<p>The paper examines the principles of applying regression analysis to prediction and analytical problems, as well as the specifics of predicting the consequences (effects) of controlling a modeling object. The following types (groups) of problems are identified: “passive” prediction of the target variable; retrospective reconstruction of the target variable’s value; prediction of causal effects; estimation of structural coefficients and causal influence coefficients; diagnosis and identification of cause-and-effect relationships; counterfactual analysis. The distinctive feature of active (causal) prediction is that it concerns a changed “world” that occurs after an intervention (the forced assignment of values to certain variables). The model of the object after the intervention differs from the original model in terms of local deformations. In some circumstances, selecting an appropriate set of regressors is sufficient to produce a correct active prediction. In more complex cases, this requires multi-step strategies and procedures that take into account knowledge about the structure of causal relationships. In particular, we describe a generalized version of the “back-door” criterion, an extended instrumental variable method for linear models, etc. We demonstrate the role of causal information and latent confounders.</p>