Someone’s got it backwards. Can you hear the actual conversation at Romney headquarters?
“He wants to be deployed.”
“No. No.”
“I know, but look, he gave 7 figures, and you know what he’ll do if we ignore him.”
“OK, OK, OK. Crap. OK. We could send him to, like Mississippi or something.”
“Right. Here’s the phone. He’s on line 2.”
This morning, I fired up the Twitter engine to find this gem:
“Research findings correlate an attitude of gratitude with higher wages, stronger immune system, more friends, fewer addictions, better sleep”
The perky tweeter means, I think to help us greet the new day with the suggestion that if we all would just do the work of cultivating a feeling of gratefulness, all our wages would go up, we would get sick less often, people would like us more, we would shake the monkeys off our backs, and sleep well at night.
Who knew that’s all that required to be successful in life?
There is an alternative explanation, though: It might be that people who get better sleep than average, aren’t addicts, have many friends, are healthy, and are rich have more to feel grateful for. I think they’re planning on having a couple of grad students run the numbers on this unlikely interpretation, just to eliminate the possibility.
Google Correlate is a new research tool used to find out where in the United States the entry of certain search terms is concentrated, and which search terms’ levels of popularity are correlated with one another across the 50 states. The first idea is pretty straightforward: where are people more or less likely to be entering a search term? In other words, where are the cultural ideas of America located? The second idea is less intuitive: which terms are searched for more often in the same sets of states, and less often in the same sets of states too? In other words, what ideas are culturally aligned in the same places?
For a little while now, I’ve been interested in the contentions (to use a charitable term), urban myths and legends (to use a descriptive term) and hoaxes (to use a judgmental term) surrounding President Barack Obama. Setting aside the content of his actual policies for a moment, his presidency has provoked a number of emotionally-charged reactions, each of which is centered around a story about the covert actions of President Obama or his alleged fellow travelers. These have spread widely around the internet and persist despite fact checking. I decided to use Google Correlate to answer two questions:
1. Is it possible to use information about geographic clustering to find new examples of urban legends and hoaxes I’ve never heard of before?
2. Are most of the major urban legends and hoaxes regarding Barack Obama prevalent in the same set of states, or are they divided somehow?
To answer these questions, I generated a snowball sample starting with a particular political hoax message and spreading out to include others. The hoax I started with is a popular but false claim that a bill called HB 1388 has just been passed and signed into law by Barack Obama to give Hamas 20 billion and collect Hamas militants for resettlement in communities of the United States. Messages spreading the untrue claim are now in their third year of making the rounds and show no signs of dying out. Using the Google Correlate service, I generated a list of other search terms which are popular in the same states that the search “HB 1388″ are popular (and which are unpopular in the same states that the search “HB 1388″ is unpopular), using a correlation of at least +0.85 as a cutoff for inclusion on the list. From that list I identified search terms associated with other political hoaxes and urban legends, and using those terms made new lists of search terms fitting similar geographic profiles. I continued the process until there were no new hoaxes or urban legends listed, generating a set of 11 hoaxes and urban legends.
The following sociogram shows which hoax/urban legend search terms were generated from the snowball sample starting with “HB 1388,” and which pairs of terms are strongly correlated in search strength on a state-by-state basis:
The included terms are:
Was Google Correlate a good way to find out about hoaxes and urban legends related to Barack Obama? Definitely. Of the 11 Obama hoaxes or urban legends Google Correlate pointed me to, I’d never heard a word about 8.
Are the Obama hoaxes and urban legends popular in similar sets of states? The set of 11 identified here appear to be. Between these 11 hoaxes and urban legends there are 55 relationships, 55 opportunities for a correlation of at least +0.85 between the two to appear. In 32 out of the 55 pairs, a correlation this strong did appear. In 8 out of the 55 pairs, the correlation reached a level of at least +0.90. What are these states? You can find out for yourself by Google Correlate query, or you can take my word for it: they’re the same states whose electoral votes went to John McCain in the 2008 election.
But the appearance of this dense cultural web of hoaxes and myths, all concentrated in the McCain states, is at least partially rigged, an product of the snowball sample itself. After all, the hoaxes and myths were found because searches for them across the 50 states occurred with a pattern highly correlated to the pattern of “HB 1388″ searches; it shouldn’t be too surprising that the geographic pattern of the hoaxes and myths are correlated with one another, too.
The bigger question is, are there hoaxes and urban legends surrounding Barack Obama that weren’t dredged up by Google Correlate, ones that aren’t popular or unpopular in the same sets of states? One urban legend is conspicuous in its absence: the claim that Barack Obama is somehow secretly a Muslim. The geographic distribution of that claim isn’t strongly correlated with searches for “HB 1388″ or any of the other 10 hoaxes described above. Google searches asking “is Obama a Muslim” are concentrated in the Bible Belt, and are correlated with a different series of myths and legends that are religious in theme: false claims that Barack Obama had canceled the National Day of Prayer, questions regarding Obama’s identity as the Anti-Christ, searches for signs of The End and people looking for descriptions of dinosaurs in the Bible. As the sociogram below shows, these are correlated with one another to a significant extent; none of them are strongly correlated with searches for “HB 1388″:
Are there other hoaxes, urban legends or myths regarding Barack Obama that you know of but that don’t appear on either of these two lists, that could be associated with another geographic cluster of anti-Obama culture?
To date, Congressman John Murtha of Pennsylvania has been a central congressional figure in the FBI’s investigative hairs this year: Murtha is suspected of doling out military contracts in the form of earmarks to the clients of lobbyist Paul Magliocchetti, in exchange for campaign contributions to Murtha from Magliocchetti, his family, his employees and his corporate clients.
As the noose closes tighter around the necks of Paul Magliocchetti and John Murtha, you can expect the following excuse to be trotted out: “oh dear, these contributions to John Murtha were not bribes! No, no, Rep. Murtha simply has shown strong interest in issues related to defense and the space community, and we military contractors want him re-elected for that pre-existing interest. That’s the ticket!”
What would John Murtha’s legislative activity look like if this were true? To answer this question, we need to recognize two facts. First, John Murtha is not alone in receiving Magliocchetti money. Second, Magliocchetti’s clients are corporations that build weapons systems and work on other military projects. This means that we should look at John Murtha’s legislative behavior on military-related bills and compare it to the legislative behavior of other members of congress regarding those same bills. Is receipt of Magliocchetti money associated with similarity in patterns of support for military-related bills?
Possibility One: If Magliocchetti money is being doled out simply as bribery to take some particular action, then the recipients of those bribes don’t necessarily have to have any similarity in their positions on military issues in general. They’re just doing as they’re told on specific bills, then going their own way when it comes to other bills. We shouldn’t expect that other recipients of Magliocchetti money are especially similar to John Murtha in their positions on military-related bills overall.
Possibility Two: If Magliocchetti money isn’t a bribe, but really just support for the sort of member of Congress who supports the military issues Magliocchetti and his clients care about, then by gum we should expect to see some kind of similarity within the members of Congress who have received Magliocchetti money. We should expect that other recipients of Magliocchetti money have greater similarity to John Murtha in their positions on military-related bills than members of Congress who didn’t receive Magliocchetti money.
Which possibility is reflected in reality? To assess these possibilities, I’ve identified 146 bills in the House of Representatives this year that are military-related in their subject matter. Then I’ve paired up John Murtha with every other member of the House of Representatives (431 pairings, since there are 432 members of the House who are currently active). For each pair of some Representative and John Murtha, I’ve calculated a correlation to measure the extent to which John Murtha and that colleague behave the same way on each of these military-related bills. I measure correlation with Murtha because he is the supposed congressional ringleader of all this corruption, the alleged big stinky cheese. Murtha is certainly a big recipient of Magliocchetti money: $72,800 from Magliocchetti and his family members, not even counting employee or client contributions.
In particular, I study cosponsorship of these bills: the act of formally signing one’s name in support. When John Murtha cosponsors one of these bills, does his colleague also? When John Murtha refrains from cosponsorship, does his colleague also? If this happens all the time, we’d see a correlation between the two of +1.00. If, on the other hand, a colleague of Murtha’s always does the opposite of Murtha (cosponsoring bills Murtha doesn’t and failing to cosponsor bills Murtha does cosponsor), then we’d see a correlation between the two of -1.00. The actual range of correlations between John Murtha and his 431 colleagues ranges from -0.06 (a mildly oppositional relationship to Rep. Tammy Baldwin on military issues) to 0.70 (cooperative relationships at a level tied for first with Rep. Kathy Dahlkemper, Rep. Thomas Petri and Rep. John Shimkus).
Just these results tell you something interesting: the person to whom the Democrat John Murtha stands most opposed on military issues is also a Democrat, and two out of the three politicians with whom John Murtha is most coordinated are Republicans. Patterns of coordination with Murtha on military-related bills are not necessarily partisan.
Among members of the House, is coordination with John Murtha on military bills highest among those who have received the most Magliocchetti money? To find out, I’ve used the FEC to gather information on the clearly coordinated campaign contributions of Paul Magliocchetti and 9 members of his family. Here’s a plot of Magliocchetti family donations to a Representative (the y-axis) against that Representative’s correlation with Murtha on military bills:
A best fit line drawn on the above this scatterplot has a positive slope, indicating that overall those members of Congress with stronger correlations with Murtha on military bills also received more money from the Magliocchetti family. But as you can see for yourself, most observations (each Representative is a blue dot) fall off the line, and a number fall very far from it. Indeed, the nine members of Congress whose pattern of support on legislation most strongly correlates with John Murtha’s received no money from the Magliocchetti family at all, and the biggest recipients of Magliocchetti cash are centered right around the mean value of correlation of behavior with John Murtha (0.23). The R-squared statistic, multiplied by 100, tells us how much of the variation in Representatives’ correlation with Murtha is explained by variation in Magliocchetti money. Magliocchetti money explains just 0.4% of the variation in correlation with Murtha. That’s measly; it could very well be random.
A multiple regression analysis to predict correlation with Murtha on military-related bills, adding Representatives’ party, gender and Progressive Action Score as control variables, fails to alter this result. Possibility Two doesn’t look so hot.
Cosponsorship, the act of registering one’s name as an official supporter of a bill, is a common activity in the United States Senate. Although the 111th Congress is less than two months old, 1753 cosponsorships for 495 Senate Bills and 11 Senate Joint Resolutions have been officially registered. That’s an average of 3 cosponsorships per bill, which may not sound like much for a body with 99 members, but there’s actually a fair amount of variation from that mean, with 180 bills having no cosponsors and one bill (the successfully passed Lilly Ledbetter Fair Pay Act) gaining 54 cosponsors.
Cosponsorship is an important behavior not only because it is associated with the likelihood of a bill’s passage out of committee and on to floor consideration (see Wilson and Young 1997: Legislative Studies Quarterly 12:25-43), but also because it can give us a glimpse into patterns of support for a wide variety of bills. If we only reference roll call votes to characterize patterns of legislative support, then we’re getting a biased picture, missing a majority of the activity that goes on in the Senate. Most bills never get a roll call vote; by studying cosponsorship we can get detailed information regarding patterns of cooperation on issues small and large, popular and unpopular.
At That’s My Congress we’ve been looking at which members of Congress have been more or less active in cosponsorship on an individual basis, but cosponsorship is more than an individual act. It is at least potentially part of a social bandwagon effort to build congressional coalitions for a bill. Advocacy groups certainly speak of cosponsorship this way in their legislative activism. But even if cosponsorship were to be a solitary act based on independent rational calculus, it would be relational in the sense that the similarity of patterns of cosponsorship between two members of Congress indicates the similarity of their policy priorities. Cosponsorship networks — whether they model social ties, policy similarity, or both — are an important object of study.
Perhaps the most intuitive way to measure ties between two Senators in a cosponsorship network is as the number of bills which both Senators have measured. Below is a matrix showing the cosponsorship network of Alaska Senator Mark Begich, California Senator Barbara Boxer, Delaware Senator Ted Kaufman and Arizona Senator Jon Kyl. Data is valid for all Senate bills with a prefix of S. or S.J. Res. (leaving out the procedural Senate Resolutions and Concurrent Resolutions) through March 1, 2009:
| # bills both cosponsor | Begich | Boxer | Kaufman | Kyl |
| Begich | 26 | 16 | 1 | 0 |
| Boxer | 16 | 42 | 1 | 1 |
| Kaufman | 1 | 1 | 5 | 0 |
| Kyl | 0 | 1 | 0 | 15 |
Across any row, the number of shared cosponsorships is a good indicator of the behavioral similarity of various pairs of Senators. The maximum value for any row is on the “diagonal,” the cell in which a Senator is compared to himself or herself. In the language of the specified relation, we could awkwardly say that Senator Mark Begich has cosponsored 26 bills that Senator Mark Begich has also cosponsored — or we could say in plain English that Senator Mark Begich has cosponsored 26 bills. In this set, Senator Begich’s behavior is most similar to Senator Boxer’s, with the two cosponsoring 16 of the same bills. Senators Begich and Kaufman have only cosponsored one bill together, while Senators Begich and Kyl have nothing in common.
But what happens if we want to make a comparisons between rows? We know, for instance, that the Senators Begich and Kaufman have cosponsored just one bill in common, and that Senators Boxer and Kyl have cosponsored just one bill in common as well. But looking down the diagonal we know that the overall levels of cosponsorship varies from Senator to Senator: 26 for Begich, 42 for Boxer, 5 bills for Kaufman and 15 bills for Kyl. You could argue that the shared cosponsorship of 1 bill between Begich and Kaufman is more meaningful for Kaufman than it the shared cosponsorship of 1 bill between Boxer and Kyl. For Barbara Boxer, that’s just one bill out of 42, but for Ted Kaufman it’s one bill out of five. Oddly enough, the very same number of shared bills between Senators Boxer and Begich — 16 — means something different from each Senator’s perspective. 62% of the bills Mark Begich has cosponsored (16/26) are also Boxer bills, while 38% of the bills Barbara Boxer has cosponsored are also Begich bills. The share of Boxer-type bills in Begich’s cosponsorship profile is lower than the share of Begich-type bills in Boxer’s cosponsorship profile… but the raw numbers don’t tell you that.
To create a standard measure that has the same meaning across all pairs of senators, we can use Pearson’s product moment correlation for each pair of Senators instead. To find Pearson’s correlation between two Senators X and Y here’s what we’ll do:
1. For every one of the 506 Senate bills introduced as of today i, ask the dichotomous questions, “Has Senator X cosponsored bill i?” and “Has Senator Y cosponsored bill i?” We’ll call the answer to the first question Xi and the answer to the second question Yi, and give each a value of 1 if the answer is “yes” and a value of 0 if the answer is “no.”
2. Why a 0-1 variable? One reason is that the mean of Xi for all 506 i Senate bills is equal to the proportion (or, when multiplied by 100, the percentage) of all of the bills that Senator X has cosponsored.
3. Now for each bill i, we’ll subtract the mean of Xi (for all bills) from Xi (this particular bill). That’s called the “deviation” from the mean, and the result is positive if cosponsorship happens and negative if cosponsorship doesn’t happen. We’ll do the same thing for Yi and the mean of Yi to obtain the deviation for Senator Y.
4. Then we’ll multiply the deviation for Senator X on bill i times the deviation for Senator Y on bill i. Why? Well, remember that a positive deviation occurs when cosponsorship of bill i has happened, and that a negative deviation means cosponsorship of bill i hasn’t happened. A positive deviation for X, multiplied by a positive deviation for Y, gives us a positive result. So when both Senator X and Senator Y cosponsor, we get a positive number. How about when both don’t cosponsor? Well, then we get a positive number too. But when Senator X cosponsors and Senator Y doesn’t (or when Senator Y cosponsors and Senator X doesn’t), we’ll multiply a positive number and a negative number, which always gives us a negative number. The end result of all this is a positive number if Senators X and Y do the same thing regarding bill i, and a negative number if they do different things regarding bill i.
5. We’ll do that same calculation over and over again, once for every one of the 506 Senate bills existing as of today. The result is that we’ll have 506 numbers for the 506 bills. Some of the bills will be positive (when Senators X and Y do the same thing), and some of them will be negative (when Senators X and Y do different things).
6. Now add all of those numbers up. If Senators X and Y tend to do the same thing regarding cosponsorship of bills more often than they differ, the sum of all those numbers will be positive. If Senators X and Y tend to diverge in their cosponsorship choice more often, the sum will be negative. If the sum is zero, then that tells you Senators X and Y differ as often as they do the same thing.
7. Here’s where I wave my hands a bit: dividing this result by the product of {the number of bills, the standard deviation of all Xi and the standard deviation of all Yi} standardizes the result so that the biggest possible positive value is 1 and the biggest possible negative value is -1. That’s so the value of the result can be interpreted the same way, no matter how many bills Senator X or Senator Y cosponsored overall.
8. Do this for all possible pairs of Senators, and you’ll have a number for every pair of Senators telling you whether the pair acts the same way regarding a bill more often than the pair acts divergently (a positive correlation) or whether the pair acts divergently more often than it acts the same way (a negative correlation).
So I did that. Rather, I had my computer do that, using a PHP program to access current reports on congressional bills via the online Thomas system and the awesome network analysis program called UCINET to run calculations.
Want to see the results? I could put them in a matrix, just as I did above, but then I’d have 99 rows and 99 columns for the 99 Senators currently seated, with 9,801 number-filled cells. How’d you like to interpret that? Blech. I sure wouldn’t, and besides, the table wouldn’t fit on your screen. Our eyes are simply better at perceiving patterns when they’re presented in the form of a picture, and our technologies are designed better to show such pictures. Pictures of relational data are called sociograms, and they work like this:
1. Every Senator is a dot.
2. Every relationship between two Senators (that meets some standard) is a line between the dots.
3. That’s it.
4. No, really, that’s it.
Let’s look at sociograms of those cosponsorship correlations between United States Senators for the 111th Congress so far. The ones you see below are generated using Netdraw.
There are two ways we can go with a sociogram. One is to look at a particular member of the Senate and to use the space of the sociogram to describe every other Senator’s relationship to that focal senator. Here’s just such a sociogram featuring Senator Mark Begich:
This form, the sociometric star, features shorter ties for higher correlations between Senators, since those indicate closer behavioral agreement. I’ve pulled aside the Senators who share absolutely no common cosponsorships with Begich. You’ll see they’re coded in blue and feature no line connecting them to Begich since they are wholly unconnected with Begich when it comes to supporting legislation in the Senate. Such unconnected individuals are called “isolates.”
The other sort of sociogram we can use draws a tie between two Senators when the strength of connection between them meets a certain threshold. In order to show relationships of strong correspondence between senators, correlations of positive 0.33 or greater merit a tie in the sociogram below:
One of the nice things about sociograms is that you can use graphic elements to display additional information. Here, I’ve given Republican senators the color red, Democratic senators the color blue, and Independent senators the color green. And finally, here at the end of our investigation, we can begin to address the assumptions behind those classic political colors. Most news reports on the Congress don’t go beyond such ideas of partisanship in describing the political coalitions in the Senate. Reporters will make blanket statements like “Republicans in the Senate are close to having no voice whatsoever” and references to “the growing strength of Democrats in the Senate”, but we don’t have to assume that the Senate is organized on a strictly partisan basis. Instead, without making any judgment about partisanship, we can look at the actual pattern of cooperation in cosponsorship and see what patterns naturally emerge.
In this sociogram, members of the Senate are made to appear close to one another not because they are members of the same political party, but because of strong cosponsorship bonds they share. Cliques of three, four, six, even ten senators are visible as well, groups within which all senators have a strong record of supporting the same sort of bills. These cliques do not encompass all members of a political party, but rather occur inside them. Senators Kay Hagan, Mark Warner, Ted Kaufman and James Webb form a strong Chesapeake-Carolina coalition of junior senators, but none of them have a strong tendency toward supporting the bills favored by a bigger clique of more senior Democratic senators including Barbara Boxer, John Kerry, Dick Durbin, Barbara Mikulski and Chuck Schumer (among many others). Conservative Senators Lindsey Graham, Johnny Isakson and Robert Bennett have a strong record of cooperation in their support of bills, but do not appear to be coordinating strongly with other conservatives like Sam Brownback or John Ensign.
That said, there’s a split in this cosponsorship network that is undeniable. All of the Republicans who have any strong correlations with other senators have those strong correlations only to other Republicans. All of the Democrats having any strong correlations in cosponsorship with other senators have those strong correlations only to other Democrats. That doesn’t look bipartisan, does it?
Well, looks may be deceiving. Yes, there are large groups of Senators whose support for bills isn’t strongly bipartisan. But remember that ties are reported only above a certain threshold of co-operation. There may be weak bipartisanship among many of these senators happening on an occasional basis. And look again at the sociogram. There are 24 senators who appear as “isolates” because they do not have a strong correlation in their pattern of bill cosponsorship with any other senator. Some of these senators, like Richard Shelby and Michael Bennet, appear as isolates because, frankly, they’re not doing much of anything but marking time and enjoying the Capitol Hill cafeteria.
But other “isolates” have earned low correlations with other senators not for their inactivity but for their high level of independent, eclectic activity. Senator John McCain, for instance, has not only introduced a number of his own bills but has cosponsored a number of bills written by and supported largely by Senate Democrats, Democrats as fiercely liberal as Russell Feingold. These cosponsorships have the effect of canceling out cosponsorships shared with fiercely conservative senators like Jim DeMint, giving John McCain a strong correlation with the cosponsorship record of no one else. That makes McCain, yes, a Maverick. It gives him weak behavioral ties to others. But these are, putting substantive judgment aside, bridging ties, ties connecting parts of the Senate in ways that the more insular ties of strictly liberal or conservative senators don’t.
There’s a lot to uncover in these cosponsorship networks, and a lot to explain as well. Look for more updates, information and analysis of congressional cosponsorship from Irregular Times and That’s My Congress.
