Metaculus Mondays is a series where we at Global Guessing answer and reflect on a select set of questions from Metaculus.com: a community forecasting platform. Every Monday, we plan to answer 2 - 3 questions that people from the community posted while discussing what the community thinks and examining the questions themselves.
In this first volume of the series we will be looking at three questions which relate to our website’s overarching focus on geopolitics.
The first question deals with the prevalence of the mutated COVID variants in the United States by the end of March. Given Global Guessing’s previous work on both COVID-19 and the emerging variants, as well clear geopolitical consequences of a worsening pandemic due to these variants, the question was a perfect fit for the inaugural volume of Metaculus Mondays.
The second question asks what the likelihood is that a nuclear weapon attack will be carried out by 2050, while the third asks about the chances that a ‘World War III’ by 2050. We considered these questions together due to the fact that their resolution time frames are identical, and that both eventualities, while fairly unlikely, demonstrate a similar magnitude of change to the international order. Furthermore, both questions are clearly related to our focus on geopolitics, and present a new event-type for Global Guessing to predict!
Q1: The Dominance of Mutated COVID-19 Variants
The first question is asking us to predict what percentage of all virus sequences reported to the GISAID database from the United States will be from the new mutated COVID variants during the week of March 29, 2021. The variants of concern are the ones with the N501 mutation, including the ones detected in the United Kingdom (B.1.1.7), South Africa (B.220.127.116.11), and Brazil (P.1).
According to the Metaculus community (as of 10:05pm MT, 2/7/21), the median prediction is that 78% of all viruses sequenced and reported to the GISAID database will be from S:N501 variants. The community prediction has a lower 25% of 76% and an upper 75% of 89% of all viruses. Because this question is so new, there is no community comments supporting predictions and therefore it is difficult to understand the reasoning behind the community prediction thus far.
Before going into our prediction and reasoning, we first want to comment briefly on the question itself. Out of the three questions we answered in today's volume, this one was by far the best constructed. The question as presented on Metaculus provides clear framing, outlines what the S:N501 variants are and which notable variants count, links to useful data sources to start tackling the problem, and lays out clear criteria for resolution which relies on the most credible data source (GISAID). Although this question currently has no comments (we plan to convert this into one shortly!), we hope it gains traction because it is both important and well constructed.
To reach our prediction, we considered the following observations:
1) A new paper projects that the UK-variant (B.1.1.7) will make up ~70% of all new COVID-19 cases in the United States by March 29, while the CDC forecasts it will become the dominant variant by the end of March.
2) The current and historical growth data of S:N501 viruses in countries from Covariants supports rapid growth and dominance of new variants.
3) The current Rt (transmission rate) of the virus in the United States is just below the 1.0 threshold, meaning the pandemic could see exponential growth again but this time driven by new variants–and thus increasing their prevalence further.
4) Based on data from cov-lineages on the South African (B.18.104.22.168) and Brazilian (P.1) variants, as well as the geographic distribution of the new variants in the United States, it is highly likely that these other two variants will also spread. Based on current geographic distribution, the South African variant appears more worrisome since it is located on the East Coast.
Based on this information, we determined:
- The U.K. variant would make up, most likely, 70% of all sequences by March 29, however it could likely be more including above 76% of all sequences.
- The South African and Brazilian variants will make up at least 6.4% of uploaded sequences to GISAID based on the growth rate of the B.1.1.7 variant, and up to 15.6%.
- Given the number of viruses the United States sequences, it is possible for the number of uploaded sequences to be different from the true prevalence in the population, thus increasing uncertainty.
- At the same time, it is likely that these variants will gain extra relative dominance in the coming weeks due to their relative immunity to existing antibodies and vaccines.
We therefore calculated that 81% of all sequenced viruses uploaded to GISAID from the United States during the week of March 29 would be from the S:N501 mutated variants.
Based on the above factors, we set the lower 25% at 76% and the upper 75% at 89%. At the same time, however, we also recognize that it is more likely for the viruses to make up nearly all or all the cases than it was to make up less than 70% of all cases. As a result, we crafted our probability density graph to represent a leveling out for the 90%+ possibilities.
Given that this question will remain open for the following week, we plan to follow the growth of the new variants closely. We are particularly interested in the following questions:
- What is the trend of S:N501 prevalence in Israel since its vaccination campaign? How has it changed in the past week? How does this trend compare to other countries?
- Are there any changes in US vaccine projections?
- Is the US making any major changes with respect to sequencing that would take effect before March 29?
- What is the trend of lockdown orders across the country?
The answers to these questions will help us better understand how the trajectory of these variants might look in the United States (and other countries, for that matter) in the coming weeks.
Q2-3: Nuclear War and World War 3
The second question asks whether ‘World War III’ will break out before 2050. In this case, ‘World War III’ is defined as a conflict that:
- Either (a) involves countries totaling 30%+ of GDP or (b) 50% of the global population
- Ten million people will have died in the conflict.
(The question also indicates that World War 3 is a ‘hot’ war, but it is unclear whether cyber-warfare or other forms of sub-threshold conflict would qualify.)
For this question, the median Metaculus forecast is an 18% likelihood of WW3 based on 559 predictions as of writing.
Some forecasters weighed similar issues to the previous question, namely specific ongoing conflicts between states that could eventually lead to a war. Comments from military leaders and current events were also popular sources of confidence and information on this prediction.
One interesting phenomenon that can be observed with this question, and with forecasting in general, is recency bias. Global Guessing finds the current community prediction of 18% to be very high. The community had predicted last April that the chances of a ‘World War III’ were 9% last April.
It is very possible that this high likelihood percentage generated by the Metaculus community is the result of the extremely sensationalist media and political volatility witnessed in the United States since that time. It follows that these stimuli may have resulted in reactionary predictions which present the world as much closer to a ‘World War III’ than Global Guessing believes we are.
The third question asks whether a nuclear weapon will be used in an act of war between now and 2050. Importantly, the question requires that the attack be deliberate and carried out by a nation-state. The mentaculus community’s median forecast is a 20% likelihood based on 88 predictions thus far.
Several factors affecting the outcome of this prediction have been discussed in the comment section of the prediction, and are important to consider.
Some forecasters have put a focus on specific bilateral conflicts that exist today, that have the potential to lead to nuclear engagement. Others have brought up the trajectory of technological innovation, and its dampening effects on the appeal of nuclear engagement in the future. And still others have evoked the popular deterrence theory, and its principals as an argument that it is not in states’ best interest to ever engage in nuclear warfare, as it is ultimately a zero-sum game.
What is fascinating about these two questions is that the Metaculus community has forecasted the likelihood of both questions to be very similar. This makes sense, although at first struck us as puzzling. Although it is most likely for a nuclear weapon to be denoted in a World War (the base-rate would back you up here), there are still numerous other scenarios where one could be used in a war that would neither start nor take place in a World War. So while Global Guessing finds the percentages given for each outcome to be too high, the fact that they track each other speaks to the aligned thinking and consistency of the Metaculus community.
For the question of a potential ‘World War III’ we first looked at the community prediction value for this question before its spike in April. The 9% forecasted was a number we felt comfortable with before examining the forecast results, and that number aligning with what the community had forecasted reified our confidence in a value near that number.
Furthermore, given the criteria provided on Metaculus we only identified a select number of current conflicts that could possibly result in a World War. Those conflicts were the ongoing power struggle between China and the United States, and the ongoing conflict between India and Pakistan (both nuclear powers), which could pull in the likes from China, the United States, and possibly Russia in the next three decades.
Aggregating the community forecast with our own outlook on the current makeup of the international arena, we felt that 9% was a strong prediction for the chances of a ‘World War III’ breaking out before 2050, per the criteria provided on Metaculus.
For the question of a nuclear weapon attack in an act of war, we first started on the conditional probability of nuclear detonation given an outbreak of World War III. We calculated this to be 60%, leading to a total likelihood of 5.4%.
Next, we considered the current suite of nuclear powers and their relative appetites for war. Weighing the strength of the current rules-based order alongside deterrence theory and logic, we were able to rule out a number of states, including the likes of the United Kingdom and France. We then considered areas of high nuclear weapon concentration, and areas of great power conflict to be the most probable backdrops for this prediction to occur. This includes areas in Eastern Europe, like Russia, as well as Asian powers like China, India, and Pakistan. Additionally, given their global influence, the United States would potentially be a relevant player as well.
Given the current trajectory of some of the aforementioned conflicts in addition to the potential for a ‘sign-of-strength’ nuclear attack in the Eurasia region in the next 30 years, we felt that an additional 4.26% likelihood was an accurate measure, based on an estimated yearly detonation likelihood of 0.20%. Combining that with our conditional probability...
We forecasted a 10% chance that a nuclear weapon is detonated as a deliberate act of war by a nation-state by 2050.
Like many questions, wording can be problematic. As previously mentioned, the qualifications of ‘hot’ war in the WW3 question remains undefined. Moreover, the question regarding nuclear weapons only requires a single nuclear weapon attack as an act of war to be carried out before 2050 to resolve. The wording of the question, however, can be confusing as it has sounded to some forecasters as a call for two or more nuclear weapons strikes as an act of war before that date. For some reason, the question was phrased in terms of total nuclear weapons ever used in war during human history...rather than the usage of nuclear weapons in between the present and 2050. The likelihood of the former is much easier to work with relative to the latter, marking an important distinction for this prediction.
Generally, however, we found both questions to be thorough although inferior to the COVID one. The long resolution windows for each question means that our habit of tracking our prediction accuracy will have to be sunset for these questions.
About the Series
Metaculus Mondays will be a series, every week on Global Guessing. As of now, we only plan to make forecasts on other people's questions. However, as we answer, review, and critique more questions, we hope to start asking some of our own questions.
We hope that this series will make us better forecasts and ask better questions while also getting us involved into the growing Metaculus community. We think the platform has a lot of potential, and from what we've observed so far the community seems great.
We are also aware that Scott Alexander is doing a series by a very similar name–Metaculus Monday (note the lack of an s)–on his new substack, Astral Codex Tex. Although we've long planned this series, Scott's started first. At the same time, we really like the name and...it's not particularly original anyways. So we decided to keep the name Metaculus Mondays, and you should also check out his own take on the series. Instead of making prediction, he goes over what the community is saying on a larger number of questions while also making comments about the questions themselves.
Finally, we are also considering a similar series for Facebook's new platform, Forecast, although we are also looking to see if other forecasting communities are out there too. If you know of any, let us know!
To get the next volume of Metaculus Monday's delivered to your inbox, subscribe below to receive Global Guessing articles as a newsletter.