I had the good fortune to catch a small part of a conference at Harvard yesterday on text analysis. Good fortune, because I was there long enough to hear Justin Grimmer‘s talk on his dissertation, Measuring Reputation Outside Congress. Grimmer is interested in an important – and tough to answer – question: how responsive are the people we elect to their constituents?
We could look for ways to answer this question by studying the voting record of legislators (qualitatively or quantitatively), examining their work in Washington (through Congressional literature) or through examining their communications with constituents at home. This latter set of questions is referred to as the “Home Style” of a politician, following the work of Richard Fenno (1978).
Home style tells us something about politicians that their voting record often doesn’t, Grimmer tells us. He invites us to compare Senators Jeff Sessions (R-Alabama) and Richard Shelby (also R-Alabama). If we consider them simply in terms of their voting behavior, they look nearly identical – they vote together the vast majority of the time and both can be described, in voting terms, as conservative republicans.
But anyone who knows Alabama politics will tell you that Sessions and Shelby are vastly diferent guys. Grimmer characterizes Session as “an intense policy guy” who will bore you to tears with incredibly long, thorough explanations of issues when all you wanted was a photo with him. Shelby, on the other hand, is all about bringing home the bacon… and there are Shelby Halls at two Alabama universities to prove it.
Evidence suggests that representational style – policy versus pork, heavy versus light communicators – cuts across party lines. And it’s likely that politicians have diverse, stable, nonpartisan home styles. If we can find ways to characterize these differences – Grimmer proposes studying the difference in communications with constituents that claim credit and those that discuss policy – we have the opportunity to compare across senators, and connect these differences to what senators do within the institutions of power.
When Fenno studied the “home style” of politicians in 1978, he engaged in “soaking and poking” – intense participant observation, which involved following 18 members of Congress over 8 years. This method, Grimmer observes, is expensive, underrepresentative (and really hard to replicate as a graduate student.) Instead, we might study texts produced by senators. One candidate is newspaper articles… but editorial bias makes it hard to use editorials as representative of senatorial communications. We might use the constituent newsletters produced by Senate offices… but they’re sent using the constitutional Franking privleges and are very hard to get hold of.
Instead, Grimmer has been studying the press releases that senate offices produce – over 64,000 in all. The average senator issues 212 press releases per year, and while the quantity produced has a wide range (some produce only a few dozen, while Hillary Clinton’s senate office produced over a thousand a year), there’s no strong correlation between political party and usage of the tool.
After collecting the releases, Grimmer used machine learning techniques to separate transcripts of floor statements (which are usually released as press releases) from pure press releases, which let him study how a senator chooses to speak to her constituents. Once that sorting has taken place, the task is pretty simple – determine the topic of a press release. This is simplified by the fact that congressional aides try hard to ensure that press releases are on a single topic.
Grimmer’s work clusters senators by the topics discussed in their press releases. His research reveals four basic clusters:
– Senate Statespersons. These folks speak like they’re running for president… and they may well be. Their releases discuss the Iraq war, intelligence issues, international relations and budget issues. John McCain’s office communicates this way.
– Domestic policy. These senators are also policy wonks, but their focus is domestic – the environment, gas prices, DHS, and consumer safety.
– Pork and policy – Communication from these senators includes discussions of water rights grants, but also has serious discussion of health and education policy. Sometimes this is because the office simply issues lots and lots of releases – (former) Senator Clinton’s office fits in this camp.
– Appropriators – These guys communicate about the grants they’ve won – fire grants, airport grants, money for universities, and for police departments.
As well as clustering press releases based on topic, Grimmer’s work considers another metric – how often a press release claims credit for an appropriation. There turns out to be a vast spectrum, ranging from John McCain, who basically only issues statements about policy, and a guy like Mike DeWine, an Ohio Republican, where virtually every press release claims credit for an appropriation. There’s a very strong correlation between the topic clusters in releases and the percentage of releases claiming credit. (That’s at least in part because claiming credit is one of the topic clusters – you’re correlating between, in part, the same factor. Interesting nevertheless.)
What’s most interesting is that this classification – either by type of politician or by their place on the credit spectrum – is tightly correlated to their voting behavior on a particular issue: votes on appropriations rules, or as Grimmer puts it, “How do legislators self-regulate the porkbarrel”. These votes aren’t partisan – the late Ted Kennedy voted with Richard Shelby on these sorts of votes, which suggests truth to the truism that there are three parties in Congress: Democrats, Republicans and Appropriators. In other words, the way a Senator communicates with constituents is strongly predictive of their legislative behavior, specifically on how they allocate funds.
I thought this was excellent stuff – I hadn’t seen someone take a large database of political communications and subject it to automated analysis, and I thought the demonstration of this “third party” was particularly compelling.