reunalaskelmat

Reunalaskelmat is a Finnish term used in statistics to refer to estimates that are obtained under constraints that restrict the parameter space, placing the estimate on the boundary of that space. The word combines reuna (boundary) and laskelmat (calculations/estimates). In constrained estimation, parameters must satisfy inequalities or other restrictions, which can force the estimator to lie on a boundary.

Reunalaskelmat arise in several settings, including constrained maximum likelihood estimation, Bayesian analyses with parameter constraints, and nonparametric methods with restricted support. They are especially common when the true parameter is near or on a boundary, such as probabilities constrained to [0,1] or variances constrained to be non-negative.

Statistical properties of boundary estimates differ from unconstrained ones. The usual asymptotic normality may not hold, and the distribution of estimators or test statistics can be non-standard. For example, likelihood ratio tests can have mixture distributions rather than a simple chi-square distribution when the null parameter lies on a boundary. Inference often requires alternative approaches, such as constrained optimization, reparameterization to remove the boundary, penalized likelihood, bootstrap methods, or profile likelihood intervals that respect the constraint.

Common examples include estimating a success probability with p in [0,1], where the maximum likelihood estimate can lie at p = 0 or p = 1, and variance components that must be non-negative. Reunalaskelmat highlight the need for careful interpretation of estimates, standard errors, and p-values in the presence of parameter constraints.

See also: constrained estimation, maximum likelihood estimation, boundary value problems.