Open edvinf opened 4 years ago
Will be discussed in Nov.
In the work with selectionmethods, another possibility for solving this issue was suggested. We may require the complete recording of non-response in order for probabilistic selection methods to be recorded. If we add that requirement to the use of probabilistic selection-method, the DEhierarchyCorrect could be clearly defined to only indicate if the hierarchy is well identified.
We have used the field DEhierarchyCorrect to encode if the hierarchy is clearly identified, and whether estimations can be done based on the design. Interpretation seem to differ. Some use this field only for encoding the correctness of the hierarchy. Others use it also to encode if they do not think design based estimates are rigorously possible. I think we have experienced over the last few years that there is great value in identifying the hierarchy even in the case when sampling is not probabilistic. Most other concerns that needs to be documented in order to ensure that design based estimation can be done are handled by other fields, such as selectionmethod and fields for design variables. In my opinion we also need to document that non-response or missing samples have been put into the data. I therefore suggest that we change the definition of DEhierarchyCorrect to only identify whether the hierarchy is clearly and correctly defined, and introduce a new field to encode if the recording is complete also with respect to "missing" samples (non-response etc.)