The ECONOMIST polling model has really been shaken up.

[brooklynite]

Link to tweet

[Turin_C3PO]

1. It's looking like

it’ll be a race between Bloomberg and Sanders. My two least favorites. I was hoping for Joe or Warren. Still, Biden could make a comeback if he does well in SC.

[Demsrule86]

2. I am waiting to see what happens in South Carolina

[LeftTurn3623]

3. I hope Biden makes a comeback

I want nothing to do with Bloomberg

[PatSeg]

5. My two least favorites as well

I'm having trouble recognizing the Democratic party some days.

[redstateblues]

7. Joe has to win SC convincingly or he is in trouble

He has bet his whole campaign on SC. I sent him another donation last night hoping he can. At the end of the day we have to beat Trump

[Mike 03]

4. My god, Bloomberg's looks like a missile launch. nt

[DanTex]

6. That's because the Economist uses a bizarre smoothing model.

For instance, those lines you see are not actually a graph of the Economist's own forecasts over time. In October, for example, they had Warren and Biden very close to each other, but now that graph has a sizeable gap even at the closest point.

What they do is apply a smoothing to the entire set of polls, and their statistical technique grossly overemphasizes the trend versus the moving average. That results in severe distortions on the boundaries of the dataset where there is data on one side but not the other. In other words, their estimate for, say, last August might be reasonable, because there's data from July, August, and September. But looks like they've done nothing at all in that model to deal with boundary effects.

And this is why they've got Bloomberg tied with Sanders even though Bernie has been significantly ahead of Bloomberg in every single poll so far.