Restusikkerhet

Restusikkerhet, or residual uncertainty, refers to the part of uncertainty in a forecast or estimate that remains after accounting for the factors explicitly included in a model. It reflects random variation, unobserved influences, and potential misspecification of the model form. In practice, restusikkerhet is what the error term in a statistical model represents.

Restusikkerhet is distinct from the uncertainty explained by known variables. It encompasses measurement error, sampling error, omitted variables, and structural uncertainty in the model. The concept is closely linked to the idea that no model perfectly captures reality, and some variability cannot be explained by the chosen inputs or the assumed relationships.

Quantifying restusikkerhet often involves examining model residuals. In regression, the variance or standard deviation of residuals provides an estimate of the unexplained error variance. Prediction intervals are built from this residual uncertainty to express a range of plausible future values. Residual diagnostics, such as residual plots and tests for heteroskedasticity or normality, help assess whether restusikkerhet is adequately captured by the model assumptions.

Managing restusikkerhet entails reducing its sources where possible, and transparently communicating what remains. This can include improving data quality, adding relevant variables, and using more flexible models. Techniques such as cross-validation, bootstrapping, Bayesian methods, and model averaging help quantify and incorporate restusikkerhet into decisions.

Applications span economics, finance, engineering, and policy analysis. A central point is that some uncertainty is irreducible due to inherent randomness or unknown factors, while other parts arise from model limitations and data gaps.