in South Carolina and Super Spectacular Tuesday.

It's relevant even today in MI. How did they get it so wrong?

I dunno how they got it so wrong but they need to check their polling.

for predicting the results so incorrectly.

enough said.

More than enough said!

I'm surprised they weren't off even more than they were. I'd trust polls conducted by Alphonse & Gaston or Abbott & Costello more than these two.

don't trust polls.

or do these ridiculous polls and articles always come with caps lock on? It seems that BS news has to be yelled at the rest of us in order to hear it, or something. Whatever it is it's offensive.

I always check the source when all caps headlines show up. I hope you didn't take offense by my questioning. It wasn't directed at you in any way.

Watch for it. It won’t stop. Trumputin at work.

for helping us get our sh#t together. We've come a long, long way in one short week

Details are irritating and confusing, and usually left out.

So that +/- 4.5 pt difference is the calculated error--no claim as to the real error--and is an estimate. Could be higher, but odds are strongly that it's 4.5 or less--but those are the odds, not an absolute truth.

The calculated error is also for the poll as a whole. The error for subgroups tends to be higher since the number of respondents in each subgroup is less. That detail's often not reported.

Poll structure matters. Was it a forced choice poll, where you couldn't say, "IDK"?

Did they ask, "If it's just Biden and Sanders, who would you vote for?" Because a lot of polls were things like 25-28-16-15-... where the top two have a clear difference of 3%! But in the actual election, some of those 16% and 15% preference candidates vanished. The people who preferred a not-Sanders candidate didn't just apportion themselves in accordance with the preferences of those who only were choosing between Sanders and Biden.

Then there's the structure of the electorate, the rock upon which more than one poll has crashed. If the poll assumes "likely voters" are whatever they get, and they poll 50% "likely (D) voters" under 25 and the other 50% from 25 to 30 years old, they really have no clue what the over 30-years-old set would say. Now, *every* poll includes some sort of model to weight their responses and fill in gaps and level hills in the response rate. That's an assumption that isn't included in calculating the error rate.

I can't find the information built into that Hill/HarrisX poll, which means I tend to regard it as less useful than reading tea-leaves. It might mean something, might not, can't tell and it's a waste of time--except that if you're having your tea leaves read, at least you got a cup of tea out of it.

and became useless.