The Shiva stuff has some issues, although still worth looking at the details.

The issue here is that the y axis is dependent on the x axis. It is IMPOSSIBLE for precincts to lay under the trump minus republican line (although somehow some do in the charts presented). You should draw this in your mind, or paint over the screengrab.

So for instance, at the 5% vote straight republican point in the x-axis, a precinct cannot lay at any point under -5% in the y axis, while they can go all the way up to 95%. Vice versa at the opposite extreme.

What this means is that there has to be a small downward slope of some sort. It is impossible for the values to be 'flat' on the y axis like Dr Shiva suggests we should see, at least at the extremes of the x-axis.

Also a problem is regression to the mean: More extreme results in the x axis are more likely to correspond with less extreme results in the trump variable.

However, I'd expect more of a S curve here, not the linear trend we see, so it does look a bit odd. We seem to get the left part of an S curve, but not the right part of it. So that part absolutely could be suspicious. If so, the place to look is at the strongest straight R precincts.

The Shiva stuff has some issues, although still worth looking at the details.

The issue here is that the y axis is dependent on the x axis. It is IMPOSSIBLE for precincts to lay under the trump minus republican line (although somehow some do in the charts presented). You should draw this in your mind, or paint over the screengrab.

So for instance, at the 5% vote straight republican point in the x-axis, a precinct cannot lay at any point under -5% in the y axis, while they can go all the way up to 95%. Vice versa at the opposite extreme.

What this means is that there has to be a small downward slope of some sort. It is impossible for the values to be 'flat' on the y axis like Dr Shiva suggests we should see, at least at the extremes of the y-axis.

Also a problem is regression to the mean: More extreme results in the x axis are more likely to correspond with less extreme results in the trump variable.

However, I'd expect more of a S curve here, not the linear trend we see, so it does look a bit odd. We seem to get the left part of an S curve, but not the right part of it. So that part absolutely could be suspicious. If so, the place to look is at the strongest straight R precincts.

The Shiva stuff has some issues, although still worth looking at the details.

The issue here is that the y axis is dependent on the x axis. It is IMPOSSIBLE for precincts to lay under the trump minus republican line (although somehow some do in the charts presented). You should draw this in your mind, or paint over the screengrab.

So for instance, at the 5% vote straight republican, a precinct cannot lay at any point under -5% in the y axis, while they can go all the way up to 95%. Vice versa at the opposite extreme.

What this means is that there has to be a small downward slope of some sort. It is impossible for the values to be 'flat' on the y axis like Dr Shiva suggests we should see, at least at the extremes of the y-axis.

Also a problem is regression to the mean: More extreme results in the x axis are more likely to correspond with less extreme results in the trump variable.

However, I'd expect more of a S curve here, not the linear trend we see, so it does look a bit odd. We seem to get the left part of an S curve, but not the right part of it. So that part absolutely could be suspicious. If so, the place to look is at the strongest straight R precincts.

The Shiva stuff has some issues, although maybe not entirely, still worth looking at the details.

The issue here is that the y axis is dependent on the x axis. It is IMPOSSIBLE for precincts to lay under the trump minus republican line (although somehow some do in the charts presented). You should draw this in your mind, or paint over the screengrab.

So for instance, at the 5% vote straight republican, a precinct cannot lay at any point under -5% in the y axis, while they can go all the way up to 95%. Vice versa at the opposite extreme.

What this means is that there has to be a small downward slope of some sort. It is impossible for the values to be 'flat' on the y axis like Dr Shiva suggests we should see, at least at the extremes of the y-axis.

Also a problem is regression to the mean: More extreme results in the x axis are more likely to correspond with less extreme results in the trump variable.

However, I'd expect more of a S curve here, not the linear trend we see, so it does look a bit odd. We seem to get the left part of an S curve, but not the right part of it. So that part absolutely could be suspicious. If so, the place to look is at the strongest straight R precincts.

The Shiva stuff has some issues, although maybe not entirely, still worth looking at the details.

The issue here is that the y axis is dependent on the x axis. It is IMPOSSIBLE for precincts to lay under the trump minus republican line (although somehow some do in the charts presented). You should draw this in your mind, or paint over the screengrab.

So for instance, at the 5% vote straight republican, a precinct cannot lay at any point under -5% in the y axis, while they can go all the way up to 95%. Vice versa at the opposite extreme.

What this means is that there has to be a small downward slope of some sort. It is impossible for the values to be 'flat' on the y axis like Dr Shiva suggests we should see, at least at the extremes of the y-axis.

Also a problem is regression to the mean: More extreme results in the x axis are more likely to correspond with less extreme results in the trump variable.

However, I'd expect more of a S curve here, not the linear trend we see, so it does look a bit odd. We seem to get the left part of an S curve, but not the right part of it. So that part absolutely could be suspicious. If so, the place to look is at the hardest straight R precincts.

The Shiva stuff is mostly B.S. unfortunately (although maybe not entirely, still worth looking at the details)

The issue here is that the y axis is dependent on the x axis. It is IMPOSSIBLE for precincts to lay under the trump minus republican line (although somehow some do in the charts presented). You should draw this in your mind, or paint over the screengrab.

So for instance, at the 5% vote straight republican, a precinct cannot lay at any point under -5% in the y axis, while they can go all the way up to 95%. Vice versa at the opposite extreme.

What this means is that there should be a small downward slope of some sort. It is impossible for the values to be 'flat' on the y axis like Dr Shiva suggests we should see, at least at the extremes of the y-axis.

Also a problem is regression to the mean: More extreme results in the x axis more likely to correspond with less extreme results in the trump variable.

However, I'd expect more of a S curve here, not the linear trend we see, so it does look a bit odd. We seem to get the left part of an S curve, but not the right part of it. So that part could be suspicious.

The Shiva stuff is mostly B.S. unfortunately (although maybe not entirely, still worth looking at the details)

The issue here is that the y axis is dependent on the x axis. It is IMPOSSIBLE for precincts to lay under the trump minus republican line (although somehow some do in the charts presented). You should draw this in your mind, or paint over the screengrab.

So for instance, at the 5% vote straight republican, a precinct cannot lay at any point under -5% in the y axis, while they can go all the way up to 95%. Vice versa at the opposite extreme.

What this means is that there should be a downward slope of some sort. It is impossible for the values to be 'flat' on the y axis like Dr Shiva suggests we should see.

Also a problem is regression to the mean: More extreme results in the x axis more likely to correspond with less extreme results in the trump variable.

However, I'd expect more of a S curve here, not the linear trend we see, so it does look a bit odd. We seem to get the left part of an S curve, but not the right part of it. So that part could be suspicious.

The Shiva stuff is mostly B.S. unfortunately (although maybe not entirely, still worth looking at the details)

The issue here is that the y axis is dependent on the x axis. It is IMPOSSIBLE for precincts to lay under the trump minus republican line (although somehow some do in the charts presented).

So for instance, at the 5% vote straight republican, a precinct cannot lay at any point under -5% in the y axis, while they can go all the way up to 95%. Vice versa at the opposite extreme.

What this means is that there should be a downward slope of some sort. It is impossible for the values to be 'flat' on the y axis like Dr Shiva suggests we should see.

Also a problem is regression to the mean: More extreme results in the x axis more likely to correspond with less extreme results in the trump variable.

However, I'd expect more of a S curve here, not the linear trend we see, so it does look a bit odd. We seem to get the left part of an S curve, but not the right part of it. So that part could be suspicious.

The Shiva stuff is mostly B.S. unfortunately (although maybe not entirely, still worth looking at the details)

The issue here is that the y axis is dependent on the x axis. It is IMPOSSIBLE for precincts to lay under the trump minus republican line (although somehow some do in the charts presented).

So for instance, at the 5% vote straight republican, a precinct cannot lay at any point under -5% in the y axis, while they can go all the way up to 95%. Vice versa at the opposite extreme.

What this means is that there should be a downward slope of some sort. It is impossible for the values to be 'flat' on the y axis like Dr Shiva suggests we should see.

Also a problem is regression to the mean: More extreme results in the x axis more likely to correspond with less extreme results in the trump variable.

However, I'd expect more of a S curve here, not the linear trend we see, so it does look a bit odd.

This is definitely kind of weird, but I'd like to see examples of this not being the case elsewhere (like for instance red states we don't expect any fraud) before calling this the smoking gun.

Also, I'd guess that race interacts very significantly with this (black/hispanic voters switching to trump are not republican)