Revisiting the jungle

In Louisiana there's another statewide election scheduled. And not the one scheduled for December 6, which will decide whether incumbent Sen. Mary Landrieu will keep her seat for another six years or if Rep. Bill Cassidy can take it from her. In 2002, the runoff fight for this same seat was termed "Operation Icing on the Cake" by Republican operatives; the only thing a victory in Louisiana would have done for Republicans was pad their majority from 51 seats to 52. And this year it's similar, although since Republicans are more likely than not to lose several seats in the 2016 elections--which favor Democrats for the same reasons 2014 favored Republicans--every seat they can get will be helpful. But despite the dearth of polls since Election Day, there's not much reason to suspect that Landrieu has any better chances of keeping her seat after the December runoff. Although her effort to bring the Keystone XL pipeline to a vote on the Senate floor--where Landrieu claims to have a filibuster-proof sixty-seat majority--has finally succeeded, it has really lost much of its importance election-wise since 1) it turns out a lot of voters didn't really care about her seniority when casting their votes and 2) her seniority wouldn't mean much in the 114th Congress anyway, seeing as she'd be demoted to the ranking Democratic member of the Energy Committee, from the chairmanship that she holds now. (Meanwhile, Sen. Lisa Murkowski of Alaska appears to be very pleased about her own rise to the chairmanship.) The DSCC has pulled its spending from the state and Landrieu looks to still be deep in it. LIKELY REPUBLICAN

That's all we have to say about that. The real competitive election in the Pelican State is next year, when Louisiana (as well as Kentucky and Mississippi) holds its gubernatorial election. Not even counting the fact that this election is not held in a midterm year (it's one of five states that holds its state elections in odd-numbered years), there are some key differences between 2015's gubernatorial and 2014's Senate election that make next year's more complicated to evaluate:

• No incumbency advantage. Republican Gov. Bobby Jindal, in the middle of his second consecutive term, is not allowed to run again in 2015, so the field is wide open for Republicans. By contrast, this year Mary Landrieu was running for re-election, so no credible Democratic challengers rose to the occasion.
• No clear front-runner for second place. One Suffolk University survey from October shows that Republican Sen. David Vitter is the front-runner but is polling well below the 50% needed to avoid a runoff. It shows New Orleans Mayor (and younger brother of the U.S. senator) Mitch Landrieu in second place, but Landrieu has not officially declared his candidacy, and at least two other candidates could conceivably take second place while keeping Vitter under 50%.

For now we'll evaluate the model using four candidates: Vitter, Landrieu, Republican Lieutenant Governor Jay Dardenne, and Democratic House Minority Leader John Bel Edwards. As we did for the Senate race in South Dakota this year, we expanded the simulator to allow for four candidates, which we're using for the Republican primary for Governor of Kentucky (which is a lot simpler, lacking the French-style jungle primary laws Louisiana has). However, we've expanded it even further for Louisiana because we'd like to know the probability of a particular candidate making it to the runoff, as well as the probabilities that each candidate wins an outright majority (29% of voters are still undecided, which would be enough to send Vitter or Landrieu to outright victory on October 24, when the elections are held, if they broke heavily for one candidate). In 2014 we were able to avoid having to do this, since Landrieu and Cassidy led the rest of the pack considerably and we could safely assume that they would advance to a runoff. But 2015 gets a little more complicated.

What new simulations have to account for are 1) the probability that one candidate wins an outright majority, 2) the probability of each candidate winning an outright majority, and 3) the probability of each candidate making it to the runoff. #3 in particular is a little tricky, since it equates to the probability of each candidate making first or second place, conditional on the probability that no candidate wins above 50% of the vote. The first steps of the simulator are analogous to the old simulator--use the polling average and approval ratings (we don't have fundraising data yet) to draw simulated vote shares.

The first thing it does is check to see if one candidate has received more than 50%; if it does, it assigns that candidate a win outright, adds it to the candidate's running total, and returns to the beginning of the loop, where the process repeats. For example, if Vitter receives more than 50% of the vote, it assigns the variable VV (I'll come back to why it's called that) a value of 1. 

If no candidate receives more than 50%, the simulator then compares the candidates' simulated vote shares and orders them from greatest to lowest: so, for example, Vitter > Landrieu > Dardenne > Edwards. It then records the results of the top two performers and assigns another variable a value of 1--in this case, Vitter and Landrieu come in first and second place, respectively, so it would assign the variable VL a value of 1. Note that it treats the above result and Vitter > Landrieu > Edwards > Dardenne as the same result, because the two who advance to the runoff are still the same. It then adds that to a running count for each variable.

This gets repeated 99,999 more times. You may have noticed that the variable names are simply the initials of the first and second place finishers, in that order--so for Landrieu > Vitter > Edwards > Dardenne, the variable assigned a value of 1 is LV, not VL. When a candidate wins a majority, the variable assigned a value of 1 is his initial twice--so a Vitter majority is VV, a Landrieu majority is LL, etc. The totals are divided, of course, by 1000 to get percentages, which are then organized into a matrix in R:

Merasmus <- c(DDprob, EDprob, LDprob, VDprob, DEprob, EEprob, LEprob, VEprob, DLprob, ELprob, LLprob, VLprob, DVprob, EVprob, LVprob, VVprob)
    BonusDucks <- matrix(Merasmus,4,4)
    BonusDucks
}

> toptwo(9.80, 3.80, 22.60, 31.60, 0.58, 0.58, 0.58, 0.58, 3.00, 0.58, 0.240, 0.188, 0.248, 0.265)
       [,1]  [,2]   [,3]   [,4]
[1,]  0.000 0.305  1.959  3.250
[2,]  0.251 0.000  0.679  1.045
[3,]  3.634 1.635  6.061 16.718
[4,] 10.653 5.019 29.181 19.610
 

On the rows are the candidate who wins first place, and on the columns are the candidate who wins second. They're arranged in alphabetical order--so for example, element [1,2] is the probability that Dardenne comes in first and Edwards comes in second. On the diagonal are the probabilities that each candidate wins a majority without needing to go to a runoff. You'll notice that [1,1] and [2,2] are both zero--in no simulated election did Dardenne or Edwards win a majority of the vote in the jungle primary, which makes sense, considering that their polling average plus the full number of undecided votes was still under 50%. 

What this matrix can basically tell you is the probability of a particular matchup in a runoff. If you want to see the probability that Mitch Landrieu (#3) and David Vitter (#4) advance to the runoff, you look at elements [3,4] and [4,3], since those are the elements that have both Landrieu and Vitter in them, and add them up, so that probability is about 46%.

That's just how we'll be reporting these probabilities as the election approaches; it's the most efficient way to present them, and not too hard to understand. There's barely any polling out for any 2015 elections, anyway, so it'll be a while.