Scientific Telephone Samples (STS) STS Sampling Tips Click here for RDD Sampling: Highly representative sample of the population.
Click here for Listed Sampling: Target specific sub-groups
Click here for Business Sampling: Target large or small businesses.

Join Our Mailing List

Enter your email address:
You will receive special offers, discounts, newsletters, and more!

Defining Replicate Size

There are no hard and fast rules as to how many records should be included in a single replicate subsample.

However, potential guidelines are evident, including observations from our ongoing industry experience and by looking at the purchase patterns of hundreds of sample buyers.

  • The consensus of research industry opinion – particularly among those with an academic orientation -- appears to be that replicates are not only good, but required to ensure proper administration of the sample and to help control the quality of data collection.

  • A surprising number of buyers, on the other hand, clearly indicate they do not wish to have their RDD samples subdivided into replicates.

Some buyers appear to customize and distribute the replicate subsamples themselves. Often, this occurs after a sorting procedure such as by time zone, census division, etc. Others claim to simply open the entire sample at the start of the job and proceed. We assume some of these rely on the CATI system itself to provide the appropriate sample management function.

For the record, if all numbers ordered for a project are dialed to exhaustion, it can be argued the replicate structure would be essentially unnecessary from the standpoint of data quality. However, some statisticians use the variability among the replicate subsample estimates to arrive at a more accurate standard error of the full-sample estimate.

  • Almost all buyers who do use replicates expect an equal number of records in each replicate. This is normally accomplished via a systematic procedure. It is possible, though, to establish unequal replicate sizes using random selection. As far as we can tell, this practice is essentially non-existent.

  • If the sample is screened for disconnects, modems, faxes, etc., sample records identified for exclusion may be marked and left in the sample, segregated from the “good” numbers (normally via sorting), or simply deleted entirely.

If records are deleted after screening, the replicate sizes will vary, although each will still contain the same number of “good” numbers as before.

  • Interestingly, the vast majority of clients prefer to simply eliminate the telephone numbers identified by the screening procedure.

This suggests the primary concern is mainly for sample efficiency.

  • Experience shows most RDD sample orders specify 50, 100, 250, 500 or 1,000 numbers in each replicate.

Much larger replicate sizes appear to be associated with very large projects, possibly in an effort to reduce staff time involvement while managing the sample.

  • There are theoretical arguments to suggest smaller replicate sizes are methodologically better. However, taken to extremes, very small replicate sizes may create an unrealistic expenditure of supervisor or IT staff time to manage the sample.

  • It is true that small replicate subsamples may help save money at the end of certain jobs. The reason for this is that some end users require all records within a replicate to be fully worked to the maximum number of attempts specified once the replicate is opened – even if just one record has been dialed.

Thus, from a cost standpoint, as the project draws down to the last few interviews, there may be a strong financial benefit to complete the job without opening much more sample. The smaller the replicate size, the less sample will have to be opened and fully worked to the maximum number of attempts.

  • It has also been suggested to base the replicate size on an easy-to-think-about fractional portion of the number of dialing attempts which your call center could make during a typical interviewing shift for the project.

The idea is to evenly match a shift’s dialing capacity for a single day’s work on a study with one or more replicates which can be dialed all the way through.

This may help to simplify the administration of the samples and reduce the number of decision points for the on-site supervision staff.

  • From the viewpoint of call center management, running out of sample – even for just a few minutes – can be a confusing and expensive occurrence.

A good rule-of-thumb, therefore, is to make sure the replicate size is large enough so that opening just one or two replicates will provide enough fresh sample records for immediate needs.

A good starting place to think about the size of replicate subsamples is to …

Do it the good “old-fashioned” way.

Many companies set the quantity of RDD numbers in each replicate equal to the ending sample size for the study. Thus, if the number of completed interviews will be 400, and it is estimated that ten numbers will be needed for each complete, this method suggests ordering ten replicates of 400 numbers each.

This seems to be a simple and useful method to quickly scale replicate size to the scope of the work. Small studies would use small replicate sizes. Larger studies will have more numbers per replicate.

This method also has theoretical elegance in that the first replicate is “the sample” and every effort is made to obtain a completed interview from each telephone number on that list.

(Isn’t it interesting how “best practices” often come back to historical roots? The current vogue in research circles is all about “address based sampling.” Address based sampling is not new. It is just a rededication to original research principles – define your sampling frame (all households in a geographic area), draw a sample of those households, and then work that sample to exhaustion using multiple techniques and approaches.)

A final note about quality considerations while using replicates…

Even though cost is often an overriding consideration, it might be useful to sacrifice a small amount of dialing efficiency by making sure replicate #1 is open for the entire span of the project. That way, a substantial portion of the completed interviews will actually come from that original replicate – just like the more conservative, traditional approach.

STS will be happy to discuss your project in more detail, including replicate subsample sizes. Please call if you would like help on this subject.

 

How to Order a Sample
Placing your RDD, listed or business sample order is as easy as a phone call or an email to STS.

Phone: (800) 944-4-STS
Email: info@stssamples.com
or, request a quote here.

An STS representative will be happy to assist you.

Copyright 2013 Scientific Telephone Samples | Terms of Use | Privacy Policy | Site Map | Contact