Article in English.
Abstract and further information, follow this link:
With overlapping questionnaires design, called split questionnaire design by, the total sample is subdivided, with different sub-sample exposed to different, but overlapping sections of the total qustionnaire. the design is a flexible tool to collect complex data without exhausting the respondents. Overlapping questionnaire design may also be used to coordinate several related surveys.
Data collected by overlapping questionnaires have many occurrences of missing data because each unit is only exposed to a subset of all items. These missing data are missing completely at random. The treatment of such missing data is d iscussed in the context of estimating generalized linear models. Because of the liearity, regression imputation involving covariates only comes naturally. Weights reflecting the relevant uncertainty in the imputation should be used when fitting the generalized model. When the imputation is linear in the observables, imputation is avoidable, and a weighted mean of estimates from fitting the model in the observables in groups of units with the same pattern of missingness, will be asymptotically efficient in the linear case. For binary regression, the probit link is attractive.
Experiences from a survey of physician health and welfare conducted by an overlapping questionnaire design are reported.
Contact us for more information.