Polarization under rising inequality and economic decline: with an interview by Mohammed El-Said

It's the end of semester, and the paper I was working on when Aylin Caliskan came to talk to me about working on my semantics project is finally published. Polarization under rising inequality and economic decline, with Alex Stewart and Nolan McCarty, out now in Science Advances. Here's a short news story on it by Hertie School, who also recorded a bunch of videos of me talking about the paper, two of which are linked there.

decent press shot of me explaining stuff
Gratuitous crop from one of the videos, so the blogpost about our paper serves a picture of me instead of Mohammed. Sorry Mohammed!

I wanted to blog about our new paper, but I was running out of things to say after talking to the press office so much. But then I got interviewed about some new things, so with permission I'm sharing that interview here, which otherwise will only be available in Arabic. 

Questions are by Mohammed El-Said, the science correspondent of the Daily News Egypt, and contributing science writer for Scientific American Arabic Edition, Sapiens, Al Araby Al Jadeed and
AGU's Eos

Would you please summarise for us the conclusions of the study? and what big question is being addressed in the paper?

It's important to understand that this is mostly a theory paper, rather than a data paper. Science advances by producing theories, testing them with data, and then improving the theories.

In this case, there was already data showing that political polarisation AND identity politics both tend to increase and decline (though not always) with inequality. But nobody knew exactly why, though there are some known feedback processes such as regulatory capture (people with money getting more political influence when they have more money.) What we created was the first clear explanation of what might be happening, and then we used a computer simulation to test whether that explanation was coherent, and we discovered it was better than we thought. Then we further tested it with some more data.

Our theory starts from another observation: we know that in general diverse teams do better than homogenous teams. So we start with the assumption that when inequality is low, we would choose to do the thing that works better, which is working together with a broad, diverse range of colleagues. Then our new idea was that maybe even though a more diverse team is most likely on average to give you better results, maybe it is also more risky -- so your expected outcome of working with diverse group is better on average, but there's some chance it will just fail / do nothing at all. So if you're fortune is failing (as is often the case for most people when inequality is increasing) and you are in real danger, for example of losing your home, or business, or not being able to feed your family then you may be better off putting all your trust and attention on those more like yourself, just to scrape by.

Our model showed that this guess does make sense, indeed in simulation this can be the problem.

What is the most fresh and new thing about the results? and what has been done before?

As I mentioned, we have the first explanation for why inequality might prompt polarisation – AND why it doesn't always. If a government is careful to ensure that the poor are still doing better, not worse, then a country can experience inequality without experiencing polarisation. This has happened recently in China and Germany, for example.

Also, we showed two other super interesting things. One is that if things are REALLY bad, so you know that working with others like yourself is not going to be enough, you should go back to working with everyone available again. This is called "gambling for redemption" -- when you are in real trouble you may take a big risk, because you know there is no choice.

The other thing is about getting rid of polarisation. You might think that if you just lower inequality or improve the well being of the worst off, our model shows that polarisation should go away. And you would be right, but only in the simplest case. If we make the model just slightly more complicated -- if we assume that in order to get cooperation that you need BOTH sides of an agreement to agree to take the chance of trusting each other, then polarisation becomes "sticky". Even if the economy gets better, people don't just start trusting each other again. This seems to correspond with real data, where often it takes some kind of a major event to remind people that they are doing well and that they need each other, for example a war or a pandemic. 

What makes the results of the study important? 

Polarisation leads to all kinds of really negative things, it tends to be very destructive of societies and the institutions that benefit us all. Of course, the point is that if inequality is growing, then our institutions aren't benefiting all of us, only some of us. So this study should be a wake-up call to politicians and those who influence them -- if you want a stable, healthy society, there needs to be enough redistribution and institution building to be sure that everyone is benefiting.  That is, the wealthy should pay their taxes, not evade them, and the powerful should use their power to ensure benefits are well spread around.

 Could you please explain in a simple way the relationship between polarization and economic decline?

I hope I did this above. I was talking about inequality, but really our model is mostly about economic decline, which may be suffered by an entire society. In fact, very unequal societies tend to experience economic decline overall, not only for the poor.  But we did also run the model for the condition where the economy stays the same and inequality increases, and the results are the same as what I described above.  

What are the implications of the findings of the paper? 

I believe I answered this under "what makes the study important"

Which methods and experiments have you used to reach the results of the study? Please explain how the work on the study was done.

We used a framework of numerical simulation (agent or individual based modelling) for the theory part of the study I just described.

For the new data contribution, we had meant to test affective polarisation worldwide, but right before our paper came out someone else did that, so we just made a more modest contribution by looking at the USA by state across the last (but one!) three presidential elections, using public data from the e American National Election Study (ANES) and the Cooperative Congressional Election Study (CCES), we show that inequality and affective polarization are correlated across US states, consistent with similar findings that inequality and affective polarization are correlated in a panel of developed democracies, which is work done by Noam Gidron and colleagues which have been presented at conferences but I believe are still in preprint.

Do the results of the study apply to the countries of the Middle East and North Africa? How?

Yes, we'd expect them to apply anywhere. Of course, exactly how varies since the Middle East and North Africa is a very diverse place. In general, places with a low GDP, where people are having trouble finding jobs and sustaining their families, are also going to find it hard to support cooperation on public goods. So I think this study may also explain the famous Herrmann, Thöni, and Gächter findings of 2008 that showed that people in (among other places) several Middle Eastern cities were more inclined than those in say Northern Europe, the USA, or the Far East to attack those who contributed to the public good. I think that's a sign of polarisation that results from the economic situation including increasing inequality. One study I'd like to run is to see whether the kinds of interventions people often run in the Middle East especially, for example bringing together different groups to reconcile, work better in areas where the economy is more stable and the least well off have been taken care of. Our "stickiness" results indicate that even if you've done the right thing and helped the poor out, you may have to make extra efforts to build trust, but you can't build trust without having first fixed the economic situation.


Are there any age or gender differences among the affected people?

Our study doesn't touch on age or gender. But there is some famous older work by Paul Van Lange and colleagues (from 1997) showing that at least for the Dutch at that time, as you grew older you were more likely to be watching out for other people. And also, if you have a sister! Your own gender didn't matter, but whether you had a sister did.

What do you personally find most surprising or exciting about these results? 

It's more exciting that surprising when you have a theory and then you find evidence for it :-).  But I was especially excited that not only did it look like we'd been thinking clearly (it's hard to be sure a complicated theory is right without testing it through simulations like we did), but that it also might explain these other phenomena I mentioned– about when things get very seriously bad you start working together again, or otherwise that it might take an outside event to remind you that you do better working together.

What I'm most excited about is that this joins the growing literature on the importance of addressing inequality. I hope we can persuade people to address the core problems soon, as we need a functioning, cooperating world to address hard problems like pandemics and sustainability, and even how to govern with digital technology and so much cross-border influence. Changing our economy so it stops hurting the planet is going to be a much bigger challenge than the challenge of the pandemic has been, and I don't say that lightly. What I most fear is that some autocrats will read this and realise that it's also a recipe for how they can keep getting people to vote for populism. But polarisation is unstable, and populism is not guaranteed to help any one person for very long. Normally, populists really do tend to help those who voted for them, but Donald Trump was a big exception, passing a tax bill that significantly increased inequality in the USA. But that didn't seem to help him, though it may have helped make the situation there even crazier.

What might you caution the lay public to not misunderstand about your work?

I wouldn't want people to think this means it's better to work only with people like yourself. It may be less risky, because you know what to expect, but it's very unlikely to bring you to a better place. People need to band together and find a way to be able to take the risks required to building greater solutions together.

How long did this work take?

For me, I started working on this in 2016. I had been asked on live television whether AI was causing inequality and I didn't know the answer. So I went to a talk about inequality and it was given by the second author, Nolan McCarty. From him I learned both that inequality was correlated with polarisation, and that polarisation looked like some of my models that I had thought had a bug in them. Why wouldn't you work together if it would help everyone? So we spent a couple of years working together, then I realised we needed another better modeller than either of us (and we're pretty good!) So I asked around and found Alex Stewart. He's an absolute genius and did most of the work that's actually published in the paper. But Nolan McCarty had been gathering and studying the data that the paper is based on for decades. 

What specific directions do you think your research might or should go from here?

As I mentioned, I'd like to test it further with data on what interventions work in what situations. I would also like to go back and test whether (as I said earlier) this might explain some of the strange results other people already had about when you do and don't cooperate, and when you even attack those who cooperate. I also just want to draw more attention to the problem and its explanation, so we can get policy in place to address it.