The causal interpretation of estimated associations in regression models