Distributional biases in language families

This paper introduces a method (the Family Bias Method) that estimates statistical biases in diachronic developments on the basis of synchronic samples. Estimates of developments are sought from their expected synchronic results: if a structure S outnumbers non-S significantly in a family, a change toward S in this family was more likely than a change away from it – either because S was there in the protolanguage and then hardly ever got lost, or because S was not there and then it was was innovated early or often in the family. If structures are balanced in a family, no signal can be inferred: unless we know the protolanguage, diversification could have proceeded in any direction. Using extrapolation methods, signals for diachronic biases can also be estimated for isolates and small families, which represent over half of the known families. If most families show a bias, but there are as many biases toward S as against S worldwide, this suggests genealogical stability. If most biases are in the same direction, this suggests a diachronic universal or, if they are limited to certain geo-historical regions, areal trends or hotbed effects. Evidence for the theoretical assumptions of the Family Bias Method comes mainly from demonstrating that synchronic distributions in families cannot be successfully explained by alternative approaches, such as those assuming general stability indices.