What should happen to a candidate distribution assigning zero density to an uncensored observed value?
StatisticsLevel: IntermediateLikelihood and fitted distributions
What should happen to a candidate distribution assigning zero density to an uncensored observed value?
Hints
- One zero factor kills the product.
- Check support violations.
Answer
Its likelihood is zero and log likelihood is negative infinity. Numerical implementations may use a large penalty; that is an optimization device, not statistical evidence.
Source and licence: SciPy — distribution fitting and likelihood · BSD-3-Clause · answer: Quantoros editorial