If I have a binomial maximum likelihood function

$$ L(\theta | k) = \Pi_i^N p(\theta)^{k_i}(1-p(\theta))^{(n_i-k_i)}$$

I know that the index terms can be made into a sum, simplifying the expression for calculation purposes. I have a model which predicts the probability of success as a function of a variable (x) and a parameter. I have measured k (knowing n) for a number of values of x and now I want to find the best fit value for my parameter which I am trying to determine experimentally. Am I right in thinking my likelihood function is in this case

$$ L(\theta | x, k) = \Pi_i^N p(\theta, x_i)^{k_i}(1-p(\theta, x_i))^{(n_i-k_i)}$$

Assuming this is the case (if not can someone correct me), is there anyway I can convert my product to a sum, or, do I have to keep it in product form?


2 Answers 2


Same way as before, surely: by taking logs to form the log-likelihood.


You can use the Poisson approximation to Binomial and compute the log-likelihood convert multiplication to sum.

$$ \text{ln}( L(\theta | x, k)) = \text{ln}( \Pi_i^N (n_ip(\theta, x_i))^{k_i}\frac{e^{-n_ip(\theta, x_i)}}{k_i!})$$

Then you can omit the factorial since it doesn't depend on $p$ nor $x$

$$ \text{ln}( L(\theta | x, k)) = \sum_i^N k_i\text{ln}(n_ip(\theta, x_i))^{}+\sum_i^Nn_ip(\theta, x_i)$$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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