Influence bargaining modeling - a new tool for forecasting

One of the hardest parts about forecasting politics is keeping track of all the moving pieces.

Take, for example, what climate policy might emerge out of Congress next year. Any eventual legislation will be the product of negotiations and bargaining among interest groups, the House, the Senate, the White House, and all the different factions within them.

Even if we are to make assumptions about who wins the elections in November, that still leaves 435 Representatives, 100 Senators, multiple Cabinet members, a President, and dozens of key external stakeholders whose moves we need to forecast. Add in the fact that all of these players have different policy positions, different levels of influence, and care about climate change in different amounts, may interact with each other in different ways, and the task becomes incomprehensible.

At that point, accurate forecasting become a near-impossible task - but one that is still necessary for many organizations.

Two Lanterns has developed a new tool for this, an model that includes political influence in a bargaining model that can simulate a baseline outcome, a range of possibilities, and quickly adapt to show what would happen in different scenarios.

Game theory and clustering

Bruce Bueno de Mesquita is a political scientist who started to tackle this problem in the 1980s. His approach used game theory and stylized negotiations as a series of rounds in which all players talk to bargain into a mutually acceptable position. He wrote a book about it and was profiled in the New York Times about his approach to predictions.

His approach appears to have been useful, but it has its drawbacks, of course.

Bueno de Mesquita’s code is a black box, meaning that verification is difficult. It also assumes a theory of expected utility that may not bear out in real life. People can be irrational, they can be motivated by other things than the matters at hand, and institutional levers can be important but difficult to include in a formal model. My own PhD dissertation examined how the United States had influence in the European Union: it was not a full member, but it was able to ensure some American red lines were not crossed through informal negotiations and using shared membership in other institutions as a source of leverage.

This is not to criticize Bueno de Mesquita’s work. From a political science perspective, his “selectorate theory” is a genuinely important concept that fully merits tenure at any institution, and his prediction work to view ideological disputes as game theoretical pursuits of utility help to make the future more understandable.

Luckily, there are other approaches that can generate some of the same outcomes with fewer computational costs.

In a paper from 2016, Martina Grabau of the University of Siegan and Simon Hegelich of the Technical University of Munich developed a clustering approach to solve the same forecasting problem (and, in fact, compared their eventual model with Bueno de Mesquita’s). In this method, they viewed a negotiation as a series of coalitions being formed. Rather than Bueno de Mesquita’s game theory approach of simulating an interaction between every paid of actors (which run into the millions of simulations for less than a dozen players), they argue that the most similar players will agree on a common position, then go to the next-most similar player to grow their coalition by a bit more, and so on. Eventually everyone has been included in this coalition, and the final position is the result of the negotiation.

This approach offers a couple of key advantages.

First, it resembles how politics actually works. In any political environment, not everyone views everyone else equally. Legislators are much more likely to talk with their close allies first and figure out what they should do, before talking with leadership, before talking with the other party. Treating deal-making as comprised of lots of mini-deals (but not too many) is a simple way to view politics that is roughly representative of how the real world works.

Second, a coalition-building approach allows us to inject extra variables that help nudge the coalitions towards a more realistic picture. Arizona Senator Kyrsten Synema might hold economic views closer to Susan Collins’ than Elizabeth Warren’s (at least if DW-Nominate is to be believed), but she and Warren vote the same way on most bills, given that they’re both Democrats. Their position on an issue is a less important predictor on how they’ll vote than their partisanship.

Influence Bargaining Model

To harness this approach, Two Lanterns has built what we’re calling an “influence bargaining model.”

It captures what we think are the most important variables for a policy issue and simulates how we think the eventual negotiation between parties will play out. In the below example, we model how a negotiation on climate policy will play out in the next Congress, assuming that Joe Biden is the President and the Democratic Party has a 1-seat majority in the Senate.

Dendrogram of the coalition-forming process in climate negotiations.

In this model, Speaker Pelosi and House Moderates immediately form a coalition based on their shared position (or in an alternate explanation, the House moderates agree to go along with Pelosi’s position). They are quickly joined by Senate moderates, and that coalition then joins with President Biden and Senate Majority Leader Schumer, before negotiating with Progressives and then Blue Dogs in both chambers. Republicans are not brought into the process until the end (when we anticipate they mostly vote against the bill).

We can also use this model to project what we expected the result of the negotiations to be. On a spectrum from loosening fossil fuel regulations to full implementation of the Green New Deal with a carbon tax, we identify all the players’ preferred positions and then see how the bargaining moves them.

The players in Washington have a wide range of initial policy positions, but through negotiations they get moved to a final bill.

In the above chart, we see the coalition-building process moves all the players to varying amounts. Progressives (the line starting at 100) have to move significantly downwards through negotiating with Moderates, Leadership, and the White House, but they are still effective enough to pull the bill closer towards the Green New Deal end. In fact, they are effectively cancelling out the Blue Dogs in the House and Senate, which is a considerable feat given how we expect West Virginia Senator Joe Manchin to be the swing vote. Republicans have little effect on the overall bill and their inclusion in the final “coalition” is more of a reflection of the policy they have to live with than their voting for the legislation.

Adding Monte Carlo to the model

One of the best parts of a computationally light approach is that we can easily replicate it a few thousand times.

A constant challenge in political forecasting is the fact that uncertainty exists. We can say that right now Joe Biden is at an 80 on the scale (supporting a large-scale green infrastructure plan), but what happens if his future Administration changes its mind. Or what happens if the Speaker has to turn her attention to economic recovery matters and lets progressives in the House drive initial negotiations.

By adding uncertainty ranges to each actor’s component factors, we can simulate how the future might unfold while reflecting that not all political actors are the same. Rep. Alexandria Ocasio-Cortez is much less likely to have her position on climate policy change than a moderate House member, who would be more likely to adopt a different position between now and next January.

This allows us to see not only the baseline expectation, as above, but to see what might be the varying outcomes of the model.

Histogram of possible climate results in a Democratic-held Senate, ranging from nothing more than rejoining the Paris Agreement to a massive infrastructure plan, 5,000 runs.

Building Scenarios

This approach allows the user to quickly adjust the inputs to show what would happen in different scenarios. Below is the Monte Carlo simulations for a Democratic Senate majority of one, a majority of three, and continued Republican control.

Histogram of policy outcomes, 5,000 simulations for each scenario.

While it is easy to say that more Democratic control would lead to more pro-climate outcomes, a lot can happen in politics. We see that the bell curve shifts in that direction, but there are still a large percentage of outcomes that are lower than what could happen in a smaller majority. Similarly, even if Republicans control the Senate, this is a good chance that something like major subsidies for renewable energy (a 40 on the above scale) happens.

Organizational utility

This is, of course, not to say that this tool will give you a definitive account of what will happen in politics. Unfortunately, no such tool exists.

Instead, what this does is to help structure a huge amount of nebulous information and makes it productive. To create the above simulations, I needed to track the preferred policy positions of the major groupings in Washington, as well as how much influence they have, how much climate policy matters to them, and how important their partisan and institutional affiliations will shape negotiations.

If the model did not exist, there would be value in having that data in one place. Even more, there is value for an organization to sit down and debate what those metrics should be. It is a way of structuring internal debates (similar to matrix games) without turning it into rambling discussions. The model offers an easy excuse to share information and build internal awareness.

This model also can help turn the advice of subject-matter experts into usable products. I’m sure many of us have received a report from an expert that goes into meticulous detail about every aspect of a situation, but forgets to include their judgement of the overall picture. That report can be very useful; it can also sit in an inbox unread by those who need their judgement. This model helps distill that expertise and blend different sources of knowledge together.

Further, this model can offer a testing ground for those who seek to influence the outcome, rather than just predict it. Let’s say you’re a climate activist who want to see a carbon tax implemented. Take this model and see what is required to get the result to that point. Change the inputs to include another player, like a climate advocacy group. How much power can they truly have and shift the result? What if you change the initial position of a player? Or how much they care about climate? Then ask yourself what it would require to realistically achieve those ends within your existing constraints. You now have the first step towards your strategy.

Get in touch

This Influence Bargaining Model could be a useful approach to your political forecasting problems, or a valuable addition to an existing scenario planning exercise or policy monitor.

Let us know what policy you need and we can start the process of building a model that can help you get a view of what might happen.