Diversity and team performance: What the research says

(Photo of group of people doing a hard thing from Wikimedia user Rizimid, CC BY-SA 3.0.)

This is an extended version (more info, more sources) version of the talk I gave at EA Global San Francisco 2017. The other talk I gave, on extinction events, is  here. Some more EA-focused pieces on diversity, which I’ve read but which were assembled by the indomitable Julia Wise, are:

Effective altruism means effective inclusion

Making EA groups more welcoming

EA Diversity: Unpacking Pandora’s Box

Keeping the EA Movement welcoming

How can we integrate diversity, equity, and inclusion into the animal welfare movement?

Pitfalls in diversity outreach

There are moral, social, etc. reasons to care about diversity, all of which are valuable. I’m only going to look at one aspect, which is performance outcomes. The information I’m drawing from here are primarily meta-studies and experiments in a business context.

Diversity here mostly means demographic diversity (culture, age, gender, race) as well as informational diversity – educational background, for instance. As you might imagine, each of these has different impacts on team performance, but if we treat them as facets of the same thing (“diversity”), some interesting things fall out.

(Types of diversity which, as far as I’m aware, these studies largely didn’t cover: class/wealth, sexual orientation, non-cis genders, disability, most personality traits, communication style, etc.)

Studies don’t show that diversity has an overall clear effect, positive or negative, on the performance of teams or groups of people. (1) (2) The same may also be true on an organizational level. (3)

If we look at this further, we can decompose it into two effects (one where diversity has a neutral or negative impact on performance, and one where it has a mostly positive impact): (4) (3)

Social categorization

This is the human tendency to have an ingroup / outgroup mindset. People like their ingroup more. It’s an “us and them” mentality and it’s often totally unconscious. When diversity interacts with this, the effects are often – though not always – negative.

Diverse teams tend to have:

  • Lower feelings of group cohesion / identification with group
  • Worse communication (3)
  • More conflict (of productive but also non-productive varieties) (also the perception of more conflict) (5)
  • Biases

A silver lining: One of these ingrouping biases is the expectation that people more similar to us will also think more like us. Diversity clues us into diversity of opinions. (6) This gets us into:

Information processing 

— 11/9/17 – I’m much less certain about my conclusions in this section after further reading. Diversity’s effects on creativity/innovation and problem-solving/decision-making have seen mixed results in the literature. See the comments section for more details. I now think the counterbalancing positive force of diversity might mostly be as a proxy for intellectual diversity. Also, I misread a study that was linked here the first time and have removed it. The study is linked in the comments. My bad! —

Creative, intellectual work. (7) Diversity’s effects here are generally positive. Diverse teams are better at:

  • Creativity (2)
  • Innovation (9)
  • Problem solving. Gender diversity is possibly more correlated than individual intelligence of group members. (Note: A similarly-sized replication failed to find the same results. Taymon Beal kindly brought this to my attention after the talk.) (10)

Diverse teams are more likely to discuss alternate ideas, look at data, and question their own beliefs.

This loosely maps onto the “explore / exploit” or “divergent / convergent” processes for projects. (2)

    1. Information processing effects benefit divergent / explore processes.
    2. Social categorization harms convergent / exploit processes.

If your group is just trying to get a job done and doesn’t have to think much about it, that’s when group cohesiveness and communication are most important, and diversity is less likely to help and may even harm performance. If your group has to solve problems, innovate, or analyze data, diversity will give you an edge.

How do we get less of the bad thing? Teams work together better when you can take away harmful effects from social categorization. Some things that help:

    1. The more balanced a team is along some axis of diversity, the less likely you are to see negative effects on performance. (12) (7) Having one woman on your ten-person research team might not do much to help and might trigger social categorization. If you have five women, you’re more likely to see benefits.
    2. Remote teams are less biased (w/r/t gender). Online teams will be less prone to gender bias.
    3. Time. Obvious diversity becomes less salient to a group’s work over time, and diverse teams end up outperforming non-diverse teams. (13) (6) Recognition of less-obvious cognitive differences (e.g. personality and educational diversity) increases over time. As we might hope, the longer a group works together, the less surface-level differences matter.

This article has some ideas on minimizing problems from language fluency, and also for making globally dispersed teams work together better.

How do we get more of the good thing? Diversity is a resource – more information and cognitive tendencies. Having diversity is a first step. How do we get more out of it?

    1. At least for age and educational diversity, high need for cognition. This is the drive of individual members to find information and think about things. (It’s not the same as, or especially correlated to, either IQ or openness to experience (1)).

Harvard Business Review suggests that diversity triggers people to stop and explain their thinking more. We’re biased towards liking and not analyzing things we feel more comfortable with – the “fluency heuristic.” (14) This is uncomfortable work, but if people enjoy doing it, they’re more likely to do it, and get more out of diversity.

But need for cognition is also linked with doing less social categorization at all, so maybe diverse groups with high levels of this just get along better or are more pleasant for all parties. Either way, a group of people who really enjoy analyzing and solving problems are likely to get more out of diversity.

2) A positive diversity mindset. This means that team members have an accurate understanding of potential positive effects from diversity in the context of their work. (4) If you’re working in a charity, you might think that the group you might assign to brainstorming new ways to reach donors might benefit from diversity more than the group assigned to fix your website. That’s probably true. But that’s especially true if they understand how diversity will help them in particular. You could perhaps have your team brainstorm ideas, or look up how diversity affects your particular task. (I was able to find results quickly for diversity in fundraising, diversity in research, diversity in volunteer outreach… so there are resources out there.)

Again, note that diversity’s effect size isn’t huge. It’s smaller than the effect size of support for innovation, external and internal communication, vision, task orientation, and cohesion – all these things you might correctly expect correlate with performance more than diversity (8). That said, I think a lot of people [at EA Global] want to do these creative, innovative, problem-solving things – convince other people to change lives, change the world, stop robots from destroying the earth. All of these are really important and really hard, and we need any advantage we can get.

  1. Work Group Diversity
  2. Understanding the effects of cultural diversity in teams: A meta-analysis of research on multicultural work groups
  3. The effects of diversity on business performance: Report of the diversity research network
  4. Diversity mindsets and the performance of diverse teams
  5. The biases that punish racially diverse teams
  6. Time, Teams, and Task Performance
  7. Role of gender in team collaboration and performance
  8. Team-level predictors of innovation at work: A comprehensive meta-analysis spanning three decades of research
  9. Why diverse teams are smarter
  10. Evidence of a collective intelligence factor in the performance of human groups
  11. When and how diversity benefits teams: The importance of team members’ need for cognition
  12. Diverse backgrounds and personalities can strengthen groups
  13. The influence of ethnic diversity on leadership, group process, and performance: an examination of learning teams
  14. Diverse teams feel less comfortable – and that’s why they perform better

Evolutionary Innovation as a Global Catastrophic Risk

(This is an extended version of the talk I have at EA Global San Francisco 2017. Long-time readers will recognize it as an updated version of a post I wrote last year. It was wonderful meeting people there!)


This is a graph of extinction events over the history of animal life.

There are five canonical major extinction events that have occurred since the evolution of multicellular life. Biotic replacement has been hypothesized as the major mechanism for two of them: the late Devonian extinction and the Permian-Triassic extinction. There are three other major events – the Great Oxygenation Event, End Ediacaran extinction, and the Anthropocene / Quaternary extinction.

Let’s look at four of them. The first actually occurs right before this graph starts.

I decided not to discuss the Great Oxygenation Event in the talk itself, but it’s also an example – photosynthetic cyanobacteria evolved and started pumping oxygen into the atmosphere, which after filling up oxygen sinks in rocks, flooded into the air and poisoned many of the anaerobes, leading to the “oxygen die-off” and the “rusting of the earth.” I excluded it because A) it wasn’t about multicellular life, which, let’s face it, is much more relevant and interesting, and B) I believe it happened over such a long amount of time as to be not worth considering on the same scale as the others.

(I was going to jokingly call these “animal x-risks”, but figured that might confuse people about that the point of the talk was.)

The End-Ediacaran extinction

“Disckonsia Costata” by Verisimilius is licensed under CC BY-SA 3.0

We don’t know much about Precambrian life, but it’s known as the “Garden of Ediacara” and seems to have been a peaceful time.

The Ediacaran sea floor was covered in a mat of algae and bacteria, and ‘critters’ – some were definitely animals, others we’re not sure – ate or lived on the mats. There were tunneling worms, the limpets, some polyps, and the sand-filled curiosities termed “vendozoans”. They may have been single enormous cells like today’s xenophylophores, with the sand giving them structural support. The fiercest animal is described as a “soft limpet” that eats microbes. They don’t seem to have had predators, and this period is sometimes known as the “Garden of Ediacara”. (1)

At 542 million years ago, something happens – the Cambrian explosion. In a very short 5 million years, a variety of animals evolve in a short window.

Molluscs, trilobites and other arthropods, a creative variety of worms eventually including the delightful Hallucigenia, and sponges exploded into the Cambrian. They’re faster and smarter than anything that’s ever existed. The peaceful Ediacaran critters are either outcompeted or gobbled up, and vanish from the fossil record. The first shelled animals indicate that predation had arrived, and that the gates of the Garden of Ediacara had closed forever.

The end-Devonian extinction

Jump forward a few million years – 50% of genuses go extinct. Marine species suffered the most in this event, probably due to anoxia.

There’s an unexpected possible culprit – plants around this time made a few evolutionary leaps that began the first forests. Suddenly a lot of trees pumping oxygen into the air lead to global cooling, and large amounts of soil lead to nutrient-rich runoff, which lead to widespread marine anoxia which decimates the ocean.

Gingko trees, some of the oldest tree lineages alive. Image by Jean-Pol Grandmont, under a CC BY-SA 3.0 license.

We do know that there were a series of extinction events, so forests were probably only a partial cause. The longer climate trend around the extinction was global warming, so the yo-yoing temperature (from general warming and cooling from plants) likely contributed to extinction. (2) It’s strange to think that the land before 375 million years ago didn’t have much in the way of soil – major root structures contributed to rock wearing away. Plus, once you have some soil, and once the first trees die and contribute their nutrients, you get more soil and more plants – a positive feedback loop.

The specific trifecta of evolutions that let forests take over land: significant root structures, complex vascular systems, and seeds. Plants prior to this were small, lichen-like, and had to reproduce in water. (3)

The Permian-Triassic extinction

96% of marine species go extinct. Most of this happens in a 20,000 year window, which is nothing in geologic time. This is the largest and most sudden prehistoric extinction known.

The cause of this one was confusing for a long time. We know the earth got warmer, or maybe cooler, and that volcanoes were going off, but the timing didn’t quite match up.

Volcanoes were going off for much longer than the extinction, and it looks like die-offs were happening faster than we’d expect from increasing volcanism, or standard climate change cycles. (4) One theory points out that die-offs line up with exponential or super-exponential growth, as in, from a replicating microbe. Remember high school biology?

One theory suggests Methanosarcina, an archaea that evolved the chemical process that turned organic carbon into methane around the same time. Remember those volcanoes? They were spewing enormous amounts of nickel – an important co-factor for that process.

Methanosarcina, image from Nature

(Methanosarcina appeared to have gotten the gene from a cellulose-digesting bacteria – definitely a neat trick. (5) )

The theory goes that Methanosarcina picked up its new pathway, and flooded the atmosphere with methane, which raised the surface temperature of the oceans to 45 degrees Celsius and killed most life. (2)

This report is a little recent, and it’s certainly unique, so I don’t want to claim that it’s definitely confirmed, or sure on the same level that, say, the Chicxulub impact theory is confirmed. That said, at the time of this writing, the cause of the Permian-Triassic extinction is unclear, and the methanogen theory doesn’t seem to have been majorly criticized or debunked.

Quaternary and Anthropocene extinctions

Finally, I’m going to combine the Quaternary and Anthropocene events. They don’t show up on this chart because the data’s still coming in, but you know the story – maybe you’re an ice-age megafauna, or rainforest amphibian, and you are having a perfectly fine time, until these pretentious monkeys just walk out of the Rift Valley, and turn you into a steak or a corn farm.

Art by Heinrich Harder.

Because of humans, since 1900, extinctions have been happening at about a thousand times the background rate.

(Looking at the original chart, you might notice that the “background” number of extinctions appears to be declining over time – what’s with that? Probably nothing cosmic – more recent species are just more likely to survive to the present day.)

Impacts from evolutionary innovation

You can probably see a common thread by now. These extinctions were caused – at least in part – by natural selection stumbling upon an unusually successful strategy. Changing external conditions, like nickel from volcanoes or other climate change, might contribute by giving an edge to a new adaptation.

    1. In some cases, something evolved that directly competed the others – biotic replacement
    2. In others, something evolved that changed the atmosphere.
    3. I’m going to throw in one more – that any time a species goes extinct due to a new disease, that’s also an evolutionary innovation. Now, as far as we can tell, this is extremely rare in nature, but possible. (7)

Are humans at risk from this?

From natural risk? It seems unlikely. These events are rare and can take on the order of thousands of years or more to unfold, at which point we’d likely be able to do something about it.

That is, as far as we know – the fossil record is spotty. As far as I can tell, we were able to pin the worst of the Permian-Triassic extinction down to 20,000 years only because that’s how narrow the resolution on the fossil band formed at the time was. It might have actually been quicker.

Even determining if an extinction has happened or not, or if the rock just happened to become less good at holding fossils, is a struggle. I liked this paper not really for the details of extinction events (I don’t think the “mass extinctions are periodic” idea is used these days), but for the nitty gritty details of how to pull detailed data out of rocks.

That said, for calibrating your understanding, it seems possible that extinctions from evolutionary innovation are more common than mass extinctions involving asteroids (only one mass extinction has been solidly attributed to an asteroid: the Chicxulub impact that ended the reign of dinosaurs.) That’s not to say large asteroid impacts (bolides) don’t cause smaller extinctions – but one source estimated the bolide:extinction ratio to be 175:1. (2)

Plus, having a brain matters, and I think I can say it’s really unlikely that a better predator (or a new kind of plant) is going to evolve without us noticing. There are some parallels here with, say, artificial intelligence risk, but I think the connection is tenuous enough that it might not be useful.

If we learn that such an event is happening, it’s not clear what we’d do – it depends on specifics.

Synthetic biology

But consider synthetic biology – the thing where we design new organisms and see what happens. As capabilities expand, should we worry about lab escapes on an existential scale? I mean, it has happened in nature.

Evolution has spent billions of years trying to design better and better replicators. And yet, evolutionary innovation catastrophes are still pretty rare.

That said, people have a couple of advantages:

        1. We can do things on purpose. (I mean, a human working on this might not be trying to make a catastrophic geoweapon – but they might still be trying to make a really good replicator.)
        2. We can come up with entirely new things. When natural selection innovates, every incremental step on the way to the final result has to an improvement on what came before. It’s like if you tried to build a footbridge, but at every single step of building it, it had to support more weight than before. We don’t have those constraints – we can just design a bridge and then build it and then have people walk across it. We can design biological systems that nobody has seen before.

This question of if we can design organisms more effective than evolution is still open, and crucial for telling us how concerned we should be about synthetic organisms in the environment.

People are concerned about synthetic biology and the risk of organisms “escaping” from a lab, industrial setting, or medical setting into the environment, and perhaps persisting or causing local damage. They just don’t seem to be worried on an existential level. I’m not sure if they should be, but it seems like the possibility is worth considering.

For instance, a company once almost released large quantities of an engineered bacteria that turned out to produce soil ethanol in large enough quantities to kill all plants in a lab microcosm. It appears that we don’t have reason to think it would have outcompeted other soil biota and actually caused an existential or even a local catastrophe, but it was caught at the last minute and the implications are clearly troubling. (9)

  1. Ediacaran biota: The dawn of animal life in the shadow of giant protists
  2. On the causes of mass extinctions
  3. Terrestrial-Marine Teleconnections in the Devonian: Links between the Evolution of Land Plants, Weathering Processes, and Marine Anoxic Events
  4. The Permo-Triassic extinction
  5. Methanogenic burst in the End-Permian carbon cycle
  6. Natural Die-offs of Large Mammals: Implications for Conservation I’m pretty sure I’ve seen at least a couple other sources mention this, but can’t find them right now. I had Chytridiomycosis in mind as well. This seems like an important research project and obviously has some implications for, say, biology existential risk.
  7. Rather sensationalized description from Cracked.Com

Talking at EA Global

I’m speaking at both lightning talk sessions (Saturday and Sunday afternoon) at the Effective Altruism Global conference SF this weekend. Catch me talking about evolutionary innovation and extinction on Saturday (5:15), and diversity in teams on Sunday (4:00).

On the off chance that you’ll be at the conference but haven’t already met me, or perhaps know me and want to chat more, feel free to comment on this post or send me an email at eukaryotewritesblog (at) gmail.com to arrange meeting up and saying hello.

I was at a party the night before and got into at least six different conversations about the existential risk / biology overlap, so I’m expecting this weekend to be a really good time. See you there!

(If you can’t make it, I’ll post the talks and longer versions of what I talked about here afterwards.)