Der Kontext ist eine Akademie über Kausalität (Mathematik). Hier ein Auszug aus https://stat.ethz.ch/people/jopeters/index/edit/causalityHomepage/causality_files/scriptChapter1-4.pdf:
Causal questions also appear in biological data sets, where we try to predict the effect of interventions (e.g. gene knock-outs). Kemmeren et al. [2014] measures genome-wide mRNA expression levels in yeast, we therefore have data for p = 6170 genes. There are nobs = 160 “observational” samples of wild-types and nint = 1479 data points for the “interventional” setting where each of them corresponds to a strain for which a single gene k ∈ K := {k1, . . . , k1479} ⊂ {1, . . . , 6170} has been deleted.
Was heißt "interventions" in diesem Kontext auf deutsch?
edit: Später, in der selben PDF:
We use the SEM (structured equation model) to define not only the distribution of observed data but also so-called interventional distributions (see Remark 2.2.5, for example). These are formally defined in Definition 2.2.1.
und dann
We are now ready to use the structure of SEMs to construct the “other distributions” P˜X from P X. Definition 2.2.1 [Intervention Distribution] Consider a distribution P X that has been generated from an SEM S := (S, P N). We can then replace one (or more) structural equations (without generating cycles in the graph) and obtain a new SEM S˜. We call the distributions in the new SEM intervention distributions and say that the variables whose structural equation we have replaced have been “intervened on”. We denote the new distribution by P X S˜ = P X | do(Xj=f˜( PA˜ j ,N˜ j )) S . The set of noise variables in S now contains both some “new” N˜’s and some “old” N’s and is required to be mutually independent. When ˜f( PA˜j, N˜j ) puts a point mass on a real value a, we simply write P X | do(Xj=a) S and call this a perfect intervention. An intervention with PA˜ j = PAj is called imperfect. It’s a special case of a stochastic intervention [Korb et al., 2004], in which the marginal distribution of the intervened variable has positive variance.