Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It only takes a minute to sign up.
Sign up to join this community
Let the density function be given by
$$ f(x;a,b) = \frac{a + 2 b g(x) + (1-a-b) g(x)^2}{(1-x)(2 b g(x) + (1-a-b) g(x)^2)}$$
where $a$ and $b$ are parameters of interest and $g(x)$ is a known function.
I was told that using this density function in maximum likelihood, the parameters $a$ and $b$ are identified.
The concept of identification is clear to me, but what are the rigorous mathematical considerations to conclude identifiability here?
In Chapter of 14 [p. 2] of Greene's book it's stated that for a likelihood function:
The parameter vector $\theta$ is identified (estimable) if for any other parameter vector, $\theta^* \neq \theta$, for some data $y$, $L(\theta^∗|y)\neq L(\theta |y)$.
Based on that, it's not hard to check that for different values of $a$ and $b$ you'll have different values for the likelihood of the given density.
EDIT: When I say different values of $f$ for different values of $a$ and $b$ I should actually say the likelihood function must be a one-to-one correspondence (or bijective, if you prefer).
I think a reasonable approach to a rigorous proof would be to assume the contrapositive -- that is, assume there exists some $a'$ and some $b'$ such that $$ f(x;a',b') = f(x;a,b)$$ and demonstrate based on this assumption that $a=a'$ and $b=b'$.
