Which polls will be off by how much 3/15?

Michigan’s epic polling fail raises the specter that polls this year can be off and off big.

The following are the current Tuesday’s RCP rolling poll averages (subject to change with more data).

NC Clinton +20.2
FL Clinton +31.5
IL Clinton +30.5
MO Only very old data.
OH Clinton +20

In each case the peaks for 538’s “polling plus” method give Clinton bigger leads.

How will polling do this time?

Personally I am guessing that Florida, being a closed primary, will be pretty spot on. But the others will be smaller than poll prediction Clinton wins, not on the Michigan scale however.

I’m guessing we’ll see a lot of pollsters rejigger their models to account for a higher turnout of young voters, so I’d expect to see Clinton’s margins of victory narrowing, if only a little.

There’s a preliminary autopsy at 538 that tryies to explain what may have happened. The main point is that multiple things went wrong and any on of the following factors would not have been enough on its own to explain what happened.

[li]Pollsters underestimated youth turnout. [/li][li]Pollsters underestimated Sanders’s dominance among young voters.[/li][li]Pollsters underestimated the number of independent voters who would participate in the primary.[/li][li]Pollsters underestimated Sanders’s support among black voters.[/li][li]Pollsters missed a late break to Sanders by not polling after Sunday.[/li][li]Some Clinton supporters chose to vote in the Republican primary.[/li][li]Pollsters had little recent history to work with.[/li][li]This is an outlier, a perfectly rotten combination of bad luck and bad timing.[/li][/ul]

Many of those are correctable by the pollsters and at least one is unique to Michigan. I suspect that most polls we see between today and 3/15 will have made some adjustments and some will have over-corrected. All in all I think the polling averages will shift a noticeable amount in the next week, but the polling averages on 3/15 will on average give us a pretty good idea of what to expect.

I do wonder how much the MI mega-fail had to do with Michigan specific items. They have not had a well attended recognized primary Democratic primary in a while (given that 2012 was uncontested and in 2008 they had that whole going early against the rules and only Clinton actually running there bit) … maybe their likely voter lists and screens were impacted? Or maybe the means by which Team Clinton does outreach were?

This one is interesting to me.

[li]Some Clinton supporters chose to vote in the Republican primary.[/li][/ul]

I’m pretty sure it’s minor in the grand scheme of things, but there’s no way to know for sure. The reason why it caught my eye is that there is poster here who is a Clinton supporter and not only says that he’ll crossover vote for Trump but also has a threadadvocating that others do the same. I also believe this poster is voting in Missouri on 3/15.

electoral-vote.com also has a portmortem in which they blame the screen for “likely voters”


So far at least the polls are not narrowing as pollsters rejigger. Florida and Illinois polling averages are both moving towards bigger Clinton wins; Ohio no new data. NC marginally more to Clinton.

I will still be surprised if they are not a wee bit underestimating how Sanders will do; I am expecting more modest Clinton wins in these states.

But how likely do you think this result is?

For this thread the subject is not the tactic of downplaying delegates and raising the importance of how many states won and selectively naming particular ones as the ones that matter (Virginia and Florida … not important at all). Just if that statement “After Michigan …” is a reality based in any way.

Hey finally a Missouri poll! Only 145 LV and who knows what cred the polling house has. FWIW though Clinton +7 with a MOE of 8.

I believe at least Illinois and Ohio have had early voting going on which I think throws wrenches in the polling models as well.

Primary polls always have a large margin of error. Michigan is one of the largest we’ve ever seen, but 10 points off is not at all unusual. Trump has pretty consistently underperformed everywhere but the South, and Sanders has pretty consistently overperformed except in the South.

Fivethirtyeigtht.com gives the weighting in their polling algorithm. They give it around a .22 compared to another recent poll which has a .77. Though what specific factors affect those numbers is left unstated.

(The second link goes to the polling site’s front page, which, as of this post, has a press release announcing the results of the poll. I was unable to find a direct link to the press release or to a graphic showing the poll.)

Thank you for that.

Meanwhile at least one polling house is breaking from the pack and showing a narrowed Sanders deficit in Ohio and a LEAD in Illinois! Florida though not much different.

Ohio Clinton +9
Illinois Sanders +2
Florida Clinton +28

On the Illinois side another pollster has also shown it narrowing, albeit still a Clinton lead: down to Clinton +6. For Ohio and Florida though they are with the rest of the pack.

Again for this thread the interest is why the divergence between pollsters and if any will capture reality. Was Michigan a one-off polling epic fail or are the pollsters structually missing something more broadly?

Well as far as why the divergence, I think it’s common knowledge that the poll industry is in a bit of flux right now. They are still trying to learn how to deal with decreased participation, the ongoing death of landlines and how to do a useful internet survey.