As I did early for the GOP, I have built a Democratic Delegate Projection Model. It is available here.
This table shows the basics:
| Current Polling | Sanders wins Iowa | |||||
| Clinton | Sanders | Clinton | Sanders | |||
| Delegates before Super Tuesday | 156 | 103 | 53 | 80 | 76 | |
| Delegates On Super Tuesday | 980 | 705 | 275 | 538 | 440 | |
| Deep South | 715 | 517 | 162 | 422 | 306 | |
| Other | 250 | 153 | 97 | 116 | 134 | |
| Total - Before March 2nd | 1136 | 808 | 328 | 618 | 516 | |
| Total - states after March 2 | 2916 | 1914 | 1002 | 1429 | 1486 | |
| Total | 4052 | 2722 | 1330 | 2047 | 2002 |
These are for elected delegates only. There are 713 Super Delegates, and Clinton has a lead among them.
The only way to do a decent delegate projection is to model the results AT EVERY CD. Democratic delegates are in two ways:
1. Delegates awarded based on state results
2. Delegates awarded based on CD results
The second becomes important because of how results are rounded. This leads to unexpected results, which usually benefit the front runner.
So where do the Sanders numbers come from if he wins Iowa? How do you know?
This model builds on work that I did in 2007, and that was subsequently cited by Nate Silver. This is a pretty significant upgrade both of his and my work, in part because the polling data behind it is significantly more robust. The polling detail is here.
There are 13 instances since 1976 in which a front runner or someone tied for the lead has been beaten in Iowa in NH. I have excluded a couple (eg 1976 Democratic) for a variety of reasons (in ‘76 the front runner did not contest NH) If you run a linear regression comparing their national polling before and after the front runner is beaten, you get the following results:
| Intercept | 5.903613917 |
| Prior National Poll | 0.614479111 |
| Won either Iowa, NH or both (either yes or no) | 16.28478719 |
What the hell does that mean? Well, it means you can model the result of a front runner by multiplying the follows
Using that information, we can estimate the effect of a Sanders win in Iowa as follows:
| Clinton | 56 | 40.4 |
| Sanders | 30 | 40.6 |
What if Sanders loses Iowa, but wins New Hampshire?
This is a complicated question, and I will explore this in a later post. Bottom line Iowa’s effect on New Hampshire is not as simple as win or lose — exceeding expectations matters greatly as well. The best example of this is Gary Hart, who beat exceptions by 10 but still lost Iowa by 49-16. However, beating expectations gave him a bounce that led to his New Hampshire victory 8 days later. Here are some examples of front runners winning Iowa and the follow on effect in NH:
| Mondale | 36.75 | 48.9 | 37.75 | 27.9 |
| Hart | 7.53 | 16.5 | 23.25 | 37.3 |
| Dole | 24.3 | 26 | 24.4 | 26 |
| Buchanan | 13.6 | 23 | 23.7 | 27 |
| Bush | 34.8 | 41 | 31.4 | 30.3 |
| McCain | 38.2 | 4.67 | 37.2 | 48.5 |
Gore | 43 | 62.8 | 50.2 | 49.7 |
| Bradley | 44.57 | 36.3 | 42.4 | 45.6 |
it is worth noting the results in 2000. Gore destroyed Bradley in NH, and did see a bounce in NH that subsequently partially receded. Bush won Iowa decisively, but it had little impact on New Hampshire as his win was expected.
If Iowa is close, I suspect Bernie will benefit. The problem, though, is that losing Iowa may cause the press to focus on the GOP race more than the Democratic race (this happened in 2000).
Sanders, Clinton and the South
By far the biggest challenge facing Sanders is the African American vote. To account for this, I adjusted the projections in two ways:
1. Where no state polling is available for southern states, I use the average of SOUTHERN polls, not national polls.
2. I adjusted the projected bounce Sanders would receive in Southern States by the percentage of vote in that state that was cast by African Americans. This adjustment is not perfect — there is little empirical date behind it. But it does ensure the model is accounting for the challenge Sanders faces in the South.
You can see this on this page, where I compute the averages for each state.