Gjettingsparameter

Gjettingsparameter is a term used in statistics, machine learning, and decision modeling to describe a parameter that governs the influence of guessing or random exploration in a predictive process. The concept is language-agnostic; in Norwegian contexts the word gjetting means guessing, and the term is used to denote the degree to which outcomes are driven by guesswork rather than informative signals.

In a probabilistic model that combines an informed response with a baseline guessing distribution, the gjettingsparameter g is a mixing weight between the guessing component and the signal-driven component. Formally, if y is a response and s is a latent state, the model can write P(y) = g P(y|guess) + (1 - g) P(y|s). The guessing distribution P(y|guess) is often uniform over possible responses. The parameter g is constrained to [0,1] and reflects the extent of uncertainty or risk-taking in the decision policy.

Estimation and interpretation: g can be estimated from data by fitting the model with maximum likelihood or Bayesian methods. It plays a role similar to a temperature or exploration parameter in other frameworks, and high values indicate substantial random guessing or exploration, while low values indicate reliance on latent information. Identifiability can be challenging when the guessing distribution closely resembles the informed distribution, or when the data do not elicit clear state-dependent differences.

Applications: The concept appears in psychometrics, cognitive modeling, and adaptive experiments, as well as in any setting that deliberately blends random choice with informed action. Related ideas include the guessing parameter c in three-parameter logistic models and mixture-model formulations that separate guessing from state-informed responses.