(08-18-2016 11:18 AM)Max Power Wrote: If your candidate can't produce a single poll showing him not losing, he's losing.
Statistics is your friend. I get what you are saying but if one is winning within the margin of error the poll isn't really saying that person is winning. They could be winning, or they could be tied, or they could be losing.
Statistics is your friend too :) Your mistake is that you're looking at the poll in a vacuum. If all the other evidence supports the real result being in the MOE extended in one direction, there isn't an equal likelihood that the truth is in the other.
Example:
If 10 polls show a generic candidate +7, and one poll finds him only +2, it's far more likely that the part of the 95% confidence interval the truth lies is in the part that's higher than 2. Because otherwise, all of those aforementioned polls showing her +7 were WAY off, outside of all of their MsOE. So it's a near certainty that the generic candidate is winning, because if he wasn't, at least ONE of the 11 polls would show him ahead. You can look at the +2 and reasonably conclude it understates his support.
Now, if we take that +2 and say there's 5 polls showing him +4 and 5 showing it tied, then yeah, maybe we can extend the margin of error equally both ways.
(This post was last modified: 08-18-2016 02:26 PM by Max Power.)