Administering Replicates
Most samples are divided into a number of "replicates". Each by itself is an independent sample, comparable in all ways to each of the other replicates. Within replicates, the numbers are randomly shuffled to eliminate calling order bias. These procedures ensure that no complicated distribution scheme is needed. Just open or load the sample and begin with the first replicate.
For best results, work all of the way through the first replicate before moving to the next. If you are using paper sample, the pages of a replicate may be separated and passed out individually to interviewers. The sample can be returned to its original state at the end of the project by observing the page and replicate numbers printed on each page.
If your sample cuts across several time zones and you have had the sample sorted by time zone you should be particularly careful to fully work each replicate. Otherwise, your completed interviews may have a time zone bias - normally due to making more callback attempts in your own time zone. This same degree of care should be taken if the sample is sorted in any other manner than in a purely random fashion.
An appropriate callback scheme (using different days and times of day) should be implemented to ensure the agreed upon number of attempts are made to all "live" numbers. For private sector studies, four attempts is a commonly accepted standard. Typically, you will find that the second attempt to working numbers will yield the highest interviewer production rate. This is because you already will have eliminated the disconnected numbers and many of the fax, modem, and business numbers.
If the time constraints for your survey are such that callback attempts cannot be made on different days, some effort should be made to perform at least a second attempt on all numbers to help reduce potential non-response bias. Also remember you may need to order extra numbers if you are not able to make the full four attempts to the entire sample.
|
