Last week, we looked at a simple demographic model that has had a strong relationship with the vote share won by Bernie Sanders in various states so far in the Democratic Party primary contests. The model had four variables – the percentage of black population in a state, whether the state conducts a caucus or a primary, whether the state is located in New England and the percentage of Hispanic population in the state.
Since then, the results of four primaries and caucuses have come out – Louisiana, Nebraska, Kansas and Maine. The results of Louisiana and Nebraska were along the expected lines, with Sanders over-achieving his predicted vote share by 2% in the former and almost tying his predicted vote share in the latter. However, in Kansas and Maine, his numbers were significantly different from those predicted by the model – in Kansas, his numbers were higher by around 14% whereas in Maine, they were lower by around 12%. Overall though, the two extreme deviations almost negated each other, indicating that Sanders is still not consistently over-performing or under-performing the model.
I made two changes to the model published in the last post. Firstly, I incorporated the results of the four states that voted over the weekend. This resulted in only modest changes in the outputs. Secondly, instead of using the vote share of Sanders as the dependent variable in the model, I used the difference of vote share between Hillary Clinton and Sanders as the dependent variable. Thus, now we obtain the difference in vote shares between the two candidates across various states as the output of the model, which can easily be used to predict the percentage votes obtained by the two candidates by assuming that the votes obtained by any third candidate would be too small to warrant any delegate in any state. This is an improvement over the previous model which was allocating votes to Clinton by assuming everyone else not voting for Sanders would be voting for her.
The predicted vote share as per the revised model has been mentioned in the following table:
|State||Date of Election||Unpledged Delegate Count||Sander’s Vote Share||Clinton’s Vote Share|
|District of Columbia||14-06-2016||20||7%||93%|
If the performance of the two candidates exactly matches their predicted vote share in each of the states as well as the individual congressional districts within each state, Sanders will end up winning around 1166 of the remaining delegates on offer, as opposed to 1697 delegates projected to be won by Clinton.
Sanders is currently trailing among pledged delegates by a margin of 195 delegates (480 delegates as against 675 delegates of Clinton). If he wants to make this up from the remaining states, his vote share needs to increase by around 13 percentage points and that of Clinton needs to be reduce by around 13 percentage points.
That is, however, easier said than done. It means Sanders will not only have to crush Clinton in the western caucus states and in the remaining New England states, but he will also have to more or less sweep the Mid West (by winning in states like Michigan, Missourie, Ohio, Wisconsin and Illinois), Appalachia (West Virginia and Kentucky) and also in the delegate rich states of north-east like New Jersey and Pennslylvania. On top of that, he will have to register a healthy win in California.
Even then, it might not be enough as Clinton has a massive lead among super-delegates who are not bound to vote according to the results of their respective primaries or caucuses. If Sanders were to overcome the delegate lead among both pledged and unpledged delegates, he would need a vote swing of around 21% in his favour and a similar vote swing against Clinton, which is almost impossible.
One major unknown factor in the equation is that we still do not know how the large Mid-west states like Michigan, Ohio, Wisconsin, Indiana and Illinois are going to vote. Our model projects Sanders to lose all of them; however, Sanders need to win these states in order to retain any chance of taking the nomination. These states are different from the mid-western states that have voted so far (Iowa, Minnesota, Nebraska, etc.). Not only they have a large population of disaffected blue collar white voters (who are expected to be more inclined to support Sanders), they also have a larger proportion of African American voters (who have supported Clinton overwhelmingly so far). It is possible that such states will behave in a way which is markedly different from how the states have voted so far.
This is why today’s vote in Michigan is so important. Our model currently shows Sanders winning around 39% vote in the state. However, if he were to be on course to win the nomination, he will need to get a vote share of around 52% in the state and win it narrowly. If he manages to win the state, he will show that he may be competitive across the entire Rust Belt, possibly the most critical collection of states yet to vote in the Democratic Primary.
On the other hand, if Sanders remains stuck in the high 30s or the low 40s, it will mean that voters in industrial Midwest are voting mostly the same way as rest of the country. This will mean Sanders will also lose each of the five large states slated to vote on 15th March (Florida, Missouri, Illinois, North Carolina and Ohio), which together award 691 delegates. His task will become more or less impossible after that and the rest of the primary contest will remain only of academic interest.