It Sucks but You have to Do the Math.

Yesterday, we discussed The New York Times’ difficulties in dealing with delegate math.

538’s sins are more serious. In plotting out both Democratic candidates’ paths to the nomination, 538 relies on the short cut described as the following “based on recent polling demographics, fundraising and Facebook data” rather than digging into the differences among states, territories, districts and even areas within districts.

In doing so, it breaks four important rules.

Rule number one on the Democratic side has to be--and this needs to repeated over and over again--caucuses are not primaries and primaries are not caucuses. The New Hampshire primary had three times the participation of the Iowa caucus, and the Iowa caucus, because of first-in-the-nation status and tradition, has a higher participation rate than any other caucus state. Thus, it doesn’t make sense to use national polling to make any assumptions about caucus states with low participation.

This is particularly true for today’s Nevada caucus. Yes, Nevada is important. The demographics should be good enough for Clinton to overcome the ways in which caucuses disadvantage her and advantage activist candidates. But if they are not, it speaks more to the caucus process than to the overall contest.

Rule number two is that territories count. 538 makes the horrendous decision to simply not mention the territories, such as Guam, probably because they are difficult to poll and project. This is unacceptable. The territories have 99 delegates combined, including Democrats aboard. They can matter. Puerto Rico, for example, has 60 delegates.

An example from a territory also illustrates how important it is to understand individual contests. The Virgin Islands has seven delegates: St. Thomas with four delegates and St. Croix with three. Because St. Thomas has four delegates, a candidate needs to receive 62.5% +1 of the vote to earn an extra delegate there. Because St. Croix has three delegates, the winner of the St. Croix caucus automatically gets an extra delegate. Given how few people participate in the St. Croix caucus (746 in 2008), an individual voter on St. Croix might have more influence than any other voter in the country.

Ignoring territories is a mistake that particularly disadvantages Clinton who is likely to do very well in Puerto Rico (Clinton won 68 to 31 in the vote, 38 to 17 in delegates, even though the race was effectively over at that point in 2008).

The third rule is that the Democratic contest is not really contests in a state or a territory as much as it is individual contests within those states and territories. On the Democratic side, there are more than 500 individual contests and each of those contests can produce a different number of delegates. The only way to meaningfully predict the outcome in a state is to understand each of these contests.

Wednesday, we did the math for Nevada and South Carolina. The net reward for winning Nevada at the statewide level is two delegates. The net reward for getting 56.25% in South Carolina’s 6th district[Where Secretary Clinton just picked up the Congressman], which has eight delegates, is also two delegates. Getting this deep into the weeds might not be fun but it is also the only way to truly get this race.

Here is how the delegate math works in a bigger state to give a sense of how complicated this can be and why winning a state does not mean winning its delegates.

The 538 Chart says Sanders can lose Ohio by 2% to have an even shot at the nomination. Here’s the problem with that. Ohio’s districts are overwhelming even numbered; out of 16 districts,15 award an even number of districts. That means, Sanders and Clinton start off very likely to split the delegates evenly, regardless of outcome. The statewide buckets are odd so a win statewide is worth at least two delegates. However, the one odd-numbered district, the 11^th district, has 17 delegates. So winning 55.8% of the vote there is worth three delegates, creating a 10-7 split compared to just two delegates for a statewide win.

So Clinton winning by even 2% seems to be worth at least three delegates and possibly five delegates. Clinton could win statewide and lose the 11th in theory, but that seems highly unlikely given that it has a large African American population and current polling showing Clinton’s strength among that group. Since no other districts have an odd number of delegates and because only three other districts have more than four delegates, the odds of a split beyond that, particularly those that benefit Sanders, are small. The Cook Political Report’s Democratic Delegate chart, which 538 endorsed, says Sanders needs to lose Ohio by a single delegate. Sure, the difference between a one-delegate win and a five-delegate win may not seem all that large but these differences, added up over time, make things difficult for Sanders.

The fourth and final rule is the value of the African American vote in the Democratic race. 538 minimizes that by making the assumption that winning a state is the same as winning the delegates in that state, thus overlooking the power of particular districts with strong African American majorities. Ohio and its 11^th Congressional district again provide a good example. Gerrymandering and Democratic rules have increased the value of the African American vote. The 11th Congressional district is 54% African American and its primary electorate is probably closer to 60%. A candidate receiving 61.7% of the vote in this 17-delegate district would be entitled to an 11-6 split. A candidate cracking the 11th at this rate could lose the statewide vote and still win the delegate count in Ohio so long as they took 43.75% in each district and got greater than 45% statewide.

2008 is a good guide. Clinton won by 9 points statewide, which netted just five at-large delegates (today it would be worth four delegates because of a change in the number of delegates). Clinton won 13 of 18 Congressional districts yet netted just two delegates at the district level. One of the primary reasons is that cracking in four-delegate districts is hard. Clinton didn’t crack in any of the seven such districts in 2008, Obama cracked in one. This year, 12 of Ohio’s 16 districts have four delegates. To give you an idea of the scale required, Sanders’ margin of victory in New Hampshire is not large enough to crack a four-delegate district. In addition, the five-delegate districts received the extra delegate as a reward for strong Democratic turnout.

This is the kind of analysis that is required for every single state and district. And we will provide a best guess to every single delegate so long as the race is contested.

538 cobbled together a heuristic model based on national polling and other data. It simply is not good enough to provide an accurate representation of where the race is. It sucks that you have to do the math, but you have to do the math.