Reply #18: margin of error is a "confidence interval"

[Teaser]

First off, I hope you didn't take those capitals as being offensive. I'm simply trying to draw attention to the post, because so many people fail to get this concept (mostly because the media that reports it reports it WRONG EVERY GODDAMN TIME).


Anyway, as for MoE, or confidence intervals, they represent your confidence that the true value falls in between the mean value + or - some function of the standard deviation (variability) of the data.

Most polls, if not all these days, are conducted with 95% confidence intervals, meaning that you can say, with 95% assurance that you are correct, the two values (in this case, poll readings for candidate 1 and candidate 2 are correct.

Now in this case, the error bars for the MOE overlap, so we don't have 95% confidence (or greater) that the two values are different. But we may well have 90% confidence that they are different. And we most certainly have 60% confidence they are different (it depends on the size of the sample). If there is a lead of ANY SIZE AT ALL, it is more likely than not that the lead is real. The size of the lead only increases your confidence that it is real.
This is why I say there is *no such thing as a statistical tie*. There just isn't. Either it's exactly tied or it's not.

The real thing one has to take care of here are outliers. There is a 1/25 chance that a given poll (with 95% confidence) will show a reverse ordering of the true results (as implemented in the population). The media also fail to understand the nature of outliers in their data. Aberrant polls are not reported as aberrant, they are reported as "reversals of fortune."

It really is hard when you butter your bread with this kind of stuff, to watch how bad the media is, and how badly they misinform the public.