[newfreep]

From Sundance’s Conservative TreeHouse

1299 respondents: Trump 40%- 11/19
981 respondents: Trump 38.5%- 11/20
636 respondents: Trump 42.4%- 11/21
464 respondents: Trump 42.9%- 11/22
441 respondents: Trump 42.6%- 11/23
381 respondents: Trump 38%- 11/24
362 respondents: Trump 37%- 11/25
358 respondents: Trump 31%- 11/26
346 respondents: Trump 31%- 11/27

Notice how Trump’s numbers significantly decreased once they began to lower the sample size?

[grey_whiskers]

That says to me something about the reproducibility of the study: for surveys of this type, the relative size of the error bars scale as 1/SQRT(N). Which means the error of the 346 respondents is 5.37%, that of 358 respondents is 5.28%, that of 362 respondents is 5.25%, but that of 1299 respondents is 2.77%.

The difference in Trump's results between the largest and smallest samples, is basically RIGHT AT the margin of error.

Either they are trying to find a sample to minimize Trump's numbers, and having to go to sample sizes which invalidate the study to do it, OR, Trump lost 9% in 8 days following the Paris attacks and the consequent focus on border security and immigration...

I call bullshit.

Nice try, though.