Variance estimation in Complex Surveys

Specialeforsvar ved Christina Strand Blaabjerg 

Titel: Variance Estimation in Complex Surveys 

Abstract: The main object of this thesis is the investigation of Replication Methods in the context of variance estimation in complex surveys. When advancing a survey to non linear estimators the standard textbook variance estimator no longer suffices. One approach to this problem is the use of a linearisation method where an approximate solution is achieved. In this thesis the Taylor Series Method is presented. Another approach is the use of replication methods. The replication methods all produce a set of sub samples from the original sample which estimates a parameter of interest and together produce a variance estimator for the original sample proportional to the sum of squares. The Random Groups Method, The Jackknife Method, The Balanced Repeated Replication Method and The Bootstrap Method is presented. These methods will prove easy to apply and after being treated under different sampling designs they also prove flexible for many different situations. The above mentioned replication methods will be accepted based on an investigation on their empirical performance, their linear performance and their asymptotic properties. Last it is presented that replication methods also helps protect the identity of participants in a survey. The level of confidentiality gained from replication methods is briefly investigated. 

 

Vejleder: Anders Milhøj, Ø.I.
Censor:   Birger Stjernholm Madsen, Novozymes