Assessing bias in demographic estimates from joint live and dead encounter models