A few days ago, I wrote a post describing the frequency distribution of bill cosponsorship in the Senate and House of Representatives for the 111th Congress so far. More particularly, I considered how many bills had gained how many cosponsors, then listed the five most cosponsored bills in each house of Congress. Today, let’s consider a separate but related question: who cosponsors more bills in the House and Senate? After we consider the question of who, we’ll begin to consider the question of why.

The following charts display the frequency of cosponsorship in the 111th Congress (as of February 18, 2009). Each dot represents one Senator or Representative, and the y-axis represents the number of substantive bills (H.R. or H.J.Res. in the House, S. or S.J.Res. in the Senate) cosponsored by a member of Congress.

The five members of the House of Representatives who have cosponsored the most bills are:

1. Raul Grijalva: 92 bills
2. Dan Burton: 83 bills
3. Maurice Hinchey: 81 bills
4. Bob Filner: 67 bills
5. Janice Schakowsky: 66 bills

The five members of the U.S. Senate who have cosponsored the most bills are:

1. John Kerry: 43 bills
2. Dick Durbin: 41 bills
3. Chuck Schumer: 40 bills
4. Barbara Boxer: 38 bills
5. Debbie Stabenow: 38 bills

If you’re familiar with the personalities of the Congress, you’ll probably have noticed that 9 out of these 10 top cosponsors in the House and Senate are members of the Democratic party — the one exception being Rep. Dan Burton of Indiana. Is it fair to say that Democrats are heavier cosponsors than Republicans in the 111th Congress? There’s a fair intuitive justification for that statement: the majority party controls the agenda more strongly than the minority party and is more likely to shepherd successful legislation through a house of Congress. Another intuitive justification for more cosponsorship among Democrats to assert that Democrats believe more in the use of government to provide solutions for social problems, and are therefore more likely to support bills that take government action.

The problem with intuition, though, is that it’s possible to think up intuitive accounts for exactly the contrary: majority parties may not need to use bill cosponsorship to demonstrate support and try to get the attention of legislative leaders, while minority party members may have to resort to cosponsorship as a way of getting others on the bandwagon for a bill through a visible demonstration of support (like a social movement in the Congress, you might say). Another problem with making a quick conclusion about party differences from the top 10 congressional cosponsors is that there are 525 other members of Congress sitting below these high fliers. Is the pattern less pronounced or even different for other members?

To answer this question, I’ve run bivariate and multivariate regression analyses to predict levels of cosponsorship activity. Regression analysis is, to put highfalutin language aside, the discovery of a “best fit” line that most accurately describes patterns in the data. Just as with the y=mx+b formula for a line you learned in high school, the formula for describing our outcome (y) (here, the number of cosponsorships by a member of Congress) is predicted by an intercept (b) plus a series of slopes (m) multiplied by our predictors (x). As you learned in high school math, the intercept is the expected value of y when x=0, and as you also learned in high school math, the slope can be interpreted as the increase in y you get with a one-unit increase in x. The only difference is that in multivariate regression, there are multiple slopes, as many slopes as there are predictors.

Let’s start with a simple regression by describing the best-fit line for the equation:

Number of Cosponsorships by a member of Congress = m*(Democrat?)+b

“Democrat?” is a variable that equals 1 if the member is a Democrat, and 0 otherwise. Here are the best-fit results for the data for the House and Senate:

House of Representatives (number of observations: 433)
Intercept b: 22.29
Slope m for “Democrat?”: -3.43 *
R-squared:: 0.02

Senate (number of observations: 99)
Intercept b: 12.40
Slope m for “Democrat?”: 4.00 *
R-squared:: 0.04

(* slope is statistically significant with a p-value of .05 or less)

The “statistical significance” of these best-fit lines’ slopes means they aren’t so small that they likely appeared by chance. There appear to be meaningful differences between Democrats and others in bill cosponsorship. But the R-squared values should give us pause. R-squared tells us how closely the actual data clusters around the best-fit line. Is the best-fit line a really good fit to the data, or is the best only a very rough approximation? Multiplied by 100, R-squared gives us the percent of the variation in bill cosponsorship explained by political party. For the Senate, that is .04*100=4%, and for the House, that is .02*100=2%. We shouldn’t expect any one variable to offer a 100% explanation when it comes to human behavior, but 2-4% explanations are really pretty dismal. Political party doesn’t predict the amount of cosponsorship activity to a great extent.

What’s more, if we interpret the values of the slopes substantively, the case for party as a driver of cosponsorship activity gets even more dismal. Because the variable “Democrat?” is set to 1 for Democrats and 0 for non-Democrats, and because its impact on cosponsorship in a y=mx+b is equal to the slope(m) times the variable(x), we can easily figure out the predicted number of cosponsorships for Democrats and Republicans:

Number of Cosponsorships by a Democratic member of the House = (-3.43*1)+22.29 = 18.86
Number of Cosponsorships by a non-Democratic member of the House = (-3.43*0)+22.29 = 22.29

Number of Cosponsorships by a Democratic member of the Senate = (4.00*1)+12.40 = 16.40
Number of Cosponsorships by a non-Democratic member of the Senate = (4.00*0)+12.40 = 12.40

In the Senate, at least Democrats are predicted by the best-fit line to cosponsor more bills. But according to the best-fit line for the House, Democrats cosponsor fewer bills! Simple partisanship isn’t doing well as a predictor of the level of cosponsorship activity in the Congress.

It’s time to consider some other variables. What if we follow one line of intuition mentioned above? If we think that members of Congress might cosponsor bills more because they are liberals who presumably believe more in government as a solution to social problems, why bother measuring liberalism as political party? Why not measure it more directly? In our House and Senate indices, we track members of Congress and measure the percentage of a slate of liberal policies they support, either through cosponsorship or roll-call voting through our Progressive Action Index. We measure the percentage of a slate of illiberal policies supported by members of Congress, too, through our Regressive Action Index.

Following the conventional intuition when it comes to congressional politics (liberal=more government action, conservative=less government action), we might expect higher scores on the Progressive Action Index to be associated with more cosponsorship, and higher scores on the Regressive Action Index to be associated with less cosponsorship.

Can you spot a practical problem with this hypothesis? People who have higher index scores get them either by voting a certain way… or by cosponsoring bills. Definitionally, then people with higher index scores must have cosponsored some bills! This trivial association is avoided by not counting the bills in our indices when calculating the number of bills cosponsored.

New Number of Bills Cosponsored Measure, Adjusted = (Total Number of Bills Cosponsored – Number of Index Bills Cosponsored.)

Here are the regression results predicting number of cosponsorships (adjusted) when we add index scores to political party:

House of Representatives (number of observations: 433)
Intercept b: -4.59
Slope m1 for “Democrat?”: -1.41
Slope m2 for Progressive Action Score: 0.81 *
Slope m3 for Regressive Action Score: 0.65 *
R-squared:: 0.31

Senate (number of observations: 99)
Intercept b: -4.03
Slope m1 for “Democrat?”: 1.29
Slope m2 for Progressive Action Score: 0.40 *
Slope m3 for Regressive Action Score: 0.26 *
R-squared: 0.26

(* slope is statistically significant with a p-value of .05 or less)

When terms measuring the extent of liberally- and conservatively-directed action in the Congress are included, the effect of political party on the number of bills cosponsored becomes statistically insignificant. That is, the slope of the effect of party becomes so small that the chances are pretty good their value showed up by chance. Again, party doesn’t seem to matter. The values of our action scores do, on the other hand, seem to be good predictors of cosponsorship activity. Not only are the slopes for the action scores statistically significant, but their inclusion also really kicks up R-squared, the measure of the goodness of a line’s fit to the actual data. In the House, the new model accounts for 31% of the variation in the amount of cosponsorship, and in the House, the new model accounts for 26% of the variation in cosponsorship activity. That’s quite an improvement.

But… look at the value of the slopes. We shouldn’t be worried that the numerical magnitude of the slopes is small: the units of these variables measure from 0 to 100, so that Senator Barbara Boxer‘s Progressive Action Score of 100 contributes +40 (100*.40) to her predicted number of cosponsorships. But then consider Senator David Vitter, whose Regressive Action Score contributes +26 (100*.26) to his predicted number of cosponsorships. The effect for Senator Boxer is in the direction we predicted, but the effect for Senator Vitter is opposite that predicted based on the liberal=big government, conservative=small government model.

What’s going on here? My hunch is that what both index scores measure, besides ideology, is activism, the tendency to engage in legislative action. Both high liberal scores and high conservative scores measure strong activist tendencies, while low liberal scores and low conservative scores measure legislative lassitude.

Do our indices not just uncover the direction of ideas but also identify the players versus the benchwarmers? Or is something else going on here? What other variables do you think might explain the variation in level of cosponsorship activity in the House and Senate? I’d love to read your thoughts.