 |
|
|
|
||||||||||
|
|||||||||||||
|
|
|
||||||||||||||||||||
|
||||||||||||||||||||||
|
How Should One Specify Replicate Sizes? The decision of how many RDD numbers should be included with a single replicate is a matter of individual choice. The only limitation is that there should be an equal count of numbers in each replicate. Looking at the way various companies specify their replicates, there seem to be three methods in common use: Option #1: Make The "Rep" Size Equal To The Sample Size Many companies specify that the quantity of RDD numbers within each replicate be equal to the sample size set for the study. Thus, if the sample size of completed interviews will be 400, and it is estimated that six numbers will be needed for each complete, they would order six replicates of 400 RDD numbers each. The theory here is that the first replicate is "the sample" and every effort is made to obtain a completed interview from each telephone number on that list.
Option #2: Have The "Rep" Size Based On The Sample Size, But Inflate It To Account For Both Incidence And The Expected "Hit" Rate A second method is to base the replicate size on the eventual sample size, but to anticipate NQR's based on the expected incidence of qualifiers and the likely "hit rate" expected for the sample. The thought here is that even though it is "nice" to try, it is unrealistic to expect a completed interview from each number in the primary sample. Therefore, why start with an impossible number as the designated "sample." The procedure for this method is to divide the ending sample size by the net effective incidence and then again by the "hit rate." For instance, for a 400 sample size with a 60% net effective incidence and a 70% "hit" rate, the replicate size should be about 952.
Option #3: Base The "Rep" Size On About Half The Estimated Number Of Dialings Anticipated During A Calling Session This method is particularly good for large telephone banks. The idea is to work through the first replicate before adding a second, the second before the third, etc. This means the replicate size should be formulated with some idea of how many numbers are used by the telephone bank within a fixed period of time. Depending on the questionnaire length, incidence, etc., figure that each interviewer will dial about 20 to 40 numbers or so per hour for a typical job. With 20 interviewers working on a single project during a four hour evening shift, therefore, between 1,600 to 3,200 RDD numbers would be utilized in one night. Of these, perhaps half — 800 to 1,600 — would be available for subsequent calling the following evening. That would mean that about 800 to 1,600 "new" numbers would need to be brought in from "reps" not used the first night. If you work the mathematics out for your size of telephone bank and the length of your calling periods, you may find that this "one half" algorithm will work fine. All things considered, the size of the replicate is not that crucial a consideration. In fact, if all of the numbers ordered for a project are used to the point where they are essentially exhausted, the replicate structure would be meaningless. Also, most CATI programs draw from available numbers in either a random or interval fashion, somewhat mitigating the need for replicates at all. If maintaining replicate structure is important to you, make sure to have your CATI system pull from the sample by replicate. STS will be happy to suggest replicate sizes for you. Please do not hesitate to call your representative if you would like their help.
|
||||||||||||||||||||||
