Problem with your analysis is that in Shivs’s analysis the graph is constant for a while and then starts dropping, so your analysis wouldn’t be accurate either.
This is just a model to illustrate why a negative slope is expected.
Trump won a certain pct of Dems and a certain pct of Repubs. When the overall Dem pct is high, he's going to outperform Repubs. When the overall Repub vote is high, he's going to underperform Repubs. This is just a mathematical certainty.
A negative slope is expected with random values, the point is the correlation between Republican straight party vote and Trump single vote is not random, you would expect it to be a fairly constant ratio, aka a straight line.
My argument here is that you wouldn't expect a straight line. It's a bit counterintuitive until you start playing with the numbers yourself. The problem is that the y-axis represents different in pct not a pct. So as the pct on the x-axis increases, the difference between the pcts will increase.
That's not true, the X axis is (# of people that voted straight party republican ticket option / total number of republicans and dems that voted straight party ticket option). The Y axis is (# of people that voted Trump but didnt vote straight party ticket option/ total # of people that didnt vote straight party ticket option) - X axis data. The Y and X axis are independent data sets (excluding the subtraction of the Xaxis data) , one does not include the other, one doesnt have to go down if the other goes up and vice versa.
No. The y-axis is (# of people that voted Trump but didnt vote straight party ticket option/ total # of people that didnt vote straight party ticket option) - (# of people that voted straight repub ticket/total number of Dems that voted straight party ticket). It's not a percentage. It's a difference of percentages with different denominators.
I think you are misunderstanding the data. From what I understand, Shiva looks at Oakland county which is broken up into precincts. The X-axis is Straight Republican Party (SRP) vote percentage for that county. So that is the percentage of all votes voting straight republican. Y-axis is percentage of individual Trump votes in the county minus the percentage of SRP votes. So what his chart is saying is that in the places where there was low SRP (aka blue city type places) he overperforms SRP votes. On the other hand, in counties with a higher SRP (so very Red counties) he underperforms the SRP vote. That would only make sense if you thought Dems in cities secretly favor Trump and Republicans in Red counties secretly dislike Trump, which I highly doubt is the case.
No. That is exactly what the mathematical relationship between the numbers would predict. As an area gets more and more Republican, there are fewer and fewer Dem votes to "flip" your way. So you are left with just the Repub votes you got, which will always be less than the straight ticket votes.
While true, the probability of anyone voting straight Republican and NOT voting Trump is extremely low. Can't see that happening at all, especially in super red counties. Definitely not to the degree of -30% underperformance vs SRP. I agree that the slope would be negative, but the value of the slope is just way too big to pass the smell test.
But that's what I'm getting at. If you have a 100% Republican area, the best Trump can do is 0% relative to the other Republicans. If ANY Republican doesn't vote for him, he will get a negative margin of the Repub vote. A negative margin is expected in a heavily Republican area.
Yes, but look at the underperformance, it is too big. If the trend continued, at 100% SRP Trump would AT LEAST lose 30% of SRP voters, that is something that I can't find believable. He has 90%+ favorability with republicans, at most he would underperform by 10%.
The graph should asymptotically approach -10% at worst.
You are deliberately letting less and less people vote on Trump (compared to Republican). This is not what normal data looks like. What they showed was a random cluster in non-manipulated districts or heavy Dem districts.
But that's the thing, the x-axis is not total votes, it's Repub vote pct of total vote, so Trump WILL underperform Repubs more and more as they take more and more of the total vote.
You literally just plugged numbers you chose into an Excel spreadsheet and created a graph from it that you knew would create a downward slope. Shiva and his associates took raw data to graph it out. And, being a percentage above or below a straight Republican ticket, you would expect it to be around that straight Republican mark. Will you potentially get a greater number of votes not for Trump in more Republican counties? Sure. But the percentage should be around the same.
Your plot has 0-100% on the X axis, but only 0-10% on the Y axis. Also if you assume an equal % of Republican share and Trump share % then it would be a straight line at y = 0. Your data itself has a downward trend in that you start with 10% more Trump share on x = 0, and end with 5% less Trump share linearly at X = 1
Remember what Dr. Shiva's y-axis is: the Trump vote pct minus the Repub straight ticket vote pct.
If the Dem vote pct is very high, Trump will still get a certain pct of those Dem votes and, thus, outperform Republicans. If the Repub vote is high, Trump will get a certain pct of those Repub votes, but not enough Dem votes remain for him to outperform the Repubs.
In the analysis Trumps share of total votes doesn't include the votes from republican only ballots, its just the % of people that didnt vote straight party but did vote Trump. the % that voted dem wont effect either column.
Yo, you have just put some numbers in a table and then said "SEE?!". Not a great presentation to make your point.
Like to someone that doesn't know what your initial question was, you have made values that decrease then plotted that they decrease.
What values are derived from what values, why, what does plotting them show. What would we see if Shiva was correct? What would we see if Shiva was incorrect?
I could open Excel and make a table with the line y=x, or y=x^2 parabola, or y = -x. So what?
Also a screenshot of a table of float values lolwut sans any explanation, which columns are inputs, which are calculated from inputs, etc.
Can you just do a better job of making your point, I'd like to understand it.
Come on, man!
EDIT: specifically
What is the source of the values in column C? just made up random numbers? What if you randomly made up different numbers lol. You have apparently plotted "Look what happens when I type random numbers that I chose, that change in an x-dependent manner, and then take the delta between that and evenly spaced numbers. much wow"
It's easier to explain with an example. Remember what Dr. Shiva's y-axis represents: the difference between Trump's vote pct and the straight Repub ticket pct. Now let's plug in a few numbers to see what happens.
Let's say a given precinct was 100% Dem. In other words, there are no straight ticket Repub votes. But let's also say Trump got 10% of the Dem vote. This means the graph coordinates for this precinct are (0%,10%).
Let's say a different precinct is 100% Repub. In other words, every voter is straight ticket Repub. The graph coordinates for this precinct are (100%, 0%).
Let's say a different precinct is 50% Repub. In other words, 50% is straight ticket Repub. But let's also give Trump 10% of the Dem vote. The graph coordinates for this precinct are (50%, 5%).
See the pattern yet? And that's assuming every Republican votes for Trump. In reality, he likely got 90-95% of the Repub vote, making this slope even more negative. The reason is that as the Repub vote share increases, there are fewer and fewer non Repub votes for Trump to outperform the Repub ticket with.
There is a lot of "Let's say" happening, where if you "Let's say something else" then you get the opposite result.
This is not evidence that your graph is expected, this is evidence that your graph is expected, given your assumptions--which might be totally incorrect.
I'm not saying my graph matches the data, I'm saying the mathematical relationship in observed in the data is expected. It would be weird if there wasn't a negative slope. That's what I'm saying. In fact, the fact that Wayne County DOESN'T have a negative slope tells me there is likely fraud there.
I'm saying the mathematical relationship in observed in the data is expected
The correct way to phrase this is
I'm saying the mathematical relationship in observed in the data is expected [GIVEN THE ASSUMPTIONS THAT I AM MAKING ABOUT THE REAL WORLD, WHICH MAY IN FACT BE THE INVERSE OF REALITY]
The mathematical relationship isn't an assumption. The example I use to illustrate that relationship is an assumption, sure. But the mathematical relationship is a consequence of the weird y-axis choice. It's baked into the way he's chosen to present the data.
The mathematical relationship isn't an assumption.
Right, and I'm not calling that relationship into question. If you assume your assumptions are right then you get that line. If you assume your assumptions are wrong, then you get different lines.
There is not a lot of information content in your analysis which is fine. Make some assumptions about the world, see what you'd observe if so. Nothing wrong with that in general.
The example I use to illustrate that relationship is an assumption, sure.
Right. As were the 2 additional examples I showed you where the conclusion is fully reversed or eliminated.
I proved this to you by showing you how you'd end up with an positive slope, or zero slope. Here again: https://i.imgur.com/gq1KIPz.png
There is nothing about the choice of axes that guarantees a negative slope.
What were your assumptions in tbe positive slope cases? And the zero slope case looks like you had to force the Trump Repub share to 100% which is possible but not the most likely outcome.
I was curious about what claim your data was making re: "how well Trump does among republicans, as a function of the repub % of total"
The answer is that you end up with a line that shows that as Repubs do better in an area, Trump does worse among those repubs. I don't see why that assumption is warranted.
If you instead assume that areas that are densely republican, are loyal to Trump (he does better in these areas, than right-leaning moderate areas, among republicans) which makes more sense to me, then you end up with a positive slope
If you make no assumptions, and that the % Trump vote is directly proportional to the % Repub vote, then obviously, you end up with a flat line.
So my problem with your post is that you have made assumptions that might be totally unwarranted. If you had made different assumptions you'd get the opposite result. Not very conclusive IMO.
Here is that visualized, plugging in 100,000 votes arbitrarily for visualization of the scanarios.
No. Trump doesn't do worse among those Republicans. It's just a quirk of the wonky y-axis choice. Remember that the y-axis is NOT a percentage. It's a difference between two percentages with different denominators. It's no longer a percentage once you subtract them like that.
And again, I'm not suggesting the slope couldn't be positive or even zero. It's just that a negative slope is the most likely result. Not a zero slope. In fact, a zero slope is the least likely result.
This is the same reason the media tries to force climate change crap down our throats. It's an appeal to authority. I have advanced engineering degrees, too. No one is immune from making mathematical errors.
So do I. And I've worked with PhDs in the maths department at top universities.
They dont make bold claims lightly.
Im not saying he is wrong or you are wrong. There is a lack of communication between you two and Im sure a simple meeting of the minds would instantly clear it up.
Problem with your analysis is that in Shivs’s analysis the graph is constant for a while and then starts dropping, so your analysis wouldn’t be accurate either.
This is just a model to illustrate why a negative slope is expected.
Trump won a certain pct of Dems and a certain pct of Repubs. When the overall Dem pct is high, he's going to outperform Repubs. When the overall Repub vote is high, he's going to underperform Repubs. This is just a mathematical certainty.
A negative slope is expected with random values, the point is the correlation between Republican straight party vote and Trump single vote is not random, you would expect it to be a fairly constant ratio, aka a straight line.
My argument here is that you wouldn't expect a straight line. It's a bit counterintuitive until you start playing with the numbers yourself. The problem is that the y-axis represents different in pct not a pct. So as the pct on the x-axis increases, the difference between the pcts will increase.
The real world correlation says that as the X axis % increases, so will the Trump vote on the Y axis, thus negating it.
But the y axis isn't Trump's vote pct. Its the difference between his vote pct and the Repub vote pct. That's the big problem with the analysis.
That's not true, the X axis is (# of people that voted straight party republican ticket option / total number of republicans and dems that voted straight party ticket option). The Y axis is (# of people that voted Trump but didnt vote straight party ticket option/ total # of people that didnt vote straight party ticket option) - X axis data. The Y and X axis are independent data sets (excluding the subtraction of the Xaxis data) , one does not include the other, one doesnt have to go down if the other goes up and vice versa.
No. The y-axis is (# of people that voted Trump but didnt vote straight party ticket option/ total # of people that didnt vote straight party ticket option) - (# of people that voted straight repub ticket/total number of Dems that voted straight party ticket). It's not a percentage. It's a difference of percentages with different denominators.
I think you are misunderstanding the data. From what I understand, Shiva looks at Oakland county which is broken up into precincts. The X-axis is Straight Republican Party (SRP) vote percentage for that county. So that is the percentage of all votes voting straight republican. Y-axis is percentage of individual Trump votes in the county minus the percentage of SRP votes. So what his chart is saying is that in the places where there was low SRP (aka blue city type places) he overperforms SRP votes. On the other hand, in counties with a higher SRP (so very Red counties) he underperforms the SRP vote. That would only make sense if you thought Dems in cities secretly favor Trump and Republicans in Red counties secretly dislike Trump, which I highly doubt is the case.
No. That is exactly what the mathematical relationship between the numbers would predict. As an area gets more and more Republican, there are fewer and fewer Dem votes to "flip" your way. So you are left with just the Repub votes you got, which will always be less than the straight ticket votes.
While true, the probability of anyone voting straight Republican and NOT voting Trump is extremely low. Can't see that happening at all, especially in super red counties. Definitely not to the degree of -30% underperformance vs SRP. I agree that the slope would be negative, but the value of the slope is just way too big to pass the smell test.
But that's what I'm getting at. If you have a 100% Republican area, the best Trump can do is 0% relative to the other Republicans. If ANY Republican doesn't vote for him, he will get a negative margin of the Repub vote. A negative margin is expected in a heavily Republican area.
Yes, but look at the underperformance, it is too big. If the trend continued, at 100% SRP Trump would AT LEAST lose 30% of SRP voters, that is something that I can't find believable. He has 90%+ favorability with republicans, at most he would underperform by 10%.
The graph should asymptotically approach -10% at worst.
The absolute numbers are just made up. That's not the important part. The important thing is that the negative slope is baked into the y-axis choice.
Yeah. That's the big problem with his analysis. You can't just subtract percentages like that and expect to geta anything meaningful from them.
You are deliberately letting less and less people vote on Trump (compared to Republican). This is not what normal data looks like. What they showed was a random cluster in non-manipulated districts or heavy Dem districts.
But that's the thing, the x-axis is not total votes, it's Repub vote pct of total vote, so Trump WILL underperform Repubs more and more as they take more and more of the total vote.
Yeah I did the calculation with a stable 90% repubs voting Trump and 20% Dems and you are right.
Does this happen when you plot for biden and the dem share of vote?
It should. It's just the mathematical relationship between party share and candidate share of vote.
You literally just plugged numbers you chose into an Excel spreadsheet and created a graph from it that you knew would create a downward slope. Shiva and his associates took raw data to graph it out. And, being a percentage above or below a straight Republican ticket, you would expect it to be around that straight Republican mark. Will you potentially get a greater number of votes not for Trump in more Republican counties? Sure. But the percentage should be around the same.
Your plot has 0-100% on the X axis, but only 0-10% on the Y axis. Also if you assume an equal % of Republican share and Trump share % then it would be a straight line at y = 0. Your data itself has a downward trend in that you start with 10% more Trump share on x = 0, and end with 5% less Trump share linearly at X = 1
Remember what Dr. Shiva's y-axis is: the Trump vote pct minus the Repub straight ticket vote pct.
If the Dem vote pct is very high, Trump will still get a certain pct of those Dem votes and, thus, outperform Republicans. If the Repub vote is high, Trump will get a certain pct of those Repub votes, but not enough Dem votes remain for him to outperform the Repubs.
In the analysis Trumps share of total votes doesn't include the votes from republican only ballots, its just the % of people that didnt vote straight party but did vote Trump. the % that voted dem wont effect either column.
The negative slope is still present if you only count Trump's share of the Dem vote. Try it yourself!
Yo, you have just put some numbers in a table and then said "SEE?!". Not a great presentation to make your point.
Like to someone that doesn't know what your initial question was, you have made values that decrease then plotted that they decrease.
What values are derived from what values, why, what does plotting them show. What would we see if Shiva was correct? What would we see if Shiva was incorrect?
I could open Excel and make a table with the line y=x, or y=x^2 parabola, or y = -x. So what?
Also a screenshot of a table of float values lolwut sans any explanation, which columns are inputs, which are calculated from inputs, etc.
Can you just do a better job of making your point, I'd like to understand it.
Come on, man!
EDIT: specifically
It's easier to explain with an example. Remember what Dr. Shiva's y-axis represents: the difference between Trump's vote pct and the straight Repub ticket pct. Now let's plug in a few numbers to see what happens.
Let's say a given precinct was 100% Dem. In other words, there are no straight ticket Repub votes. But let's also say Trump got 10% of the Dem vote. This means the graph coordinates for this precinct are (0%,10%).
Let's say a different precinct is 100% Repub. In other words, every voter is straight ticket Repub. The graph coordinates for this precinct are (100%, 0%).
Let's say a different precinct is 50% Repub. In other words, 50% is straight ticket Repub. But let's also give Trump 10% of the Dem vote. The graph coordinates for this precinct are (50%, 5%).
See the pattern yet? And that's assuming every Republican votes for Trump. In reality, he likely got 90-95% of the Repub vote, making this slope even more negative. The reason is that as the Repub vote share increases, there are fewer and fewer non Repub votes for Trump to outperform the Repub ticket with.
There is a lot of "Let's say" happening, where if you "Let's say something else" then you get the opposite result.
This is not evidence that your graph is expected, this is evidence that your graph is expected, given your assumptions--which might be totally incorrect.
I'm not saying my graph matches the data, I'm saying the mathematical relationship in observed in the data is expected. It would be weird if there wasn't a negative slope. That's what I'm saying. In fact, the fact that Wayne County DOESN'T have a negative slope tells me there is likely fraud there.
The correct way to phrase this is
The mathematical relationship isn't an assumption. The example I use to illustrate that relationship is an assumption, sure. But the mathematical relationship is a consequence of the weird y-axis choice. It's baked into the way he's chosen to present the data.
Right, and I'm not calling that relationship into question. If you assume your assumptions are right then you get that line. If you assume your assumptions are wrong, then you get different lines.
There is not a lot of information content in your analysis which is fine. Make some assumptions about the world, see what you'd observe if so. Nothing wrong with that in general.
Right. As were the 2 additional examples I showed you where the conclusion is fully reversed or eliminated.
I proved this to you by showing you how you'd end up with an positive slope, or zero slope. Here again: https://i.imgur.com/gq1KIPz.png
There is nothing about the choice of axes that guarantees a negative slope.
What were your assumptions in tbe positive slope cases? And the zero slope case looks like you had to force the Trump Repub share to 100% which is possible but not the most likely outcome.
Column C is Trump's share of Dem vote multiplied by Dem share of total vote plus Trump's share of Repub vote multiplied by Repub share of total vote.
I was curious about what claim your data was making re: "how well Trump does among republicans, as a function of the repub % of total"
The answer is that you end up with a line that shows that as Repubs do better in an area, Trump does worse among those repubs. I don't see why that assumption is warranted.
If you instead assume that areas that are densely republican, are loyal to Trump (he does better in these areas, than right-leaning moderate areas, among republicans) which makes more sense to me, then you end up with a positive slope
If you make no assumptions, and that the % Trump vote is directly proportional to the % Repub vote, then obviously, you end up with a flat line.
So my problem with your post is that you have made assumptions that might be totally unwarranted. If you had made different assumptions you'd get the opposite result. Not very conclusive IMO.
Here is that visualized, plugging in
100,000votes arbitrarily for visualization of the scanarios.LINK TO NEW PLOTS
No. Trump doesn't do worse among those Republicans. It's just a quirk of the wonky y-axis choice. Remember that the y-axis is NOT a percentage. It's a difference between two percentages with different denominators. It's no longer a percentage once you subtract them like that.
Anyway, as I demonstrated to you, if you make different assumptions than you made, you get the opposite outcome.
You can play with your excel sheet to discover that I am correct, if you don't trust mine.
And again, I'm not suggesting the slope couldn't be positive or even zero. It's just that a negative slope is the most likely result. Not a zero slope. In fact, a zero slope is the least likely result.
Oh, then we agree. That is grand.
I hate to piss on your parade - but a PhD from MIT has got us all beat in the math and statistics department. Easily.
He wouldnt publicly stake his reputation and career on analyzing the data improperly.
Most people at Harvard and MIT are (wealthy) fucking stupid OrangeManBads.
You need to be a person that does not rely on credentials, and instead focus on the strength of reasoning and evidence.
This is the same reason the media tries to force climate change crap down our throats. It's an appeal to authority. I have advanced engineering degrees, too. No one is immune from making mathematical errors.
Maybe you should contact Dr. Shiva with your findings and let him check his work.
Something tells me this was intentionally misleading. Lot of grifters out there.
So do I. And I've worked with PhDs in the maths department at top universities.
They dont make bold claims lightly.
Im not saying he is wrong or you are wrong. There is a lack of communication between you two and Im sure a simple meeting of the minds would instantly clear it up.
Plop these numbers into excel and play with them yourself. A negative slope is the expected result.
But why does Shiva's negative slope begin at 20%? Why is the negative slope not there for democrat districts?
Came to say this as well. This is not what Dr. Shiva’s graph looked like.
This is a model so it won't reflect reality 100%. It just illustrates why a negative slope is expected on his graph.
It must be something with the independent vote, or the assumptions just breaking down at really low Republican percentages.