The first problem addressed is definition of the sampling frame when repeated sampling is attempted and part of the population is transient. Sampling the transient population may lead to analytical bias if the minimum length of survey participation is not satisfied. Practical methods to reduce longitudinal sampling problems include (1) sample replenishment of new subjects with matching characteristics, (2) sample compensation, (3) shortening the preenrollment survey period, (4) using crossover designs that permit experimental subjects to transfer from control groups to treatment groups, and (5) connecting cross-sectional surveys with a panel survey. Another problem, disproportionate sampling of populations of greater interest or with certain characteristics, can be avoided by carefully reviewing survey modeling assumptions and the likelihood of misclassification. In addition, the allocation of subjects to treatment condition to obtain optimal survey by sampling from certain portions of a variable distribution if a functional form is known (e.g., linear responses in experimental findings that provide minimum variance estimates through balanced distribution of high and low responses). However, risks are involved in trusting the nature of the responses, and crossover designs rarely can be used to advantage in social experiments. A discussion of the fourth problem, identified as balancing covariates when assigning subjects to treatment modes in the presence of field constraints, concludes that blocking, proportional stratification, and the Finite Selection Model can all be used to improve the balance and precision of the estimates from an experiment. However, the gains that they provide relative to simple random sampling are reduced by random nonacceptance rate. Seven references, footnotes, tabular data, and a technical appendix are provided. (Author abstract modified)