's Law as was done with Iran a couple years ago, as if it could say anything at the precinct level, or with the small sample sizes at larger levels, or as if multiplying everything by 1.1 would do anything to that distribution. Of course this, as well as anything that might have worked, has been stymied by Waukesha's apparent refusal to release precinct-level data, which is also absurd. The Chapel Hill Newspaper used to publish that data Wednesday morning when I was a kid. Dane County publishes it now. (Those are post-canvass numbers, but the unofficial ones were there Wednesday.) Waukesha county used to publish it, I guess until Nickolaus implemented the new super-secret system. Again, this does not reflect well on her job performance.
So Story Two can't be detected with advanced math until the base data comes out (and probably not even then), which it appears will never happen. However, it can be detected with basic math, because there are lots of people out there who do have the original data.
For example, I know that Story Three is false. I know this because the unofficial Brookfield tally was published in the local paper on Wednesday morning, and it is exactly the same as the official tally we're seeing now. Attempts to study Story Three are actively harmful to the study of Story Two, because they hide the fact that we know these things.
And if we know the preliminary results from Brookfield City, we can know all the preliminary results. The Brookfield City clerk gave these to the local paper Tuesday night; it follows that every city and township clerk had their results Tuesday night, and in many cases probably gave them to the local papers. All one has to do is track down all the city and town clerks and ask them for the pre-canvass, Tuesday night numbers. Then publish these for all to see; we can all add them up for ourselves. The results may be off by a couple hundred from the true totals, but they will be off by roughly 7500 from those predicted by either Story One or Story Two.
If the numbers agree with Story One, but Story Two is true, then every clerk whose district was modified is necessarily in on the conspiracy; this is either blindingly obvious from turnout or in moon-landing territory.
Why has nobody done this (or, if they have, why can't I find it)? I can't; I don't know what all the cities are, and probably can't get the right people to talk to me. Certainly it's not possible for every Wisconsinite to do it individually. But this seems to be exactly the sort of thing that reporters are supposed to do all the time.
Instead, everything I see about this is trying to analyze Story Three. ARGH!
Anyone still reading this probably already knows the context, but here it is anyway. Wisconsin, like way too many other states, chooses judges by popular election. Normally this is a waste of everybody's time and money as people either vote for the incumbent or do the rough equivalent of coin-flipping. Now apparently there's a one-vote conservative majority on the state Supreme Court, and David Prosser, one member of that majority, was up for re-election last week. Normally this would be a snoozer, but between the massive labor uprising in the state, and a couple of comments suggesting that Prosser would prejudge politically charged cases, and some other comments calling his general fitness for office into question, the thing started looking close. Naturally, everyone decided this was a proxy fight, and both sides of the labor fight threw lots of resources at it. (I was rooting pretty hard for the challenger, Joanne Kloppenburg, about whom I knew and know nothing.)
Anyway, on election night - or, rather, Wednesday morning - unofficial counts suggested the challenger had won by roughly 200 votes. Then, during the canvass Thursday evening, the Waukesha county (exurban Milwaukee) clerk announced that she had forgotten to count the city of Brookfield, the second-largest city in the county. Adding them back in swung 7500 (=11000-3500) votes to Prosser.
Naturally, Democrats cried foul. There were reasons for this, other than the obvious: First, the clerk in question is a known Republican partisan who had been involved in election probes before. Second, she was using a highly nonstandard, jealously guarded setup to do the counting (so the public couldn't see what was going on; this made errors like omitting an entire city possible, and also may have made wholesale fraud possible), for which she'd come under heavy fire in the recent past. Third, 7500 votes was almost exactly what Prosser needed to move out of automatic-recount territory (Kloppenburg made up ground elsewhere in the canvass, so that didn't quite happen).
So there are two equally implausible stories out there now:
1. Kathy Nickolaus made a major data entry error Tuesday night, and didn't notice until much later. The official version here is that she failed to save (but that's apparently impossible in Microsoft Access, so more likely she failed to paste or equivalent, and doesn't know the difference), discovered the error around noon Wednesday, then didn't tell anyone until early Thursday evening.
2. Kathy Nickolaus decided to count the votes from Brookfield City twice, then spent all day Wednesday and most of Thursday covering her tracks.
There's also a third story, which is even more absurd, about which more later:
3. Kathy Nickolaus intentionally modified the count from Brookfield City, while leaving the rest of Waukesha County untouched ... and spent all day Wednesday and most of Thursday covering her tracks.
If Story One is true, then Nickolaus is at least the moral equivalent of criminally negligent, for creating an opaque system where this sort of error is possible. If mistakes of this magnitude can happen, smaller mistakes can go undetected, and how can anyone in Wisconsin trust the results of a close election ever again? She needs to be replaced, and quickly. Meanwhile, Story Two needs to be debunked so people can have confidence in the results of this election. That means all relevant records have to be made public, immediately. This hasn't happened, which again amounts to negligence on Nickolaus' part, and is another reason she needs to be replaced.
If Story Two is true...
Here's how I would steal an election, assuming I had access to a completely closed database of votes. First, I'd decide how many votes my candidate needed, then I'd go find a precinct or city or something which was giving him as close to that margin as possible. Now I need to go about double-counting that city. Suppose the city actually cast {K1,K2} votes, and the rest of the county cast {N1,N2} votes for the respective candidates. I need to distribute those K votes among the rest of the county, so I'm going to multiply all the precincts by (K1+N1)/N1 and (K2+N2)/N2. Of course this won't give integers, which is a problem, so now I'll round everything down, then distribute the missing votes at random. At this point I'm reasonably confident that I can't be caught with the power of Statistics (any statisticians know?), but I'm worried for two reasons. First, this is ridiculously easy to reverse-engineer, and I'm afraid that the number theorists or whoever might still be able to find a signature. Second, I have a sense that some form of variance was probably supposed to have increased as the numbers went up, and it almost certainly didn't. So, I'm going to shoot a couple thousand votes back and forth among the precincts (one at a time, again at random), in an attempt to kill any signature I didn't know about. At this point everything should be more or less exactly as in line with demography and local history as the true results, except that turnout may be a little high - but that's unavoidable. If turnout in the copied city is now obviously low, I'll give it the same treatment. If not, better leave well enough alone. Now I announce that I mysteriously didn't count my city, and release the new numbers. I'm reasonably confident that the mathematicians won't be able to touch get me ... provided that nobody anywhere has any of the original data.
So of course people are (or were at first) trying to use advanced math to prove that something's wrong. There was talk, for example, of using Benford's Law as was done with Iran a couple years ago, as if it could say anything at the precinct level, or with the small sample sizes at larger levels, or as if multiplying everything by 1.1 would do anything to that distribution. Of course this, as well as anything that might have worked, has been stymied by Waukesha's apparent refusal to release precinct-level data, which is also absurd. The Chapel Hill Newspaper used to publish that data Wednesday morning when I was a kid. Dane County publishes it now. (Those are post-canvass numbers, but the unofficial ones were there Wednesday.) Waukesha county used to publish it, I guess until Nickolaus implemented the new super-secret system. Again, this does not reflect well on her job performance.
So Story Two can't be detected with advanced math until the base data comes out (and probably not even then), which it appears will never happen. However, it can be detected with basic math, because there are lots of people out there who do have the original data.
For example, I know that Story Three is false. I know this because the unofficial Brookfield tally was published in the local paper on Wednesday morning, and it is exactly the same as the official tally we're seeing now. Attempts to study Story Three are actively harmful to the study of Story Two, because they hide the fact that we know these things.
And if we know the preliminary results from Brookfield City, we can know all the preliminary results. The Brookfield City clerk gave these to the local paper Tuesday night; it follows that every city and township clerk had their results Tuesday night, and in many cases probably gave them to the local papers. All one has to do is track down all the city and town clerks and ask them for the pre-canvass, Tuesday night numbers. Then publish these for all to see; we can all add them up for ourselves. The results may be off by a couple hundred from the true totals, but they will be off by roughly 7500 from those predicted by either Story One or Story Two.
If the numbers agree with Story One, but Story Two is true, then every clerk whose district was modified is necessarily in on the conspiracy; this is either blindingly obvious from turnout or in moon-landing territory.
Why has nobody done this (or, if they have, why can't I find it)? I can't; I don't know what all the cities are, and probably can't get the right people to talk to me. Certainly it's not possible for every Wisconsinite to do it individually. But this seems to be exactly the sort of thing that reporters are supposed to do all the time.
Instead, everything I see about this is trying to analyze Story Three. ARGH!