shintasama wrote:
Your basic understanding of probability is so wrong and the degree of which you've missed the boat is so high, I'd likely have to teach you from scratch. I have neither the time or the patience to do so. If you want to learn basic statistical analysis the links I posted are a good place to start.
Really? Wow. It's so great that we have a resident super-genius to tell us all how none of us can read a simple poll result properly! Thank you so much super-statistics-guy!
Wait! On second thought, I still think you're spouting BS.
Quote:
The "whole picture" has absolutely nothing to do with under reporting/over reporting/non-poisson distributions/whatever other nonsensical garbage you think I'm talking about. There is no reason to think the distribution is not normal.
You're the one claiming that it is though, right? If it were "normal", then we could simply subtract the two numbers and conclude that the poll results indicate a decrease in popularity for Obama. So now you're contradicting yourself.
Quote:
As stated above the differences in average likelihood comes from the centers of the distributions "current" and "1 week ago" polls being different, so the subsets of "1 week ago" and "current" required to meet "popularity rises" and "popularity falls" are different, ...
You have absolutely no reason to assume this though. It's the same poll, taken by the same polling organization, using the same methodology. It's possible for the distributions to be
randomly different from week to week, but that's as far as we can go here.
Quote:
... with the values in the subsets that meet "popularity rises" being comparatively fewer and further from the center of the distribution making this outcome less likely (but still reasonably possible). I cannot emphasize enough that "more likely" does not mean "this is the case."
Ok. Here's where you go off the rails... again. Why do you assume that (the first part, not the second)? And, more importantly, why assume that for this weeks results, but not a similar factor present in the previous week? Remember, we're calculating a delta between poll results taken over time. I suppose you could say that factors have changed over that week making "popularity rises" less likely and "popularity drops" more likely, but then isn't that exactly what the damn poll tells us right off the bat?
I suspect you are way over thinking this in order to make a result not mean what it clearly means. Which is pretty darn silly in this case.
Quote:
The "whole picture" is that you cannot say "the president's popularity rose" or "the president's popularity fell" with any sort of real confidence in this case.
Of course I can. The president's popularity fell. See! I said it. :)
Kidding aside, I absolutely can say that the odds that the presidents popularity fell over the course of that week is
greater than the odds that it increased. Which again, is the point of the poll.
Quote:
The change between now and one week ago is not statistically significant due to the limited power of the study. This is not to say that his popularity is necessarily exactly the same although this is also not unlikely. This instead means that you cannot draw any conclusions one way or the other without first increasing the statistical power of the study (by polling a lot more people) or observing a larger change.
No poll is perfectly accurate. I think we all get that. What's your point here? Let's just chuck out the practice of taking sampled polls for anything ever? That's kinda silly, isn't it?
Edited, Jun 23rd 2011 1:09pm by gbaji