There is a recommended diary right now on delegate math that claims the Slate delegate projector is flawed because it assumes a proportional allocation of delegates, whereas the actual delegate process is highly non-linear -- you have to cross specific percentage thresholds (like 62.5% of the two-candidate vote) to gain any extra delegates. Since Clinton is unlikely to cross these thresholds on average , the diary argues that even if Clinton goes on to win, for example, every single remaining contest by 55-45%, her delegate margin would end up much less than her popular vote margin.
The argument is quite misleading, because it looks at averages but ignores the all-important cross-district variance. In fact, the diary is completely wrong because of that omission. Don't just believe me -- follow me to the flip and I will illustrate.
Clinton won the TX popular vote by 3% and Burnt Orange shows her with a 4 delegate lead out of 126 delegates. How did she do that -- a 3% win in a given district gets you nothing. Furthermore, TX was supposedly unusually bad for Clinton, because the Hispanic "Clinton Districts" had a low number of delegates.
TX is just one example, but Slate points out that across states the delegates have in fact on average been allocated pretty close to proportionately. Not in every case, of course, but on average quite close.
How does this happen? What is wrong with the math in the recommended diary? Consider a state with two 4 delegate districts, each with the same number of primary voters. Say that Clinton wins the state 55%-45%. Now, a 55-45 split in a given district wins you 2 of 4 delegates, so Clinton and Obama split the state delegates, right?
No. It depends on the cross-district variance in the votes. Say that Clinton wins district 1 by 65-35 and loses district 2 by 45-55. Her overall margin is 10%, but she wins 2 delegates in district 2 (because 45 rounds to 50) and wins 3 delegates in district 1 (because 65 rounds to 75). She gets 5 of 8 delegates on a 10% win!
The recommended diary is wrong, because it makes the assumption that all vote outcomes will be equal (no cross-district variance). In TX, Clinton got massive margins in Hispanic areas and crossed the necessary thresholds to get a more-than-proportional number of delegates. This offset her other disadvantages and gave her an approximately proportional number of delegates.
Now, in a small state you could see a strange outcome, in either direction. If things go just so, you could see a strange result in a big state as well, but it often requires a very particular set of circumstances. Across many states and/or districts, these things tend to average out, which is why Slate can correctly aruge that its calculator predicts past contests pretty well on average. Doing better than "on average" requires so much detail on exact nature of the cross-district variance that it is very hard to do at all (see again: very sophisticated but ultimately incorrect pre-TX analysis.)
Slate's on-average proportionality argument can only be correct if there is a lot of cross-district variance. But look at the results for any big state -- you will see all kinds of blow-outs in both directions. The "equal margins across districts" assumption is just really bad.
In particular, it is a bad mistake to plug equal 55-45 assumptions into all districts and claim that an on average a 10% margin gains a candidate less than 10% of delegates. In practice, cross-district variance is typically quite big and it matters a lot. In fact, in past primaries, on average, a 10% vote margin got you about 10% of the delegates. Proportional delegate allocation isn't a bad assumption going forward and pre-TX analysis shows how hard it is to beat.