Though the voting public focuses more on bills that are debated, passed, or defeated on the floor of Congress, government scholars are also interested in what bills do and do not make it to the floor, known as agenda control.

In “Bill Text and Agenda Control in the U.S. Congress” (forthcoming at The Journal of Politics), SPA Assistant Professor Andrew Ballard examines the concept of agenda control on and before the floor in the House and Senate, based on estimates of how members would have voted had the bills made it to a vote.

“The choices that are presented to people directly affect what actually happens,” said Ballard. “If you have two versions of a relatively similar bill, and one of them gets a vote and one of them doesn't, the differences between those bills might affect the votes of some of the members of Congress. The idea of agenda control is using various political and institutional strategies to control what gets on the agenda.”

This process is referred to as either negative agenda control (keeping items off the agenda) or positive agenda control (ensuring items make it onto the agenda).

“Sometimes, this happens almost by default,” said Ballard. “There are lots of bills that members of Congress or members of other legislatures introduce [for messaging purposes] that are dead on arrival.”

In other instances, agenda control is exercised strategically by gatekeepers at each level, such as the Speaker of the House. “There are lots of examples where one party, subgroup, or bipartisan coalition wants to do something,” said Ballard, “but people in certain important positions don't think that would be good policymaking, or good for their party's brand or their own individual re-election chances. So they keep that from happening.”

The House can pass legislation with a simple majority, and the Speaker of the House has significant agenda control power, influencing the Rules Committee (which makes rules for debate on the House floor) and deciding which committees will take up the bill. The Senate, meanwhile, has fewer institutional rules enabling positive agenda control, though its higher pass threshold leads to strong negative agenda control.

“It is easier for a minority [or bipartisan] coalition to stop things from happening in the Senate. It is, all else equal, bad for the majority party to bring something to the floor and then have it defeated. So they're thinking ahead, thinking, well, do we have the 60 votes to bring this to the floor? And if they don't, that makes it relatively easier for a minority coalition to stop things from happening.”

To determine these preferences, Ballard analyzed the text of bills introduced in the House and Senate from 1994 (the earliest year in which digital versions are available) to 2016, using machine learning methods for Natural Language Processing to train computers to understand and predict language as well as humans. Using an algorithm called doc2Vec, he was able to parse through thousands of pages of bills relatively quickly, yielding values to be used as predictors in a regression model.

“We're actually getting to the point where if a human can classify and predict something then a computer can too,” Ballard continues. “A decent way to get a quantitative representation of this is with a set of machine-learning methods called vector embedding models, which runs some text . . . through what's called a neural network, which in its simplest form is a number of computational layers for algorithmically training some task. What spits out at the end fits with the way that we actually think about language.”

Using the set of bills that got a vote, Ballard trained another model to predict individual members’ preferences, and then applied that predictive model to those bills that did not make it to the floor.

“At least in the published literature, the model that I built is the most accurate to date in predicting votes in both the House and the Senate. So we could be relatively sure that, when applying it to bills that didn't actually get a vote, it would also be reasonably accurate.”

He found high degrees of negative agenda control in both chambers, but a much higher level of positive agenda control in the House.

“If the majority party in the House likes something, it's much more likely to reach the floor than if the majority party in the Senate likes something,” he said. “In both chambers, if the majority doesn't like something, it is really unlikely that's going to see the floor. But the House is better at pushing things through.”

“These implications are even stronger for divided government. [Under the previous administration], the House passed a lot of bills, but the Senate took up almost none of it. It's really hard for the Democrats to get their coalition together in the Senate, particularly with more moderate members who perceive their electoral fortune to be improved if they come out publicly against the party's more progressive stances.”

The process built upon Ballard’s academic goals of contributing both substantive insights into government and methodological innovations to the field of political science.

“This work has two purposes,” he said. “One is to learn something substantive about Congress, and the other is to advance the ways that we measure and quantify things. One of the roadblocks to testing substantive questions has been that we only see preferences expressed on bills that actually get a vote, a tiny percentage of the actual bills. In order to get a fuller picture, we would need to know something about how members feel about bills that didn't actually make it to the floor. I wanted to measure that better, and to provide a new measure to people to further research in this area.”

These findings help explain the frustrating reality of gridlock, which can contribute to losses in the president’s party in the mid-term elections. The checks and balances of the Constitution introduce institutional transactions costs that may appear to the American public to be Congressional apathy.

“When one party has control of both chambers and the White House, the American public expects policymaking to happen. But there's often a disconnect between these expectations and what is feasible,” said Ballard.

“It's not a very satisfying answer,” he added. “The American public doesn't like gridlock. But the reasons for it are not necessarily that politicians or elected representatives don't want to do things.”