The fade away effect of initial nonresponse bias in regression analysis