Advanced survey estimation methods for treatment of non-sampling errors 1
|Convenor||Dr Alina Matei (University of Neuchatel, Switzerland )|
|Coordinator 1||Professor Giovanna Ranalli (University of Perugia, Italy)|
In this work we consider estimation techniques from dual frame surveys in the case of estimation of proportions when the variable of interest has ordinal outcomes. We propose to describe the joint distribution of the class indicators by an ordinal model. Ordinal model assisted estimators and ordinal model calibration estimators are introduced for class frequencies in a population by using the two different approaches to estimation. Theoretical properties are investigated for these estimators. An application in a complex survey is also included.
Indirect sampling is an alternative approach to classical sampling theory in dealing with the problem of overlapping sampling frames on survey estimates. In this paper, a comparison is made between the optimal estimator of Deville and Lavallée and two classes of estimators - Domain Membership estimator and Unit Multiplicity estimator - translated into the context of Indirect Sampling. Next, a comparison is made of these estimators, taking in consideration the different kinds of cases that occur in a dual frame context, and the variance of the optimal estimator Deville and Lavallée is presented. Finally, the results of a simulation study
A new blocked-imputation method will be presented to deal with missing data in longitudinal surveys. Usually, information on why a unit is missing is recorded (e.g. refusal, noncontact, out-of scope), is recorded, but this information is then never used in the imputation process. The new approach used this information to improve the quality of the imputation process. The approach is illustrated with a simulation study and data from the British Household Panel Survey.
In the case of nonignorable nonresponse, the unit response probabilities
depend on the variable of interest. It is assumed that the variable of interest is sampled from a population which can be described as a mixture of some hidden subpopulations; a typical example of such a variable is income. Given the relationship between the variable of interest and the response probabilities, the latter can also be described by a such mixture. We consider that auxiliary information is available for all the sampled units. Two approaches that underline the mixture structure are considered to estimate the response probabilities through logistic regression.