Abstract:
|
In this panel, we plan to discuss the following:
- How do you decide which inconsistent patterns are allowable? How do you know if you're over-editing or under-editing? - Do any of the following criteria affect the amount of editing you do? - Whether you release micro-data. How likely is it that data users will point out complicated inconsistent patterns and ask you why you allowed them? - The number of affected cases. - The primary uses of the data. It seems that if the data are primarily used for estimates, then editing a few cases will probably not have much impact. If the data are primarily used for modeling, or for estimates for small subpopulations, perhaps more editing is called for. If the data are primarily used to prepare a dataset for public use, then perhaps extensive editing is called for. - Where should your edits be applied: by the survey instrument, or after the data are collected? If the former, how vigorously should the instrument enforce the edit rules? If the latter, how do you handle errors in skip logic (e.g., the respondent answers the filter question "No" but gives a valid response to the follow-up question)?
|