NullLiftWert

NullLiftWert is a statistical concept used in the analysis of association rules and lift-based measures. In the sense of a baseline, it designates the lift value expected under the null hypothesis of independence between two events or items.

Definition and calculation: The lift of two events A and B is defined as lift(A,B) = P(A,B) / [P(A) P(B)]. Under the null hypothesis that A and B are independent, P(A,B) = P(A)P(B), hence the theoretical NullLiftWert equals 1. In empirical data, the observed lift is compared to this baseline. The null distribution can be estimated by permutation, bootstrap, or analytical approximation to assess statistical significance.

Interpretation and use: A NullLiftWert close to 1 indicates no detectable association. Observed lift values greater than 1 suggest positive association; values less than 1 suggest negative association. Significance testing against the NullLiftWert is common in rule mining and market basket analysis.

Related concepts and context: The concept relates to confidence, support, leverage, and conviction, and to statistical tests of independence such as chi-square. It is used in domains including retail analytics, recommender systems, and text mining. Caveats include sensitivity to imbalanced marginals and sample size; NullLiftWert is not a universal constant across datasets. See also lift, association rule learning, and permutation testing.