Correct Predictions, Correct Method? COVID Variants Spread Globally and de Sousa Wins in Portugal
Public Health and Pandemics Elections Forecast Retrospective

Correct Predictions, Correct Method? COVID Variants Spread Globally and de Sousa Wins in Portugal

Clay Graubard
Andrew Eaddy
Clay Graubard, Andrew Eaddy

In the past two weeks, two of our predictions at Global Guessing have resolved. Both were correct, but was our method correct? Let's examine what happened and what we think about our predictions.

UK COVID Variant Spread

The first prediction was on the spread of the new, mutated coronavirus variant first detected in the United Kingdom. On December 21, 2020, Global Guessing predicted:

  • There was a 82.5% chance [89% CI = 75-90%] that at least 4 of the following 5 countries identified the new COVID mutation from the U.K. within their borders in the next two weeks: United States; Spain; Germany; United Arab Emirates; and France.
  • There was a 85% chance [89% CI = 75-90%] that at least 3 of the following 4 countries identified the new strain in the next two weeks: India; Switzerland; Portugal; and Ireland.
  • There was a 75% chance [89% CI = 70-85%] that at least 4 of the following 6 countries identified the new strain in the next three weeks: Austria; Qatar; Sweden; Greece; Japan; and South Africa.
  • And there was a 73.5% chance [60%, 80%] that at least 4 of the following 10 countries identified the new strain within their borders in the next four weeks: Canada; Egypt; Russia; Turkey; Hong Kong; Maldives; Nigeria; Poland; Romania; and Singapore.

All four of these predictions came true.

Portugal's Next President

The second prediction was on the presidential election in Portugal which took place on January 24th. We predicted that Marcelo de Sousa was the overwhelming favorite to win the election.

  • With a calculated polling average over 63%, Global Guessing gave President Marcelo de Sousa a 99.42% chance to win, making him nearly certain to win.

On January 25 it was announced that incumbent President Marcelo Rebelo de Sousa cruised to re-election. Despite a troubling election lead-up which saw de Sousa contract the coronavirus and the country’s coronavirus cases surge, the former president earned an impressive 60.7% in his victory.

  • By comparison, de Sousa won the presidency in 2016 with 52% of the vote.

Global Guessing forecasted that de Sousa would win with 59.94% of the vote, .76% off of the exact tally. Moreover, the most likely individual outcome in our model was de Sousa winning 60-61% of the vote.

Socialist candidate Ana Gomes came second in the presidential election with 12.9% of the vote, while the nationalist candidate Andre Ventura came in third with 11.9% of the vote. Despite Ventura’s third place finish, some see his vote share in this election as evidence of the potential rise of a far-right culture in Portugal.

  • Chega, the far-right political party Ventura started in 2019, has historically espoused incredibly divisive and targeted language.
  • As a result, Ventura’s political future and de Sousa’s response to this phenomenon will be under great scrutiny by the international community.

Reflecting on Predictions

Overall, we are pleased with these two predictions. However, two correct predictions does not mean the predictions were reached perfectly. Although the Portugese prediction was highly accurate, we nevertheless feel that our model is ill-equipped to forecast vote distribution likelihoods.

  • Although we were mildly concerned following our predictions in Ghana and Niger, the limitation of our model better revealed itself when the distance between the base-rate and current-position is great.

Similarly, we feel our prediction for the spread of the B.1.1.7 COVID variant was extremely well constructed for the world of December 21, 2020. The issue, however, was the static nature of our predictions.

  • Normally this has posed a minimal issue with our predictions since we release predictions close to the event. However, with the COVID prediction not only was the event on-going, but the amount of information we learned in late December about B.1.1.7 and other variants was exponentially increasing daily.
  • Moving forward, we are working on a way to efficiently update predictions as new information comes out.

Reflecting on our COVID Variant Tracker

Finally, our COVID-19 prediction live-update article (and accompanying Twitter threads) was a complete success and was our first attempt at efficiently updating information as it comes out. For a month, our live update provided the most up-to-date and comprehensive tracker of the UK and South African COVID variants, staying ahead of major news outlets such as Al Jazeera and others.

  • Although we had originally planned on updating the variant spread manually for a few days, with the understanding that the mainstream media outlets would create their own, faster updating ones shortly, the project continued until our predictions closed. Thankfully, around this time other reliable and regularly updated articles emerged.

If you are interested in tracking the COVID-19 variants, we recommend:

  1. CoV-Lineages: UK Variant (B.1.1.7); South African Variant (B.1.3.5.1); Brazilian Variant (P.1) | Most up-to-date information.
  2. Recombinomics: Google Maps Tracker | Relies on virus sequence data from GISAID, and as a result countries might not show up for days leading to an update lag. On the other hand, it is a good way to visualize the geographic spread due to map and sequence plotting.

NextStrain is also an excellent resource for tracking the new mutations of the virus.

If you would like to learn more about the variants, we recommend these articles on the UK Variant, South African Variant, and Brazillian Variant.