Editor's note: Beth Wallace is manager of product marketing at Survey Sampling, Inc.

The 1990s will bring an increasingly cost-conscious and competitive environment for providers and buyers of survey research. Researchers will face additional pressure to maintain research quality while containing or reducing research costs.

One area that should be of top concern to researchers is the efficiency of the random digit samples they use to conduct telephone research. The efficiency level will significantly affect sampling and data collection costs and affect the overall research study budget. The more efficient the sample, the less the sample will cost to purchase and administer in the field.

Various techniques have been employed over the last decade to increase efficiency levels, including stratification techniques and purging business numbers. Recently, a new technology has been introduced by Survey Sampling, Inc. to push the efficiency level of random digit samples to 80% or better.

How sampling efficiency affects research costs

With random digit telephone samples, the level of efficiency refers to the proportion of sampling units that will reach residential households. This proportion is usually referred to as the "working phones rate" of the sample. Naturally, the nature of randomizing phone numbers to contact unlisted households will result in disconnected, unassigned, and business telephone numbers being created and included in your sample. It is precisely this group of unproductive telephone numbers that needs to be kept to a minimum which, conversely, will provide a high working phones rate.

The efficiency of random digit samples affects research costs in two main areas-initial sampling costs and data collection costs. Each of these areas requires some explanation.

Initial sampling costs are immediately affected by the working phones rate of the sampling methodology. The lower the working phones rate, the more sampling units will need to be purchased to complete the desired number of completed interviews. The hypothetical situation presented below illustrates the point.

   

Sample Type I

   

Sample Type II

Completed interview quota

1000

   

1000

Working phone rates

55%

   

65%

Anticpated incidence

80%

   

80%

Anticipated cooperation

40%

   

40%

Sampling units required

5682

   

4808

Sample Type I, having an expected working phones rate of 55%, will require 18% more sampling units than Sample Type II, which has an expected working phones rate of 65%. Clearly, a bigger sample is needed if a lower working phones rate is anticipated.

Data collection costs are directly affected by unproductive numbers included in the random digit sample. Dialing attempts made to unproductive numbers may cost more than many researchers think. Continuing with our hypothetical situation will demonstrate the waste associated with dialing bad numbers.

   

   

Sample Type I

   

Sample Type II

Sampling units dialed

   

5682

   

4808

Unproductive dialing disconnects

   

29% or 1648

   

21% or 1010

Business
times 3 attempts

   

11% or 625
1875

   

8% or 385
1155

Total unproductive dialings

   

3523

   

2165

Due to its lower working phones rate, Sample I will require 1358 more dialings to unproductive numbers than required with Sample II. Dialings to business numbers exacerbate the situation since consumer samples are usually dialed in the evening hours and business numbers will, in most cases, not be discovered to prevent wasted callbacks.

The cost to make unproductive dialings will vary from company to company and location to location. Interviewer salary, supervisor time and related overhead all must be considered. Estimates by large and small research companies suggest the cost of a dialing ranges from 50 cents to 90 cents. If 69 cents, a well-documented figure presented to Survey Sampling, is used to continue our example, the cost impact is clear:

   

   

Sample Type I

   

Sample Type II

Total unproductive dialings

   

3523

   

2165

Estimated cost of dialing

   

$.69

   

$.69

Cost to dial unproductive numbers

   

$2430.87

   

$1493.85

By using the less efficient Sample Type I, an additional $937.02 in data collection costs will be incurred.

Greater efficiency can provide other benefits as well. With fewer unproductive numbers to dial, studies can be completed with shorter field times. In addition, encountering fewer bad numbers can boost interviewer morale.

Methodologically, efficiency has peaked

Given the significant impact of efficiency on research costs, it is no wonder that research and development to identify additional ways to improve efficiency continues.

In terms of random digit sampling methodology itself, efficiency has peaked. During the last decade, when computers allowed sampling to become a sophisticated state of the art process, various methods were employed to provide highly efficient random digit samples, including:

  • frequent database updates to freshen area code, exchange, and working block data,
  • cleaning out low utilization working blocks from the database,
  • stratification to area code, exchange, and working block in proportion to known directory listed households,
  • purging known business numbers using business database sources.

Further increases in efficiency cannot be developed methodologically without significantly risking the statistical integrity of the sample. We at Survey Sampling believe in maintaining the representativeness and projectibility of random digit samples and therefore have not moved in this direction. Rather we have broadened our scope and investigated ways to improve sample efficiency during the data collection phase of a study.

New inroads to greater efficiency

A new technology has been developed to identify a large portion of the unproductive numbers in random digit telephone samples without affecting the samples' integrity. Survey Sampling recently introduced the new "Sample Screening Service" after nearly two years of research and development.

The Sample Screening Service employs a proprietary methodology to identify disconnected and unassigned numbers in a sample before the sample is dialed during data collection. Importantly, the service involves a post-production process so the integrity of the sample is not compromised in any way. Random digit samples are selected by standard methodologies, then the standalone process is applied to identify and mark unproductive numbers.

The service identifies, on average, half of the disconnected or unassigned numbers in a sample, which results in a 10-15% increase in sample efficiency. Survey Sampling's normally high working phones rates of 65-75% is improved to an effective working phones rate of 75-85%.

The 10-15% additional increase in efficiency achieved by utilizing the Screening Service extends the many benefits of sample efficiency discussed above. Further cost savings during the data collection phase of the research study will be realized and so will shortened field times, better interviewer morale, and a generally more efficient operation.