Forecasting the Olympic medal table from socioeconomic data
- Article
- Sports
Before the Tokyo 2020 Olympic Games, could you have said how many medals each country would take home? Christoph Schlembach, Sascha L. Schmidt, Dominik Schreyer, and Linus Wunderlich built a model to do exactly that, published in Technological Forecasting and Social Change. Their argument is that medal forecasts are useful to more than bettors: governments deciding how much to fund athletic programmes, sponsors choosing where to put money, and media outlets planning coverage all have a stake in knowing where the medals are likely to go.
The model
The forecast uses a two-stage random forest. The first stage predicts whether a country wins any medals at all; the second predicts how many, for the countries that do. The predictors are socioeconomic rather than sporting: GDP adjusted for purchasing power parity, population, the number of athletes a country sends, and the incidence of and deaths from lower respiratory disease as a health indicator. Where data was missing, the authors filled it by extrapolation and by benchmarking against similar countries in the region. The model also included the effect of the COVID-19 pandemic, which had delayed the Games by a year.
Only total medals per country were modelled, not the split into gold, silver, and bronze.
Results
The comparison baseline was the naive model, which predicts that every country will win the same number of medals it did at the previous Games. That baseline is harder to beat than it sounds, because medal tables are quite stable from one Olympics to the next. The two-stage random forest showed a significant improvement over it, and reached an area under the ROC curve of 0.95 on the first-stage question of whether a country medals at all, where 1 would be perfect.
For Tokyo, the model forecast the United States, China, and Great Britain at the top with 120, 87, and 74 medals respectively, followed by Russia at 63 and Japan at 51. Comparing runs with and without the pandemic effect, the authors found that among the top 20 countries COVID-19 made minimal difference to the forecast.
Why it matters
The authors describe this as the first model to beat the naive forecast across the Summer Games from 2008 to 2020, and they suggest that finer-grained data on individual athletes and on medal types could improve it further. For the post, the lesson was more general: a country’s Olympic result is, to a surprising degree, a function of its wealth, size, and health, and a model that knows those three things can see the medal table coming.