Editorial—The Anna Karenina Bias: Which Variables to Observe?

The opening of Count Lev Nikolayevich (Leo) Tolstoy's novel inspired linguist, molecular physiologist and biogeographer Jared M. Diamond's eponym for the Anna Karenina principle (Diamond 1997). The principle suggests that no one property guarantees success but many guarantee failure. The Anna Karenina (TAK) bias is a logical consequence. TAK bias is more insidious than the kindred Survivor bias, which cautions that measured variables for passively observed survivors often differ from easily overlooked nonsurvivors. TAK bias, in contrast, cautions that the observed variables themselves might differ for survivors. The most revealing variables might exhibit negligible variation among survivors because survivors are necessarily alike. Perhaps variability is inversely related to the variable's importance for survival. TAK bias is more problematic for descriptive research, in contrast to normative (i.e., prescriptive) research, which seeks the true causal variables. Normative research only offers conditional aid to the decision maker on specific variables. To avoid TAK bias, we must not passively let accountants decide which variables we observe. We must actively collect data guided by predictions from deductive theory.