楼下奔驰野马做的Excel让我觉得很奇怪,不论你用什么输入假设, 都是对的。 所以我就推了一下公式。 结果发现, Dr. Shriva制作的曲线, 是(非常简单的)数学规律, 根本就没什么好奇怪的。 如果不这样, 才成问题。
摘要:
Dr. Shriva给出了
X= Vote by Party for Trump
Y=Vote by Candidates for Trump – Vote by Party for Trump
然后画了个Y相对X 的图, 说曲线斜率是负的, 所以作弊。
实际上在正常情况下, Y是X的减函数, 斜率是负的是正常现象; 如果斜率不是负的, 那才有问题。
Matt的一句关键的推理出现了明显而且严重的数学错误, 把相关性等同于斜率为0. 基础数学不过关。 “ If they are correlated, there should be very small differences between them. and therefore the line should be flat.”
以下是初等数学证明。
下面这个表格是基本假设:
Table 1. Vote by party % vs Vote by Candidate %
|
Republican |
Democrat |
Vote by Party |
gt% |
db% |
“Vote by Candidate” |
gb% |
dt% |
X = Vote by Party for Trump= gt%
Y= Vote by Candidates for Trump – Vote by Party for Trump= dt% - gt%
如果你看视频6:59, Matt 给出了他的基本假设 (Almost Verbatim):"If you look at a precinct and if you look at % people who vote by party and for republican, that gives you a sense of how people vote, which means, should correlate with candidates who also picked Trump"
翻译成初等数学就是: gt% ~ dt%. 那我们就给他一个比例: dt% = a x gt% .
于是我们有 Y = dt% - gt% = (a-1) gt% = ( a -1 ) X
由于 Cross party vote 比例应该低于Straight Vote, 所以 a 小于1. a-1<0. 也就是说Y是X的减函数,。
而且你从图里可以看到, Dr. Shriva给的图的纵坐标截距很接近于0. 也符合上述公式。
Matt在上述假设后还有一句话 (7:20): If they are correlated, there should be very small differences between them. and therefore the line should be flat. 这是对基本数学原理的理解错误。