But in large sets they would be close. DIstricts are geographic areas. The percentage or party line voters is relatively small. But people tend to vote the way their neighbors vote. You would not have a huge variance in probability of people that voter individual vs people that vote party line. The party line is a geographic affinity.
The gist is this: Dr. shiva plotted a difference between Trump non-tickst share and straight R ticket share on the y-axis. This difference will mostly likely grow as Repub vote share increases because the pcts themselves increase.
This is a model, not data. The data will have other idiosyncracies and randomness that this model does not include. But as the Repub vote increases, the trend will be almost inevitable.
I don't care which side is right - none of this is admissible in court. Trumps lawyers would need an expert and that expert would need data that came from discovery - not from some rando off the internet. This whole exercise is completely moot.
This isn't right. Your calculation for column D doesn't seem right.
Take row 5 as an example. 30% of straight ticket ballots went to Reps (and 70% went to Dems). You calculated column D using the theoretical 10% Dem votes to Trump times the 0.7 straight ticket to Dems to get 7%. This is not what Shiva calculated. His non-straight ballot calculation (aka column D) was a percentage of the non-straight-party ballots - column D should not depend at all on column B, since the data Shiva used for each of these columns is separate.
How did Dr. Shiva get his numbers? To me, it looks like he just took the number of Trump votes and subtracted the number of votes for some other R candidate, like John James. Regardless, the number of non-R-ticket Trump voters very much DOES depend on the number of Dem voters. You are much more likely to get a crossover vote when there are more possible votes to cross over. In other words, as the pool of Dems decreases, the number of non-R-ticket Trump votes decreases with it. Hence, the negative slope.
The data provided by the Michigan counties includes the straight party and non-straight party numbers separately. He did not extrapolate or subtract using senate numbers, like you mentioned.
And yes, there is more cross-over possible, however Shiva addressed this in his analysis. Since this is precinct-by-precinct data, all located closely geographically, if there was any cross-over, you'd expect it to be relatively consistent across the different precincts. What he showed is that, apparently there was more cross-over to Biden the more straight party Republicans, and it was increasing linearly. Someone else also plotted the Dems, and showed that the more Dem straight party votes, the more Trump individual ballots appeared. See here: https://thedonald.win/p/11Q8SpHsTj/dr-shiva-algorithm-data-for-mi-c/
That's just the same relationship inverted. In heavy Repub areas, you'd expect Biden to outperform straight ticket D because there will be far fewer straight ticket Ds and far more Repubs to poach votes from.
That still doesn't make sense. In the graph I linked, take a look at the 50% Rep (straight party) plot. Since this is 50/50 - how can we explain that among the individual ticket votes, Trump underperformed by 5%, and Biden overperformed by 5%?
I'd expect, among the precincts that were 50/50, sometimes Trump would overperform, and sometimes he would underperform. But in this case, he always underperforms?
It's all in what he chose to plot on the y-axis. It's not a pct, its a different between two pct. That means, as the Repub vote increases, the margin between Trump's pct and the Repub vote also increases.
It doesn't matter if it is the difference between two pcts.
show me the spread of the residuals. if the spread isn't uniform, the data shows it is inorganic.
noooo it isn't ahhhhhhh. If you compare the ones with the ones without voter fraud, the one with voterfraud has high correlation and can be modeled, while the one without cannot.
You modeling the spread (for example D13) shows that it can be modeled and is not organic.
No it isn't but it uses the same principles.
What is you background by any chance? I am just wondering if you taken a couple courses in statistics.
From the website I shown:
Why do they want the scatter plots uniform? by removing the Y and the Yexpected, all you have is a random scatter plot.
You can translate this to shiva's plots. He shows there is a correlation where there shouldn't be. If you are looking for a relationship, then this is good! This is how scientific findings are found. But we are trying to prove there isn't a relationship, so the exact opposite.
Here's a simple example that illustrates the point:
Imagine a precinct with 0% Republican voters. Assume 10% of these Dem and Indy voters vote for Trump anyway. That gives you a y-axis value on Dr. Shiva's graph of 10%.
Imagine another precinct with 50% Republican voters. Assume 90% vote for Trump and another 10% of the remaining non R voters vote for Trump. That gives you a y-axis value of 5-45= -40%.
Imagine another precinct with 100% Republican voters. Assume 90% vote for Trump. That gives you a y-axis value of -90% on the graph.
See the pattern? Dr. Shiva's graph is highly non-intuitive and I think that was by design.
You don't have the axes correct. There is no flaw in the math.
The flaw might be that the votes that are not straight party tickets are expected to track the strength of the straight party ballot bias.
However, it clearly was not correlated in Wayne county (not the some line) and it also was not tightly correlated if you looked at the democrat side of the coin from Dem Straight Party =40%-60% of the vote
It would be EXTREMELY surprising if Trump outperformed the Straight ticket R vote in even moderately Republican areas. Think about it. Straight R ticket means they voted for Trump. To outpeform that, he has to get a LOT of Dem vote in highly Repub areas. In a majority Repub area, it's virtually impossible to get anything over 0.
I trust the man that with 4 degrees from MIT analyzing real data over some rando online and 11 rows of made up data.
That's an appeal to authority. Do the math yourself if you don't believe me.
AGAIN -you are decoupling the straight R and individual voter Trump voter vote. Why would you imagine they would decouple like that?
Republicans rarely do party line vote anyway.
They are decoupled by definition. Straight ticket R votes and individual non-straight-ticket Trump votes are mutually exclusive.
But in large sets they would be close. DIstricts are geographic areas. The percentage or party line voters is relatively small. But people tend to vote the way their neighbors vote. You would not have a huge variance in probability of people that voter individual vs people that vote party line. The party line is a geographic affinity.
You have that flipped. Party line voters are by far the most common. Non party line voters are less common.
Does this have an explanation to go along with it?
The gist is this: Dr. shiva plotted a difference between Trump non-tickst share and straight R ticket share on the y-axis. This difference will mostly likely grow as Repub vote share increases because the pcts themselves increase.
Shiva's graph doesn't trend down until 20-30% Republican share of vote...
This is a model, not data. The data will have other idiosyncracies and randomness that this model does not include. But as the Repub vote increases, the trend will be almost inevitable.
I don't care which side is right - none of this is admissible in court. Trumps lawyers would need an expert and that expert would need data that came from discovery - not from some rando off the internet. This whole exercise is completely moot.
This isn't right. Your calculation for column D doesn't seem right.
Take row 5 as an example. 30% of straight ticket ballots went to Reps (and 70% went to Dems). You calculated column D using the theoretical 10% Dem votes to Trump times the 0.7 straight ticket to Dems to get 7%. This is not what Shiva calculated. His non-straight ballot calculation (aka column D) was a percentage of the non-straight-party ballots - column D should not depend at all on column B, since the data Shiva used for each of these columns is separate.
How did Dr. Shiva get his numbers? To me, it looks like he just took the number of Trump votes and subtracted the number of votes for some other R candidate, like John James. Regardless, the number of non-R-ticket Trump voters very much DOES depend on the number of Dem voters. You are much more likely to get a crossover vote when there are more possible votes to cross over. In other words, as the pool of Dems decreases, the number of non-R-ticket Trump votes decreases with it. Hence, the negative slope.
The data provided by the Michigan counties includes the straight party and non-straight party numbers separately. He did not extrapolate or subtract using senate numbers, like you mentioned.
And yes, there is more cross-over possible, however Shiva addressed this in his analysis. Since this is precinct-by-precinct data, all located closely geographically, if there was any cross-over, you'd expect it to be relatively consistent across the different precincts. What he showed is that, apparently there was more cross-over to Biden the more straight party Republicans, and it was increasing linearly. Someone else also plotted the Dems, and showed that the more Dem straight party votes, the more Trump individual ballots appeared. See here: https://thedonald.win/p/11Q8SpHsTj/dr-shiva-algorithm-data-for-mi-c/
That's just the same relationship inverted. In heavy Repub areas, you'd expect Biden to outperform straight ticket D because there will be far fewer straight ticket Ds and far more Repubs to poach votes from.
That still doesn't make sense. In the graph I linked, take a look at the 50% Rep (straight party) plot. Since this is 50/50 - how can we explain that among the individual ticket votes, Trump underperformed by 5%, and Biden overperformed by 5%?
I'd expect, among the precincts that were 50/50, sometimes Trump would overperform, and sometimes he would underperform. But in this case, he always underperforms?
Explanation of each column:
A is just a set of example Republican voter percentages from 0 to 100%.
B is the opposite: Dem voter percentages. Just 1 minus A.
H1 is a made up example percentage of Dems that Trump won.
H2 is a made up example percentage of Repubs that Trump won.
C is the straight R ticket pct based on the H2 and A numbers. This is the x-axis on Dr. shiva's graph.
D is the non-R-ticket Trump vote pct based on the H1 and B numbers.
E is the difference between D and C. This is the y-axis on Dr. Shiva's graph.
I don't get? Just having a correlation is not good. the spread has to show no correlation to suggest it being organic.
It's all in what he chose to plot on the y-axis. It's not a pct, its a different between two pct. That means, as the Repub vote increases, the margin between Trump's pct and the Repub vote also increases.
It doesn't matter if it is the difference between two pcts. show me the spread of the residuals. if the spread isn't uniform, the data shows it is inorganic.
That's not what he plotted, though. This is what he plotted.
noooo it isn't ahhhhhhh. If you compare the ones with the ones without voter fraud, the one with voterfraud has high correlation and can be modeled, while the one without cannot.
You modeling the spread (for example D13) shows that it can be modeled and is not organic.
Use this as a reference.
https://online.stat.psu.edu/stat462/node/117/
The expected value is the x-axis.
That's not what he plotted at all.
No it isn't but it uses the same principles. What is you background by any chance? I am just wondering if you taken a couple courses in statistics.
From the website I shown: Why do they want the scatter plots uniform? by removing the Y and the Yexpected, all you have is a random scatter plot. You can translate this to shiva's plots. He shows there is a correlation where there shouldn't be. If you are looking for a relationship, then this is good! This is how scientific findings are found. But we are trying to prove there isn't a relationship, so the exact opposite.
I took one class on statistics in college. I'm an aerospace engineer. Dr. Shiva's plot is not a plot of residuals.
If you want to prove Dr. Shiva wrong, try to make the spread uniform.
What you are showing is proving Dr. Shiva right.
Here's a simple example that illustrates the point:
Imagine a precinct with 0% Republican voters. Assume 10% of these Dem and Indy voters vote for Trump anyway. That gives you a y-axis value on Dr. Shiva's graph of 10%.
Imagine another precinct with 50% Republican voters. Assume 90% vote for Trump and another 10% of the remaining non R voters vote for Trump. That gives you a y-axis value of 5-45= -40%.
Imagine another precinct with 100% Republican voters. Assume 90% vote for Trump. That gives you a y-axis value of -90% on the graph.
See the pattern? Dr. Shiva's graph is highly non-intuitive and I think that was by design.
You don't have the axes correct. There is no flaw in the math.
The flaw might be that the votes that are not straight party tickets are expected to track the strength of the straight party ballot bias.
However, it clearly was not correlated in Wayne county (not the some line) and it also was not tightly correlated if you looked at the democrat side of the coin from Dem Straight Party =40%-60% of the vote
It would be EXTREMELY surprising if Trump outperformed the Straight ticket R vote in even moderately Republican areas. Think about it. Straight R ticket means they voted for Trump. To outpeform that, he has to get a LOT of Dem vote in highly Repub areas. In a majority Repub area, it's virtually impossible to get anything over 0.
This isn't a time series graph. This is percentage graph.
A downward trend is what you expect no matter who wins.
I posted this earlier, but I corrected the numbers here to more closely match what Dr. Shiva was plotting on his y-axis.
Dr. Shiva is a charlatan.