Sunday, March 30, 2014

How to Report the Invisible Primary

Thus far in the invisible primary of 2016 we've seen plenty of terrible reporting from all quarters of the press that are frankly too numerous to count, much less link up here.  However, there's some really good stuff going on out there.  For my money, Robert Costa of the Washington Post is quickly establishing himself as my personal favorite, as his coverage has really focused on what matters in things invisible primary.  A hallmark of the junk tends to be heavy on the various match ups for 2016 or even worse, the results of horrifically unrepresentative straw polls such as what we saw coming out of CPAC several weeks ago.  Not so for Costa...

Whilst multitasking between Fareed Zakaria, watching the T20 cricket championship (Australia, what are you doing?), and scanning headlines, I ran into this article by Costa and his colleague Phillip Rucker.  Kudos to them both.  

Let's set aside Jeb Bush's chances for winning the nomination for the moment, and focus on the coverage itself, as this is good stuff from the perspective of political scientists.  They quite literally are covering all the bases that matter.  First, they recognize that this isn't just a matter of a candidate trolling for support within the various parts of the party establishment, but the various parts of the party establishment actively vetting a candidate, or as they put it a "draft" is underway.  Second, they recognize the players here are a diverse network and are deeply concerned about both policy and the political feasibility of the noisiest candidate of the moment, Senator Rand Paul.  Third, they hit the importance of money at this point, and not just ability of Bush stuff his own coffers, but as I've noted before, make it rain for other Republicans across the country.  Fourth, they have a nice discussion on the strengths and weaknesses Bush brings to the table and how they will play to the various factions of the party.  If reporting the invisible primary was a four-banger of a motor, they'd be firing on all cylinders.

And here's where I really appreciate their work and where they separate themselves even more from most of their competition: While they eventually get around to some poll numbers and the (possible) issue of "Bush fatigue," it's at the tail end of the article and well embedded in a nuanced discussion of the factors that matter.  And they're not referencing some idiotic "Jeb vs. Hillary" face-off that litters most coverage at this point, but focusing on factors unique to Gov. Bush.  When we cannot even begin predict what the fundamentals of the 2016 contest will be, those sorts of head to head polls are superfluous.  

Overall, nicely done gentlemen!  I for one, will be following both these people's work rather carefully, as should you all.

 

Tuesday, March 18, 2014

Modeling the Endorsement Decision

For political scientists, March means one thing: conference season!  Though the macdaddy of conferences is labor day weekend, the ones I always look forward to the most (for a variety of reasons) are the MPSA and WPSA.  Even though I was crushed that I cannot attend the midwest this year (regional Model UN conferences... sigh), I'm in that sweet spot of a project where I'm furiously working on the models for my latest offering for the western.  It's really rough at this stage, mind you, and will need a significant amount of work in the coming months, but thus far it has some mildly interesting findings that are all about the invisible primary.

Back in late 2011 whilst still at Briar Cliff in Iowa, I was struck by what I thought was an invisible primary oddity.  Rep. Steve King took some time off blaming undocumented Hispanics for all of the nation's ills and expanding upon Todd Akin's legitimate rape comments, to rub elbows with a few of the presidential contenders; or perhaps more accurately, they interrupted his demagoguery to cozy up to him.  The prize being a coveted endorsement from an incumbent Iowan representative.  As we know, the success of an invisible primary candidacy can be predicted rather well by the number of endorsements they accrue in the pre-primary process.  What struck me as particularly odd about this case was that King never gave anyone the nod, despite the high profile hobnobbing.  Which begged a question of "why?"  As I puzzled over his decision not to endorse, I began thinking in broader terms: what factors out there can we associate with this decision?  What is it that drives this important behavior?  Luckily, the clever minds behind The Party Decides have a nifty data base of endorsements that can give a digital ream or two of information necessary to model this decision.  With the help of an undergraduate research monkey, I began to append some vital bits of data to every Representative and Senator in the 2000 presidential election cycle to build a logit model or two.

Though it's taken some time, thought, and rethinking, I'm beginning to see some results.  For those of you who wince at the sight of Stata outputs and the like, kindly avert your eyes and skip down the post a ways.  I modeled each party separately, as I reckoned there may be some differences between them.  The key independent variables are (in order presented): what week the presidential primary/caucus is in the MC's home state, state level presidential margin in 2000, the same margin squared, the electoral margin in the MC's last election, that margin squared, David Mayhew's state party culture (regular organization as reference), DW-NOMINATE dimensions 1 and 2 (common space, as I intend to go time series soon), years in office, age, and age squared.  The d.v. is a simple 0,1; no endorsement, endorsement.  Please forgive the cumbersome variable names.  Though they probably make no damn sense to you, they do to me, and that's what counts at this stage! 

Here are the Democrats:


And the Republicans:


As you can see there aren't any earth shattering results, but there are some interesting contrasts.  My suspicion that Democrats and Republicans approach the endorsement decision differently is seemingly supported.  Briefly, Republicans seem more sensitive to state level electoral factors (as expressed in an interesting quadratic relationship to the presidential election margin in their home state) and Democrats to the electoral calendar and the enigmatic 2nd dimension of DW-NOMINATE.   Some noteworthy, but non-statistically significant associations are found with Republicans and ideology (both dimensions); and for Democrats, the years served in office and a quadratic effect of age.  Both parties were sensitive to party culture, though in slightly different ways.  And if you're wondering, I did suspect the possibility of non-linear relationships in each of these variables before I even opened Stata on this project.  

So what does all this mean?  Apart from the little story told above that I'll be hashing through and thickening in the next few weeks, I'm not 100% sure at this stage.  Clearly Democratic and Republican MCs see the decision to endorse, at least in 2000, in quite different terms, some of which may speak towards the invisible primary process.  However, as the results suggest, this model is far from complete.  What could be missing from this mix of variables?  I'm also a bit perplexed by the direction of the relationship between 2nd dimension of DW-NOMINATE and the d.v.  As Keith Poole has written on several occasions, the 2nd dimension might be reflecting an insider-outsider dimension within the parties, as the more mavericky/Tea Party types tend to score on the negative side of things.  This was why I included the 2nd dimension in the first instance.  Presumably those on the inside would be more likely to engage in this critical party conversation than outsiders looking in.  While it might not be the perfect measure of this inside-out split, it's a convenient proxy.  Yet for some reason, that initial sense I had appears to be back asswards.  Higher 2nd dimension scores (the more "inside" an MC is), are associated with a lower likelihood of endorsing amongst Democrats, and possibly Republicans as well.  I suspect that it's acting as a proxy for a yet-to-be-accounted-for variable.

So in the spirit of our age and latest trends of the interwebs, I've decided to try my hand in a bit of crowd sourcing on these two key questions:

1) Are there any obvious omissions that come to your mind?  I'm all for clean, short, and sweet models, but glaring omissions need to be attended to...

2) What's with the 2nd dimension of DW-NOMINATE here?  Assuming this is a good measurement of "inside-outside" status, I suppose one could hypothesize that "outsiders" due to their outsider position, might feel more compelled to interject into the party conversation with an endorsement than a real insider.  Frankly, I'm not too happy with that line of reasoning.  Something is clearly going on here, but I'm not convinced I'm on to it just yet.

So crowd, get to it.  Any productive discussion is warmly welcomed and appreciated.  Moreover, I hope to see many of you in Seattle!

******
Day After Addendum: a technical note you should be aware of is that the dependent variable is an endorsement during the pre-primary period, prior to the Iowa caucuses.  As you may know, most every MC eventually makes an endorsement at some point, but often it's after the nomination's all but sewn up.  This is especially true of Democrats whose role as super delegates in the convention gives them a slightly different incentive structure in the endorsement decision.  One of a few reasons I model the parties separately.