I just wanted to put this out there, as I don't think things are as straightforward as some of those alleging Dr. Shiva is wrong are implying.
According to the methodology Dr. Shiva gives, the x-axis represents the percentage of RSP (Republican straight party) votes out of the total of straight party votes. Obviously this ranges between 0 and 100%. So does the percentage of TIC (Trump individual candidate) votes out of the total of individual candidate votes. Note that RSP and TIC are separate and together make up the total number of Trump votes, and we must have 0 <= RSP% + TIC% <= 200%.
The y-axis, we are told, is TIC%-RSP%. So the sum x+y of the coordinates of a data point (x,y) is just TIC%. It cannot exceed 100.
So, for example, if you have RSP at 100%, the y-value can't exceed 0. If RSP is 0%, the y-value can't be less than 0. If RSP is 50%, the y-value has to range between -50% and 50%.
Hence this image that someone arguing against Dr. Shiva made on Twitter, which shows these constraints.
All of this appears true enough. But does this imply anything probabilistic about what real plots will look like?
The key question is, why does Dr. Shiva expect his "normal case"to hold, which is equivalent to the assumption that RSP% and TIC% should be roughly equal, and do the aforegoing facts contradict it? Votes are not just random coin tosses, and RSP and TIC are not causally independent of one another, so Dr. Shiva's assumption doesn't appear unreasonable to me (not to say it's correct), while the argument that as RSP% increases, the y-value decreases, so a plot like the ones he obtained are more likely, that doesn't seem to be sound.
Thoughts?
I just wanted to put this out there, as I don't think things are as straightforward as some of those alleging Dr. Shiva is wrong are implying.
According to the methodology Dr. Shiva gives, the x-axis represents the percentage of RSP (Republican straight party) votes out of the total of straight party votes. Obviously this ranges between 0 and 100%. So does the percentage of TIC (Trump individual candidate) votes out of the total of individual candidate votes. Note that RSP and TIC are separate and together make up the total number of Trump votes, and we must have 0 <= RSP% + TIC% <= 200%.
The y-axis, we are told, is TIC%-RSP%. So the sum x+y of the coordinates of a data point (x,y) is just TIC%. It cannot exceed 100.
So, for example, if you have RSP at 100%, the y-value can't exceed 0. If RSP is 0%, the y-value can't be less than 0. If RSP is 50%, the y-value has to range between -50% and 50%.
Hence this image that someone arguing against Dr. Shiva made on Twitter, which shows these constraints.
All of this appears true enough. But does this imply anything probabilistic about what real plots will look like?
The key question is, why does Dr. Shiva expect his "normal case"to hold, which is equivalent to the assumption that RSP% and TIC% should be roughly equal, and do the aforegoing facts contradict it? Votes are not just random coin tosses, and RSP and TIC are not causally independent of one another, so Dr. Shiva's assumption doesn't appear unreasonable to me, while the argument that as RSP% increases, the y-value decreases, so a plot like the ones he obtained are more likely, that doesn't seem to be sound.
Thoughts?