Sampling Pool

Sampling pool is a survey operations term, one that statisticians sometimes refer to as the designated sample size, which was proposed by Paul J. Lavrakas in the 1980s to refer to the set of elements selected from a sampling frame that may or may not all be used in completing data collection for a given survey project. The value of using this term is to be able to have a unique term to differentiate the sampling pool that a researcher starts with from the final sample (i.e. the final sample size) the researcher finishes with. Traditionally, survey researchers have used the word sample to refer to both the final number of completed interviewers a survey is striving to attain and the number of elements used to gain those completed interviews. Because noncontacts, nonresponse, and other reasons (e.g. ineligibility) cause many elements in a sampling pool to not end as completed interviews, the final sample is essentially always smaller in size than the sampling pool and, in many cases, is substantially smaller, for example, 1/10 or 1/20 the size of the sampling pool.

For example, if researchers have estimated that they will need 10,000 telephone numbers for a random-digit dialing (RDD) survey that has a goal of completing 800 interviews, the survey call center that does the interviewing may not need to activate all of those numbers during the data collection period. That is, their processing of the RDD numbers toward the goal of 800 completions may be more efficient than expected, and they may not need to activate all the numbers that were selected for the sampling pool. To allow the sample coordinator the ability to closely manage the sampling, typically all the numbers in the sampling pool will be divided into sample replicates. If, for example, the sampling pool contained 10,000 RDD numbers made up of 100 replicates, then each replicate would contain a random subset of 100 of the 10,000 numbers. The sample coordinator may start data collection by releasing half of the replicates (thus a random half of the numbers in the sampling pool) on the first day of the survey's field period. Then the coordinator might observe for the next day or two how efficiently the interviewers are able to process the released numbers in achieving completed interviews. If the efficiency is better than the researchers anticipated, then the coordinator may only need to release another 30 replicates (another 3,000 numbers) to attain the final sample size goal of completed interviews for this survey. Thus, in this example, 2,000 numbers (i.e. 20 replicates) from the sampling pool would never be dialed by interviewers. Of further note, these unreleased numbers would not be considered in any response rate calculations the researchers performed after data collection was completed.

To estimate the size of the sampling pool a given telephone survey needs, Lavrakas advised use of the following formula:

Here FSS stands for the final number of completed interviews the survey must attain; HR stands for the hit rate, or the estimated proportion of the sampling pool that will reach residences; REC stands for respondent exclusion criteria, or the estimated proportion of households that will be deemed ineligible for the particular survey; and LE stands for loss of eligi-bles, or the estimated proportion of eligibles that will end as nonresponders. For example, if in an RDD survey 1,000 completed interviews are desired and the HR is known to be about .65 (65% of the numbers will be households), REC is .05 (5% of households will not have an eligible adult in residence), and LE is .75 (75% of the eligible household will not complete an interview due primarily either to refusals or non-contacts), the estimated size of the sampling pool needed to complete this survey would be (1000)/((.65)(1 − .05)(1 − .75)) or 6,478 RDD numbers. Thus to be [Page 794]on the safe side, the researchers might decide to start with a sampling pool of 8,000 numbers.

...