Future outcomes still carry uncertainty
How many successes might occur in the next 5, 20, or 50 trials? Move from estimating a success rate to predicting future outcomes.
← Seven labs / LAB 02About two sources of predictive variation
Uncertainty in the rate and randomness in outcomes
The success rate is uncertain, and outcomes would still vary by chance even if that rate were known. A posterior predictive distribution includes both sources of variation.
Change the number of future trials to compare predictions
This example uses a Beta(12, 8) posterior from a Beta(5, 5) prior and seven successes in ten trials.
Change the number of future trials and compare the predicted range of successes.
The posterior mean success rate is 60.0%. Predict successes in the next 20 trials.
The horizontal axis always spans 0–60 successes. The bars include uncertainty in the success rate; the dashed line is a binomial distribution with the rate fixed at its posterior mean. Their difference is not an additive breakdown of probability. The bars below compare 80% predictive intervals on the same scale. Because counts are discrete, an interval may contain more than 80% of the probability.
Compare 5, 20, and 50 trials numerically
What about just one more trial?
With one future trial, there are just two bars: zero or one success. With 20 trials, the chart assigns a predictive probability to each count from zero to 20.