Latest Poll Numbers Suggest Sanders Could Pull Off More Michigan-style Wins (if we GOTV)

[GreatGazoo]

His biggest gains are in Illinois, where the Vermont senator was once down by as much as 37 points. Now he’s leading Clinton by 2 points (48 to 46 percent) in the latest CBS poll, out Friday.

In Ohio, Clinton still holds a lead. But that advantage has narrowed from about 30 points a week ago to about 6 points (49 to 43 percent) in the latest polls to come out, conducted by Quinnipiac, PPP, and CBS News/YouGov.

And in Missouri, where reliable polling has been scarce, Clinton appears to be leading by 7 points (47 to 40 percent) in a Fort Hays State University poll conducted of the race — but that’s within the poll’s huge 8-point margin of error.

http://www.vox.com/2016/3/14/11219906/bernie-sanders-polls

[saltpoint]

1. An 8-point margin of error

seems well, less than entirely useful.

Maybe increase the sampling size, just for starters. What did they do? Drive up to a bus stop and ask 4 or 5 people standing there who they were leaning toward, call in the tally and publish the results?

[Kalidurga]

2. That would probably give us more accurate results than we have been getting.

[saltpoint]

3. : ) Point taken, Kalidurga. The

polling organizations are having nearly as bad a year as Rick Perry. Well, not quite that bad, but still, kinda lousy.

[reformist2]

4. So, did 30-40% of voters change their minds in one week, or are these polls complete garbage??

I say the latter. These latest numbers are probably more accurate, now that all the pollsters were chastened by Michigan voters. But what it really tells me is that pollsters basically can create whatever result they want.

[GreatGazoo]

5. other articles break it down further

Wired says it is a combination of over use of landlines, over sampling of older Americans, undersampling of 20-somethings (who aren't as likely to answer calls from unknown numbers) and inaccurate weighting of those polled.

When we see a poll and says it was based on "1,023" respondents it means that they got 1023 people to answer (obviously) but what it doesn't say is that they break those answers out by the age and other demographics of those respondents and then weight them to match the actual demographics they expect at the ballot box.

In other words, if you they had polled human beings inside of a grade school and asked what is your favorite TV show. 10 kids say "Thomas the Tank Engine" and 5 teachers say "Downton Abbey" then the raw poll data is saying 67% TTE and 33% Downton Abbey. But once they weight it to say 450 students and 50 teachers you get: 90% TTE and 10% DA. So that formula, the weighting of the raw data, is key to making the final poll numbers as reflective as possible of the general population.

The Wired article:
http://www.wired.com/2016/03/sanders-michigan-win-shows-pollsters-bernie-blindspot/