Identification of causal effects in linear models: beyond instrumental variables