Good point about the slope not being guaranteed to be negative, because the slope of the split ticket quantity can be greater than 1! I’ve revised the article to instead say the expected “flat” line would be rare, and that a line of SOME sort would always fall out.
A chief complaint of Ayyadurai is that there was a “regular line”, which is why he wanted to push claims of algorithmic interference. Instead it really is just a function of what he’s plotting.
Further more, in actual data, the slope of % split ticket vs. % straight ticket is in fact less than 1, which I’ve now also included.
As for whether or not the interpretation for Trump vs. Biden would be the same — doesn’t super matter. We’ve shown that the line Shiva plots will be negative in all cases where the slope of % split ticket vs. % straight ticket is less than 1. In real data this would be true of most non-party affiliated people, and in fact is. It’s not a good discriminating quantity for identifying fraud, it’s just fluff.