Determining How Many RDD Numbers To Purchase

There are two basic approaches toward deciding how many RDD numbers you will need.

Traditional Industry Approach

The traditional industry approach is to multiply the expected "hit rate" (the % of a sample which will be working residential telephones) times the net effective incidence (NEI) times an expected cooperation rate (varies from 30% to 40% depending on the telephone bank). Then, you must divide the resulting figure into 100 to arrive at an estimate for how many numbers will be needed. Normally, it is assumed the "hit rate" from major sampling suppliers will range from 65% to 70%.

For instance, at a hit rate of 70%, a 60% incidence and a 30% respondent cooperation rate the calculations would look as follows:

• 70 (hit rate) * .60 (incidence) * .3 (coop rate) = 12.6
• 100 divided by 12.6 = 7.93 numbers needed per complete

This 7.93 figure is then adjusted judgmentally to account for factors such as questionnaire length, topic of the study, time of year (are people Christmas shopping?), whether or not the questions are phrased in an interesting manner which is not likely to result in a terminate, any special characteristics of the geographic area to be sampled (are refusal rates typically higher in this location?), the degree to which call backs will be employed, and so forth.

Approach Based On A Regression Model

The second approach is to use the results of a multiple regression performed on sample dispositions from a large telephone bank. The formula from this analysis yields the following equation:

• RDD #s = a constant of  3.5 + the number 295 divided by the net effective incidence

Using the previous example, about 8.42 RDD numbers per completed interview would be needed as follows:

• 295 divided by 60 = 4.92
• 4.92 + 3.50 = 8.42

Once again, this figure would be adjusted to account for the various factors such as questionnaire length, the number of call backs, the subject matter of the interview, etc.

Then what should I pick?

The chart on the next page compares both of these techniques. Notice the approaches are very close to each other in the mid-ranges of incidence.

Given the fact there always will be language problems, over-quota situations, minimum and/or maximum age qualifications, and so forth, which should be figured into the net effective incidence (NEI), there is rarely any project with more than a true 90% incidence. When the incidence is high, our advice is always to buy a few more numbers than you feel you will need. The incremental cost will not be great and — at 5¢ a number or less — it is best to have a safety margin rather than starting through the order process once again.

On the other hand, the number of RDD numbers you might need is quite difficult to predict below 5% incidence. Our suggestion is to always "under-buy" when you are looking for very hard-to-find respondents. You can always order more numbers as the project reaches the mid-point and the true level of incidence can be predicted.

Per Number Guidelines For RDD Samples

Net Effective Incidence	Regression Approach	Traditional Approach  (30% Cooperation)	Traditional Approach  (35% Cooperation)	Traditional Approach  (40% Cooperation)
5%	62.5 #'s	98.0 #'s	84.0 #'s	73.5 #'s
10	33.0	49.0	42.0	36.8
15	23.2	32.7	28.0	24.5
20	18.3	24.5	21.0	18.4
25	15.3	19.6	16.8	14.7
30	13.3	16.3	14.0	12.3
35	11.9	14.0	12.0	10.5
40	10.9	12.3	10.5	9.2
45	10.1	10.9	9.3	8.2
50	9.4	9.8	8.4	7.4
55	8.9	8.9	7.6	6.7
60	8.4	8.2	7.0	6.1
65	8.0	7.5	6.5	5.7
70	7.7	7.0	6.0	5.3
75	7.4	6.5	5.6	4.9
80	7.2	6.1	5.3	4.6
85	7.0	5.8	4.9	4.3
90	6.8	5.5	4.7	4.1

Remember: There is rarely a project with more than a true 90% incidence.

Click here to download this article.

Contact Us:
Scientific Telephone Samples
27121 Towne Centre Drive
Suite 290
Foothill Ranch, CA 92610

(800) 944-4-STS
(949) 461-5400
Fax (949) 609-4577

email us at:
info@stssamples.com
or, stssamples@aol.com
 

[Capabilities] [RDD Pricing] [RDD Sampling Tips]