As members of the House of Representatives begin returning to the Capitol over the coming months, it is essential that Congress do everything in its power to prevent the spread of COVID-19 among members and staff. Even as we start to return to a degree of normalcy, health experts warn that some level of social distancing will be required for at least another year.
House Republican Leader Kevin McCarthy suggested that Congress implement a “block” or “staggered” schedule to reduce interaction among members while they attend committee hearings. In a block schedule, committees are distributed among several groups; these groups are then assigned specific durations of time during which they can meet. By staggering committees throughout the day or week, we can reduce the frequency with which members interact.
BPC has long been a proponent of block scheduling and previously recommended a specific block schedule that would minimize the number of scheduling conflicts that members face. This research showed that implementing an optimized block schedule, informed by data science, could reduce scheduling conflicts by up to 86 percent.
Of course, the concern now is not so much conflict as it is close social interaction between members. Considering this novel dilemma, BPC applied similar optimization logic to approximate a distribution of committees among blocks that would best maintain social distancing. We found that an optimized block schedule could reduce direct interaction between members by 36 percent.
While promising, it is important to note that, if taken alone, this measure would not be effective. Rather, only in combination with other strict social distancing precautions, such as wearing personal protective equipment, would an optimized block schedule prove useful.
Using a greedy algorithm, we were able to identity a committee distribution that significantly reduces interactions between members of Congress. While we refer to the given assignment of committees as the “optimized solution,” it is important to note that this is only an approximation of the true optimum. However, after testing over 50,000 random block assignments, the distribution of outputs suggests that this is negligibly close to the global optimum.
The optimized distribution of committees below would naturally create social distancing with the potential to reduce the risk of direct member-to-member spread of the coronavirus by 36%.
|Block A||Block B||Block C|
|Committee on Energy and Commerce||Committee on Appropriations||Committee on Agriculture|
|Committee on Ways and Means||Committee on Financial Services||Committee on Armed Services|
|Permanent Select Committee on Intelligence||Committee on House Administration||Committee on Foreign Affairs|
|Committee on Ethics||Committee on Rules||Committee on Natural Resources|
|Joint Committee on Taxation||Committee on Science, Space, and Technology||Committee on Small Business|
|Joint Economic Committee||Committee on the Budget||Committee on the Judiciary|
|Committee on Oversight and Government Reform||Committee on Transportation and Infrastructure|
|Joint Committee on Printing||Committee on Veterans' Affairs|
|Joint Committee on the Library||Committee on Education and the Workforce|
|Select Committee on the Modernization of Congress||Committee on Homeland Security|
|Select Committee on the Climate Crisis|
If all members were present at the same time without a block schedule in place, there would be a total of 92,235 pairwise interactions.1 Dividing committees into the three blocks produced by the algorithm yielded a total of 58,903 pairwise interactions, a sizable decrease from the original 92,235.
An optimized schedule might be interesting on its own, but it is useless without proper implementation. To maximize the effectiveness of a block schedule, the House of Representatives should also do the following:
- Stagger scheduling blocks throughout the week, with each block meeting on a specific day. This would not only minimize the chances that members in different blocks interact unnecessarily, but it also allows time for committee rooms to be disinfected each day, thereby reducing the risk of spreading the virus between blocks. One option for how this might look in practice is holding floor activities on Monday and/or Friday and then reserving one weekday for each scheduling block:
|Floor Activities||Block A Committee Meetings||Block B Committee Meetings||Block C Committee Meetings||Floor Activities|
- Keep strict social distancing between members of Congress. In a separate model, BPC sought to identify the optimal distribution of committees when secondary and tertiary interactions between members were treated as equal in weight to primary interactions (see Figure 1 in Endnotes for more information). In this model, no committee distribution could be found which reduced risk. In other words, without social distancing to stop the spread of the virus from one potentially asymptomatic member to another, there is no stopping a ripple effect in which the virus ultimately spreads to all members. In the example below with only four members, we see that Member B is assigned to both Block One and Block Two. If we treat secondary interactions as carrying a risk of spreading the virus, then all members in that example could be infected by Member B’s presence, highlighting the importance of social distancing even within the same block.
The purpose of a block schedule is to make social distancing easier and more effective, but it is not intended to replace social distancing measures altogether. Strict social distancing, even between members assigned to the same block, will be necessary to keep the virus at bay.
BPC set out to minimize the frequency of direct, pairwise interactions between members of Congress. In our model, a pairwise interaction is defined as any time when two members of Congress are both present in a scheduling block. In the figure below, a single interaction between two members is represented through a blue arrow. Using a simple example in which there are only four members, we see that without a block schedule, there are six total blue arrows, or six total pairwise interactions.
However, breaking the members up into two blocks (One and Two) reduces direct interactions by one third (from six to four). To reflect the reality of multiple committee assignments per member of Congress, this example assumes that Member B has a committee assignment in both Block One and Block Two.
The model that we used sought to distribute committees among three scheduling blocks such that the total number of direct, pairwise interactions (blue arrows) would be minimized.
Controls were set in place such that if the same two members of Congress interacted in two different blocks, this would count as only a single interaction. This ensures that the model minimizes the number of new interactions, rather than focusing on interactions that already happened.
With three blocks and 27 committees, there are over 7.6 trillion different ways to assign committees to blocks (3^27). As such, testing every schedule alternative is hardly feasible. Instead, BPC used a modification of Kruskal’s algorithm, adjusting the classic graph-theoretic approach to accommodate the more difficult problem.
The algorithm starts by putting each committee in its own block. It then repeatedly calculates how many new pairwise interactions are created by combining any two blocks and merges the two that introduce the fewest new interactions. This repeats until there are only three blocks left. This approach allows us to approximate the optimal solution without trying all 7.6 trillion alternatives.
In a time of crisis, Congress needs all the help it can get to find a safe and sustainable way to operate. Where possible, Congress should rely on the tools of data science to inform their decision making. As BPC’s new research demonstrates, an optimized block schedule could sizably reduce the risk of spreading COVID-19 among members. It is Congress’ responsibility to keep its members safe, and if it implements a block schedule, we believe that one informed by optimization modeling is the best way forward.
The author would like to thank Spencer McCall, BA Computer Science and Economics, UC Berkeley, for research support.
1 Where n is the number of Congressmembers (430 with committee assignments) and r in the number in a pair (2):