I've read a number of different explanations trying to understand the likelihood function, and I understand the purpose of it, but some statements sound contradictory.

Consider observed data X, model parameters M, likelihood function L(M|X), and probability P(X|M).

I keep seeing it written that L(M|X) = P(X|M). At the same time, they say that the likelihood is not the same thing as the probability and L is not a probability density function (pdf).

What does it mean to say they are not the same thing but equal? How could L be equal to a pdf, but not be a pdf?

Reference to one of the places I've read this: http://www.stat.cmu.edu/~larry/=stat705/Lecture6.pdf


We say that the likelihood is not the same thing as the probability because, $\mathcal{L} (M\mid X )\neq P(M \mid X) $.

But it is true that $\mathcal{L} (M\mid X )= P(X \mid M) $. So why do we say that the likelihood is not a pdf? This is because the likelihood is seen as a function of $M$ rather than of $X$. As a function of $M$, the likelihood does not obey the laws of probability: the integral of the likelihood over the parameter space can be greater than 1.

If you want to define a pdf on the parameter space, you need to use Bayes' theorem: $P(M \mid X) = \frac{\mathcal{L} (M\mid X )\times P(M)}{P(X)}$

For a more complete view, have a look at What is the difference between "likelihood" and "probability"?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.