Maximum Likelihood Method(redirected from Maximum likelihood)
Also found in: Dictionary, Medical, Acronyms, Wikipedia.
maximum likelihood method[′mak·sə·məm ′līk·lē‚hu̇d ‚meth·əd]
Maximum Likelihood Method
a method of finding statistical estimates of the unknown parameters of a distribution. According to the maximum likelihood method, we select as the estimates of the parameters those values for which the data resulting from observations are “most likely.” It is assumed that the results of observations X1, …, Xn are mutually independent random variables with identical probability distributions, all depending on the same unknown parameter θ ε ө, where ө is the set of admissible values of θ. To assign an exact meaning to the concept of “most likely,” we proceed by introducing a function
L(x1, …, xn; θ) = p(x1; θ) … p(xn; θ)
where p(t;θ) for a continuous distribution is interpreted as the probability density of the random variable X, and in the discrete case as the probability that the random variable X takes the value t. The function L(X1, …, Xn;θ) of the random variables X1,…, Xn is called the likelihood function, and the maximum likelihood estimate of the parameter θ is that value (X1, …, Xn) (which is itself a random variable) of θ for which the likelihood function attains the largest possible value. Since the maximum point for log L is the same as that for L, it is usually sufficient to solve the so-called likelihood equation
in order to find the maximum likelihood estimates.
The maximum likelihood method does not always lead to acceptable results but in some sense is the best method for a broad set of cases of practical importance. For example, we may assert that if there exists an efficient unbiased estimate θ* for the parameter θ in a sample of size n, then the likelihood equation will have the unique solution θ = θ*. In dealing with the asymptotic behavior of maximum likelihood estimates for large n, it is well known that the maximum likelihood method leads under certain general conditions to a consistent estimate that is asymptotically normal and asymptotically efficient. The definitions given above can be generalized to the case of several unknown parameters and to the case of samples from multivariate distributions.
The maximum likelihood method in its modern form was proposed by the British statistician R. Fisher in 1912, although particular forms of the method were used by K. Gauss; even earlier, in the 18th century, J. Lambert and D. Bernoulli came close to the idea of the method.
REFERENCESCramer, H. Matematicheskie metody slalisliki. Moscow, 1948. (Translated from English.)
Rao, C. R. Lineinye statisticheskie metody i ikh primeneniia. Moscow, 1968. (Translated from English.)
Hudson, D. Statistika dlia fizikov. Moscow, 1970. (Translated from English.)
A. V. PROKHOROV