In order to obtain valid inference, the analysis of survey data requires special approaches to account for sampling design features. This is particularly true for analyzing complex survey data in which inclusion probabilities are not constant-as is the case, for example, in the National Educational Panel Study (NEPS). One way to achieve proper results-even when a statistical method does not explicitly account for survey design features-is the method of balanced repeated replication. This

more »

... logy provides a correct assessment of the variances for a wide range of estimators from stratified multistage sampling designs. Balanced repeated replication can be applied without further ado if so-called replication weights are available. To facilitate an unbiased variance estimation for NEPS data users, the NEPS methods group provides specific replication weights for the students participating in the first wave of the NEPS fifth-grade sample and for students participating in the first and the second wave of the NEPS ninth-grade sample as well. Additionally, replication weights are provided for fifth and ninth graders, for whom an interview with one parent could be realized. In this paper we describe how these weights have been derived and how they can be used to yield valid variance estimates.