Yet another look at the omitted variable bias