SASAI LAB / INTERACTIVE STATISTICS
BAYESPLAY
Each observation
changes the picture.
Represent what remains unknown with a distribution.
How does its shape change when data arrive?
Move the values and compare the curves to build intuition for Bayesian statistics.
Explore Bayesian statistics from observation to prediction in StatPlay.
Seven successes. Is the success rate 70%?
Move through steps 1 → 2 → 3 and compare the curves.
Uncertainty remains after the update.
The horizontal axis is the unknown success rate; the vertical axis is probability density. The dashed likelihood is rescaled to compare shapes. The band shows a 95% credible interval.
One update leads to more questions.
New to the sequence? Start with Lab 01.
Or choose the question that interests you.
01 → 03 / UPDATING AND PREDICTION
Update a success rate
Add successes and failures one at a time. Compare the prior, likelihood, and posterior.
BETA × BINOMIALPredict future results
How does uncertainty in the success rate affect the next 20 outcomes?
POSTERIOR PREDICTIVECompare 7/10 with 70/100
Both are 70%. How does the number of observations change the width of the curve?
LIKELIHOOD04 → 05 / NEW SETTINGS
Which way does the mean move?
Change the sample size and spread to see how the prior and observed means are weighted.
NORMAL × NORMALTen events in how much time?
Combine event counts and observation time to estimate events per hour.
GAMMA × POISSON06 → 07 / SHARING INFORMATION
What can a small sample tell us?
See how shared information moves an extreme rate from a small sample.
SHRINKAGEBorrow information while retaining differences
Use the overall center and differences among groups to explore partial pooling.
PARTIAL POOLINGLook at the estimate
and the uncertainty that remains.
Bayesian statistics represents uncertainty about unknown values with distributions and updates those distributions using data.
- 01 / PRIOR
Prior distribution
A distribution over plausible values before observing the data.
- 02 / LIKELIHOOD
Likelihood
How well each possible value explains the observed data.
- 03 / POSTERIOR
Posterior distribution
The updated distribution obtained by combining the prior and likelihood, then normalizing its area to one.