(2) With the exception of New Hampshire, Obama’s polls numbers have tended to rise as an election approaches. This could be because the Democratic electorate is already very familiar with Clinton and any movement in support would go from Clinton to Obama (even though undecided voters may break evenly). Moreover, Obama’s appeal to non-traditional constituencies has allowed him to bring in voters that pollsters aren’t accustomed to accounting for in a primary. While Obama met expectations in New Hampshire, he exceeded them in every other contest.
That’s why Obama and Edwards are RISING! If you follow me to the end, there’s a method to the madness that is these projections.
.
In a previous diary entry, I broke down the delegate process with special attention on the delegates at stake on ‘Super Tuesday’. 22 states hold elections on February 5th and there are 1678 pledged delegates up for grabs during this contest. By my current estimates, Clinton is slated to pick up 893 delegates, followed by Obama with 657 and Edwards with 128.
Where did this numbers come from? Not of out of thin air, if that is your worry. In advance of the coming deluge of polling data, I would like to outline a methodology for projecting delegate counts in each of the 22 states that hold elections on February 5th. Hopefully, this diary can serve as a valuable reference for future projections based on the most up-to-date polling data.
.
Introduction:
With the way the process works, we aren’t assured of a straightforward correlation between a candidate’s popular vote percentage and that candidate’s delegate share. The relationship between the popular vote percentage and the delegate share is complicated by a couple of factors.
a) Explicit Thresholds
As most of you are aware, caucuses have a 15% viability rule. A candidate who fails to receive 15% of the vote in a given precinct cannot be awarded delegates from that precinct.
In addition, some primaries have their own version of the viability rule. States like California, New Hampshire and Mississippi have explicit thresholds that a candidate must meet in order to receive ANY pledged delegates. In almost all cases the threshold is set at 15% of the vote, just like in the caucuses. For example, if Edwards receives 14% of the vote statewide in California, he will receive ZERO at-large delegates. The delegates are then apportioned among the remaining candidates who exceed the 15% threshold.
b) Implicit Thresholds
This applies mainly to the awarding of district-level delegates and also to the awarding of some at-large delegates in smaller states. District level delegates are awarded on the basis of the popular vote within a given congressional district. Usually, there are 4-5 delegates allocated to each Congressional district. Let’s take the hypothetical case of a district with 4 delegates.
A candidate with 25% of the popular vote in that district is guaranteed at least one district-level delegate. However, when that candidate’s vote drops below 25% in this district, he or she is no longer guaranteed a delegate. Winning even one delegate now depends on the candidate’s performance relative to the field. In a crowded field, this candidate stands a high chance of winning at least one district-level delegate because of his or her relative strength. However, the converse is true in a small field. With two evenly matched leading opponents, a third candidate pulling in 20% of the vote would receive zero delegates in this district.
This implicit threshold also applies to some of the smaller states which have only a few at-large delegates.
The 2004 Democratic Primary Elections
The dynamic in the 2004 race is very similar to the dynamic in the current race. In 2004, John Kerry was the presumptive nominee after only a few early victories. Still, John Edwards did win a sizeable share of the vote in many elections until Kerry clinched the nomination. Although the current race differs with respect to the relative dominance of Clinton over Obama, much of the vote is expected to be split between only two candidates, just as it did in 2004. Moreover, both races featured a third candidate receiving a small, but significant portion of the popular vote (Edwards in 2008, Dean/Clark in 2004).
Does the 2004 data give us some hints about projecting delegate counts from the popular vote? Take a look at the following graphs which relate the candidates’ delegate share(%) to their share of the popular vote(%) in all the 2004 contests.
* - the first graph is data from large states (50 or more pledged delegates) and the second one is data from small states (50 or fewer pledged delegates)
.
.
.
While it is clear that there isn’t a one-to-one correlation between a candidate’s delegate share and his or her popular vote percentage, there is a distinct relationship. Can we tease out this relationship? I believe so.
.
Methodology:
Big States vs. Small States
If you pay close attention to the two graphs above, you will notice some differences in the region where the popular vote percentage is between 5% and 20%. In smaller states, a candidate who got less than 16% of the vote received no delegates. However, in larger states, a candidate garnering even 8% of the vote could be expected to win at least some share of the delegates. Why the difference?
The reason is because of the awarding of district-level delegates. Let’s take a look at a quick example to see why. Assume candidates A, B and C receive 45%, 45% and 10% of the popular vote respectively in BOTH North Dakota and California. What share of the delegates could they be expected to receive in each state?
In the case of North Dakota, our task is pretty simple since there is only one congressional district with 8 district-level delegates. Candidate C, with 10% of the statewide vote received ZERO district level delegates because of the explicit 15% threshold in North Dakota. Candidates A and B each split North Dakota’s district-level delegates with 4 a piece.
In California, the picture isn’t quite so simple. While candidate C received 10% of the statewide vote, we still don’t know what percentage of the vote candidate C received in each of the state’s 53 congressional districts. Candidate C could have received 30% of the vote in some districts and 2% in others. We simply don’t know. Therefore, it is eminently plausible that candidate C did well enough in at least a few congressional districts so as to receive some district-level delegates. The above data shows that this scenario is not only plausible, but that it bore out in a number of large states in 2004.
Therefore, even with the same statewide popular vote margins, a candidate will receive different delegate shares!
Data Fit
Based on the above graphs, I decided to treat the relationship between the delegate share and the popular vote percentage as a piece-wise function, with two pieces. For large states, I assumed that a popular vote percentage below 6% would not translate into any delegates. I performed a series of fits (straight line, logarithmic, exponential, power) for data points above a popular vote percentage of 6%. Similarly, for small states, I set my threshold at 14% and performed a series of fits for data above that threshold. Here are the results (best data fits):
.
.
.
For Large States,
a) if the popular vote is 6% or fewer, a candidate’s delegate share
= 0%
b) if the popular vote is greater than 6%, a candidate’s delegate share
= [-8.034 + (1.380 * (Popular Vote in percent))]%
For Small States,
a) if the popular vote is 14% or fewer, a candidate’s delegate share
= 0%
b) if the popular vote is greater than 14%, a candidate’s delegate share
= [-104.6 + (42.8 * LN(Popular Vote in percent))]
.
Going from a Model to Projections:
Defining ‘Large’ and ‘Small’ States
For my purposes thus far, I defined a ‘large’ state as one having more than 50 pledged delegates and a ‘small’ state as having 50 or fewer pledged delegates. This cutoff was arbitrary and on closer inspection I found that states with between 35 and 60 delegates could easily be described by either model. Preferring to use a straight-line model as often as possible, I now place the cutoff between ‘large’ and ‘small’ states at 35 pledged delegates.
Preservation of Delegate Shares
One obvious flaw in the above model is that if you have two candidates splitting the vote 50%-50%, simply plugging in for these popular vote percentages gives you delegate shares of above 60% for each candidate. Clearly, a total delegate share of 100%, across all candidates, will not be preserved by a simple plug-and-chug using the above formulas.
Adjustment #1
In a three candidate dynamic, I calculate the delegate share of the person with the lowest popular vote percentage using the formulae above. With this delegate share, I ascertain how many delegates this candidate would receive from the state in question.
Adjustment #2
With delegates ‘won’ by the third candidate accounted for, now apportion the remaining delegates amongst the top two candidates in proportion to their popular vote percentages relative to each other. No messy formulas needed here.
.
Comparison of Model with Results to Date:
Iowa Results
Popular Vote - Obama 38%, Edwards 29%, Clinton 29%
Delegates – Obama 16, Edwards 14, Clinton 15
Projections – Obama 17, Edwards 14, Clinton 14
New Hampshire Results
Popular Vote - Obama 37%, Edwards 17%, Clinton 39%
Delegates – Obama 9, Edwards 4, Clinton 9
Projections – Obama 9, Edwards 4, Clinton 9
Nevada Results
Popular Vote - Obama [Clinton] 51%, Edwards 4%, Clinton [Obama] 45%
Delegates – Obama [Clinton] 12, Edwards 0, Clinton [Obama] 13
Projections – Obama [Clinton] 13, Edwards 0, Clinton [Obama] 12
South Carolina Results
Popular Vote - Obama 55%, Edwards 18%, Clinton 27%
Delegates – Obama 25, Edwards 8, Clinton 12
Projections – Obama 25, Edwards 8, Clinton 12
The projections were off by a delegate each in Nevada and Iowa. However, overall the agreement with results was quite good.
.
Projecting ‘Super Tuesday’:
There are 22 states up for grabs and post-South Carolina polls are only just starting to trickle in. Not wanting to reconcile any ‘bounce’ between the pre and post-South Carolina polls, I will confine my projections only to post-South Carolina polling data.
The sizeable number of undecided voters in many of the polls adds another layer of complexity to the projections. For the purposes of my projections I have ‘pushed’ undecided voters into the Clinton or Obama columns in proportion to the decided Clinton-Obama vote. My justifications for doing so are two fold:
- The assumption that undecided voters will prefer to pick between the candidates who have a realistic shot at winning the nomination
- Edwards’ lack of financial resources and free media coverage will prevent him from reaching uncommitted voters in the way that Clinton and Obama can.
I will be more than happy to alter these assumptions based on informed suggestions from Kossacks or the movement of polls in the coming days. I am not out to dismiss Edwards or diminish his campaign. I am trying to work with what I think are the most reasonable set of assumptions.
.
Unveiling the Projections [T minus 7 days]:
Many of the states haven’t been polled since the South Carolina Primary. To make up for this temporary lack of polling data, I am substituting poller.com’s nationwide trends for those states with out polls. As polls come in over the next few days, the state-by-state accuracy will improve dramatically. For now, here are the projections:
.
.
Note: Relevant state polls not yet available for IL, NJ, MA, GA, MN, TN, AL, CT, AR, KS, NM, UT, ID, DE, AL, ND
Yes, I acknowledge that lack of polling in many states hurts the accuracy of these projections. But polls are due out for many of the big states in the next couple of days and these projections will improve in accuracy leading up to ‘Super Tuesday’. If you don’t like the numbers, I would appreciate it if you confined your criticism to perceived faults in the method laid out and not make things personal. As an Obama supporter myself, it should be abundantly clear than I’m not weighting my methods to make his chances appear better than they are.
.
Causes for Optimism as Polls Come Out:
(1) Using the nationwide numbers for Edwards serves to undercut his true expected delegate haul. This is because his support isn’t 12% uniformly across the nation. There are states where he has strong support and states where he has weaker support. Consider an extreme case – Edwards’ 12% is the result of him garnering 24% of the vote in half the ‘Super Tuesday’ states (weighted) and 0% in the remaining states. In this scenario Edwards would get 20%-30% of delegates in half the ‘Super Tuesday’ states, and none in the other half. Therefore, my current projections undersell his delegate haul in states without polls.
(2) With the exception of New Hampshire, Obama’s polls numbers have tended to rise as an election approaches. This could be because the Democratic electorate is already very familiar with Clinton and any movement in support would go from Clinton to Obama (even though undecided voters may break evenly). Moreover, Obama’s appeal to non-traditional constituencies has allowed him to bring in voters that pollsters aren’t accustomed to accounting for in a primary. While Obama met expectations in New Hampshire, he exceeded them in every other contest.
.
DATA
Data_Primary_Democrat_2004
ST_Delegate_Projections
Permalink | 81 comments