Modeling the population in survey sampling problems continues to be controversial. An important reason is that the likelihood principle makes it somewhat necessary to model the population. Estimating the population total in two-stage survey sampling is considered, making use of a "superpopulation" model. The problem is then really one of predicting the unobserved part of the total. The concept of predictive likelihood is studied. A non-Bayesian likelihood approach to prediction problems in general is used. Prediction intervals and a predictor for the population total are derived for the normal case, based on predictive likelihood. (TJH)