March 26, 2016
Our goal for today is to estimate margins for each of the caucuses. Bernie Sanders is very likely to sweep the day. That’s a starting point, not a finishing one. He’s trailing by a pile of delegates. In order to begin catching up, begin reversing the narrative, he needs 3 landslides to follow his Idaho and Utah stompings.
Our one survey from early January has Hillary ahead of Bernie 44 to 41. It’s time to play the Poll Adjustment Game. Leaving aside the far more intensively contested and polled Iowa and Nevada events, Bernie always does significantly better than the limited data indicates.
How much should we expect here?
Let’s look back so we can predict forward. We’re going to ignore Iowa and Nevada. Both had multiple current polls at the time of voting. Hillary had a major organizational advantage in both states.
Her campaign began building on the ground months ahead of Sanders. This isn’t the case in the other caucus states. There isn’t the history of participation of Iowa, the institutional influence of Nevada.
One more adjustment. In November/December Bernie wasn’t doing very well in polls. Hillary had recovered from her September slump, wasn’t bruised yet by Trump. Anything from mid-January to the present is roughly equivalent.
Remember, Bernie does better in open caucuses. These polls may not have well accounted for crossover votes. We would expect more voting day inflation in open events. Alaska is closed.
Maine and Nebraska were data free, so we have five examples:
Poll: Hillary +28
Adjustment: subtract 15
Result: Bernie +19 (up 32 from adjusted poll)
Poll: Hillary +34
Result: Bernie +23 (up 57)
Poll: Hillary +10
Result: Bernie +35 (up 45)
Poll: Bernie +2
Result: Bernie +57 (up 55)
Poll: Bernie +8
Result: Bernie +59 (up 51)
In closed caucuses, he added 39 points to the polling total. In open/mostly open caucuses, Bernie improved by 54. Alaska starts at minus 3. The poll was early enough in January for us to give a 5 point adjustment, moving him to plus 2. Add the closed caucus average, and voila!
Bernie Sanders 70.2%
Hillary Clinton 29.1%
Yes, this breaks a thousand rules about sample size. There’s a reason FiveThirtyEight doesn’t do projections based on one 2.5 month old poll and a few recent voting fragments. Nate Silver is getting paid for accuracy. This is free. Plus, you can chuckle to yourself when this winds up dead-on.