Reply #87: extend that square root

[OnTheOtherHand]

The standard error is sqrt(pq/n), not pq/sqrt(n).

TIA's shortcut is perfectly handy (when used judiciously!) -- to get a margin of error, use 1/sqrt(n). Either way, around 5%, or 10% on the margin.

Given what we can see of house effects (or whatever they are) in races with multiple polls, I would be leery about relying too much on any one survey result regardless of MoE.

Of course if all the close races seem to break the same way, that would still raise eyebrows. I have to admit that my eyebrows aren't up -- could be because I came in by a different analytical route. I do know that I haven't lost my capacity for quantitative alarm, because Sarasota set me off right quick.