M

🚀Metaculus Open-Source Rewrite is Live!

You're now enjoying a rewritten, open-source Metaculus, with a range of improvements and new opportunities for collaboration. Learn more, and share your feedback and questions here

How to forecast on Metaculus

Binary and Multiple Choice Questions

Examples: "Who will be Japan's next Prime Minister?", "Will NASA's Artemis 2 launch be successful?", …

To predict, share the probability you give the outcome as a number between 0.1% and 99.9%. On the question page, simply drag the prediction slider until it matches your probability and click “Predict”. You can also use the arrows to refine your probability or select the field and type the probability.

Binary prediction example

Multiple choice questions ask about more than two (Yes/No) possibilities. Predicting works the same, except your predictions should sum to 100%. After inputting probabilities, select auto-sum to guarantee they do.

Multiple choice prediction example

The higher the probability you place on the correct outcome, the better (more positive) your score will be. Give the correct outcome a low probability and you'll receive a bad (negative) score. Under Metaculus scoring, you'll get the best score by predicting what you think the actual probability is, rather than trying to “game” the scoring.


Numerical and Date Questions

Examples: "When will humans land on Mars?", "What will Germany's GDP growth be in 2025?", …

To predict, provide a distribution, representing how likely you think each outcome in a range is. On the question page, drag the slider to change the shape of your bell curve, and focus your prediction on values you think are likely.

Numerical prediction example

If you want to distribute your prediction in more than one section of the range, you can add up to four independent bell curves to build your distribution and assign a weight to each of them.

Multiple bell curve prediction example

The higher your distribution is on the value that ultimately occurs, the better your score. The lower your distribution on the actual value, the worse your score. To get the best score, make your distribution reflect how likely each possible value actually is.